death to auto-widening yellow brick roads

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

Agreed, mostly. I derailed on my weight goal not that long ago, so got
some new perspective on this.

The road width should continue to be related to the variability.

I sometimes have a 1kg delta (in either direction) from day to day, for no
obvious reasons. So I’m glad that even when the old can’t-lose-widening isn’t
in effect, the road width isn’t strictly governed by my slope, but
accommodates random fluctuations. (90% of all daily variance, I believe we
claim somewhere.) My current (below the road) road lane-width is 0.75kg,
which tells you something about my daily weight fluctuations.

My “weight loss” slope is -0.1 kg per week, enough to ensure that I don’t
weigh 300lbs when I hit 50, and gentle enough that a series of small
sustainable life changes are sufficient to keep me on the road.

I’m trying to imagine this goal without a variable width. Under a
fixed-width regime, the road feels it’d be like what the roadwidth post
described as a “fixed bright line”. Each lane on my road would be 14 grams
wide. The Withings scale publishes my weight in 50 gram increments. It
seems unlikely that I will see any graph colours other than red and green.
(Maybe it works better for more aggressive slopes, and my perspective is
atypical.)

To be 90% safe, I’d need to stay stupidly far below the road, with about 50
"safe-days". With variable road-width but no magical-widening, that 90%
safe line is the centre-line. (Not that I stay anything like 90% safe
today, even with the wider lanes I spend more time in the wrong lane than
anywhere else.)

I fully agree that I only take drastic measures (like exercising) when I’m
in the red. So maybe 90% is the wrong number, but I do think that there’s
value in having a variance-based roadwidth. [1] Basing the width on slope
will pretty much eliminate the value of having lanes and turn the yellow
brick road into a fixed bright line.

But I’ve been wrong about how other proposed changes[2] will feel in
practice. I’m willing to be a guinea pig.

Slope-based widths work really well for do-more goals, where the units
measured are directly related to the units of slope. (That’s badly
expressed, sorry, because obviously the literal units of measure are the
same.) There may be a qualitative difference related to one being a
measure of the whole, and one being a measure of a delta. (I’ll ponder
that more and try again later.) With do-more there’s a direct connection
between going for a run and getting back on the good side of the
distance-run graph. The connection with weight loss is less direct.

Philip

[1] for consistency, that’s the unit that should be used to calculate safe
days, and the assumed-gain when flatlining. Similar to our set-a-limit
aspirations. Flatlining doesn’t really apply to weight loss goals; like
set-a-limit, the non-act of non-reporting should rapidly result in
derailment, just to force a datapoint.

[2] one of my favourite features is pcrc [3], for example, and I initially
worried that it would feel like being on a runaway train.

[3] pre-commit to re-commit. q.v. http://blog.beeminder.com/recommit/

On Saturday, 27 July 2013 02:50:52 UTC+1, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

A lot of good points here! Especially your point about how a very
shallow road will be razor thin and all your weight loss datapoints
will tend to be either green or red. I may be willing to bite the
bullet on that. At least you get the proper color transitions (green
-> blue -> orange -> red) when not reporting at all (ie, flatlining –
blog.beeminder.com/glossary#flatline ).

I also like your idea of applying pessimistic presumptive reports to
weight loss roads. First we need to deploy set-a-limit pessimistic
reports globally (it’s still only turned on for a handful of intrepid
souls right now, due to some untied loose ends with it).

One thing to keep in mind while debating this is The First Rule of Beeminder:
http://blog.beeminder.com/catchup/
Key quote: Better to push it / procrastinate up to a bright,
unambiguous line than to push it / procrastinate up to a line whose
exact location is shrouded [or confused or complicated].

So any cleverness with variable road width can only hurt you in the
long run. We did it originally to address a specific psychological
issue but that issue doesn’t exist anymore.

Last point for now: There’s certainly value in a visual indicator of
your variance, but I think the turquoise swath could better fill that
role.

On Fri, Jul 26, 2013 at 11:55 PM, pjh philip@hellyer.net wrote:

Agreed, mostly. I derailed on my weight goal not that long ago, so got some
new perspective on this.

The road width should continue to be related to the variability.

I sometimes have a 1kg delta (in either direction) from day to day, for no
obvious reasons. So I’m glad that even when the old can’t-lose-widening
isn’t in effect, the road width isn’t strictly governed by my slope, but
accommodates random fluctuations. (90% of all daily variance, I believe we
claim somewhere.) My current (below the road) road lane-width is 0.75kg,
which tells you something about my daily weight fluctuations.

My “weight loss” slope is -0.1 kg per week, enough to ensure that I don’t
weigh 300lbs when I hit 50, and gentle enough that a series of small
sustainable life changes are sufficient to keep me on the road.

I’m trying to imagine this goal without a variable width. Under a
fixed-width regime, the road feels it’d be like what the roadwidth post
described as a “fixed bright line”. Each lane on my road would be 14 grams
wide. The Withings scale publishes my weight in 50 gram increments. It
seems unlikely that I will see any graph colours other than red and green.
(Maybe it works better for more aggressive slopes, and my perspective is
atypical.)

To be 90% safe, I’d need to stay stupidly far below the road, with about 50
"safe-days". With variable road-width but no magical-widening, that 90%
safe line is the centre-line. (Not that I stay anything like 90% safe
today, even with the wider lanes I spend more time in the wrong lane than
anywhere else.)

I fully agree that I only take drastic measures (like exercising) when I’m
in the red. So maybe 90% is the wrong number, but I do think that there’s
value in having a variance-based roadwidth. [1] Basing the width on slope
will pretty much eliminate the value of having lanes and turn the yellow
brick road into a fixed bright line.

But I’ve been wrong about how other proposed changes[2] will feel in
practice. I’m willing to be a guinea pig.

Slope-based widths work really well for do-more goals, where the units
measured are directly related to the units of slope. (That’s badly
expressed, sorry, because obviously the literal units of measure are the
same.) There may be a qualitative difference related to one being a measure
of the whole, and one being a measure of a delta. (I’ll ponder that more
and try again later.) With do-more there’s a direct connection between
going for a run and getting back on the good side of the distance-run graph.
The connection with weight loss is less direct.

Philip

[1] for consistency, that’s the unit that should be used to calculate safe
days, and the assumed-gain when flatlining. Similar to our set-a-limit
aspirations. Flatlining doesn’t really apply to weight loss goals; like
set-a-limit, the non-act of non-reporting should rapidly result in
derailment, just to force a datapoint.

[2] one of my favourite features is pcrc [3], for example, and I initially
worried that it would feel like being on a runaway train.

[3] pre-commit to re-commit. q.v. http://blog.beeminder.com/recommit/

On Saturday, 27 July 2013 02:50:52 UTC+1, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
"Akratics Anonymous" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

On Sat, Jul 27, 2013 at 10:47 PM, Daniel Reeves dreeves@beeminder.comwrote:

So any cleverness with variable road width can only hurt you in the
long run. We did it originally to address a specific psychological
issue but that issue doesn’t exist anymore.

I kept debating whether to jump in on this discussion, but I guess I will
now. Because this bit - the psychological issue not existing anymore - is
something I just don’t understand (yet?).

My caveat: I’m mostly a fan of the auto-widening road because of the way
I’ve been using it for a daily measure of blood glucose control, and that’s
something that cannot be fixed with the ‘daily minimum’ change, because the
measure is always calculated over a whole day. But this is a corner case,
obviously. (…And my views in this case are complex because I’m also not
convinced that the auto-widening road isn’t too generous, in this case. A
really bad day has the potential to make the road really wide, and it takes
too long (although no longer with retroratchet, perhaps!) for the road to
narrow again.)

But aside from my corner case, I also don’t quite see how the ‘daily
minimum’ change fixes the problem for weight-loss goals. Sources I’ve read
in the past* say that first-thing-in-the-morning weight tends to be the
lowest of the day, and that’s been my (anecdotal) experience. Short of
unhealthy solutions that I don’t even want to suggest, I don’t see how
people can “claw your way back onto the road by the end of the day.” But
perhaps I’m missing something?

Jana

*A quick Google now isn’t giving me anything with a credible cited source,
but it is confirming that this advice seems to be standard (e.g.,
herehttp://www.medicinenet.com/script/main/art.asp?articlekey=55489and
herehttp://www.lifehack.org/articles/lifestyle/do-you-make-these-10-common-mistakes-before-weighing-yourself.html).
So, grain of salt.

Ooh, this may be a corner case but it’s an awesome one. Maybe we can
keep auto-widening around as an option for custom goals, along with
custom road widths.

You bring up a great example, though, of why auto-widening is really
hard (or perhaps impossible) to get right. If we can’t do it right
then it’s probably better to just be explicit and transparent, namely,
it’s just up to you to set your own hard cap, to set the yellow brick
road high enough to accommodate your fluctuations. We’ve heard good
arguments why that’s not very satisfying to have almost all your
points below the road, but it may be the lesser of two evils. (Bonus
for us if we get rid of the expectation that the road accommodates
random fluctuations: not having to field support emails about how a
derailment wasn’t legit because it didn’t!)

I did just tick off one objection: You don’t need the road itself as a
visual indicator of your variance anymore, because the width of the
turquoise swath (aka blue-green aura) now uses that algorithm.

As for clawing oneself back onto the road, I’ll quote myself from last
summer, when I was skating the edge of my weight road. (Interestingly,
I haven’t had to do that since beeminding my sugar consumption. And,
yes, everything below is not officially recommended!)

At 9:40 this morning I weighed 71.85 kg (158.4 lb). That was a pound
or so above my weight road, bmndr.com/d/mass .
So I jumped around the house and did squats and burpees and things
until I had sweated out about 2 soda cans’ worth of sweat.
Here’s what our scale sent to Beeminder for the hour and half that I
repeatedly weighed myself to see if I was on the road yet:

24 71.85 "auto-entered from withings scale at 09:40"
24 72 "auto-entered from withings scale at 09:40"
24 72.05 "auto-entered from withings scale at 09:52"
24 72.05 "auto-entered from withings scale at 09:55"
24 71.85 "auto-entered from withings scale at 10:15"
24 71.85 "auto-entered from withings scale at 10:21"
24 71.95 "auto-entered from withings scale at 10:27"
24 71.75 "auto-entered from withings scale at 10:32"
24 71.85 "auto-entered from withings scale at 10:33"
24 71.9 "auto-entered from withings scale at 10:37"
24 71.9 "auto-entered from withings scale at 10:40"
24 71.85 "auto-entered from withings scale at 10:42"
24 71.65 "auto-entered from withings scale at 10:47"
24 71.4 "auto-entered from withings scale at 10:54"
24 71.55 "auto-entered from withings scale at 10:56"
24 71.4 "auto-entered from withings scale at 11:00"
24 71.55 "auto-entered from withings scale at 11:03"
24 71.5 "auto-entered from withings scale at 11:05"
24 71.25 “auto-entered from withings scale at 11:09”

That last one put me back on the yellow brick road, barely.

What I’m wondering is if there’s a good rule of thumb for how to game
Beeminder like that without endangering one’s health.
I’m thinking something like this:

  1. Don’t overdress in order to sweat more. You want the side effect
    of burning a lot of calories.
  2. Don’t do this multiple days in a row. Your true weight might be
    deviating more and more from the road, which means eventually you’ll
    have to dangerously dehydrate yourself to get back on.
  3. Listen to your body. If you feel thirsty, drink.

PS: I would’ve just not eaten all day and weighed in again at night
but Bethany made chocolate chip muffins so I was pretty motivated to
get on my road before leaving the house so I could eat them.

On Sun, Jul 28, 2013 at 12:31 PM, Jana Beck jana.eliz.beck@gmail.com wrote:

On Sat, Jul 27, 2013 at 10:47 PM, Daniel Reeves dreeves@beeminder.com
wrote:

So any cleverness with variable road width can only hurt you in the
long run. We did it originally to address a specific psychological
issue but that issue doesn’t exist anymore.

I kept debating whether to jump in on this discussion, but I guess I will
now. Because this bit - the psychological issue not existing anymore - is
something I just don’t understand (yet?).

My caveat: I’m mostly a fan of the auto-widening road because of the way
I’ve been using it for a daily measure of blood glucose control, and that’s
something that cannot be fixed with the ‘daily minimum’ change, because the
measure is always calculated over a whole day. But this is a corner case,
obviously. (…And my views in this case are complex because I’m also not
convinced that the auto-widening road isn’t too generous, in this case. A
really bad day has the potential to make the road really wide, and it takes
too long (although no longer with retroratchet, perhaps!) for the road to
narrow again.)

But aside from my corner case, I also don’t quite see how the ‘daily
minimum’ change fixes the problem for weight-loss goals. Sources I’ve read
in the past* say that first-thing-in-the-morning weight tends to be the
lowest of the day, and that’s been my (anecdotal) experience. Short of
unhealthy solutions that I don’t even want to suggest, I don’t see how
people can “claw your way back onto the road by the end of the day.” But
perhaps I’m missing something?

Jana

*A quick Google now isn’t giving me anything with a credible cited source,
but it is confirming that this advice seems to be standard (e.g., here and
here). So, grain of salt.


You received this message because you are subscribed to the Google Groups
"Akratics Anonymous" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

I just joined this list, and I’ll probably have something more substantial
to say about weight goals later after reading more of the discussion, but I
wanted to reply now because I object pretty strongly to the idea of letting
you weigh in multiple times in one day and taking the lowest value. I had
no idea (and now wish I still didn’t know) it worked that way now. I always
weigh myself only once, first thing in the morning. Later in the day when
I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5% of my weight for me -
it would be more for a larger person. If I’m off my road when I first weigh
in in the morning, I don’t think there’s anything healthy I can do to get
back on later in the day, and I definitely don’t want to incentivize myself
to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my gym-going
emergency day coming, but I won’t know that it’s a weight beemergency until
I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused people
to dread standing on the scale. So they’d not weigh in daily, which meant
they’d lose sight of their progress (or regress), which meant they’d be
more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument that
I’ll still step through the colours on my way to derailing doesn’t apply to
me. (And wouldn’t occur if presumptive-pessimism is implemented.) I think
that the turquoise swath itself might be vestigial; it used to clearly show
the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead of
paying slightly more attention to my diet when I’m orange, that will need
to become a rule like ‘if today I’m heavier than yesterday’ or ‘if fewer
than 2 safe days’ or something less obviously derived from the graph and
its colouring.

But I’m still doubtful that this is for the best, and still welcome being
mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on top
of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative roads,
which are thinner in the new regime, which will cause more derailments,
etc. Which screws with the benefit of beeminding. (And is reminiscent of
the dreaded-weigh-in loop that lead us to auto-widening roads in the first
place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more substantial
to say about weight goals later after reading more of the discussion, but I
wanted to reply now because I object pretty strongly to the idea of letting
you weigh in multiple times in one day and taking the lowest value. I had
no idea (and now wish I still didn’t know) it worked that way now. I always
weigh myself only once, first thing in the morning. Later in the day when
I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5% of my weight for me -
it would be more for a larger person. If I’m off my road when I first weigh
in in the morning, I don’t think there’s anything healthy I can do to get
back on later in the day, and I definitely don’t want to incentivize myself
to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight
loss. Since January I’ve had a my weight +/- 3 lbs road which seems to
accommodate my womanly fluctuations and also water retention due to
medicines that I take.

I also agree that taking the lowest weight of the day is beneficial because
the very first weight of the morning isn’t always the lowest. I’m not
really sure what the optimal solution is with regards to weight loss goals
for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh philip@hellyer.net wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my gym-going
emergency day coming, but I won’t know that it’s a weight beemergency until
I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused people
to dread standing on the scale. So they’d not weigh in daily, which meant
they’d lose sight of their progress (or regress), which meant they’d be
more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument that
I’ll still step through the colours on my way to derailing doesn’t apply to
me. (And wouldn’t occur if presumptive-pessimism is implemented.) I think
that the turquoise swath itself might be vestigial; it used to clearly show
the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will
need to become a rule like ‘if today I’m heavier than yesterday’ or ‘if
fewer than 2 safe days’ or something less obviously derived from the graph
and its colouring.

But I’m still doubtful that this is for the best, and still welcome being
mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on top
of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative roads,
which are thinner in the new regime, which will cause more derailments,
etc. Which screws with the benefit of beeminding. (And is reminiscent of
the dreaded-weigh-in loop that lead us to auto-widening roads in the first
place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/**glossary/#flatlinehttp://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Thanks Jill.

I should probably have re-iterated that I also think that a somewhat wider
road is the right answer, but with a width based on the variability of the
data, not on the road slope. Auto-widening definitely increased the games
I could play and still stay on the road, so I’m all for turning that off.

It’s a good point that each of us is figuring out how to employ beeminder’s
tools in a way that works for us as individuals. Simultaneously the
Beeminder team is trying to make the tools work for most people
out-of-the-box. The dynamic tension in that relationship is one of the
things that helps both sides keep improving.

Philip

On Friday, August 2, 2013 4:02:36 PM UTC+1, Jill Renaud wrote:

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight
loss. Since January I’ve had a my weight +/- 3 lbs road which seems to
accommodate my womanly fluctuations and also water retention due to
medicines that I take.

I also agree that taking the lowest weight of the day is beneficial
because the very first weight of the morning isn’t always the lowest. I’m
not really sure what the optimal solution is with regards to weight loss
goals for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh <phi...@hellyer.net <javascript:>>wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my gym-going
emergency day coming, but I won’t know that it’s a weight beemergency until
I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused people
to dread standing on the scale. So they’d not weigh in daily, which meant
they’d lose sight of their progress (or regress), which meant they’d be
more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument that
I’ll still step through the colours on my way to derailing doesn’t apply to
me. (And wouldn’t occur if presumptive-pessimism is implemented.) I think
that the turquoise swath itself might be vestigial; it used to clearly show
the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will
need to become a rule like ‘if today I’m heavier than yesterday’ or ‘if
fewer than 2 safe days’ or something less obviously derived from the graph
and its colouring.

But I’m still doubtful that this is for the best, and still welcome being
mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on top
of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative roads,
which are thinner in the new regime, which will cause more derailments,
etc. Which screws with the benefit of beeminding. (And is reminiscent of
the dreaded-weigh-in loop that lead us to auto-widening roads in the first
place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/**glossary/#flatlinehttp://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+u...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

TLDR: I’ve bolded the parts that I think matter; so that you don’t have to
read so much of my yammer to get to my points.

I’m a fan of the auto-widening road, but I think there might be another way
to accommodate my concerns about losing it. The problem is that it uses
custom goal settings, to which most users won’t have access. Can I
recommend that if you get rid of auto-widening roads, you make some of
the custom features available for the “lose/gain weight” type of goals
?
(Maybe just one of these could be allowed per user so it’s not a loophole
around not having access to custom goals if you’re not a subscriber.)

Today is an example case of what worries me. I’ve recently deleted my “max
safe days” on most of my goals (for unrelated reasons). Thank goodness,
because today when I weighed in, I weighed in at 2.2 lbs less than
yesterday
. When I got on the scale and saw that, I didn’t think “Hmm, I
wonder if I’m dehydrated” or “I wonder if this is closer to my real weight
(seeing as the most recent point were a little higher than the “group” of
nearby points) and I’m balancing out again”. Instead,* I thought “Shit!
That’s going to screw up my graph” This is not the right response to have
elicited by a low weigh-in.* (I wasn’t sure when exactly auto-widening
would die, or if it perhaps already has, so I wasn’t sure if I’d be
affected yet.)

See, if I had kept my max-safe-days (which I want to use to keep myself
pushing forward rather than coasting if/when I have particularly good
months) then I’d be in trouble because it would have tightened my road and
I would derail tomorrow or the next day when a balanced weigh-in occurs
again.
This is solved, probably, by custom road width and using “edgy”(“put the initial point on the road edge instead of centerline”) so that if
the max-safe-days kicks in the new point will (I assume) be at the bottom
of the road so that it won’t shove my road so far down that I’m screwed,
and it also won’t require me to have a road width 2x the necessary size.

This also would help what Jill has called “womanly fluctuations” as well as
medical situations. I have a thing that makes it such that some days I
weigh much more than my “weight”, as well, and so having the ability to set
my own road width could be really helpful in allowing me to have a road
that is as steep as I want and also as wide as I want. (Something I can do
with a custom goal, but new customers can’t.)

All of this still doesn’t really fix the one thing that I think is more
important, though. I think the focus on the “daily” part is kind of off. The
idea that if we’re over the line one day we can try to claw our ways back
down that day to correct it is kind of silly (and seems an unhealthy way to
look at weight loss).
[1] If I’ve reached the point that I’m over the
line, I may have already lost (unless it was a weird, random upward
fluctuation). *Today’s datapoint should matter less in weight loss goals
than the trend. Today should matter only in so far as it’s a piece of that
trend. I would greatly prefer having a road that is calibrated to derail
according to an average. I think that’s way more stable a measurement for
weight loss datapoints (because random fluctuations only effect it slightly
since the surrounding points matter too) and it is, I believe, a healthier
way to think of weight loss. *Today matters, because it could push my
average over, but it’s not the focus. Most Beeminded goals are behaviour
(read more of this, eat less of that, run more, lift more, smoke less,
whatever); weight loss is not a behaviour, it’s an outcome of behaviour.
Outcomes are less predictable and less in our direct control than
behaviour, and I think a system that measures an outcome too tightly is a
little off. Behaviour should be measured tightly; an outcome should be
measure more loosely, and thought of as being representative of behaviour
that needs to be altered.
("Hmm… my average is trending upwards; perhaps
12 twinkies a day is too many… I’m going to create an “eat fewer twinkies
goal”)

The solution I will use to adapt to this will be one that I think you
guys… will not love, to say the least, but this post is stupid-long
already, so I’ll save that for another time.

Ess

[1] Note that I’m not saying that you should get rid of multiple weigh-ins;
I think they’re useful for other reasons.

On Friday, August 2, 2013 11:02:36 AM UTC-4, Jill Renaud wrote:

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight
loss. Since January I’ve had a my weight +/- 3 lbs road which seems to
accommodate my womanly fluctuations and also water retention due to
medicines that I take.

I also agree that taking the lowest weight of the day is beneficial
because the very first weight of the morning isn’t always the lowest. I’m
not really sure what the optimal solution is with regards to weight loss
goals for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh <phi...@hellyer.net <javascript:>>wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my gym-going
emergency day coming, but I won’t know that it’s a weight beemergency until
I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused people
to dread standing on the scale. So they’d not weigh in daily, which meant
they’d lose sight of their progress (or regress), which meant they’d be
more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument that
I’ll still step through the colours on my way to derailing doesn’t apply to
me. (And wouldn’t occur if presumptive-pessimism is implemented.) I think
that the turquoise swath itself might be vestigial; it used to clearly show
the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will
need to become a rule like ‘if today I’m heavier than yesterday’ or ‘if
fewer than 2 safe days’ or something less obviously derived from the graph
and its colouring.

But I’m still doubtful that this is for the best, and still welcome being
mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on top
of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative roads,
which are thinner in the new regime, which will cause more derailments,
etc. Which screws with the benefit of beeminding. (And is reminiscent of
the dreaded-weigh-in loop that lead us to auto-widening roads in the first
place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/**glossary/#flatlinehttp://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+u...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

Wow, yes, this. If weight roads were based on the moving average
instead of the individual data points, it would resolve my concerns
and make me a very happy bee.

I love that Beeminder shows a moving average a la Hacker Diet, but I
think it misses the point a little bit. Hacker Diet is all about
focusing on the trend and pretty much ignoring the individual data
points. http://blog.beeminder.com/weighly/ seems to me like it’s
saying “Weigh every day, focus on the trend, and ignore the individual
data points. Oh, but your goals are centered entirely around the
individual data points, so actually just focus on those.”

Maybe I should track data points and calculate a moving average in a
separate system, and then just beemind the calculated average.

On Tue, Aug 6, 2013 at 4:22 AM, Essentiae essentiae@gmail.com wrote:

TLDR: I’ve bolded the parts that I think matter; so that you don’t have to
read so much of my yammer to get to my points.

I’m a fan of the auto-widening road, but I think there might be another way
to accommodate my concerns about losing it. The problem is that it uses
custom goal settings, to which most users won’t have access. Can I recommend
that if you get rid of auto-widening roads, you make some of the custom
features available for the “lose/gain weight” type of goals? (Maybe just one
of these could be allowed per user so it’s not a loophole around not having
access to custom goals if you’re not a subscriber.)

Today is an example case of what worries me. I’ve recently deleted my “max
safe days” on most of my goals (for unrelated reasons). Thank goodness,
because today when I weighed in, I weighed in at 2.2 lbs less than
yesterday. When I got on the scale and saw that, I didn’t think “Hmm, I
wonder if I’m dehydrated” or “I wonder if this is closer to my real weight
(seeing as the most recent point were a little higher than the “group” of
nearby points) and I’m balancing out again”. Instead, I thought “Shit!
That’s going to screw up my graph” This is not the right response to have
elicited by a low weigh-in. (I wasn’t sure when exactly auto-widening would
die, or if it perhaps already has, so I wasn’t sure if I’d be affected yet.)

See, if I had kept my max-safe-days (which I want to use to keep myself
pushing forward rather than coasting if/when I have particularly good
months) then I’d be in trouble because it would have tightened my road and I
would derail tomorrow or the next day when a balanced weigh-in occurs again.
This is solved, probably, by custom road width and using “edgy” (“put the
initial point on the road edge instead of centerline”) so that if the
max-safe-days kicks in the new point will (I assume) be at the bottom of the
road so that it won’t shove my road so far down that I’m screwed, and it
also won’t require me to have a road width 2x the necessary size.

This also would help what Jill has called “womanly fluctuations” as well as
medical situations. I have a thing that makes it such that some days I weigh
much more than my “weight”, as well, and so having the ability to set my own
road width could be really helpful in allowing me to have a road that is as
steep as I want and also as wide as I want. (Something I can do with a
custom goal, but new customers can’t.)

All of this still doesn’t really fix the one thing that I think is more
important, though. I think the focus on the “daily” part is kind of off. The
idea that if we’re over the line one day we can try to claw our ways back
down that day to correct it is kind of silly (and seems an unhealthy way to
look at weight loss). [1] If I’ve reached the point that I’m over the line,
I may have already lost (unless it was a weird, random upward fluctuation).
Today’s datapoint should matter less in weight loss goals than the trend.
Today should matter only in so far as it’s a piece of that trend. I would
greatly prefer having a road that is calibrated to derail according to an
average. I think that’s way more stable a measurement for weight loss
datapoints (because random fluctuations only effect it slightly since the
surrounding points matter too) and it is, I believe, a healthier way to
think of weight loss. Today matters, because it could push my average over,
but it’s not the focus. Most Beeminded goals are behaviour (read more of
this, eat less of that, run more, lift more, smoke less, whatever); weight
loss is not a behaviour, it’s an outcome of behaviour. Outcomes are less
predictable and less in our direct control than behaviour, and I think a
system that measures an outcome too tightly is a little off. Behaviour
should be measured tightly; an outcome should be measure more loosely, and
thought of as being representative of behaviour that needs to be altered.
("Hmm… my average is trending upwards; perhaps 12 twinkies a day is too
many… I’m going to create an “eat fewer twinkies goal”)

The solution I will use to adapt to this will be one that I think you
guys… will not love, to say the least, but this post is stupid-long
already, so I’ll save that for another time.

Ess

[1] Note that I’m not saying that you should get rid of multiple weigh-ins;
I think they’re useful for other reasons.

On Friday, August 2, 2013 11:02:36 AM UTC-4, Jill Renaud wrote:

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight loss.
Since January I’ve had a my weight +/- 3 lbs road which seems to accommodate
my womanly fluctuations and also water retention due to medicines that I
take.

I also agree that taking the lowest weight of the day is beneficial
because the very first weight of the morning isn’t always the lowest. I’m
not really sure what the optimal solution is with regards to weight loss
goals for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh phi...@hellyer.net wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the red
within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my gym-going
emergency day coming, but I won’t know that it’s a weight beemergency until
I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and ran
back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused people
to dread standing on the scale. So they’d not weigh in daily, which meant
they’d lose sight of their progress (or regress), which meant they’d be more
likely to derail, etc. That was ‘fixed’ by the road-widening guarantee. So
Danny’s absolutely right that we’ve piled one hack on top of another, and
that it’s high time we identify & resolve the underlying problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument that
I’ll still step through the colours on my way to derailing doesn’t apply to
me. (And wouldn’t occur if presumptive-pessimism is implemented.) I think
that the turquoise swath itself might be vestigial; it used to clearly show
the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without Beeminder.
Seemingly spurious beyond-my-control derailments don’t fall in that
category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will need
to become a rule like ‘if today I’m heavier than yesterday’ or ‘if fewer
than 2 safe days’ or something less obviously derived from the graph and its
colouring.

But I’m still doubtful that this is for the best, and still welcome being
mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on top
of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email, something
else suffers. And so on. One of the roles that Beeminder plays for me is
to ensure that I keep all the important plates spinning a little, regardless
of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative roads,
which are thinner in the new regime, which will cause more derailments, etc.
Which screws with the benefit of beeminding. (And is reminiscent of the
dreaded-weigh-in loop that lead us to auto-widening roads in the first
place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+u...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

I think the key point here is this:
Behaviour should be measured tightly; an outcome should be measure more
loosely, and thought of as being representative of behaviour that needs to
be altered.

*
*
I’ve stopped measuring long term outcomes, and only measure things I can
automatically measure and realistically fix. Ideally, those are very
granular and loophole free.

R

On Tue, Aug 6, 2013 at 4:22 AM, Essentiae essentiae@gmail.com wrote:

TLDR: I’ve bolded the parts that I think matter; so that you don’t have to
read so much of my yammer to get to my points.

I’m a fan of the auto-widening road, but I think there might be another
way to accommodate my concerns about losing it. The problem is that it uses
custom goal settings, to which most users won’t have access. Can I
recommend that if you get rid of auto-widening roads, you make some of
the custom features available for the “lose/gain weight” type of goals
?
(Maybe just one of these could be allowed per user so it’s not a loophole
around not having access to custom goals if you’re not a subscriber.)

Today is an example case of what worries me. I’ve recently deleted my
“max safe days” on most of my goals (for unrelated reasons). Thank
goodness, because today when I weighed in, I weighed in at 2.2 lbs less
than yesterday
. When I got on the scale and saw that, I didn’t think
“Hmm, I wonder if I’m dehydrated” or “I wonder if this is closer to my real
weight (seeing as the most recent point were a little higher than the
“group” of nearby points) and I’m balancing out again”. Instead,* I
thought “Shit! That’s going to screw up my graph” This is not the right
response to have elicited by a low weigh-in.* (I wasn’t sure when exactly
auto-widening would die, or if it perhaps already has, so I wasn’t sure if
I’d be affected yet.)

See, if I had kept my max-safe-days (which I want to use to keep myself
pushing forward rather than coasting if/when I have particularly good
months) then I’d be in trouble because it would have tightened my road
and I would derail tomorrow or the next day when a balanced weigh-in occurs
again.
This is solved, probably, by custom road width and using “edgy”(“put the initial point on the road edge instead of centerline”) so that if
the max-safe-days kicks in the new point will (I assume) be at the bottom
of the road so that it won’t shove my road so far down that I’m screwed,
and it also won’t require me to have a road width 2x the necessary size.

This also would help what Jill has called “womanly fluctuations” as well
as medical situations. I have a thing that makes it such that some days I
weigh much more than my “weight”, as well, and so having the ability to set
my own road width could be really helpful in allowing me to have a road
that is as steep as I want and also as wide as I want. (Something I can
do with a custom goal, but new customers can’t.)

All of this still doesn’t really fix the one thing that I think is more
important, though. I think the focus on the “daily” part is kind of off. The
idea that if we’re over the line one day we can try to claw our ways back
down that day to correct it is kind of silly (and seems an unhealthy way to
look at weight loss).
[1] If I’ve reached the point that I’m over the
line, I may have already lost (unless it was a weird, random upward
fluctuation). *Today’s datapoint should matter less in weight loss goals
than the trend. Today should matter only in so far as it’s a piece of that
trend. I would greatly prefer having a road that is calibrated to derail
according to an average. I think that’s way more stable a measurement for
weight loss datapoints (because random fluctuations only effect it slightly
since the surrounding points matter too) and it is, I believe, a healthier
way to think of weight loss. *Today matters, because it could push my
average over, but it’s not the focus. Most Beeminded goals are behaviour
(read more of this, eat less of that, run more, lift more, smoke less,
whatever); weight loss is not a behaviour, it’s an outcome of behaviour.
Outcomes are less predictable and less in our direct control than
behaviour, and I think a system that measures an outcome too tightly is a
little off. Behaviour should be measured tightly; an outcome should be
measure more loosely, and thought of as being representative of behaviour
that needs to be altered.
("Hmm… my average is trending upwards;
perhaps 12 twinkies a day is too many… I’m going to create an “eat fewer
twinkies goal”)

The solution I will use to adapt to this will be one that I think you
guys… will not love, to say the least, but this post is stupid-long
already, so I’ll save that for another time.

Ess

[1] Note that I’m not saying that you should get rid of multiple
weigh-ins; I think they’re useful for other reasons.

On Friday, August 2, 2013 11:02:36 AM UTC-4, Jill Renaud wrote:

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight
loss. Since January I’ve had a my weight +/- 3 lbs road which seems to
accommodate my womanly fluctuations and also water retention due to
medicines that I take.

I also agree that taking the lowest weight of the day is beneficial
because the very first weight of the morning isn’t always the lowest. I’m
not really sure what the optimal solution is with regards to weight loss
goals for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh phi...@hellyer.net wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my
gym-going emergency day coming, but I won’t know that it’s a weight
beemergency until I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused
people to dread standing on the scale. So they’d not weigh in daily, which
meant they’d lose sight of their progress (or regress), which meant they’d
be more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument
that I’ll still step through the colours on my way to derailing doesn’t
apply to me. (And wouldn’t occur if presumptive-pessimism is implemented.)
I think that the turquoise swath itself might be vestigial; it used to
clearly show the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will
need to become a rule like ‘if today I’m heavier than yesterday’ or ‘if
fewer than 2 safe days’ or something less obviously derived from the graph
and its colouring.

But I’m still doubtful that this is for the best, and still welcome
being mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on
top of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative
roads, which are thinner in the new regime, which will cause more
derailments, etc. Which screws with the benefit of beeminding. (And is
reminiscent of the dreaded-weigh-in loop that lead us to auto-widening
roads in the first place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/glossary/#flatlinehttp://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google
Groups “Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send
an email to akratics+u…@**googlegroups.com.

For more options, visit https://groups.google.com/**groups/opt_outhttps://groups.google.com/groups/opt_out
.


You received this message because you are subscribed to the Google Groups
“Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Bingo! Thanks Ess. I think you’ve hit on what I was trying to express about
there being an essential difference between measuring weight and counting
do-more tasks. For me, the key quote is this one: “weight loss is not a
behaviour, it’s an outcome of behaviour”

Behaviours (and not-doings) are catered for by do-more and set-a-limit
goals. Fat-loser and weight-gainer goals are outcomes, which we once
described on the site as having values that were expected to vary but to
have a particular trend.

If you’re keen for Beeminder to use the trend rather than the individual
datapoints, you could upvote this UserVoice suggestion:

Looking back over my weight-goal graph, the skinny purple line is above the
(current-width) yellow-brick-road in 2 places, and is a good reflection of
where I ought to have derailed.

Speaking of road-width, the skinny purple line moves slowly enough that the
width of the road is somewhat less important. I suspect that it should
still be tied to the variability of the data, but the implications of that
haven’t settled in my head yet.

I’m also uncertain about what this will do for beemergency days—possibly no
difference, just teeny-tiny effects on the purple line. But as Essentiae
said, by the time you’re over the road on a weight goal, you may already
have lost.

Philip

On Tuesday, 6 August 2013 12:22:16 UTC+1, Essentiae wrote:

TLDR: I’ve bolded the parts that I think matter; so that you don’t have to
read so much of my yammer to get to my points.

I’m a fan of the auto-widening road, but I think there might be another
way to accommodate my concerns about losing it. The problem is that it uses
custom goal settings, to which most users won’t have access. Can I
recommend that if you get rid of auto-widening roads, you make some of
the custom features available for the “lose/gain weight” type of goals
?
(Maybe just one of these could be allowed per user so it’s not a loophole
around not having access to custom goals if you’re not a subscriber.)

Today is an example case of what worries me. I’ve recently deleted my
“max safe days” on most of my goals (for unrelated reasons). Thank
goodness, because today when I weighed in, I weighed in at 2.2 lbs less
than yesterday
. When I got on the scale and saw that, I didn’t think
“Hmm, I wonder if I’m dehydrated” or “I wonder if this is closer to my real
weight (seeing as the most recent point were a little higher than the
“group” of nearby points) and I’m balancing out again”. Instead,* I
thought “Shit! That’s going to screw up my graph” This is not the right
response to have elicited by a low weigh-in.* (I wasn’t sure when exactly
auto-widening would die, or if it perhaps already has, so I wasn’t sure if
I’d be affected yet.)

See, if I had kept my max-safe-days (which I want to use to keep myself
pushing forward rather than coasting if/when I have particularly good
months) then I’d be in trouble because it would have tightened my road
and I would derail tomorrow or the next day when a balanced weigh-in occurs
again.
This is solved, probably, by custom road width and using “edgy”(“put the initial point on the road edge instead of centerline”) so that if
the max-safe-days kicks in the new point will (I assume) be at the bottom
of the road so that it won’t shove my road so far down that I’m screwed,
and it also won’t require me to have a road width 2x the necessary size.

This also would help what Jill has called “womanly fluctuations” as well
as medical situations. I have a thing that makes it such that some days I
weigh much more than my “weight”, as well, and so having the ability to set
my own road width could be really helpful in allowing me to have a road
that is as steep as I want and also as wide as I want. (Something I can
do with a custom goal, but new customers can’t.)

All of this still doesn’t really fix the one thing that I think is more
important, though. I think the focus on the “daily” part is kind of off. The
idea that if we’re over the line one day we can try to claw our ways back
down that day to correct it is kind of silly (and seems an unhealthy way to
look at weight loss).
[1] If I’ve reached the point that I’m over the
line, I may have already lost (unless it was a weird, random upward
fluctuation). *Today’s datapoint should matter less in weight loss goals
than the trend. Today should matter only in so far as it’s a piece of that
trend. I would greatly prefer having a road that is calibrated to derail
according to an average. I think that’s way more stable a measurement for
weight loss datapoints (because random fluctuations only effect it slightly
since the surrounding points matter too) and it is, I believe, a healthier
way to think of weight loss. *Today matters, because it could push my
average over, but it’s not the focus. Most Beeminded goals are behaviour
(read more of this, eat less of that, run more, lift more, smoke less,
whatever); weight loss is not a behaviour, it’s an outcome of behaviour.
Outcomes are less predictable and less in our direct control than
behaviour, and I think a system that measures an outcome too tightly is a
little off. Behaviour should be measured tightly; an outcome should be
measure more loosely, and thought of as being representative of behaviour
that needs to be altered.
("Hmm… my average is trending upwards;
perhaps 12 twinkies a day is too many… I’m going to create an “eat fewer
twinkies goal”)

The solution I will use to adapt to this will be one that I think you
guys… will not love, to say the least, but this post is stupid-long
already, so I’ll save that for another time.

Ess

[1] Note that I’m not saying that you should get rid of multiple
weigh-ins; I think they’re useful for other reasons.

On Friday, August 2, 2013 11:02:36 AM UTC-4, Jill Renaud wrote:

Philip,

I really like your response here. I also agree with your opinion that a
razor thin road will lead to more derailments with regards to weight loss
roads. That said, I think the auto-widening feature can be detrimental to
weight loss. I’ve found that having a fixed width (relatively wide) road
has been working wonders for me. When the road was variable width I’d play
games that were ultimately bad for my progress. If I was on the right side
of the road one day, I’d go hog wild and pig out to keep my road wide.
Obviously going hog wild and pigging out is not beneficial for weight
loss. Since January I’ve had a my weight +/- 3 lbs road which seems to
accommodate my womanly fluctuations and also water retention due to
medicines that I take.

I also agree that taking the lowest weight of the day is beneficial
because the very first weight of the morning isn’t always the lowest. I’m
not really sure what the optimal solution is with regards to weight loss
goals for everyone, but I think after 5 years of beeminding/kibotzing I’ve
figured out what works for me.

-Jill

On Fri, Aug 2, 2013 at 10:43 AM, pjh phi...@hellyer.net wrote:

One of the benefits of Beeminder graphs is the predictability that they
bring to my life. At the extreme, I react to beemergency days and JFDI the
task. At its best, I will notice that my gym-going graph will be in the
red within 3 days, so I plan a gym morning into my schedule. Being able to
predict when the emergency days are going to hit, and take steps to make my
life less reactive, is one of the intangible benefits of Beeminder.[1]

Weight graphs are not predictable in the same way. I can see my
gym-going emergency day coming, but I won’t know that it’s a weight
beemergency until I stand on the scale in the morning. [2]

This morning was such a day, and happily I had already intended to go to
the gym. (Turns out that it’s possible to get back on the road without
resorting to Danny’s dozen weigh-ins. :slight_smile: My first weigh-in was well into
the red. I had a bit of breakfast, ran to the gym, swam, stretched, and
ran back, adequately hydrated throughout. And weighed-in at 0.3kg lighter.
Sufficient to get me back on the road & safely achieved.)

We used to only allow one weigh-in each day, and I think it caused
people to dread standing on the scale. So they’d not weigh in daily, which
meant they’d lose sight of their progress (or regress), which meant they’d
be more likely to derail, etc. That was ‘fixed’ by the road-widening
guarantee. So Danny’s absolutely right that we’ve piled one hack on top of
another, and that it’s high time we identify & resolve the underlying
problem.

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

I don’t have a problem with doing my daily weigh-in, so the argument
that I’ll still step through the colours on my way to derailing doesn’t
apply to me. (And wouldn’t occur if presumptive-pessimism is implemented.)
I think that the turquoise swath itself might be vestigial; it used to
clearly show the difference between actual data slope and road slope. With
auto-re-railment, most of my data points follow the road, and the turquoise
swath does the same. (I’ve turned the swath back on for a few goals to see
whether it’s more interesting than I remember.)

I’ve always liked that we treat different types of graph differently.
Weight loss graphs are for absolute measures that vary, but where the goal
is a downward trend. This is substantively different from counting units
of progress that increase monotonically.

It seems that a razor-thin road will force me into exactly the situation
that Danny wants to avoid. My reality will be governed by an indiscernible
and invisible line somewhere below the road. I predict an increase in
derailments.[3]

The occasional derailment on the path to a goal is fine, provided that I
think I’ve made good progress that wouldn’t have occurred without
Beeminder. Seemingly spurious beyond-my-control derailments don’t fall in
that category. Double plus bad, on so many levels.

Having said all of that:

It’s possible that a presumptively-pessimistic view of safe-days will be
enough to let me see where I really am with respect to the road edge.

It certainly would change the games that I can play with myself; instead
of paying slightly more attention to my diet when I’m orange, that will
need to become a rule like ‘if today I’m heavier than yesterday’ or ‘if
fewer than 2 safe days’ or something less obviously derived from the graph
and its colouring.

But I’m still doubtful that this is for the best, and still welcome
being mistaken.

Philip

[1] My life seems to be in a continual state of imbalance. If I’m on
top of my physical health, my work suffers. If I’m on top of conference
speaking, my other commitments suffer. If I’m caught up on email,
something else suffers. And so on. One of the roles that Beeminder plays
for me is to ensure that I keep all the important plates spinning a little,
regardless of where my impetuous focus currently lies.

[2] For this same reason, auto-data goals need to do their first fetch
before I check my graphs in the morning. Can’t be finding out part-way
through my day that something is going to derail at midnight. (I suppose
that I could force this behaviour by setting a stupidly early morning
panic-email time, but it should be standard.)

[3] Worse, unexpected derailments will encourage more conservative
roads, which are thinner in the new regime, which will cause more
derailments, etc. Which screws with the benefit of beeminding. (And is
reminiscent of the dreaded-weigh-in loop that lead us to auto-widening
roads in the first place…)

On Friday, August 2, 2013 2:47:40 PM UTC+1, Katherine Baxter wrote:

I just joined this list, and I’ll probably have something more
substantial to say about weight goals later after reading more of the
discussion, but I wanted to reply now because I object pretty strongly to
the idea of letting you weigh in multiple times in one day and taking the
lowest value. I had no idea (and now wish I still didn’t know) it worked
that way now. I always weigh myself only once, first thing in the morning.
Later in the day when I’m hydrated and fed, I am ~5lbs heavier. That’s 3.5%
of my weight for me - it would be more for a larger person. If I’m off my
road when I first weigh in in the morning, I don’t think there’s anything
healthy I can do to get back on later in the day, and I definitely don’t
want to incentivize myself to do unhealthy things just to stay on my road.

I’m adding fine print to my weight goal that says I am only allowed to
record the first weight I measure on a particular day.

Why on earth does it work this way?

On Friday, July 26, 2013 9:50:52 PM UTC-4, Daniel Reeves wrote:

This is madness: blog.beeminder.com/roadwidth

At the time we kind of painted ourselves into that corner and didn’t
see a way out. We thought we needed auto-widening to solve a
psychological problem with weight loss (or more generally, akrasia
with respect to a noisy metric) where you try to skate too close to
the edge and sooner or later get burned by a random fluctuation.

Auto-widening was a (too) clever solution given the constraints we
started with: your first weigh-in in the morning is your official
weight, and you don’t get to peek at what it’s going to be before
reporting it. In that universe, you need guidance from Beeminder on
when to panic. The answer was, if you’re in the wrong lane today then
you have to panic because you might be doomed when you step on the
scale in the morning. Auto-widening meant it was time to panic if and
only if you found yourself in the wrong lane, which is the important
bit of psychology. There has to be a way to save up the panic and hit
you with it all at once so that you actually do panic. (No
wolf-crying.) Auto-widening does that. If you’re in the right lane
then you don’t have to panic yet. You can’t lose tomorrow. If you’re
in the wrong lane, you can lose tomorrow.

(Also it sounds really good – “Beeminder uses fancy maths to
automatically adapt to random fluctuation in your weight” – which
will make this hard to throw away!)

But all the strategery is different in the current universe. Beeminder
takes the min weight of the day, and you can keep weighing in all day
long, till midnight. So now there’s no true cause for panic unless
you’re actually in the red, in which case you have to claw your way
back on the road by the end of the day. For a true akratic doing the
absolute bare minimum to cling to the edge of their road,
auto-widening is now meaningless. If I’m not in the red I just don’t
give a shit.

A huge amount of cruft and confusion and corner cases (not to mention
at least one long-standing bug) in Beeminder will go away when we drop
auto-widening.

Yay for (eventual) progress! And thanks so much to everyone on this
list who helped us hash this out! (And the hashing can continue, of
course.)

Technical addendum:

  1. We’ll keep the first-week leniency: Your weight loss road starts at
    the max weigh-in from your first week of data.

  2. The can’t-lose-tomorrow guarantee will still obtain, in a sense,
    but only applies for flatlining [1]. The width of each lane of the
    road is always equal to the current daily rate of the road [2]. So
    being just into the green today and reporting nothing means blue
    tomorrow, orange the next day, and red the day after that.

[1] http://blog.beeminder.com/**glossary/#flatlinehttp://blog.beeminder.com/glossary/#flatline

[2] Confusing exception: flat spots have road width equal to max of
the width of the preceding and subsequent segments.


http://dreev.es – search://“Daniel Reeves”
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google
Groups “Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send
an email to akratics+u...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Thanks so much for the input on this, everyone. (And keep it coming!)
I’m thinking hard about it.

First, I think I was confusingly conflating 2 things before:

  1. the “can’t lose tomorrow” guarantee with immediate auto-widening
    (road accommodates any jump today as long as you were in the blue
    yesterday)
  2. determining road width based on variance of the data (road width is
    such that if your true weight followed the centerline then random
    fluctuations would put you above the road 5% of the time and below the
    road another 5% of the time)

Those are orthogonal and I think it may be uncontroversial to drop #1.
Jill gives a great example of why #1 is actually bad (purposefully
pigging out to widen your road!).

But I’m also not sold on the value of #2. I think we have a lot of red
herrings in this debate. Remember that if you’re truly akratic then
nothing really matters but the top edge of your yellow brick road. (If
you’re not truly akratic then just stay comfortably below the road and
ignore all this!) So I’m not buying the concern about a razor-thin
road. Right now you have a huge yellow area below you but it’s that
razor-thin top edge that’s all that matters. And remember that the
turquoise swath, as of a few days ago, has width determined by the
algorithm in #2 above. So you’ve got that guidance. Make the top edge
of your aura touch the top edge of your road and you have about a 5%
chance of a fluctuation derailing you (assuming no emergency-day
heroics).

This is probably getting too theoretical and it’s time to just try things!

Specific replies follow:

(@Katherine: You can, by emailing support, have us change how multiple weigh-ins are counted or aggregated. I think “first of the day” is still an option.)

Yup, and it’s available to users in advanced settings for custom
goals, though that’s a premium feature.
I think it’s too late for us to backpedal on the “min weight of the
day” rule as the default but I certainly see where you’re coming from
on that, and aggday=first will make Beeminder do what you want.
Another reason we bowed to pressure originally on the min-weight rule
(which I also originally objected to) was that sometimes your scale
can give an anomalous reading, or maybe you forget to go to the
bathroom first. But then I came to actually like the ridiculous gaming
of running around the block until you sweat your way back onto the
road. I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss as well as the short term goal of clawing your way back onto the
yellow brick road.

To Essy’s point that you should beemind the moving average, not the
individual weight:

The moving average is the right thing to look at to see the trend but
I don’t think it will help to beemind it. Here’s my concen, something
Philip hinted at:
Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road. It’s an emergency day and you
need to lose .5kg to get back on. Except you’re beeminding the moving
average so now you need to lose 10 times that (or whatever) to
actually bring the average down enough. You’re actually requiring a
bunch more foresight to keep from derailing, which I think is a
nonstarter. In general, being akratic and beeminding the moving
average still means that you’re skating the edge and every day there’s
some magic weight that you must weigh in at to keep from derailing.
That’s fundamental to beeminding weight. No matter how you try to
smooth the data or widen the road, if you eked by yesterday then there
will necessarily be a magic weight that you absolutely must hit today
to stay on track. It’s like a no-free-lunch theorem. Whatever we do
ends up equivalent to having a single bright line that your weight is
not allowed to cross every day.

Essy’s point about beeminding inputs vs outputs:

This is astute and is a good general principle, but I think weight is
an exception. For one, it’s so easy to measure. And as long as you
have a way to account for the randomness (which hopefully Beeminder
will continue to do) then you do have sufficient control over it. I’d
actually advocate going for the best of both worlds: beemind your
weight but keep it conservative and the stakes low – just make sure
it’s going in the right direction or staying in a reasonable range –
and also beemind all the inputs (sugar, exercise, vegetables,
anything you’re akratic about) and beemind those aggressively and with
high stakes. Beemind the inputs so that weight becomes a no-brainer.
But still beemind it, since it’s all too easy to delude yourself.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

Thin vs. wide roads
"Remember that if you’re truly akratic then nothing really matters but the
top edge of your yellow brick road. (If you’re not truly akratic then just
stay comfortably below the road and ignore all this!) So I’m not buying the
concern about a razor-thin road."

I think it’s right that it’s the top of the road that matters. That said,
the yellow brick road provides valuable visual feedback on how you’re
doing: when you’re in risky territory (orange), when you’re sailing
smoothly (blue), and when you’re rockin’ it (green). Having only green and
red points loses that valuable visual data, as well as the safety one has
earned when getting onto a green point. Beeminder doesn’t just help
akratics stop doing dumb $#!~ today, it also sort of teaches akratics the
value of thinking and working ahead. I’ve certainly noticed a tendency to
start thinking things like “I should get X goal above my road by a few days
cause event Y is coming up and I don’t want to have to think about it
during that time.” I’ve always ~thought~ that… but now I’m doing it with
some goals cause I have a visual representation of my effort. That’s a
fundamental shift. I think people who are doing better than their roads
have “earned”, in a way, the feeling of safety on the green/blue side. So
road width matters, IMO. I think custom road width for weight, if it’s not
going to have auto-widening, allows that. (BUT fix it so we can’t change
the road width on the fly; there should be a 7-day countdown for road width
changes. Otherwise: “Argh, I’m over by .2lbs today… guess I’ll change my
road width…”)

Saving yourself on a given day
"Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision course
with the road. After 9 days of 100kg each day, you find yourself on the
wrong side of the road. It’s an emergency day and you need to lose .5kg to
get back on."

Then you’ve already lost.

If you have to try to lose weight in a day (when you’re going to be (and
OUGHT to be) putting food and water into your body from the moment you get
up), I think you’re in a bad place. I’m really leery of the idea of trying
to recover on any single given day. But that may be because of how I use my
weight goal. I use it to force me to tighten the behavour goals that result
in weight loss/maintenance (Daniel mentioned a few ideas later on in his
post) rather than as a “today” thing. If my weight’s not coming down, I
need to tweak some of my behavour goals, on penalty of derailing later.
(For this reason, I would love my weight goal to be set up with a 30-day
akrasia horizon, instead of a 7-day akrasia horizon.)

I think having a weight loss goal without having behavioural goals is
symptomatic of the akratic problem when connected to weight loss. Weight
goals are fundamentally different from goals like getting more work done or
limiting TV hours, and the akratic problem for that type of goal is
different from the akratic problem for the “omgz just do a little of this
today you procrastinator” type goals.

Real-time vs average
So… sometimes I talk of the world of the “ideal” about things that don’t
actually matter ~a ton~ to me. This is one of those cases. I really don’t
care that much whether the road is measuring individual datapoints or the
average, and I think the fact that it would mean making what might be some
pretty significant changes to Beeminder behind-the-scenes probably makes it
not worth it. The reason I would support it (in lieu of the auto-widening
roads) is for the unreasonable derailment cases (though, auto-widening
roads still didn’t fix many of the long-term, regular, legit derail cases).

I believe very strongly that genuine “am not losing weight so am over
today” datapoints = you’ve already lost… pay the hell up. Weight goals are
think-ahead goals and we akratics need to learn that. Losing water by
running around or veggy fasting for a day is not a healthy solution, IMO.
You’ve already lost… pay up, go have a glass of water, a normal breakfast,
and make healthier choices this week. So, whether it’s impossible to move
the moving average down after that or not means nothing to me. What worries
me, are those days that are not representative of the true trend. If you
are genuinely on the right track, you shouldn’t be penalized for something
like water retention, pms retention, or other things best left to your
imagination. Auto-widening roads or using an average doesn’t ~really~ help
that, cause in the water-retention or pms cases, we’re probably looking at
several “above goal” days even if you’re actually doing well and even if
you’re losing fat during that time.

I don’t really care how the goal is set up as long as it a) forces me to
plan my behaviour ~enough~ to meet my targets, b) doesn’t leave me thinking
about hitting the sauna to weigh in lower, c) doesn’t punish me for
non-representative changes.

For me, a longer akrasia horizon for weight goals + some different way of
counting when a derail happens (whatever it is) that forgives spikes that
aren’t representative is enough to do that… whatever those details are.

This is what I’ve done from time to time (and might do again… I’m sorry
Daniel, you’re going to hate this). I would enter my datapoints, but fail
to report my above the road points. If they were genuinely above the road
b/c I was actually failing, I would derail by not entering for several days
and running out of time. If they weren’t, I would pop back down before that
happened and the trend would continue just fine. This was fair, to me, b/c
it meant I wouldn’t be unfairly penalized… as long as the road was steep
enough to matter or the derail point was ~before~ the akrasia horizon.
Otherwise it’s too easy to change the road rate and never derail…

The problem with that is that I lose the QM value. BUT, what if weight loss
goals were set with different derail conditions. This is off the top of my
head, so there might be stupid things I’m not thinking of hidden here:

  1. Custom road width. (I want to earn the blue points, dammit! hehe. But
    more importantly, I also think visual representation if someone is losing
    ~too much~ weight is important. losing over 2-3lbs/week has gallbladder
    implications, so someone who sets their road to 2lbs with 1 safe day and
    who has a trend that is clearly steeper has a visual cue that they should
    eat a little more to stay on the healthy side of their efforts. I realize
    these are the minority cases, and not my own case, but it is one of the
    ways Beeminder can help people who are a little overzealous about it to
    keep from hurting themselves)

  2. Users (more typical than those mentioned in 1) setting longer safe day
    buffers (I like 14 because some of the legit spikes can be 10-days long but
    can catch right back up afterwards; men might prefer something like 5-7
    days, all depending on personal patterns/medical considerations). See
    attached as an example of what was a legit spike.

  3. A farther akrasia horizon for this goal (I like 30 days, but it could be
    a customizable value, maybe?… with changing that value subject to the
    ~currently~ set akrasia horizon value so that I can’t just weasel),

  4. (sorry) changing it so that it allows me to enter my above the line
    points for QM purposes, but derails as though I hadn’t. Make it so that
    some recent point matters, but not necessarily the most recent point. Maybe
    the lowest day in the “max safe days” range. I.e., if I have a 10-day max
    safe day range, then the lowest day in the last 10 days is what counts for
    derailing. BUT, it doesn’t count as though it’s today’s point, it counts as
    though I haven’t entered a point since. That way, if there fails to be
    positive change in [max safe day range] number of days, I lose. (But count
    the Hard caps by using today’s point and it’s distance from the point I
    need to be back to when I need to be back on the road.)

  5. Allow only 1 weight goal per user so that people don’t start using
    weight goals in order to create “debts” for other kinds of goals. (“I’ll
    just create a weight-type goal for my reading because then I can choose to
    catch up if I don’t keep on track” <-- FAIL!)

This could allow users who know their legit fluctuation details to plan for
them with safety, but not to use their safety net too loosely. And,
although I get your point about akratics and the upper line or the road,
weight goals are just different. They can’t be treated like other goals b/c
they’re just not like other goals. Part of heading in the right direction
is realizing that and adapting to it, I think. AND if you want to treat it
like other goals then create it like another goal: custom goal for the no
eating + sauna win.

Sorry for yet another way-too-long post!

You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road.

If I’ve been on a collision course for 9 days and haven’t done
anything about it, I’m okay with derailing. I am okay with needing to
think ahead for a weight goal.

I’d either like to beemind the purple line itself, or have a guarantee
that I won’t derail if the purple line is in the right lane.

I derailed a weight goal once due to water retention happening 2 days
in a row, even though my purple line was solidly in the right lane. I
found it so frustrating that I just stopped beeminding my weight.

And a side note:

I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss

I disagree! I don’t think running around the block when you haven’t
had anything to eat or drink that day, or eating only vegetables for a
day are healthy.

On Wed, Aug 7, 2013 at 11:25 PM, Daniel Reeves dreeves@beeminder.com wrote:

Thanks so much for the input on this, everyone. (And keep it coming!)
I’m thinking hard about it.

First, I think I was confusingly conflating 2 things before:

  1. the “can’t lose tomorrow” guarantee with immediate auto-widening
    (road accommodates any jump today as long as you were in the blue
    yesterday)
  2. determining road width based on variance of the data (road width is
    such that if your true weight followed the centerline then random
    fluctuations would put you above the road 5% of the time and below the
    road another 5% of the time)

Those are orthogonal and I think it may be uncontroversial to drop #1.
Jill gives a great example of why #1 is actually bad (purposefully
pigging out to widen your road!).

But I’m also not sold on the value of #2. I think we have a lot of red
herrings in this debate. Remember that if you’re truly akratic then
nothing really matters but the top edge of your yellow brick road. (If
you’re not truly akratic then just stay comfortably below the road and
ignore all this!) So I’m not buying the concern about a razor-thin
road. Right now you have a huge yellow area below you but it’s that
razor-thin top edge that’s all that matters. And remember that the
turquoise swath, as of a few days ago, has width determined by the
algorithm in #2 above. So you’ve got that guidance. Make the top edge
of your aura touch the top edge of your road and you have about a 5%
chance of a fluctuation derailing you (assuming no emergency-day
heroics).

This is probably getting too theoretical and it’s time to just try things!

Specific replies follow:

(@Katherine: You can, by emailing support, have us change how multiple weigh-ins are counted or aggregated. I think “first of the day” is still an option.)

Yup, and it’s available to users in advanced settings for custom
goals, though that’s a premium feature.
I think it’s too late for us to backpedal on the “min weight of the
day” rule as the default but I certainly see where you’re coming from
on that, and aggday=first will make Beeminder do what you want.
Another reason we bowed to pressure originally on the min-weight rule
(which I also originally objected to) was that sometimes your scale
can give an anomalous reading, or maybe you forget to go to the
bathroom first. But then I came to actually like the ridiculous gaming
of running around the block until you sweat your way back onto the
road. I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss as well as the short term goal of clawing your way back onto the
yellow brick road.

To Essy’s point that you should beemind the moving average, not the
individual weight:

The moving average is the right thing to look at to see the trend but
I don’t think it will help to beemind it. Here’s my concen, something
Philip hinted at:
Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road. It’s an emergency day and you
need to lose .5kg to get back on. Except you’re beeminding the moving
average so now you need to lose 10 times that (or whatever) to
actually bring the average down enough. You’re actually requiring a
bunch more foresight to keep from derailing, which I think is a
nonstarter. In general, being akratic and beeminding the moving
average still means that you’re skating the edge and every day there’s
some magic weight that you must weigh in at to keep from derailing.
That’s fundamental to beeminding weight. No matter how you try to
smooth the data or widen the road, if you eked by yesterday then there
will necessarily be a magic weight that you absolutely must hit today
to stay on track. It’s like a no-free-lunch theorem. Whatever we do
ends up equivalent to having a single bright line that your weight is
not allowed to cross every day.

Essy’s point about beeminding inputs vs outputs:

This is astute and is a good general principle, but I think weight is
an exception. For one, it’s so easy to measure. And as long as you
have a way to account for the randomness (which hopefully Beeminder
will continue to do) then you do have sufficient control over it. I’d
actually advocate going for the best of both worlds: beemind your
weight but keep it conservative and the stakes low – just make sure
it’s going in the right direction or staying in a reasonable range –
and also beemind all the inputs (sugar, exercise, vegetables,
anything you’re akratic about) and beemind those aggressively and with
high stakes. Beemind the inputs so that weight becomes a no-brainer.
But still beemind it, since it’s all too easy to delude yourself.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups “Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send an email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Another consideration from a business perspective as well as from an
exhausted, likely to fail persons perspective, is ease of use. The
simplicity of the explanation. “Messaging”

I know we all like to optimize these systems, but every exception makes it
harder to explain. Ideally, half a sentence should be enough to explain
everything, and every exception will confuse newcomers. With added special
math, we might have a better system, but only a very few people like to use
it, or understand it.

While I think I am clever, once I am sleepy and tired, I am also among the
confused ones.

So, one option I think I could suggest is using the color scheme that’s
used for the frames of the images. And a “hard line” at the bottom, that is
based on the ramp. Go below it, it’s red, you fail by the end of the day.
Orange zone, likely to fail tomorrow, but good for today, green zone, good
for today and likely also tomorrow.
Any and all widening will push you further away from the line, there’s no
going over the line. (Orange and red areas get wider).
The “road” may widen automatically, if stddev is increasing, but not narrow.
You can widen the road yourself. You can not narrow it.

Essentially, what I am saying is to make the minimum edge of the yellow
brick road the main line, and stack things on top of it. And all widening
is towards more results, not from the center of the road.

The easy explanation is:

  1. The line is never to be crossed. Ever. Period.
  2. The colors are guides, not guarantees. You can adjust them if you like,
    but the line is the line. Green - good, Orange, watch it, Red - you are
    pretty much dead unless you really work on it today, and your rate line is
    hard. If it’s too little, adjust your margins, tough luck.

Initial values for the borders (for example for weight) can take into
account the historical fluctuations of people with similar goals, or be
based on maybe 3*stddev of the first, flat week.

On Wed, Aug 7, 2013 at 11:25 PM, Daniel Reeves dreeves@beeminder.comwrote:

Thanks so much for the input on this, everyone. (And keep it coming!)
I’m thinking hard about it.

First, I think I was confusingly conflating 2 things before:

  1. the “can’t lose tomorrow” guarantee with immediate auto-widening
    (road accommodates any jump today as long as you were in the blue
    yesterday)
  2. determining road width based on variance of the data (road width is
    such that if your true weight followed the centerline then random
    fluctuations would put you above the road 5% of the time and below the
    road another 5% of the time)

Those are orthogonal and I think it may be uncontroversial to drop #1.
Jill gives a great example of why #1 is actually bad (purposefully
pigging out to widen your road!).

But I’m also not sold on the value of #2. I think we have a lot of red
herrings in this debate. Remember that if you’re truly akratic then
nothing really matters but the top edge of your yellow brick road. (If
you’re not truly akratic then just stay comfortably below the road and
ignore all this!) So I’m not buying the concern about a razor-thin
road. Right now you have a huge yellow area below you but it’s that
razor-thin top edge that’s all that matters. And remember that the
turquoise swath, as of a few days ago, has width determined by the
algorithm in #2 above. So you’ve got that guidance. Make the top edge
of your aura touch the top edge of your road and you have about a 5%
chance of a fluctuation derailing you (assuming no emergency-day
heroics).

This is probably getting too theoretical and it’s time to just try things!

Specific replies follow:

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

Yup, and it’s available to users in advanced settings for custom
goals, though that’s a premium feature.
I think it’s too late for us to backpedal on the “min weight of the
day” rule as the default but I certainly see where you’re coming from
on that, and aggday=first will make Beeminder do what you want.
Another reason we bowed to pressure originally on the min-weight rule
(which I also originally objected to) was that sometimes your scale
can give an anomalous reading, or maybe you forget to go to the
bathroom first. But then I came to actually like the ridiculous gaming
of running around the block until you sweat your way back onto the
road. I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss as well as the short term goal of clawing your way back onto the
yellow brick road.

To Essy’s point that you should beemind the moving average, not the
individual weight:

The moving average is the right thing to look at to see the trend but
I don’t think it will help to beemind it. Here’s my concen, something
Philip hinted at:
Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road. It’s an emergency day and you
need to lose .5kg to get back on. Except you’re beeminding the moving
average so now you need to lose 10 times that (or whatever) to
actually bring the average down enough. You’re actually requiring a
bunch more foresight to keep from derailing, which I think is a
nonstarter. In general, being akratic and beeminding the moving
average still means that you’re skating the edge and every day there’s
some magic weight that you must weigh in at to keep from derailing.
That’s fundamental to beeminding weight. No matter how you try to
smooth the data or widen the road, if you eked by yesterday then there
will necessarily be a magic weight that you absolutely must hit today
to stay on track. It’s like a no-free-lunch theorem. Whatever we do
ends up equivalent to having a single bright line that your weight is
not allowed to cross every day.

Essy’s point about beeminding inputs vs outputs:

This is astute and is a good general principle, but I think weight is
an exception. For one, it’s so easy to measure. And as long as you
have a way to account for the randomness (which hopefully Beeminder
will continue to do) then you do have sufficient control over it. I’d
actually advocate going for the best of both worlds: beemind your
weight but keep it conservative and the stakes low – just make sure
it’s going in the right direction or staying in a reasonable range –
and also beemind all the inputs (sugar, exercise, vegetables,
anything you’re akratic about) and beemind those aggressively and with
high stakes. Beemind the inputs so that weight becomes a no-brainer.
But still beemind it, since it’s all too easy to delude yourself.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google Groups
"Akratics Anonymous" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

I don’t have feedback that’s as helpful as Katherine’s or Essy’s, but I
wanted to say that I agree with their concern about unhealthy weight loss
strategies. Keep in mind how unhealthy a relationship many people have with
their weight. You can write nothing all week and churn out a blog post on
the last day, but weight loss shouldn’t work like that.

On Thursday, August 8, 2013 9:03:30 AM UTC-5, Katherine Baxter wrote:

You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road.

If I’ve been on a collision course for 9 days and haven’t done
anything about it, I’m okay with derailing. I am okay with needing to
think ahead for a weight goal.

I’d either like to beemind the purple line itself, or have a guarantee
that I won’t derail if the purple line is in the right lane.

I derailed a weight goal once due to water retention happening 2 days
in a row, even though my purple line was solidly in the right lane. I
found it so frustrating that I just stopped beeminding my weight.

And a side note:

I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss

I disagree! I don’t think running around the block when you haven’t
had anything to eat or drink that day, or eating only vegetables for a
day are healthy.

On Wed, Aug 7, 2013 at 11:25 PM, Daniel Reeves <dre...@beeminder.com<javascript:>>
wrote:

Thanks so much for the input on this, everyone. (And keep it coming!)
I’m thinking hard about it.

First, I think I was confusingly conflating 2 things before:

  1. the “can’t lose tomorrow” guarantee with immediate auto-widening
    (road accommodates any jump today as long as you were in the blue
    yesterday)
  2. determining road width based on variance of the data (road width is
    such that if your true weight followed the centerline then random
    fluctuations would put you above the road 5% of the time and below the
    road another 5% of the time)

Those are orthogonal and I think it may be uncontroversial to drop #1.
Jill gives a great example of why #1 is actually bad (purposefully
pigging out to widen your road!).

But I’m also not sold on the value of #2. I think we have a lot of red
herrings in this debate. Remember that if you’re truly akratic then
nothing really matters but the top edge of your yellow brick road. (If
you’re not truly akratic then just stay comfortably below the road and
ignore all this!) So I’m not buying the concern about a razor-thin
road. Right now you have a huge yellow area below you but it’s that
razor-thin top edge that’s all that matters. And remember that the
turquoise swath, as of a few days ago, has width determined by the
algorithm in #2 above. So you’ve got that guidance. Make the top edge
of your aura touch the top edge of your road and you have about a 5%
chance of a fluctuation derailing you (assuming no emergency-day
heroics).

This is probably getting too theoretical and it’s time to just try
things!

Specific replies follow:

(@Katherine: You can, by emailing support, have us change how multiple
weigh-ins are counted or aggregated. I think “first of the day” is still an
option.)

Yup, and it’s available to users in advanced settings for custom
goals, though that’s a premium feature.
I think it’s too late for us to backpedal on the “min weight of the
day” rule as the default but I certainly see where you’re coming from
on that, and aggday=first will make Beeminder do what you want.
Another reason we bowed to pressure originally on the min-weight rule
(which I also originally objected to) was that sometimes your scale
can give an anomalous reading, or maybe you forget to go to the
bathroom first. But then I came to actually like the ridiculous gaming
of running around the block until you sweat your way back onto the
road. I don’t find it hard to stick to the rule that you can only
engage in craziness (like running around the block, or eating only
vegetables) that’s in line with the long term goal of healthy weight
loss as well as the short term goal of clawing your way back onto the
yellow brick road.

To Essy’s point that you should beemind the moving average, not the
individual weight:

The moving average is the right thing to look at to see the trend but
I don’t think it will help to beemind it. Here’s my concen, something
Philip hinted at:
Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision
course with the road. After 9 days of 100kg each day, you find
yourself on the wrong side of the road. It’s an emergency day and you
need to lose .5kg to get back on. Except you’re beeminding the moving
average so now you need to lose 10 times that (or whatever) to
actually bring the average down enough. You’re actually requiring a
bunch more foresight to keep from derailing, which I think is a
nonstarter. In general, being akratic and beeminding the moving
average still means that you’re skating the edge and every day there’s
some magic weight that you must weigh in at to keep from derailing.
That’s fundamental to beeminding weight. No matter how you try to
smooth the data or widen the road, if you eked by yesterday then there
will necessarily be a magic weight that you absolutely must hit today
to stay on track. It’s like a no-free-lunch theorem. Whatever we do
ends up equivalent to having a single bright line that your weight is
not allowed to cross every day.

Essy’s point about beeminding inputs vs outputs:

This is astute and is a good general principle, but I think weight is
an exception. For one, it’s so easy to measure. And as long as you
have a way to account for the randomness (which hopefully Beeminder
will continue to do) then you do have sufficient control over it. I’d
actually advocate going for the best of both worlds: beemind your
weight but keep it conservative and the stakes low – just make sure
it’s going in the right direction or staying in a reasonable range –
and also beemind all the inputs (sugar, exercise, vegetables,
anything you’re akratic about) and beemind those aggressively and with
high stakes. Beemind the inputs so that weight becomes a no-brainer.
But still beemind it, since it’s all too easy to delude yourself.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com


You received this message because you are subscribed to the Google
Groups “Akratics Anonymous” group.
To unsubscribe from this group and stop receiving emails from it, send
an email to akratics+u...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

Quick clarification: You probably mean “QS” as in Quantified Self.
Were you thinking QM as in quantified-mind.com ?

And, yes, a common soap box of mine is that Beeminder should be
foremost a quantified self tool. So any hackery involving suppression
of data points or entering values that don’t correspond to an actual
thing being measured should be a non-starter. Actually I hugely
appreciate hearing about cases (like Essy not recording anomalously
high weights) where you’re tempted to deviate from the QS ideal
because (as with the case of auto-widening or smoothing or minding the
moving average, or whatever the answer to this problem is going to
be!) it tells us how we need to fix Beeminder to get rid of that
temptation.

On Thu, Aug 8, 2013 at 6:21 AM, Essentiae essentiae@gmail.com wrote:

Thin vs. wide roads
"Remember that if you’re truly akratic then nothing really matters but the
top edge of your yellow brick road. (If you’re not truly akratic then just
stay comfortably below the road and ignore all this!) So I’m not buying the
concern about a razor-thin road."

I think it’s right that it’s the top of the road that matters. That said,
the yellow brick road provides valuable visual feedback on how you’re doing:
when you’re in risky territory (orange), when you’re sailing smoothly
(blue), and when you’re rockin’ it (green). Having only green and red points
loses that valuable visual data, as well as the safety one has earned when
getting onto a green point. Beeminder doesn’t just help akratics stop doing
dumb $#!~ today, it also sort of teaches akratics the value of thinking and
working ahead. I’ve certainly noticed a tendency to start thinking things
like “I should get X goal above my road by a few days cause event Y is
coming up and I don’t want to have to think about it during that time.” I’ve
always ~thought~ that… but now I’m doing it with some goals cause I have a
visual representation of my effort. That’s a fundamental shift. I think
people who are doing better than their roads have “earned”, in a way, the
feeling of safety on the green/blue side. So road width matters, IMO. I
think custom road width for weight, if it’s not going to have auto-widening,
allows that. (BUT fix it so we can’t change the road width on the fly; there
should be a 7-day countdown for road width changes. Otherwise: “Argh, I’m
over by .2lbs today… guess I’ll change my road width…”)

Saving yourself on a given day
"Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision course
with the road. After 9 days of 100kg each day, you find yourself on the
wrong side of the road. It’s an emergency day and you need to lose .5kg to
get back on."

Then you’ve already lost.

If you have to try to lose weight in a day (when you’re going to be (and
OUGHT to be) putting food and water into your body from the moment you get
up), I think you’re in a bad place. I’m really leery of the idea of trying
to recover on any single given day. But that may be because of how I use my
weight goal. I use it to force me to tighten the behavour goals that result
in weight loss/maintenance (Daniel mentioned a few ideas later on in his
post) rather than as a “today” thing. If my weight’s not coming down, I need
to tweak some of my behavour goals, on penalty of derailing later. (For this
reason, I would love my weight goal to be set up with a 30-day akrasia
horizon, instead of a 7-day akrasia horizon.)

I think having a weight loss goal without having behavioural goals is
symptomatic of the akratic problem when connected to weight loss. Weight
goals are fundamentally different from goals like getting more work done or
limiting TV hours, and the akratic problem for that type of goal is
different from the akratic problem for the “omgz just do a little of this
today you procrastinator” type goals.

Real-time vs average
So… sometimes I talk of the world of the “ideal” about things that don’t
actually matter ~a ton~ to me. This is one of those cases. I really don’t
care that much whether the road is measuring individual datapoints or the
average, and I think the fact that it would mean making what might be some
pretty significant changes to Beeminder behind-the-scenes probably makes it
not worth it. The reason I would support it (in lieu of the auto-widening
roads) is for the unreasonable derailment cases (though, auto-widening roads
still didn’t fix many of the long-term, regular, legit derail cases).

I believe very strongly that genuine "am not losing weight so am over today"
datapoints = you’ve already lost… pay the hell up. Weight goals are
think-ahead goals and we akratics need to learn that. Losing water by
running around or veggy fasting for a day is not a healthy solution, IMO.
You’ve already lost… pay up, go have a glass of water, a normal breakfast,
and make healthier choices this week. So, whether it’s impossible to move
the moving average down after that or not means nothing to me. What worries
me, are those days that are not representative of the true trend. If you are
genuinely on the right track, you shouldn’t be penalized for something like
water retention, pms retention, or other things best left to your
imagination. Auto-widening roads or using an average doesn’t ~really~ help
that, cause in the water-retention or pms cases, we’re probably looking at
several “above goal” days even if you’re actually doing well and even if
you’re losing fat during that time.

I don’t really care how the goal is set up as long as it a) forces me to
plan my behaviour ~enough~ to meet my targets, b) doesn’t leave me thinking
about hitting the sauna to weigh in lower, c) doesn’t punish me for
non-representative changes.

For me, a longer akrasia horizon for weight goals + some different way of
counting when a derail happens (whatever it is) that forgives spikes that
aren’t representative is enough to do that… whatever those details are.

This is what I’ve done from time to time (and might do again… I’m sorry
Daniel, you’re going to hate this). I would enter my datapoints, but fail to
report my above the road points. If they were genuinely above the road b/c I
was actually failing, I would derail by not entering for several days and
running out of time. If they weren’t, I would pop back down before that
happened and the trend would continue just fine. This was fair, to me, b/c
it meant I wouldn’t be unfairly penalized… as long as the road was steep
enough to matter or the derail point was ~before~ the akrasia horizon.
Otherwise it’s too easy to change the road rate and never derail…

The problem with that is that I lose the QM value. BUT, what if weight loss
goals were set with different derail conditions. This is off the top of my
head, so there might be stupid things I’m not thinking of hidden here:

  1. Custom road width. (I want to earn the blue points, dammit! hehe. But
    more importantly, I also think visual representation if someone is losing
    ~too much~ weight is important. losing over 2-3lbs/week has gallbladder
    implications, so someone who sets their road to 2lbs with 1 safe day and who
    has a trend that is clearly steeper has a visual cue that they should eat a
    little more to stay on the healthy side of their efforts. I realize these
    are the minority cases, and not my own case, but it is one of the ways
    Beeminder can help people who are a little overzealous about it to keep from
    hurting themselves)

  2. Users (more typical than those mentioned in 1) setting longer safe day
    buffers (I like 14 because some of the legit spikes can be 10-days long but
    can catch right back up afterwards; men might prefer something like 5-7
    days, all depending on personal patterns/medical considerations). See
    attached as an example of what was a legit spike.

  3. A farther akrasia horizon for this goal (I like 30 days, but it could be
    a customizable value, maybe?… with changing that value subject to the
    ~currently~ set akrasia horizon value so that I can’t just weasel),

  4. (sorry) changing it so that it allows me to enter my above the line
    points for QM purposes, but derails as though I hadn’t. Make it so that some
    recent point matters, but not necessarily the most recent point. Maybe the
    lowest day in the “max safe days” range. I.e., if I have a 10-day max safe
    day range, then the lowest day in the last 10 days is what counts for
    derailing. BUT, it doesn’t count as though it’s today’s point, it counts as
    though I haven’t entered a point since. That way, if there fails to be
    positive change in [max safe day range] number of days, I lose. (But count
    the Hard caps by using today’s point and it’s distance from the point I need
    to be back to when I need to be back on the road.)

  5. Allow only 1 weight goal per user so that people don’t start using weight
    goals in order to create “debts” for other kinds of goals. (“I’ll just
    create a weight-type goal for my reading because then I can choose to catch
    up if I don’t keep on track” <-- FAIL!)

This could allow users who know their legit fluctuation details to plan for
them with safety, but not to use their safety net too loosely. And, although
I get your point about akratics and the upper line or the road, weight goals
are just different. They can’t be treated like other goals b/c they’re just
not like other goals. Part of heading in the right direction is realizing
that and adapting to it, I think. AND if you want to treat it like other
goals then create it like another goal: custom goal for the no eating +
sauna win.

Sorry for yet another way-too-long post!


You received this message because you are subscribed to the Google Groups
"Akratics Anonymous" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to akratics+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com

Yup, I meant QS. (“QM” is stuck in my mind from quantitative methods.)

I don’t envy your task of trying to get the mechanics just right for this
kind of goal. (Tough enough so that it doesn’t just allow… everything…
but also not so tough that there’s a pms-tax, hehe.) If anyone’ll figure
out how to do it fairly, though, it’s you guys.

This is partly a user issue, too. I mean, if we want to track it only
semi-harshly, we can always have it set up using the “don’t gain, maybe
lose” method by having a nearly imperceptible road rate with a safety
buffer just a little larger than the size of the largest safety buffer
we’ll need. I mean, if this type of goal is different and we need to think
about it differently… then… maybe ~we~ ought to think about it
differently. Plan to lose/gain, Beemind the related behaviours tightly
(but reasonably), and if we bust through our safety zones at thatpoint… well that’s on us. We get the QM value, we keep from going in the
wrong direction or losing hard won gains, and you guys need less Advil. In
a way, we may be kind of contradicting ourselves anyway: This kind of goal
is different and we need to think of it differently… but I want to think
of it the same so rig the graphs up so that I can do that and still not
lose too badly.

On the other, other hand, just spit-ballin’ a little more: You know how
there’s a countdown for dropping back down from a higher pledge? Have a
user-defined period in the settings (for this type of goal only) and a
"this is an anomalous datapoint" button. Clicking it begins a countdown
clock. You can’t derail while on the countdown clock, but you see that
you’re over and need to be back down below the line before it runs out in
3/5/7/10/14/whatever days. If it’s a legit anomaly, you will be and all
will be well. If it’s not, it’s just a delayed derailment. (I think that,
in that case, quitting during your “time off” period should trigger the
delayed derailment, and the akrasia horizon should be pushed to [7 +
countdown time left] days (to prevent weaselling like archiving before
your countdown time runs out or changing the road rate to weasel). Changing
the number of days on the safety clock should also be subject to the
akrasia horizon (maybe after the first month and the user is more familiar
with individual patterns). Otherwise, the road works just like all other
goals and is double the size of the road rate (or user-defined in custom…
with changes subject to the akrasia horizon).

On Thursday, August 8, 2013 11:58:53 PM UTC-4, Daniel Reeves wrote:

Quick clarification: You probably mean “QS” as in Quantified Self.
Were you thinking QM as in quantified-mind.com ?

And, yes, a common soap box of mine is that Beeminder should be
foremost a quantified self tool. So any hackery involving suppression
of data points or entering values that don’t correspond to an actual
thing being measured should be a non-starter. Actually I hugely
appreciate hearing about cases (like Essy not recording anomalously
high weights) where you’re tempted to deviate from the QS ideal
because (as with the case of auto-widening or smoothing or minding the
moving average, or whatever the answer to this problem is going to
be!) it tells us how we need to fix Beeminder to get rid of that
temptation.

On Thu, Aug 8, 2013 at 6:21 AM, Essentiae <esse...@gmail.com <javascript:>>
wrote:

Thin vs. wide roads
"Remember that if you’re truly akratic then nothing really matters but
the
top edge of your yellow brick road. (If you’re not truly akratic then
just
stay comfortably below the road and ignore all this!) So I’m not buying
the
concern about a razor-thin road."

I think it’s right that it’s the top of the road that matters. That
said,
the yellow brick road provides valuable visual feedback on how you’re
doing:
when you’re in risky territory (orange), when you’re sailing smoothly
(blue), and when you’re rockin’ it (green). Having only green and red
points
loses that valuable visual data, as well as the safety one has earned
when
getting onto a green point. Beeminder doesn’t just help akratics stop
doing
dumb $#!~ today, it also sort of teaches akratics the value of thinking
and
working ahead. I’ve certainly noticed a tendency to start thinking
things
like "I should get X goal above my road by a few days cause event Y is
coming up and I don’t want to have to think about it during that time."
I’ve
always ~thought~ that… but now I’m doing it with some goals cause I have
a
visual representation of my effort. That’s a fundamental shift. I think
people who are doing better than their roads have “earned”, in a way,
the
feeling of safety on the green/blue side. So road width matters, IMO. I
think custom road width for weight, if it’s not going to have
auto-widening,
allows that. (BUT fix it so we can’t change the road width on the fly;
there
should be a 7-day countdown for road width changes. Otherwise: “Argh,
I’m
over by .2lbs today… guess I’ll change my road width…”)

Saving yourself on a given day
"Imagine your road is sloping down and you’re below it at first but you
weigh in every day at the same weight, 100kg. You’re on a collision
course
with the road. After 9 days of 100kg each day, you find yourself on the
wrong side of the road. It’s an emergency day and you need to lose .5kg
to
get back on."

Then you’ve already lost.

If you have to try to lose weight in a day (when you’re going to be (and
OUGHT to be) putting food and water into your body from the moment you
get
up), I think you’re in a bad place. I’m really leery of the idea of
trying
to recover on any single given day. But that may be because of how I use
my
weight goal. I use it to force me to tighten the behavour goals that
result
in weight loss/maintenance (Daniel mentioned a few ideas later on in his
post) rather than as a “today” thing. If my weight’s not coming down, I
need
to tweak some of my behavour goals, on penalty of derailing later. (For
this
reason, I would love my weight goal to be set up with a 30-day akrasia
horizon, instead of a 7-day akrasia horizon.)

I think having a weight loss goal without having behavioural goals is
symptomatic of the akratic problem when connected to weight loss. Weight
goals are fundamentally different from goals like getting more work done
or
limiting TV hours, and the akratic problem for that type of goal is
different from the akratic problem for the “omgz just do a little of
this
today you procrastinator” type goals.

Real-time vs average
So… sometimes I talk of the world of the “ideal” about things that don’t
actually matter ~a ton~ to me. This is one of those cases. I really
don’t
care that much whether the road is measuring individual datapoints or
the
average, and I think the fact that it would mean making what might be
some
pretty significant changes to Beeminder behind-the-scenes probably makes
it
not worth it. The reason I would support it (in lieu of the
auto-widening
roads) is for the unreasonable derailment cases (though, auto-widening
roads
still didn’t fix many of the long-term, regular, legit derail cases).

I believe very strongly that genuine "am not losing weight so am over
today"
datapoints = you’ve already lost… pay the hell up. Weight goals are
think-ahead goals and we akratics need to learn that. Losing water by
running around or veggy fasting for a day is not a healthy solution,
IMO.
You’ve already lost… pay up, go have a glass of water, a normal
breakfast,
and make healthier choices this week. So, whether it’s impossible to
move
the moving average down after that or not means nothing to me. What
worries
me, are those days that are not representative of the true trend. If you
are
genuinely on the right track, you shouldn’t be penalized for something
like
water retention, pms retention, or other things best left to your
imagination. Auto-widening roads or using an average doesn’t ~really~
help
that, cause in the water-retention or pms cases, we’re probably looking
at
several “above goal” days even if you’re actually doing well and even if
you’re losing fat during that time.

I don’t really care how the goal is set up as long as it a) forces me to
plan my behaviour ~enough~ to meet my targets, b) doesn’t leave me
thinking
about hitting the sauna to weigh in lower, c) doesn’t punish me for
non-representative changes.

For me, a longer akrasia horizon for weight goals + some different way
of
counting when a derail happens (whatever it is) that forgives spikes
that
aren’t representative is enough to do that… whatever those details are.

This is what I’ve done from time to time (and might do again… I’m sorry
Daniel, you’re going to hate this). I would enter my datapoints, but
fail to
report my above the road points. If they were genuinely above the road
b/c I
was actually failing, I would derail by not entering for several days
and
running out of time. If they weren’t, I would pop back down before that
happened and the trend would continue just fine. This was fair, to me,
b/c
it meant I wouldn’t be unfairly penalized… as long as the road was steep
enough to matter or the derail point was ~before~ the akrasia horizon.
Otherwise it’s too easy to change the road rate and never derail…

The problem with that is that I lose the QM value. BUT, what if weight
loss
goals were set with different derail conditions. This is off the top of
my
head, so there might be stupid things I’m not thinking of hidden here:

  1. Custom road width. (I want to earn the blue points, dammit! hehe. But
    more importantly, I also think visual representation if someone is
    losing
    ~too much~ weight is important. losing over 2-3lbs/week has gallbladder
    implications, so someone who sets their road to 2lbs with 1 safe day and
    who
    has a trend that is clearly steeper has a visual cue that they should
    eat a
    little more to stay on the healthy side of their efforts. I realize
    these
    are the minority cases, and not my own case, but it is one of the ways
    Beeminder can help people who are a little overzealous about it to keep
    from
    hurting themselves)

  2. Users (more typical than those mentioned in 1) setting longer safe
    day
    buffers (I like 14 because some of the legit spikes can be 10-days long
    but
    can catch right back up afterwards; men might prefer something like 5-7
    days, all depending on personal patterns/medical considerations). See
    attached as an example of what was a legit spike.

  3. A farther akrasia horizon for this goal (I like 30 days, but it could
    be
    a customizable value, maybe?… with changing that value subject to the
    ~currently~ set akrasia horizon value so that I can’t just weasel),

  4. (sorry) changing it so that it allows me to enter my above the line
    points for QM purposes, but derails as though I hadn’t. Make it so that
    some
    recent point matters, but not necessarily the most recent point. Maybe
    the
    lowest day in the “max safe days” range. I.e., if I have a 10-day max
    safe
    day range, then the lowest day in the last 10 days is what counts for
    derailing. BUT, it doesn’t count as though it’s today’s point, it counts
    as
    though I haven’t entered a point since. That way, if there fails to be
    positive change in [max safe day range] number of days, I lose. (But
    count
    the Hard caps by using today’s point and it’s distance from the point I
    need
    to be back to when I need to be back on the road.)

  5. Allow only 1 weight goal per user so that people don’t start using
    weight
    goals in order to create “debts” for other kinds of goals. (“I’ll just
    create a weight-type goal for my reading because then I can choose to
    catch
    up if I don’t keep on track” <-- FAIL!)

This could allow users who know their legit fluctuation details to plan
for
them with safety, but not to use their safety net too loosely. And,
although
I get your point about akratics and the upper line or the road, weight
goals
are just different. They can’t be treated like other goals b/c they’re
just
not like other goals. Part of heading in the right direction is
realizing
that and adapting to it, I think. AND if you want to treat it like other
goals then create it like another goal: custom goal for the no eating +
sauna win.

Sorry for yet another way-too-long post!


You received this message because you are subscribed to the Google
Groups
"Akratics Anonymous" group.
To unsubscribe from this group and stop receiving emails from it, send
an
email to akratics+u...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.


http://dreev.es – search://"Daniel Reeves"
Goal tracking + Commitment contracts == http://beeminder.com