A little over a month ago I wrote about my experiment in triangular beeminding for reducing drinking. This is basically a one month progress report for how it’s gone:
The first note is that it has done exactly what I designed it to do, in that it has significantly reduced the number of nights where I have a large number of drinks. So there’s that.
However I’m getting hit by the perverse incentives this goal has set up hard. I think this is more a function of this sort of very tight do less goal structure than it is the drinking, as I found more or less the same when I was tracking coffee like this last year, but this goal setup seems to have more perverse incentives than typical.
The following are the perverse incentives I’ve encountered, with patches where I’ve figured them out.
- If I derail it becomes a “free day” where I can drink as much as I want. This actually causes me to drink significantly more than I otherwise would. Patch: On a suggestion from @drtall any data points after the roll-over are counted against the next day. So far the threat of this has been enough to stop me going over.
- My drinks have crept up in size as a result of this because the unit is “drink” irrespective of size of drink. Patch I am currently about to try: Measuring units instead. I’m worried this may be harder to figure out on the fly, but hopefully it will be OK.
- I do drink slightly more days of the week than I otherwise would have because one point of drinks is so “cheap” in comparison it feels a shame not to. I’m still not drinking every night, but I think I have two nights off per week right now in compared to 3-4 previously. I do not yet have a patch but I’m considering creating a second goal which just tracks total units per week and keeping that low too so that a day has a higher perceived cost. If I do this it may make sense to retire the triangle units and use square units (triangles are a multiple of n^2 + n, so if I’m already minding n it makes more sense to mind n^2 rather than n^2 + n. This isn’t quite true because different weightings but I don’t really want to think about those different weightings).
I think those are the main ones.
Anyway, as a result of the above I think the amount I drink as measured in units / week has probably gone up slightly since starting this goal even though the intended effect of fewer days drinking >= 3 drinks has been achieved.
I’m going to continue experimenting and see if I can patch this to make it work, as given that it has achieved the desired goal I think it’s worth seeing if I can achieve the desired goal without the side effects, but for now I remain on the fence about its validity as a technique.
Thanks for the report, this is a great read. For your last caveat regarding lots of 1pt days, perhaps a more direct way to mind this would be for a second goal of any drinking whatsoever, regardless the magnitude.
It would be handy if you could fill this programmatically from the first, so you only have to input data a single time.
The IFTTT integration is perfect for this, using a recipe like this one to count the datapoints as they fly past.
Yeah, I’ve thought about doing similar.
Another patch I’ve thought of trying but haven’t is to start all days with drinking at one unit as a “Well I’m probably rounding down a bit” fudge factor on the drinks. This makes a single drink day more expensive than it otherwise would have been.
Right now though I’m trying not to change too much at once, and I’m not that worried about the more days drinking thing, so I’m not going to tinker with it for another week or two other than trying to keep up tracking units.
- If I derail it becomes a “free day” where I can drink as much as I want.
This actually causes me to drink significantly more than I otherwise
would. Patch: On a suggestion from @drtall
any data points after the roll-over are counted against the next day.
So far the threat of this has been enough to stop me going over.
What are the mechanics of this? Do you just implement this manually with fine-print? How does it interact with the fact that you can actually be in the red (sometimes) with do-less goals and not derail?
Manually with fine print. The condition is “be in the red” not “be derailed”. though now that I think about that that’s slightly problematic with the way the triangular aggregation works because it can turn a day I would have derailed on into a day I would have derailed on.
In practice the way this works is that once I’m in the red I stop drinking.
I’ve been holding off on trying this system out because I’m afraid of this side effect, but as of today I’m going to start an experiment. I’m going to be doing triangular drink tracking, except that the 1st drink of a day costs 2 points (instead of 1) and the 2nd drink of a day costs 1 point (instead of 2).
I think this will encourage me to find days where it makes sense to abstain from drinking, and the 2 point “bonus” from abstaining is more substantial. I don’t mind that I’m more likely to have double drink days, because 1) the 3rd drink of a day is still very expensive and 2) Mitchell and Webb fans will agree that slightly less than two drinks is where one should stop anyway.
I’ll report back after the experiment.
I have in fact just given up on triangle alcohol. I’m going to keep up the alcohol tracking, but yesterday I turned it down to linear aggregation, set the number of units per week to something sensible and retroratchetted to the same number of days of buffer as I had before.
I may turn it back on, or add a secondary copy of the alcohol tracking with a square alcohol, but for now what I want is a baseline for how much of this is just the effects of alcohol tracking and how much is the specific incentives of the triangles. I suspect the latter.
Just to report back, I’ve been having great success with the triangular 2,1,3,4,… system. It feels good to skip a day because +2 points for a day off is pretty meaningful.
Is anyone still doing alcohol tracking? I’m interested in setting up a goal for this, and am torn between just a >X drinks per week goal or a triangle goal. What has worked fo you?
I have a do less goal to track units of alcohol consumed, and a do more goal to track alcohol-free days.
No triangles here, just straight counting, with a script to help with the math of calculating units…
I do the exact same. I tried triangular scoring on consecutive dry days, but it was a nightmare to reason about near deadlines.
I’ve been doing a version of this with a sort of “inverse triangular” scheme related to the type of drinks I have. My goal was primarily to have more nights of the week where I don’t drink, and secondarily to reduce the amount of beer I drink, mostly for the reduced calories.
I’ve found that I don’t respond well to goals which limit the number of drinks I have per night; there’s a lot of reasoning upthread about why that’s the case which I agree with. I’ve had much more success, though, with the following system:
Start a “Do more” goal with the units “Days without drinking.” Set the rate appropriately .
Give yourself the following points:
1 point for a day with no alcohol
.5 for a day with no beer and only one drink
.25 for a day with no beer
I’ve found that my incentives are to drink less beer (good!), to drink more wine and liquor on nights when I have a social reason to drink (less good, but good overall for my total calories), and to have more weeknights when I have no alcoholic drinks (great!).
 Mine is set to 2.8, which is the appropriate rate to force Weekends Off to behave in such a way that I’m required to have 2 days during the week without alcohol. Other appropriate values are 1.4, 4.2, 5.6, or 7. YMMV, and you may not care about having weekends free to drink whatever you like, so of course everyone’s ideal behavior will vary.
Ok, I am going to try to have a Do Less goal for units and a Do Less goal for days consumed alcohol. I chose Do Less instead of Do More as it makes it easier to feed from the units goal with IFTTT.
You’ll probably want to turn off the ‘pessimistic presumption’ for those goals, otherwise you might find yourself getting more points than you’d expect to.