How do you measure this? I think if you could measure subjective suffering, the situation may be better. Did you assume that people needed the extra $50 for something?
1) 1% of people earn $1 per day, and 99% of people earn $100 per day 2) Everyone earns $1.01 per day
I think most people, myself included, would take their chances with the first option. The Veil of Ignorance is a useful thought experiment, but I don't think many people would consider a simple Minimax algorithm to be reasonable, and would take into account the well-being of the average person as well.
Do you have any evidence for the claim that minmaxing causes a steep drop off in overall expected utility? I'm not sure if research has been done about this with traffic lights, but I think that could be a good case study about this. Minmax for traffic light timings would mean minimizing the maximum wait time at a traffic light, though I could see how such a strategy might hurt the overall throughput of the intersection. In the extreme case, you have one granny trying to cross a 6-lane stroad during rush hour; in the pure throughput model, the road would never stop for granny and in the minmax model, she would only need to wait about as long as it takes her to cross the road. Obviously, different models would still allow granny to cross the road eventually and maybe with less of a throughput hit than pure minmax. All of this is speculation though and I'm curious what the throughput hit would be for minmax.
My evidence is just experience with optimization/utility theory. If you're looking for a paper describing this one takeaway out of thousands from this body of theory, I'm not sure what it's called.
In general, any strategic constraint reduces EV, unless the constraint happens to already be a component of the optimal strategy. That doesn't tell you how steep the dropoff is, just that it exists. Determining the variance/EV tradeoff precisely depends entirely on the structure of the optimization space.
However, any complex optimization space exhibits essentially the same behavior; crushing the outcome distribution to be super narrow (latency, wealth, ROI, test scores, etc.) is typically only possible at great cost to the median and average cases.
Minimax is emphatically not even remotely close to EV maximization, especially for high-dimensional optimization spaces with sharp risk/return curves as I described.
Unsure why you would think I was expecting a NYT article (except as some weird cultural signaling point?), I was thinking more along the lines of a welfare economics or econometrics journal article.
Apologies if this seems harsh, but people's habit of asking for a source on every original or propositional claim on this website really bugs me. Sometimes people are sharing new observations!
Not going to keep responding, I've said all I wanted to.
I am not really clear on what background you could have such a background but be unable to generate this fact, Wikipedia seems to suggest it strongly when it points out that "In non-zero-sum games, this is not generally the same as minimizing the opponent's maximum gain, nor the same as the Nash equilibrium strategy" [2].
This isn't a personal attack, but you really could have got this one solved by looking at either the thread or Wikipedia, but chose to assign someone else research homework instead. If you're curious, check! It's really fun and you learn a lot.
[1]: https://news.ycombinator.com/item?id=31588384 [2]: https://en.wikipedia.org/wiki/Minimax#Maximin