In yesterday’s post, I had a great back and forth with Lobo on decision strategies:
While the usual thinking is to get to the optimum plan, actually the less cognitive intensive method is to choose something that satisfies 80% of the most important areas. Optimizing the other 20% is usually not worth the cost… There’s a reason for insurance brokers, finance experts, etc. Maybe this is a place for AI. Plug in your situation and have the algorithm spit out the 3-5 best fits.
I’ve mentioned this many times in the past. I am firmly convinced that if I was facing an array of 20, 30, 40, 50, 121 distinct health insurance choices on Healthcare.gov that I would only pick the optimal plan by random, dumb luck. I am firmly convinced that I can avoid the worst plans. And this is me; I’m someone who spends most of my professional life thinking about health insurance, benefit designs, payment models and the right hand skew of risk. I like this shit. I’m weird. And if I am convinced that I’m unlikely to make the optimal choice when there is a large, uncurated choice menu, I’m very confident that unrestricted choice lists will lead to sub-optimal choice on a regular basis.
I agree that this is a place for AI; Jon Gruber et al have a 2020 working paper that looked at the improvements in choice selection once an AI assist was added to a Medicare Advantage exchange:
The addition of AI-based decision support improves outcomes by $278 on average and substantially reduces heterogeneity in broker performance. Experts efficiently synthesize private information, incorporating AI-based recommendations along dimensions that are well suited to AI (e.g. total expected patient costs), but overruling AI-based recommendations along dimensions for which humans are better suited (e.g. specifics of doctor networks)… While AI is a complement to skill on average, we find that it is a substitute across the skill distribution; lower quality agents provide better recommendations with AI than the top agents did without it.
I want to pull back a little more and talk about the different choice strategies people can use.
There is optimization. This is a search for the “best” plan without regard to the search costs. An individual makes a comprehensive list of all relevant attributes and has a model of substitution, trade-offs and minimal acceptable values. And then they look through the entire universe of choice to find the best thing possible. This works pretty well when looking for a candy bar at the super market check-out line (Mounds!) but on complex products, it is cognitively expensive.
There is satisficing. This is a “good enough” solution. With satisficing, the chooser sets up a list of minimal acceptable criteria. And once they find something that meets that acceptable criteria, they stop searching. This is, at least, a cousin to the “80%” solution Lobo mentions above. Satisficing acknowledges search and information costs and that these costs can be high for minimal gain. An optimal choice is possible, but depending on the size of the choice menu, it could be unlikely. But good enough is often good enough as we all know from the pre-COVID days of looking for a parking space in the mall parking lot at 2:00pm on the Saturday before Christmas.
Minimizing-maximum regret is another decision choice strategy. In mini-max, the objective is not to choose the best plan. The objective is to pick out the least bad outcome. This is a way of dealing with uncertainty. In the insurance context, this probably leads to higher premiums in the pursuit of lower out of pocket maximums, but that is a reasonable trade-off. Mini-max when applied to a large choice menu is very unlikely to pick the optimal plan. Instead, it is a trade to get the least downside at the cost of giving up a lot of upside.
And then there are decision strategies that range from listening to what someone said on Facebook, information overload, inertia and best guesses.
From a choice architecture view, one of the most important things that a choice architect can do is eliminate the objectively hideous choices from the choice menu. Most HR departments do this; Duke offers four employee health insurance plans even as there are 50+ plans offered on the ACA Exchange in the local community. When I worked at UPMC Health Plan, HR offered either four or five plans. HR did most of the hard work of getting rid of the objectively bad choices.
And there is a lot of pay-off in getting rid of the worst choices. Abaluck and Gruber found that restricting the choice menu reduced consumer costs and improved average choice quality. Abaluck et al found in the Medicare Advantage space that eliminating the bottom 5% of Medicare Advantage plans on the metric of difference from expected mortality would lead to thousands of additional life-years for beneficiaries. The reseach on dominated plan choice where there is an objectively inferior plan being offered consistently shows several hundred dollars in additional costs that should not be paid but are.
So when making choices, having a good (or at least a good enough) strategy is important at the individual level, but good choice curation and architecture is important as well.