Satisficing is choosing a solution that meets minimally defined criteria because the search costs of finding the optimal solution to a problem are too high and we are humans with bounded rationality and limited focus and attention spans. A common case of satisficing is finding and taking the first parking space seen at the mall on December 23rd. Sure it might be half a mile from the entrance and a mile from the one store you really need to get to and there might be a spot three spaces from the entrance that will be available just as you drive by but the spot is good enough so you take it.
What does this have to do with health insurance?
In my personal mental model of the firm, I believe that there are very few problems that are optimized and many problems that are satisficed. This includes a subset of problems that have the trappings of an optimization problem but there are numerous good enough assumptions built into the model so the optimization is either a GIGO optimization or more likely a good enough local near maximization instead of a universal optimization. Big company wide strategic decisions are likely to be nearly optimized while day to day decision making is often an accretion of satisficing decisions within multi-objective cross cutting frameworks of constraints.
This includes Human Resources and Employee Benefits. Under a strict optimization strategy, everyone would be paid their marginal value. Everyone would receive just enough of a total compensation package to match their actual contributions to the firm. That does not happen. There are some bullshit artists who have talked their way up. There are some people who have not been promoted for years and as soon as they leave for the competitor at the other end of town two entire departments fall apart because they were the off the org chart glue that made all of the kludges work smoothly.
Health insurance and benefits is also a problem. Typically the constraints of a problem space from the point of view of the benefit manager will be a fairly hard upper boundary on cost, a moderately hard constraint to not piss off very senior management by making their lives or more importantly their spouses’ lives more difficult, a softer constraint of keeping the offering vaguely competitive with peer offerings. Those will be the formal constraints. There can also be informal constraints. One of the big ones is keeping the number of people yelling at the benefit manager throughout the year to a minimum and making sure that people occassionally talk to the benefit manager when there is not a problem with benefits.
And this is where satisficing comes in.
We could build an optimization model and enter every single carrier that has a network and a product in a region into that model. For very large firms that basically happens with requests for proposals (RFPs). But for smaller firms in the small and mid-size group markets they might be working with a broker (who is trying to balance the principal’s constraints with the agent’s own constraints) or a couple of people in house who need to present options.
The decision process uses total premium costs as the first cut. Plan offerings that are significantly above the proposed benefit budget are often rejected out of hand. After this it is a series of trade-offs. Wider networks with less restrictions such as a broad PPO will mean a higher premium. Going to a narrow network HMO will save 17%. However that narrower HMO will piss off the husband of the CFO so that is outside of the solution space. Switching to a carrier that has a lower administrative burden and a higher MLR leads to $7 per life per month less in costs but almost guarantees more employee complaints. The final benefit package is an accretion of non-optimal but minimally satisfactory trade-offs within the organizational’s solution space. Seldom is cost the only boundary of the solution space.
This is different than on-Exchange as there price is either the primary driver of the decision maker or health status is the primary driver of the decision maker. ESI has a much more muddled set of drivers for the actual decision makers.
And finally, it is HR, they are the ones who get to put together videos like the one below:
Ah, there’s a fancy word for that now: satisficing. A former colleague once told me that the motto he was trying to follow in his struggle against his tendency towards perfectionism was ‘good enough is perfect’.
What Have the Romans Ever Done for Us?
Some economists never give up on the idea of optimizing…because it’s completely ingrained in their thinking from Econ 101. You’re not satisficing, you’re optimizing with an additional consideration of the marginal information costs of searching further for an even more optimal solution, which may or may not exist. The solution you have found is the optimal one given consideration of information costs.
I think in practice people satisifice all the time. They also fall back on search images/patterns to limit nearly unlimited choices. Opportunities for actual optimization are few and far between, but that doesn’t keep economists from fantasizing about them.
The best part of that Archer video was Mitch McConnell wearing a blue mask and speaking Russian at the end of it.
@What Have the Romans Ever Done for Us?: But at what point is adding another implied constraint on optimization problems merely adding an epicycle?
My intellectual history has always been strongly influenced by bounded rationality and satisficing ( https://balloon-juice.com/2014/07/21/system-success-and-people-with-options/ https://balloon-juice.com/2013/09/26/i-aint-got-no-satisficing/ )
That was just the environment where I was trained plus a mindset in that I deal with fuzzy very well and clearly distinct lines poorly.
As someone who has been managing benefits at my medium sized company for 18 years, the cost component is a never ending upward spiral of higher and higher premiums.
At some point there is nothing to do but cut benefits, even if it pisses people off.
If we ever got healthcare costs under control, and going down, we’d see better benefits, which would result in less money going from workers to the healthcare sector and more money for the rest of the economy.
I really do wonder how much money rising healthcare costs have sucked out of the rest of the economy.
@What Have the Romans Ever Done for Us?:
I think the idea of modeling satificing as an optimization problem with additional constraints is a reasonable one. It lets them fit it into their existing paradigm, which makes it more intellectually acceptable and means they have a big set of tools already developed to analyze it. More importantly, it lets you treat satisficing quantitatively, answering questions like how much effort people put in before taking what they can, rather than as an ad hoc assumption.
I’m currently on the phone with my medical carrier, and they’re telling me that I’m having a billing problem because while the hospital I took my two-day-old to (this is back in July, she’s good now) was in network, but the ER doc was not. The carrier’s working with me on this, but what gives? (If you’ve already covered this and I missed it, my apologies.)
@Roger Moore: One problem with modeling satisficing as optimization with additional constraints is that it can just make finding the mathematical solution impossible, especially if they insist on actual optimal solutions. Explicitly recognizing satisficing means making clever tools to do support exactly that and is usually related to deal up-front-and-center that not all desires / goals / constraints even are captured adequately by mathematical equations. These, along with heuristic solution techniques and Decision Support techniques were certainly almost the real focus when I was being beat about the head by LP and OR techniques in grad school late 80’s early 90s.
Richard, you really have to stop with the clickbaity titles. It makes you seem cheap.
” But at what point is adding another implied constraint on optimization problems merely adding an epicycle?”
IMHO, when you have to repeatedly blow up the dimensionality of the system to get a domain large enough to accommodate behavior that cannot be explained as a solution to any possible optimization problem in a lower dimensional space. If this habit becomes routine, then you get “well, OK, the choice is a solution to an optimization problem ‘all things considered’, and for this problem we will jigger up the ‘all things considered’ we need to model it as an optimization problem”. . For any empirical science, ‘ALL things considered’ is way too many to appeal to on a routine basis. But you see that often in microeconomics.
There is to put it politely, a tension between existence theorems that prove pretty much anything can be rationalized as a (often uncomputable) solution to an optimization problem and the existence of algorithms that can compute them. Of course, you can always go black box and appeal to revealed preference type thinking and say it is not economics’ or decisions theory’s problem how or what people compute, but we just study their solutions. Then the problem is kicked out to whether you observe behavior that can rationalized in an empirically verifiable way on any stable and operationally definable domain. For some important problems, like general equilibrium economy with intertemporal assets and decisions (let alone production), the answer seems to be no.
For some reason, the literature on this topic that gets to the core of the issue is (as far as I can tell) very small. A good behavioral economist wrote a nice intuitive and expository article on it, but the name escapes me now. Will post when I remember.
@jl: People’s preferences violate the triangle inequality all the time, why their decisions should be modeled as optimization problems always seemed at best a heuristic in itself.
” For some important problems, like general equilibrium economy with intertemporal assets and decisions (let alone production), the answer seems to be no. ”
To clarify, what I mean is that any behavior you observe can be reationalized as some kind of optimization problem if you blow up the domain (which includes inherently unobservable parameters like subjective preferences of the agents) enough. So, this approach leads to empirically empty theories unless you impose a lot of seperability conditions that applied work indicates are violated.
It is true that if you are content to live in an econ 101 world of pure flow consumption and production with no dependence on what happens at different times, then we are cooking with gas and we have optimization based theories with empirical content.
@scav: I agree, maybe it is a heuristic notion that is so deeply embedded in our goal oriented common sense that it seems like an obvious fundamental principle. Microeconomists have been flailing away at the approach for over 140 years now, with not much to show for it, at least in terms of producing a stable empirical theory. As a modelling strategy it can produce temporary successes and (depending on your mathematical tastes) dazzling technique and insights, or boring applied math with no application.
@What Have the Romans Ever Done for Us?:
My project management oriented sister introduced my wife and I to the concept of satisficing vs. optimizing a couple of years ago. I think that we always did a lot of satisficing, but I think now we save a few more decision cycles just knowing that were looking for an acceptable choice, not the best of all worlds choice.
I’d also agree that people limit their options deliberately. I need someone to help me plan landscaping. My sister just had a job done. I get her recommendation. We do that all the time.
I took a positive psychology course and one of the keys to happiness was satisficing. Essentially, the more things in your life you can put into the satisficing category, the less unhappiness you will have. Also, satisficing means setting a floor on what you’ll accept, not a limit.
@Highway Rob: What state — let me respond at length after I get a tooth pulled this afternoon
Using satisfying logic rather than optimizing also seems to make sense in a Darwinian evolutionary context. Evolution really seems to operate with a sharper edge by pruning the non-satisfactory solutions (they get eaten or die). Generally satisfactory critters are the bulk of the breeding population, which can be a right mixture of somewhat variable critters. What they do is provide a baselien of alternative solutions to being that critter. Pure optimization evolution can get critters into dead ends where they are optimized for one niche, one set of conditions and when things change, it may be hard to back out of it. Having a breeding pool of good enoughs with variation is pretty much insurance and a head-start on adapting to changing conditions. The optimizing function, the better adapted critters having more babies is still important for understanding the spread of better adapted characteristics throughout the population, but getting the balance between satisfying and optimizing seems important.
https://balloon-juice.com/2015/06/04/cracking-down-on-the-invisible-pears/ a decent starting point… Families USA is a good resource too
@Richard Mayhew: Tx was in TX. Carrier is BCBS-IL.
@Highway Rob: ok you are fucked
@Richard Mayhew: Thanks. That’s the exact shell game I encountered. A shame, too, ’cause the hospital itself is one of the best pediatric facilities there is, I hate to see them playing it this way.
What Have the Romans Ever Done for Us?
@Roger Moore: The real problem with optimization is the underlying utility theory which can be used to justify any decision as utility maximizing. Buy a new car, you’re just maximizing your utility…get hooked on heroin, ditto. It becomes a tautalogical explanation for any decision,which means it really doesn’t explain why anyone makes any particular decision
@Wapiti: I use patterns all the time. I buy, like, the same 4 kinds of cereal repeatedly. Is there a cereal I might like better out there? Sure, I haven’t tried them all…but to pick new ones I’d have to read the nutrition labels, etc. and it’s just not worth the effort or time…so I grab the yellow Cheerios box and move on to another isle.