Progressing depth of uncertainty
Individuaization is digging yourself out the hole by taking or inventing a small set of stories, terms, ideas, and assuming they work. You do not have enough time to evaluate these fully in your lifetime. In a granular society, many ideas are assumed, so there is a better chance that one of them works best. And knowledge is improved. In a mostly homogenous society, you can carefully watch you peers who share mostly similar assumptions, but with a few key differences. You can then watch what works better or worse. However, you have no hope of learning about the problems with the things everyone you know has assumed. So progress is low dimentional, but potentially much quicker than in a more open society.
If there is a more homogenous soceity, and the assumptions they hold work, they tend to do better than a granular society. A granular society tends to do better than a homogenous one when the assumptions don’t work well.
What does work well mean?
Note that the first can hold for a long time even if it does poorly for individual’s fitness. However, in the very long run, the second should eventually win out.
What can a water molecule tell of its fate? It can look to the left, and see another molecule about to collide with it. It can look to the right, and tell that there is nothing there, suggesting it will be pushed in that direction soon. It can tell if it is closely packed with and connected to other molecules in a liquid or bouncing around loosely in a gas.
If this is an observant and well informed molecule, perhaps it can even look at neighboring water droplets in the cloud, and see that their density and size is increasing, suggesting it is about to fall to the ground as rain. But after it falls to the ground, where will it end up? Will it end up mixing will salt in the ocean? Being absorbed by a tree on land? Or something else entirely, something out of the realm of experience and knowledge of this molecule and all its neighbors?
This last case is deep uncertainty, this uncertainty that defies all attempts at estimation and probability. That dives off not only the long tail of the distribution, but out of the dimensional scale of a multidimensional analysis entirely.
Now, for a student of probability, this idea seems lazy and unnatural. Everything fits on the dimensional scale by definition. Anything independent of the variables is treated as error or noise, and is treated formally in the analysis. But what sorts of assumptions are required to treat it formally? These assumptions are the form deep uncertainty takes in mathematical treatments of uncertainty.
Take p values, the central statistical tool in the social sciences. p values suggest how probable a known variable correlates with the outcome, given that there are unknowns creating noise in the outcome. This is incredibly powerful tool for knowledge because it allows us to determine correlations in the world. The problem? It assumes that the noise is normally distributed. If the noise has a long tail, then the estimated variance can be much lower than the actual variance, creating much better p values than should actually occur.
Worse still, the assumption that the data is independent. What a ridiculous assumption in the social sciences. One of the best evidenced theories in education is that teacher quality can be evaluated effectively using standardized tests, and that removing the worst teachers can dramatically improve student performance. This has been shown consistently by a number of large scale randomized controlled trials, and even larger multivariate analysis and natural experiments. Yet there is a problem: Teachers hate it, and teacher organizations and political lobbies fight it, and tend to destroy its effectiveness. These political forces only get into gear when policies are enforced across large areas. So the assumption of independence is false, despite random sampling of schools across the country, because the political force correlates with the scope of the policy.
Worst of all, the implicit assumption that population under study is uniform and unreactive. Take nutrition. Nutrition and diet is a very important part of each person’s life. Adults have a long list of dietary habits and preferences. These habits are difficult to change psychologically, and attempts to change them can sometimes lead to dietary collapse like cheat days, relentless snacking, etc. Not as obviously, it seems plausible that these habits have a dynamic relationship with the bacteria in your gut, which respond to what you eat and also mysteriously seem to affect every almost every aspect of your bodily function, including psychological conditions like depression and anxiety, which in turn can cause dietary collapse. Meanwhile, a scientist comes along with a hypothesis that eating 50g of whole grains a day reduces risk of heart disease on average. On its own, this sounds like a solid, concrete hypothesis to test, and given enough resources, it seems likely that a statistically significant result can be found, on way or the other, by randomized controlled testing. But is this fact generated by RCTs really useful? Is it useful to the person who starts lathering on trans-fat laden margarine on their whole wheat toast? Both were supposed to be a healthy alternative to traditional food. More importantly, is it useful to the person who starts dumping sugar on whole grain breakfast cereal, or to the person who starts eating ice cream after their whole wheat pasta? Probably not. And importantly, these people probably know that the dietary change is not good for them. And what about the people who it does work for? Well, they will probably start feeling better in a few weeks: more energy, less fatigue, etc. Finally, what about people who have no interest in reducing their risk of heart disease? People who’s lifestyle or food culture is more important that living a couple more years? Are we really to say that this is irresponsible behavior that should be ostracized? Especially if we don’t actually know that whole grain is good for them in particular?
This rabbit hole of catering to the mean goes extremely deep. College raises people’s income on average, therefore everyone should go to college. Barack Obama made this argument despite the fact that we already know that people who don’t go to college are on average less suited for it, and so we have no idea if it will actually raise their incomes or not. Lets say you do a randomized controlled trial where you take random people and force them through college, and it turns out that these people got higher wages than their peers. You still don’t know that everyone should go to college because the job market might be flooded with people with college degrees, lowering wages for entry level degree jobs, and raising wages for entry level non-degree jobs. Not to mention that people may not be entirely happy about pursuing a completely different career than the people they grew up with.
There is also catering to the observables. All great colleges in this country (Ivy leagues, Stanford, etc) have beautiful, green campuses, excellent sports venues, excellent cafeterias or eating options. Since these are all aristocratic institutions, these are a must to appeal to the sensibilities of the elites they serve. Meanwhile the real characteristics which impact your education, are your peer group and professors, specifically in your specialty. But you are just a confused high school graduate who has no idea what you want to specialize in, you have no idea what professors and peers are like at different colleges, and your parents want to reduce the risk of you ending up somewhere bad or disreputable. So you go for the colleges with the greenest lawns, despite the fact that you are not an aristocrat, and care little for these surface features.
Now, imagine you are an employer. You are looking around for solid, respectable junior employees who know what they are doing, and can be counted on. You are a smaller company, so you can only have career fair booths in a few colleges. You have little idea what colleges are good or bad, all you know is what colleges your friends went to–and also where the greenest lawns are. And of course, the most promising students got sent to these green lawn schools, and so they are also the most promising graduates.
So now green lawns have an actual effect on employment outcome. So the cycle continues. Meanwhile, colleges that have less green lawns are falling behind, so they need to spend on greener lawns to stop their peer group from degrading entirely. So they raise tuition to cover that cost. So colleges are signaling to students and employers, students are signaling to colleges and employers, and costs rise for no reason.
Note that this whole thing is a “worst of all worlds” scenario. There are many solutions to this pathological spending, the easiest being a national university system where the universities rankings and specialties is set by decree, so they cannot compete over signals like green lawns. However, this is still suboptimal. If some really insightful and knowledgeable counselor could direct students towards colleges which are most suited for them, and really knowledgeable, insightful consultants could direct employers to places to look for students which best meet their needs and resources, this would create much more optimal solutions.
Appeals to the mean is also a powerful force in housing dynamics. Gentrification is a drive to move towards an area other rich, successful people are. Urban and rural decay are forces where successful people move away from areas where people aren’t as successful. These forces create poverty and drive the evils of inequality: crime, resentment, etc. And these forces aren’t even always beneficial to the rich. For example, the richest school in the area may not always be the best, and certainly not the best for every kind of student. Lower performing graduates from rich neighborhoods may be lost in the competitive field of cooperate management, but may be a great fit for a government role that they never heard about.
Now lets take healthcare. Healthcare does not an appeal to the mean, but to the extreme. If you are a doctor that makes the right call 95% of the time, you may still be prosecuted if you make the wrong call and your patent dies 5% of the time. So extra tests are ordered, extra care in sanitation is required, etc. Worse still, hospital administrators are wandering around, looking for problems to solve before they appear. They are assessing risk over an entire hospital, not just a single doctor, making even more cautious. This leads to decisions like: sterilizing medical tools can be botched by laziness, so just throw the tools away (yes, this is common practice). Patients must be separated by more distance to reduce risk of infectious spread. Tests must have 99.5% accuracy, not just 98%, even if they cost 10 times as much. The newest equipment and medicines must be used, in case someone accuses the hospital of not trying their best to save someone’s life, even if that new medicine has not been proven to be more effective. Doctors must fill out extensive paperwork regarding all this administration to make sure their asses are covered in case of a lawsuit, decreasing their capacity for taking care of patients in the best way they can.
Patients that do not visit the hospital because of cost are of course, not the hospital’s problem. Patents who die because they feel lost and uncared for in an overly process oriented medical system are unlikely to generate cause to sue.
These risk reducing measures are particularly pernicious in healthcare, but they appear in every bureaucratic system. If you have ever been in a large organization, you have probably attended a meeting where you review recent failures. Failures in many different areas are documented, some policies are created to try to reduce the risk of these failures happening again, and someone is tasked with enforcing these policies.
But how are these policies justified? They are justified by what people know for certain, and can communicate in a meeting, not by the weird urges, feelings and instinct that people use to guide their personal lives. For example, say a newspaper company measures journalist output by number of articles that make it past the editor. The editor is attracted so a junior journalist, and starts holding the articles hostage to a sexual relationship. The journalist refuses, and is fired due to low output. Now, a number of senior employees at this meeting may have felt weird vibes about this situation, and feel that something is wrong, but because all the concrete evidence happened behind closed doors, nothing is said, and the editor continues to hire some new young woman.
Meanwhile, if the evidence is communicable, and falls under a familiar story, excessive or meaningless action can be taken. A meeting about a sexual abuser who gets caught may initiate a whole company wide program to educate employees about how to prevent sexual abuse, whether or not this program has any actual effect at all. Because what is important to the executives in the meeting is not the sexual abuse itself, but how employees feel like they are being protected and taken care of to the “best of their ability”.
Take TSA security at airports. 9-11 was a horrible attack which left an entire country shaken and insecure. People were desperate for any solutions that would stop a future attack. The systems installed have been an enormous pain to all travelers, especially infrequent fliers who aren’t accustomed to the situation and naturally feel assaulted when they are patted down and their stuff is inspected. But most travelers feel it is worth if to stop another horrible attack. But does it actually work? Luckily, the people who created the system were wise enough to perform stress tests of the system on a regular basis to assess its effectiveness at stopping people from smuggling weapons and bombs onto airplanes. And it fails every single time by absurd margins. So is the system removed? No, because frequent fliers have convinced themselves for so long that these are necessary and no one is bothered to take the effort to convince them otherwise, (probably because they aren’t the ones paying for it).
The military is especially bad about creating useless procedures. Following procedures carefully and precisely is very important in life and death scenarios soldiers find themselves in. Senior officers and politicians in the military start getting used to the idea of policies and procedures as a fix to every problem. The naval vessels are getting rusty and need to be cleaned? Create highly refined rotations of duties with multiple levels of enforcement. People don’t want to join the navy because it is all about cleaning boats? Not their problem.
So lets talk about useless policies. Policies are different from procedures in that there is even more separation between politicians, voters, and the people being impacted. The same is true in authoritarian regimes, where the authoritarian may be very effective at handling issues close to their attention, but very poor at handling issues not in their clear focus. Understanding this great degree of separation, the founding fathers created a system where policy would only move slowly and in limited areas so that the federal state law had limited impact and only if it involved issues that were widely agreed upon.
But schoolteachers and populists seem to engage in this absurd idea that government is there to take care of any and all social problems that are around. After all, if social problems exist, that must mean existing social structures are not enough to solve them, so we need to get together and create new structures to solve them, and how else can this be done other than government policy?
The problem is that problems are easy to identify, and solutions easy to argue for, but difficult to analyze and check their effectiveness. Take poverty. Poverty is a horrible, self-perpetuating economic, social and psychological condition that increases the chance of committing crime and decreases the chance of their children having good lives. No one wants poverty. So give the impoverished people money, right? The government has enough money to pay for basic food and shelter, and with that, they should be out of poverty, right? This is the easy story behind all anti-poverty programs.
But what if they use this money for drug and alcohol use instead of food and shelter? What if they start to feel entitled and
These problems can be well described and argued about in mathematical forms. Statisticians have developed new methods similar to p values but which have much weaker assumptions for the noise distributions. Whole branches of study, such as psychology, have started deemphasizing the value of p values and other statistical measures, as they tell you little about the magnitude of correlations, only about their existence. Randomized controlled trials have started falling out of fashion due to the fact that they are rarely taken seriously by political forces; perhaps because politicians, unlike academics, understand the political and economic pushback that large scale execution creates.
But there is a
These failures are fairly well understood.
After all, these things are independent from the data, i.e. you don’t have any way of studying them, so why bother? Well the reason you bother is that these uncertainties are often far more important to your future success than the thing you are studying.