The Case Against Education: Splitting the Education Premium Pie and Considering IQ

post by Zvi · 2018-04-29T12:40:00.219Z · LW · GW · 5 comments

Contents

5 comments

Previously: The Case Against Education: Foundations, The Case Against Education

VI

Bryan Caplan makes several cases against education. His primary case revolves around the question of where the education premium comes from – no one doubts that the better educated make more money. As we noted last time in Part V, there are three categories of explanation: Human capital (school teaches you valuable things, so you produce more value), signaling (school sends a valuable signal, so you get better opportunities and better pay), and ability bias (those who would always have earned more money attend more school, and school is taking credit for effects it didn’t cause).

As I noted in part V, it’s undeniable (despite frequently being denied) that ability bias plays a large part in the story, as does signaling, but there’s large uncertainty on how much. It’s a difficult problem, worth looking at from multiple angles.

One way is to ask, what are the relevant measures of human capital? What do we learn in school that helps us be more productive? If we can measure that, we know how much of the story is human capital, leaving the rest to signaling and ability bias.

Another way in is to start with ability bias. If we can correct our measures for underlying ability via a natural experiment or finding good proxy measures for underlying ability, the difference between the adjusted and non-adjusted numbers represents ability bias. There’s lots of types of ability, but what we can measure and correct for would represent a safe lower bound on ability bias.

If you had a good measure of human capital, you could combine these two approaches. Measure human capital before school, then measure it again after school, and compare it to the measure of those who didn’t attend. The additional increase in human capital from school represents human capital, and the starting difference is ability bias. The rest is signaling (or unmeasured abilities and capital, so one must be careful).

We could also start by measuring signaling directly. If we can find explicit signaling effects we can measure, that’s a safe lower bound on the value of signaling.

Another alternative is to compare personal benefits, which include both signaling and human capital, to societal benefits, which should include only human capital (plus the benefits of signaling by the society, which is interesting to ponder; Bryan doesn’t get into that much). The difference is signaling.

Or you could look at exactly what causes increased earnings, and ask what stories that’s compatible with.

Additional dangers include questions such as, what if going to school’s ability bias is in the form of those who intend to work to increase their human capital? That’s tricky as hell to measure and sounds plausible. Or, if you want to measure societal benefit to schooling, what if the active ingredient is sorting the schooled for socialization purposes, in a would-have-happened-anyway, zero-sum (or worse) fashion?

It’s like any math or research problem – if you do valid things, to the extent they are valid, you’ll get a valid piece of the puzzle. Bryan takes every approach he can find and compares the implications, which is an excellent tactic here.

When he corrects for ability, what does Bryan find?

Correcting for ability, college grads earn 40% more than high school grads.

Citations needed and work must be checked and all that, and what ability means in context is worth thinking about carefully, but yeah. Wow.

Bryan begins with IQ as a source of human capital and signaling:

Almost all the research has two conclusions in common. First, IQ pays. Holding education constant, an extra point of IQ raises earnings by about 1%. Second, holding IQ constant, the education premium shrinks but never vanishes. In 1999, a comprehensive review of earlier studies found that correcting for IQ reduces the education premium by an average of 18%.

The mystery isn’t why IQ pays, but why it pays so little. I’ve known about these results for a long time, and understanding why this effect is so small seems important.

Mathematical ability might account for more than all of this, from a certain point of view, which is even more odd:

Correcting for mathematical ability may tilt the scales even more; the most prominent researchers to do so report a 40–50% decline in the education premium for men and a 30–40% decline for women.

There are a few stories that seem relevant and plausible. They are related, so there’s overlap.

Story 1: Most jobs are about being a cog in a machine, being in a given place at a given time to do a thing, or physically doing certain tasks. Being smarter might help a little, but mostly this is a threshold effect. If you’re good enough to be a retail clerk, you’re not going to produce much more value being a really smart retail clerk, and your being bored out of your mind is likely to backfire. The person fetching coffee needs to be smart enough to make and fetch coffee, but being smarter than that isn’t obviously good, if you’re assembling a car you need to be smart enough to know how, and so on for the butcher, the baker and the candlestick maker. There’s honor in all of it, but why should IQ pay off in most jobs?

This story implies those people also don’t need much education. They need to learn to submit, to learn to be cogs in a machine and do rote tasks. Of course, school is all about that.

Story 2: Smart people demand fulfilling, interesting work. Regular people just want steady work and a solid paycheck. Smart people get a much better basket of goods from their job, in the form of interesting things, but that lets employers pay them less money. 

Because economics, these smart people don’t produce that much more value for others, they mostly amuse themselves. But regular people were just as amused without all that work. So once again, investing in making people smarter slash teaching them more things isn’t worth all that much.

Story 3: Smart people have a hard time relating to (relatively) dumb people, and vice versa. You want your friends to be within one level of you [LW · GW], or at most two levels. At three levels, communication is all but impossible, an intelligence version of the SNAFU principle. A lot of many jobs is relating to others, so being too smart becomes a handicap even if you aren’t bored to tears.

Related, but not identical, to that is the story that no one likes a smart-ass, and people punish you for being smart, or judge you against a higher standard and thus pay you less. Even when they like that you’re smart, paying you for being smart feels unfair. They’re ‘working hard’ while you’re working smart. It’s natural to think that pay should follow who works hard rather than who produces more, even before inequality considerations or status considerations (stories 4 and 5).

This is especially true for those who create innovations and improvements and original works, the entrepreneurs and inventors and content creators. It’s notoriously hard for even corporations to capture that much of the value their new ideas create. Trying to capture the value as an individual inside an organization, or create a new one and keep that much of the equity in the one that you create, is even harder.

This story is bullish on making people smarter, teaching them how to think and be free, and on real education and skill development in general. Human capital is awesome! The few who generate worthwhile new things create tons of value. From some perspectives, they create more than all of the value, period. But they don’t capture it, so society benefits massively more than the pay statistics would suggest. We should subsidize such education, and such actions!

Alas, that doesn’t mean we should be subsidizing school. School doesn’t teach you to innovate and be free; it teaches you how not to think, and to be a cog. Those who create these types of value tend to be rebels who reject school. So it’s a strong argument against school, except insofar as it does actually raise intelligence as such as a side effect of its other actions.

Story 4: People hate inequality. The idea that the best programmer is worth 100 times the average programmer, who in turn is worth 10 times the average worker, is something we are highly motivated to reject out of hand, whether or not it is true. Besides, everyone has roughly the same needs. So the IQ 140 employee produces ten times what the IQ 110 employee produces, and gets paid 30% more.

This is definitely a huge thing. It’s related to the idea that you should be rewarding working hard or even working for X hours rather than producing value. But it’s also its own thing that says people aren’t allowed to have radically different value, people just won’t stand for it. You’re taxed on your extra value on tons of levels. People refuse to pay you for much of the extra value you capture, then they tax you, then they expect you to help the ‘less fortunate’ around you and around the world, and use any number of other techniques on you. Among them is, they forced you to consume more school, taking years off your life and saddling you with debt, and because you’re smart you need to do the same for your children. This can all easily add up to tax rates well in excess of 100%.

This story is bullish on building up intellectual human capital. Its value doesn’t show up fully in the pay statistics.

Story 5: Productivity doesn’t primarily determine pay. Primate status and social dominance, office politics and systematized oppression determine pay. If you’re not at the top, what matters is how you suck up to the boss. In extreme cases like government jobs and union shops, pure title and seniority entirely determine pay. So we’re asking the wrong questions (on many levels).

This is again bullish on human capital because its value isn’t paid for. The world gets the benefits mostly for free.

I see some truth in all of these stories. In any case, Bryan notes that intelligence enhancement can’t possibly be the bulk of what school does, because the effect size is too small:

Suppose each year of school permanently made you a whopping 3 IQ points smarter. According to standard estimates, this would raise your earnings by about 3%, leaving a supermajority of the education premium unexplained.

What of the theory  that one goes to school primarily to signal intelligence? His response is that there’s an easy and at least as effective way to test for intelligence alone: IQ tests are totally a thing.

Yes, there’s the minor catch that they’re illegal in theory:

Thanks to the landmark 1971 Griggs vs. Duke Power case, later codified in the 1991 Civil Rights Act, anyone who hires by IQ risks pricey lawsuits. Why? Because IQ tests have a “disparate impact” on black and Hispanic applicants. To escape liability, employers must prove IQ testing is a “business necessity.”

But, he argues, mostly legal in practice:

Only 4% of federal discrimination cases brought between 1987 and 2003 alleged disparate impact. That amounts to under a thousand annual cases against any form of employment testing. If disparate impact cases cost the usual amount, employers’ total test tax is under $200 million a year.

Which means that companies could just pay up when necessary:

Compared to the total upcharge a nation of employers pays for college graduates, this is a pittance. If IQ testing really let employers hire college-quality workers for high school wages, prudent employers would freely test IQ and treat the occasional lawsuit as a minor cost of doing business. Remember: correcting for ability, college grads earn 40% more than high school grads. If IQ laundering were a central function of higher education, courts could raise the test tax a hundredfold—and IQ testing would remain profitable.

Thus, employers must care mostly about other things, or they’d just pay up. In particular, they care about conscientiousness and conformity, both of which school signals strongly.

This bothered me enough that I spun off a post about this structure of argument, which is here. I’m also going to talk a bit about how I see the employers’ perspective and how I’d think about giving an IQ test.

If you Google “I don’t pay you to think” you get 147 million results. If you Google “I pay you to think” you get 51 million. The top of both searches is a TVTropes link to I Don’t Pay You To Think. So the amount of actual paying-for-thinking is not especially large. Chances are, I need you to not be a situational moron, and pass a threshold for intelligence, but don’t care about much beyond that. But lets say this is one of the exceptions, and I do want to pay you to think.

If I give an IQ test, it’ll be awkward. Applicants will see me test them and think it is weird, turning them off. Employees will see me doing it, and think it’s weird and uncaring and potentially legally dangerous. Everyone around could sue me, or more importantly could threaten to sue me, or perhaps even more importantly everyone knows that everyone is worried that anyone could implicitly threaten to sue me. So now I’m afraid of publicity, I’m afraid of anyone getting mad and wanting to go after me, I’m afraid of my own shadow. Any reporter who gets a whiff of this might shift their story to my terrible practices.

Even if none of that is a concern, I’ll have to be continuously reassuring employees, applicants, investors and everyone else that this isn’t an issue. If everyone thinks it’s an issue, it’s a huge issue, even if I never get sued and no one ever threatens to sue me. And all of them will treat this as a huge negative signal, that I take stupid legal risks and make myself vulnerable, and that I waste my weirdness points, and that I care about all the wrong things. What other similar things am I doing? Will I still be in business in a few years? Should anyone rely on me for anything? Will others think the same ways, even if they shouldn’t, or think others will think in those ways?

And of course, IQ tests have a bad reputation among Those Who Protest Things, which can itself be highly damaging.

So this is all super expensive, even if it amounts to nothing. 

And if I actually get sued, it’s a disaster that is likely to take down the whole business. Work will grind to a halt. Expensive lawyers will be brought in, stress levels will be sky high, no one will talk about anything else. Fishing expeditions take place in depositions. Employees may bolt. For the rest of time, people will remember that I was sued, that my business was sued, and I will be tainted. The actual payout cost of the suit is a small fraction of the damage it will due to me and my business. Lawsuits are nuclear weapons.

Perhaps I’ll just give this one a rest, even if I mostly do care about intelligence (and I almost certainly do care a lot about conscientiousness, and probably conformity, and definitely lots of other things, many of which I’m not supposed to care about). Plus, if I give people a numerical test, I won’t be able to use my judgment on who to hire, and no series of papers saying my judgment sucks is going to tell me I should stop using it. At most, I’ll rely on it somewhat less.

Alternatively, I can use a skills test that doubles as an IQ test. If you’re going to be a coder, I can listen to you think as you code. If you’re going to be a trader, I can give you trading situations or probability questions, and do the same. If challenged, I’m testing your applicable skills, so even if it is ruled an IQ test, it’s meeting a clear ‘business necessity.’ That’s not just a technicality, it’s actually doing that. So that seems better.

Using the IQ test combines all the disadvantages of a strict test. You’re opening yourself up to legal challenges on multiple levels, taking away your judgment. And you don’t get the most important benefit of strict testing, which is to appear to be doing a fair, sensible and responsible thing. You’re not doing something no one could blame you for. Quite the opposite. You’re actively courting mild social disapproval and awkwardness, rather than avoiding it. So why are you even doing it?

 

 

 

5 comments

Comments sorted by top scores.