Bryan Caplan  

Education's Selfish and Social Returns

The Case Against Education<... The Minimum Wage Harm that Few...
I've created a resource page for The Case Against Education's calculations of education's selfish and social returns.  This work is the tentatively final version of what I blogged last June.

Start with the slideshow.  If that piques your interest, the remaining resources allow you to carefully check my number crunching. 

I'm interested in all comments and criticism.  But my first priority is rooting out demonstrable errors in my formulas.  I will happily treat to lunch anyone who alerts me to one or more non-trivial demonstrable errors. 

Remember: The selfish and social value of correcting me now is far greater than it will be after publication!

P.S. I've also updated my c.v.

Update: Before uploading the files, I converted them to the latest version of Excel.  I just noticed that this latest version of Excel refuses to calculate negative internal rates of return.  I'll replace the current files with the prior versions on Monday.

Comments and Sharing

COMMENTS (4 to date)
Kevin Dick writes:

I didn't find any gross errors. And I generally agree with you. However, I have an analytic and pedagogical suggestion.

Your conservative to reasonable jump on the amount of signalling is too high. If those are your only levels, you open yourself to attack as assuming your conclusion when you choose 80% as the "reasonable" level. Better to have at least three levels.

Moreover, using discrete levels at all muddies the analysis on slides 14 and 15. What you really want to see are _graphs_ showing where the social returns vs the level of signalling, with a series for each ability level. Then you see where different levels of signalling force each ability level into negative return territory.

This approach naturally leads to a nice lengthy discussion of the crossover points for each ability level and how likely it is that reality is close to each point. You can also be extra clever and say, "So here's a cheap experiment we could run to exclude these possibilities because now we know what we're looking for."

BC writes:

Slide 2: "At the margin, signaling raises pay but not productivity"

Is this an assumption of the signaling model or an empirical finding? If signaling identifies ability, then can't it increase productivity by better matching workers to jobs?

Consider a basketball team and a pre-season training camp. Suppose that the camp does not increase any player's skills but only allows coaches to evaluate those skills (pure signaling). One might naively believe that the only gain from the camp is to allow players to get the playing time that they "deserve" once the season starts (private return), but overall net team gains (social returns) are zero because no player's skills were improved. Obviously, though, that would be incorrect since the team plays better when they have the right players in the right roles; signaling can improve productivity.

Does Caplan find empirically that productivity does not improve when educational degrees help employers identify workers' abilities (or at least that productivity gains are less than educational costs) or does he assume this?

Rick writes:

Agree with comment by BC above. Signaling has social value if it reduces search costs for employers. As noted, educational achievement does correlate with IQ and diligence. Your unstated assumption is that there must be a more efficient way to match people to jobs than to put them through years of school. But to be credible you will need to identify this alternate mechanism and compare the results.

Vitalik Buterin writes:

I think the weakest (or maybe least well-explained) part of the argument that stands out right now is: where do the 45% ability bias and the 3.4x and 6.7x sheepskin multipliers come from?

For the 45% multiplier, I see from one of the appendices, "In the NLSY, cognitive ability and socio-economic status have a .552 correlation"; that seems like the source of the 0.45 value. However, you can't just take a correllation and turn it into a multiplier like that. Particularly, note that a 0.55 correlation is, theoretically, perfectly compatible with 0% ability bias: suppose that your innate intelligence has zero impact on job performance (education only matters), but it has full impact on _your decision_ of whether you go to school: smart people do it, stupid people drop out because they have poor impulse control and are unwilling to put in the hard work of school. This extreme case is obviously false, but the effect may well exist to some extent, and it should be somehow statistically measured and accounted for.

For the 3.4x and 6.7x sheepskin multipliers, in the charts they seem to be given exogenously, and I can't find the source for them. The presence or absence of a sheepskin effect is an important cornerstone in the evidence here, so it should be justified much more front and center (and the justification for the 0.45 ability bias should also be more prominently explained).

Comments for this entry have been closed
Return to top