Bryan Caplan  

Sacerdote Replies

PRINT
Michele Bachmann Gets It; Ron ... Ideas for teaching economics...

Bruce Sacerdote has graciously agreed to let me post this reply to my last post.  Here's Bruce:



Dear Bryan, thanks for the heads up!  I am perfectly 
happy with your description. I would make two key
points. First, my effects are indeed significantly
larger than the 0 effects found and trumpeted in the
BG literature. More importantly, most economists
including me believe that the entire effort to
engage in this sort of variance breakdown is
misguided. The very quote by Pinker gives one
reason why.

50% is genetics and 50% is some mysterious own
environment factor that we don't know what it is ?
And the mysterious environment factor is not shared
family environment. I mean come on. No parent
believes this stuff. These strong results stem
from the strong assumptions needed to parametrize
a BG model. But don't believe me...read Goldbergers
work from the 1970s. Many have shown how modest
tweaks to the model give you a different variance
breakdown.

The more scientific and accurate point that
economists make is that knowing the heritability of
something tells you nothing about whether a given
environmental intervention is worth doing. To take
Goldberger's famous example, knowing that eyesight
is 90% heritable tells us nothing about whether
provision of eyeglasses is a good idea. This second
point is largely accepted by economists and probably
by many or most behavioral geneticists.

My paper was really about simplifying the problem to
simply ask "what is the treatment effect from being
assigned to a particular type of family."

Thanks for the kind post and opportunity to respond.

Yours truly,
Bruce


Here's my reply:

50% is genetics and 50% is some mysterious own environment factor that we don't know what it is?  And the mysterious environment factor is not shared family environment.  I mean come on.  No parent believes this stuff. 
I'm a parent, and I basically do believe this stuff - especially for long-run effects.  So do most of the dads I know who are familiar with BG.  The moms are another story, of course...
The more scientific and accurate point that economists make is that knowing the heritability of something tells you nothing about whether a given environmental intervention is worth doing.  To take Goldberger's famous example, knowing that eyesight is 90% heritable tells us nothing about whether provision of eyeglasses is a good idea.   This second point is largely accepted by economists and probably by many or most behavioral geneticists.
This seems a little hasty to me.  A variance decomposition does tell you something useful.  Suppose your original cost-benefit analysis assumed that shared family environment explained 50% of the variance, so a 1 SD improvement in environment leads to a .7 SD improvement in outcome.  Then you learn that the variance explained by shared environment is only 10%, so a 1 SD improvement in environment leads only to a .3 SD improvement in outcome.  This means that a given environmental improvement is a lot less likely to pass a cost-benefit test.

When I was arguing about this with Bill Dickens, he kept saying that it's the regression coefficient, not the variance explained, that matters for policy.  My reply to him is that both are useless without price tags attached (either $/unit or $/SD), and both are equally useful with price tags attached.  Isn't that right?

Comments and Sharing





COMMENTS (8 to date)
Steven writes:

I have no position in the larger dispute, but the point you make here is incorrect.

A couple of extreme examples will show your error.

Suppose radical egalitarians succeed in equalizing everyone's environment, so that 100% of observed variations are due to the environment.

Now consider a group of 1 million clones, with 100% of the variation in outcomes being explained by environmental factors (either shared or not).

The effect of an intervention could be the same for the two groups (e.g., an improvement that increases longevity by 1 year for all people), despite the radically different variance decompositions.

The key point: the results of a variance decomposition for any outcome will depend greatly on the amount of environmental and genetic variation in the underlying population, even in cases where neither of theses sources of variation has any effect on the impact of an intervention. Bill Dickens is right -- what matters for interventions of this sort is the regression coefficients (and the prices in terms of $/unit).

It is true that for some interventions, the results will depend on the characteristics of the population. In these cases we would like to know the distribution of relevant population characteristics. But even in these cases, the variance decomposition is not what we're looking for. Rather we want the distribution of treatment effects, which entails finding the distribution of the treatment coefficient.

Ryan writes:

"Suppose radical egalitarians succeed in equalizing everyone's environment, so that 100% of observed variations are due to the environment."

I believe you meant to say that 100% of observed variations are due to genetics, correct?

Steven writes:

"I believe you meant to say that 100% of observed variations are due to genetics, correct?"

Correct -- I can't believe I made that mistake.

RL writes:

"Correct -- I can't believe I made that mistake."

Don't feel bad. It was probably your environment... :-)

t_c969 writes:

How many kids does Sacerdote have? I'm guessing one.

t_c969 writes:

From his homepage, looks like two.

Steve Roth writes:

Also:

>knowing that eyesight is 90% heritable tells us nothing about whether provision of eyeglasses is a good idea.

Wearing eyeglasses in childhood does not affect adult vision. This is a specious analogy.

Steve Sailer writes:

Keep in mind that many of these 50%-50% findings are for personality. It's not clear, however, what is a better personality. That seems quite situation-dependent.

Comments for this entry have been closed
Return to top