Dec 16, 2008

Circular causality

It amazes me how much confusion there is in economics papers and presentations about the different types of endogeneity that exist and the biases they create. I blame this on bad econometrics classes that focus on technicalities rather than intuition.

I guess in most cross country studies the problem most of the time is one of omitted variable bias, or simultaneity, as country specific unmeasurable factors explain the relationship between your two variables of interest. For example, corruption and female penetration in government are corrlated even though none cause another. How should the bias be in this form of endogeneity? What bias?! We're talking about a causality that doesn't exist!

Now consider the reverse causality of self-enforcing mechanisms, such as between institutions and growth. Most economists talk about over-estimated coefficients in this case!? just becuz mechanisms are self-enforcing coefficients have an upward bias?!? Nope. Let's say institutions cause growth by a factor of 3 whereas growth causes institutions by a factor of 2. Rgeressing growth on institutions with OLS you estimate a factor of 2.5 (which is the correlation, not the causality)...this is a downward bias...

2 comments:

Pascal said...

Is this rigotnometrics ? My textbook asymptotics tell me there is an upward bias...available upon request!

Pascal said...

oooops, PL convinced me that the sign is ambiguous...coffee on me!! viva rigotnometrics!