Saturday, March 8, 2008

Ross on Oil, Islam, and Women's Rights

Michael Ross has a provocative article in the current issue of the APSR. He argues that women's rights in the Middle East have been stymied more by oil than by Islamic cultural norms (gated link here). Here's the abstract:
Women have made less progress toward gender equality in the Middle East than in any other region. Many observers claim this is due to the region's Islamic traditions. I suggest that oil, not Islam, is at fault; and that oil production also explains why women lag behind in many other countries. Oil production reduces the number of women in the labor force, which in turn reduces their political influence. As a result, oil-producing states are left with atypically strong patriarchal norms, laws, and political institutions. I support this argument with global data on oil production, female work patterns, and female political representation, and by comparing oil-rich Algeria to oil-poor Morocco and Tunisia. This argument has implications for the study of the Middle East, Islamic culture, and the resource curse.

The evidence is pretty compelling, although I do have some critiques. The regressions in Tables 1 and 2 include income and working age, which are endogenous to the other variables in the model, in which case the coefficients suffer from a type of posttreatment bias. The models reported in Table 4 that include the political system variables have the same problem. It doesn't seem like the results in Table 4 change much as a result, so it might not be a problem. We can't see the consequences in Tables 1 and 2 because all models include the endogenous variables. The concern is that Islam => lower income and lower income => less female labor participation; if so, purging out the effect of income prevents us from seeing all of the possible effect of Islam on depressing labor participation. If some kind of control for income is desired, it would have to be some measure of income that includes variation due to things that are exogenous to Islam and oil. One might also argue that income, working age, and polity are endogenous to the outcomes of female labor force participation and female parliamentary representation. If that is true, then the bias on coefficients on those variables may be substantial, and that bias may propagate bias in the coefficients of interest (on Islam and Oil). (At least I think that may be that case...need to think through that a bit more, perhaps in terms of partial regression...) It's frustrating to see that these problems still arise in top-notch poli sci journals, especially when running the "right" regressions is so easy.

All that being said, though, Figures 3-6 are pretty startling in the patterns that they reveal. So there really does seem to be something to this. A few simple regressions that avoid posttreatment bias could help make the case that much stronger.


Anonymous said...

Hello Cyrus -- What book would you recommend to the first-year student who does not know how to comment intelligently about these stat/regression/variable-oriented issues? I see you link to Andrew Gelman's blog, his book perhaps? Thanks for your answer.

kerimcan said...

Why are you surprised by endogeneity problems? It is easy to run the right regressions only if you have the instruments (I suppose you have 2SLS regressions in mind), but finding good instruments are tricky.

Cyrus said...

Books that I would recommend for the first-year student:

Gelman and Hill, Data Analysis Using Multilevel/Hierarchical Models.

Goldberger, A Course in Econometrics.

Manski, Identification Problems in Social Sciences.

I also just picked up Morgan and Winship, Counterfactuals and Causal Inference, which looks really promising.

Fr. said...