So, I have categorical variables for "most dominant religions in the world" (Islam, Protestant, Catholic, Orthodox and Others). So, when I am writing the regression model, how should I write it?

% of women in parliament= B0+ B1 (Most Dominant Religion)+ B2 (Education)+e ?

Or % of women in parliament= B0+ B1 (Islam)+ B2 (Protestant)+ B3 (Catholic)+ B4 (Orthodox)+ B5 (Other religions)+ B6 (% of population with secondary education)+e So precisely, my question is: in case of categorical variables (race/ religion); do we consider each race/ religion as a single variable? And I should run the bivariate between % of women in parliament and Islam?

  • 1
    $\begingroup$ I get the idea of your 2nd specification. For each religion except one, you' create a dummy variable. You should read up on using dummy variables, but you must exclude ONE category from the regression for it to work (and all the categories are relative to the excluded category). The 2nd specification essentially treats each religion as a separate fixed effect and estimates relationship between women in parliament and education using edu variation within each religion. On the 1st spec, I don't know what "most dominant religion" is? $\endgroup$ – Matthew Gunn Dec 9 '17 at 0:56
  • $\begingroup$ Dummy variables won't work here because they are percentages. It's not individual data. $\endgroup$ – Peter Flom Dec 9 '17 at 14:24
  • $\begingroup$ Can you tell us what a typical set of values would be for one country for each of you two models? At the moment it is not clear to us what they are. $\endgroup$ – mdewey Dec 9 '17 at 15:15

Which model you should run depends on what you want to find out. You need to think about how you want to conceptualize the religious make-up of a country.

One way would be to use the percentages as compositional data - that is, data where the total is constrained to be 100%. This method has been discussed here, see this tag.

Another would be to come up with some index (or possibly more than one) that captures something about countries that you think will be related to the % of women in parliament. For instance, maybe you think it is the % of people who profess no religion. Or maybe you think it is how dominant a particular religion is. Or something else.

Finally, your dependent variable is also bounded and you need to account for that; one method would be beta regression.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.