Oaxaca Module: Normalizing to Solve Omitted Dummy Variable Problem

[Brad Foster]

Dear Statalist:

I'm using the Oaxaca module to decompose changes in migration propensities (from the IPUMS version of the Current Population Survey) in the U.S. over time as a function of individual characteristics. For example, I might construct the following model:

oaxaca migrate female nhwht nhblk hispanic own married hhchild unemp collegeplus age_18to24 age_25to34 age_35to44 age_45to54 age_55to64 age_65to74 [pw = wtsupp], by(time) cluster(serial) swap noisily detail,

where "migrate" is a dichotomous variable indicating migration (or not), female through age_65to74 are dummy variables with reference groups omitted, wtsupp is a probability weight, time is two different years, and serial is a household grouping indicator.

Ideally, I would use the normalize() command within the list of dummy variables to normalize coefficients for any particular set of dummies around a grand mean, rather than referencing the omitted category. However, if I run the following simplified model:

oaxaca migrate normalize(male female), by(time),

I get the following error message:

"factor variables and time-series operators not allowed."

To be sure, this is not a data issue - male and female uniquely identify all of my cases.

Is there any obvious reason why the normalize() command isn't working? Any help is much appreciated!

Thanks,

- brad

[Brad Foster]

Not sure what the etiquette is for answering your own question, but I finally stumbled upon my error and found a solution.

normalize() and/or categorical() is an option in the Oaxaca module and, therefore, goes after the comma in the command. Dummy variables are to be listed in the basic command. For example, to see the effect of gender at two the two chosen points in time:

oaxaca migrate female, categorical(male female).

For multiple categorical predictors, separate lists in the categorical() option by a comma:

oaxaca migrate female married, categorical(male female, married single).

[Nick Cox]

It's fine to answer your own question.