Wednesday, July 06, 2011

The GSS samples a disproportionate number of females

Folks who use General Social Survey (GSS) data should know that, except for the first year of the survey--1972--the sample has always been lopsided toward women. GSS workers knock on doors during the daytime hours and are more likely to find women to interview.

The percent of the sample that is female is typically in the mid-to-high 50s; men are typically in the low-to-mid forties. NORC has evidently made no effort in almost 40 years to get a balanced sample.

Analysts should take this into account, especially when they generate national estimates that differ significantly by gender. For example, if you are estimating the percent of adult Americans who favor the death penalty, your figure will be low since men disproportionately support it but are underrepresented in the sample.  On many political attitudes, the country is a bit more conservative than the GSS indicates.


bgc said...
This comment has been removed by the author.
bgc said...

It's an important point. In practice, precisely representative random samples are impossible without compulsion.

A full enumeration - i.e. a census - is often simpler - but even then there will be omissions.

Which only emphasizes that social science research should be based on causal theories, not pure 'inductivism'!

I try to explain this here:

ziel said...

I was struck by the very low levels of self-described homosexuality in the GSS - in the 2% range from the latest survey. Could this be an artifact of this female oversampling?

Timothy said...

Here, I do not actually think this is likely to have success.