2
$\begingroup$

Say I am predicting happiness in a sample of college students and my predictors are age, family income, gender, and ethnicity. For my categorical variables (gender and ethnicity), I want to specify ahead of time the minimum number of respondents in a level (e.g., ethnicity = Asian) for me to report that coefficient. For example, if I have an overall sample of 300 respondents and only 2 are Asian, I don't think that dummy variable's coefficient would be reliable and I wouldn't want to report it. If 100 of my 300 respondents were Asian, I'd feel comfortable reporting that coefficient. But where is the cut-off?

Is this something that could be solved with a standard power analysis? For example, if I expect the effect of being Asian vs. my reference group is d = .2, would I need n = __ Asian students to report on that level?

Thank you!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
3
+50
$\begingroup$

No, there's no simple way. It depends on effects of your variable of interest both on the outcome as well as your other covariates (age, family income, etc.).

Improve this answer
$\endgroup$
4
  • $\begingroup$ Do you have any guidance on how one would conduct a "power analysis" for effects like this? $\endgroup$
    – ila
    Commented May 30, 2021 at 14:49
  • 1
    $\begingroup$ You can try a simulation approach where you model the data generating process. But, you'll have to make a number of assumptions on how the variables are related to each other. $\endgroup$
    – parasu
    Commented May 30, 2021 at 17:18
  • 1
    $\begingroup$ Take a look at this article: psycnet.apa.org/record/2000-16737-004 But that still requires the covariance matrix for all your variables $\endgroup$
    – parasu
    Commented May 30, 2021 at 17:26
  • $\begingroup$ Thank you for the article! And yes, I was thinking about a simulation approach. On the whole it seems like there's no simple way to approach this, as you say, but I appreciate the suggestions $\endgroup$
    – ila
    Commented May 30, 2021 at 20:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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