significant linear trend in proportion of males: some of same people found in different years

[mjkop56]

I am trying to determine if there is a significant linear trend over time in the proportion of males. This post is very helpful, and the suggestion to use a logistic regression with gender as the outcome and year as the predictor:

https://communities.sas.com/t5/Statistical-Procedures/Test-for-significant-difference-in-proportion-...

I have some of the same people who are found in different years (but not in the same year). Can I use the same set up, or does the logit model need to be modified?

[sbxkoenk]

Hello,

See the answers of @PGStats and @SteveDenham in this post : 

Testing Trend in Proportion Over Time
Posted 06-25-2020 04:42 PM
https://communities.sas.com/t5/Statistical-Procedures/Testing-Trend-in-Proportion-Over-Time/td-p/665...

Test your errors for autocorrelation if you use a regression modelling approach!!

Koen

[mjkop56]

Thanks, I'm trying to figure our how to test for autocorrelation with a logistic regression.

I have individual level data, for example, and the outcome (gender) does not change within a person.

ID  year  gender

1    2005    M

1    2006    M

1    2007    M

2    2005    F

3    2006    M

3    2007    M

3    2008    M

Could a model with cluster robust standard errors fix the issue (with same person in multiple yrs), or does one need to go to a random intercept or random coefficient model?

[sbxkoenk · Accepted solution]

Hello,

You could consider fitting a GEE model in PROC GENMOD and using TYPE=AR or MDEP(1) to examine the estimated correlation between adjacent observations.

GEEs = Generalized Estimating Equations 

GEEs should cope with having data that are correlated in clusters.

Generalized Estimating Equations

https://support.sas.com/rnd/app/stat/topics/gee/gee.pdf

Cheers,

Koen

[mjkop56]

I will try that. Many thanks, Koen!