Papke & Wooldridge (2008) coding for panel data methods for fractional response variables

[Kristin Bevilacqua]

Hello,

I have been working on a comparative interrupted time series analysis looking at the difference-in-differences of the proportion of incidents of IPV reported to police between white and Latina women before an after the Trump presidential campaign (2015). The outcome variable is the proportion of incidents reported by year by ethnicity (Latina and white). So, each incident was yes/no reported to police and then I calculated the weighted proportion of incidents reported by year by ethnicity.

On my last post, Dr. Wooldrige suggested that this was the same analysis in his 2008 paper and so I am trying to replicate that coding using the .do file provided on Dr. Papke's website.

The code runs but I get three notes after the glm code:
note: 1.TRUMP omitted because of collinearity.
note: 1.TRUMP#1.LATINAW omitted because of collinearity.
note: prop_report has noninteger values

Because of the colinearity, the code does not produce coefficients for TRUMP nor for TRUMP#LATINAW


I wanted to see if this means that the (complete) output for the xtgee model is not valid and, if so, if there is an error in the glm model code that needs to be address.

Below is the code (the keep code is just because the data set is so large, restricting to the variables I am using helps it run):

[xtset ID

keep V2117 V2118 WGTHHCY IDPER WGTPERCY SERIES_WEIGHT exclude_outUS dummy V3018 V3014 IPV YEAR LATINAW prop_report TRUMP V3040 EDU MARITAL PROXY ID YEARQ

glm prop_report TRUMP##LATINAW V3014 i.EDU i.MARITAL if [(exclude_outUS==0) & (dummy==0) & (PROXY == 0) & (V3018 ==2) & (IPV==1) & (inrange(LATINAW,0,1)) & (YEAR > 2014)] , fa(bin) link(probit) cluster(IDPER)
mat b = e(b)
xtgee prop_report TRUMP##LATINAW V3014 i.EDU i.MARITAL if [(exclude_outUS==0) & (dummy==0) & (PROXY == 0) & (V3018 ==2) & (IPV==1) & (inrange(LATINAW,0,1))] , fa(bi) link(probit) corr(exch) robust from(b,skip)]

Thanks in advance for any feedback/guidance!

Best,
Kristin

[Nefer Titi]

Kristin Bevilacqua I am desperately looking for the data set used in Papke and Wooldridge (2008). Dr. Papke's website at Michigan State University is down so I can't access it. Would you know where to find the data set and perhaps the stata do.file?

[Jeff Wooldridge]

Nefer: Our department recently transitioned to a new web page, and some of us are trying to catch up (including me). Here's a link to a Dropbox folder that should have what you're looking for. It also implements an alternative method that I published in 2019 with my former student Wei Lin.

PW (2008)

[Nefer Titi]

This is exactly what I was looking for; thank you very much!