Dear Statalists,
I am struggling with a R&R and I want to ask for some help / suggestions on how to deal with a methodological question, as I do not have experience with the topic.
In my project I used repeated cross-sectional data collected in a large number of countries (23) and waves (3) and I constructed a pseudo panel, that is, I constructed synthetic social groups based on gender, education and birth cohort and I followed them through time. My pseudo panel dataset had the structure: social groups followed in time and nested in countries. My main interest was to estimate the contextual effect as well as the effect of changes in time in a particular contextual characteristic on the DV. I used Stata 13 for estimating the pseudo panel models and xi: xtmixed command for hybrid models following Allison, P. D. (2009). Fixed effects regression models. Los Angeles: Sage.
One of the reviewers suggested that instead of constructing these synthetic groups in this way, I should use propensity score matching. The exact phrasing of the suggestion is:
“I recommend to use Propensity Score Matching instead of a pseudo panel design. With PSM also the problem of causality is taken into account, but you can stay at the individual level, and examining the impact of contextual and longitudinal effects. By first using PSM, and then using the weighted data in your multilevel model, you will have one integrated model…”.
If I understand correctly, the reviewer suggests to determine individuals that are “comparable” in each wave / country and then follow them through time, while applying a multilevel model in order to estimate the contextual effects of interest, correct? If so, is this a valid way to deal with repeated cross-sectional samples? I was not successful in finding any example of research using such a framework. Can such a model be estimated in STATA?
Best regards and thank you in advance for any suggestion!
Ioana
I am struggling with a R&R and I want to ask for some help / suggestions on how to deal with a methodological question, as I do not have experience with the topic.
In my project I used repeated cross-sectional data collected in a large number of countries (23) and waves (3) and I constructed a pseudo panel, that is, I constructed synthetic social groups based on gender, education and birth cohort and I followed them through time. My pseudo panel dataset had the structure: social groups followed in time and nested in countries. My main interest was to estimate the contextual effect as well as the effect of changes in time in a particular contextual characteristic on the DV. I used Stata 13 for estimating the pseudo panel models and xi: xtmixed command for hybrid models following Allison, P. D. (2009). Fixed effects regression models. Los Angeles: Sage.
One of the reviewers suggested that instead of constructing these synthetic groups in this way, I should use propensity score matching. The exact phrasing of the suggestion is:
“I recommend to use Propensity Score Matching instead of a pseudo panel design. With PSM also the problem of causality is taken into account, but you can stay at the individual level, and examining the impact of contextual and longitudinal effects. By first using PSM, and then using the weighted data in your multilevel model, you will have one integrated model…”.
If I understand correctly, the reviewer suggests to determine individuals that are “comparable” in each wave / country and then follow them through time, while applying a multilevel model in order to estimate the contextual effects of interest, correct? If so, is this a valid way to deal with repeated cross-sectional samples? I was not successful in finding any example of research using such a framework. Can such a model be estimated in STATA?
Best regards and thank you in advance for any suggestion!
Ioana
Comment