Hi everyone,
I am estimating the effect of time-varying and time-invariant variables on an outcome using correlated random effect (CRE) model using STATA 19, but the problem I am facing right now is the inability to account for time fixed-effect (FE) and population differences. Since the model decomposes the variables into within-units and between-units components (see https://www.stata.com/new-in-stata/c...effects-model/), including time FE will lead to collinearity and hence, the coefficients for time will be omitted because time is static across the panels (beat_id). I could have considered the traditional FE model, but my covariates are time-invariant (they do not really change over time - 12 months). Relatedly, I could not use random effect, but it assumes that the individual effects (the random component) are uncorrelated with the explanatory variables, and this assumption is hard to be met given the nature of my analysis. So, my question is how can I treat "i.time" in a way that CRE won't decompose it into within-units and between-units component? For population exposure, it is possible to do this using "exposure(population)" in traditional FE, but how can I do the same in CRE model? CRE does not allow for exposure.
Note: I use this code "xtreg, DV IVs, cre vce(cluster, beat_id)" for estimation and it can be done in STATA 19 only. I do not mind using a different approach to achieve similar results. Thank you for your time!
I am estimating the effect of time-varying and time-invariant variables on an outcome using correlated random effect (CRE) model using STATA 19, but the problem I am facing right now is the inability to account for time fixed-effect (FE) and population differences. Since the model decomposes the variables into within-units and between-units components (see https://www.stata.com/new-in-stata/c...effects-model/), including time FE will lead to collinearity and hence, the coefficients for time will be omitted because time is static across the panels (beat_id). I could have considered the traditional FE model, but my covariates are time-invariant (they do not really change over time - 12 months). Relatedly, I could not use random effect, but it assumes that the individual effects (the random component) are uncorrelated with the explanatory variables, and this assumption is hard to be met given the nature of my analysis. So, my question is how can I treat "i.time" in a way that CRE won't decompose it into within-units and between-units component? For population exposure, it is possible to do this using "exposure(population)" in traditional FE, but how can I do the same in CRE model? CRE does not allow for exposure.
Note: I use this code "xtreg, DV IVs, cre vce(cluster, beat_id)" for estimation and it can be done in STATA 19 only. I do not mind using a different approach to achieve similar results. Thank you for your time!
Comment