Hi,
I have a question with regards to the use of a difference-in difference model when I have multiple groups, 20 years and I can observe each group only for two years whereby the first observation is the pre-treatment time period, and the second observation is the post-treatment time period. More specifically: I use yearly repeated cross sectional survey data whereby each individual (=group) is observed in two consecutive surveys before being dropped out of the survey. I want to use this very short panel dimension to see how switching occupation (selective treatment) between period 0 and period 1 affects the change in working hours for switchers (treatment group) compared to non-switchers (control group). My groups are the individuals which appear in two consecutive surveys and can either select themselves into treatment (switching occupation) or not. Although individuals are only in the survey for two years, I want to estimate the average effect from 2003 to 2022 by using all surveys combined. So let us say for example individual 1, 2 and 3 enter in 2003 (pre-treatment) and drop out after 2004 (post-treatment). At the same time individuals 4,5 and 6 enter the survey in 2004 (pre-treatment) and leave the sample after 2005 (post-treatment) and so on and so forth till the last survey in 2022. I was wondering if I could use a simple difference-in-difference approach to look at the AET by setting the data to panel data with -xtset individual_id year and then use the following Stata command: xtdidregress (hours wagerate) (treatment_switch) [pw = final_linked_wtfinl], group(individual_id) time(year). I also control for changes in the wagerate between the two observation periods for each individual as this variable is time-varying and individual-specific. What are my other possibilities to consistently estimate the effect of switching occupation on working hours in the explained setting. Thanks!
I have a question with regards to the use of a difference-in difference model when I have multiple groups, 20 years and I can observe each group only for two years whereby the first observation is the pre-treatment time period, and the second observation is the post-treatment time period. More specifically: I use yearly repeated cross sectional survey data whereby each individual (=group) is observed in two consecutive surveys before being dropped out of the survey. I want to use this very short panel dimension to see how switching occupation (selective treatment) between period 0 and period 1 affects the change in working hours for switchers (treatment group) compared to non-switchers (control group). My groups are the individuals which appear in two consecutive surveys and can either select themselves into treatment (switching occupation) or not. Although individuals are only in the survey for two years, I want to estimate the average effect from 2003 to 2022 by using all surveys combined. So let us say for example individual 1, 2 and 3 enter in 2003 (pre-treatment) and drop out after 2004 (post-treatment). At the same time individuals 4,5 and 6 enter the survey in 2004 (pre-treatment) and leave the sample after 2005 (post-treatment) and so on and so forth till the last survey in 2022. I was wondering if I could use a simple difference-in-difference approach to look at the AET by setting the data to panel data with -xtset individual_id year and then use the following Stata command: xtdidregress (hours wagerate) (treatment_switch) [pw = final_linked_wtfinl], group(individual_id) time(year). I also control for changes in the wagerate between the two observation periods for each individual as this variable is time-varying and individual-specific. What are my other possibilities to consistently estimate the effect of switching occupation on working hours in the explained setting. Thanks!
Comment