Hello dear forum members,
The main objective of my research is to determine which psychological factors (45 in total) significantly affect cancer-related outcome. The outcome is a rate ranging between 21 and 604.9. In order to be able to understand whether the determinants of the outcome change at different levels of the dependent variable (and also at the extreme values), I am using quantile regression.
My data consists of almost 3000 observations (at the county-level) collected in 2013 and 2014. Although I am aware of recent developments in quantile regression with fixed effects (e.g., -qregpd- or -xtqreg-), I agree with Joao Santos Silva's opinion that "there is very little information on the performance of these estimators, and I would say that at the moment there is no established method to address this problem" (https://www.statalist.org/forums/for...and-panel-data). Therefore, I turn to the "standard" estimation approach using:
Next, I use -test- to test for the differences in the estimated coefficients of the significant regressors.
Recently, however, I came across the following paper de Luca, F., & Boccuzzo, G. (2014). What do healthcare workers know about sudden infant death syndrome?: the results of the Italian campaign ‘GenitoriPiù’. Journal of the Royal Statistical Society: Series A (Statistics in Society), 177(1), 63-82. Interestingly, rather than using an index with bounds (outcome) that may vary depending on the data, they consider an index varying between 0 an 1 for ease of interpretation of its values and of the results. As such, they use quantile regression for bounded outcomes as implemented in -lqreg-. Assuming this approach might also be appropriate for my case with rate, I also tried it using:
Overall, I observe relatively high consistency of the estimates obtained using -sqreg- and -lqreg- (with minor differences).
My questions: (1) Are both of these approaches appropriate for modeling rate? (2) Does the -lqreg- approach have any advantages over -sqreg- in my case? (3) What is the correct interpretation of the estimates in case of -lqreg-?
Thankfully,
Anton
The main objective of my research is to determine which psychological factors (45 in total) significantly affect cancer-related outcome. The outcome is a rate ranging between 21 and 604.9. In order to be able to understand whether the determinants of the outcome change at different levels of the dependent variable (and also at the extreme values), I am using quantile regression.
My data consists of almost 3000 observations (at the county-level) collected in 2013 and 2014. Although I am aware of recent developments in quantile regression with fixed effects (e.g., -qregpd- or -xtqreg-), I agree with Joao Santos Silva's opinion that "there is very little information on the performance of these estimators, and I would say that at the moment there is no established method to address this problem" (https://www.statalist.org/forums/for...and-panel-data). Therefore, I turn to the "standard" estimation approach using:
Code:
sqreg y x1-x45, reps(500) q(.01 .05 .10 .25 .50 .75 .90 .95 .99)
Recently, however, I came across the following paper de Luca, F., & Boccuzzo, G. (2014). What do healthcare workers know about sudden infant death syndrome?: the results of the Italian campaign ‘GenitoriPiù’. Journal of the Royal Statistical Society: Series A (Statistics in Society), 177(1), 63-82. Interestingly, rather than using an index with bounds (outcome) that may vary depending on the data, they consider an index varying between 0 an 1 for ease of interpretation of its values and of the results. As such, they use quantile regression for bounded outcomes as implemented in -lqreg-. Assuming this approach might also be appropriate for my case with rate, I also tried it using:
Code:
generate y_bound = (y-21)/(604.9-21) lqreg y_bound x1-x45, quantiles(01 05 10 25 50 75 90 95 99) seed(123) cluster(id)
My questions: (1) Are both of these approaches appropriate for modeling rate? (2) Does the -lqreg- approach have any advantages over -sqreg- in my case? (3) What is the correct interpretation of the estimates in case of -lqreg-?
Thankfully,
Anton
Comment