Hi,
I am trying to tease out causality between the X and Y or X and Z or X and Y*Z. I included the data below.
X - Dependent Variable - Continous variable
Y - Independent variable - Log transformed Count variable as it is highly right-skewed
Z - Independent Variable - continuous variable
Control variables - all continuous control variables - 9 variables
I am trying to analyze panel data in Stata 17.0 on Mac. The panel data have state and week-level observations and is strongly balanced. The data has 1462 observations with 40 missing values in the dependent variable. In the panel data N<T where N=34 and T=43.
It has the following issues:
1. small N and large T - too many time fixed effects in the model
2. Autocorrelation - AR1 is significant, but AR2 is insignificant when two lags of the dependent variables are used
3. Reverse causality between X and Y - where the lag of X has a positive influence on Y
4. Heteroskedasticity
I tried the following models. - Fixed effects (xtreg with time fe) , xtreg with lag of the dependent variable and time fixed effects, the dynamic panel data models - diff GMM and sys GMM (xtabond2 ). I have applied clustered robust standard errors or robust standard errors to control for the issues. I have the following questions.
1. Can we use the GMM models with such a small N? Are the results valid as the instruments are higher than my N?
2. What models should can I apply for analysis to ensure the robustness of my results apart from fixed effects?
3. I read in one of the Stata lists that we need to use FE with a lag of the dependent variable than the GMM models as the bias produced is much less than the GMM models. If so, how many lags should I use in my FE models?
Any suggestions regarding the models to apply for the analysis of my data will be very helpful for my analysis. Thank you in advance for all your help.
I am trying to tease out causality between the X and Y or X and Z or X and Y*Z. I included the data below.
X - Dependent Variable - Continous variable
Y - Independent variable - Log transformed Count variable as it is highly right-skewed
Z - Independent Variable - continuous variable
Control variables - all continuous control variables - 9 variables
I am trying to analyze panel data in Stata 17.0 on Mac. The panel data have state and week-level observations and is strongly balanced. The data has 1462 observations with 40 missing values in the dependent variable. In the panel data N<T where N=34 and T=43.
It has the following issues:
1. small N and large T - too many time fixed effects in the model
2. Autocorrelation - AR1 is significant, but AR2 is insignificant when two lags of the dependent variables are used
3. Reverse causality between X and Y - where the lag of X has a positive influence on Y
4. Heteroskedasticity
I tried the following models. - Fixed effects (xtreg with time fe) , xtreg with lag of the dependent variable and time fixed effects, the dynamic panel data models - diff GMM and sys GMM (xtabond2 ). I have applied clustered robust standard errors or robust standard errors to control for the issues. I have the following questions.
1. Can we use the GMM models with such a small N? Are the results valid as the instruments are higher than my N?
2. What models should can I apply for analysis to ensure the robustness of my results apart from fixed effects?
3. I read in one of the Stata lists that we need to use FE with a lag of the dependent variable than the GMM models as the bias produced is much less than the GMM models. If so, how many lags should I use in my FE models?
Any suggestions regarding the models to apply for the analysis of my data will be very helpful for my analysis. Thank you in advance for all your help.
Code:
* Example generated by -dataex-. For more info, type help dataex clear input float(s t x y yz) 1 1 0 3.178054 2.0726438 1 2 0 5.560682 .7299737 1 3 0 5.983936 -.5439942 1 4 .10846561 5.78996 .3019305 1 5 .10962963 6.507277 -.3404405 1 6 .05624143 5.32301 -.6784229 1 7 .01112347 5.834811 -1.865086 1 8 .007829977 4.795791 .8392634 1 9 .010362694 4.969813 1.772451 1 10 .008665511 4.962845 1.0484884 1 11 .016587678 6.654152 -1.8288187 1 12 .005568814 4.990433 .4443536 1 13 .010438413 5.068904 -.6416335 1 14 .003556188 3.73767 .54697603 1 15 .008531383 4.543295 .4885263 1 16 .002701486 3.6888795 .1891733 1 17 .00713128 5.129899 1.43515 1 18 .0094513 5.164786 1.0092111 1 19 .00599952 5.081404 .19055265 1 20 .01040724 4.7361984 1.8860967 end format %tw t label var s "group(statename)" label var t "group(week_of_year)" label var x "dependent variable" label var y "continous Independent variable"
Comment