Dear statalist,
I hope this finds you well.
I am working on a small-n large-T (n=7, weekly dataset from 2015-2020) data, and I have been using fe to estimate two endogenous variables's inverted u-shaped effect (x1 and x3; x2 and x4 are their squared terms)
I have read the recommendation by one post thus I chose xtscc command to conduct the main analysis. However, while trying to conduct an 2SLS analysis with instrumental variables, I noticed that if I just use xtivreg2, the first stages would not produce DK standard error, and only robust SE.
So I tried to manually do 2sls step-by-step, I realized that it is hard to get key statistics such as Hansen J statistic or Kleibergen-Paap rk LM statistic(Underidentification) unlike the xtivreg2,
Below are my commands, here are my questions:
1) Is this a legitimate approach to conduct IV analysis?
2) Is there any possible way to attain the above statistics with the xtscc?
3) This is only the main effect; in my main analysis, for the mediation analysis, I also estimated the quadratic mediation analysis using (Hayes & Preacher 2010) SPSS macro medcurve besides Baron and Kenny's approach. Do I need to also address the endogeneity issue for the quadratic mediation analysis? It feels like an uncharted area, I have not found out any established approach can do that. Do you think I need to do anything about this before submitting my paper?
Any insight or suggestions for any of the above questions would be greatly appreciated.
Best regards,
A struggling junior researcher
I hope this finds you well.
I am working on a small-n large-T (n=7, weekly dataset from 2015-2020) data, and I have been using fe to estimate two endogenous variables's inverted u-shaped effect (x1 and x3; x2 and x4 are their squared terms)
I have read the recommendation by one post thus I chose xtscc command to conduct the main analysis. However, while trying to conduct an 2SLS analysis with instrumental variables, I noticed that if I just use xtivreg2, the first stages would not produce DK standard error, and only robust SE.
So I tried to manually do 2sls step-by-step, I realized that it is hard to get key statistics such as Hansen J statistic or Kleibergen-Paap rk LM statistic(Underidentification) unlike the xtivreg2,
Below are my commands, here are my questions:
1) Is this a legitimate approach to conduct IV analysis?
2) Is there any possible way to attain the above statistics with the xtscc?
3) This is only the main effect; in my main analysis, for the mediation analysis, I also estimated the quadratic mediation analysis using (Hayes & Preacher 2010) SPSS macro medcurve besides Baron and Kenny's approach. Do I need to also address the endogeneity issue for the quadratic mediation analysis? It feels like an uncharted area, I have not found out any established approach can do that. Do you think I need to do anything about this before submitting my paper?
Any insight or suggestions for any of the above questions would be greatly appreciated.
Code:
//step by step approaches – using DK SE for both steps //First stages xtscc x1 IV1 control_list month, fe predict x1_hat,xb xtscc x2 IV1_sq control_list month, fe predict x2_hat xtscc x3 IV2 IV3 control_list month, fe predict x3_hat xtscc x4 IV2_sq IV3_sq control_list month, fe predict x4_hat //second stages xtscc log_MC c.x1_hat##c.x1_hat c.x3_hat##c.x3_hat control_list month, fe
Best regards,
A struggling junior researcher
