Dear all,
I am new with panel data, and I need your help.
The reviewer suggested me the methodology applied in paper Fabrizi and Mussida (2020). The methodology applied is correlated random effects with endogenous initial conditions. I used command xtprobit in order to estimate model. xtprobit command has two options, random effects and population averaged. There is no FE with xtprobit model. The authors wrote the following
Random effects have more desirable properties than fixed effects if the assumptions underlying them are satisfied. For this reason we applied Hausman specification test Hausman (1978) and Woutersen and Hausman (2018) which confirmed that the random effect model is not biased and that this is the correct estimation procedure in comparison to the fixed effect model.
How can I pre-test if I should use FE or RE? For Hausman test I need to estimate model with FE and RE and then to compare them. But it is not possible to estimate xtprobit FE. I can estimate xtlogit with FE and RE, and then performa Hausman test. Additionally, I don't know how to perform Woutersen and Hausman (2018) specification test.
I estimate probability that the child is poor, with lagged dependent variable (poverty in previous state), with initial conditions (poverty in first period) and parents labour marekt status in initial period, and few explanatory variables. Due to several variables being time-inveriant (initial condition variables) and also several variables are with low variability (e.g. household type, education level of parents), is it possible that FE model could be appropriate. Please find attached the log file with estimation results of xtprobit.
To resume, should I care about some testing procedures for comparing FE and RE, or should I only proceed with the suggested methodology?
Best regards,
Aleksandra
I am new with panel data, and I need your help.
The reviewer suggested me the methodology applied in paper Fabrizi and Mussida (2020). The methodology applied is correlated random effects with endogenous initial conditions. I used command xtprobit in order to estimate model. xtprobit command has two options, random effects and population averaged. There is no FE with xtprobit model. The authors wrote the following
Random effects have more desirable properties than fixed effects if the assumptions underlying them are satisfied. For this reason we applied Hausman specification test Hausman (1978) and Woutersen and Hausman (2018) which confirmed that the random effect model is not biased and that this is the correct estimation procedure in comparison to the fixed effect model.
How can I pre-test if I should use FE or RE? For Hausman test I need to estimate model with FE and RE and then to compare them. But it is not possible to estimate xtprobit FE. I can estimate xtlogit with FE and RE, and then performa Hausman test. Additionally, I don't know how to perform Woutersen and Hausman (2018) specification test.
I estimate probability that the child is poor, with lagged dependent variable (poverty in previous state), with initial conditions (poverty in first period) and parents labour marekt status in initial period, and few explanatory variables. Due to several variables being time-inveriant (initial condition variables) and also several variables are with low variability (e.g. household type, education level of parents), is it possible that FE model could be appropriate. Please find attached the log file with estimation results of xtprobit.
To resume, should I care about some testing procedures for comparing FE and RE, or should I only proceed with the suggested methodology?
Best regards,
Aleksandra
