Quantile regression with fixed effects

[Hale Isaac]

I would like to investigate the distributional effects of a set of independent variables (e.g. food insecurity) on my outcome variable (low, moderate and high mental health). For example, are individuals who report low initial mental health more/less affected by food insecurity. I have a panel data set with two waves and 2,000 observations.

I am aware of Kit Baum’s xtqreg (SSC) which estimates quantile regression with fixed effects, but I am not sure if I can run this analysis with only two waves of data?

[Alex Caridyas]

Hi Hale,

Have you checked out @Matthew J. Baker , Travis Smith & David Powell's qregpd package on SSC? It implements the quantile estimator for panel data detailed here - says its consistent for small T, although I have not used this package before.

[Joao Santos Silva]

Dear Hale Isaac,

Unfortunately, our estimator implemented in xtqreg does not work with T=2; it requires large T. Powell's estimator suggested by Alex in #2 is valid for small T, but it is not really a fixed effects estimator. With T=2, your best bet is to use correlated random effects.

Best wishes,

Joao

[Hale Isaac]

Hi Alex and Joao, thank you both for your suggestions. After having read the Powell(2016) paper I think I will be using their package. It does eliminate fixed effects, but seems to have various other advantages ("estimates of the coefficients are robust even if density of exposure is sparse at either end of the mental health distribution").