I am working with panel data (flights per month of several airlines). I suppose that I need to cluster standard errors since OLS errors are not valid. Is there a way to test whether I indeed need to cluster errors and how to check across which dimension the clustering is required (e.g. airlines/type of aircraft/destinations etc)? Looking forward to any suggestions.
$\begingroup$
$\endgroup$
13
-
$\begingroup$ It looks like you’re after the Hausman test for fixed or random effects. en.wikipedia.org/wiki/… $\endgroup$– Guilherme MartheCommented Apr 30, 2021 at 16:26
-
$\begingroup$ @GuilhermeMarthe thanks for your reply! However, as far as I understand, the Hausman test could indicate whether the fixed effect model is preferred over the random effect model. But this does not directly relate to clustering? $\endgroup$– AnnaCommented Apr 30, 2021 at 16:30
-
$\begingroup$ It depends. What is the goal of your research? $\endgroup$– Thomas BilachCommented Apr 30, 2021 at 17:03
-
$\begingroup$ @ThomasBilach thanks for your response. I am studying flight delay propagation, in particular, which factors affect whether a delay is going to propagate. So as a dependent variable I have propagated delays for 10 airlines, and some airline/airport-specific characteristics as explanatory variables. To see which factors are actually significant I want to make sure that the standard errors are valid, thus I want to make sure that clustering is needed and to possibly check across which dimension. $\endgroup$– AnnaCommented Apr 30, 2021 at 17:23
-
$\begingroup$ I assume delays in the early part of the day have some downstream effects. Maybe some airlines don't include enough of a buffer period in between flights. Are you observing propagated delays over many time periods? Do you suspect any "within-airline" correlation in propagated delays over time? $\endgroup$– Thomas BilachCommented Apr 30, 2021 at 20:24
|
Show 8 more comments