No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Fixed and random effects WITHOUT using clustered robust standard errors (CRSE)

    Hi Statalist users.

    I'm new to this forum and would love to get some help from you guys.

    I'm working with a panel dataset consisting of 36 countries and 828 observations.
    I'm trying to estimate the effect on corruption by using BCI (a corruption measure) on infant mortality. I'm doing this by both using a simple OLS and a fixed effects method.
    I have read that you should not use CRSE if you have fewer than 40clusters, which I have. When I use the -xtreg- command followed by -robust- my standard errors automatically gets clustered. To avoid this i specify my model the following way:

    reg infant mortality BCI controls i.year, robust.

    My first question is if this is an okey specifikation and if it is okey to use robust standard errors instead of clustered standard errors in the case of panel data?

    Secondly, I'm wondering if I can do the simple OLS with my panel data set without getting the CRSE? Can I somehow re-write the -xtreg- command into a -reg- command and still get my Random effect (OLS)?

    Thanks in advance!

    Kind Regards,

  • #2
    welcome to this forum.
    About your query:
    - the number of clusters that is considered ok for clustering is actually unknown; that said, 40 clusters seem enough. I assume that you cluster your standard errors for a sound methodological reason (that is, heteroskedasticity and/or autocorrelation of the idiosyncratic error);
    - as the options -robust- and -cluster- available under -xtreg- do the very same job, no wonder that the result does not change, as they both produce clustered robust standard errors;
    - unless you suspect that you do not actually have evidence of panel-wise effect (-xtreg- tests it for you with the F-test appearing as a footnote of the outcome table, but with default standard errors only), you should go -xtreg-.
    Eventually, I fail to get why you started your post with -fe- but ends it mentioning -re- specification.
    Kind regards,
    (Stata 16.0 SE)


    • #3
      Thank you for your fast reply!