0
$\begingroup$

I suspect reverse causality between Y and X1, therefore I'm applying an instrumental variable approach to panel data. My main regression consists of country- and year-fixed effects. Do I have to also apply country- and year-fixed effects to my first stage regression where I regress the endogenous variable (X1) on the instrument (Z) and the other supposedly exogeneous control variables (W) or would it be OK to specify the model differently (e.g. no fixed effects at all)?

$\endgroup$
3
  • 1
    $\begingroup$ Yes, your first stage should include all the exogenous components of your second stage, including all "fixed effects". Are you implementing this by hand? $\endgroup$
    – Durden
    2 days ago
  • $\begingroup$ @Durden Thank you for your quick reply. No, I'm implementing it using the plm-package in r (something like this: fixed1_nc_imp_rb_1 <- plm( log(y) ~ log(x1) + x2 + x3 | log(z) + x2 + x3, model = "within", data = df, index = c("country","year"), effect = "twoways", vcovHC = "HC3") $\endgroup$
    – TFT
    2 days ago
  • 1
    $\begingroup$ I was asking because a well-developed software implementation would keep track of the fixed effects in either of your equations. I presume plm does. $\endgroup$
    – Durden
    2 days ago

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.