Identifying dropped observations in xtregar panel model with -fe

[Uriel Stettner]

Dear all,

I am fitting a fixed-effects model using xtregar in Stata 13.0 on Win 10.

xtregar performance patent_experience l.marketsize_sic l.rndintensity l.revt, fe.

The performance variable and all ivs/controls are continuous. The full, unbalanced panel contains 3545 obs, but due to missing values, e(sample) returns 1630 obs. However, the command output specifies only1474 obs.

FE (within) regression with AR(1) disturbances		Number of obs	=	1474
Group variable: gvkey		Number of groups	=	153
R-sq: within = 0.2492		Obs per group: min	=	1
between = 0.0113		avg	=	9.6
overall = 0.0239		max	=	16
		F(12,1309)	=	36.21
corr(u_i, Xb) = -0.8013		Prob > F	=	0.0000

Question:
1. How can I identify the 156 obs that were listwise deleted (so that I may, for example, produce a correlation table on the proper sample)?
2. Does the fe-model drop the 156 observation because the dependent variable lacks within-group variance? Are there other explanations?


Thank you for your help!

[Carlo Lazzaro]

Simon (full given and family names, although not mandatory, are welcomed on this forum that, in turns, welcome you. Thanks).
Under -fe- specification, -xtregar- removes the first observation for each panel (see -xteregar- entry in Stata .pdf manual; page 468); this is, in all likelihood, the explanation for what you experienced, because Stata applies listwise deletion to observations with missing value(s) in any variable (and so they're excluded from subsequent analyses).

[Uriel Stettner]

Thank you Carlo!
Indeed, when I drop the first observation of each panel I am left with 1474 - as reported by xtregar output.

[Carlo Lazzaro]

Simon:
thanks for sharing the result of your check.