{smcl} {* 28feb2008}{...} {hline} help for {hi:fese} {hline} {title:Standard errors for fixed effects} {title:Syntax} {p 6 16 2} {cmd:fese} {vars} [{cmd:if} {it:exp}] [{cmd:in} {it:range}] [{cmd:,} {cmd:s(}{it:string}{cmd:)} {cmdab:o:only} {cmdab:a:bsorb(}{it:varname}{cmd:)} {it:{help areg:other areg_options}} ] {pstd} {marker rd_description}{title:Description} {p}{cmd:fese} implements a fixed-effects regression using {help areg} and saves the estimated fixed effects and their standard errors as new variables on the data. Note that {help areg} produces identical results to {help xtreg} with the {cmd:fe} option except when the {help vce_option:cluster-robust VCE} is requested, in which case {help areg} uses a different degrees of freedom adjustment, which is more appropriate for the present case. {help areg} is always equivalent to running {help regress} with an exhaustive set of indicator variables for panels. {p}These standard errors are not usually computed in a fixed-effects regression, but we may need them. One example takes student test scores as the dependent variable and teacher assignments as the explanatory variables, as in Rothstein (2007), where the fixed effects measure the assumed additive effect of a teacher on her students' test scores. The variance of estimated fixed effects captures both the variance of true fixed effects and the variance of the estimator: the variance of true fixed effects (i.e. how disparate are teachers' apparent impacts on students' scores) can be estimated as the observed variance in estimated fixed effects less the best estimate for the variance of the estimator, which is the mean of squared standard errors. {p}The estimated fixed effects, and OIM and heteroskedasticity- and cluster-robust standard errors (See {help vce_option} and {hi:[U] 20.15 Obtaining robust variance estimates}) are saved by default, as new variables which vary only across panels. See Nichols and Schaffer (2007) for more on cluster-robust variance estimators. Note, however, that the asymptotic justification for a cluster-robust variance estimator requires that the number of clusters m approaches infinity, which implies we cannot get consistent estimates of the fixed effects or the variance of the estimates, though if the number of observations per panel also increases without bound, the properties of the estimator may be acceptable. In the example above, it is implausible that the number of students per teacher is increasing without bound, but with strongly balanced clusters of 25 or so students per class, the behavior of the estimator may rest on how close 25 is to infinity for our purposes. {marker options}{title:Options} {p 0 4}{cmd:s(}{it:string}{cmd:)} specifies the stub to prefix to new variable names. {p 0 4}{cmdab:o:only} requests that only OIM SEs be calculated (speeding up calculations). {p 0 4}{cmdab:a:bsorb(varname)} specifies the panel variable over which to calculate fixed effects. When this option is not specified, {cmd:fese} reads the panel variable from the data {help char:characteristic} {cmd:iis} (often set by either {help xtset} or {help tsset}). {marker s_macros}{title:Remarks and saved results} {p}{cmd:fese} leaves the results from {help areg} in {cmd:e()} and saves matrices of standard errors: Matrices {col 4}{cmd:ov}{col 18}OIM SE vector {col 4}{cmd:rv}{col 18}heteroskedasticity-robust SE vector {col 4}{cmd:cv}{col 18}cluster-robust SE vector {marker s_examples}{title:Examples} {hline} {p 8 12}{stata "ssc install fese, replace" : ssc install fese, replace }{p_end} {p 8 12}{stata "webuse grunfeld, clear" : webuse grunfeld, clear } {p_end} {p 8 12}{stata "fese inv mval kst, s(fe_)" : fese inv mval kst, s(fe_) }{p_end} {p 8 12}g b=. {p_end} {p 8 12}g se=. {p_end} {p 8 12}g hrse=. {p_end} {p 8 12}g crse=. {p_end} {p 8 12}qui tab com, g(d_) {p_end} {p 8 12}qui reg inv mval kst d_* {p_end} {p 8 12}qui forv i=1/10 { {p_end} {p 8 12} replace b=_b[d_`i'] {p_end} {p 8 12} replace se=_se[d_`i'] {p_end} {p 8 12}} {p_end} {p 8 12}qui reg inv mval kst d_*, r {p_end} {p 8 12}qui forv i=1/10 { {p_end} {p 8 12} replace hrse=_se[d_`i'] {p_end} {p 8 12}} {p_end} {p 8 12}qui reg inv mval kst d_*, cl(com) {p_end} {p 8 12}qui forv i=1/10 { {p_end} {p 8 12} replace crse=_se[d_`i'] {p_end} {p 8 12}} {p_end} {hline} {p 8 12}{stata "webuse nlswork, clear" : webuse nlswork, clear}{p_end} {p 8 12}{stata "fese ln age, s(o) oonly" : fese ln age, s(o) oonly}{p_end} {hline} {marker s_refs}{title:References} {p 0 4}Nichols, Austin and Mark E. Schaffer. 2007. ``Cluster-robust and GLS Corrections." Unpublished Working Paper. {p 0 4}Rothstein, Jesse. 2007. ``Do Value-Added Models Add Value? Tracking, Fixed Effects, and Causal Inference." Working Paper, Princeton University. {marker s_citation}{title:Citation of fese} {p}{cmd:fese} is not an official Stata command. It is a free contribution to the research community, like a paper. Please cite it as such: {p_end} {phang}Nichols, Austin. 2008. fese: Stata module calculating standard errors for fixed effects. {browse "http://ideas.repec.org/c/boc/bocode/s456914.html":http://ideas.repec.org/c/boc/bocode/s456914.html}{p_end} {title:Authors} Austin Nichols Urban Institute austinnichols@gmail.com {title:Also see} {p 1 14}Manual: {hi:[U] 20 Estimation and post-estimation commands}, esp. {hi:[U] 20.15 Obtaining robust variance estimates}{p_end} {p 10 14}{hi:[R] areg}{p_end} {p 10 14}{hi:[XT] xtreg}{p_end} {p 1 10}On-line: help for {help areg}, {help xtreg}, {help vce_option}{p_end}