Missing observations - exploratory factor analysis efa

[Cecily Josten]

I am running an exploratory factor analysis (EFA) using an individual-level survey data set. The survey answers are coded from 1-5 (agree, agree strongly..) where "does not apply" is coded as missing value. When I run the EFA using Stata's "factor" command, Stata only shows me the output for the observations with non-missing variables. But I would still like to use the responses of an individual even if they clicked "does not apply" for one or more questions or if they only completed part of the survey. How can I do that? I'm losing valuable information here...

[Joseph Coveney]

I would still like to use the responses of an individual even if they clicked "does not apply" for one or more questions or if they only completed part of the survey. How can I do that?

Something like the following might do in a pinch. It uses -factormat- with pairwise correlations from -pwcorr-. Begin at the "Begin here" comment.

.ÿ
.ÿversionÿ17.0

.ÿ
.ÿclearÿ*

.ÿ
.ÿ//ÿseedem
.ÿsetÿseedÿ81044616

.ÿ
.ÿtempnameÿCorr

.ÿmatrixÿdefineÿ`Corr'ÿ=ÿJ(5,ÿ5,ÿ0.7)ÿ+ÿI(5)ÿ*ÿ0.3

.ÿquietlyÿdrawnormÿl1ÿl2ÿl3ÿl4ÿl5,ÿdoubleÿcorr(`Corr')ÿn(200)

.ÿ
.ÿforvaluesÿiÿ=ÿ1/5ÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿegenÿbyteÿo`i'ÿ=ÿcut(l`i'),ÿgroup(4)
ÿÿ3.ÿÿÿÿÿÿÿÿÿquietlyÿreplaceÿo`i'ÿ=ÿo`i'ÿ+ÿ1
ÿÿ4.ÿÿÿÿÿÿÿÿÿquietlyÿreplaceÿo`i'ÿ=ÿ.ÿifÿruniform()ÿ>ÿ0.9
ÿÿ5.ÿ}

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿpwcorrÿo?

ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿo1ÿÿÿÿÿÿÿo2ÿÿÿÿÿÿÿo3ÿÿÿÿÿÿÿo4ÿÿÿÿÿÿÿo5
-------------+---------------------------------------------
ÿÿÿÿÿÿÿÿÿÿo1ÿ|ÿÿÿ1.0000ÿ
ÿÿÿÿÿÿÿÿÿÿo2ÿ|ÿÿÿ0.7185ÿÿÿ1.0000ÿ
ÿÿÿÿÿÿÿÿÿÿo3ÿ|ÿÿÿ0.6719ÿÿÿ0.6553ÿÿÿ1.0000ÿ
ÿÿÿÿÿÿÿÿÿÿo4ÿ|ÿÿÿ0.7084ÿÿÿ0.7108ÿÿÿ0.7118ÿÿÿ1.0000ÿ
ÿÿÿÿÿÿÿÿÿÿo5ÿ|ÿÿÿ0.7137ÿÿÿ0.6665ÿÿÿ0.6755ÿÿÿ0.7570ÿÿÿ1.0000ÿ

.ÿtempnameÿC

.ÿmatrixÿdefineÿ`C'ÿ=ÿr(C)

.ÿ
.ÿquietlyÿcount

.ÿlocalÿNÿ`r(N)'

.ÿ
.ÿfactormatÿ`C',ÿfactors(1)ÿn(`N')ÿforcepsd
(obs=200)

Factorÿanalysis/correlationÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿ=ÿÿÿÿÿÿÿÿ200
ÿÿÿÿMethod:ÿprincipalÿfactorsÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿRetainedÿfactorsÿ=ÿÿÿÿÿÿÿÿÿÿ1
ÿÿÿÿRotation:ÿ(unrotated)ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿparamsÿ=ÿÿÿÿÿÿÿÿÿÿ5

ÿÿÿÿ--------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿFactorÿÿ|ÿÿÿEigenvalueÿÿÿDifferenceÿÿÿÿÿÿÿÿProportionÿÿÿCumulative
ÿÿÿÿ-------------+------------------------------------------------------------
ÿÿÿÿÿÿÿÿFactor1ÿÿ|ÿÿÿÿÿÿ3.43724ÿÿÿÿÿÿ3.44631ÿÿÿÿÿÿÿÿÿÿÿÿ1.0758ÿÿÿÿÿÿÿ1.0758
ÿÿÿÿÿÿÿÿFactor2ÿÿ|ÿÿÿÿÿ-0.00908ÿÿÿÿÿÿ0.04113ÿÿÿÿÿÿÿÿÿÿÿ-0.0028ÿÿÿÿÿÿÿ1.0730
ÿÿÿÿÿÿÿÿFactor3ÿÿ|ÿÿÿÿÿ-0.05021ÿÿÿÿÿÿ0.01969ÿÿÿÿÿÿÿÿÿÿÿ-0.0157ÿÿÿÿÿÿÿ1.0573
ÿÿÿÿÿÿÿÿFactor4ÿÿ|ÿÿÿÿÿ-0.06990ÿÿÿÿÿÿ0.04321ÿÿÿÿÿÿÿÿÿÿÿ-0.0219ÿÿÿÿÿÿÿ1.0354
ÿÿÿÿÿÿÿÿFactor5ÿÿ|ÿÿÿÿÿ-0.11311ÿÿÿÿÿÿÿÿÿÿÿÿ.ÿÿÿÿÿÿÿÿÿÿÿ-0.0354ÿÿÿÿÿÿÿ1.0000
ÿÿÿÿ--------------------------------------------------------------------------
ÿÿÿÿLRÿtest:ÿindependentÿvs.ÿsaturated:ÿÿchi2(10)ÿ=ÿÿ697.31ÿProb>chi2ÿ=ÿ0.0000

Factorÿloadingsÿ(patternÿmatrix)ÿandÿuniqueÿvariances

ÿÿÿÿ---------------------------------------
ÿÿÿÿÿÿÿÿVariableÿ|ÿÿFactor1ÿ|ÿÿÿUniquenessÿ
ÿÿÿÿ-------------+----------+--------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿo1ÿ|ÿÿÿ0.8340ÿ|ÿÿÿÿÿÿ0.3045ÿÿ
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿo2ÿ|ÿÿÿ0.8127ÿ|ÿÿÿÿÿÿ0.3395ÿÿ
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿo3ÿ|ÿÿÿ0.7981ÿ|ÿÿÿÿÿÿ0.3630ÿÿ
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿo4ÿ|ÿÿÿ0.8627ÿ|ÿÿÿÿÿÿ0.2558ÿÿ
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿo5ÿ|ÿÿÿ0.8367ÿ|ÿÿÿÿÿÿ0.3000ÿÿ
ÿÿÿÿ---------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

I'm losing valuable information here...

You might want to compare the factor structures that you get with listwise omission and pairwise correlation coefficients. It's not necessarily a given that respondents who refuse to respond or who click "NA" on various subsets of items will share the same correlation pattern with those who complete the entire form.