You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1091 lines
34 KiB
Plaintext

7 months ago
*! v.2.0.2 Confirmatory factor analysis, by Stas Kolenikov, skolenik at gmail dot com, 08 Sep 2009
program define confa, eclass properties( svyr svyb svyj ) sortpreserve
version 10.0
if replay() {
if ("`e(cmd)'" != "confa") error 301
Replay `0'
}
else {
Estimate `0'
}
end
program define Estimate, eclass properties( svyr svyb svyj )
syntax anything [if] [in] [aw pw iw/ ], [ ///
UNITvar(str) /// provides the list of factors where unit variance identification is used
FREE /// estimate all parameters as free -- the user provides identification through constraints
CONSTRaint(numlist) /// see the previous one
FROM(str) /// starting values, compliant with -ml init- syntax
LEVel(int $S_level) ROBust VCE(string) CLUster(passthru) /// standard errors and inference
LOGLEVel(int 1) /// logging level
CORRelated(string) /// correlated measurement errors
SUBtractone /// subtract one from the sample size in places
USENames ///
MISSing /// allow for special treatment of missing data
SVY * ]
* preliminary work
global CONFA_loglevel = `loglevel'
cap bothlist a \ b, global( CONFA_t )
if _rc==199 {
* listutil not installed
di as err "listutil not found, trying to install from SSC..."
ssc install listutil
}
if "`subtractone'"~="" local subtractone -1
*** MISSING
tempvar touse
marksample touse, zeroweight
global CONFA_touse `touse'
if $CONFA_loglevel > 2 tab $CONFA_touse
* weights?
if "`weight'" ~= "" {
global CONFA_wgt [`weight'=`exp']
global CONFA_wgti [iw=`exp']
}
* initial values?
if "`from'"~="" {
if "`from'" == "iv" | "`from'" == "IV" | "`from'" == "ivreg" | "`from'" == "2SLS" {
if "`unitvar'" ~= "" {
di as err "cannot specify from(`from') and unitvar at the same time"
CleanIt
exit 198
}
else global CONFA_init IV
}
else if "`from'" == "ones" global CONFA_init ones
else if "`from'" == "smart" global CONFA_init smart
else {
gettoken isitmat isitnot : from , parse(",")
cap confirm matrix `isitmat'
if _rc {
di "{err}Warning: matrix `isitmat' not found"
* do something sensible instead
global CONFA_init smart
}
else {
* do nothing --- let the hell break loose
global CONFA_init
}
}
}
* vce?
if "`vce'" ~= "" {
gettoken vce1 rest : vce, parse(" ,")
CheckVCE , `vce1'
local lvce = length("`vce'")
if `"`vce'"' != substr("sbentler", 1, max(2, `lvce')) ///
& `"`vce'"' != substr("satorrabentler", 1, max(3, `lvce')) {
local vceopt vce(`vce')
}
}
if $CONFA_loglevel > 2 di as text "Parsing..."
cap noi Parse `anything'
if _rc {
CleanIt
exit 198
}
* copy everything down -- -ivreg- cleans -sreturn-
local obsvar=s(obsvar)
global CONFA_obsvar `obsvar'
local nobsvar : word count `obsvar'
local nfactors = s(nfactors)
forvalues k=1/`nfactors' {
local indicators`k' = s(indicators`k')
local name`k' = s(name`k')
local factorlist `factorlist' `name`k''
}
* begin collecting the equations, starting values, bounds, and model structure
if $CONFA_loglevel > 2 di as text "Setting the structure up..."
Structure, unitvar(`unitvar') correlated(`correlated') `usenames'
local nicecorr $CONFA_t
* produces a bunch of globals
* did we need all this starting values business at all?
gettoken isitmat isitnot : from , parse(",")
cap confirm matrix `isitmat'
if ~_rc | strpos("`from'",",") {
* the user has provided the starting values
global CONFA_start `from'
}
else {
* not a matrix, no comma: use our ugly computations
global CONFA_start $CONFA_start, copy
}
* if "$CONFA_bounds" ~= "" global CONFA_bounds bounds($CONFA_bounds)
if $CONFA_loglevel > 3 di `" ml model lf confa_lf $CONFA_toML $CONFA_wgt, constraint($CONFA_constr `constraint') `svy' `robust' `cluster' init($CONFA_start) bounds($CONFA_bounds) `options' maximize"'
tempvar misspat touse1
global CONFA_miss `misspat'
qui gen byte `touse1' = 1-$CONFA_touse
if "`missing'" != "" {
if $CONFA_loglevel > 1 di "{txt}Working on missing values..."
* cycle over the observed variables, create missing indicators
forvalues k=1/`nobsvar' {
local thisvar : word `k' of `obsvar'
tempvar miss`k'
qui gen byte `miss`k'' = mi( `thisvar' ) if $CONFA_touse
local misslist `misslist' `miss`k''
}
* sort by pattern: relevant observations first
* when $CONFA_touse==0 `misslist' will be missing
qui {
bysort `touse1' `misslist' : gen long $CONFA_miss = (_n==1)
replace $CONFA_miss = sum( $CONFA_miss )
replace $CONFA_miss = . if mi( $CONFA_touse )
}
cap assert $CONFA_miss == 1 if $CONFA_touse
local anymissing = _rc
if !`anymissing' {
di "{txt}Option missing specified, but no missing data found"
}
else {
qui tab $CONFA_miss
di _n "{txt}Note: {res}" r(r) "{txt} patterns of missing data found"
}
if $CONFA_loglevel > 3 li $CONFA_miss `misslist'
}
* if "`anymissing'"=="0" | "`missing'" == "" {
else {
* -missing- option is omitted
if $CONFA_loglevel > 1 di "{err}NOT {txt}working on missing values"
markout $CONFA_touse `obsvar'
qui gen byte $CONFA_miss = 1 if $CONFA_touse
if $CONFA_loglevel > 2 {
sum `obsvar' if $CONFA_touse
tab $CONFA_miss, missing
}
}
cap noi ml model lf confa_lfm $CONFA_toML $CONFA_wgt if $CONFA_touse, ///
constraint($CONFA_constr `constraint') `svy' `robust' `cluster' `vceopt' ///
init($CONFA_start) bounds($CONFA_bounds) `options' `missing' ///
maximize
local mlrc = _rc
if `mlrc' {
CleanIt
error `mlrc'
}
* parametric matrices
tempname bb
mat `bb' = e(b)
global CONFA_loglevel -1
* to indicate to CONFA_StrucToSigma() that the matrices should be posted to Stata
qui mata : CONFA_StrucToSigma(st_matrix("`bb'"))
global CONFA_loglevel `loglevel'
* now, post all those matrices to ereturn
mat rownames CONFA_Sigma = `obsvar'
mat colnames CONFA_Sigma = `obsvar'
mat rownames CONFA_Lambda = `obsvar'
mat colnames CONFA_Lambda = `factorlist'
mat rownames CONFA_Phi = `factorlist'
mat colnames CONFA_Phi = `factorlist'
mat colnames CONFA_Theta = `obsvar'
mat rownames CONFA_Theta = `obsvar'
ereturn matrix Sigma = CONFA_Sigma, copy
ereturn matrix Lambda = CONFA_Lambda, copy
ereturn matrix Phi = CONFA_Phi, copy
ereturn matrix Theta = CONFA_Theta, copy
if "`missing'"!= "" ereturn local missing missing
eret local observed `obsvar'
eret local factors `factorlist'
if "`unitvar'" ~= "" {
* need to unwrap the contents of `unitvar'...
* or change its defintion from passthru to string
if "`unitvar'" == "_all" eret local unitvar `factorlist'
else eret local unitvar `unitvar'
}
forvalues k=1/`nfactors' {
eret local factor`k' `name`k'' : `indicators`k''
}
if "`correlated'"!="" eret local correlated `nicecorr'
if "`svy'`cluster'`exp'`robust'" == "" & "`vce'"!="robust" & substr("`vce'",1,2)!="cl" & "`missing'"=="" {
* if the data are not i.i.d., LRT is not applicable
* don't know what to do with missing data
tempname S Sindep trind
qui mat accum `S' = `obsvar' $CONFA_wgti if $CONFA_touse, dev nocons
mat `S' = `S' / ( e(N) `subtractone' )
mat `Sindep' = diag(vecdiag(`S'))
* degrees of freedom
local nconstr = `: word count $CONFA_constr' + `: word count `constraint''
local pstar = `nobsvar' * (`nobsvar' + 1) / 2
local df_m = rowsof(CONFA_Struc) - `nobsvar' - `nconstr'
ereturn scalar pstar = `pstar'
* test against independence
mat `trind' = trace( syminv(`Sindep') * `S' )
local trind = `trind'[1,1]
ereturn scalar ll_indep = -0.5 * `nobsvar' * e(N) * ln(2*_pi) - 0.5 * e(N) * ln(det(`Sindep')) - 0.5 * e(N) * `trind'
ereturn scalar lr_indep = 2*(e(ll)-e(ll_indep))
ereturn scalar df_indep = `pstar' - `nobsvar'
ereturn scalar p_indep = chi2tail(e(df_indep),e(lr_indep))
* goodness of fit test
ereturn scalar ll_0 = -0.5 * `nobsvar' * e(N) * ln(2*_pi) - 0.5 * e(N) * ln(det(`S')) - 0.5 * `nobsvar' * e(N)
ereturn scalar df_u = `pstar' - `df_m'
ereturn scalar lr_u = cond(e(df_u)==0,0,-2*(e(ll)-e(ll_0)))
ereturn scalar p_u = chi2tail(e(df_u),e(lr_u))
* make the g.o.f. test the default test
ereturn scalar df_m = `df_m'
ereturn local chi2type LR
ereturn scalar chi2 = e(lr_u)
ereturn scalar p = e(p_u)
* other crap
ereturn matrix S = `S'
if `"`vce'"'==substr("satorrabentler",1,max(3, length("`vce'"))) ///
| "`vce'" ==substr("sbentler",1,max(4, length("`vce'"))) {
* repost Satorra-Bentler covariance matrix
* not defined for complex survey data,
cap noi SatorraBentler, constraint(`constraint') `missing'
if _rc {
di as err "Satorra-Bentler standard errors are not supported; revert to vce(oim)"
}
else {
tempname SBVar SBV Delta Gamma VV U trUG2 Tdf
mat `SBVar' = r(SBVar)
mat `Delta' = r(Delta)
mat `Gamma' = r(Gamma)
mat `SBV' = r(SBV)
mat `VV' = e(V)
mat `SBVar' = ( `VV'[1..`nobsvar',1..`nobsvar'], `VV'[1..`nobsvar',`nobsvar'+1 ...] ///
\ `VV'[`nobsvar'+1...,1..`nobsvar'], `SBVar'[`nobsvar'+1...,`nobsvar'+1...] )
ereturn repost V = `SBVar'
ereturn matrix SBGamma = `Gamma', copy
ereturn matrix SBDelta = `Delta', copy
ereturn matrix SBV = `SBV', copy
ereturn local vce SatorraBentler
ereturn local vcetype "Satorra-Bentler"
* compute the corrected tests, too
* only takes care of the covariance structure
* Satorra-Bentler 1994
mat `U' = `SBV' - `SBV'*`Delta'*syminv(`Delta''*`SBV'*`Delta')*`Delta''*`SBV'
ereturn matrix SBU = `U'
mat `U' = trace( e(SBU)*e(SBGamma) )
ereturn scalar SBc = `U'[1,1]/e(df_u)
ereturn scalar Tsc = e(lr_u)/e(SBc) * (e(N) `subtractone' ) / e(N)
ereturn scalar p_Tsc = chi2tail( e(df_u), e(Tsc) )
mat `trUG2' = trace( e(SBU)*`Gamma'*e(SBU)*`Gamma')
ereturn scalar SBd = `U'[1,1]*`U'[1,1]/`trUG2'[1,1]
ereturn scalar Tadj = ( e(SBd)/`U'[1,1]) * e(lr_u) * (e(N) `subtractone' ) / e(N)
ereturn scalar p_Tadj = chi2tail( e(SBd), e(Tadj) )
* saddlepoint approximation comes here!!!
* Yuan-Bentler 1997
ereturn scalar T2 = e(lr_u)/(1+e(lr_u)/e(N) )
ereturn scalar p_T2 = chi2tail( e(df_u), e(T2) )
}
}
}
* are we done yet?
ereturn matrix CONFA_Struc = CONFA_Struc
ereturn local predict confa_p
ereturn local estat_cmd confa_estat
ereturn local cmd confa
Replay
CleanIt
end
program define CleanIt
* just in case
return clear
* release the constraints
constr drop $CONFA_constr
* clear the globals
if $CONFA_loglevel < 3 {
global CONFA_constr
global CONFA_init
global CONFA_loglevel
global CONFA_toML
global CONFA_start
global CONFA_bounds
global CONFA_args
global CONFA_constr
global CONFA_obsvar
global CONFA_t
global CONFA_wgt
global CONFA_wgti
}
end
program define Parse, sclass
* number of factors?
local input `0'
mata: st_local("nfactors",strofreal(CONFA_NF(`"`input'"')))
if `nfactors' == 0 {
* something terrible happened
di as err "incorrect factor specification"
exit 198
}
sreturn local nfactors = `nfactors'
tokenize `input', parse("()")
local k = 0
while "`1'"~="" {
* right now, `1' should contain an opening bracket
if "`1'"~="(" {
* the first character is not a "("
di as err "incorrect factor specification"
exit 198
}
else {
* the first character IS a "("
mac shift
* right now, `1' should contain a factor-type statement
local ++k
local factor`k' `1'
mac shift
* right now, `1' should contain a closing bracket
if "`1'"~=")" {
* the first character is not a ")"
di as err "incorrect factor specification"
exit 198
}
else mac shift
* it may contain a space, I guess
* then -mac shift- it again
if trim("`1'")=="" mac shift
}
}
forvalues k=1/`nfactors' {
* now, parse each factor statement
tokenize `factor`k'', parse(":")
sreturn local name`k' `1'
* `2' is the colon
unab indicators : `3'
sreturn local indicators`k' `indicators'
local obsvar `obsvar' `indicators'
}
cap uniqlist `obsvar'
if _rc == 199 {
* uniqlist not found
di as err "uniqlist not found, trying to install from SSC..."
ssc install listutil
uniqlist `obsvar'
}
local obsvar = r(list)
* mata: st_local("obsvar",CONFA_UL(`"`obsvar'"'))
sreturn local obsvar `obsvar'
sreturn local nobsvar = `: word count `obsvar''
end
program define Structure
syntax , [unitvar(str) correlated(str) usenames]
* implement usenames:
* the parameters go along with the factor and variable names
* rather than matrix indices
* utilize all the sreturn results
if $CONFA_loglevel > 3 sreturn list
* copy everything down -- -ivreg- cleans -sreturn-
local obsvar=s(obsvar)
local nobsvar : word count `obsvar'
local nfactors = s(nfactors)
forvalues k=1/`nfactors' {
local indicators`k' = s(indicators`k')
local name`k' = s(name`k')
local factorlist `factorlist' `name`k''
}
if "`unitvar'" == "_all" {
local unitvar `factorlist'
}
* set up the labeling system
if "`usenames'" != "" {
* give the parameters varname labels
forvalues k=1/`nobsvar' {
local o`k' : word `k' of `obsvar'
}
forvalues k=1/`nfactors' {
local f`k' `name`k''
}
}
else {
* give the parameters numberic lables
forvalues k=1/`nobsvar' {
local o`k' `k'
}
forvalues k=1/`nfactors' {
local f`k' `k'
}
}
* returns:
* - ML equations
* - ML bounds
* - the structure matrix
* - ML statement for the likelihood evaluator
* initialize everything
local eqno = 0
global CONFA_toML
global CONFA_start
global CONFA_args
global CONFA_constr
global CONFA_bounds
mata : CONFA_Struc = J(0,4,.)
* process the means first
tokenize `obsvar'
forvalues j=1/`nobsvar' {
* 1. equations to ML
local ++eqno
global CONFA_toML $CONFA_toML (mean_`o`j'':)
* 2. starting values
sum ``j'', mean
global CONFA_start $CONFA_start `r(mean)'
* 3. confa_lf arguments
global CONFA_args $CONFA_args mean_`o`j''
* 4. CONFA structure
mata : CONFA_Struc = CONFA_Struc \ (1, `eqno', `j', 0)
}
* next, process lambda's
forvalues k=1/`nfactors' {
* determine if unitvar is needed here
bothlist `name`k'' \ `unitvar', global(CONFA_t)
if "$CONFA_t" ~= "" {
* identification by unit variance, no scaling variables
local scalevar
}
else {
* identification by the scaling variable: the 1st one on the list
local scalevar : word 1 of `indicators`k''
}
forvalues j=1/`nobsvar' {
* determine whether `k'-th factor loads on `j'-th variable
if strpos( "`indicators`k''", "``j''") {
* 1. equations to ML
local ++eqno
global CONFA_toML $CONFA_toML (lambda_`o`j''_`f`k'':)
* 2. starting values
local r2_`j' = 0.5
if "``j''" == "`scalevar'" {
* the current one is the scaling variable
* set up the constraints, initialize to 1
global CONFA_start $CONFA_start 1
constraint free
local nconstr = r(free)
constraint `nconstr' [lambda_`o`j''_`f`k'']_cons = 1
global CONFA_constr $CONFA_constr `nconstr'
}
else if "$CONFA_init" == "IV" {
* initialize by a simple version of instrumental variables
* use the remaining indicators of this factor as instruments
dellist `indicators`k'', delete(`scalevar' ``j'')
local ivlist = r(list)
if "`ivlist'" == "." {
di as err "Warning: no instruments available for ``j''"
local ivl = 1
}
else {
qui ivreg ``j'' (`scalevar' = `ivlist')
local ivl = _b[`scalevar']
}
global CONFA_start $CONFA_start `ivl'
if !mi(e(r2)) local r2_`j' = e(r2)
}
else if "$CONFA_init" == "ones" {
global CONFA_start $CONFA_start 1
}
else {
* no init options
global CONFA_start $CONFA_start 0
}
global CONFA_bounds $CONFA_bounds /lambda_`o`j''_`f`k'' -100 100
* 3. confa_lf arguments
global CONFA_args $CONFA_args lambda_`o`j''_`f`k''
* 4. CONFA structure
mata : CONFA_Struc = CONFA_Struc \ (2, `eqno', `j', `k')
}
}
}
* next, process Phi matrix
forvalues k=1/`nfactors' {
local scalevar1 : word 1 of `indicators`k''
foreach kk of numlist `k'/1 {
* 1. equations to ML
local ++eqno
global CONFA_toML $CONFA_toML (phi_`f`kk''_`f`k'':)
* 2. starting values
if `k' == `kk' {
* diagonal entry
bothlist `name`k'' \ `unitvar', global(CONFA_t)
if "$CONFA_t" ~= "" {
* identification by unit variance
constraint free
local nconstr = r(free)
constraint `nconstr' [phi_`f`k''_`f`k'']_cons = 1
global CONFA_constr $CONFA_constr `nconstr'
local v`k' = 1
}
else {
* identification by the scaling variable
if "$CONFA_init" == "smart" | "$CONFA_init" == "IV" {
qui sum `scalevar1'
local v`k' = r(Var)*0.5
}
else if "$CONFA_init" == "ones" local v`k' = 1
else local v`k' = 0
}
global CONFA_start $CONFA_start `v`k''
global CONFA_bounds $CONFA_bounds /phi_`f`k''_`f`kk'' 0 1000
}
else {
* off-diagonal entry
if "$CONFA_init" == "smart" | "$CONFA_init" == "IV" {
local scalevar2 : word 1 of `indicators`kk''
qui corr `scalevar1' `scalevar2'
local v = 0.5*r(rho)*sqrt(`v`k''*`v`kk'')
}
else if "$CONFA_init" == "ones" local v = 0.5
else local v = 0
local vv = 1.5*abs(`v') + 0.01
global CONFA_start $CONFA_start `v'
global CONFA_bounds $CONFA_bounds /phi_`f`kk''_`f`k'' -`vv' `vv'
}
* 3. confa_lf arguments
global CONFA_args $CONFA_args phi_`f`kk''_`f`k''
* 4. CONFA structure
mata : CONFA_Struc = CONFA_Struc \ (3, `eqno', `kk', `k')
}
}
* residual variances
forvalues j=1/`nobsvar' {
* 1. equations to ML
local ++eqno
global CONFA_toML $CONFA_toML (theta_`o`j'':)
* 2. starting values
if "$CONFA_init" == "ones" {
local v_`j' = 1
}
else if "$CONFA_init" == "IV" | "$CONFA_init" == "smart" {
qui sum ``j''
local v_`j' = r(Var)*(1-`r2_`j'')
}
else local v_`j' = 0.01
global CONFA_start $CONFA_start `v_`j''
global CONFA_bounds $CONFA_bounds /theta_`o`j'' 0 1000
* 3. confa_lf arguments
global CONFA_args $CONFA_args theta_`o`j''
* 4. CONFA structure
mata : CONFA_Struc = CONFA_Struc \ (4, `eqno', `j', 0)
}
* the error correlations
while "`correlated'" != "" {
gettoken corrpair correlated : correlated , match(m)
gettoken corr1 corrpair : corrpair, parse(":")
unab corr1 : `corr1'
gettoken sc corr2 : corrpair, parse(":")
unab corr2 : `corr2'
* make sure both are present in the list of observed variables
poslist `obsvar' \ `corr1', global(CONFA_t)
local k1 = $CONFA_t
if `k1' == 0 {
di as err "`corr1' is not among the observed variables"
CleanIt
exit 198
}
poslist `obsvar' \ `corr2', global(CONFA_t)
local k2 = $CONFA_t
if `k2' == 0 {
di as err "`corr2' is not among the observed variables"
CleanIt
exit 198
}
* will be empty @ the first call
local nicecorr `nicecorr' (`corr1':`corr2')
* 1. equations to ML
local ++eqno
global CONFA_toML $CONFA_toML (theta_`o`k1''_`o`k2'':)
* 2. starting values
global CONFA_start $CONFA_start 0
local vv = sqrt(`v_`k1''*`v_`k2'')
global CONFA_bounds $CONFA_bounds /theta_`o`k1''_`o`k2'' -`vv' `vv'
* 3. confa_lf arguments
global CONFA_args $CONFA_args theta_`o`k1''_`o`k2''
* 4. CONFA structure
mata : CONFA_Struc = CONFA_Struc \ (5, `eqno', `k1', `k2')
}
if "`nicecorr'"!="" global CONFA_t `nicecorr'
if $CONFA_loglevel > 3 {
di as text "ML input (" as res `: word count $CONFA_toML' as text "): " as res "$CONFA_toML"
di as text "Starting values (" as res `: word count $CONFA_start' as text "): " as res "$CONFA_start"
di as text "Likelihood evaluator (" as res `: word count $CONFA_args' as text"): " as res "$CONFA_args"
di as text "Constraints (" as res `nfactors' as text "): " as res "$CONFA_constr"
di as text "Correlated errors: " as res "`nicecorr'"
constraint dir $CONFA_constr
mata : CONFA_Struc
}
mata : st_matrix("CONFA_Struc",CONFA_Struc)
end
program define Replay
syntax, [posvar llu(str) level(passthru)]
* get the implied matrix
tempname bb Sigma
mat `bb' = e(b)
* mata : st_matrix("Sigma",CONFA_StrucToSigma(st_matrix("`bb'")))
mat `Sigma' = e(Sigma)
mat CONFA_Struc = e(CONFA_Struc)
* determine what kind of labeling has been used
* RATHER FRAGILE: checking for mean_1 rather than trying to find
* whether option usenames was specified
cap local whatis = [mean_1]_cons
if _rc {
* mean_1 not found => labeling by names
forvalues k=1/`: word count `e(observed)' ' {
local o`k' : word `k' of `e(observed)'
}
forvalues k=1/`: word count `e(factors)' ' {
local f`k' : word `k' of `e(factors)'
}
}
else {
* mean_1 was found => labeling by numbers
forvalues k=1/`: word count `e(observed)' ' {
local o`k' `k'
}
forvalues k=1/`: word count `e(factors)' ' {
local f`k' `k'
}
}
* header
di _n as text "`e(crittype)' = " as res e(ll) _col(59) as text "Number of obs = " as res e(N)
di as text "{hline 13}{c TT}{hline 64}"
if "`e(vcetype)'" ~= "" {
di as text " {c |} {center 15:`e(vcetype)'}"
}
di as text " {c |} Coef. Std. Err. z P>|z| [$S_level% Conf. Interval]"
di as text "{hline 13}{c +}{hline 64}"
tokenize `e(observed)'
local nobsvar : word count `e(observed)'
* let's go equation by equation
local eqno = 0
* Means
_diparm __lab__, label("Means") eqlabel
forvalues j = 1/`nobsvar' {
local ++eqno
_diparm mean_`o`j'' , label("``j''") prob `level'
}
* Loadings
_diparm __lab__, label("Loadings") eqlabel
local ++eqno // to point to the next line
forvalues k=1/`: word count `e(factors)' ' {
_diparm __lab__ , label("`: word `k' of `e(factors)' '")
while CONFA_Struc[`eqno',1]<=2 & CONFA_Struc[`eqno',4]==`k' {
local j = CONFA_Struc[`eqno',3]
_diparm lambda_`o`j''_`f`k'', label("``j''") prob `level'
local ++eqno
}
}
* Factor covariance
_diparm __lab__, label("Factor cov.") eqlabel
forvalues k=1/`: word count `e(factors)'' {
foreach kk of numlist `k'/1 {
_diparm phi_`f`kk''_`f`k'', label("`: word `kk' of `e(factors)''-`: word `k' of `e(factors)''") prob `level'
local ++eqno
}
}
* Error variances
_diparm __lab__, label("Var[error]") eqlabel
forvalues j= 1/`nobsvar' {
_diparm theta_`o`j'' , label("``j''") prob `level'
local ++eqno
}
* Error correlations
if `eqno' <= rowsof(CONFA_Struc) & CONFA_Struc[`eqno',1] == 5 {
_diparm __lab__, label("Cov[error]") eqlabel
while (`eqno' <= rowsof(CONFA_Struc) & CONFA_Struc[`eqno',1] == 5) {
local k1 = CONFA_Struc[`eqno',3]
local k2 = CONFA_Struc[`eqno',4]
_diparm theta_`o`k1''_`o`k2'', label("``k1''-``k2''") prob `level'
* range check: what am I supposed to check here? Hm...
local ++eqno
}
}
if "`e(vcetype)'"~="Robust" & "`e(missing)'"=="" {
di as text "{hline 13}{c +}{hline 64}"
di as text "R2{col 14}{c |}"
forvalues j = 1/`nobsvar' {
qui sum ``j'' if e(sample)
local r2 = (`Sigma'[`j',`j']-_b[theta_`o`j'':_cons])/r(Var)
di as text %12s "``j''" "{col 14}{c |}{col 20}" as res %6.4f `r2'
}
}
di as text "{hline 13}{c BT}{hline 64}"
if e(df_u)>0 {
di as text _n "Goodness of fit test: LR = " as res %6.3f e(lr_u) ///
as text _col(40) "; Prob[chi2(" as res %2.0f e(df_u) as text ") > LR] = " as res %6.4f e(p_u)
}
else {
di as text "No degrees of freedom to perform the goodness of fit test"
}
di as text "Test vs independence: LR = " as res %6.3f e(lr_indep) ///
as text _col(40) "; Prob[chi2(" as res %2.0f e(df_indep) as text ") > LR] = " as res %6.4f e(p_indep)
if "`e(vce)'" == "SatorraBentler" & e(df_u)>0 {
* need to report all those corrected statistics
di as text _n "Satorra-Bentler Tsc" _col(26) "= " as res %6.3f e(Tsc) ///
as text _col(40) "; Prob[chi2(" as res %2.0f e(df_u) as text ") > Tsc ] = " as res %6.4f e(p_Tsc)
di as text "Satorra-Bentler Tadj" _col(26) "= " as res %6.3f e(Tadj) ///
as text _col(40) "; Prob[chi2(" as res %4.1f e(SBd) as text ") > Tadj] = " as res %6.4f e(p_Tadj)
di as text "Yuan-Bentler T2" _col(26) "= " as res %6.3f e(T2) ///
as text _col(40) "; Prob[chi2(" as res %2.0f e(df_u) as text ") > T2 ] = " as res %6.4f e(p_T2)
}
if "`e(vce)'" == "BollenStine" {
* need to report Bollen-Stine measures
di as text _n "Bollen-Stine simulated Prob[ LR > " as res %6.4f e(lr_u) as text " ] = " as res %6.4f e(p_u_BS) ///
as text _n "Based on " as res e(B_BS) as text " replications. " ///
as text "The bootstrap 90% interval: (" as res %6.3f e(T_BS_05) as text "," ///
as res %6.3f e(T_BS_95) as text ")"
}
mat drop CONFA_Struc
end
**************************** Satorra-Bentler covariance matrix code
program SatorraBentler, rclass
syntax [, noisily constraint(numlist) missing]
if "`missing'"!="" {
di "{err}cannot specify Satorra-Bentler standard errors with missing data"
exit 198
}
* assume the maximization completed, the results are in memory as -ereturn data-
* we shall just return the resulting matrix
* assume sample is restricted to e(sample)
* preserve
* keep if e(sample)
* get the variable names
tempname VV bb
mat `bb' = e(b)
mat `VV' = e(V)
local p : word count $CONFA_obsvar
qui count if $CONFA_touse
local NN = r(N)
* compute the implied covariance matrix
tempname Lambda Theta Phi Sigma
mata : st_matrix("`Sigma'",CONFA_StrucToSigma(st_matrix("`bb'")))
* compute the empirical cov matrix
tempname SampleCov
qui mat accum `SampleCov' = $CONFA_obsvar $CONFA_wgti if $CONFA_touse , nocons dev
* divide by sum of weights instead???
mat `SampleCov' = `SampleCov' / (`NN'-1)
* compute the matrix Gamma (fourth moments)
if $CONFA_loglevel > 4 {
di as text "Computing the Gamma matrix of fourth moments..."
}
tempname Gamma
SBGamma $CONFA_obsvar if $CONFA_touse
mat `Gamma' = r(Gamma)
return add
* compute the V matrix, the normal theory weight
if $CONFA_loglevel > 4 {
di as text "Computing the V matrix..."
}
SBV `SampleCov' `noisily'
if !mi(r(needmatsize)) {
di as err "matsize too small; need at least " r(needmatsize)
exit 908
}
tempname V
mat `V' = r(SBV)
return add
* compute the Delta matrix
if $CONFA_loglevel > 4 {
di as text "Computing the Delta matrix..."
}
tempname Delta DeltaId
noi mata : SBStrucToDelta("`Delta'")
*** put the pieces together now
* enact the constraints!
SBconstr `bb', constraint(`constraint')
* zero out the rows of Delta that correspond to fixed parameters
mat `DeltaId' = `Delta' * diag( r(Fixed) )
local dcnames : colfullnames `bb'
local drnames : rownames `Gamma'
mat colnames `DeltaId' = `dcnames'
mat rownames `DeltaId' = `drnames'
return matrix Delta = `DeltaId', copy
tempname VVV
mat `VVV' = ( `DeltaId'' * `V' * `DeltaId' )
mat `VVV' = syminv(`VVV')
mat `VVV' = `VVV' * ( `DeltaId'' * `V' * `Gamma' * `V' * `DeltaId' ) * `VVV'/`NN'
* add the covariance matrix for the means, which is just Sigma/_N
* weights!
* third moments!
return matrix SBVar = `VVV'
end
* of satorrabentler
* Compute Gamma: the fourth moments matrix -- check!
program define SBGamma, rclass
syntax varlist [if] [in]
unab varlist : `varlist'
tokenize `varlist'
marksample touse
local p: word count `varlist'
forvalues k=1/`p' {
* make up the deviations; weights are used in a weird way
*** MISSING: change r(mean) to _b[whatever] ?
qui sum ``k'' $CONFA_wgti if `touse', meanonly
tempvar d`k'
qui g double `d`k'' = ``k'' - r(mean) if `touse'
local dlist `dlist' `d`k''
}
local pstar = `p'*(`p'+1)/2
forvalues k=1/`pstar' {
tempvar b`k'
qui g double `b`k'' = .
local blist `blist' `b`k''
}
* convert into vech (z_i-bar z)(z_i-bar z)'
mata : SBvechZZtoB("`dlist'","`blist'")
* blist now should contain the moments around the sample means
* we need to get their covariance matrix
tempname Gamma
qui mat accum `Gamma' = `blist' $CONFA_wgti if `touse', dev nocons
mat `Gamma' = `Gamma'/(r(N)-1)
mata : Gamma = st_matrix( "`Gamma'" )
* make nice row and column names
forvalues i=1/`p' {
forvalues j=`i'/`p' {
local namelist `namelist' ``i''_X_``j''
}
}
mat colnames `Gamma' = `namelist'
mat rownames `Gamma' = `namelist'
return matrix Gamma = `Gamma'
end
* of computing Gamma
* compute V = 1/2 D' (Sigma \otimes Sigma) D
* normal theory weight matrix, see Satorra (1992), eq (24) -- check!
program define SBV, rclass
args A noisily
tempname D Ainv V
local p = rowsof(`A')
if $CONFA_loglevel > 3 di as text "Computing the duplication matrix..."
mata : Dupl(`p',"`D'")
mat `Ainv' = syminv(`A')
cap mat `V' = .5*`D''* (`Ainv' # `Ainv') * `D'
if _rc == 908 {
* need a larger matrix
return scalar needmatsize = rowsof(`A')*rowsof(`A')
}
else {
return matrix SBV = `V'
}
end
* of computing V
program define SBconstr, rclass
* need to figure out whether a constraint has the form [parameter]_cons = value,
* and to nullify the corresponding column
syntax anything, [constraint(numlist)]
local bb `anything'
* that's the name of the parameter vector, a copy of e(b)
tempname Iq
mat `Iq' = J(1,colsof(`bb'),1)
tokenize $CONFA_constr `constraint'
while "`1'" ~= "" {
constraint get `1'
local constr `r(contents)'
gettoken param value : constr, parse("=")
* is the RHS indeed a number?
local value = substr("`value'",2,.)
confirm number `value'
* parse the square brackets and turn them into colon
* replace the opening brackets with nothing, and closing brackets, with colon
* that way, we will get "parameter:_cons", which is the format of e(b) labels
local param = subinstr("`param'","["," ",1)
local param = subinstr("`param'","]",":",1)
local param = trim("`param'")
local coln = colnumb(`bb',"`param'" )
mat `Iq'[1,`coln']=0
mac shift
}
return matrix Fixed = `Iq'
end
program define CheckVCE
syntax [anything] , [ROBust CLuster oim opg SBentler SATorrabentler BOOTstrap JACKknife]
if "`bootstrap'" ~= "" {
di "{err}vce(bootstrap) not allowed, but you can run {inp}bootstrap ... : confa ... {err}instead."
CleanIt
exit 198
}
if "`jackknife'" ~= "" {
di "{err}vce(jackknife) not allowed, but you can run {inp}jackknife ... : confa ... {err}instead."
CleanIt
exit 198
}
end
exit
if "$SOCST" == "c:\-socialstat" {
// at home, run the Mata file
do C:\-Mizzou\CONFA\confa.mata
}
else {
// for public release, add Mata code
mata : mata mlib index
}
Globals used:
CONFA_init -- initialization type
CONFA_loglevel -- detail level
CONFA_toML -- model statement for -ml model-
CONFA_start -- default starting values
CONFA_bounds -- ml search bounds
CONFA_args -- the list of parameters, to appear in -confa_lf-
CONFA_constr -- the list of constraints
CONFA_obsvar -- the list of observed variables
CONFA_wgt -- weight specification
CONFA_wgti -- iweight
CONFA_t -- temporary global for -listutil-
Structure matrix:
CONFA_Struc -- the model structure: (parameter type, equation number, index1, index2)
History:
v.1.0 -- Jan 09, 2008
-- basic formulation without -cluster-, -robust-, -weights-, -svy-
v.1.1 -- Mar 21, 2008
-- Satorra-Bentler?
v.1.2 -- Sep 16, 2008
-- Ken Higbee comments
v.1.5 -- usenames
-- Mata moved to lconfa.mlib
-- survey-compatible
v.1.6 -- listwise deletion for missing data
-- what kind of idiot should Stas be to not pay attention to this???
-- informative message about matsize in Satorra-Bentler calculations
v.2.0 -- FIML missing data
-- prepared for revision in SJ
v.2.0.1 -- fixed -if- in Satorra-Bentler calculations
v.2.0.2 -- fixed reporting of correlations with -unitvar-: confa.ado, confa_estat.ado
v.2.1 -- someday?
-- Bartlett correction: (N - 1 - (2p+4m+5)/6)
-- F-statistic in place of chi-square, both normal theory and S-B