Postestimation tools for confa
The following commands are available after confa:
command description ------------------------------------------------------------------------- estat fitindices fit indices estat aic AIC estat bic BIC estat correlate correlations of factors and measurement errors predict factor scores -------------------------------------------------------------------------
Special interest postestimation commands
These commands provide some additional post-estimation output.
estat correlate transforms the covariance parameters into correlations for factor covariances and measurement error covariances. The delta method standard errors are given; for correlations close to plus or minus 1, the confidence intervals may extend beyond the range of admissible values. Additional options are allowed: level(#) specifies the CI level bound provides an alternative CI based on Fisher's z-transform (arctanh) of the correlation coefficient. It guarantees that the end points of the interval are in (-1,1) range, which may not produce desirable results for Heywood cases.
estat aic and estat bic compute the Akaike and Schwarz Bayesian information criteria.
estat fitindices computes, prints, and saves into r() results a number of traditional fit indices. The following options of estat fitindices request specific indices:
option fit index ------------------------------------------------------------------------- aic AIC, Akaike information criteria bic BIC, Schwarz Bayesian information criteria rmsea RMSEA, root mean squared error of approximation rmsr RMSR, root mean square residual tli TLI, Tucker-Lewis index cfi CFI, comparative fit index _all all of the above indices, the default -------------------------------------------------------------------------
predict [type] newvarlist [if] [in] [, scoring_method]
predict can be used to create factor scores following confa. The number of variables in newvarlist must be the same as the number of factors in the model specification; all factors are predicted at once by the relevant matrix formula, anyway. The following methods are supported:
option factor scoring method ------------------------------------------------------------------------- regression regression, or empirical Bayes, score empiricalbayes alias for regression ebayes alias for regression mle MLE, or Bartlett score bartlett MLE, or Bartlett score, alias for mle -------------------------------------------------------------------------
bollenstine, reps(#) saving(filename) confaoptions(...) bootstrap_options
bollenstine performs Bollen and Stine (1992) bootstrap. The original data are rotated to conform to the fitted structure. By default, bollenstine re-estimates the model with rotated data, and uses the estimates as starting values in each bootstrap iterations. It also rejects samples where convergence was not achieved (implemented through reject( e(converged) == 0) option supplied to bootstrap).
The following options are supported:
reps(#) specifies the number of bootstrap replications. The default is 200.
saving(filename) specifies the file where the simulation results (the parameter estimates and the fit statistics) are to be stored. The default is a temporary file that will be deleted as soon as bollenstine finishes.
confaoptions(...) allows to transfer options to confa. Some bootstrap replications produce non-convergent samples that may never converge, so in order to speed up computations, it might make sense to limit the number of iterations, say with confaoptions( iter(20) ).
All non-standard model options, like unitvar or correlated, must be specified with bollenstine to produce correct results!
All other options are assumed to be bootstrap_options and passed through to bootstrap.
Example
. use http://web.missouri.edu/~kolenikovs/stata/hs-cfa.dta, clear . confa (vis: x1 x2 x3) (text: x4 x5 x6) (math: x7 x8 x9), from(iv) corr(x7:x8) . estat fit . estat corr . estat corr, bound . predict fa1-fa3, reg . predict fb1-fb3, bart
Also see
Online: confa, bollenstine.
References
Bollen, K. and Stine, R. (1992) Bootstrapping Goodness of Fit Measures in Structural Equation Models. Sociological Methods and Research, 21, 205--229.
Contact
Stas Kolenikov, kolenikovs at missouri.edu