Issues in structural equation modeling Hans Baumgartner Penn State University
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling Misinterpretation of overall model fit Baumgartner and Homburg (1996) showed: □ the median number of degrees of freedom in type III models was 49 (28, 124); □ The median percentage contribution of the measurement model to the total number of degrees of freedom was 93 (81, 97); □ the percentage of type III models for which R 2 for structural equations was reported was 45;
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling 22 1 AVE ( 1 ) =.51AVE ( 2 ) =.56 Discriminant validity
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling Measurement model: 2 (38)=45.16 RMSEA=.026 SRMR=.016 CFI=1.00 TLI=1.00 Latent variable model: 2 (49)= RMSEA=.088 SRMR=.09 CFI=.96 TLI=.95
Issues in structural equation modeling Measurement model: 2 (38)=45.16 RMSEA=.026 SRMR=.016 CFI=1.00 TLI=1.00 Latent variable model: 2 (49)= RMSEA=.088 SRMR=.09 CFI=.96 TLI=.95
Issues in structural equation modeling Common problems Incomplete information 2 statistic and degrees of freedom Misinterpretation of overall model fit Covariance fit vs. variance fit Reflective vs. formative indicators Discriminant validity Measurement model vs. latent variable model Questionable model modification Size of MI vs. conceptual meaningfulness Correlated errors in equations vs. directed paths
Issues in structural equation modeling
Common problems (cont’d) Baron & Kenny and SEM Pooling data from multiple samples Assessment of measurement invariance
Issues in structural equation modeling 11 22 11 33 Mediation
Common problems (cont’d) Baron & Kenny and SEM Pooling data from multiple samples Assessment of measurement invariance
Issues in structural equation modeling Common problems (cont’d) Baron & Kenny and SEM Pooling data from multiple samples Assessment of measurement invariance