Question 1

Which of the following suggests an assumption of CFA has been met?&#10;A) self-selection of units in the analysis&#10;B) skewness statistic of 8.50&#10;C) Tolerance value of .01&#10;D) VIF of 2

Accepted Answer

Another statistical method for detecting collinearity is to compute a variance inflation factor (VIF) for each observed variable (using a series of special regression analyses). The largest VIF should be less than 10 in order to satisfy this assumption (Myers, 1990; Stevens, 2009; Wetherill, 1986).

Question 2

Which statistical procedure is most closely related to confirmatory factor analysis in terms of the types of evidence it can provide?&#10;A) discriminant analysis&#10;B) exploratory factor analysis&#10;C) hierarchical generalized linear modeling&#10;D) multiple linear regression

Accepted Answer

Confirmatory factor analysis (CFA) is similar to EFA in that it can be used to provide evidence of construct validity by examining the underlying constructs of variables, allowing us to work with all items simultaneously and at the same time know something about their underlying data structure.

Question 3

Endogenous latent variables are akin to which one of the following?&#10;A) dependent variable&#10;B) exogenous variable&#10;C) independent variable&#10;D) indirect effect

Accepted Answer

Endogenous latent variables are caused from one or more variables in the model and are related to dependent variables.

Question 4

Exogenous latent variables are akin to which one of the following?&#10;A) dependent variable&#10;B) exogenous variable&#10;C) independent variable&#10;D) indirect effect

Accepted Answer

Exogenous variables are akin to independent variables.

Question 5

Model degrees of freedom equals which one of the following?&#10;A) indirect effects between latent constructs&#10;B) the correlation between observed variables that load onto the latent constructs&#10;C) the largest number of parameters to be estimated minus the actual number of estimated parameters&#10;D) the largest number of variances and unique covariances that can be estimated

Accepted Answer

The difference between the largest number of parameters to be estimated, p, and the actual number of estimated parameters, q, is the model degrees of freedom: df_m = p - q.

Question 6

Only when the following occurs can the CFA parameters be estimated?&#10;A) when the latent constructs are interrelated&#10;B) when the correlation between latent constructs is directionally specified&#10;C) when the number of elements in the variance-covariance matrix is less than the number of freely estimated parameters in the model&#10;D) when the measurement model is identified

Accepted Answer

A CFA model is identified if a unique set of parameter estimates for each unknown parameter in the model (i.e., latent construct) can be derived from the known variance-covariance matrix. It is only when the measurement model is identified that the CFA parameters can be estimated.

Question 7

Which one of the following reflects an appropriately scaled latent construct?&#10;A) five observed variables load onto a latent construct&#10;B) the metric of a latent construct has been fixed to the metric of another latent construct&#10;C) the number of elements in the variance-covariance matrix is greater than the freely estimated parameters&#10;D) the variance of a latent construct has been fixed to 1.0

Accepted Answer

Second, latent constructs must be scaled in one of two ways: (a) fix the variance of the construct to a specified value, usually 1.0; or (b) fix the metric of the construct by scaling to an observed variable (i.e., scaling an indicator variable).

Question 8

Which one of the following estimation procedures should NOT be used when the observed variables are categorical?&#10;A) maximum likelihood&#10;B) robust weighted least squares&#10;C) weighted least squares&#10;D) unweighted least squares

Accepted Answer

Although ML may be the most widely used estimation procedure, it is by far not the only from which to select. And in some cases, is not the appropriate estimation procedure to select. Recall that one of the assumptions when using ML is the distribution of the observed variables are multivariate normal and one of the conditions is that the observed variables are continuous. When these do not occur, other estimation procedures are more appropriate.

Question 9

A researcher finds an RMSEA value of .065. Which one of the following is suggested by this value?&#10;A) close model fit&#10;B) model underidentification&#10;C) multivariate nonnormality&#10;D) singularity

Accepted Answer

The RMSEA can range from 0 (perfect fit) to positive infinity (however values greater than +1.0 are uncommon). Values of .05 to .08 or smaller indicating close.

Question 10

A researcher finds a SRMR value of .05. Which one of the following is suggested by this value?&#10;A) good model fit&#10;B) model underidentification&#10;C) multivariate skewness&#10;D) multicollinearity

Accepted Answer

The SRMR can range from 0 (perfect fit) to +1.0 and will be lower as the number of estimated parameters increase and as the sample size increases. Different thresholds of acceptable model fit exist. Values of < .08 have been suggested as good model fit (Hu & Bentler, 1999) as have values < .05 (Schumacker & Lomax, 2010).

Deck 10: Path Analysis, Confirmatory Factor Analysis, and Structural Equation Modeling