APA Style JARS

Journal Article Reporting Standards

JARS–Quant | Table 7
Reporting Standards for Studies Using Structural Equation Modeling

Title Method (continued)

• Mention the basic mechanism or process reflected in the primary model to which the • For a power analysis, state the
data are fit. (Note: The complexity of the multivariate data analyzed in many structural ‒ target level of power
equation modeling (SEM) studies makes it unlikely that, in most cases, the variables under ‒ null and alternative hypotheses
investigation and the relations between them could be concisely stated in the title.) ‒ significance of key parameters
‒ fit statistics that figured in the analysis
Abstract
‒ expected population effect sizes
• Report values for at least two global fit statistics, each from a different class, and include a • Report the computer software or algorithm used if the data were generated by simulation,
brief statement about local fit (residuals). State whether the interpreted model (if any model state and justify the sizes of generated samples, and disclose whether samples were lost
is retained) is the originally specified model. because of nonconvergence or inadmissible estimates.

Introduction Results
• Describe the primary model to be fitted to the data, and include an explanation of theory • Report data diagnostics, including
or results from previous empirical studies that support the primary model.
‒ percentage of missingness (if some data are missing) and how it is distributed across
• Point out paths that are especially important, and justify directionality assumptions, cases and variables
such as the claim that X causes Y instead of the reverse. Do the same for paths of
secondary importance. ‒ empirical evidence or theoretical arguments about causes of missing data (i.e., missing
completely at random [MCAR], missing at random [MAR], or missing not at random
• State whether respecification is planned, if the primary model is rejected.
[MNAR])
Method ‒ evidence that distributional or other assumptions of estimation methods are plausible

• State whether the data were collected from research participants or generated by Missing Data
computer simulation.
• Report whether indicators of latent variables were drawn from one questionnaire or from • Indicate the statistical method used to address missingness, such as multiple imputation,
multiple questionnaires. full information maximum likelihood (FIML), substitution of values, or deletion of cases. For
multiple imputation or FIML estimates, state whether variables not included in the model
• Describe, for each questionnaire, whether the indicators are items or total scores across
homogeneous sets of items (scales, parcels), stating how were specified as auxiliary variables.
‒ scales were constructed, reporting their psychometrics
Distributions
‒ items were treated in the analysis as continuous or categorical
• Report how the target sample size was determined, including • State whether the data were evaluated for estimation methods that assume multivariate
‒ rule of thumb normality.
‒ availability of resource constraints ‒ Report values of statistics that measure univariate or multivariate skewness and kurtosis
‒ results of a priori power analysis that support the assumption of normal distributions.
‒ estimates of parameter precision used to plan the number of cases with appropriate ‒ If the data were not multivariate normal, state the strategy used to address nonnormality,
explanation such as use of a different estimation method that does not assume normality or use of
normalizing transformations of the scores.

© 2020 American Psychological Association www.apastyle.org/jars JARS–Quant | Table 7 | Page 1 of 2

 Data Summary Model Fit
• Report in the manuscript—or make available in the supplemental materials—sufficient • Report fit statistics or indices about global (omnibus) fit interpreted using criteria justified by
summary statistics that allow secondary analysis, including citation of most recent evidence-based recommendations for all models to be interpreted.
‒ covariance matrix with means, or correlation matrix with standard deviations and means • Report information about local fit, such as covariance, standardized, normalized, or
for continuous variables correlation residuals, that justify retaining the model at the level of pairs of observed
‒ polychoric correlation matrix, items thresholds, and asymptotic covariance matrix for variables for all interpreted models.
categorical variables • State the strategy or criteria used to select one model over another if alternative models were
‒ whether the case-level data are archived, and provide information about how those data compared. Report results of difference tests for comparisons between alternative models.
can be accessed by interested readers. • State the test and criterion for testing estimates of individual parameters. If parameter
estimated were compared over groups or occasions, indicate how those comparisons
Specification were made.
• Indicate the general approach that best describes the application of SEM: strictly Respecification
confirmatory, comparison of alternative models, or model generation.
• Provide the diagram for each model fitted to the data. If the diagram would be overly • Indicate whether one or more interpreted models was a product of respecification.
complex, such as when large numbers of variables are analyzed, then clearly describe If so, then describe the method used to search for misspecified parameters.
the models in text. A reader should be able to translate the text description of a model • State which parameters were fixed or freed to produce the interpreted model. Also
into a diagram. provide a theoretical or conceptual rationale for parameters that were fixed or freed after
• Give a full account of the specification for all models to be evaluated, including observed specification searching.
variables, latent variables, fixed or free parameters, and constrained parameters. • Indicate whether models for which results are presented were specified before or after
• Report sufficient information, such as tabulations of the numbers of observations versus fitting other models or otherwise examining the data.
free parameters, so that the model degrees of freedom can be derived by the reader.
• Verify that models to be analyzed are actually identified. State the basis for this claim, Estimates
including the method, rules, or heuristics used to establish identification. • Report both unstandardized and standardized estimates for all estimated parameters.
• State the basis in theory or results of previous empirical studies if a measurement model • Report the corresponding standard errors, especially if outcomes of significance testing
is part of a larger model. for individual parameters are reported. State the cutoffs for levels of statistical significance,
• Describe fully the specification of the mean structure if the model has a means component. if such cutoffs were used.
• Explain the rationale for including error correlations in the model if correlated error terms • Report estimates of indirect effects, both unstandardized and standardized. Also report
are specified. values of standard errors for indirect effects, if possible. State and justify the strategy for
• Explain how the effects are specified if the model includes interaction effects. testing indirect effects.
• Explain how nonindependence is accounted for in the model for nested data (e.g., • Report estimates of interaction effects and also results of follow-up analyses that clarify
occasions within persons, students within classrooms). the underlying pattern for interpreted interactions. Also report values of standard errors
• Describe any comparisons of parameters to be made between groups or occasions, and for such interactions.
indicate which parameters are to be compared if models are fitted to data from multiple
groups or occasions. Discussion
• Summarize the modifications to the original model and the bases, theoretical or statistical,
Estimation for doing so.
• State the software (including version) used in the analysis. Also state the estimation method • Address the issue of equivalent models that fit the same data as well as retained models
used and justify its use (i.e., whether its assumptions are supported by the data). or alternative-but-nonequivalent models that explain the data nearly as well as retained
• Disclose any default criteria in the software, such as the maximum number of iterations models. Justify the preference for retained models over equivalent or near-equivalent
or level of tolerance, that were adjusted in order to achieve a converged and admissible versions.
solution.
• Report any evidence of an inadmissible solution (e.g., error variances less than zero or
constrained by the computer at zero; estimated absolute correlations or proportions of
explained variance that exceed 1.0). Explain what was done to deal with the problem.

© 2020 American Psychological Association www.apastyle.org/jars JARS–Quant | Table 7 | Page 2 of 2