A Systematic Literature Review of Latent Variable Mixture Modeling in Communication Scholarship

ABSTRACT

Recently, latent variable mixture modeling has gained traction in many disciplines, given its unique ability to discover unknown groups within a broader population. Indeed, this method assumes that a finite number of mixtures (i.e. unknown groups) exist within the population and can be discovered by evaluating participants’ response patterns to a set of manifest indicators. Despite the intuitive approach, recommendations have been proposed to overcome some methodological concerns associated with latent variable mixture modeling. The primary purpose of this study was to understand the characteristics of latent variable mixture modeling in communication research and to evaluate the extent to which the existing research meets these recommendations. Ninety-five manuscripts published between 2010 and 2022 in 18 communication journals were identified and systematically analyzed. The review found that (1) the use of latent variable mixture modeling has increased; (2) latent class analysis and latent profile analysis are the most common models; and (3) most manuscripts did not meet the proscribed standards for random start values, auxiliary variable procedures, indicator requirements, and missing data procedures. These findings are discussed more in comparison with the proscribed standards. In addition, conceptual and applicable recommendations are provided to improve communication scholarship.

Acknowledgments

We would like to thank Monica Cornejo and the reviewers for their constructive feedback on earlier versions of this manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19312458.2023.2179612

Notes

¹ Classes containing less than 5% of the sample should not be retained because they are susceptible to “…low power and a lack of statistical precision” (Wickrama et al., 2016, p. 215).

² Other extensions of entropy have been proposed: the classification likelihood information criterion (CLIC; Biernacki et al., 2000), the normalized entropy criterion (NEC; Celeux & Soromenho, 1996), and variable-specific entropy (Asparouhov & Muthén, 2018).

³ 2019) recommends the ML for distal outcomes over the BCH, whereas Asparouhov and Muthén (2014/2021) and Bakk and Kuha (2021) recommend the BCH over the ML. Both approaches have concerns. The ML approach can experience class switching in the final step, whereas the BCH can produce negative weights.

⁴ The “manifest indicators” for LCGA and GMM are intercepts and slopes from latent growth models.

⁵ Though recommended, the result of cross-validation procedures and its extensions have been mixed (e.g., Whittaker & Miller, 2021). Researchers should use a combination approach for the class enumeration process.

Additional information

Notes on contributors

Colton E. Krawietz

Colton E. Krawietz (M. A. - University of Texas at Austin) is a doctoral candidate at the University of Texas at Austin. His research examines how people in close relationships evade intense conversations, how their motivations influence their communication behaviors, and how their perceptions of and attributions about these talks can impact their psychological distress and desire to exit the relationship.

Rudy C. Pett

Rudy C. Pett (Ph.D. - University of Texas at Austin) is an assistant professor at Saint Louis University. His research examines how individuals strategically edit their communication about distressing issues in order to maximize their individual and relational outcomes.