A systematic review of latent variable mixture modeling in social work research
Latent variable mixture modeling (LVMM) estimates latent, heterogeneous groups within a sample. This systematic review examines the use and quality of LVMM analyses in social work research literature. We screened 478 articles, which were then independently assessed by three reviews to arrive at a final sample of 32 studies meeting inclusion criteria. LVMM studies were published between 2004-2016 with a majority published after 2012 (n = 18, 56.3%). Social workers were listed as first or second authors for most studies (n = 28, 87.5%). Mplus (n = 20, 62.5%) was the most commonly used statistical software package. Samples sizes ranged from 199 to 1,002,122 (median = 533). LCA was used in most studies (n = 19, 59.4%) followed by latent profile analysis (n = 6, 18.8%) and longitudinal variants of LVMM (n = 7, 21.9%). Studies used a mean of 3.13 tests of model fit (SD = 1.24), most common were Bayesian Information Criterion (n = 29, 90.6%) and entropy scores (n = 19, 59.4%). This systematic review demonstrates the usefulness and growing popularity of LVMM studies in social work research. Researchers are encouraged to employ person-centered methods to explore unobserved heterogeneity within cross-sectional and longitudinal data.