Causal Latent Class Analysis with Distal Outcomes: A Modified Three-Step Method Using Inverse Propensity Weighting

Area of Science:

• Statistics
• Causal Inference
• Latent Class Analysis

Background:

• Latent class (LC) analysis is widely used to link class membership with outcomes.
• Estimating causal effects from observational data requires causal inference techniques due to non-randomized LC membership.
• Existing stepwise LC analyses using propensity scores have limitations in causal effect estimation.

Purpose of the Study:

• To propose and evaluate two novel propensity score-based strategies for estimating the causal effect of latent class membership on distal outcomes.
• To address confounding in latent class analysis using observational data.
• To compare the performance of the proposed methods against existing propensity score approaches in stepwise latent class analysis.

Main Methods:

• Modification of the bias-adjusted three-step latent class analysis by incorporating propensity scores in the final step.
• Two strategies proposed: inverse propensity weighting (IPW) and including propensity scores as control variables.
• Classification errors handled using BCH or ML corrections; performance evaluated via simulation and real-world data (LISS panel).

Main Results:

• Both proposed methods yielded essentially unbiased parameter estimates, outperforming previously suggested methods.
• The IPW-based approach demonstrated high variability and potential non-convergence with smaller sample sizes.
• The methods were successfully illustrated using data from the LISS panel.

Conclusions:

• The novel propensity score strategies effectively estimate the causal effect of latent class membership on distal outcomes.
• These methods provide a valuable advancement for causal inference in latent class analysis with observational data.
• Researchers should be mindful of sample size limitations when employing the IPW-based strategy.