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

Authors

• Tra T. Le | Tilburg University (First author)
• Dr. Felix Clouth | Tilburg University (First author)
• Prof. dr. Jeroen Vermunt | Tilburg University

Description

This repository contains the code to reproduce all the results for the paper "Causal Latent Class Analysis with Distal Outcomes: A Modified Three-Step Method Using Inverse Propensity Weighting"

Exemplary Data Analysis

The steps for the exemplary data analysis are as follows

1. Run create_exemplarydata.lgs to generate a synthetic data set 'exemplary.sav'.
2. Run step1model.lgs and save the posterior probabilities in 'classification.txt'
3. Run 1-stepmodel.lgs and save the posterior probabilities and classification results in 'classification2.txt'
4. Use 'classification.txt' and step3model.lgs
5. Run step3-adjusted (ML method and proportional assignment) to get propensity scores saved in 'propensity1.txt'
6. Run step3-unadjusted (no correction method and modal assignment) to get propensity scores saved in 'propensity2.txt'
7. Run schuler.lgs with 'propensity2.txt'
8. Run yamaguchi.lgs with 'classification2.txt'
9. Run braive-tra.lgs with 'propensity1.txt'

Simulation study in batch mode

1. The GenData.lgs file was used to generate the data set ‘data.sav’.
2. The Step1OneStep.lgs file with the simulated data set ‘data.sav’ were used to:
3. run Step 1 model (estimate LC model with only the response indicators) and get the posterior probabilities in ‘data0.sav’.
4. run One-step model (estimate LC model with the response indicators and the confounders) and get the posterior probabilities and classification results in ‘data1.sav’.
5. The Step3.lgs file with the posterior probabilities ‘data0.sav’ were used to:
6. run the adjusted Step 3 model (ML correction method and proportional assignment) to get the propensity scores saved in ‘data2.sav’.
7. run the standard Step 3 model (no correction method) to get the propensity scores saved in ‘data3.sav’.
8. The Schuler.lgs file with ‘data3.sav’ data set were used to run the final step of Schuler et al. (2014) method.
9. The Yama.lgs file with ‘data1.sav’ data set were used to run the final step of Yamaguchi (2015) method.
10. The BrayTraNaive.lgs file with ‘data2.sav’ data set were used to run the final step of Bray et al. (2019), our method, and the naive method.

We used R to generate the above six LatentGOLD syntaxes and run them in batch mode in a loop for 500 replications as follows: "~\ LatentGOLD6 .0\ lg60 " GenData.lgs Step1.lgs Step3.lgs Schuler.lgs Yamaguchi.lgs Bray_Tra_Naive.lgs / b / r 500

R Codes

The R scripts were provided to generate LatentGOLD syntax files, read the results from LatentGOLD, and create graphs and tables.

Empirical Application

The folder contains the LatentGOLD files to analyze an empirical dataset from the LISS panel and the R script to generate the figures and tables in the paper.