Data analysis
The 23 dichotomous caregiving ability indicators were used for Latent Class Analysis (LCA)1125 Wang J, Wang X. Structural Equation Modeling: Applications Using Mplus, 2d Edition. New York: John Wiley; 2020.. The LCA analysis was started with a single-class solution and increased the number of classes (to four) while comparing model fit. Model fit statistics and indices include the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted Bayesian Information Criterion (aBIC), Bootstrapped Likelihood Ratio Test (BLRT), Vuong-Lo-Mendell-Rubin likelihood ratio test (LMR), Lo-Mendell-Rubin adjusted LRT test(aLMR) and entropy score25. Lower values on AIC, BIC and aBIC indicate better model fit; significant p-values on the BLRT, LMR, and aLMR indicate the k class model is more preferable than the k-1 class model25. BIC and BLRT perform better than other fit statistics/indices in determining the number of latent classes. The entropy statistic measures the certainty of class classification. The values of entropy range from 0 to 1, and a value closer to 1 indicating better classification25. For a LCA model with entropy ≥0.80, the latent class membership estimated from the model can be saved as ”observed” categorical variables for further analysis22Clark SL. Mixture Modeling with Behavioral Data. Doctor thesis. University of California, Los Angeles, CA, 2010.. In the present study, we used a multinomial logistic regression model to examine the effects of demographic and clinical variables on the latent class membership. LCA was estimated using Mplus 7.0, and other analyses were conducted using SPSS 22.0. All tests of statistical significance were 2-sided at α=0.05 level.