Latent Class Analysis with Planned Missingness: Best Approach?

• Activity Number: 605
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Nadra Lisha* and Kevin Delucchi and Pamela Ling
• Companies: and UC San Francisco and UC San Francisco
• Keywords: Planned missingness ; Multiple imputation ; Latent class analysis ; Tobacco

Abstract

Real world data collection must take into account participant burden - for example, in studies such as ours, the amount of time to answer a survey must be short. One way to avoid reducing the number of items, while minimizing response time is to create planned missing data patterns. Data for the current study was taken from a larger tobacco cessation study for young adults where surveys were administered in bars. As such, only 2/3 of participants were given certain items. We focus on the 13-item Social Prioritization Index (SPI) for this analysis. We examined how latent class analysis (LCA) can be used to examine class membership based on the SPI using two imputation approaches. First, we used important variables in the dataset to impute all of the SPI items, and ran LCAs on each of the ten imputed datasets. Second, we ran the LCA on the original non-imputed dataset, and then used the class information and other variables available in the raw data to impute class membership probabilities for each participant. The optimal number of classes across all the LCAs appeared to be five. As such, for each individual we had a 5 (class) by 20 (imputation) matrix where each cell of the matri