2026 ASEE Annual Conference & Exposition

In this method full paper, we present the Majority Rules Procedure, a modification of the existing Majority Vote Procedure, for handling missing data based on multiple imputation in Latent Profile Analysis (LPA) studies. While full-information maximum likelihood (FIML) is considered standard practice for addressing missing data in LPA studies, recent research has identified limitations in its performance. The main insights from this work are that multiple imputation methods, such as the Majority Vote Procedure, are more effective in determining latent profiles within datasets. However, a major limitation still exists with the Majority Vote Procedure; namely, the procedure does not identify nor provide a single dataset for use in subsequent analysis. This limitation hinders follow-up analyses, such as correlating profile membership with critical variables not used in the LPA. While pooling may be an effective method for linear regression, this method is conceptually problematic for LPA, as it raises the question of what a student’s pooled profile membership represents. To address this issue, we proposed the Majority Rules Procedure, which not only identifies the most frequent latent profile result from multiple imputations but also designates a single representative dataset for follow-up analysis. This dataset is determined based on multiple fit indices that consider validity, reliability, and parsimony, including Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-size adjusted Bayesian Criterion (SABIC), entropy, and profile size. We tested the effectiveness of this method using a dataset we have previously used for an LPA study. This dataset contains a full dataset from 1630 first-year engineering students related to their cultural value orientations. After introducing varying levels of artificial missingness, we conducted the Majority Rules Procedure, comparing our LPA results across the different missing-data conditions. Overall, we found that several limitations for this procedure exist, including that high missingness conditions can lead to differing LPA outcomes and that profile separation may be an important factor that limits the performance of imputation techniques. We hope that this work helps to promote others to tackle important questions regarding imputation techniques in person-centered approaches.