Abstract
Latent class (LC) modeling is a model based cluster analysis technique. This dissertation applied LC modeling techniques to a measurement problem (study 1) and a substantive problem (study 2). It used secondary data from the Virginia Secondary School Climate Survey (VSSCS; Cornell et al., 2014). Study 1 contained 52,012 students from 323 high schools; whereas the second study sample consisted of 47,631 students from 323 high schools.
The first study introduced a novel technique for identifying invalid respondents in self administered questionnaires (SAQs). Respondent characteristics such as joking, lying, and/or responding carelessly could undermine the validity of the study. It is desirable to screen out such invalid respondents from the analytic sample for accurate inferences. The proposed technique was conducted in three steps: 1) creation of a response-inconsistency variable that gauged the extent to which the individual’s responses increased or decreased the coefficient alpha for the sample for each scale, 2) application of latent class modeling on these variables to examine the clustering at extreme values of the response-inconsistency variables, and 3) cross-validation of cases identified as invalid with traditionally used techniques like screening item (additional item asking students if they responded truthfully) and response time data (if students reported too fast so that they may not have read the questions carefully). Researchers across different fields of social sciences may find this technique useful.
The second study adopted a comprehensive person-centered analytic approach through the examination of profiles of student perceptions of school climate in high schools. Multilevel analyses helped unpack four meaningfully different within-school (student-level) latent profiles: positive climate class, medium climate-low bullying class, medium climate-high bullying class, and negative climate class. On average, students reporting higher levels of disciplinary structure, academic press, teacher respect for students, student’s willingness to seek help from teachers, academic engagement and cognitive engagement also reported lower levels of prevalence of teasing and bullying, general victimization, and probability of being bullied and bullying others. In addition, students reporting positive climate also reported higher academic outcomes and lower risk behaviors. Finally, the implications of within-school clusters of school climate are discussed.
