Latent Class Analysis of Survey Data Dealing With Student Attitude toward Faculty Classroom Behavior

Article By: 

Joseph Adwere-Boamah

Currently, I am a faculty member of Shirley M. Hufstedler School of Education at Alliant International University. I completed my PhD from UC Berkeley in 1970. My areas of interest are quantitative methods, educational policy and program evaluation. I retired after 35 years from the Oakland School District as an Assistant Superintendent of Planning, Research and Policy Development in 2000.


The purpose of this article is to illustrate the use of Latent Class Analysis (LCA) of survey data to identify distinct groups or clusters of students based on their responses to a set of questionnaire items. Using this analysis may help colleges to identify favorable classroom behavior and student feelings towards the class.  


The current emphasis on Data Driven decisions has encouraged colleges and school districts to collect not only student achievement data, but also all kinds of attitudinal and behavioral data on students, teachers and instructors. The emphasis on accountability mainly defined in terms of academic outcomes has led however, to extensive analysis of stated instructional outcomes to the exclusion of most of the attitudinal data that might be collected annually. Data driven decisions can be enhanced by analysis of most of the attitudinal and other school or system wide data to provide a context for interpreting student academic and or institutional outcomes. A typical analysis of school or college survey data usually focuses on frequency distributions and simple cross tabulation of responses. The inadequate time and resources of institutional researchers may discourage them from spending time to search their data for underlying patterns or structure to enable then to develop a deeper understanding of the data.
This paper analyzed survey data from a larger study of students’ needs assessment conducted in a Northern California Community college. Specifically, student ratings of five Likert- type attitudinal items on faculty classroom behavior were analyzed using LC models for identification and characterization of distinct groups or clusters of students in the general college population (Sample items were: In my classes, most instructors find ways to help students to understand the materials, In my class, I feel free to ask for help from my instructor, My instructor cares about me as a person).   
Latent Class analysis is a method for analyzing the relationship among manifest variables where a number of latent or unobserved categorical variables are used to explain the relationships among the manifest data. Unlike Cluster analysis, or K-Means approach to cluster analysis, Vermunt and Magidson (2002) describe LCA as model-based because it:

  • Uses model based posterior membership probabilities to classify cases into homogeneous groups, clusters or latent classes.
  • Provides Bayesian information criterion (BIC) among other diagnostic statistics for determining the number of latent clusters. 
  • Enables the analysis to be performed on nominal, ordinal or continuous variables as well as counts. 

           The analysis was performed with Latent Gold Version 4.5 (Vermunt & Magidson 2005).  The analysis began with an exploratory LC Cluster analysis of the data.  Different models were estimated by stepwise addition of classes, until the model that fitted the data well was obtained.
The results of the analysis show that a survey model with three classes/groups fitted the data better than one, two, or four class models. The three class model has the lowest BIC, (4737.034) compared to the others. (One class model BIC = 5321.024, two class model BIC = 4779.042, and four class model BIC = 4776.171). The model identified three distinct groups of students and provided evidence of the existence of group differences in attitude toward faculty behavior. The pattern of responses of the three groups (indicated by the conditional probabilities), suggests that the groups could be named as follows:
       Group 1, “Students with positive experiences toward their instructors’ class room    behavior”, (consisted of 79% of the sample). 
       Group 2, "The Neutrals" (14%), and,
       Group 3, "Students with negative experiences" (7%).
These results indicate that there are three subgroups of students in the college with a large majority (79%) who have positive experiences of their instructors’ classroom behavior. However, a small minority (7%) of the students registered negative experiences. Further investigations may uncover significant correlates of students’ experiences of their instructor’s classroom behavior.
In summary, the identification and characterization of attitudes of subgroups of college students toward faculty classroom behavior could provide important feedback to college faculty and shed some light on students’ feelings. Universities and Colleges are in tough times especially those in California. Because of the state’s severe budgetary problems, the Colleges are struggling amid the harshest economic climate in a generation. However, public demand for a strong system of institutional accountability requires the colleges to collect essential data for program assessment, planning and evaluation. It is imperative that we extract as much information as possible from the data we collect to:

  •  Increase our knowledge and understanding of the complex substantive problems
    facing our colleges, and
  • Enhance our decision-making capabilities

The methods of Latent Class Analysis are rigorous statistical methods that can be used to extract useful information from the different types of data usually collected by colleges.
It can be used to develop and construct suitable models of the real-world problems facing colleges today.

1. For more complete information on LCA, see 1) Papers by Jay Magidson and Jeroen Vermunt, 2.) J.A. Hagenaars and A.L McCutcheon, (2002), Applied Latent Class Analysis NY: Cambridge University Press


Hagenaars, J.A & McCutcheon, A.L (2002). Applied Latent Class Analysis. New York: Cambridge University Press
Vermunt, j.k. & Magidson, J. (2002). Latent Class Analysis in Hagenaars, J.A & McCutcheon, A.L (eds.), Applied Latent Class Analysis. New York: Cambridge University Press, pp 89- 106