Article Text
PDF
Statistics from Altmetric.com
Objective
We aim to use longitudinal latent class analysis (LLCA) to explore patterns of alcohol consumption over time, while considering the impact of associated covariates.
Background and Data
Data were collected to investigate the social impact of drinking in students at a UK University, during the period 2006–2007. The number of units of alcohol consumed each day, over a period of seven days, is the outcome measure, giving differing patterns of consumption over time (trajectories) for each student. Non-drinkers (n = 289) were excluded, giving a total of 3183 students available for analysis.
Methods
Alcohol consumption may vary according to many other factors related to the student or their course, such as: gender, age, smoking status and year of study. We use LLCA to classify the study participants into latent classes, to investigate how these trajectories are associated with covariates of interest. Instead of undertaking analysis over all students, this method simplifies by looking at natural clusterings of trajectories of alcohol consumption over time and the emerging classes then contain types of students rather than all individuals. LLCA model fit was explored comparing log-likelihood statistics and misclassification rates.
Results
As the number of latent classes is increased, the model fit continues to improve. Selecting only a few classes provides a clear picture of behaviour whereas including many classes has the ability to express more diversity in the associated alcohol consumption trajectories. To provide a balance between simplicity and sufficient expression, the model with four latent classes was chosen. The model contained one class of heavy drinkers, with a high number of units consumed daily; two classes of moderate drinkers, with differing patterns of consumption; and one class of light drinkers, with a low number of units consumed at the weekend only. Class profiles differed by student characteristics (sex, age, smoking status, ethnicity, number of dependents, UK resident status) and by course characteristics (faculty, mature student status, year of study).
Conclusions
The longitudinal latent class structure was informative: the model suggests differing natural clusterings of trajectories of alcohol consumption over time and these trajectories may be associated with characteristics of the student and/or their course. By assessment of these characteristics, there may be an opportunity to identify those students who might consume excess alcohol, and so permit the targeting of a social intervention.