# Bayesian hierarchical modelling of academic orientation and advising effects on student retention and progression: Multi-cohort evidence

## TL;DR

This study uses Bayesian modeling to show how academic orientation and advising impact student retention and progression in a South African university.

## Contribution

The novelty lies in applying Bayesian hierarchical modeling to analyze multi-cohort educational data in a Global South context.

## Key findings

- Stronger academic orientation performance is linked to higher retention and progression rates.
- Academic advising engagement improves outcomes, especially when paired with good orientation performance.
- Bayesian hierarchical models offer inferential advantages by capturing outcome correlations without sacrificing predictive accuracy.

## Abstract

Student retention and academic progression remain central concerns in higher education, particularly within contexts characterised by widening access and structural inequality. This study examines the independent and interactive effects of academic orientation performance and academic advising utilisation on first-year student retention and progression at a South African public university. Using administrative data from 8,300 undergraduate students across three entry cohorts spanning the 2023–2024 academic periods, we employ Bayesian hierarchical multivariate modelling to account for cohort-level variation and the interdependence between retention and progression outcomes. Retention is operationalised as enrolment in the subsequent academic year and is analysed only for cohorts with observable follow-up data (based on confirmed registration records), while academic progression is examined for all cohorts. Results indicate that stronger performance in academic orientation is positively associated with both retention and progression, while engagement with academic advising is associated with improved outcomes, particularly when combined with higher orientation performance. The observed interaction effects suggest complementary engagement between orientation and advising rather than differential treatment effectiveness, and all estimated relationships are interpreted as associative rather than causal. Model comparison results indicate statistically indistinguishable predictive performance between joint and univariate specifications, with the joint model offering additional inferential advantages by capturing correlations between outcomes. Overall, the findings highlight the value of integrated first-year support interventions and demonstrate the utility of Bayesian hierarchical approaches for analysing complex, multi-cohort educational data in Global South higher education contexts.