Exploratory Analysis of Multivariate Longitudinal Child Education Data

TL;DR

This paper introduces a novel 2D graphical orbit method to analyze multivariate longitudinal data of households, revealing insights into factors influencing child educational default in rural South Africa.

Contribution

It presents a new geometric orbit visualization technique for longitudinal data, enabling comprehensive analysis of household dynamics and their relation to educational outcomes.

Findings

Presence of a biological mother correlates with non-defaulting households.

Orbits effectively identify clusters and change patterns in household data.

Insufficient data to determine effects of household head age and adult death.

Abstract

We analyse binary multivariate longitudinal data of a population of households from a rural district in South Africa. Using a 2-dimensional graphical representation of longitudinal data, each household's data is transformed into a time-evolving geometric orbit. Orbits communicate complete information of change in the data over time and provide insights into the dynamics of both a household's and the population's evolution. The outcome of interest is child educational default, where defaulting is defined as having failed more than three years of schooling. A visual analysis of the impact on educational default of three household factors, namely the presence of a biological mother, the age of the household head (minor- or adult- headed household) and the death of an adult, is presented. In both the non-defaulting and defaulting households, dynamics is mainly described by the temporary in- and out-migration of biological mothers. We find that the presence of mother is more likely in the non-defaulting households. Owing to insufficient events involving change in the age of the household head and adult deaths, we have no conclusion regarding their effect. Orbits offer easily interpreted information of clusters, patterns of change, and the density of state transitions of household orbits.