MULTILEVEL MODELLING WITH APPLICATION TO CHILD DISCIPLINE PRACTICES IN GHANAIAN HOMES

Abstract

This study applied multilevel modelling to analyze data on child discipline practices from 8,156 households from the MICS 4 data collected by GSS in 2011. The analysis took into consideration intra-cluster correlation that results from the collection of such hierarchical structured data and unmeasured higher level characteristics that impact on values of the response variables. A key purpose of the study was to address the error terms of individual observations that correlate in such datasets and lead to the violation of standard statistical assumption of different observations having independent residuals. Autocorrelation which leads to underestimation of standard errors of regression coefficients and results in overestimation of t values and subsequent underestimation of p values was addressed. The results showed significant contextual effects on household responses at the regional level but not at the location (rural/urban) level. Also, younger household heads (15-21 years), single child households and wealthy households tended not to approve the use of physical discipline for correction. Ethnicity had significant effect on household’s approval of physical discipline. In actual application of the different discipline methods by households, the results further showed that, the number of children aged 2-14 in a household, the religion and education level of household heads had significant influence on a household’s likelihood to apply physical discipline methods or non-physical discipline methods or psychological aggression discipline methods relative to the reference group of random use of discipline methods. It is recommended that to totally eliminate violence against children we will need to take into consideration information provided in this study to provide parents with nonviolent alternative discipline methods.