Statistical Modelling of Child Abuse Incidence in Malaysia: A Count Data Regression Approach (2016-2021)

Abstract

Child abuse continues to be one of the fundamental social and public health issue in Malaysia, with rising reported cases despite protection under the law. This study analyses national-level secondary data at the national level from Department of Statistics Malaysia (DOSM) and the Social Welfare Department (JKM) to examine the occurrence of child abuse between year 2016 and 2021. The research is guided by three key objectives, firstly to identify the exposure factors of child abuse using Poisson Regression Modelling, then to determine the dependency between ethnicity and gender in reported abuse cases using a Chi-Square Independence Test and lastly to model child abuse and its contributing factors using both Poisson and Negative Binomial Regression. The six socioeconomic exposure variables examined were: poverty rate, divorce, early marriage, drug addiction, violent crime, and depression. In order to model count data, the study used a quantitative technique that included descriptive analysis, Chi-Square Independence Test, Poisson Regression, and Negative Binomial Regression. The Chi-Square Test result show a statistically correlation between gender and ethnicity in reported cases of child abuse (p<0.001). Overdispersion (deviance/df=368.7) disregarded model assumptions, even though Poisson Regression originally found several significant predictors. As a result, the Negative Binomial Regression model was chosen since it had a lower AIC (28.8) and BIC (288.0). Depression was shown to be the sole statistically significant predictor of all variables (p=0.093), with a positive coefficient, indicating that higher frequency of child maltreatment. This study supports Negative Binomial Regression as a suitable technique for overdispered count data in social science research and highlights the significance of mental health in child protection policies.