The data used for our analysis is drawn from the first four waves of the National Income Dynamic Study to determine the factors that influence poverty and household welfare in South Africa. Contrary to most existing studies, which have applied ordinary least squares and probit/logit models on cross-sectional data, this analysis captures unobserved individual heterogeneity and endogeneity, both via fixed effect, and via a robust alternative based on random effect probit estimation. The results from fixed effect and random effect probit indicate that levels of education of the household head, some province dummies, race of the household head, dependency ratio, gender of the household head, employment status of the household head and marital status of the household head are statistically significant determinants of household welfare. Consistent with previous research, we also found that, compared to traditional rural areas (used as reference category), households living in urban and farms are less likely to be poverty stricken, which implies that rural areas (traditional rural areas) should continue to be a major focus of poverty alleviation efforts in South Africa.