Comparative Analysis of the Classical and Robust Regression Models for Predicting Life Expectancy in Nigeria

Abstract

Macroeconomic and demographic datasets from developing economies are frequently 
affected by extreme observations, structural breaks, heteroskedasticity, and non-normal 
error distributions arising from economic instability, policy reforms, financial crises, and 
other exogenous shocks. These characteristics often violate the assumptions underlying the 
Classical Ordinary Least Squares (OLS) regression model, resulting in biased parameter 
estimates, inefficient inference, and unreliable policy conclusions. Robust regression 
techniques provide an effective alternative by reducing the influence of outliers and 
influential observations, thereby producing more stable and reliable parameter estimates 
under non-ideal data conditions. This study evaluates the suitability of robust regression for 
modeling life expectancy in Nigeria by comparing its predictive performance with that of the 
Classical OLS regression model. Gross Domestic Product (GDP) and mortality rate were 
employed as explanatory variables, while life expectancy served as the response variable. 
Both modeling approaches were fitted to the data and their estimated coefficients were 
examined to assess the consistency and reliability of the relationships between the selected 
predictors and life expectancy. The empirical results revealed that mortality rate exerted a 
negative effect on life expectancy in both the Classical OLS and robust regression models, 
indicating that higher mortality levels are associated with reduced life expectancy. However, 
notable differences emerged in the estimated impact of GDP. While the OLS model produced 
a positive but statistically non-significant relationship between GDP and life expectancy, the 
robust regression model identified GDP as having a positive and statistically significant 
influence on life expectancy in Nigeria. This disparity suggests that outliers and influential 
observations may have obscured the true relationship under the OLS framework. 
Consequently, the findings demonstrate that robust regression provides more reliable 
estimates in the presence of anomalous observations and should be preferred over OLS when 
datasets exhibit outlier contamination or violations of classical regression assumptions.