Filling the gap in food and nutrition security data: What imputation method is best for Africa's food and nutrition security?

Abstract

Our study presents the methods adopted to produce accurate imputed values for Africa's food security and nutrition (FSN). We focused primarily on the following five imputation methods for handling missing data: Mean Imputation; Multiple Imputed values using a Chained Equation (MICE); imputation based on the Conditional Distribution of a variable Diagnostics (mi); Additive Regression and Predictive Mean Matching (Hmisc); and Random Forest (missForest). We describe each method, including how they performed under MAR and MNAR using RMSE and MAE as a measure of accuracy. After these methods of imputation were examined for nonignorable missing values in the context of accurate and unbiased estimates for food and security analysis, we found that MissForest handled nonignorable missing values more effectively and with less bias, increasing the precision of the data by imputing the closest data values within the dataset. Hence the missForest is the best alternative for handling missing values for food security and nutrition concerning Africa. This study adds to the current body of knowledge on food and nutrition insecurity and provides useful information to policymakers, particularly about the imputations of missing values aimed at food security and nutrition concerning Africa, which has significant economic and social ramifications.

Keywords

• food security and nutrition
• missForest
• missing value
• multiple imputations
• Missing at Random (MAR)
• Missing Not at Random (MNAR)