Principal Component and Multiple Correspondence Analyses in Dimensionality Reduction: A Study on Aflatoxin Contamination of Peanuts in Kenya Henry

In a study about the factors that contributed to the risk of aflatoxin contamination of peanuts in the Peanut CRSP project in Kenya, contingency table analysis (Pearson’s chi-square) was used to analyze a large mixed data set from a survey. The data was collected between March and July 2009 from three provinces in Kenya namely Nairobi, Western and Nyanza. Data analysis with contingency tables has limitations since it cannot allow for testing of statistical significance, variables with many categories produce large tables that were difficult to read and the Chi-square test cannot provide predicted values and can only be used to analyze the effect of a single categorical variable on the response. This study was intended to identify more sensitive statistical methods that could overcome the above limitations by analyzing the data using multiple regression analysis, analysis of variance (ANOVA), Principal component analysis (PCA) and Multiple correspondence analysis (MCA). With such methods, 12 factors were identified as having played a significant role in enhancing aflatoxin contamination of peanuts. Principal component analysis was useful in reduction of the large data set of 37 variables into a lower dimension of six variables and in constructing data composites for MCA. Multiple correspondence analysis was applicable in the interpretation of aflatoxin contamination of peanuts by establishing associations for more than two categorical variables in a low-Euclidean dimensional space and was an excellent heuristic for getting into complex multi-factorial data than contingency tables. There is need for further studies on some of the variables that were identified as having played a significant role in aflatoxin contamination of the peanuts, especially those to do with peanut storage and housing conditions in order to qualify the findings.