Techniques for Determining Classifications in Multiple Multivariate Data: Application to Prices of Local Food Items on Markets in Ghana

Abstract

The study makes use of time-dependent displaying components and structural equation modelling to determine the price levels of key local food items on several markets in Ghana over non-consecutive time periods. The application of the techniques to this multiple multivariate complex data structure identifies suitable dimensions along which to assess the major influence of the price data over a time-period and highlight possible extreme prices simultaneously. It also examines multiple sets of data-generating variables that may be put together in a single model. These variables include a ‘vectorised’ factor solution and some market-feature covariates. The displaying components comprise the principal component and the outlier displaying component (ODC). The study has provided necessary extensions that would make the components suitable for the study. The plots for the first five components in addition to the preliminary results give the set of suspect outlying markets. Using this set, the Modified 1-ODC is applied based on the pooled reduced sample Sum of Squares and Cross Product matrix. Markets 17 and 65 are clearly identified as the most consistently low and high priced, respectively, over the period. Using the factor solution and two covariates, a structural model is obtained for determining the price levels. Even though the factors constitute a significant model by themselves, they are not significant in the model that contains significant covariates, which are ‘Region’ and the ‘number of days’ of trading. The model shows that extreme markets, which are few, are predominantly associated with large number of market days. Equitable and increased production of cereals and spices in particular in all regions could reduce price variations across markets and enhance well-being.