Abstract

Data arising from repeated measurements of experimental units occur in many occasions in forestry and related fields. Very often such data are analysed without considering their several peculiarities, like correlation between successive measurements and heterogeneity of variances, which may lead to erroneous conclusions. The present study was undertaken with the objective of identifying appropriate methods of analysis of data from long term trials characterised by repeated measurements on experimental units. In this study, three different methods of analysing repeated measures viz., two way analysis of variance, univariate mixed model analysis of variance and multivariate analysis of variance were discussed with respect to their suitability in different contexts. These three methods were compared using data collected from certain typical situations in forestry and the appropriate method of analysis to be followed in respective cases were identified. Specifically, data collected from a study on several soil properties observed from multiple core samples from 0-15, 15-50 and 50-100 cm layers under six different types of vegetation and another study on annual yield of latex from rubber trees in three years were used for comparing the appropriatenessof the methods. The study revealed that multivariate analysis of variance is the most appropriate methods of analysis for majority of the soil properties. This was found justifiable because the extent of residual variation in individual soil properties at different depths was not of the same order and also the correlation between values at different layers were not of the same magnitude. Multivariate analysis of variance was found suitable for analysing data on annual yield of latex from rubber trees aswell. This was so due to the heterogeneity in variannces and covariances of variance-covariance matrix of errors