MSc.

The present study entitled “Comparison of statistical methods for control of error in long term experiments in rice (Oryza sativa L.)” was conducted at College of Agriculture, Vellayani during 2015 - 17. Objective of the study is to compare different parametric and non-parametric statistical approaches in the analysis of field experiments over years and seasons in long term experiments in rice and to identify the most suitable method.
Data on a field experiment on rice (var. Aiswarya) viz. ‘Permanent plot experiments on integrated nutrient supply system for a cereal based crop sequence’ conducted at Integrated Farming System Research Station (IFSRS), Karamana for the period from 1985 - 86 to 2013 - 14 were used for the study. The field experiment consisted of 12 different treatments on modified fertilizer doses based on the recommended dose including a control T1 (no fertilizers and no organic manures) and T12 (farmer’s practice). Randomized block design (RBD) with four replications was used for kharif and rabi seasons for all these years. The main observations collected were grain yield, straw yield, plant height, total number of tillers, number of productive tillers, dry matter production and harvest index. The descriptive statistics and the usual RBD analysis of variance (ANOVA) were carried out for all the biometric characters and detailed study were made on grain yield data of (kharif, rabi and yearly data) by different approaches.
Pooled analysis of raw and transformed (square root and logarithmic) grain yield data indicated highly heterogeneous estimates of error variances, ie. mean sum of squares for error (MSE), (5.22 to 35.7 for kharif, 5.74 to 32.04 for rabi and 12.25 to 90.06 for yearly data). Weighted analysis was then attempted which produced non-significant year × treatment interactions which indicated that more refined statistical procedures are needed for effective conclusions. So exploratory statistical analysis was attempted. The data were subjected to univariate normality tests and those years
with more than ten outliers were discarded and hence 21 years data were used for further study. The statistical procedures ordinary pooled analysis, split plot type of analysis, analysis of covariance (ANCOVA), time series (serial correlations) regression analysis and a non-parametric method (Friedman’s test) were conducted.
Ordinary pooled analysis of the data indicated homogeneity of error variances with a pooled error of 8.42, 8.92 and 20.16 for kharif, rabi and yearly data respectively and year × treatment interactions were found to be significant. The treatment T6 [50% RDN of NPK through fertilizers + 50% through FYM for kharif, 100% RDN of NPK through fertilizers for rabi and (50% RDN of NPK through fertilizers + 50% through FYM + 100% RDN of NPK through fertilizers for yearly data)] was obtained as highest yield during many of the years or seasons.
Then the data were subjected to Split plot type of analysis, the treatments were taken in main plot and years or seasons in subplots. In this, the sub plot error variances obtained were 10.23, 9.65 and 23.09 for kharif, rabi and yearly data respectively, which were higher than that of ordinary pooled analysis.
A correlation study was conducted with grain yield and the other characters, to identify those characters having high correlation with grain yield and treated them as covariates for ANCOVA. It is observed that, as the number of covariates increased there was not much changes in the error variances but there is a declining tendency for treatment variances. So it is inferred that the variable having high correlation with grain yield (viz. straw yield) can be taken for covariance analysis.
Time series regression analysis and serial correlations were attempted for specific treatments. It was found that neither serial correlations nor partial regression coefficients were found to be significant for kharif, rabi as well as yearly data.
Non parametric analysis is one of the best methods for non normal data. The treatment means were ranked for each year and subjected to Friedman’s test for two way classified data. Significant treatment differences were obtained and treatment T6 obtained best score.
Hence it is concluded that treatment T6 maintained the highest yield over the years and seasons. Ordinary pooled analysis of data was found to be the best under the exploratory data analysis. Analysis of covariance with one covariate was found to be equally good with adjusted MSE almost equal to that of MSE of ordinary pooled analysis.