Analysis of Maize Yield Data of the 50 Years Old Long-Term Experiments in Martonvasar.

The long-term experiments (LTE) set up in Martonvásár (HU) are now 50 years old. The 14 LTEs consist of a total of over 400 plots, covering almost 20 hectares. The most important of these experiments involve crop rotation vs. monoculture trials, the comparison of fertilisation systems, studies on the interactions and carry-over effects of organic and mineral fertilisers, fertiliser rate experiments and polyfactorial experiments. Valuable scientific results are obtained from these experiments regarding the reasons for yield depression in monocultures, the yield-increasing effect of crop rotations, the comparative benefits of organic and mineral fertilisation, the agronomic responses of genotypes, the sustainability and yield stability of crop production techniques, and the interaction between various crop production factors. The sustainability of various cropping systems can only be properly investigated in LTEs.

Data from LTE are in the form of repeated measurements on each treated plot. Repeated-measures data present a special challenge for statistical analysis. Among the most common approaches are: (i) the summary statistics, (ii) multivariate ANOVAs, and (iii) modelling the correlation structure. The summary statistics approach to repeated measures is to reduce the multiple measurements on each plot to one or more summary statistics that measure some phenomenon of interest. One analysis approach that has gained popularity in recent years is to model the serial correlation associated with the repeated measures, and then to base inferences on a mixed model ANOVA that incorporates the estimated serial correlation structure. Serial correlation here refers to the correlation between measurements taken on the same plot across time. Stability analysis is a suitable method for the interpretation of the significant experiment ´ treatment interactions observed in variance analysis of long-term experiments. Time series analysis and forecasting are important tools in the interpretation of long-term data.