Design and Analysis of On-Station and On-Farm Agricultural Research Experiments: A Revisit

Abstract

In every scientific investigation, the variability in the experimental material is taken care of by arranging the experimental units in blocks such that the experimental units within a block are as homogeneous as possible. However, in majority of experiments conducted in National Agricultural Research System, the block mean squares have been found to be small as compared to the error mean square meaning thereby that the very purpose of forming blocks has been defeated. It is, therefore, desirable to develop a methodology of deciding shape and size of plots and blocks, so that in new blocking structure, the block mean square is large as compared to error mean square. The present study has been undertaken in this spirit. Several uniformity trials data were analyzed from various angles and it has been observed that the nested block structures with smaller block sizes are beneficial in comparison to usual blocks that are generally rectangular or square strips, in the sense that it improves the precision of the treatment comparisons.

On-Farm Research Experiments are an important component of agricultural research. In these experiments, the technologies developed at research stations are tested by taking into account the realistic environment and with farmers’ participation in specific recommendation domain. For on-Farm Research, resolvable block designs have been identified that are quite useful in taking care of the (a) plot to plot variation and (b) variation in farmers’ managerial skills. A catalogue of resolvable block designs for v<=10 has also been prepared.

In on-farm research trials, farmers' practice (most common practice adopted by farmers) is taken as one of the treatments and the experiments are conducted to compare new treatments with the farmers' practice. The designs for making test treatments-control treatment comparisons have been found to be useful for these situations. In many experimental situations, identification of one common farmers' practice may not be feasible. For such situations, it has been recommended that for a given farming situation, instead of identifying one common farmer’s practice to be used as control, group the farmers on the basis of practices used and as many controls should be used as number of groups. In such situations, reinforced blocks/nested block designs are useful. Here a reinforced resolvable design is a resolvable design in test treatments and different control treatments are added once to every block of the resolvable group (or bigger block). For the trials designed and managed by farmers, with some intervention of the researcher, treatments can be partitioned into two groups, viz. (i) core treatments and (ii) farmer decided treatments. Every farmer should use core treatments. Augmented designs have been recommended for such experimental situations.

In most of the On-farm trials, the farmers are selected using a multistage random sampling design. Hence, some of the effects in the model are random. At present analysis is generally carried out using a fixed effect model. Therefore, a linear, nested mixed effect model has been developed for analysis of on-farm trial data currently being generated by PDCSR, Modipuram. This is capable of identifying specific developmental blocks that favour one treatment over the other that is not possible through comparing treatment effects averaged over all development blocks alone.