dc.description.abstract | There is a lot of emphasis now for on-farm than on station trials. The big challenge
here is how to address carefully the high variability in farmers fields while designing an
experiment to ensure that the results from such studies are of high quality. In practice
researchers fit blocks along contours in a field perceived to be uniform without necessarly
subjecting the soil sample for laboratory testing to identify homogeneous sections. Chapter
1 we have Introduction, literature review, problem statement, objective and methodology
of the study. Previous studies in on-farm trials have given very high Coefficient of variation
(CV) of over 40% and R2 of less than 0.5, which is attributed to careless designing. The
objectives of this study were to: find out if plots in the same contour across a farm are
always homogeneous and if the contour is an appropriate and fast criteria for blocking in
Randomised Complete Block Design (RCBD) experiment in on-farm trials, use Exploratory
data analysis method to identify underlying structure of the data collected; fit several models
to the data and identify the best model and recommend the length of a block that guarantees
homogeneity within the same contour on-farm. The study was conducted in Meru and Embu
Districts in 32 farms, 16 in each district. RCBD design was used with each block divided
into 8 plots. Blocks were el1hr 16m, 24m, 32m or -40m long. In chapter two we show
how data is modelled in data anaysis. The data collected was subjected to exploratory
data analysis in chapter three and was found to be normally distributed. It was then
analysed by fitting linear and proc mixed models i~.S-Plus and SAS respectively. The
result in chapter four shows that 'plots within blocks of 16m long did not give significantly
different mean yield (p = 0.2105) while those within blocks of 24m, 32m and 40m long
were significantly different at p = 0:6001,0.0256 and 0.0005 respectively. The same result
was obtained in the height analysis. Plots in the blocks o-flength 16m long did not give
significant difference in maize height. But blocks of length 24,32 and 40 meters long were
significantly different at p = 0.0018,0.119 and 0.0011 respectively. Therefore field of land
(on-farm) is not significantly different within short distance of not exceeding 20m. Contour
can be used as criteria for blocking in the field perceived to be uniform so long as blocks are
kept small. In chapter five I have recommend that blocks of maximum 15m long would be
be ideal length to ensure homogeneity within blocks otherwise when the size of the block
exceeds 20m long then balanced incomplete block design would be ideal. | en |