BREEDING AND GENETICS Statistical Evaluation of the Cotton Regional Breeders Testing Network (RBTN)

In the U.S., Ted Wallace coordinates a Regional Breeders Testing Network (RBTN) for cotton that is sponsored by Cotton Incorporated. The objective of this program is to provide testing sites for public breeders and geneticists across a regional basis. This study looked at various statistical aspects of field testing with emphasis on lint yield. Ten years of data (107 environments) were examined. The objectives were to: 1) determine any relationship between error variance and mean lint yield, 2) establish a procedure for rejecting less precise data, and 3) discern the most optimum testing sites in the program. To achieve the first objective the natural log of er ror variances were regressed on the natural log of mean environmental lint yields and tested for significance. The “b” value of 0.85 was significant indicating that the error variance increased with increasing yield levels. Using a procedure previously published on rejecting less precise data, the second objective was met. Five of the 107 environments were deemed imprecise and should not have been included in across-location tables. A genotypic index regression method was followed to ascertain the most desirable test sites in the program. Twelve of 23 test sites did an acceptable job of discriminating the entries. Thus, by eliminating nearly half of the test sites more reliable data can be produced, less seed would be required, and more efficient use of resources would be achieved. Sites with less than desirable tests might contribute by collecting data on disease tolerance, morphological traits, or insect resistance, etc. and thus might still be valuable in the cotton RBTN. T he cotton (Gossypium hirsutum L.) Regional Breeders Testing Network (RBTN) was established as a means of providing a range of testing sites for public breeders and geneticists. Their elite breeding lines are tested in diverse environments in exchange for providing a testing site at their location. Over the years the program has had sites ranging from Virginia to California. The number of sites within a particular year might vary from eight to 15 but the tendency has been to include more sites. This paper reports on a statistical study of the testing program with the objective of finding useful suggestions to improve efficiency. It is generally accepted that certain locations have more desirable resources and are probably more suited to field testing. Various statistical aspects of field testing will be covered in this paper. It is generally thought that error variance increases with mean yield (Snedecor and Cochran, 1967), but Bowman and Rawlings (1995) showed that for three maturity groups of corn (Zea mays L.) and one maturity group of soybean (Glycine max L.) there was not a significant relationship. There were significant relationships in barley (Hordeum vulgare L.), oat (Avena sativa L.), wheat (Triticum aestivum L.), and for one maturity group of soybean (Bowman and Rawlings, 1995). Allen et al. (1978) also showed a relationship between error variance and yield. Even though these latter relationships were significantly different from zero, they were also significantly different from 2.0, which is the threshold one needs to use the coefficient of variation (CV) as an indica tor of experimental validity (Bowman and Watson, 1997). The relationship between error variance and lint yield in cotton has not been well documented. Combining data across locations has been a topic of several papers over the years starting with Yates and Cochran in 1938. The issue is combining data from locations with different levels of precision. The RBTN attempts to cover environments representative of the growing area. Even though these trials are nearly identical in size, they are not always similar in precision. Due to soil variability and other factors, results from these trials can be quite variable. The assumptions of the analysis of variance are not always