Multiple comparison of means in empirical agricultural researches

The article considers the carried out analysis of two approaches to mean comparison - the conventional one, which is based on pair-wise comparison according to the least significant difference criterion and arguably rare used in practice of agricultural researches of the approach based on the theory of single-factor analysis of variance. With specific reference of empiric research of influence of probiotic “Olin” on the basic zootechnic indicator of chickens growing - broiler live weight - it is demonstrated that multiple comparison of means with respect to the conservative test of Scheffe, realized by the procedure of generalized linear model of statistical efforts package of social sciences data analysis of SPSS Base. It provides a scientist with comprehensive information on statistical conclusion reliability. Conventional wisdom insufficiency of agricultural researches initial data in the format of «mean SD», without indicating replication number is discussed. To imitate three replications of experiments the initial mean value variety of the studied indicator is suggested to complete with two multitudes: array of «means plus standard deviation» and array of «means minus standard deviation» and in case of six-fold replication, it is necessary to add twice this series. The process stages of generalized linear model are discussed, starting from disperse analysis, estimating statistical significance of means difference generally and finishing with calculation of confidence intervals of their differences and differentiating the homogeneous subgroups of means, within which means are statistically imperceptible. It is proved that the useful effect of the procedure of generalized linear model is estimation of effects of the experiment variants. In terms of the experiment of influence of probiotic “Olin” on live weight of broiler chickens, the discussion of the model parameters and their confidence intervals is held.