Population studies with wheat: I. Sampling

Abstract

The design and field technique of two large scale experiments, laid down to investigate the problems of sampling and “propinquity”, are described in detail. These experiments were designed so that the plant number, stem number, ear number, straw weight and grain weight for 7200 lengths of 6 in. of drill row, together with the position of each observation, might be obtained.The lowest sampling error, expressed as a percentage of the mean, is obtained by using the smallest sampling unit, but the large number of sampling units of this size that would have to be taken make it impracticable. The optimum sampling unit consists of 6 ft. of drill row, taken as 3 ft. of two adjacent rows. The size of the sampling unit is of greater importance than the shape in determining its accuracy, and little s lost by using sampling units of 18 in. of five adjacent rows.The nature of the observation affects the degree of sampling. In the experiments examined the plant number, stem number and ear number require very similar sampling, but grain weight requires a sample which is about twice as large as that required by the other observations.The percentage of the plot that must be taken as a sample (the sampling percentage) to obtain any given accuracy is related to the size of plot: the larger the plot, the lower the sampling percentage necessary. To obtain an error of 5% of the mean the sampling percentage for grain weight is approximately 5% in 1/20 acre plots, 15% in 1/100 acre plots and 43% in 1/600 acre plots if the plots are not subdivided. These percentages are similar to those calculated from various randomized block experiments in Cambridge.