Online Ring Testing to Elicit the Cause of Underdispersion in Seed Germination Tests

Abstract

According to International Rules for Seed Testing, a germination test is performed with 400 seeds, subdivided into four replicates of 100 seeds. The empirical variance between the four replicate results was found by many authors to be smaller than random sampling variation. Hypotheses for sources for this underdispersion were developed, but no experimental proof was possible, as the population's true germination value is unknown. To overcome this obstacle, we performed two online ring tests with computer‐generated seedling images. Seedlings had to be classified as normal, if the shoot/root ratio was between 0.5 and 2.0, otherwise as abnormal. Results showed significantly smaller variances between empirical replicate results than between true replicate values, confirming underdispersion. Rates of false normal and false abnormal classification of seedlings were 19.5 and 3.5%, respectively. Error rates are different between analysts and depend on the true replicate value. In replicates with many abnormal seedlings, the false normal rate is higher and the false abnormal rate is lower than in samples with few abnormal seedlings. This is the source of underdispersion and a direct result of decisions made by seed analysts. Pooling of replicate results of all analysts and recombining showed that underdispersion can be reduced significantly when each replicate is tested by a different analyst. Hence, replicates should be blinded and every replicate evaluated by another analyst.