REVIEW OF DATA-ANALYSIS METHODS FOR SEED-GERMINATION

Numerous procedures for data analysis of seed germination responses are scattered throughout the literature. Here an attempt was made to summarize the existing methods, to compare the information they provide and to examine their strengths and weaknesses. The methods reviewed include the percent germination, germination index, coefficient of velocity, median response time, probit analysis, curve-fitting of cumulative germination, heat sums, survival analysis with life tables, logistic regression, proportional hazards regression and accelerated failure time analysis. Comparisons among these methods are discussed and illustrated with data from germination responses in tomato (Lycopersicon esculentum Mill.). Seed germination involves not only qualitative responses of individuals, but also population responses which are distributed over time. Standard analysis of variance or regression methods are appropriate for a data analysis where germination of all viable seed is observed, but they are inappropriate when some viable seeds fail to germinate. Such missing (censored) data complicates statistical analysis and subsequent interpretation. Germination tests should be designed to determine the nature of censored responses, which can be subsequently accommodated by several statistical procedures referred to as survival analysis.