ABSTRACT
This chapter examines methods for testing the significance of any observed, increasing trend in the response over the ordered levels of the independent variable, x. the test of trend, while significant, is based upon a model in which the functional relationship between the response variable, number of offspring, and the predictor variable, nitrofen concentration, is misspecified. A powerful approach for trend testing with discrete data was proposed by Cochran and Armitage – and, in a more general form, by Yates – and is known as the Cochran–Armitage trend test. Trend testing with the generalized estimating equations statistic is known to exhibit very stable false positive error, and also good power to detect true increases in trend with overdispersed proportions, at least in large samples. Emphasizing simple tests for trend, consideration is given to both continuous and discrete data structures, including overdispersed settings for the discrete case.
