The Berg hypothesis proposes that specialized-flower traits experience stronger stabilizing selection than non-floral structures and predicts that variation in specialized-flower traits will be mostly uncorrelated with variation in non-floral traits. Similarly, adaptive-accuracy theory predicts lower variation (as a proportion of the mean) in floral traits than in non-floral ones. Both hypotheses can be extended to comparisons between floral traits, where different parts of the flower can be expected to experience different strengths of stabilizing selection, resulting in contrasting patterns of variation.

Armbruster and Wege test these ideas by analysing variation/covariation in those floral traits influencing the location of pollen placement on, and stigma contact with, pollinators (‘pollination-mechanics traits’, PMTs) in relation to variation/covariation in non-floral traits and floral traits not directly involved in the mechanics of pollination. The prediction was that PMTs are canalized (buffered against genetic and environmental variation) relative to attraction traits, as manifested in lower variances and modular independence. The authors measured floral and inflorescence structures of ten species of triggerplants (Stylidium, Stylidiaceae) in south-western Australia, and analysed the data using multivariate and bivariate approaches to detect modular structure of floral and non-floral traits and assess evidence for canalization of PMTs.
Only six of the ten species had PMTs with smaller correlation coefficients than attraction traits, in contrast to the Berg expectation. However, allometric and variance patterns were generally consistent with the predictions of an extended Berg hypothesis and adaptive accuracy. There was modular separation of most floral traits from non-floral traits and clear intra-floral modular structure. PMTs showed lower proportional variation and shallower allometric slopes than pollinator-attraction traits in nine and eight, respectively, of ten species.