Several surveys have documented an association among species between habitat type and seed mass, suggesting that habitat attributes impose a direct selective force on seed mass. Previous comparative surveys, however, have not controlled for the statistical effects of shared phylogenetic history (at the genus or family level) and life form when evaluating the relationship between habitat and seed mass. This study of the Indiana Dunes angiosperm flora provides statistical 'control' of genus and family membership by: (i) partitioning out the statistical effect of genus membership prior to measuring the effect of habitat on seed mass, and (ii) seeking an association between habitat and seed mass within eight genera (206 species) and ten families (366 species). To measure the associations between ecological factors, taxonomic membership and seed mass, I examined life form, phenological schedules and seed mass among species in 8 genera distributed among 13 habitat types (assigned to 1 of 4 categories of inferred water and light availability). One-way ANOVAS indicated that genus, life form, habitat, water/light category, the onset of flowering and the duration of flowering accounted for 71%, 51%, 10%, 4%, 14% and 14% of the variance in seed mass, respectively. However, multi-factor ANOVAS measured the variance in seed mass accounted for by each variable independently of the others: only genus explained a significant proportion (11%). Genus membership is strongly associated with the other ecological factors, accounting for the difference between one-way and multi-factor ANOVAS. Within the ecologically widespread genera and families of this study, there was no significant association between water/light category and seed mass, even though this association can be detected across taxa. Among congeners and confamilials, interspecific variation in seed mass (measured as the coefficient of variation) was as high within habitat types as among them, suggesting that habitats do not provide upper limits to the range of seed mass exhibited by the species within them. A previous study of 648 Indiana Dune species showed that species segregate among habitats on the basis of seed size; large-seeded species tend to occupy closed habitats and small-seeded species tend to inhabit open habitats. This segregation creates the ecologically meaningful observation that low-light habitats support larger-seeded species than high-light habitats, even though this pattern cannot be detected independently of taxonomic membership. Generalist taxa may occupy a wide range of habitats for reasons other than their seed size. If this is a common feature of ecological generalists, it may not be possible to detect an association between habitat and seed size independently of taxonomic membership. © 1990 Chapman and Hall Ltd.
