A question of data quality—Testing pollination syndromes in Balsaminaceae

Pollination syndromes and their predictive power regarding actual plant-animal interactions have been controversially discussed in the past. We investigate pollination syndromes in Balsaminaceae, utilizing quantitative respectively categorical data sets of flower morphometry, signal and reward traits for 86 species to test for the effect of different types of data on the test patterns retrieved. Cluster Analyses of the floral traits are used in combination with independent pollinator observations. Based on quantitative data we retrieve seven clusters, six of them corresponding to plausible pollination syndromes and one additional, well-supported cluster comprising highly divergent floral architectures. This latter cluster represents a non-syndrome of flowers not segregated by the specific data set here used. Conversely, using categorical data we obtained only a rudimentary resolution of pollination syndromes, in line with several earlier studies. The results underscore that the use of functional, exactly quanitified trait data has the power to retrieve pollination syndromes circumscribed by the specific data used. Data quality can, however, not be replaced by sheer data volume. With this caveat, it is possible to identify pollination syndromes from large datasets and to reliably extrapolate them for taxa for which direct observations are unavailable.