One can fall into various misinterpretations by aggregating heterogeneous sets of individuals and therefore obtaining spurious correlations. Here we are dealing with specific, pollination-related manifestations of a general statistical problem. If bees do vary substantially in performance, but are treated statistically as equivalent replicates, the interbee variation can pop out in a variety of spurious relationships. For example, Pyke (1978) hypothesized that optimally foraging bees ought to show area-restricted search, i.e., they should fly shorter distances between plants after they have just received larger than average rewards. Because it is hard to know how much nectar a bee has received, Pyke and many others have substituted the time spent at a flower as a surrogate variable for the amount of reward received. This is reasonable, as it takes more time to extract more nectar (Harder 1986; Kato 1988). Making this substitution, one then tests for area-restricted search by testing for positive correlation between the time spent at one flower and the distance flown to the next. Pyke found this pattern. Although this procedure would be trustworthy for observations of a single bee, suppose that some bees in a population -say, those with tattered wings - work all flowers more slowly, and always tend to fly shorter distances. If one then combines data from fast and slow bees, one could obtain the expected positive correlation, even if no individual bee shows area-restricted behavior (Thomson et al. 1982).
Analogous difficulties attend field studies of flower constancy. Here, an attractive hypothesis is that a flower visitor should be more willing to switch to another species of flower after having received little reward. This flexibility would allow individuals to track the relative values of different resources and concentrate on the best ones. If flower-handling time is used as a surrogate for reward, if interbee variation in constancy is correlated with variation in working speed, and if data are pooled across bees, however, spurious correlation can cause the hypothesis to be accepted when it should be rejected, or vice versa.
Problems of this sort arose in a study of constancy in many unmarked bumble bees that were followed for as long as possible as they foraged freely in a meadow with several suitable flower species (Chittka et al. 1997; L. Chittka, unpublished data). The authors initially classified flower-handling times into two categories, either above or below the grand median for all bees. In this data set, bees were significantly more likely to switch plant species if their last (several) visits had been shorter than the median, and more likely to stay constant if their last visits had been longer than the median. This seems consistent with the hypothesis that bees switch when they are dissatisfied, but when each bee's visit times were re-scaled by the median for that bee's bout (rather than the grand median), the effect disappeared. Further exploration of the data suggested that the heterogeneity causing the spurious correlation arose not so much from interbee variation as from temporal variation. In the morning, all bees handled flowers slowly, presumably because nectar levels were high, and all bees tended to be constant. In the afternoon, visits were shorter and constancy dropped overall, so the relationship between visit length and subsequent constancy could not be clearly attributed to short-term behavioral flexibility. In fact, Chittka et al. (1997) resurrected the flexibility hypothesis; they reanalyzed the data within bouts, considering not just the upper and lower halves of the visit times but the upper and lower quartiles. Then, bees were more likely to switch following very short visits, and more likely to be constant following very long visits.
This example illustrates not just the danger of spurious correlation but also a reasonable way of handling existing data to avoid problems. Although marking animals is not always feasible, more trustworthy results will be obtained by restricting analyses to comparisons within bees, as well as considering other cryptic sources of heterogeneity (such as time of day). One investigation of flower constancy that apparently did not include such precautions is a study of skippers by Goulson et al. (1997). They used exactly the procedure initially tried by Chittka et al. (1997), except that they used means rather than medians for dividing the data, and they reached the same initial conclusion. It might be worthwhile to analyze their data further, along the lines of Chittka et al. (1997), assuming that the bout lengths are long enough.
As Galen & Plowright (1985) showed, bumble bees that forage for nectar on Epilobium angustifolium visit the vertical inflorescences differently from those that seek pollen from the same plant. These authors interpreted their results in terms of reward maximization criteria, as if members of a group of equivalent bees first made a decision to specialize on pollen or nectar, then adjusted their movements accordingly. One would also like to know, however, whether parasitic infections also played a role in the food-type decisions; if so, then the population might be more profitably viewed as comprising heterogeneous groups of infected and uninfected individuals with different behaviors.
From the plant's point of view, it is clear that the adoption of pollen- or nectar-collecting behavior by a visitor can greatly change the fitness value of that visitor to the host-plant (Galen & Plowright 1985; Shykoff & Schmid-Hempel 1991; Wilson & Thomson 1991).
Ignoring pollinator individuality can lead not only to spurious correlations, it can hamper insight regarding the adaptive problems that animals or plants are "trying" to solve. Knowledge ofindividuals can lead one to pose questions that would otherwise go unasked. For example, researchers concerned with pollinators' responses to variation in plant phenotypes tend to assume that the plant's visitors are influenced only by the characteristics, such as inflorescence size, that the plant presents at the moment. However, the behavior of bees that return frequently to particular plants might also be sensitive to qualities that the plant displayed previously but no longer does. For example, Aralia hispida plants change sex from male to female phases several times during a flowering season. When floral rewards were manipulated in male-phase inflorescences (Thomson 1988), bumble bee visitation increased to the richer inflorescences. When all of the variable male-phase inflorescences were replaced with uniform female ones, simulating the natural sex change, the bees preferentially visited female inflorescences that were located where the richer males had been. This result highlighted an ambiguity in interpreting selection on floral displays in terms of sex allocation theory: nectar secreted by a flower in male phase can increase the visitation rate to that flower in female phase. Should the cost of producing that nectar be considered a male or a female cost.?
Even without special subtleties due to sex roles, early flowers can influence visitation rates to later flowers if pollinators show "trapline holdover," as bumble bees sometimes do (Thomson 1988, 1996). This effect could provide adaptive explanations for some aspects of floral biology, such as the tendency of many plants to burst into bloom with many flowers, then to taper off flower production. Here, the early flowers may benefit the plant not only through their own gametes but also by recruiting a faithful set of individual pollinators that will continue to serve the plant through its blooming period (cf. Thomson 1988). Without knowing the site fidelity of individual pollinators, one cannot fully interpret how pollinator-based selection might act on inflorescence architecture.
We have long had indications that bumble bees scent-mark flowers and respond to those marks (e.g., Cameron 1981; Kato 1988; Schmitt & Bertsch 1990), but this evidence has not yet been well incorporated into the thinking of many who study foraging primarily from an energetic point of view. The energetic viewpoint has interpreted bees' decisions at flowers as being driven mostly by direct assessment of rewards gained at a blossom, rather than indirect olfactory assessment of recent visitation. This is partly because the evidence for scent marks has been mostly indirect and partly because the interpretation has been somewhat confusing. Schmitt & Bertsch (1990) review the evidence up to that date for bumble bees and honeybees; they indicate that some chemicals deposited on flowers may serve as attractants that denote rewarding flowers, while others, probably more volatile and short-lived, may serve as repellents that signal bees not to revisit flowers that have recently been drained. Schmitt & Bertsch interpret their results as strong evidence for an attractant role. Conversely, Giurfa & Nunez (1992) found evidence that marks recently left by honeybee foragers act as repellents. More recently, Goulson et al. (1998) have reported field evidence from bumble bees for a repellent role, a finding reinforced by experimental application of extracts from bee tarsal glands to flowers (Stout et al. 1998). To date, however, it is not clear if bees use more than a single scent to mark flowers, nor whether scent-marking is an active process (Chittka et al. 1999). It is equally possible that tarsal secretions are used for adherence of bee feet to flowers, and are used as scent marks only as an epiphenomenon: bees might use the scent marks as repellents if the flowers are known to refill slowly, and as attractant if they remember the flowers as having high refill rates.
Our goal in considering scent marks in this chapter is not to resolve controversies but to show how an individualistic perspective can help clarify how these marks should be interpreted. If one adopts an adaptationist viewpoint of bees as optimal foragers that search widely for food, scent-marking is hard to understand. Of course, it is easy to see that a short-lived repellent mark might be useful in helping an individual avoid revisiting flowers that it has just probed, but it is harder to see how it could be adaptive to leave long-lived attractive marks on rewarding flowers. It would seem to require some special conditions. First, there must be an expectation that the bee who does the marking will return in time to benefit from the mark. This condition is easily met if bees use small foraging areas. Second, and more onerous, the mark must be expected to be of more benefit to the bee who left it than to other bees that may also detect it. It will do an individual little good to flag a rich resource if the primary result is to help competing bees exploit that resource. This paradox could be explained by kin selection if most of the visitors to a plant were sisters. In honeybees, which might combine scent-marking of flowers with site-specific dance information in the hive (von Frisch 1967), this may sometimes be the case. In bumble bees, however, these conditions probably do not apply: they lack a site-specific recruitment system (Dornhaus & Chittka 1999); workers range too far, workers per colony are too few, and colony densities are too high (Cumber 1953; Harder 1986), for sibling encounters to be frequent.
On the other hand, if a traplining individual is making a substantial fraction of the visits to a plant (Fig. 10.1), that bee may reap enough benefits from attractive scent marks to offset the possible advantage given to competitors. Some analogous mechanism might help explain a puzzling observation by Williams & Thomson (1998) in the 1994 data mentioned above. Modeling nectar production and removal with some simple assumptions, they estimated that the bees that visited the focal plant most often - i.e., the regular trapliners - gained more reward per plant visit than did the casual visitors that arrived less often. Interestingly, the trapliners achieved their edge not by arriving at times when the plant had more reward overall, but rather by being better at selecting the flowers that had not been visited recently by others. Positing scent cues does not in itself dispel the puzzle, for the casual visitors presumably have as much access to scent cues as the trapliners do. Conceivably, the trapliners simply pay more attention to these cues for some reason; an interesting alternative is that bees can leave some private cues that are not accessible to others. Individual-specific trail marks are known in some species of ants (Maschwitz et al. 1986). In laboratory tests, scent marks left by bees on artificial flowers have also been shown to be more efficient in repelling the individual which left them than other bees (Giurfa 1993), but whether this effect holds up in the field remains to be shown. If it does, trapliners that return at regular intervals might be able to make the best use of marks left by themselves and those left by other bees.
Was this article helpful?