Go To Contingency Table Data Sheet

Animals seeking food in natural environments face a tremendous number of choices. Natural selection is expected to favor efficient foraging patterns; this expectation forms the basis of a large body of optimal foraging theory and associated tests. Bees have been widely used in such studies. We will investigate several types of choices that bees encounter frequently while feeding at flowers, and try to establish the decision rules that the bees have developed to deal with these choices in nature.

Procedure

We will be presenting free-foraging bees with choice tests in the field, in a situation analogous to the T-maze (or Y-maze) design used in many lab experiments in behavioral psychology. We use a stick with two short pieces of plastic tubing at the end to hold two different flowers or inflorescences. By extending the stick toward a forager in such a way that the two flowers are equidistant from the animal, and observing which one the animal chooses to visit, one can gradually build up a large sample of decisions.

We can easily record two sorts of data. Most simply, we can just record the number of times that bees chose each of the two choices. We can analyze those data by a chi-squared goodness-of-fit test, against the null hypothesis that we expect equal numbers of visits to the two flowers. A significant test result indicates a feeding preference. We can also record data on the behavior of the bee after it lands on an inflorescence. These data could include things like the time spent or the number of florets probed. We will collect data (separately) for two taxa of bees, honey bees (Apis mellifera) and bumble bees (Bombus spp.). We may encounter several species of Bombus, but we will not try to distinguish them. We’ll demonstrate how to recognize the bees, carry out the interviews, and record the data.

We will consider four general questions:

1. Flower constancy. In many situations, bees become "constant" on a particular flower species, i.e., they visit it to the exclusion of other species that they encounter. This has obvious implications for the successful cross-pollination of plants. Constancy is seldom absolute, especially when two plants are very similar; also, different types of bees are thought to differ in their levels of constancy. We will determine whether bumble bees and honey bees show constancy to goldenrod species. (Depending on the availability of flowers, you will probably conduct this experiment on the first class field trip, as an independent exercise.)

Design. Find bees feeding on one goldenrod (Solidago) species (say, S. rugosa) in an area where another species (e.g., S. graminifolia) is nearby or intermingled. Give these bees a choice between inflorescences of S. rugosa (constant) and S. graminifolia (inconstant). Try to present about the same number of flower heads of each species. Ideally, do the reverse experiment with the same inflorescences (e.g., test bees from S. graminifolia, also. (Why?)) Record data separately for honey and bumble bees.

2. Floral advertising. The showy parts of flowers (usually petals) are thought to function primarily in pollinator attraction. In some composite heads, there seems to have been an evolved division of labor between the central disk florets and the peripheral ray florets. Although in some cases the ray florets retain sexual function, in many cases they are either male-sterile or completely sterile. They also often have large, brightly-colored, petal-like extensions. We will test the hypothesis that the ray florets of Centaurea maculosa play a role in attraction.

Design. Cut two heads of C. maculosa, taking them from the same plant and matching them as nearly as possible for age and size (i.e., number of florets). With your fingers, remove the ray florets only from one head; leave the other intact. Again, record data separately for the two types of bee.

3. Nectar rewards. Many animal-pollinated flowers produce nectar as a reward, and pollinators are known to modify their foraging behaviors in several ways in response to variations in the nectar they find (or don't find) in flowers they investigate. We are interested in several questions here: (i) Can bees detect the difference between nectar-rich and nectar-poor heads of C. maculosa remotely, i.e., without visiting and probing the flowers? (ii) Do bees behave differently on rich vs. poor heads?--specifically, do they probe different numbers of flowers? and (iii) Do bees that have just encountered rich heads modify their subsequent foraging moves, relative to those that have just visited poor heads?

Design. We will have bagged a number of flower heads to prevent visitation for 24 hrs previous to the experiment. Pair one of these heads with a head of similar size and age which you have just seen a bee visit. Record not only which head receives the first visit by a bee, but also the number of probes that the bee makes with its tongue. This will require close observation. Then also record whether or not the bee then moves over to visit the second head of the pair, and if it does, record the number of probes there as well.

4. Pheromone attraction. BeeScent is a commercial product designed to attract honey bees to crop plants that need pollination. It is presumably an artificially synthesized version of mandibular gland pheromone. Does it work? If it works on honey bees, does it also work on bumble bees?

Design. Work with a small group of classmates. What can you come up with?

Analysis

Because our design offers a pairwise choice, most of our data (Experiments 2, 3i, and 4) should be tested against the null hypothesis that bees pick each of the choices the same number of times, i.e., the expected value for the number of choices to either inflorescence would be the total number of choices observed, divided by two. If our observed results match this null expectation exactly, we know that the bees showed no preference. A perfect match is very unlikely, however. Assuming that we do not see a perfect match of observed and expected values, how do we judge whether the deviation is due to a real preference or to mere chance? Consider: If each bee were mentally flipping a coin to decide which inflorescence to visit, there would be no real preference, but simply by chance, there might be a excess of "heads" or of "tails." We need a statistical test to tell us how likely it is that we could have obtained our observed results if the null hypothesis of "no preference" is true. If the likelihood of this is very small, we can reject that null hypothesis and conclude that the bees are really making non-random choices.

Because our data are frequencies, or counts of the number of times we observe a particular outcome in a series of independent trials, we can use the chi-squared test statistic to perform a goodness-of-fit test on the data from each set of pairwise trials.

The full analysis of constancy (Experiment 1) is slightly more complicated than a pairwise choice test, because we must consider not only which of the two species the bee visited in the bouquet but also which of the two species the bee was coming from in the field before making its choice. Here, the null hypothesis is that the plant species that the bee chose in the bouquet does not depend on the species that the bee was coming from in the field. To do this test of independence, we use the chi-squared statistic again, but this time to test a 2 x 2 contingency table. A similar approach can evaluate Experiment 3iii.

Finally, to analyze Experiment 3ii, we can take several approaches. Although the numbers of probes are integer values, strictly speaking, the will have too large a range to consider each probe number a separate category for an analysis of frequencies by chi-squared. We can either create a small number of categories (e.g., 0-5 probes, 5-10 probes, etc.), or we can simply treat the numbers of probes as continuous variables and use a t-test or a non-parametric equivalent such as a Mann-Whitney U-test. The latter is probably preferable because the numbers of probes are unlikely to meet the assumptions of normality.

Thought questions

1. To stay constant on flower species A, a bee presumably will sometimes encounter but not visit another rewarding species B. Wouldn't it be more energetically efficient, in terms of food gain/travel time, for the bee to stop an feed at B rather than to fly past it? Can constancy produce energy-efficient foraging? If not, why would we ever see animals doing it?

2. Why would a bee consistently prefer to visit one flower species over another?

3. If bees show preferences for certain flowers, do you think they are born with those preferences or acquire them through experience? Outline a set of experiments that would begin to answer this question.

4. If sprays of "pheromone" really produce a response in honey bees, would you expect a similar response in bumble bees? Why or why not? What sorts of functions have pheromone systems evolved to perform?

5. Experiment 3 is designed to show whether bees make predictions about the state of rewards they will find in the next flower they visit, based on their success or failure at finding rewards in the flower they just visited. Explain how you would interpret different possible experimental outcomes to draw conclusions about whether bees are in fact making such predictions or not.

References.