AGREEMENT ON BEST SITE - Honeybee Democracy - Thomas D. Seeley

Honeybee Democracy - Thomas D. Seeley (2010)


Love quarrels oft

in pleasing concord end.

—John Milton, Samson Agonistes, 1671

In the previous chapter we saw how the quarrels among scout bees, like those among human lovers, “oft in…concord end.” Now we will see if the agreements reached by the bees are “pleasing.” That is, when the dancing bees reach a consensus about their new homesite, are they apt to have chosen the best site? The answer is yes! But before looking at the evidence that a swarm usually chooses the best of the many candidate sites the scouts discover, let’s first consider the structure of the choice problem faced by the house-hunting bees. This will sharpen our appreciation of a honeybee swarm as a democratic decision-making body.

A swarm of bees selecting a nest cavity faces a decision-making problem akin to what a person faces when choosing a place to live. This is a complex choice problem, for it is one where there are numerous alternative solutions (e.g., houses or apartments) and each one has many attributes (e.g., neighborhood, number of bedrooms, and such). And as is true for all decision-making problems, finding a good solution is a twofold process: first identify the possible alternatives, then choose among them. In an ideal world, the decision maker would be able to learn about all the alternatives and all the attributes of each, calculate the value of each alternative in light of all its attributes, and rationally choose the one with the highest value. Doing all these things will produce optimal decision making. In the real world, however, truly optimal decision making rarely happens because decision makers must pay costs in time, energy, and other resources to acquire and process information, and these costs usually preclude making the decision using all the relevant information. For example, someone hunting for an apartment in a large city would have to expend excessive time, money, and mental effort to survey the entire market of available rental properties, evaluate them all, and make the perfect choice.

Given that decision makers do not possess unlimited time, boundless resources, and infinite powers of reason, psychologists and economists now recognize that real world decision making—often called bounded rationality—relies on simplified mechanisms of choice, termed heuristics. These generally involve reducing either the breadth or the depth, or both, of the decision maker’s consideration of the alternatives. For example, the decision-making heuristic called satisficing reduces the breadth of the search for alternatives. It takes the shortcut of setting an acceptance threshold and ending the search for alternatives as soon as one is encountered that exceeds this threshold. Imagine, for example, someone who has just moved to a distant city, is hunting for an apartment, and can’t search the housing market broadly because she needs to start work immediately at her new job. If she simply picks the first apartment that is acceptable, which almost certainly is not the best one available, then she will have used the satisficing heuristic. Another decision-making heuristic, called elimination by aspects, reduces the depth of the decision-making task. Someone using this heuristic to find an apartment first decides what attribute is most important (say, commuting distance), sets an acceptance limit (say, no more than a 20-minute commute), and eliminates all the apartments that exceed the limit. The process is then repeated, attribute by attribute (no more than $1,000 per month; a park for jogging within five blocks) until either a choice is made or the set of possibilities is narrowed sufficiently to switch over to a thorough evaluation of the finalists. This person probably won’t choose the apartment that would be best overall—she will not consider an apartment that has a 22-minute commute even if it has a low rent and a beautiful park nearby—but she has certainly diminished the time, expense, and mental effort needed to find a place to live.

Given that humans and other animals usually make decisions by drawing on a toolbox of heuristics, it is remarkable that a honeybee swarm does not use these shortcut methods of decision making and instead selects its new living quarters by taking a broad and deep look at the bee housing market. As we have seen in chapter 4, a swarm makes its decision only after its scout bees have discovered numerous alternative nest sites and have performed a multifaceted inspection of each site. In the full-size, natural swarms studied by Martin Lindauer, the mean number of candidate sites reported by scouts on a swarm was 24 (range 13 to 34), and even in the small, artificial swarms studied by Susannah Buhrman and me, the mean number was 10 (range 5 to 13). And as we have seen in chapter 3, each candidate site is evaluated with respect to at least six attributes (e.g., cavity volume, entrance height, and entrance size). Thus a honeybee swarm pursues an unusually sophisticated strategy of decision making, one that involves nearly all of the information relevant to the problem of choosing the best place to build its new nest. (Note: even a honeybee swarm is not all knowing, for even though it sends out hundreds of scouts to search for candidate nest cavities, these bees probably don’t discover all the available dwelling places.) A swarm is able to be so thorough in choosing its home because its democratic organization enables it to harness the power of many individuals working together to perform collectively the two fundamental parts of the decision-making process: acquiring information about the alternatives and processing this information to make a choice. We will now look at the evidence that honeybee democracy does indeed achieve nearly optimal decision making.

Best of N?

To investigate whether a swarm’s scouts usually reach agreement on the best available site, I needed to go beyond observing their dances for natural nest sites on the swarm cluster. Specifically, I needed to present them with an array of artificial nest sites that differed in quality, and I needed to do so in a location lacking natural nest sites so that the scouts would focus their attention on my artificial dwelling places. With such a setup, I could find out if a swarm’s scout bees consistently choose the best nest site out of a set of alternatives—biologists refer to this as solving a “best-of-N” choice problem—or if they don’t actually achieve such optimal decision making.

One can imagine various ways in which a swarm’s choice of its new home could fall short of perfection. As we saw in chapter 4, a swarm’s scout bees do not enter all the candidate sites into their debate at the same time, but instead do so over many hours or even a few days. If the best site happens to be presented late in the debate, its supporters might have difficulty overtaking those advocating a poorer site that was presented early on and has gained much support. Or even if the best alternative is entered into the debate at the start, it could lose out if the bees advertising it fail to tout its high quality. (How scout bees indicate a site’s quality in their waggle dances will be discussed in chapter 6.) Yet another way that the best site might lose out in the scout bees’ debate, even if it is reported promptly and correctly, is if this site is especially hard to find by recruits, maybe because it is far away or because it has an obscure entrance opening. Either situation will hamper the mustering of support for the best site. Given the many situations that seem like they could cause swarms to perform suboptimally in choosing a home, I wondered if swarms really are so skilled at solving the best-of-N choice problem. To find out, I needed to test their decision-making skills with controlled experiments.

Mediocrity in 15 Liters

To perform these experiments, I returned in the summer of 1997 to Appledore Island, in the Gulf of Maine, where some 20 years before I had good luck in getting swarms to be interested in artificial nest sites. In the intervening years, I had studied the bees mainly at the Cranberry Lake Biological Station, deep in the forested Adirondack Mountains of northern New York State, where flowers are sparse and bees are eager to forage from artificial food sources. Studying the bees in the north woods had been thrilling; each summer my students and I had uncovered secrets about the beautiful inner workings of a honeybee colony, particularly those that enable a colony to gather its food efficiently. Also, I had fallen in love with swimming in the lake’s crystalline water, watching the northern lights glow in the midnight sky, and falling asleep to the haunting calls of loons. But by 1997 I was ready to return to the brilliant sunshine, ferocious gulls, lush poison ivy, and invigorating salty air of Appledore Island.


My first goal was to figure out how to make an artificial nest site that was acceptable but not ideal to the bees. If I could solve this problem, then I could test whether or not swarms achieve optimal decision making. The design of the test called for presenting swarms (one at a time) with an array of five nest boxes that would offer four acceptable homesites and one ideal homesite, and then seeing how reliably swarms would select the best of the five nest boxes. From my studies in the mid-1970s of the bees’ nest-site preferences, I knew that bees prefer nesting cavities that have a large volume (40 liters) and a small entrance (15 square centimeters), so I decided to see if I could dilute the goodness of a nest box by decreasing its cavity volume or increasing its entrance area. Figure 5.1 shows the design of the nest boxes that I built. Each one had a cavity volume of 40 liters, but this could be reduced to 20, 15, or 10 liters by placing an inner wall in the appropriate location, as shown. Similarly, I could enlarge each box’s entrance from 15 cm2 to 30 or 60 cm2 by replacing one entrance reducer with another. It was essential that the nest boxes differed only in cavity volume or entrance area, so I positioned each nest box inside an open-sided shelter (fig. 5.2). These shelters all faced the same direction so all five nest boxes had identical exposures to wind, sun, rain, and…gull poop.


In early August, I loaded my pickup truck in Ithaca with five nest boxes, five shelters, the swarm stand I had used for video recording the scout bees’ debates, and three hives of bees for making artificial swarms. After driving to Portsmouth, New Hampshire, I loaded my equipment on the R/V John M. Kingsbury, the workhorse research vessel of the Shoals Marine Laboratory. It would ferry me and my 60,000 “co-workers” from the dock in Portsmouth, down the Piscataqua River, and out to the cluster of offshore islands known as the Isles of Shoals, of which the 96-acre Appledore is the largest. My 13-year-old nephew, Ethan Wolfson-Seeley, had joined me as research assistant. Soon we were off. Standing in the brilliant sunshine, drinking in the beauty of coastal New England, I felt exhilarated to be returning to one of my favorite outdoor haunts, where I had made some of my first scientific discoveries.

But I also felt slightly apprehensive to be returning to experiments with swarms, which I remembered were extremely difficult, even on Appledore Island. I had heard that Rodney Sullivan, my lobster fisherman friend, had left the island and sold his cottage. Would the new owners allow me to screen off their chimney to deter my scout bees? I also knew that over the past 20 years the Shoals Marine Laboratory had built several new dormitories and laboratories. Would these new buildings contain attractive homesites for the bees? And I wondered if I had designed my experimental nest boxes correctly, so that they could be tuned to the right settings of cavity volume and entrance area to produce a mediocre but still acceptable nest site. Would these nest boxes work? I soon stopped worrying, however, reminding myself that I’ve always made progress in my studies whenever I’ve watched the bees closely, paid close attention to unexpected results, and treated “failed” attempts to reach a goal as fingerposts indicating a better way to go forward. Certainly the remote setting of Appledore Island, 640 kilometers (400 miles) from Cornell University and 10 kilometers (6 miles) out in the Atlantic Ocean, would give me a perfect opportunity to focus my attention on the bees.

In a few days, Ethan and I had set up a swarm on the porch of one of the laboratory buildings and had placed two nest boxes in grassy sites on the north half of the island, both of them 250 meters (820 feet) from the swarm but in slightly different directions (sites A and B in fig. 5.3). To help gain the scout bees’ interest, the nest boxes were set up with a large cavity volume (40 liters) and a small entrance opening (15 cm ). Already I had introduced myself to the new owners of the Sullivan cottage (from Massachusetts and without shotgun), had explained why I wanted to put a screen over the top of their chimney, and had done so with their blessing. Now we sat patiently by the swarm, watching for bees performing waggle dances on the swarm’s surface, to see what the scout bees would report. All bees announcing one of the two nest boxes would be left alone, but any bee indicating some other site would be plucked off with forceps, dropped into a small cage, and later put in a freezer. This censorship of the scout bees’ communications turned out to be critical to our success. From time to time a bee would appear on the swarm dancing excitedly for a “rogue” site, and if we did not remove her quickly there would soon be an unstoppable escalation of interest in the distracting site as bees recruited to the site would come back and recruit still more bees there. Such snowballing of the scout bees’ interest in an unintended site actually happened three times that summer. In two cases, we managed to find the place of interest by reading the bees’ dances to determine the direction and distance to the site they were excitedly advertising, plotting its estimated location on a map of the island, and then searching there for scout bees flying in and out of some small opening. One site was a space beneath a pile of old boards and the other was a small cave in a stone wall. I rendered both sites worthless to the bees by opening them up. The third time, however, our search-and-destroy operation failed even though we poked around for hours in the right general area, which was among the three old houses on the south shore of the island. Evidently, a scout bee had discovered first-rate living quarters somewhere in the evil-looking jungle of poison ivy behind these houses, a place that we didn’t dare explore. Because we could not eliminate this site, we could not extinguish the bees’ raging interest in it, so all we could do was remove the swarm with its errant scout bees and start over with a new bunch of bees.


Fortunately, all our other swarms focused their house-hunting efforts on our nest boxes, and in doing so they taught us how to make one into a mediocre but acceptable dwelling place. The first lesson we learned was that I had guessed wrong about doing so by enlarging the entrance opening to 30 or 60 cm2. If we gave a swarm of bees a 40-liter nest box with a 15-cm2 entrance, they showed great interest in the box, as indicated by a rapid buildup of scout bees at the box soon after its discovery. For example, on August 10, 1997, one such nest box was found shortly before 1:00 p.m., and by 2:30 p.m. there were more than 10 bees crawling and flying about outside this nest box. There could be no doubt that the scout bees had judged this box to be highly desirable and had recruited others to it. In fact, around 1:00 p.m., back at the swarm cluster, we had observed several bees advertising the box with vigorous waggle dances. But after we enlarged the entrance opening to 60 cm2 at 2:30 p.m., the number of bees outside the box plummeted, falling to just one or two bees by 3:00 p.m. This sudden abandonment of the box suggested that the scouts were no longer attracted to it. At 3:00 p.m. the entrance was reduced back to 15 cm2 and the number of scout bees outside the box shot up as before, reaching more than 12 bees by 4:30 p.m. But after the entrance was enlarged again to 60 cm2 at 4:30 p.m., the counts of the scout bees plummeted again, dropping to less than one bee by 6:00 p.m. The next day we observed the same pattern of strong buildup of bees outside the box when its entrance was 15 cm2 and a steep crash in their number after we enlarged the entrance to just 30 cm2. These results, confirmed by those obtained from a second swarm a few days later, taught us that scout bees judge a nest box with a 30 or 60 cm2 entrance opening to be a low-quality, probably even unacceptable, homesite. They also showed us how easy it was to conduct an opinion poll of a swarm’s scouts: simply count the bees outside each nest box (fig. 5.4).


We next tried to create a medium-quality nest site by reducing the cavity volume to something less than 40 liters. This approach worked well. Our first trial started with scout bees discovering both nest boxes late in the day on August 13, 1997. The next morning, we set the volume of one box at 40 liters and that of the other box at 15 liters; both boxes had the entrance opening set to 15 cm2. As is shown in figure 5.5, the number of scout bees outside the 40-liter box rose steadily throughout the morning and reached nine bees by early afternoon.


Meanwhile, the number outside the 15-liter box stayed low at just one or two bees. It was clear that the scout bees were treating the 40-liter box as a high-quality site. But were they treating the 15-liter box as a medium-quality site, that is, one not highly desirable but certainly acceptable? To see if the bees would accept the 15-liter box, at 12:30 p.m. we enlarged the entrance opening of the 40-liter box to 60 cm2 to render it unacceptable, and we watched to see if the bees would now accept the 15-liter box. They did! While the number of bees at the 40-liter box plunged, the number at the 15-liter box climbed to a high level and at 1:28 p.m. the swarm took off to fly to the 15-liter nest box. (The reason for the sharp drop in number of scouts at the chosen site shortly before swarm takeoff will be discussed in chapter 8.) Thus this first trial yielded evidence that we could present our bees on Appledore with a mediocre but acceptable dwelling place if we gave them one of our nest boxes with the volume set at 15 liters and the entrance set at 15 cm2.

Additional trials made with two other swarms produced results similar to those from the first swarm. When given a choice between two nest boxes with different volume settings, a swarm’s scout bees would build up much more strongly at a 40-liter box than at a 15-liter one so long as both boxes had a small (15 cm2) entrance opening. But when the 40-liter box was severely degraded by enlarging its entrance opening to 60 cm2, the scout bees would become numerous at the 15-liter box and eventually would accept this box for their future home.

Window on a Bee’s Mind

Further evidence that we had found the right formula for creating a mediocre but acceptable homesite came from observations made at the swarm cluster rather than at the nest boxes. At the swarm cluster, we could see scout bees performing dances simultaneously for the 40-liter and 15-liter nest boxes (when both had small entrances). We could also identify which nest box each dancing bee was advertising by noting the angle of the waggle runs in her dance, for we had carefully positioned the nest boxes so that their directions differed by 30° (see fig. 5.3). (We were most grateful to the bees for saving us the trouble of giving the scouts from the two sites different labels!) Now, it is well known that when a bee performs a waggle dance to recruit hive mates to a food source, she decides how strongly she should dance based on the desirability of her flower patch. For instance, a bee advertising flowers brimming with sweet nectar might perform a strong dance that contains 100 dance circuits and lasts for 200 seconds, whereas a bee reporting on a poorer nectar source might produce a rather weak dance that contains only 10 dance circuits and lasts just 20 seconds. This correlation between flower desirability and dance strength (number of dance circuits) means that the waggle dance provides us with a “window” on a bee’s mind, especially on her sense of the quality of what she is reporting to her hive mates.

Assuming that this window works for bees advertising nest sites as well as food sources, we decided to look through it to see how scout bees advertising our 40-liter and our 15-liter nest boxes judged the quality of each as a prospective home. We did so by video recording the dances performed side by side on a swarm by two groups of scouts, those reporting on our 40-liter nest box and those reporting on our 15-liter nest box. The fact that both nest boxes elicited dancing by the scout bees told us that both were of considerable interest to the bees. But even more telling was what we learned by carefully reviewing the video recordings and measuring the strength of each bee’s dance. We found that bees reporting the 40-liter box performed strong dances that on average contained about 35 circuits and lasted about 85 seconds, whereas the bees reporting the 15-liter box performed weaker dances that on average contained only about 14 circuits and lasted only about 45 seconds. These findings strongly support the conclusion that the bees judged our 15-liter nest box to be an acceptable but mediocre homesite. The indication of acceptability is that the scouts produced dances for the 15-liter nest box (we don’t expect scouts to advertise an unacceptable site), and the indication of mediocrity is that the scouts produced relatively weak dances for the 15-liter nest box.

Critical Experiment

On Appledore Island, during the sunny days of August 1997, I had learned from the bees how to tune my experimental nest boxes so that I could present a swarm with a choice among five possible homesites, four of them fixer-uppers and one a dream home. So at this point I was ready to present swarms with the best-of-5 choice problem, and I was extremely eager to do so, but alas I had to wait until the following summer to conduct this critical test of their decision-making skills. Fall semester classes start at Cornell in the last week of August, and each fall semester I’m part of a team that teaches a popular class in animal behavior, so I needed to get back to Ithaca to give my lectures. I also needed to set up the special glass-walled hives of bees that we use in the course to introduce students to the pleasures of watching and wondering about bee behavior.

In June of 1998, I returned to Appledore Island. With me was the smart and dedicated Cornell undergraduate student, Susannah Buhrman, who had helped me the summer before in documenting the scout bees’ debates. Our goal now was to test the decision-making skills of swarms by giving them the best-of-5 choice test. Administering this test required two people working as a team, one sitting at the swarm to eliminate any scouts performing dances for sites other than the nest boxes (fig. 5.6), and one circulating among the nest boxes to count the scouts visiting them. As is shown in figure 5.3, we set up the nest boxes in a fan-shaped array on the east side of the island, so that the boxes were about the same distance (approximately 250 meters) but in different directions (at least 15° apart) from the swarm. We began each trial of the experiment by arranging the inner walls inside the five nest boxes so that one offered a 40-liter nesting cavity and the others offered a 15-liter cavity. Next, we mounted the swarm to be tested on the swarm stand. Once the swarm had formed its cluster and scout bees had begun flying from the cluster, one of us started monitoring the scouts’ dances to remove reports of sites other than our five boxes, while the other person started checking the nest boxes, visiting each box every half hour and counting the bees there. We performed five trials of the experiment, each with a different swarm of bees and each with a different location for the excellent nest box. It should be noted that we did not change the location of the excellent nest box between trials by moving one excellent box around, rather we did so by leaving the five boxes in place and adjusting their volume settings. Thus in each trial a different box was given the 40-liter setting that made it the excellent option.

The full results of the five trials of this experiment are shown in figure 5.7. It shows for each trial how many scout bees were counted outside each nest box over the course of the trial. We can see that in all five trials the swarm’s scouts found all, or nearly all, five nest boxes, which means that each swarm had knowledge of most of the candidate sites. We can also see that in each trial the scouts did not find the nest boxes simultaneously—though they did find them all on the same day—and that they never found the excellent nest box first. For example, in Trial 1, scout bees were seen at the four mediocre nest boxes in the morning but not at the excellent nest box until the afternoon. Furthermore, we can see that sometimes a substantial crowd of scout bees had formed at one or more of the mediocre boxes before even one scout had found the excellent box. In Trial 2, for example, the number of scouts outside the mediocre nest box 1 grew steadily between 11:30 a.m. and 2:00 p.m. and had reached more than five bees by the time the excellent nest box 2 was discovered, shortly before 2:00 p.m.


Given that the excellent nest box was never found first and so always started out behind in the race to gain supporters, it is impressive that in four out of the five trials (1, 2, 3, and 5) the excellent nest box eventually gained the most supporters and became the chosen site. So, the five swarms did not achieve a perfect 5-for-5 score in this choice test, but they did demonstrate impressive skill in decision making. To see why this is so, consider the probability of getting the observed outcome purely by chance. If the swarms had chosen at random among these five nest boxes, then the probability that they would have chosen the best box in four out of five trials is vanishingly small, just 0.0064. In other words, one would expect to get the observed outcome of four correct choices and one incorrect choice simply by chance only one time in 156 repetitions of the experiment (1/156 = 0.0064). It is clear, therefore, that compared to relying on chance, the democratic decision-making process found in a honeybee swarm greatly increases a swarm’s likelihood of selecting for its future home the best of the candidate sites located by the intrepid scout bees.


I often find it useful to ponder instances where the bees behave unexpectedly, asking myself, “What is this surprise telling me?” Trial 4 of the best-of-5 choice test, in which the swarm chose a mediocre site, was a good eye-opener about how a swarm’s knowledge of each prospective nest site is at first extremely fragile and easily lost. We see from figure 5.7 that in the four other trials in which the swarms chose the best site, the counts of the scout bees suddenly changed in two ways after the excellent site was found: they rose rapidly at the excellent site and they fell steadily at the mediocre sites. In Trial 4, however, neither change occurred following the discovery of the excellent site. Why not? For some reason, neither of the two scouts that discovered the excellent site ever produced a waggle dance to announce her discovery. It is puzzling that neither bee reported her find, because in Trial 2 and Trial 3 the bees that found the nest box in this location (at the north end of the array) had produced waggle dances, even though they had found only a mediocre, 15-liter nest box there. It seems clear, therefore, that there was nothing wrong about the location per se. Whatever the cause of the puzzling nondancing by scouts from the excellent nest box in Trial 4, the consequence was clear: the swarm “overlooked” the best possible dwelling place on the island. Meanwhile, a slow buildup of scout bees persisted at one of the mediocre nest boxes and eventually the swarm chose this second-rate nesting site. This anomalous outcome shows us how it is critically important to the success of a swarm’s decision making that when a scout discovers a prospective homesite she reports it so that it becomes one of the options debated on her swarm. In the next chapter, we will see that the bees have a nifty rule of house-hunting behavior that normally results in every respectable housing option found by a swarm’s scout bees getting included in their debate. Good decisions require good information.

Swarm Knows Best

One might question whether the results of the experiment just described really show that honeybee swarms are good decision makers. After all, to draw this conclusion from the best-of-N experiment, one has to assume that a 40-liter cavity with a 15-cm2 entrance is indeed a high-quality nest site, and that a 15-liter cavity with a 15-cm2 entrance is indeed a medium-quality nest site, so that in choosing the former over the latter a swarm improves its ability to survive and reproduce. This seems to me to be a reasonable assumption, for why would honeybees have a preference for 40-liter cavities over 15-liter ones unless natural selection has favored having this preference? Certainly, studies of other animals—including various birds, reptiles, insects, and fish—have found that the nest-site preferences of these animals enhance their reproductive success.

In 2002, I decided to test my assumption that the housing choices of honeybee swarms really are good choices, ones that help colonies survive and reproduce. Regrettably, this test required an experiment in which many colonies would die, for I needed to compare the survival probabilities of colonies living in hives embodying what the bees do and do not prefer in a home. To do this, I installed artificial swarms in hives of two different sizes in the spring, then left them alone all summer, and saw whether the two types of colonies differed in probability of surviving the following winter. (As discussed in chapter 2, most honeybee colonies living in nature starve during their first winter.) Each swarm contained approximately 10,000 bees, a typical population size for natural swarms. For the two sizes of hives, I chose ones that held either 5 or 15 of the rectangular wooden frames that hold the beeswax combs in a hive, because these are the numbers of frames needed to hold the amount of comb that bees will build inside a 15-liter or a 45-liter tree cavity. Natural swarms occupy empty tree cavities and must invest heavily in comb building, so to give my artificial swarms the same challenge I installed them in hives containing empty frames in which they had to build their combs. (I did install a sheet of beeswax “foundation” in every other frame, to induce the bees to build their combs neatly within the wooden frames.) Each year that I have conducted this experiment, I have set up five colonies of each type in early June and then followed them for the next twelve months to see which ones would survive to the following spring.

To date, I have performed three replicates of the experiment—in 2002-2003, 2003-2004, and 2004-2005—so I have followed the fates of 30 colonies. For colonies in the 15-frame hives, the probability of winter survival has been 0.73 (11 out of 15 colonies), but for the colonies in the 5-frame hives this probability has been only 0.27 (4 out of 15 colonies). The large difference in colony survival between the two treatments has only a tiny probability (p = 0.02) of arising simply by chance. Almost certainly, the colonies in larger hives survived better because they amassed larger stockpiles of honey to sustain them through winter. I can make this claim because I weighed each colony’s hive at the start of the experiment in June and again in October, after heavy frosts ended the bees’ foraging for the year, and I recorded widely different average weight gains for the two sizes of hive: 23 versus 10 kilograms (51 versus 22 pounds), most of which is honey. Also, when I examined the combs of the colonies that died in this experiment, almost always I found them empty of honey. The poor bees starved. These stark statistics on colony survival as a function of hive roominess are solid evidence that swarms really do know best about their housing needs, and in exercising their nest-site preferences they really do make good decisions. They also make clear why honeybee swarms go to so much trouble to find good homes.