~—~——-—- ‘v— v I‘ - . ’1: v ,4 ' A_ r—u my IEIIJIIII’I'W"WWI"' ‘ 3.41%; -.w-:, «W IIIIIIIiéif'Ififffff .lx.» THE FEEDING DIVERSITY 0F DEER MICE Dissertation for the Degree of Ph. D- MICHIGAN STATE UNIVERSITY LINCOLN CLIFTON GRAY 1977 .. .rn ‘ TY L BRAR‘ES " iiiiii IIIIIIIIIIII I III IIIIII 3 1293 10692 3299 "P w- - " '_"H" .r,- - ~.. H‘ v~.‘.l"’. ." ) k l ; we Uni \r' c rsi t: y This is to certify that the thesis entitled THE FEEDING DIVERSITY OF DEER MICE presented by Lincoln Clifton Gray has been accepted towards fulfillment of the requirements for Ph-D- degree in .Neumsniences—Zoology M; Ci. Ides 0 Major professor \ Date WW/ /? /?7f 4’ / 0-7 639 llellIIIl [III ABSTRACT THE FEEDING DIVERSITY OF DEER MICE By Lincoln Clifton Gray The diversity of foraging preferences of laboratory—reared Peromyscus maniculatus boreaIis, a northern form, is significantly greater than the foraging diversity of P.m.blandus, a southern desert form. Foraging preferences are observed under controlled laboratory conditions in three different tests of motor activity and food preferences. Diversity is quan- tified by Shannon's index, calculated from percentage preferences of the various choices within each day and then averaged over days for each individual. Consistent differences between the two groups in these three independent tests suggest that behavioral diversity is the most predict— able parameter for characterizing the resource utilization strategies of these mice. Desert rodents exploit various arrays of seed types with predictable diversities. After experimental manipulation of a particular cue, changes in the diversity index show whether the hypothesized prominent resource axis was necessary and/or sufficient to determine foraging strategies. The hypothesis that desert rodents partition their food resources on the basis of seed size is examined in a series of experiments. Lab-reared and wild-caught desert rodents were offered choices among different- and same-sized seeds. Food size is necessary and sufficient to determine feeding diversity, but it is neither necessary nor sufficient to determine the item of maximal preference. The ecological applications of a psychophysical scaling technique Lincoln Clifton Gray called unfolding are explained. All of the information contained in a population of individuals' ordinal preference rankings for an array of resources can be preserved in an unfolded form. Key questions about foraging strategies can now be visualized: what is the dimension along which the animals partition available resources; and are there species— specific differences in which resources are maximally utilized? Seed preference data from desert rodents are unfolded. The dimension of difference among the seeds is consistent both within and between experiments, but not between lab-reared and wild-caught groups. Food size is the dimension along which sympatric wild-caught kangaroo mice and deer mice partition seed resources if we allow the possibility of bell- and U—shaped utilization functions. The dimension of resource selection is a better characterization of a species' foraging strategy than what items they maximally prefer. THE FEEDING DIVERSITY OF DEER MICE By Lincoln Clifton Gray A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Neurosciences - Zoology T977 ACKNOWLEDGEMENTS The support of my advisor, John A. King, made this study possible. Helpful criticism on all phases of these experiments by Martin Balaban, Donald Beaver, John King, James Zacks, and the graduate students of the behavior group at Michigan State University is gratefully acknowledged. Bill Foran, Tom Haviland, Jim Stanton, and Jeff Witcher assisted in the collection of data. Dorsey Green, Bernie Mathes, and Janet Wilson helped in the preparation of data and manuscripts. I thank them all. Computer facilitites were provided by the university. Financial support was provided by a Graduate Office Fellowship from the College of Natural Sciences and by a teaching assistantship in the Department of Zoology. TABLE OF CONTENTS LIST OF FIGURES INTRODUCTION CHAPTER 1 THE FEEDING DIVERSITY OF TWO SUBSPECIES OF PEROMYSCUS MANICULATUS IS CONSISTENT AND PREDICTABLE . INTRODUCTION MOTOR PREFERENCE TEST . Methods and Materials: Analysis of the Data: Results and Discussion: SAN D BOX TEST Methods and Materials: Results and Discussion: FOOD PREFERENCE TEST Methods and Materials: Results and Discussion: THE PREDICTABLE PARAMETERS OF FORAGING STRATEGIES THE MEASUREMENT OF BEHAVIORAL NICHE WIDTH CHAPTER 2 THE USE OF BEHAVIORAL DIVERSITY TO EXAMINE THE RESOURCE DIMENSIONS CONTROLLING FORAGING STRATEGIES FOOD PREFERENCE TEST Methods and Materials: Analysis of the Data: . Results: . . Page om“ 16 16 I7 19 19 20 23 28 30 3O 33 Page SIZE CONTROL TEST . . . . . . . . . . . . . . . . . 33 Methods and Materials: . . . . . . . . . . . . . . . 33 ResuHs: . . .. . . . . .. . . . . .. . . . . .. 3S [Nscussknr . . . . .. . . . .. . . . .. . . . .. 3S INTRINSIC CUE CONTROL TEST . . . . . . . . . . . . 36 Methods and Materials: . . . . . . . . . . . . . . . 37 ResuHs: . . .. . . . . .. . . . . .. . . . . .. 37 [Nscussknr . . . .. . . . .. . . . .. . . . .. . 38 FOOD PREFERENCE OF SYMPATRIC DESERT RODENTS . . 39 Methods and Materials: . . . . . . . . . . . . . . . 40 Results and Discussion: . . . . . . . . . . . . . . . 140 BRAZHJWITTEST . .. .. . .. .. .. . .. .. . 44 Methods and Materials: . . . . . . . . . . . . . . . all Results and Discussion: . . . . . . . . . . . . . . . 45 SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . '49 CHAPTER 3 THE USE OF PSYCHOPHYSICAL UNFOLDING THEORY TO DETERMINE PROMINENT RESOURCE AXES . . . . . 51 PART I THE PSYCHOPHYSICAL SCALING TECHNIQUE OF UNFOLDHK). .. .. .. .. . . 5H F%Hding . . . . . . . . . . . . . . . . . . . . . . . 54 Lhfiokflng .. . .. . .. . .. . .. . .. . .. . . 56 PART II PROMINENT RESOURCE AXES OF SEED SELECTION WIDESERT'RODENTS . .. . .. . .. .. . . 61 Tests With Laboratory—Reared Deer Mice A . . . . . . . 61 Tests With Sympatric Desert Rodents . . . . . . . . . 66 SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . 73 SUMMARY .. .. .. .. .. .. .. .. ... .. .. .. .. 75 APPENDIX THE PROCESS OF UNFOLDING . . . . . . . . . . . . . 76 BHMJOGRAPHY . .. .. .. . .. .. .. .. .. .. .. .. . 79 FIGURE 1. FIGURE 2. FIGURE 3. FIGURE 4. FIGURE 5. FIGURE 6. FIGURE 7. FIGURE 8. FIGURE 9. FIGURE 10. FIGURE 11. LIST OF FIGURES Three Dimensional Structure of Data Matrix Behavioral Diversities from the Motor Preference (MP) , Sand Box (SB), and Food Preference (FP) Tests . Mean Individual Diversities in the Food Preference (FP), Size Control (SC), and Intrinsic Cue Control (ICC) Tests Mean Individual Diversities in the Food Preference (FP), and Brazil Nut (BN) Tests Population Preferences in the Food Preference Test . Population Preferences in the Brazil Nut Test An Illustration of Folding . An Example of Unfolding Population Scales from Lab-reared P.m.borealis (G) and P.m.blandus (S) Population Scales from Microdipodops (M) and Peromyscus with Regular (P) and Inverted (P') Individual Scales . Population Scales from Microdipodops (M) and Peromyscus with Regular (P) and Inverted (P') Individual Scales . Page II 3LI LII 43 47 55 59 63 68 71 INTRODUCTION In this dissertation, I investigate the feeding behavior of deer mice. I use controlled experimental conditions in an attempt to identify and quantify the parameters which predictably characterize this behavior. The research is presented in three related but independent sections, written as chapters. A general introduction is presented at the beginning of Chapter I. It is my opinion that the goal of science is predictability, and that this understanding is best gained by attempting to objectively identify patterns in data. Thus, any models of foraging strategies developed in this research are merely shorthand attempts to describe what I see as the important structure in the data. This research is an attempt to let the animals tell me how they see their world. CHAPTER I THE FEEDING DIVERSITY OF TWO SUBSPECIES OF PEROMYSCUS MANICULATUS IS CONSISTENT AND PREDICTABLE 3 INTRODUCTION Three parameters form the basis of modern quantitative descriptions of resource utilization: dimensionality, variability, and central tendency (MacArthur, 1972; Levins, 1968; Schoener, 1974). These parameters derive from Hutchinson's (1957) model of the ecological niche. The niche is composed of many dimensions, each corresponding to a requirement of the species (Vandermeer, 1972) . When a measure of resource utilization is plotted against one of these dimensions, each species theoretically exhibits a characteristic utilization function (Root, 1967) . Ideally, the utilization functions of all relevant dimensions would be combined to give an imaginary exploitation space, or n-dimensional hypervolume, mapping the species' potential utilization in real space. However, for simplicity, investigators usually attempt to predict the most prominent environmental factor along which animals partition their resources, and the utilization data are plotted against that single axis. Thus utilization functions are usually depicted as probability density distributions along one axis with a particular mean and variance (Roughgarden, 1974b; MacArthur, 1972) . We can now visualize the three parameters of exploitation strategies that will form the basis of this discussion. The parameter of central tendency reflects the optimal or most utilized items. The parameter of variability quantifies the spread or breadth of utilization. The terms generalist and specialist are often used to refer to end points along a continuum of behavioral niche width and have fre- quently been measured using diversity indices (Morse, 1970 and 1971; KIOpfer, 1967 and 1973; Levins, 1968; Pianka, 1973; Colwell and Futuyma, 1971; MacArthur, 1972; Crowell, 1962). The parameter ofdimensionality It identifies the prominent axis, both its subjective scale as perceived by the animals and its physical attributes. Chapters 2 and 3 explore the prob- lem of dimensionality. This research uses controlled preference tests to identify and measure the parameters that most predictably characterize a population's resource utilization. In particular, I consider the behavioral niche width or the variability of foraging behaviors and ask if a measure of behavioral diversity is a more predictable characterization of the foraging strategy of deer mice than a measure of central tendency or the most preferred item. In order to minimize the chances that any discovered differences in foraging behaviors are the result of morphological differences that might exist between families (as in Klopfer, 1973) or even species (as in Rough- garden, 1974a; or Pianka, 1973) , I chose to study two geographically isolated subspecies of the same species. The deer mouse, Peromyscus maniculatus, includes approximately 28 geographical subspecies (Hooper, 1968) which inhabit a wide variety of habitats throughout North America from the boreal forest to the sub-tropics (Baker, 1968) . I used ecological niche theory in the initial selection of two subspecies, since the environment presumably influences the breadth of foraging strategies. Two hypotheses derived from niche theory propose that climatic stability and competition affect behavioral diversity. Although these presumed determinants of niche breadth enabled me to correctly predict the direction of differences in the foraging diversity of the two subspecies I selected, these experiments do not test the niche theory arguments which follow. My educated guess of which subspecies to study could have been merely lucky. The climatic stability hypothesis states that the more stable the 5 environmental parameters, the more specialized the species that occur in that environment. Several models of community structure predict the exis- tence of greater numbers and diversities of species in stable environments than in fluctuating environments (Connell and Orias, 1964; Sanders, 1969; Pianka, 1966; MacArthur, 1965). The decrease of species diversity with increasing latitude is one of the most imposing biogeographical features on earth (Fischer, 1960). Klopfer (1973) believes this increased species diversity is due to a decrease in the average niche width, a trend toward specialization. However, since niches can only be defined in retrospect, Slobodkin and Sanders (1969) warn that this attractive connec- tion between species diversity and behavioral diversity is circular. Field data are conflicting (Orians, 1969; Pianka, 1967). However, many models about foraging strategies at the level of individual species predict that niches will be wider in environments that tend to be widely fluctuating and unpredictable (Emlen, 1968; Futuyma, 1973; Griffiths, 1975; MacArthur and Levins, 1967; Klopfer and MacArthur, 1960; Oster and Heinrich, 1976). Some data exist to support these models (Randolph, 1973) . The problem with the argument above is two-fold. First of all we have no operational definition of the dependent variable of niche width. The problem of behavioral specialization is multi-dimensional. An organism may be a specialist for some resources and a generalist for others (Morse, 1971) . How the investigator partitions the environment will profoundly influence the dependent measure of diversity (ColweII and Futuyma, 1973) . Secondly, we have no operational definition of the independent variable. The predictability of an organism's environment is dependent in part on the organism itself. If an organism is satisfying all of its biological needs by using a constant or predictable resource even in a widely variable 6 climate, that organism's environment is stable (Slobodkin and Sanders, 1969) . It seems we have no way to quantify relevant environmental predictability, biotic or physical, without looking at the responses of the animals themselves. We are then trapped in another circular argument, this time between predictability and specialization. These problems make the climatic stability hypothesis virtually untest- able, but as an initial guess I chose to look for a generalist in the north where the climate and hence the resource base are more likely to be widely fluctuating and unpredictable. Thus, I selected P.m.borealis, a sub- species from the northern prairies of Alberta, for the experiments and predicted it to be a generalist. The competition hypothesis is an elaboration of the climatic stability hypothesis, susceptible to the same problems in the definition of the dependent and independent variables. This hypothesis states that if an environment is sufficiently predictable, and if its inhabitants face keen com- petition from sympatrics with neighboring niches along a resource axis, then evolution should force a decrease in niche width, i.e. specialization (Miller, 1967; Root, 1967; Pianka, 1966). Thus I selected a second sub- species, P.m.blandus, from the southern deserts of Arizona, predicted to be a specialist. P.m.blandus exists in an area populated by a host of other rodents: cricetids including P.eremicus, and Reithrodontomys; and heteromyids including Dipodomys merriami, D.spectabilis, and Perognathus flavus and P.penicillatus. Although it is doubtful that the ephemeral environment of the desert approaches the stability of the theo- retical models, the seeds produced in that environment are a predictably available resource (Brown, 1973; Reichman, 1975). In addition, the motor activities the animals use to get that food, foraging near the surface 7 for wind-blown seeds for example, are probably also reasonably constant. Thus, I predict P.m.blandus to be a specialist in the motor activity and seed selection experiments to follow. MOTOR PREFERENCE TEST The northern environment of P.m .borealis would presumably require the mice to use a wide variety of motor patterns to obtain food: climbing or walking along stalks for seeds, jumping or swimming puddles in the spring, gnawing through obstacles, or digging through snow in the winter. The desert environment of P.m.blandus requires relatively stereotyped motor activity. The design of the tasks used in this experiment depended, first, on whether I could build them and whether the mice would use them; secondly, on an attempt to represent motor activities I believed the mice might have to use in order to obtain food; and thirdly, to create four tasks as different as possible from each other. Methods and Materials: Each mouse was given a choice of four different motor activities which would lead it to food. In this situation the mice could choose among balancing, climbing, jumping, and swimming to get food. Each motor task led to identical goal platforms (35x55mm) with a block of food bolted to the far end. The platforms were connected to microswitches sensitive to the weight of a mouse. The number of times each platform was depressed, called "counts"; the accumulated time a mouse was on each platform, called "duration"; and sequential events were recorded in another room. Blocks of food between 3.0 and 4.5 grams were weighed in and bolted to the platforms every day. Since a mouse eats about three grams of food a day, 8 this amount did not force generalization on any mouse. During the next day the counters and clocks were read and the remaining food weighed. Each motor task was available in a separate unit constructed of acrylic plastic cages (28 cm long, 13 cm wide, and 15 cm tall). The balance, climb, and jump units were two cages high. Briefly, balance required walking 165 mm along a 5 mm diameter horizontal dowel 160 mm above the floor; climb was 235 mm up a 10 mm diameter rope; jump required a 98 mm leap 160 mm above the floor; and swim required travelling 75 mm in 40 mm deep water. The light cycle (15: 9) was identical to the colony room. Small lights burned at night to make observations possible. The position of each motor activity unit was varied randomly each day. Two P.m.borealis and two P.m.blandus were run concurrently with alternating shelf positions between runs. The subjects were F1 to F2 laboratory-reared males of wild-caught parents, between the ages of three months and one year. The genetic diversity and closeness to wild-caught parents in all of the following experiments were the maximum allowed by the breeding colony. The entire apparatus was thoroughly washed between mice to avoid any olfactory cues from previous subjects. Analysis of the Data: Figure 1 shows the three dimensional structure of the data collected in these experiments. Ten animals (shown as N along the third dimension) of each subspecies (represented by the two cubes) were given simultaneous choices among four motor activities or food types (shown as items along the horizontal x-axis). The presentations were continued for six consecutive days (shown along the vertical y-axis) . The important depen- dent variable, the amount of each food resource consumed per day, $ 0 '5“ ITEMS 2 3 4 ' if (D2 ‘- >-3 _ :4 ... 5 — 6 s. s, 3,3, ‘LH --) HI FIGURE 1. Three Dimensional Structure of Data Matrix 10 measured to the nearest 0.1 gram, filled the matrix of Figure I. In the analysis of behavioral diversity, these weights are used to calculate the percentage of food of each category eaten per day, per organism. These percentages are used to calculate indices of diversity to quantify not only the breadth but also the evenness of an animal's response to the available resources. Two common indices are used. Shannon's information theory index of diversity, H'=-; piloge(pi) , where pi is the proportion of con- sumption in the ith categb1ry (Shannon and Weaver, 1963; Pielou, 1966) , is used throughout the study. For the most important results Simpon's index, 8:”; piz, is also used to show that the significant difference does not dep-elid on which index is used. Values of H' vary between 0, when the mouse eats 100% of its food at one category, to 1.39, when the consumption is divided equally among all four categories. H' is calculated for each day for each animal as shown to the right of the cube in Figure 1. The daily H' is then averaged over the six day baseline to arrive at a mean diversity for each animal, shown as FI'. The two populations of scores are then described by means of these means (X of H') , and compared with the Mann-Whitney U statistic. Properties of the diversity index are discussed later. Results and Discussion: P.m.borealis, X=.66, s=.20, was significantly more diverse in its motor activity preferences than P.m.blandus, X: .19, 5: .18, as measured by H' calculated for each day and then averaged over days for each mouse (U=9ll/100, p<.001). These within-day diversities are shown under MP, for motor preference, on the left side of Figure 2. This result is not unique to one particular diversity index. Using Simpson's B, P.m.borealis 11 WITH IN DAYS SUBS PECIES L4 up 33 re MP se FP I.3 ' . O 1.2 0 II 0 LO 3: I I t 1 ° '5 .7 I } If. 2 .6 ° 5 ,4 "t 3 X Isa. 2 0 2 m.§grgolis . 1 m. .l O pCOOI p(.025 p(.0l FIGURE 2. Behavioral Diversities from the Motor Preference (MP), Sand Box (SB), and Food Preference (FP) Tests 12 is significantly more diverse than P.m.blandus (U=98/100, p<.001). Notice that there are two ways an animal can be diverse: over days and within days. For example, if a P.m.blandus ate 100% of its food at jump on one day, its daily H' would equal zero. But if on the next day it ate 100% of its food at climb, it would still have a daily H' of zero. The mean of these would be zero. The animal would be diverse over days but not within days. The analysis of this problem of switching involves a comparison of diversities within and over days. A new diversity measure, ignoring days, can be calculated by summing the amount of food eaten at each station over all six days and using these sums, shown as Si at the bottom of the cube in Figure 1, to calculate one H' for each mouse. If P.m.blandus were daily switchers but specialists within days, the difference between the groups would disappear in this analysis. This is not the case. P.m.borealis is significanly more diverse than P.m.blandus as measured by H' calculated from the sums of food eaten at each motor pattern over all six days for each mouse (U=97/100, p(. 001) . The amount of food eaten at each station was just one of the three mea- sures recorded by the apparatus. ' Diversities can be calculated from both counts and duration in the same way as described above for food. Counts. the number of times the mice depressed the platform, is somewhat difficult to interpret. If a mouse never visited a platform, it would register no counts. Unfortunately, however, positive counts do not measure the number of times the mice emitted the motor activity. Mice were observed to jump around on the platforms giving multiple counts for each motor activity completion. Total counts of more than 100 per night were not uncommon. The accumulated duration probably reflected less of this erratic behavior as no mice were observed motionless on a platform for 13 more than a few seconds at a time. Clocks were available to measure durations on only 7 mice of each subspecies. The switches jammed shut periodically, but event records of all switches allowed me to deter- mine when this happened (an average of 1.05 times for each mouse) and the mechanical data for that day were discarded. The same difference in behavioral diversity is found regardless of what dependent variable is used to measure resource utilization. Using counts, P.m.borealis, X=.93, s=.17, was significantly more diverse than P.m.blandus, X=.37, s=.29 (U=97/100, p<.001). Using duration, P.m. borealis , X: .65, s=.30, was significantly more diverse than P.m.blandus, X=.23, s=.20 (U=lI3/ll9, p<.01) as measured by H' calculated within days and averaged over days. A generalist is sometimes called a "jack-of—aII—trades-master—of—none" whereas a specialist is supposedly much more efficient at what it does. However, it has never been shown that generalists are less efficient in procuring resources than specialists (Morse, 1971). The relationships among food, counts, and duration can be examined in this regard. One model of an efficient mouse is one that only visits and spends time at stations where it eats. This model would predict high correlations among food, counts, and time for efficient mice. Linear regressions were cal- culated using percentage preferences of the three variables calculated within each day. Measures of central tendency are reported as corre- lation coefficients for all the usable data points for the subspecies as a whole. Tests for differences between the two forms use nonparametric statistics among the individual correlation coefficients calculated sepa— rately for each mouse. Since the mice had to spend a considerable amount of time chewing on 14 the bolted—down food lumps, the correlation of %food to %duration is, not surprisingly, high, and the two subspecies are not different (P.m.borealis r=.927; P.m.blandus, r=.957; U=33/Ll7, p).IO). However, in the regres- sion of %food to %counts P.m.borealis, r=.617, had a significantly lower correlation coefficient than P.m.blandus, r=.897 (U=82/100, p<.01). In the regression of %counts to %duration the correlations for P.m.borealis, r=.567, were also significantly lower than P.m.blandus, r=.945 (U=l12/ll9, p<.025) . These differences indicate that the more diverse strategist, the P.m.borealis, visited stations where it did not eat or spend time. This suggests that the P.m.borealis may be using a more energetically expensive strategy, and the specialists, P.m.blandus, are more efficient as predicted by the model above. This result is consistent with the hypothe- sis that the strategy of P.m.borealis is an adaptation for a widely fluctua- ting and unpredictable environment, perhaps in order to gain the advan— tages of constantly censusing alternative food sources in the environment, even though these resources are not immediately consumed. According to the hypothesis that the strategy of P.m.borealis is better adapted to changing environments, we would further predict that the P.m.borealis would adapt more quickly to the new test apparatus than P.m.blandus. For example, P.m.borealis should eat more food on the first day of the test, but the "neophobia" exhibited by the P.m.blandus should disappear by the end of the six days. The data support this pre- diction. P.m.blandus ate significantly less food on the first test day (U=82.5/100, p<.01) , but by the sixth day there was no difference (U=50/100). This result is also consistent with the hypothesis that the strategy of P.m.borealis is better suited for changeable environments. This result, however, is not replicated in all later tests. In all three 15 experiments the total amount of all food eaten over all six days by indi- vidual P.m.borealis and P.m.blandus was not different. Several theories (MacArthur and Pianka, 1966; MacArthur, 1968; MacArthur and Wilson, 1967; Schoener, 1974) and data (Royama, 1970) indicate that patch utilization may be more responsive to compression and expansion than the range of foods eaten within those patches. The dif— ference between P.m.borealis and P.m.blandus in this motor preference test is clearly greater than the differences in the following tests. Only in this test were the different resource items separated into physically different "patches". The greater difference between subspecies in this test probably includes a component of patch-utilization diversity. In summary, this experiment has demonstrated that two geographically isolated subspecies of Peromyscus maniculatus differ in the diversities with which they will exploit an available array of resources. I realize that foraging mice do not normally encounter horizontal dowels or hanging ropes. These tests are not an attempt to mimic natural conditions. Rather, they abstractly represent different motor activities that mice might have to use in order to obtain food. If natural selection in Alberta and Arizona has led to the existence of generalist and specialist strategies, then these differences should appear even in controlled laboratory conditions where they can be carefully mea- sured. If this interpretation is correct, then similar differences in behav— ioral diversity should appear in different and independent tests of for- aging preferences. 16 SAND BOX TEST This experiment was designed as an independent test of the diversity of motor activities used to exploit food resources. The priorities considered in the design of this experiment were the same as in the previous experi— ment. Methods and Materials: Four different motor activities, climb, dig, gnaw, and reach, were available simultaneously in a large arena, 59x30x30 cm. After each motor activity the animal had access to a small round 190 mg. "Noyes" food pellet. Sixteen pellets were available to the mouse from each motor task. Eight "climb" ladders, 10x240 mm made of 4x14 hardware cloth, hung in two rows along the middle of the arena ceiling. Two wooden hoppers containing one pellet each were attached to the t0p of each ladder, oriented in opposite directions, and covered by inverted eight-ounce party glasses to prevent the mice from jumping between hoppers. For "dig", sixteen pellets were buried under 5017 cm3 (approx. 2.5 cm deep) of sand. For "gnaw", pellets were placed in 2.5 cm diameter holes bored in a row of eight in each of two 35 cm lengths of 2x3 lumber and covered by a standard note card with a hole punched in the middle. The mouse had to gnaw through the card to reach the pellet. These "gnaw boxes" were placed at midline along the sides of the arena on top of the sand. "Reach" re- quired the mouse to reach into a 10x10x5 mm cup with one forepaw and remove a pellet. The food cups were located in two rows of eight each, 6 cm above one of two 2.6xlll cm platforms. The platforms were 9 cm 17 above the gnaw boxes and could be reached by metal ramps or short wire ladders. A narrow sloping roof was built 1 cm above the row of cups to prevent the mice from sitting on the cups or walking from cup to cup collecting food. I A home cage with nesting box and ad libitum water was connected to one side of the arena by a short tube. Two P.m.«borealis and two P.m. blandus were run concurrently in four apparatus in alternating positions. The entire apparatus was washed between mice. The light cycle, 15: 9, was the same as in the animal room. to F , lab-reared males between 1 2 the ages of three months to one year. The dependent variable which filled Subjects were previously untested F the matrix of Figure 1 was the number of pellets eaten from each motor activity. Hoarded and scattered pellets, identifiable by tiny markings, were not included in the measure of consumption. Data were collected every 24 hours for a six-day baseline and any missing pellets were replaced. Results and Discussion: P.m.borealis, X=.92, s=.23, was significantly more diverse than P.m.blandus, 52:.72, s=.18 (U=80/100, p<.025), as measured by H' calculated within days and averaged over days. These diversities are shown in Figure 2 under SB. The same difference is observed using Simpson's B (U=78/100, p<.025). Although significant, the difference between P.m.borealis and P.m. blandus is much less in this sand box test than in the previous motor preference test. One possible explanation is that the Noyes pellets were very friable. Perhaps the pellets crumbled and the powder became 18 unavailable in the sand as the mice attempted to eat. If this were the case, then the 3.05 grams of food available from each motor activity would not be sufficient to sustain a mouse for one night, and a specialist by choice would be forced to diversify. Data of total food consumption support this conjecture. In the previous motor preference test the mice ate an average of 2.7 grams of food a day. However, in the sand box test the mice ate an average of 24 pellets or 4.6 grams of food per day. It appears that this food inadvertently forced a more generalist strategy on any specialist mouse, and thus minimized any difference that might exist between the groups. Still, the important difference in behavioral diversity endured. In the examination of diversities calculated from data summed over days, the groups fell slightly short of statistical significance (U=70/100, .05.10). In summary, the difference in behavioral diversity between P_._m_ borealis and P.m.blandus is robust enough to endure in two separate motor activity preference tests. If this measure of H' calculated within days quantifies a general difference in foraging strategies between the two subspecies, the difference should appear in food preference as well as motor activity preference tests. 19 FOOD PREFERENCE TEST The climate of Alberta is likely to require that P.m.borealis select foods which differ widely along several resource dimensions. For this test of food preference diversity, I selected four easily available foods which the mice would eat. I realize that foraging mice do not normally encounter piles of the food used here. However, these four foods differ in many characteristics such as size, texture, and nutrition to which mice likely attend in selecting their natural diet. Methods and Materials: P.m.borealis and P.m.blandus were presented with a simultaneous choice of raw shelled Spanish peanuts, wheat germ, shelled sunflower seeds and millet. Truncated conical porcelain feeding dishes (45mmx43mm OD at the top) were placed inside cylindrical metal dishes (45mmx82mmOD) that served to collect any spilled food, and positioned in the corners of large plastic cages (20x48x38 cm). A small wooden nest box with bedding and access to water was placed in the center of each cage. The experiment was conducted in the colony room. The experiment measured food consumption for six days with 5i.2 grams of each food in a dish at the start of each night. The next day the remaining food was weighed and new food added to 51.2 g. The amount of food that was consumed was calculated and filled the matrix of Figure 1. Consumption did not include any hoarded or scattered food. Recent data from desert rodents show that hoarded food may be considerably differ— ent from the food that is eaten (Reichman, 1975) , thus supporting the procedure of subtracting hoarded food from the measure of consumption. The positions of the foods were randomized each day. 20 Seventeen males and three females, at least one of each subspecies, F1 to F , P.m.borealis and P.m.blandus, three to twelve months of age 2 were used as subjects. Because it is well known that obese humans and rats are much more sensitive to external cues in food choice situations (Schachter and Rodin, 1974) , all the adult mice in the colony were weighed, and only mice within one standard deviation of the colony mean were used as subjects. Two to three P.m.borealis and P.m .blandus were run con— currently with alternating cage positions. All apparatus were thoroughly washed between m ice. Results and Discussion: P.m.borealis, X: .93, s=.17, was significantly more diverse in its food preferences than P.m.blandus, X=.66, s=.23 (U=81/100, p<.01), as measured by mean H' calculated within each day for each mouse averaged over the six day baseline. Figure 2 shows these mean diversities under FP. The same difference is found using Simpson's B (U=81/100,p<.01). P.m.borealis is also significantly more diverse than P.m .blandus (U= 77/100, p<.025) as measured by H' calculated from the raw data summed over days for each individual. This means there is no bias in the previous conclusion from a daily switching of preferences. The 51:29 of each type of food were more than enough to avoid forcing any strategy on individual mice. The mice ate an average of 2.8 grams of food per day. The amount of food eaten on the first day was different between the two groups in the predicted direction (P.m.borealis ate more, U=76. 5/100, p<.05) , and there was no difference in consumption on the sixth day (U: 56 5/100, p).10). 21 THE PREDICTABLE PARAMETERS OF FORAGING STRATEGIES The results of these three experiments indicate a strong and consistent phenomenon. P.m.borealis has been shown to be more diverse in its utilization of available arrays of resource items than P.m.blandus. An unexpected result of this research is that whereas the measure of individual diversity is highly predictable, the type of motor activity or food an individual chooses is not predictable. Since the difficulties of the motor activities and the types of food were basically chosen for my convenience, with rough pilot data to insure the mice would exploit the resources, it might be predicted that one of the items would be easier or better than the others, and the mice would con- sistently show a preference for it. In fact, I originally predicted that there would be a consistent hierarchy of preferences for the available items and that P.m .borealis would just sample more of the least preferred items. This was clearly not the case. For example, among the 20 mice in the motor preference test, at least three different individuals showed a maximum preference for balance, climb, jump, and swim. The analysis of these differences in preferences among individuals involves a third measure of diversity (as in Hespenheide, 1975; Hurtubia, 1973; Pielou, 1972). In this measure the raw data in grams are summed over all days and all individuals for each subspecies. These sums are used to calculate four percentages, shown as Pi at the back of the cube in Figure 1, reflecting the percentage of each item used by the subspecies as a whole. The percentages are used to calculate a single diversity index for each subspecies. These subspecies measures are expected to be higher than the individual diversity measures because of variation between individuals, but if all the individuals of each subspecies had roughly the same 22 preferences, the measures would be similar. However, it is possible to obtain high subspecies diversities and low individual diversities if dif- ferent individuals preferred different items. Figure 2 shows this to be the case. In every experiment the subspecies diversities are much larger than the individual diversities calculated within days. Individuals within each subspecies had widely varying preferences but consistent diversi— ties. Stated differently, these laboratory data indicate that it is crucial to avoid lumping individuals' data in the analysis of foraging diversity. There is a major difference between the model of foraging strategies indicated by this research and other models currently in the literature. Most theories (MacArthur and Pianka, 1966; Schoener, 1971; Cody, 1974; May, 1973; Roughgarden, 1974b; Emlen, 1968; Werner and Hall, 1974) start with the assumption that natural selection has led animals to have a maximal preference for some optimal array of items. Then, given local changes in abundance or distribution of the foods, or competition from niche neighbors, the breadth of those choices should change. That is, they predict that some measure of central tendency should be more consis- tent than a measure of the breadth of utilization for characterizing a species' foraging behaviors. This research indicates exactly the opposite: diversity is more consistent than what particular item the animals will prefer. That is, natural selection has led to either diverse or specialized feeding strategies, but what the animals choose depends on proximal cues that I do not now understand. In this model we can picture a world of opportunistic feeders constrained in the breadth of their exploitation strategies but changing their preferences depending on the immediate situation. 23 THE MEASUREMENT OF BEHAVIORAL NICHE WIDTH Means and variances are appropriate measures of foraging strategies only if interval scale values can be assigned to the choice items. An interval scale assumes that arithmetic differences between numbers meaningfully reflect differences between categories (Stevens, 1975) . Examples of interval scales include prey size (Roughgarden, 1974a) or seed size (Brown and Lieberman, 1973) . Only if we accept this scaling as an accurate representation of stimulus differences as the animal sees them can foraging parameters be represented by means or variances. Many fewer assumptions are involved in using nominal scales to measure resources. Nominal scales assume only that different numbers refer to different items. Examples include numbers on the backs of football play— ers and the index of items used in these experiments, i=1,2,3,4. Diversity is an appropriate measure of variability for nominally scaled data (Pielou, 1975) , and the mode, or item of maximal preference, is an appropriate measure of central tendency. H' and B are two commonly used measures of diversity. No one index has any preeminent advantage (Hill, 1973) , although H' is slightly more sensitive to changes in rare categories, while B is more sensitive to changes in common categories (Peet, 1974) . l have shown that the im- portant results do not depend on which index is used, and I present values of H' only in dedication to the paper where I first encountered the concept of diversity (MacArthur and MacArthur, 1961) . There are three important factors to consider in the interpretation of diversity indices as measures of niche width. However, the following problems are not unique to nominally scaled data, since any measurement scheme must at least be a nominal scale. In fact, many additional 24 assumptions are necessary for interval scaling of foraging behaviors. The concept of information in this sense applies not to the individual messages or sets of behaviors, but rather to the situation as a whole (Shannon and Weaver, 1963) . That is, these indices are not directly comparable if the conditions are different (Peet, 1974; DeBenedictis, 1973; Pielou, 1975). We reach the conclusion of McNaughton and Wolf (1970) that generalists and specialists are only meaningfully defined in relation to each other. Thus, in this paper comparisons have been made only between subspecies in identical test conditions and not between studies. All units assigned to a particular category are assumed to be equal (Peet, 1974) . This assumption is clearly met in the two motor activity tests. A gram of food eaten at one location is identical to a gram of food eaten at another. However, the assumption is not met in the food preference test. A gram of peanuts is not identical to a gram of sunflower seeds. It is conceivable to measure food consumption with many dependent vari— ables other than grams: volume, dry weight, calories, protein, isoleucine, niacine, ... The decision of which one of these myriad measures to use involves the difficult question of the universal currency of animals' choices (Dawkins, 1976). My aprigridecision was to use grams as a measure of consumption because it reflects some combination of many food attributes. Grams are used throughout the three independent tests with consistent results. It is comforting that in the motor preference test the other measures of resource utilization, counts and duration, gave similar differences in diversity. No other measure was possible in the sand box test. Many current theories of foraging strategies (Schoener, 1971) assume animals 25 are attempting to maximize caloric intake. Critics of the use of grams might have preferred the use of calories as the measure of consumption in the food preference test. It makes no difference. Grams of each food were converted to numbers of calories using calorie contents from the literature (Bowes and Church, 1970) . The important result was nearly identical. P.m.borealis, X=.90, s=.l9, was significantly more diverse in its food preferences than P.m.blandus, X=.61, s=.24 (U=82/100, p<.01) , as measured by H' calculated within days and averaged over days for each mouse, using calories instead of grams. A final consideration necessary for the interpretation of H' is that all of the categories are assumed to be equally different (Peet, 1974) . All of the categories in these tests were physically different. We know, in addition, that the animals perceived them as being different because individual diversities were considerably less than the maximum of 1.39. Resources perceived as being equal would be exploited equally. The hypothesis that the specialists perceive finer differences among the resource categories than the generalists, and the hypothesis that both groups possess equal discriminative capabilities but differ in their attention to the cues associated with their diet, are not distinguishable in these preference tests. This distinction is not important in this study. The difference in behavioral diversity described in this paper is a robust phenomenon. It is replicable in several different tests of foraging preferences, and it makes no difference how it is measured. Although I used theory from community ecology to lead me to the prediction that P.m.borealis should be more diverse than P.m.blandus, the use of the concepts of competition, and environmental stability and predictability is less important at this point than the fact that I found a consistent 26 difference in foraging behaviors. These data are a test of the hypothesis that P.m.borealis is more diverse in foraging preferences than P.m.blandus. Differences in motor activity or food preferences recorded in the field or with natural resources in the laboratory would not be a test of that hypothesis. Measures of foraging diversity could not be meaningfully compared because the available resource arrays would be different for the two groups. Since I have hypothesized that natural selection in Alberta and Arizona has led to the existence of generalist and specialist foraging strategies, it is important to discover if these laboratory behaviors are consistent with behaviors observed in the field. These field data will be very dif— ficult to interpret because resource availabilities could force specialist or generalist strategies on either group. However, forced generalization of P.m.blandus in the field due to limited resource availabilities and individual specialization in these laboratory tests with super-abundant resources is easier to imagine than the reverse for P.m.borealis. Thus, for the most important check of these results in the field, P.m.borealis should at some times exhibit a variety of motor patterns to get a variety of foods. Upon finding this difference in foraging diversities, I faced a choice of research priorities. I could have done a test of environmental variability of food resources in space and time and the concurrent structural com- plexity that may also be present in Arizona and Alberta. I could have measured the behavioral diversities of other groups of Peromyscus whose habitats differed in latitude, spatial heterogeneity, numbers of competi— tors, or resource predictability. This might have enabled me to do a large multivariate correlation to find those parameters that most influenced 27 the measure of behavioral diversity. Partially because of the difficulty in unambiguously quantifying these independent environmental variables, I chose first to use this discovered difference in diversity to explore the underlying behavioral mechanisms. The ecological or evolutionary reasons for this difference may best be revealed after the behavioral process provides an operational definition of niche width. CHAPTER 2 THE USE OF BEHAVIORAL DIVERSITY TO EXAMINE THE RESOURCE DIMENSIONS CONTROLLING FORAGING STRATEGIES 28 29 Resource utilization functions are usually depicted as probability density distributions along a prominent resource axis (Root, 1967; Roughgarden, 1974b; MacArthur, 1972) . Tests of foraging strategies and resource partitioning depend critically on the decision of what axis the data are plotted against (Schoener, 1974) . If the dimension is estimated by the biologist, then the particular prominent axis is based on human rather than the animals' perceptions of salient resource differences. In addition, Colwell and Futuyma (1971) warn that resource states should not be mea- sured in terms of physical or chemical variables measured in ordinary units, but in terms of a "subjective effect" on the organism in question. Thus, methods are needed to determine how the animals perceive differ- ences among available resources. This chapter will show how a measure of behavioral diversity can be used to test the prominent resource dimensions which control foraging strategies. The experiments reported here involve only an examination of seed selection by size in desert rodents, but this method of analysis may prove generally useful in the examination of other sensory dimensions to which animals may attend in the selection of their diet. A specialized feeder must utilize perceived differences between the available items in its decision of what item to eat. The hypothesis that specialists perceive finer differences among the food items than generalists, and the hypothesis that both groups possess equal discriminative capabil- ities but differ in their utilization of the cues associated with their diet, are not distinguishable in preference tests. Regardless of how the diet is - selected, some resource dimension must be more salient to specialist than generalist feeders in the control of their resource utilization stategies. The salient resource dimension can be examined by measuring the 30 diversity of diet selection after specific manipulations of certain physical attributes of the food items. If the manipulation significantly increases the diet diversity, a necessary cue to which the animals attended in selecting their diet was removed. Conversely, if low diet diversities endure after the presentation of resources which differ only along the hypothesized prominent resource axis, then that cue is sufficient to determine the diet breadth of the animal. The demonstration that a particular cue was neces— sary and sufficient would be convincing evidence that a salient resource axis to the animals had been found. FOOD PREFERENCE TEST The first chapter established that two geographically isolated subspecies of Peromyscus maniculatus differed in the diversity of their diet selection among available arrays of resources. In a food pref— erence test the desert-inhabiting P.m .blandus selected a more specialized diet from among a choice of different seeds than P.m.borealis, a northern prairie form. This test is described again here because the methods and data are the basis of the experiments which follow. Methods and Materials: P.m.borealis and P.m.blandus were presented with a simultaneous choice of four easily available foods which I had previously determined the mice would eat: raw shelled Spanish peanuts, wheat germ, shelled sunflower seeds, and millet. In this test the foods were presented as purchased. Each food was placed in truncated conical porcelain feeding dishes (45x43mmOD at the top) which were inserted in cylindrical metal dishes (45mmx82mmOD) that collected any spilled food. The dishes were 31 positioned in the corners of large plastic cages (20x38x48cm). A small wooden nest box with bedding and access to water was placed in the center of each cage. The experiments were conducted in the animal room with relatively constant temperature (70i50F) and a 15:9 light cycle. The experiment measured baseline food consumption for six days with 5i.2 grams of each food in a dish at the start of each night. The next day the remaining food was weighed to the nearest 0.1 g and new food added to 5i.2g. The amount of food consumed was calculated and entered into the matrix of Figure 1. Consumption did not include any hoarded or scattered food. Recent data from desert rodents show that hoarded food may be considerably different from the food that is eaten (Reichman, 1975) , thus supporting the procedure of subtracting hoarded food from the measure of consumption. The positions of the foods were randomized each day. Seventeen males and three females, at least one of each subspecies, F to F from wild-caught P.m.borealis and P.m.blandus, three to 1 2 twelve months of age, were used as subjects. The parental generation of P.m.borealis had been caught in the prairies near Calgary, Alberta, and the P.m.blandus in the desert near Portal, Arizona. Because obese humans and rats are unusually sensitive to external cues in food-choice situations. (Schachter and Rodin, 1974) only subjects within one standard deviation of the mean weight for the mice in the colony were used. Two to three P.m.borealis and P.m.blandus were run concurrently with alternating cage positions. The apparatus were thoroughly washed between mice. 32 Analysis of the Data: Figure 1 shows the three dimensional structure of the data collected in these experiments. Ten animals (shown as N along the third dimension) of each of two groups (represented by the two cubes) were given simulta- neous choices among four food types (shown as items along the horizon- tal x-axis) . The presentations were continued for six consecutive days (shown along the vertical y-axis). The dependent variable, the amount of each food consumed per day measured to the nearest 0. lg, filled the matrix of Figure l. The percentage of food of each type eaten per day per mouse was calculated. The percentages were used to calculate Shannon's (Shannon and Weaver, 1963) information theory index of diversity, H'=-§1piloge(pi) , where pi equals the proportion of consumption of the ith ro—od type. H' quantifies not only the breadth but also the evenness of the animals' diet selection. Values of H' vary between 0 when a mouse eats 100% of its food at one category, to 1.39 when the mouse divides its consumption equally among all four categories. H' was calculated for each day for each animal as shown to the right of the cube in Figure 1. The daily H' was then averaged over the six-day baseline to arrive at a mean diversity for each animal, shown as H'. Properties of the diversity index are discussed in Chapter 1. The two populations of mean diver- sities were then compared by tand U tests. Non-parametric statistics are used throughout this and the previous chapter for testing differences between groups of mean H's. This is a conservative approach. It could be argued by the central limit theorem that since H' is averaged over six days, this measure is normally distributed and appropriate for parametric statistics. In statements which accept the null hypothesis the use of non—parametric statistics might increase the chance of a Type II 33 error. Thus in all of these situations I have also calculated Student's t and found a probability greater than .05. Results: In the food preference test where the foods were presented whole, the desert P.m.blandus, X: .66, $2.23, were significantly more specialized than P.m.borealis, X=.93, s=.17 (U=81/100, p<.01), as measured by H' calculated within days and averaged over days for each mouse. Figure 3 shows these diversities. SIZE CONTROL TEST Size of resource items has been hypothesized to be an important dimen- sion in the diet selection of desert rodents (Brown and Lieberman, 1973; Reynolds, 1950; Smith, 1942) and other animals (Werner and Hall, 1974; Roughgarden, 1974a) . The debate about the importance of food size will be discussed later. If size were the cue to which the desert Peromyscus were attending in selecting their diet from among the four different-sized foods in the preceding experiment, and if that cue of seed size were removed, then these animals would perceive the foods as being more similar and select a more diverse diet. That is, if the different foods are made the same size, and the specialists become generalists, then food size (or a confounded cue) was necessary for their foraging strategies. Methods and Materials: In the size control test two screens (2.00 and 0.42 mm openings) were found that removed approximately 5% of the coarsest and finest particles of the wheat germ, the smallest food in the food preference test. 34 MANIPULATION OF DIVERSITY IA, FP SC ICC I 8‘ § ,: .— é } u, .6 2 O .4, Yilss .2. 02.111.12.9me 0136919313.: 0 FIGURE 3. Mean Individual Diversities in the Food Preference (FP) , Size Control (SC), and Intrinsic Cue Control (ICC) Tests 35 This provided the standard size for the other foods. The three other seeds were coarsely milled and sifted between the two screens. Ten, naive, F1 to F2, three to twelve month old males of each subspecies were used as subjects. All other details were identical to the food preference test. Results: In the size control test where the foods were ground and screened to the same size, the P.m.blandus, X=.84, s=.14, were not different from P.m.borealis, X=.88, s=.34 (U=62/100, p).10). The shift in foraging diversity occurred as predicted with the previously determined specialists, P.m.blandus, becoming significantly more generalized (U=74/100, p<.05) in the size control test as compared to the food preference test. The diver- sity of P.m.borealis did not change (U=51/100, p).10). Figure 3 shows these diversities. The specialists became generalists. The item most preferred by the P.m.blandus, peanuts, did not change between the two tests, showing that food size is not necessary to determine the central tendency of the population. Discussion: Food size is actually a conglomerate of cues including diameter, cross-— sectional area, volume, weight per particle, texture, surface to volume ratio, .. . , which have been concomitantly manipulated. The term food size will refer to this conglomerate. The ability to increase the foraging diversity of the desert specialists shows, with one minor qualification, that food size is a necessary resource cue to which the animals attend in selec— ting their diet. It is possible, though I believe unlikely, that the manip— ulation of grinding and screening changed some sensory dimension 36 other than food size (such as the intensity of olfaction or gustation or the ratio of edges to volume or width to length) to which the animals might have attended. These mice used some cues other than size in the decision of which foods to eat. The degree to which the mice see the foods as being differ— rent is quantified by the diversity of their selection. The upper 95% confidence limits on the mean individual diversities for P.m.blandus (.93) and P.m.borealis (1.28) are below the maximum diversity of 1.39. That is, we know the ground and screened foods were still perceived as somewhat different because the mice did not sample equally from all available items. Many cues other than size, such as taste, smell, and color remain for the mice to use in the determination that the foods are different. By manipu- lating these dimensions one at a time, it would be theoretically possible to partition the animal's attention to these various cues. The salience of the cue would be quantified by the decrease In selectivity after each manipu- lation, as was done with food size. Unfortunately, the manipulation of cues such as taste, smell, and color, while holding all other cues constant, is practically impossible. However, it is possible to determine the relative salience of the combination of these myriad other resource dimen- sions from another approach. INTRINSIC CUE CONTROL TEST If the mice were offered four identical foods, there would be no resource cues to differentiate the items. The degree to which the mice attend to all of the possible resource dimensions other than size can be measured by the increase in diversity after the presentation of identical items Other external cues would remain: the position of the feeding cup, any 37 olfactory markings the mice might have left in the dishes or food, or any small irregularities in the dishes. Any attention to these external cues can be measured by the degree to which the diversity of selection among identical foods deviates from the maximum. Methods and Materials: Four dishes of whole peanuts were presented to 10 P.m .borealis and 10 P.m.blandus. Peanuts were used because they were the most preferred category over all mice in both the food preference and size con— trol tests. The amount of hoarded food was partitioned among the four cups in the proportion to which food had been removed from each cup. The subjects were previously untested F to F males between the ages of 1 2 three months and one year. All but one P.m.borealis (+1.25) were within one standard deviation of the colony mean in weight. All other details were identical to the previous experiments. Results: P.m.blandus, X=.99, s=.24, was no more specialized than P.m.borealis, X=1.02, s=.18, as measured by H' calculated within days and averaged over days (U=54/100, p).10). Neither the P.m.blandus (U=71/100, p).05) nor the P.m.borealis (U=60/100, p).10) increased significantly in diver— sity in the intrinsic cue control test as compared to the size control test. These diversities are shown in Figure 3. These data indicate that intrinsic cues other than size, such as color, taste, and smell, do not contribute significantly to the animals' perceptions that the foods are different, even in the aggregate. That is, only size (or a confounded cue) is a necessary resource dimension. In a different situation any one of these cues (such as 38 smell as in Drickamer, 1970) might be sufficient to determine a specialized diet. The upper 95% confidence limit on mean individual diversity (1.13 for both groups) is still below the maximum diversity of 1.39. This indicates that external cues contribute significantly to the diet selectivity of these mice. Discussion: There is considerable debate over food size as the important resource dimension underlying desert rodent foraging strategies. Several authors have stated that desert rodents are Opportunistic, with preferences gener— ally determined by the availability of food items (Reichman, 1975; Blair, 1937; Monson, 1943) especially Peromyscus maniculatus (Meserve, 1976; Cogshall, 1928) . Since in these experiments the foods were presented in equal abundances, these hypotheses would predict equal sampling from all foods. The Peromyscus studied here are obviously not attending solely to abundance. Other authors have suggested that habitat dimensions may be more important than seed size for desert rodents (Smigel and Rosenzweig, 1974; Lemen, 1976; MacMahon, 1976) and other animals (Royama, 1970; Smith and Sweatman, 1974; MacArthur and Pianka, 1966) . If we accept a tenuous analogy between these "habitat dimensions" and the "external cues" of this research, the data suggest that external and intrinsic resource cues are about equally important. However, among all of the possible intrinsic dimensions, food size (or something else changed by grinding and screen— ing) is the only cue necessary to the desert P.maniculatus in selecting their diet. 39 From studies with other animals other properties have been suggested as prominent resource cues, such as secondary plant compounds (Freeland and Janzen, 1974) , rate of intake of digestible material (Smith and Follmer, 1972) , handling time (Willson and Harmeson, 1973) , or unknown qualities (Willson, 1971). These cues are not necessary in the system of these experiments. Again, this does not mean that any of these other cues might not be sufficient in other situations. In summary, the manipulation of grinding and screening removed a cue(s) necessary for the desert deer mouse, P.m.blandus, in determining the breadth of their diet. Furthermore, this cue(s) was the only significant resource dimension used by the mice, but other factors external to the resource items contributed significantly to their diet selection. This demonstration that food size is a necessary cue for the diet selection of these mice is only half of the evidence required for a convincing argument that food size is a prominent resource axis. The demonstration of sufficiency is the other half. Before such a test is presented , a background experi- ment is necessary. FOOD PREFERENCE OF SYMPATRIC DESERT RODENTS It is widely believed that environmental contingencies influence the breadth of foraging strategies in some way (Klopfer, 1973; Randolph, 1973; Cody, 1974; Morse, 1971; MacArthur, 1972; Futuyma, 1973; Grif- fiths, 1975; Oster and Heinrich, 1976; Miller, 1967; Jaeger, 1974). In the first chapter I hypothesized that the difference in behavioral diversity between P.m.blandus and P.m .borealis in the food preference test may have been due to the different environments from which they came. If there is an environmental factor that affects foraging strategies, either 40 biotic or physical, then two sympatric species facing the same environ— mental constraints should exhibit equal diversities. It is true that even in the same geographical location the "relevant" environments of two sympatric species may be different, but my initial prediction was that two sympatric rodents which utilized similar resources in a similar manner should have equal diet breadths. Furthermore, by the competitive ex- clusion principle (Hutchinson, 1959; Morse, 1971; MacArthur, 1972; May and MacArthur, 1972) the two sympatric species should prefer different items, if one assumes food is a limiting resource. Methods and Materials: P.m.sonoriensis and a similar sized heteromyid, Microdipodops pallidus, the kangaroo mouse, were wild—caught in Fish Lake Valley, Nevada. The mice were acclimated to the conditions in the animal room for several weeks and then tested for food preferences and diversity of feeding strategies using whole peanuts, sunflower seeds, millet, and wheat germ. Other details of the procedure were identical to the pre- ceding tests. Four Microdipodops and eight Peromyscus survived to complete the experiment. All mice were used regardless of sex, age, or weight. Results and Discussion: The sexes of either group did not differ and their data were pooled. The diversity of the feeding strategies of Peromyscus, X=.53, s=.27, and the Microdipodops, X=.45, s=.22, as measured by H' calculated within days and averaged over days for each mouse, did not differ signif— icantly (U=18/32, p>.10). These diversities are shown in Figure 4. 41 DIVERSITY OF DESERT SYMPATRICS |,4_ FP BN '0 l L '0: l OI? DIVERSITY, H' j..o.. a L .._x __.| O '__x_—| it“... X 111931129122! 0 Mom FIGURE 4. Mean Individual Diversities in the Food Preference (FP) , and Brazil Nut (BN) Tests 42 In Figure 5 population preferences of the two species are plotted against seed size. In this analysis the raw data in grams are summed over all days and all mice for each group. These sums are used to calculate four percentages, shown as Pi at the back of the cube in Figure 1, reflecting the percentage of each food item that is eaten by the population as a whole. These percentages are then graphed as a function of the middle of the modal size category determined by sifting the foods through a series of screens (12.7,11.1,9.5,7.8,6.4,4.0,2.0,1.4,1.0, .84, .50, .42, .35, .24, .20 mm opening). As expected, the sympatric rodents had equal diversities but different preferences, and these utilization functions remarkably resembled theoretical utilization curves (MacArthur, 1972) . The maximally preferred seed size for the Microdipodops in the experiment, 1.70 mm, is nearly identical to that in published field data. Brown and Lieberman (1973) snap-trapped wild rodents in Fish Lake Valley, Nevada, and plotted the distribution of seed sizes recovered from their cheek pouches. For Microdipodops pallidus they report maximum seed size frequency centered at 1.65 mm (Figure 3, p. 790). For estimating food size in Peromyscus, they graphed a utilization function based solely on body size and concluded that their preferred seed size was nearly identical to Microdipodops. In contrast, my research supports Smigel and Rosenzweig (1974) indicating that body size need not be correlated with food size preferences. These data are consistent with the hypothesis that food size is the resource axis along which these animals partition their food resources but are not a test of that hypothesis. Food size could merely be correlated with another unknown dimension. 43 PERCENT UTILIZATION FOOD SIZE in m. WWW FIGURE 5. Population Preferences in the Food Preference Test 44 BRAZIL NUT TEST A direct test of the hypothesis that food size is a prominent resource axis would require the presentation of food items that were identical in all respects except for size. If food size alone is sufficient to determine the diet selectivity of these animals, the diversities for both groups would remain unchanged from the previous experiment where the different sizes were presented as different seeds. Methods and Materials: One year after the previous food preference test additional Microdipo— dops pallidus and P.m.sonoriensis were wild-caught in Fish Lake Valley, Nevada. After one to six months of acclimation to laboratory conditions they were tested for diet selection among four sizes of Brazil nuts. Brazil nuts were chosen because they are a readily available homogeneous large nut meat that can be easily crafted to various sizes. The nuts were skinned, chopped, rounded in a device similar to a rock polisher, and then screened. Four sizes were used in the experiment: large, 12.7 - 11.1 mm in diameter; large medium, 9.5 - 7.9 mm; small medium, 6.4 - 4.0 mm; and small, 2.0 — 1.4 mm. The large size was chosen to determine if the utiliza— tion function of the Peromyscus (Figure 5) was a bell-shaped distribution truncated on the right as might be predicted by current foraging theories (MacArthur, 1972). Large medium was the size of peanuts, small medium the size of sunflower seeds, and small the size of millet. This experiment was continued for twelve instead of the usual six days. Mice were used regardless of their sex, age, or weight. Other experimental details were identical to previous experiments. 45 Results and Discussion: Microdipodops did not adapt as well to the Brazil nut diet as they had to the mixed-food preference test. Of 17 wild-caught Microdipodops that were put into the Brazil nut study, slightly over half of them (9) re- fused to eat or ate very little and were removed from the experiment within the first six days, usually in torpor, to save the animals. Only one Peromyscus out of 13 had to be removed. My impression was that those mice that did start to eat did not establish consistent eating patterns as quickly as in the previous tests. Thus the presentations of Brazil nuts were continued for 12 successive days. The data support my initial impres— sion. The eight Microdipodops that finished the experiment ate only 17.5% of the total food consumed during the 12 days of the experiment during the first three days, but 46% of the food in the first six days. If the animals ate at a constant rate from the beginning of the experiment, 25% and 50% of total food consumption would be expected after the first three and six days respectively. The Peromyscus ate at a more constant rate, consuming 24.5% and 50.5% of their food in the first three and six days. Plots of per- centage food consumption over days for the Microdipodops appeared to increase until reaching asymptote on around day six. Thus, in the analysis of these data I used only the last six days (7-12) in order to have the best possible comparison within and between studies. Eight Microdipodops and 12 Peromyscus finished the experiment. The data from each sex in both groups did not differ, and their data were pooled. The diversities of the Peromyscus, )_<=.59,s=.34, did not change from the food preference test (U=55/96, p).10) , nor did the diversities of the Microdipodops, X=.43,s=.26 (U=17/32, p).10). The two species did not differ in their diversities in the Brazil nut test (U=64/96, p). 10). 46 There was virtually no change in the pattern of diversities from the food preference test as shown in Figure 4, indicating that seed size alone is sufficient to determine the diet selectivity of these mice. Population preferences as a function of seed size are shown in Figure 6. Three Microdipodops and one Peromyscus were observed to be attending more to the position of the food dishes than to the size of the nuts in the dishes. These animals were excluded from the population preference analy— sis because it makes no sense to include the food size preferences of animals who were attending to position. Position tracking was determined by assembling a new matrix where the columns refer to the corners of the cages in which the foods were placed on each day. A new diversity measure is calculated from this position matrix and the data matrix of Figure 1, where the columns refer to the different food types. This measure is found by summing the amounts of food eaten in each column (size or corner) over all six days and using these sums, shown as Si at the bottom of the cube in Figure 1, to calculate two H's for each mouse, one from each matrix. If the animals attended more to size than to the position of the dish, then the H' from the size matrix will be smaller than the H' from the position matrix. When the reverse was true, the animal was considered a position tracker and was discarded from the analysis of population preferences. Data from the position trackers were included in the analysis of individual diversities because H' calculated within days and averaged over clays remains the same regardless of whether the animal tracks position or size. No position trackers were found among the wild-caught rodents in the previous food preference test. The population preferences of the two subspecies show utilization functions on food size in Figure 6 that are very different from those of PERCENT UTILIZATION 47 4o- 30“ .0 x 20- IO-* I To ii ('2 L FOOD SIZE in mm. -x- Microdipodops, N=5 ..o.. Peromyscus,N=ll FIGURE 6. Population Preferences in the Brazil Nut Test 48 previous tests (Figure 5 from the food preference test and Brown and Lieberman, 1973) where the different sizes were presented as different seeds. Thus seed size alone is not sufficient to determine which item the animals will maximally prefer, although it is sufficient to determine the breadth of diet selection. These data indicate that the parameter of diver- sity is a much more predictable characterization of a species' resource utilization strategy than a measure of what the animals maximally prefer. The same conclusion was reached in the first chapter with different data and for a different reason. The point of minimum preference for the Peromyscus in the food pref- erence test (Figure 5) coincides with the point of maximum preference for the Microdipodops. Although the population preferences of Figure 5 resemble displaced bell-shaped distributions on first glance, the curve of the Peromyscus might be a U-shaped function truncated on the left instead of a bell-shaped distribution truncated on the right. The pref-— erence curve of the Peromyscus in the Brazil nut test (Figure 6) supports this conjecture. The substantive issue is whether the animals are foraging with separated bell—shaped distributions as predicted by many current foraging theories or whether they are foraging with centered bell— and U-shaped utilization functions. Bell- and U—shaped utilization functions have been proposed for other niche dimensions (Grant, 1972; and Jaeger, 1974) but never, to my knowledge, for foraging data. It is very difficult to decide which underlying structure better fits these data from the population preference curves, but in the third chapter I will use psychophysical scaling techniques to show that Microdipodops and Pero- myscus more likely forage with bell- and U-shaped utilization functions respectively. Perhaps the more morphologically specialized 49 Microdipodops out-scramble other rodents for seeds of a particular size and the Peromyscus learn to prefer the only other available seeds. A process similar to imprinting (lmmelmann, 1975) could explain these learned preferences. SUMMARY AND CONCLUSIONS One difficulty in this study is that the demonstration of necessity used F1 to F2 P.m.blandus while the demonstration of sufficiency used wild- caught P.m.sonoriensis. For several reasons, I doubt that the conclu- sion of necessity and sufficiency for diversity are threatened by the use of these two groups. First, P.maniculatus from Arizona and Nevada are probably very similar. In fact, according to Hall and Kelson (1959) the P.m.blandus used in this experiment could as accurately be called P.m. sonoriensis. Second, the diversities of the two groups in the food pref— erence test were not different (U=54.5/100, p).20). Finally, diversities have been shown repeatedly in my experiments to be a predictable charac— terization of a species' strategy. Wild-caught P.maniculatus were used for the Brazil nut test because these mice had shown population preferences along food size (Figure 5) that were of theoretical interest. The lab-reared mice, however, did not select their foods along the dimension of food size (see Chapter 3) even though the diversites were identical. l surmise that while diversities may be genetically determined, the dimensions to which the mice attend in selecting items within that specified range are the result of early experience. Thus, the interesting question is not whether food size is sufficient for the lab-reared mice but rather whether it is necessary for wild-caught mice. This experiment is impossible without further support. 50 In summary, this chapter has shown how a measure of behavioral diver— sity can be used to determine the cues to which animals attend in selec— ting their diet. Food size, or some other confounded cue changed by grinding and screening, is necessary to desert P.maniculatus in the determination of their diet selectivity. No other intrinsic resource cue was necessary, but other cues external to the food items contributed significantly to the animals' perceptions of resource differences. Food size is not necessary to determine the item of maximal preference. Two sympatric desert rodents had equal diet diversities and divergent pref— erences. Although food size is sufficient to determine the foraging diver— sity of desert Peromyscus and Microdipodops, it is not sufficient to deter- mine the item of maximal preference. In conclusion, diversity is a much more predictable parameter of foraging strategies than a measure of central tendency which represents the most preferred food item. The resource cues which determine the parameter of diversity are different from those cues that determine central tendency. In this model we can picture a world ofopportunistic feeders, attending to particular resource dimensions in determining the breadth of their exploitation strategies but changing their preferences depending on the immediate situation. CHAPTER 3 THE USE OF PSYCHOPHYSICAL UNFOLDING THEORY TO DETERMINE PROMINENT RESOURCE AXES 51 52 Implicit in any statement about resource partitioning either between (MacArthur, 1972; Schoener, 1974; May and MacArthur, 1972) or within species (Roughgarden, 1974a) , is the assumption that the investigator knows the prominent resource dimension (5) to which the animals attend in selecting their diet and partitioning the available resources. Usually investigators make intelligent guesses about this dimension based on competent observation and analyze their data accordingly. However, these guesses are based on human rather than the animals' perceptions of salient stimulus differences. Occasionally it is possible to perform the experimental manipulations necessary to demonstrate the prominence of a particular cue (Werner and Hall, 1974; Chapter 2). These are time con— suming experiments; only one hypothesized prominent resource axis can be tested at a time and negative results are not very instructive. Experimental manipulation must by definition involve a change in a mea- surable independent variable, but Colwell and Futuyma (1971) warn that resource states should be measured not in terms of physical or chemical variables measured in ordinary units, but in terms of a "subjective effect" on the organism in question. To circumvent these problems of guesses based on human perceptions, trial and error experimental manipulations, and the need to measure resource states along a subjective scale to the consumer, techniques are needed where salient resource dimensions are determined from the animals' utilization data. The field of sensory psychophysics is concerned with precisely this problem: how to determine the dimension (5) of subjective difference among various stimuli. The field has been primarily concerned with human perceptions of physical dimensions such as the intensity or wavelength of a light or the amplitude or frequency of a sound (Stevens, 1975) . 53 However, psychophysicists have developed elegant and simple techniques which when applied to resource utilization data can determine the prominent axis along which the animals partition the available resources. In the first part of this chapter I describe a simple psychophysical scaling technique called unfolding. Raw preference data from several animals are used to determine the prominent axis along which the pop- ulation perceived differences among the available items. The non-metric unfolding described in this chapter was developed by Coombs (1964) and requires only a pencil and paper. Further discussion of the details of actually carrying out this technique can be found in the Appendix and in Coombs (1964) . The technique is appropriate when the investigator believes the animals perceive the physical stimuli along a single basic dimension. Other more complicated techniques (Coombs, 1964; Carroll and and Wish, 1974; Shepard, 1972) , which will not be considered here, are appropriate for scaling data that may have a multi—dimensional structure. The assumption that resource utilization data have an underlying uni— dimensional structure is very common in ecology. In order to simplify complicated niche metrics, investigators usually attempt to predict the most prominent environmental factor along which animals partition their re- sources. Thus, utilization data are usually plotted along a single resource axis (e.g. Roughgarden, 1974 a and b; MacArthur, 1972; Root, 1967; Brown and Lieberman, 1973; Werner and Hall, 1974; May and MacArthur, 1972). Unfolding uses data from the animals to define that single most prominent resource axis. In the second part of this chapter I show how the technique can be used on seed preference data from desert rodents to determine the sensory dimensions along which the animals perceived 54 differences among the available seeds. PART I THE PSYCHOPHYSICAL SCALING TECHNIQUE OF UNFOLDING In order to present an intuitive understanding of unfolding, I first describe its inverse, folding. We start with three major assumptions about the underlying structure of the utilization function and derive what the animals' raw preferences would be if the assumptions were true. Folding In the illustrative example of folding shown in Figure 7 I consider fish feeding on minnows, but the method is applicable to any dimension along which animals respond differentially: foraging behavior, communi- cation, or habitat selection. For the purposes of this example assume that we know that fish attend only to the size of minnow prey in selecting their diet. This is the first assumption: size is the sensory dimension to which these animals attend in partitioning food resources. Furthermore, let us assume that there are four different sizes of minnows in the lake (called A, B, C, and D). It is a simple matter to determine the ordinal ranking of the items along the scale. In this case minnow A is the largest and D is the smallest. This is the second assumption: the ordinal ranking of the items along the dimension which the population uses to rank the resource items is ABCD or DCBA, depending on how you look at the scale. This will be called the "population scale". Finally, let us assume that we know the items of maximal preference for each animal. The big fish (1) prefers the largest minnows, the small fish (3) the smallest, and the medium— sized fish (2) prefers intermediately sized prey. This is the third assumption: we know the optimal resource category for each predator. 55 FOLDING IDEAL > >4 POINTS 4 POPULATION e- r 4 : t l I | | . SCALE j I I I I ' l ' I j l I > l>© -® ITEMS I s 3 n o« A "0 F1 POPULATION ad SCALE I c" FOLDED .. AROUND a «0 EACH IDEAL j .Ponrr B‘r "l’c A" «II-D Arr tip «it a DA ABCD CODA ocaA ' gm, '- FIGURE 7. An Illustration of Folding 56 These will be called the "idealpoints". The location of each individual's ideal point is shown in Figure 7. The closer an item is to the ideal point the greater the preference for that item. Knowing these three things, it is possible to "fold" the population scale around each individual's ideal point, and reading from the fold toward the ends, determine the ordinal preference ranking of each individual for the available prey. These are called "individual scales". The individual scales which result from this folding are shown at the bottom of Figure 7. Since we are only concerned with ordinal rankings, the exact location of ideal points is unimportant until they cross midpoints between resource items. For example, the derived individual scale for the largest fish would have been identical for any ideal point to the left of the midpoint between items A and B. If its ideal point occurred on the right side of this imaginary midpoint, the folded individual scale would have been BACD instead of ABCD. This idea, that the locations of ideal points are only meaningfully bounded by midpoints between items, will be important later. To more formally sumarize, in the operation of folding we assume that we know 1) the particular axis along which a population of animals per— ceives resource differences, 2) the ordinal ranking of resource items along that axis, and 3) the location of each individual's ideal point. These three assumptions determine the ordinal preference rankings of each individual for the available items. Unfolding In unfolding we do the reverse of folding. Starting with a population of individual ordinal preference rankings for a certain number of different items (several individual scales), we "unfold" these scales to determine 57 the population scale from which they came: 1) the axis along which those animals perceived the difference between the items, 2) the ordinal ranking of the items along that axis, and 3) the optimal resource items for each animal. Behaviorists and ecologists will realize that individual ordinal rankings are data that we can or already have collected. All of the infor- mation contained in one form of the data, folded or unfolded, is preserved in the other form. In the unfolded form important ecological relationships that were obscure in the raw data can now be clearly visualized: what is the dimension along which animals partition their food resources, and are there species— or individual—specific differences in the locations of ideal points of maximally preferred categories? The necessary raw data can be collected by any method that measures differential preferences such as volume (Pianka, 1973) or weight (Rough- garden, 1974a; Reichman, 1975) of stomach contents, cheek-pouch contents (Brown and Lieberman, 1973) , fecal analysis (Meserve, 1976; Smigel and Rosenzweig, 1974) or the consumption of provided foods (Smith and Follmer, 1972; Willson, 1971; Willson and Harmeson, 1973; Werner and Hall, 1974; Chapter 2) . Habitat as well as food preferences could be scaled using smoked cards (M'Closkey, 1975) or counting foraging visits (Morse, 1970). Ordinal preference rankings of the available items for each indi— vidual animal are easily derived from these data. These are the individ— ual scales to be used in unfolding. If all of the individual animals used the same population scale to select their resources, then their individual scales must end in one of only two items regardless of the location of their ideal points. Notice that it is impossible to fold the population scale of Figure 7 in such a way that the derived individual scales do not end in either the largest or the smallest 58 minnow. Figure 8 shows an example of eight different individual scales. All end in either A or D. We do not know which item is first or last, but we know they are at the ends of the population scale. There can be two and only two individual scales which begin with one of these items and end with the other, and these two different individual scales must be mirror images of each other. In the example of Figure 8 these scales are ABCD and DCBA. This order determines the ordinal ranking of the resource items. Again, we do not know in which direction to read the scale, but we know the ranking of the items from most to least of something. The biologist must now look at that population scale and determine what physical attributes are consistent with the ordinal ranking. Usually there are several possibilities, but many can be eliminated. If the items were the minnows of Figure 7, possible physical dimensions would include body size, body weight, size of tail fin, surface to volume ratio, etc. This list of possibilities would not include color, chemical composi- tion, or body shape. The location of each individual's ideal point can be located within a region bounded by the midpoints between items. There are six midpoints among four items which delimit seven regions, shown in Figure 8. A limited amount of metric information can be obtained from the un- folding of four items. The form of the middle individual scale implies the relative distance between the lst and 2nd items as compared to the 3rd and 4th items. In Figure 8 the middle individual scale is assumed to be CBAD, implying that the distance along the population scale between A and B is less than the distance between C and D, symbolized as ABEoLod can :2: “300033.225. Eot mo_mum catfisaom .3 $50.“. sum ....Exmm. 2. .2 m A. mean. .8... _ 8222 Incraz: $2. .2 .... m n. 222 . Ia. manna L233. mhzmoom hmmmmo .uozmmwuwmn— 000“. 69 Peromjscus and Microdipodops were foraging with U- and bell-shaped utilization functions respectively, a consistent unfolding would be impos- sible. The least preferred items of the two groups would be different, and the first condition of unfolding would not be met. Some scale would result because of noise in the data, but it would not explain much of the data, and it would not be easily interpreted. lnverting the ordinal preference rankings (changing PWSM to MSWP for example) of the animals hypothesized to have the U-shaped utilization functions, the Peromyscus, would correct this problem. The items of least preference for the Peromyscus are now analyzed as if they were the most preferred and vice-versa. The "ideal points" of the deer mice are now actually points of least preference. If the two groups foraged with bell— and U~shaped utilization functions, then the two species should now have similar individual scales ending on one of only two items and their ideal points should coincide. The unfolded population scale should explain more of the data and be easily interpreted. The inverted individual scales of the Peromyscus coincide closely with the regular individual scales from the Microdipodops; unfortunately, they coincide too closely. All of the rodents except one Peromyscus now have one of only two individual scales that are not mirror images of each other: MWSP and SMWP. It is virtually impossible to meaningfully unfold a population of individual scales that are this similar. (See the discussion of degenerate data in Step 1 of the Appendix.) A population scale of food size, including even the known metric information (the difference in size between peanuts and sunflower seeds is greater than the difference in Isize between millet and wheat germ), can be folded to explain all of these raw data except for the single odd scale, and the "ideal points" coincide as shown in the bottom of Figure 10. However, the argument I wish to make 70 is weakened considerably by the fact that there is another population scale (PWMS, PVT/(MS) that can be folded to include 100% of the raw data. None of the physical scales listed above are consistent with this ordinal ranking. Because of the similarity in the individual scales of 92% of the animals used in this experiment, this second population scale is basically deter- mined by the individual scale of the single "outlier": MWPS. The difference between the two population scales is the result of one mouse. Perhaps its preference ranking is not characteristic of the population. These data indicate that either the animals are foraging with separated unimodal distributions along a prominent resource axis consistent with phenylalanine or niacine concentration; or more likely, foraging with bell- and U-shaped distributions along a dimension consistent with food size, or an unknown dimension. It is not possible to distinguish among these hypotheses with these data, but the alternatives can be distinguished in the Brazil nut test where the four different items differed only in size. The unfolded population scale from the Brazil nut test is shown at the top of Figure 11. This ordinal ranking is not consistent with seed size. In fact, this population scale is nonsense; it is very hard to imagine any physical scale consistent with the subjective ranking of small medium, small, large medium, and large, or vice-versa. To test the alternatives of the underlying structure of these data, the ordinal preference rankings of the Peromyscus were inverted as discussed above. There were a larger number of sufficiently diverse indi- vidual scales in these data to allow a meaningful unfolding solution. This yields a population scale along food size as shown at the bottom of Figure 11. The population scale explains more of the data and is easily interpreted. 71 mo_mom _man. c. 2.52: can An: Lm_:mom 5:2, moom>EoLod new :2: 3000 596:2 E9... mo_mom cozflaaom .: Mano—u o\o mm m 55 .2... I. P b 1 3.8.2: + h . .9... .83. .182 .r has. 162825 Av\emn. I. 2... m 2m . b . 1 -II-- III- F - n22 6.2 .2 .&2+n.a.2. A. 36:63.. 2.32 I__N