## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

THE STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY Author(s): RICHARD LEVINS Reviewed work(s): Source: American Scientist, Vol. 54, No. 4 (DECEMBER 1966), pp. 421-431 Published by: Sigma Xi, The Scientific Research Society Stable URL: http://www.jstor.org/stable/27836590 . Accessed: 02/01/2013 00:29

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

.

Sigma Xi, The Scientific Research Society is collaborating with JSTOR to digitize, preserve and extend access to American Scientist.

http://www.jstor.org

This content downloaded on Wed, 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions

and evolution is usually as population or species is treated at sumed to occur in a constant environment. 54. population biology must deal tionary time with genetic. had developed with quite mathematical different assumptions and techniques. while mathematical biogeography is essentially a new field. the result expressed in the form of quotients of Even (c) sums of products of parameters would have no meaning for us. and age heterogeneity within simultaneously and evolving systems changing demographically species of multispecies a in other of influences under the fluctuating heterogeneous species some are still only there are too many parameters to measure. This would require using perhaps 100 simultaneous partial differential equa tions with time lags. if soluble. on the other hand. the most areas of population biology. as are treated the species genetically homogeneous. describes populations The environment is allowed to vary but and densities. one-to-one reflection of this complexity. many would require measurement. sys ecology. brute force approach would be to set up a mathematical model which is a faithful. in a way that preserves the Clearly we have to simplify the models essential features of the problem. 4. a lifetime each for their vaguely defined. a population of genotypes without reference to the age distribution. phys tems. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . measuring hundreds of parameters. so that evolution is environment. physiological.american scientist. recognizes multispecies Population in terms of their age distributions. is specified by the frequencies For population genetics. or population density. However: (a) and then measuring these pre ignored. 1966 THE STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY By RICHARD LEVINS biology arises from the coming together of population what were previously independent clusters of more or less co Modern herent theory. The problem is how to deal with such a complex system. physiological state as a reflection of past history. A single a time. solving the equations to get numerical predictions. dictions against nature. iological states. (6) The equations are insoluble analytically and exceed the capacity of even good computers. Population genetics and population ecology. The naive. time and evolu But there is increasing evidence that demographic are Thus commensurate. The difference between legitimate and 421 This content downloaded on Wed.

This approach is favored (1965) and myself. and precision toward the overlapping but not identical goals of understanding. Therefore. Starting with precision they hope to increase realism. a selection coefficient that varies between .g. means instead of specifying the mathematical form of an equation. But the problem at hand was: that environments this tradition which involves setting up quite general equations from which precise results may be obtained.422 AMERICAN SCIENTIST It is of course desirable to work with manageable models which max imize generality. Sacrifice generality to realism and precision. ofmany fishery biologists. in any case.. greater or less than some value. make fairly accurate measurements. For instance. and end with precise testable pre dictions applicable to these particular situations. physiological states. and the effect of a species' population density on its own rate of increase. realism. This that the predictions we can make are also expressed as inequalities This content downloaded on Wed. and ofWatt (1956). the way in which nature departs from theory will suggest where further complications of their organism. 1959). Sacrifice realism to generality and precision. Since we are really concerned in the by MacArthur rather than quantitative results (which are long run with qualitative only important in testing hypotheses) we can resort to very flexible often graphical. predicting. But this cannot be done. and would be an unnecessary complication. increasing or decreasing. They expect that many of the unrealistic assumptions will cancel each other. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions .001 and . The early pioneering work in population genetics by Haldane. convex or concave. illegitimate simplifications depends not only on the reality to be de scribed but also on the state of the science. is an essential in But. which generally assume that functions are models. and modifying nature. and most physicists who enter population biology work in will be useful. several alternative strategies have evolved : 1. (e. These workers can reduce the parameters to those relevant to the short term behavior Could weak natural selection account for evolutionary change? For the purposes of this problem. Kerner (1957).01 will have effects somewhere between constant selection pressures at those values. Sacrifice precision to realism and generality. and that. and Wright all assumed a in the models although each author was aware constant environment are not constant. 2. solve numerically on the computer. that small deviations from realism result in small deviations in the conclusions. environmental heterogeneity gredient of the problems and therefore of our mathematical models. for us today. But these workers hope that their model is analogous to assumptions of frictionless systems or perfect gases. 3. they use the Volterra predator-prey systems which omit time lags. Their equations are clearly unrealistic. This is the approach of Holling. Fisher. Leigh (1965).

relative distances on themap correspond to relative distances in reality. even the most flexible models have artificial assumptions. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . However.STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY as between 423 geographic map. we attempt to treat the same problem with several al ternative models each with different simplifications but with a common biological if these assumption. our model treats all the curves as As an example of a robust theorem consider the proposition that. Since a mixed population of two pheno types would be represented by a point on the straight line joining their points. on a graph whose axes are Wi and W2. despite their different have what we can call a robust details of the model. system (Levins and MacArthur. such as the Robust and Non-robust identical except for the location of the peak s*. it is not always obvious when we are using too high a magnification. Hence our lies. We will use threemodels. This problem does not arise in the more familiar models. where we all know that contiguity on themap implies contiguity in reality. insular versus continental faunas. there are polymorphic populations represented by new points. The environment consists of two (easily extended to N) alternative facies or habitats or conditions. lead to similar results we theorem which is relatively free of the truth is the intersection of independent models. if the two environments are similar compared to the rate at which fitness declines with deviation (that is. Therefore. etc. But. from the actual environment. the fitness set of Levins (1962). 2. the fitnesses in environments 1 and 2. if the fitness set is convex then population heterogeneity adds no new fitness points. each phenotype is repre sented by a point as in Figure 1. Thus. a calculus of variation argument. on a concave fitness set. whereas. For each phenotype i there is a best environment w declines with the deviation of s. The set of all available phenotypes is designated the fitness set. the extended fitness set of all possible populations is the smallest convex set enclosing the fitness set. It remains to add that. 1966). and the rate at which fitness declines peak at s?. Then. the height the with the deviation from optimum. and one which specifies the genetic This content downloaded on Wed. patchy versus uniform environments. In particular. Theorems tropical and temperate species. Al ? Si) in nature may differ in the location of the though the curves W(s the of peak. in the mathematical models of population biology. There is always room for doubt as to whether a result depends on the essentials of a model or on the details of the simplifying assumptions. in an uncertain environment. assumptions. but color is arbitrary and a microscopic view of the map would only show the fibers of the paper on which it is printed. I assumes: Model and fitness 1. species will evolve broad niches and tend toward polymorphism.

But as the environments diverge the set becomes concave. an uncertain + (1-p) log Wi Same environment the optimum is the one population log Wa. lb. the fitness set will be convex. WipW21'~p. family of straight lines pWi + (l-p)Wi ures for a fine-grained stable environment. Same In for a concave fitness set. This content downloaded on Wed. fitness occurs for the pheno Optimum is represented of these lines with the fitness set. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . If the environment is uniform in space but variable in time. maximizes specialized. In an environment which is uniform in time but showing fine grained heterogeneity in space. that un lc. Fig.and fine-grained environments and therefore gave the linear expression for spatial heterogeneity in general. Id. Fig. each individual is exposed to many units ? of environment of both kinds in the proportions to 1 of their oc = C are the fitness meas la. The 1962 paper did not distinguish between coarse. The Fig. currence. type which by the point of tangency Fig.424 AMERICAN SCIENTIST similar compared to the tolerance of an individual phenotype). the rate of increase is a product of fitnesses in successive generations. 3. Here creates a broad for a concave polymorphism niche. On a convex fitness set this is monomorphic fitness set. Thus the rate of increase of the population is pWx + (1 ? p)W<?. For the two situations these alternative functions would be maximized tomaximize over-all fitness.

2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . Therefore. The rules of genetic segregation restrict the possible to points on the curve joining the two homozygous points populations and bending halfway toward the heterozygote^ point. This analysis does polymorphism directly since it discusses the assignment of the fitness of the whole curve is broader than the maximum breadth population. the populations environment the to both of type intermediately well-adapted be will all monomorphic exceeds the tolerance of If the environmental environments. The points of the three possible genotypes AiAi. The result is that. These = CP(S). so that fitness to survival in any other. ? Instead we fix II does not fix the shape of the curve W(s Model = to so C. natural selection will move toward gene frequencies which maximize the log fitness averaged over all individuals. At optimum. A2A2 are the fitness points at that locus.thefitnessis log (C) + / logP(s) -P(S)d8. if is not very diverse (convex fitness set). assigned to one contributes nothing that they coincide in their major results adds to the robustness of the theorem. But if the P(S) attainable by individual phenotypes. this means that selection maximizes ? (1 p)W^ But as the environment is the same as maximizing pWi + two axes which This content downloaded on Wed. While a continuum. The graph inFigure 2 has. diversity in results then fitness the individual (concave set) spatial diversity common habitat while temporal diversity more the to specialization results in polymorphism. the fW(s)P(s)ds tion. We already know from Fisher that. the flatter and W(S) spread not mention and the broader the niche. We did not assert that evolution will in fact establish the optimum population but only the weaker expectation that populations will differ in the direction of their optima. Thus curve more out the iable the environment. A1A2. P(S) case. Therefore. discretely While the fitness set specifies how different environments are by showing the relation between fitness in both environments for each phenotype. the optimum population would assign all its fitness to the most abundant environment while in the second case the optimum isW(S) the more var log (C) minus the uncertainty of the environment. for rather general conditions. In a fine-grained environ ? log [pWi + (1 p)W%] which ment. represent fitnesses in environments 1 and 2. In the first heterogeneity. Both models are similar in that they use optimization arguments and ignore the genetic system. inmodel III.STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY The 425 rest of the argument is given in the figure. polymorphism will be optimal. we examine a simple genetic model with one locus and two al?eles. the second treats each environment as totally different. But even this is not obvious. this restric curve that fW(s)ds the area under the Subject which for a is rate of increase. second the different permits environments. or the first allows only two models differ in several ways. as before. we maximize for and temporal / log W(s)P(s)ds fine-grained spatial heterogeneity is the frequency of environment S.

In a fine-grained environment average superiority of the hetero zygote is necessary for polymorphism. will 2a. resp. The are represented curves .) Fig. optimum strategy in the sense that a population of all heterozygotes would not be optimal. there can be no poly is closer to 0. Only when will the A1A1. A2A2. A2A2 curve at the point A1A1 then specialization (or A2A2. popula are on the curve AiAi. As the two environments become more similar. 4.5 than the slope of the curve morphism. each individual is exposed to fewer units of environment until. = zero and for values of between governed pW\ + (l-p)W<i by the equation one. 3.426 AMERICAN SCIENTIST becomes more coarse-grained. Other work on the joint evolution of habitat selection and niche breadth. and polymorphism can only come about as an imposition of the facts of segregation when the heterozygote is superior. A2A2 population become unspecialized. In a fine-grained. heterogeneity appears as an average.K' by points each with from an infinite family of curves connecting equal points samples fineness. Then the optimum is a single genotype. the is to broaden But polymorphism is an ogeneity required 2.) to the A1A1. in a coarse and hence uncertainty. Here grained environment as alternatives ? selection maximizes: logWi + (1 p) logWi orWipW2l~p. each one lives either in environment 1 or in environment 2 for the relevant parts of his life. In a coarse-grained environment the same holds?a sufficient heter now niche. If most This content downloaded on Wed. tions replace polymorphism. Wi approaches W2 for each genotype. Thus. We note the following from the figures: 1. in the limit. When than the slope of the tangent these lines are steeper (or flatter. be linear and the K-curves would stable environment. This has nothing to do with the "mixed strategy" polymorphism of previous arguments. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . Selection in an environment Possible with average heterosis. and on food getting procedures all converge in supporting the theorem that environ of the environment is of type 1. on the role of productivity of the environment. in a fine grained environment.

how simple logistic. depending we add a term ?px to the right hand side to the result can be reversed if indicate extraneous pr?dation. and K. The points of tangency one specialized and one polymorphic. Sufficient Parameters The thousand or so variables of our original equations can be reduced to manageable proportions by a process of abstraction whereby many This content downloaded on Wed. is fragile and cannot be asserted as a biological fact. tainly even in resources the are on used. Further. Thus the theorem. the carrying capacity or saturation level.We may be dealing here with a case of examining a map under the microscope. is an environmental variable. As an example of a non-robust theorem. Selection Fig. aver without environment in an uncertain 2b. the favored gene frequencies.and predator-species both results have a common explanation in the model in which high r from below but also not only increases the rate of approach toward This latter from above. (coarse-grained) curve to the curve AiAi. However without a saturation prey. r is the intrinsic rate of increase. Leigh (1965) showed that it also holds for a Volterra system of many level K. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . consider the proposition that a high intrinsic rate of increase leads to a smaller average population (productivity is opposed to biomass). although interesting. toward crash the property was cer speeds or of and in the definition not intended may not be true may r. This result can be derived from the logistic equation for population growth = dx/dt rx(K x)/K where is the population size. A2A2 are of the age heterosis.STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY 427 leads to increased niche breadth while certain but mental uncertainty diverse environments lead to specialization.

Thus the measure should have the resources equally. plicated function of the species densities which may acquire meaning for us with further study. and competition coefficients. Similarly. In other cases. Kerner dis and may matics covered a conservation law for predator-prey systems. If two populations which have equal niche breadths that do not overlap are merged. all the physiological in a genotype enter the models of population genetics only as part of "fitness. Thus. This is by no means equivalent to asserting that community properties are additive or that these sufficient parameters are terms enter into consideration only by way of a reduced number of interactions of genes higher-level entities. the sufficient parameters are formalizations of pre viously held but vague properties such as niche breadth.428 AMERICAN SCIENTIST very much fewer than the number of parameters on the lower level and which among them contain most of the important information about events on that level. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . on This content downloaded on Wed." independent." The multiplicity of genetic variance" species interactions is grouped in the vague notions of the ecological niche. It is an essential ingredient in the concept of levels of phenomena that there exists a set of what." The great diversity in populations appears mostly as "additive and "total genetic variance. the sufficient parameters arise directly from the mathe lack obvious intuitive meaning. Thus. niche overlap. If it uses two resources unequally. Two measures satisfy these re : quirements log ? ? log is the measure of relative abundance where resource or in a given habitat. niche breadth. We would like some measure of niche breadth which reflects the spread of a species' fitness over a range of environments. their joint niche breadth should be the sum of their separate breadths. by analogy with the sufficient statistic. we can call sufficient parameters defined on a given level (say community) which are Sometimes. between proposed Both are defined by us to meet heuristic criteria. It may be less if they overlap but never more. The final choice of an appropriate measure of niche breadth will depend on convenience. working with cellular metabolism and starting like Kerner with a physics background. the niche breadth measure should lie between 1 and 2. Goodwin (1963) as a biological found an invariant which he refers to metaphorically "temperature. and 1/B one = p2 of the species on a given is the "true" measure in the sense that one can decide Neither alternative structures for the hemoglobin molecule. But what is con It is a com served is not anything obvious like energy or momentum. it should following properties: if a species utilizes have a niche breadth ofN.

Method I is log proportion = ? of the given species taken where 2. produc tivity of the habitat.0 breadth 21. so that the m?ximum niche measure II 10.9 5. would is the be models which explain how to go from "uncertainty" to the components of the environment and biology of the species in question. patchiness of the environment. TABLE SEASONAL NICHE BREADTH 1 RIC?N DROSOPHILA 1962??? OF SOME PUERTO .7 11.2 8. The many-to-one nature of "uncertainty" pre vents us from going backwards.9 7. Thus general models have three kinds of imprecision : The sufficient parameter is a many-to-one transformation of lower lies its power and utility. D. If either temporal variation or patchiness or low productivity of uncer leads to uncertainty. and on the extent to which the measures lead to biological predictions based on niche breadth. In Table 1 we show some from our study of Puerto Ric?n Droso sample niche breadth measures phila populations. In our robust theorem on niche breadth we found that temporal variation.5 4.6 6. Therein source of imprecision.4 15. and mode of hunting could all have similar effects and that they did this by way of their contribution to the uncertainty of the environment.5 7.6 5.2 6. It be comes necessary to supplement our theorem with some subordinate The sufficient parameters may arise from the combination of results of more limited studies. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . Mean we should use both measures in presenting ecological data so that while. (1) they omit factors which have small effects or which have large effects but only in rare cases. measure I 14.2 13.0 12. X4 X6 The data species melanogaster latifasciaeformis dunni tristriata ananassae repleta nebulosa paramediostriata (tripunctata group) (tripunctata group) are based on 21 collections. Thus uncertainty emerges as a sufficient parameter.5 11. they may be compared and studied together.STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY 429 some new criteria which may arise. but also a new level phenomena.2 7.9 log ? and method II is 1/B in each collection. D.2 6. This content downloaded on Wed.5 15.7 11. functions (2) they are vague about the exact form of mathematical in order to stress qualitative properties.-niche breadth during D. the consequences or tainty alone cannot tell us whether the environment is variable patchy or unproductive. Therefore we have lost information. D. D. D. D.0 6. D.

a model isnot verifiable directly by experiment. 3. inadequate. Hence. standing but by a relation between the general and the particular. Broken equivalent relation among aspects of nature is likely to be true in the sense that it leave out a lot occurs (although rarely and slightly). For all models are both true and false. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . destroys in the general models are necessary but not sufficient for under nature. Models Clusters of A mathematical model is neither an hypothesis nor a theory. Yet all models sense The are validation of a in that and false. For understanding is not achieved by generality alone. Relations ecological among community. Unlike the scientific hypothesis. those not any longer live issues This content downloaded on Wed.430 AMERICAN SCIENTIST (3) the many-to-one property of sufficient parameters formation about lower level events. that it but "true" not is that it model is generates good testable hy A model may be discarded in to important problems. incomplete. potheses relevant is more but it a favor of usually simply outgrown when the powerful one. some of the components in a theory of the structure of an lines enclose alternative models. Almost any plausible proposed Measures of environmental diversity Developmental biology Response to selection in a heterogeneous environment classical ^population genetics Fig. are for which itwas designed.

121-146. Levins. Ees. On I. Nat. 1962. PNAS. function. MacArthur. The fitness set of fitness in a heterogeneous R. 361-373. Thus a satisfactory theory is usually a cluster of models. while they are essential for understanding reality. 2 Jan 2013 00:29:47 AM All use subject to JSTOR Terms and Conditions . 320. Bull. 1959. Levins. stabilityof a community.C.. 1965. heterogeneous nature and a mind that can only cope with few variables at a time. and adaptive In press. PN AS. of Canada. A statistical mechanics of interacting biological species. in Cells. This content downloaded on Wed. even by the opposing esthetic standards which emphasize the stark as against the richness and simplicity and power of a general theorem the diversity of living nature. 1957. 58 (4) 777-783. for pre models choice and solution of mathematical E. 2. 6. H. Goodwin. should not be confused with that reality itself. and . biomass. the relation between productivity. Mat. as complementary models they can cope with different aspects of the same problem and give complementary as well as overlapping results. Watt. Holling. H. 613-645. Biophys. Fisheries 13. dicting and maximizing 1956.. and R. models are restricted by technical considerations to a few components at a time. Egbert. The by a study of small components of the European entomologist. 7. selection. Competition. by the contradictory desiderata of generality. /. Bd. The multiplicity of models is imposed by the contradictory demands of a complex. 19. There fore. These conflicts are irreconcilable. Am. 1965. as hierarchically arranged "nested" models. displacement in a patchy environment. not nature. R. and character habitat R. they jointly produce robust theorems. and precision. mammal 3. Leigh. F. Press 1963.STRATEGY OF MODEL BUILDING IN POPULATION BIOLOGY 431 Unlike the theory. each provides an interpre tation of the sufficient parameters of the next higher level where they are taken as given. 96 (891). Academic Organization Temporal as revealed of pr?dation C. MacArthur. H. In Figure 3 we show schematically the relations among some of the models in the theory of community structure. 5. These models are related to each other in several ways :as coordinate alternative models for the same set of phenomena. even in systems which are complex. 91 (5) 293 pine sawfly. But the conflict is about method. S. for the individual models. Kerner. Theory environment. and R. by the need to understand and also to control. Levins. The Kenneth 8. BIBLIOGRAPHY 1. the yield of a fishery. diversity. realism. 4. 51 (3) 1207-1210. Canadian pr?dation E. the alternative approaches even of contending schools are part of a larger mixed strategy.

- Ecology & Evolution - Lecture 2
- 2species Levins Model
- Arent Greve
- The Interaction Between Predation and Competition
- Mapping Parameters & Variables
- World Prehistory _Grahame Clark
- Towards Using Genetic Algorithms in Lossy Audio Compression 2008
- jaynes_history
- 485958 - BCT-FI Extracting FI Line Items Up to the Current Date
- 1-s2.0-S2212567114003499-main(1)
- Jameson-FEAR AND LOATHING IN GLOBALIZATION
- curriculum map gagen kaitlin
- Introduction to Evolutionary Computing
- Probabilistic Fatigue Methodology and Wind Turbine Reliability_ThesisPhd_lange
- Alcatel BSS - B9 Telecom Parameters Dictionary (209240000e04)
- A System's Wave Function is Uniquely Determined by Its Underlying Physical State; Roger Colbeck, Renato Renner
- Full Text
- Confidential Interval
- gettingStartedV40_006b
- Biomems article
- Method and System for Monitoring and Analyzing Energy Consumption in Operated Chemical Plants US8554381
- Day 1 First Day Lecture and Chapter 1 Spring 2012
- Optimizing a Corrugated Horn
- ZTE - UMTS Radio Parameter
- 20349949 Construction Planning
- New Microsoft Word Document
- adztooldoc
- BS in Wildlife and Wildlands Conservation

Skip carousel

- The Optimizing Multiple Travelling Salesman Problem Using Genetic Algorithm
- Comparative Analysis of Optimization Algorithms Based on Hybrid Soft Computing Algorithm
- Survey Paper on Designing of FIR Filter using Genetic Algorithm
- tmpA05C
- tmp7B9A.tmp
- Intrusion Detection System Using Genetic Algorithm
- tmpF060

Sign up to vote on this title

UsefulNot usefulClose Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Levins 1966 StrategyEcologyModels will be available on

Loading