Anim. Behav., 1996, 51, 225–228

COMMENTARIES

How to measure inclusive fitness, revisited

JEFFREY R. LUCAS*, SCOTT R. CREEL† & PETER M. WASER* *Department of Biological Sciences, Purdue University †Rockefeller University Field Research Center

(Received 14 September 1994; initial acceptance 31 October 1994; final acceptance 13 March 1995; MS. number: -1136)

An individual’s inclusive fitness is derived by verbal definition) is equal to its Darwinian fitness augmenting its traditional Darwinian fitness by stripped of the effects of helpers. This is equival- certain components, and stripping it of others ent to the reproductive success of a breeder in (Hamilton 1964). The component to be added is an unaided pair. But in obligately cooperative the sum of all effects of the individual on his species, unaided pairs cannot rear young (e.g. neighbours, weighted by the coefficient of genetic dwarf mongooses, Helogale parvula: Creel 1990a; relationship (r) between them. In the original African wild dogs, Lycaon pictus: Malcolm & derivation of inclusive fitness, the component to Marten 1982). Under the original verbal definition be subtracted was verbally defined as ‘all compo- of e0, the inclusive fitness of breeders in groups of nents which can be considered as due to the any size would therefore be zero. Bringing the individual’s social environment’ (Hamilton 1964). verbal definition of e0 into line with its original It is not widely appreciated that this verbal defi- algebraic definition resolves the bizarre problem nition of the component to be subtracted does not of zero inclusive fitness for breeders. But as we agree with its algebraic definition. Creel (1990a) discuss here, it raises two new and conceptually used Hamilton’s algebraic definition of inclusive thorny issues. fitness to show that the component to be sub- (1) Calculation of e0 is not always irrelevant tracted, e0, is actually equal to the average effect when using Hamilton’s rule. Creel (1990a) sug- of one individual on others’ fitness (dT0,in gested that Hamilton’s rule (that help is favoured Hamilton’s 1964 formulation). Thus, the calcu- when rb>c, where b is the effect of a helper lation of e0 requires that we know the fraction of on breeder reproductive success, r is the genetic individuals in the population that provide help relatedness between helper and breeder, and c is and the mean amount of help provided per helper. the cost of helping to the helper) is unaffected by e0 is simply the product of these two numbers (see the definition of e0, because e0 is stripped from the below). Subtracting e0 resolves what has become inclusive fitness of both breeders and helpers. known as the ‘double accounting’ problem Thus ‘e0 drops out of an inequality comparing (Grafen 1984; Brown 1987). inclusive fitness for alternative strategies’ (Creel The verbal definition of e0 is important, because 1990a). The implication is that the correct defi- it is primarily the verbal definition of inclusive nition of e0 is not relevant to a correct calculation fitness that has been put to empirical use (Grafen of Hamilton’s rule. However, this statement does 1982, 1984). The flaw in the original verbal defi- not hold if inclusive fitness effects are summed nition of e0 can be seen by applying it to the case over more than a single reproductive season and if of a breeder in an obligately cooperative species alternative decisions differ in their risk of mortal- (Creel 1990a). In a group with a single breeding ity. This is true because e0 is subtracted from each pair, the breeder’s inclusive fitness (using the old individual’s fitness each time during its lifespan that direct or indirect fitness components are Correspondence: J. R. Lucas, Department of Biological measured, that is, once every breeding season. We 0 Sciences, Purdue University, West Lafayette, IN 47907- can illustrate why e will not cancel out using a 1392, U.S.A. (email: [email protected]). simple example.

0003–3472/96/010225+04 $12.00/0  1996 The Association for the Study of Animal Behaviour 225 226 Animal Behaviour, 51, 1

Assume that if an individual dispersed it would Therefore the helper’s inclusive fitness is produce, on average, 0.5 offspring as a breeder in (r"p)#b. This will be negative whenever the a new group the current year but would have zero coefficient of relationship is less than the pro- probability of surviving another year. The inclu- portion of individuals in the population that are sive fitness of this individual would be 0.5"e0. non-breeding helpers. Alternatively, assume that the individual could From published demographic data (e.g. Stacey remain in its current group, and if it did so it & Koenig 1990) it appears that negative inclusive would survive the current year as a helper, helping fitness for helpers will be common, unless helpers its sibling produce 0.5 more offspring than it often obtain hidden direct reproduction. For 12 otherwise would have produced. Also, assume cooperatively breeding species ranging from lions, that this helper then bred the next year, pro- Panthera leo, to splendid fairy wrens, Malurus ducing 1.0 offspring and dying thereafter. The splendens, mean helper–breeder relatedness falls lifetime inclusive fitness of this option is between 0.23 and 0.47 (Macdonald 1980; (0.5#r"e0)+(1.0"e0). In this case, the fitness Woolfenden & Fitzpatrick 1984; Austad & components to be subtracted from dispersers’ and Rabenold 1985; Emlen & Wrege 1988; Mills 1990; non-dispersers’ fitnesses will not cancel out. More Packer et al. 1991; Creel & Waser 1994; also see generally, differences in mortality risks between Stacey & Koenig 1990). If the dichotomy between alternatives will cause differences in expected helping and breeding were absolute, this range of longevity, which will in turn cause differences in r-values would yield negative fitness for helpers in the relative weighting of e0 on lifetime inclusive groups larger than 2.6 (for r=0.23) to 3.8 (for fitness. We emphasize that since e0 relates to the r=0.47). Many cooperatively breeding mammals average ability of individuals to give (or equiva- live in group sizes large enough to yield negative lently to receive) help, it should be calculated on a fitness for helpers (e.g. African wild dogs with per-breeding-season basis, as we use it here. mean adult pack size of 11: Fuller et al. 1992). A number of recent analyses have shown that Modal group size for most cooperatively breeding future effects are important components in the birds is less than four (two, in most cases: Stacey estimation of inclusive fitness (e.g. Brown 1987; & Koenig 1990), but even for these species Mumme et al. 1989; Creel 1990b; Lucas et al., in many helpers will have a negative fitness unless press). In addition, differences in mortality risks relatedness is high. between alternative behaviour patterns (e.g. dis- Of course, the calculation of inclusive fitness is persal versus non-dispersal, or helping versus non- affected by direct fitness effects. Direct production helping) are likely to be nearly ubiquitous (Brown of offspring by helpers has recently been dis- 1987; Waser et al. 1994). Together, these data covered in several species (e.g. dwarf mongooses: suggest that an accurate evaluation of e0 is Keane et al. 1994; stripe-backed wrens: Rabenold warranted. et al. 1990). This direct reproduction will obvi- (2) Helpers often will have negative inclusive ously increase the inclusive fitness of helpers, but fitness. Applying the definition of e0 in a dynamic at least in dwarf mongooses, the number of young optimization analysis of dispersal decisions in produced by helpers is insufficient to give them cooperatively breeding dwarf mongooses, Lucas positive inclusive fitness (Lucas et al., in press). et al. (in press) found that subordinate helpers The role of the indirect component of inclusive of many age–sex classes had negative inclusive fitness on the evolution of helping behaviour has fitness. To illustrate this, assume that helpers do been controversial. Some authors argue that in- not breed (and breeders do not help), so that a direct effects are relatively unimportant (Zahavi helper’s fitness is entirely indirect (Brown 1987). 1974; Ligon 1981; Walters et al. 1992); others Indirect fitness is the product of a helper’s effect suggest that indirect effects cannot be overlooked on breeder reproductive success (b) multiplied by (Mumme et al. 1989; Emlen & Wrege 1991; Creel its relatedness to the breeder (r). Inclusive fitness is & Waser 1994). Although we have shown here obtained by stripping e0 from this indirect fitness, that the inclusive fitness of helpers can actually be where e0 is the product of an average helper’s negative, we want to emphasize that this does not effect on breeder reproductive success (b) multi- mean that indirect fitness effects will be unimport- plied by the proportion of individuals in the ant in the evolution of helping behaviour in these population that are helpers (p) (Creel 1990a). species. For example, we show elsewhere that, Commentaries 227 despite negative indirect fitness for subordinate ness to other group members, and so on. The dwarf mongooses, indirect effects can play a alternative genotypes that compete in the real major role in the evolution of social behaviour in world are likely to be more complicated, this species (Lucas et al., in press). and might for example influence their bearers’ Non-breeders will likely suffer negative fitness, tendencies to disperse conditional on some com- irrespective of their choice of behaviour, if the bination of, for example, dominance and related- frequency of helpers in a population is large. ness. An approach that explicitly allows the Indeed, the inclusive fitness of non-helping non- modelling of such conditional traits is dynamic breeders will always be negative whenever helping optimization; we have presented a model of exists. In itself, this is not relevant to the choice of dwarf mongoose dispersal that incorporates such behaviour; instead, we need to know the payoff to conditionalities elsewhere (Lucas et al., in press). helping relative to other alternatives available to a Since its formulation 30 years ago, the idea of non-breeder. Generally, the inclusive fitness of inclusive fitness has had enormous impact, but its non-helping subordinates will be more negative empirical application remains limited. The process than that of helpers. We need not be concerned of measuring inclusive fitness completely and cor- with where a fitness value falls relative to zero, rectly has been an iterative one. Among his listing only with where it falls relative to the fitness of of ‘misconceptions of kin selection’, Dawkins alternative phenotypes. (1979) discussed sources of confusion over the Several problems may remain with the esti- definition of r. Grafen (1982, 1984) pointed out mation of inclusive fitness. One of these deals with several common mis-definitions of inclusive fitness traits whose fitnesses are frequency-dependent. as well as the ‘double accounting’ problem dis- Hamilton’s (1964) derivation of inclusive fitness cussed here. Brown (1987) suggested that inclusive did not apply to such traits, as he indicated fitness might be usefully broken into direct and explicitly. The payoffs to traits like ‘helping’ and indirect, current and future components. The ‘breeding’ are, in reality, almost certainly most difficult components of inclusive fitness to frequency-dependent, so that models of inclusive conceptualize and estimate are the future effects; fitness like the one we discuss here are only Mumme et al. (1989) and Creel (1990b) discussed approximations of reality. the ways in which ‘future indirect’ fitness compo- D. C. Queller (personal communication) has nents might arise. Emlen & Wrege (1994) and raised a further question about the nature of e0. Creel et al. (1991) suggested partial solutions to Creel’s (1990b) approach, like prior discussions of the problem of estimating future fitness compo- how to measure inclusive fitness (e.g. Grafen 1982, nents, both direct and indirect. Lucas et al. (in 1984), assumed that the value of e0 is the same for press) show how dynamic optimization can be all behavioural genotypes. For example, in the used to provide a complete, lifetime accounting. above example comparing the inclusive fitnesses We believe that methods of estimating inclusive of hypothetical dispersers and non-dispersers, fitness are converging, as they must if tests of the same value of e0 is used to estimate both behavioural ecological hypotheses are to become genotypes’ inclusive fitnesses. Queller (personal quantitative. communication) suggested that the value of e0 associated with two different genotypes should be different. This is an important issue that we REFERENCES cannot resolve. Austad, S. N. & Rabenold, K. N. 1985. Reproductive A further problem that may not have been enhancement by helpers and an experimental inquiry adequately appreciated arises when one estimates into its mechanism in the bicolored wren. Behav. Ecol. the fitness of conditional strategies (see Parker Sociobiol., 17, 18–27. Brown, J. L. 1987. Helping and Communal Breeding in 1989). In the simple dispersal/non-dispersal Birds. Princeton, New Jersey: Princeton University example used above, the strategies modelled are Press. not sensitive to the state (other than the age) of Creel, S. R. 1990a. How to measure inclusive fitness. the potential dispersers: the genes whose spread Proc. R. Soc. Lond. Ser. B, 241, 229–231. is being modelled are genes that predispose Creel, S. R. 1990b. The future components of inclusive fitness: acounting for interactions between members their bearers to ‘disperse’ or ‘stay’ whatever their of overlapping generations. Anim. Behav., 40, 127– nutritional state, dominance, group size, related- 134. 228 Animal Behaviour, 51, 1

Creel, S. R., Monfort, S. L., Wildt, D. E. & Waser, in Mammals (Ed. by N. Solomon & J. French). P. M. 1991. Spontaneous lactation is an adaptive Cambridge: Cambridge University Press. result of pseudopregnancy. Nature, Lond., 351, Macdonald, D. W. 1980. Social factors affecting 660–662. reproduction amongst red foxes. In: The Red Fox: Creel, S. R. & Waser, P. M. 1994. Inclusive fitness and Symposium on Behavior and Ecology (Ed. by E. reproductive strategies in dwarf mongooses. Behav. Zimen), pp. 123–175. The Hague: Junk. Ecol., 5, 339–348. Malcolm, J. R. & Marten, K. 1982. Natural selection Dawkins, R. 1979. Twelve misunderstandings of kin and the communal rearing of pups in African selection. Z. Tierpsychol., 51, 184–200. wild dogs, Lycaon pictus. Behav. Ecol. Sociobiol., 10, Emlen, S. T. & Wrege, P. H. 1988. The role of kinship in 1–13. helping decisions among white-fronted bee-eaters. Mills, M. G. L. 1990. Kalahari Hyaenas: The Behav- Behav. Ecol. Sociobiol., 23, 305–315. ioural Ecology of Two Species. London: Unwin Emlen, S. T. & Wrege, P. H. 1991. Breeding biology of Hyman Press. white-fronted bee-eaters at Nakuru: the influence of Mumme, R. L., Koenig, W. D. & Ratnieks, F. L. W. helpers on breeder fitness. J. Anim. Ecol., 60, 309–326. 1989. Helping behaviour, reproductive value, and the future component of indirect fitness. Anim. Behav., 38, Fuller, T. K., Kat, P. W., Bulger, J. B., Maddock, A. H., 331–343. Ginsberg, J. R., Burrows, R., McNutt, J. W. & Mills, Packer, C., Gilbert, D. A., Pusey, A. E. & O’Brien, S. J. M. G. L. 1992. Population dynamics of African wild 1991. A molecular genetic analysis of kinship and dogs. In: Wildlife 2001: Populations (Ed. by D. R. cooperation in African lions. Nature, Lond., 351, McCullough & R. H. Barret), pp. 1125–1139. 562–565. London: Elsevier Applied Sciences. Parker, G. A. 1989. Hamilton’s rule and conditionality. Grafen, A. 1982. How not to measure inclusive fitness. Ethol. Ecol. Evol., 1, 195–211. Nature, Lond., 298, 425. Rabenold, P. P., Rabenold, K. N., Piper, W. H., Grafen, A. 1984. Natural selection, kin selection, and Haydock, J. & Zack, S. W. 1990. Shared paternity group selection. In: Behavioural Ecology: an Evol- revealed by genetic analysis in cooperatively breeding utionary Approach. 2nd edn (Ed. by J. R. Krebs & tropical wrens. Nature, Lond., 348, 538–540. N. B. Davies), pp. 62–84. Oxford: Blackwell Scientific Stacey, P. B. & Koenig, W. D. 1990. Cooperative Publications. Breeding in Birds. Cambridge: Cambridge University Hamilton, W. D. 1964. The genetical evolution of social Press. behaviour. J. theor. Biol., 7, 1–16. Walters, J. R., Doerr, P. D. & Carter, J. H., III. 1992. Keane, B., Waser, P. M., Creel, S. R., Creel, N. M., Delayed dispersal and reproduction as a life-history Elliot, L. F. & Minchella, D. J. 1994. Subordinate tactic in cooperative breeders: fitness calculations reproduction in dwarf mongooses. Anim. Behav., 47, from red-cockaded woodpeckers. Am. Nat., 139, 65–75. 623–643. Ligon, J. D. 1981. Demographic patterns and communal Waser, P. M., Creel, S. R. & Lucas, J. R. 1994. Death breeding in the green woodhoopoe, Phoeniculus pur- and disappearance: estimating mortality risks associ- puratus. In: Natural Selection and Social Behavior: ated with philopatry and dispersal. Behav. Ecol., 5, Recent Research and New Theory (Ed. by R. D. 135–141. Alexander & D. W. Tinkle), pp. 231–243. New York: Woolfenden, G. E. & Fitzpatrick, J. W. 1984. The Chiron Press. Florida Scrub Jay: Demography of a Cooperative- Lucas, J. R., Creel, S. R. & Waser, P. M. In press. breeding Bird. Princeton, New Jersey: Princeton Dynamic optimization and cooperative breeding: an University Press. evaluation of future effects. In: Cooperative Breeding Zahavi, A. 1974. Communal nesting by the Arabian babbler. Ibis, 116, 84–87.