Kin Selection

How kin selection works

Darwin's problem with sterile worker bees was real. A worker honeybee lives six weeks, never reproduces, and dies stinging an attacker. A self-sacrificing allele should not reach the next generation through its bearer's offspring, because the bearer leaves none. Darwin guessed in The Origin (1859) that selection might act on the family rather than the individual. The puzzle waited a century.

William Hamilton solved it in 1964 with one inequality. An allele that costs you C personal offspring while letting you help a relative produce B extra offspring spreads when rB > C, where r is the coefficient of relatedness — the probability that the relative shares the allele by descent. A copy propagates through the relative with probability rB, the donor loses C copies through foregone reproduction. The allele wins when rB exceeds C.

The trick is to count copies of the gene, not offspring. From a gene's point of view, a sister or a daughter is a probabilistic vehicle for copies. Helping a sister produce four extra nieces costs you one daughter is a winning trade if (sister-niece r) × 4 > (own-daughter r) × 1. Selfish vs altruistic behaviour reduces to arithmetic on inheritance probabilities.

The coefficient of relatedness

Relatedness is the probability that two individuals share a given allele identical by descent. In a diploid sexual species:

• Parent and offspring share r = 0.5. Each child has half their genome from each parent.
• Full siblings share r = 0.5 on average. Each gets half their alleles from each shared parent, so the expected overlap with a sibling is 50%.
• Half-siblings share r = 0.25. Same on the shared-parent side, none on the unshared side.
• Grandparent and grandchild share r = 0.25.
• First cousins share r = 0.125. Two halvings down the family tree.
• Identical twins share r = 1. The whole genome.

Haplodiploid Hymenoptera break this scheme. Males develop from unfertilised eggs and carry one copy of every gene; females develop from fertilised eggs and carry two. A diploid female sister inherits her father's entire haploid genome (since he has no other genome to choose from) plus half her mother's. Sisters therefore share their father's side with probability 1.0 (instead of 0.5) and their mother's side with probability 0.5, giving expected sister-sister relatedness of 0.75. A worker is more closely related to a sister than to her own daughter (0.5). Hamilton's rule therefore favours helping mum produce more sisters over producing your own daughters — which is exactly what worker bees do.

Kin selection vs group selection vs reciprocal altruism vs mutualism

	Kin selection	Group selection (naive)	Reciprocal altruism	Mutualism
Who benefits	Genetic relatives	Group members regardless of relatedness	Past or future reciprocator	Both partners simultaneously
Mechanism	Hamilton's rule rB > C	Differential group survival	Tit-for-tat, memory, recognition	By-product of cooperation
Vulnerable to cheats?	Limited — kin recognition	Strongly — cheats invade groups	Yes — defectors exploit	Limited — cheats hurt themselves
Requires kin recognition?	Yes (or population viscosity)	No	No, but requires individual recognition	No
Theoretical status	Mainstream consensus since 1970s	Largely rejected (Maynard Smith 1964)	Trivers 1971, well-supported	Standard ecology
Canonical example	Sterile worker bees, ground-squirrel alarm calls	Wynne-Edwards' population self-restraint	Vampire bat blood-sharing (Wilkinson 1984)	Mycorrhizae, gut microbiome
Equivalent to inclusive fitness?	Yes by definition	Sometimes mathematically equivalent	No — separate mechanism	No

The boundaries blur in practice. Reciprocal altruism between relatives is just kin selection plus repeated interaction. Mutualism between species can look like reciprocal altruism without the gene-counting. Group selection mathematically reduces to kin selection in viscous populations because group-mates tend to be relatives. The point of the table is not to police categories but to show that altruism has multiple sustainable evolutionary routes; kin selection is the dominant one when r is high.

Worked example — Hamilton's rule for worker bees

The worker honeybee is the textbook calculation. Two reproductive options:

1. Reproduce. Lay haploid eggs (worker can't mate) that develop into sons. r = 0.5. Genetic return per son: 0.5.
2. Help the queen. Forgo personal reproduction, labour to raise sisters. r = 0.75. Genetic return per sister: 0.75.

One extra sister per worker yields return 0.75; one extra son yields 0.5. Helping wins. Across a colony of tens of thousands of workers each contributing tiny incremental benefits per sister, the inclusive-fitness payoff of the worker caste is enormous.

The arithmetic shifts when queens mate with multiple drones (typically 10–20 in honeybees): workers are full sisters only 1/N of the time, half-sisters (N-1)/N. Effective r drops toward 0.25 + 0.5/N, and colonies shift to worker policing — workers destroying eggs from other workers because those would be half-related (r = 0.25) rather than full nieces (r = 0.375). Ratnieks and Visscher (1989) measured this: in multiply-mated colonies, workers eat unrelated worker eggs almost as fast as they are laid.

Real-world cases

• Honeybee colonies (Apis mellifera). 30,000–60,000 sterile workers, all daughters of one queen. Reproductive division of labour locked in by the haplodiploid 0.75 sister-sister relatedness. Worker policing of unrelated worker eggs.
• Belding's ground squirrels. Sherman (1977) showed alarm calls are given mostly by philopatric females living next to relatives; dispersing males rarely call. The pattern matches r-weighted predicted call rates.
• Florida scrub jays. Younger birds delay breeding to help parents raise siblings; helpers measurably increase nest success. Helping is rare in unrelated populations.
• Naked mole rats. Eusocial mammals: a single breeding queen, 100–300 non-breeding workers in colonies of close relatives (inbreeding pushes effective r above 0.8). Only known eusocial mammals.
• White-fronted bee-eaters. Helpers at the nest preferentially help close kin; given a choice, they help full siblings over half-siblings, half-siblings over cousins, in proportion to r.
• Slime mould (Dictyostelium discoideum). Single-celled amoebae aggregate into a fruiting body where 20% form a non-reproductive stalk holding up the 80% that become spores. Cheats exist; populations resist them through kin-discrimination genes.
• Plants — kin recognition. Arabidopsis seedlings allocate fewer roots when grown next to siblings vs strangers (Dudley & File 2007). The mechanism uses root exudates as kinship cues.

Variants and refinements

• Inclusive fitness. Hamilton's broader formulation: an individual's total reproductive success counts personal offspring plus relatives' offspring weighted by r. Kin selection is the special case where alleles spread by altruistic acts toward kin.
• Greenbeard genes. A theoretical case where an allele encodes both the trait and the recognition cue — a green beard says "I'm a copy, help me." Documented in fire ants (Gp-9 locus) and slime moulds.
• Population viscosity. When dispersal is limited, neighbours are statistically relatives. Kin selection works on geographic proxies for r without explicit recognition.
• Multilevel selection. A formalism that decomposes selection into within-group and between-group components. Mathematically equivalent to inclusive fitness in many cases (Queller 1992).
• Ecological constraints. Helping at the nest is often favoured when independent breeding is hard — saturated habitat or dangerous dispersal raise C, making rB > C easier to meet.

Common pitfalls

• "Kin selection equals group selection." They are different in framing and history. Kin selection acts on alleles in related individuals via Hamilton's rule; naive group selection assumed groups self-restrain. Kin selection (Hamilton 1964) is mainstream; naive group selection (Wynne-Edwards 1962) was rejected by Maynard Smith's 1964 critique. They can be mathematically reformulated into each other in some cases, but conflating Hamilton with Wynne-Edwards misses the whole point of the 1964 paper.
• "Animals must consciously calculate r." They don't. Selection acts on alleles whose phenotypic effects produce behaviours that satisfy rB > C on average. The animal needs only to behave in a way that approximates the optimum; the math runs in the population, not in the animal's head.
• "Altruism evolves 'for the good of the species'." No. Species are not units of selection in the standard view. Kin selection works through gene copies; species-level benefits are usually side effects.
• "Haplodiploidy is necessary for eusociality." It is helpful but not necessary. Diploid termites and naked mole rats are eusocial with r = 0.5 (or higher under inbreeding). Ecological factors — defensible nests, valuable resources — also matter.
• "Hamilton's rule has been falsified." It hasn't. Nowak, Tarnita & Wilson (2010) argued it was redundant; 137 biologists replied that it remains the predictive workhorse for empirical work. The rule's assumptions are restrictive (linear costs and benefits, weak selection, large population) but extensions handle the deviations.
• "Kin selection explains all altruism." No. Reciprocal altruism, mutualism, manipulation, and indirect reciprocity all contribute. Kin selection is dominant when r is high; other mechanisms take over when r is low.