Appendix 1. The sequential probability ratio test The female obtains a sequence X 1 , X 2 , X 3 , of observations on male quality. These observations are assumed to be independent and identically distributed given the male type. Let f H denote their common probability density function given that the male is helpful, and let f N denote their common probability density function given the male is not helpful. Let f (X ) Z i log H i fN (Xi ) (A1.1) be the log-likelihood ratio for the ith observation. For each n 1 set n W ( n) Z i . (A1.2), i 1 which is the sum of the log-likelihood ratios for all n observations gathered thus far. Let a and b be positive constants. Then if the female implements the sequential probability ratio test with absorbing barriers at –a and b she behaves as follows. She continues to sample while the process W (n) : n 0 remains between the barriers (i.e. a W (n) b ). Once the process hits one of the barriers she ceases sampling and decides on male type: if W (n) a the female decides that the male is not helpful, if W ( n) b the female decides the male is helpful. The error probabilities under this procedure are α = Prob(process absorbed at - a | male is helpful) (A1.3) and β = Prob(process absorbed at b | male is non-helpful) (A1.4). The continuous time limit. From now on we assume that, if the male is helpful, the X i have a normal distribution with mean H and variance 2 . If the male is not helpful they have a normal distribution with mean N and variance also 2 . Then the log-likelihood ratio for the ith observation is Zi 1 2 2 ( H N )[ 2 X i ( N H )] . (A1.5) Then, given the male is helpful, this random variable has mean E H Z i 1 2 2 ( H N ) 2 (A1.6) and variance Var H Z i ( H N ) 2 / 2 . (A1.7) 2 The analogous quantities when the male is not helpful are E N Z i 1 ( H N ) 2 (A1.8) Var N Z i ( H N ) 2 / 2 . (A1.9) 2 2 and We now suppose that there are n observation per unit time. When time t is multiple of 1 / n we can set Y (t ) W (nt ) . This random variable Y summarizes the useful information up until time t. Define and 2 by n ( H N ) 2 2 2 (A1.10) and 2 n ( H N ) 2 2 . (A1.11) Then E H Yt t and Var H Yt 2 t . (A1.12) 3 Similarly E N Yt t and Var N Yt 2 t . (A1.13) We now let n tend to infinity, allowing ( H N ) 2 / 2 to tend to zero so that the product of these two terms, v, remains constant. In this limit the process Yt : t 0 tends to a diffusion process with drift v / 2 and variance σ2= v. We assume that Yt : t 0 is such a diffusion in what follows. Error probabilities. Suppose that the process Yt : t 0 has absorbing barriers at –a and b. If the male is helpful the process is a diffusion process with drift and diffusion coefficient equal to . Thus by standard results, the probability of being absorbed at the barrier at –a is A AB 1 AB (A1.14) where A exp{2a / 2 } exp{a} (A1.15) and 4 B exp{2b / 2 } exp{b} (A1.16) (e.g. Taylor & Karlin 1998). Similarly B AB . 1 AB (A1.17) We can also use these relationships to express A and B in terms of the error probabilities as A 1 (A1.18) and B . 1 (A1.19) Mean times. The mean time to be absorbed at one of the boundaries can be expressed as H b(1 A) a(1 B) A (1 AB) (A1.20) (Taylor & Karlin 1998). Thus by equations (A1.15), (A1.16), (A1.18) and (A1.19) 5 2 v 1 log . 1 (A1.21) 2 1 (1 ) log log . v 1 (A1.22) H (1 ) log Similarly N Taylor, H.M. & Karlin S. 1998 An introduction to stochastic modelling. San Diego, CA Academic Press. Appendix 2. Results when females cannot gain information We consider first the special case when females are unable to gain information on male type. In this case females do not sample, but just mate with all males encountered. Thus equations (1) – (8) hold with H = N = 0, = 0 and = 1. As we now show, even in this restricted case both male types can be maintained in the population at evolutionary stability. Consider a population in which the proportion of helpful males is p and the proportion of non-helpful males is 1 – p. In this population the mean time for a male to encounter a female, 1 /( F ) , is determined by equations (1) – (5). To emphasise the dependence on 6 p we denote this mean time by T(p). Note that this mean time is the same for both types of male. Before finding the ESS we first investigate dependence of T(p) on p. Females encounter males at rate s M , where M is the proportion of males that are searching. Since females mate on encounter, and mating and caring takes unit time, the proportion of time that females spend searching is F 1 / s M 1 . 1 (1 / s M ) 1 s M (A2.1) A searching male takes mean time T ( p) 1 F (A2.2) to encounter a female. Thus by equations (A2.1) and (A2.2) T ( p) 1 sM . (A2.3) Non-helpful males spend all of their time searching. Thus the proportion of males searching is M (1 p) p H (A2.4) 7 where H is the proportion of time that helpful males spend searching. This proportion is given by H 1 / F 1 (1 / F ) . (A2.5) Thus by equations (A2.2), (A2.4) and (A2.5) M 1 p . T ( p) 1 (A2.6) From equations (A2.3) and (A2.6) T(p) is the positive root of the quadratic equation T ( p) 2 (1 1 s)T ( p) (1 s(1 p)) 0 , (A2.7) and hence is given by T ( p) (1 1 s) 2 4 ps (1 1 s) . 2 (A2.8) Thus dT s . dp (1 1 s ) 2 4 ps (A2.9) 8 This shows that T decreases as p increases. The intuitive reason for this is as follows. As the proportion of males that care increases there is a decrease in the proportion of males that are searching at any time. Thus it takes longer for females to find mates, so that the number of females that are searching increases. This means that it is easier for any single male to find a mate, so that search times are decreased. We now examine which type of male does best in this population, and how this depends on p. By equations (6) and (7) the reward rates of the two types of male are H ( p) BH 1 T ( p) and N ( p) BN . T ( p) Thus non-helpful males have a higher reward rate if and only if BN R( p) , BH where 9 T ( p) . 1 T ( p) R( p) Since T(p) is a decreasing function of p , so is R(p). There are thus three possible cases. Case 1: BN R(0) BH In this case we have BN R( p) BH for all p so that non-helpful males always have higher fitness. Thus p* = 0 is the only ESS. Case 2. R(1) BN R(0) . BH In this case helpful males do better when rare and can invade. Similarly, non-helpful males do better when they are rare and can invade. There is a unique p* such that R(p) > 10 BN/BH for p < p* and R(p) < BN/BH for p > p*. Thus the population will always evolve to the unique equilibrium p* Case 3. BN R(1) BH In this case BN R( p ) BH for all p. and the population evolves to p* = 1. 11