Individual variation in life history traits
A striking feature of long-term studies focussing on lifetime reproductive success in long-lived species is the consistency with which they have reached the conclusion of differences in “individual quality”. Analyses based on longitudinal data from kittiwakes led to the same conclusion with the same consistency.
A possible approach to assessing individual heterogeneity is to address correlations between fitness components without using an experimental approach. For example, we looked at the relationship between breeding activity and success in year t and survival from t to t+1 . Similarly, we investigated the relationship between and breeding activity and success in year t and in the following year ( t+1 ). These analyses were conduted using multistate capture-recapture models. Our results provided evidence of a positive relationship between fitness components (as illustrated by the “regession lines” between survival (S) and reproduction (R) in the schematic figure below, or between reproduction in year t (Rt) and in year t+1 (Rt+1)). Nonbreeders (i.e., individuals with the poorest reproductive performance) have the lowest survival rate, and also the lowest probability of breeding in the following year.
This is likely to have major implications for studies of the influence of age on life history traits (e.g., senescence).
"HETEROGENEITY'S RUSES"
Several investigators have suggested that heterogeneity in survival among individuals is likely to lead to spurious conclusions concerning effects of age on life history traits. If individuals in the population have different survival probabilities (e.g., “higher quality" individuals having higher survival probability), those with lower survival probability die younger. This results in "selection" within cohorts ("within-generation phenotypic selection"). The proportion of individuals with different survival probability varies with age, more precisely the proportion of individuals with lower survival decreases. This is likely to result in apparent increase in survival with age in the population, when survival doesn’t vary with age within each individual. A phenomenon of particular interest that one suspects to be masked by "selection" is senescence. This phenomenon has been extensively studied in humans, sometimes addressed with captive insects, and virtually never with data from wild animal populations.
The process is illustrated below. Assume there are 3 groups of individuals with different survival probabilities (0.85, 0.75, 0.65) in the population. The number of individuals in each group is 100. The number of groups, their size, or the survival values don’t matter. At the time individuals enter the study (e.g., at birth, or when they recruit into the breeding segment of the population), the survival rate at the population level (i.e., the “average” survival probability) is 0.75. Importantly, survival doesn’t vary with age at the individual level. 1 time step later (e.g., 1 year) the size of the groups is smaller (some individuals died): 85, 75 and 65, respectively. Of course, the decrease in the size of the group with lower survival is more substantial. The proportion of individuals with lower survival in the population decreases (65/225 = 0.289 versus 100/300 = 0.333 initially). At the population level, survival is now 0.76. 5 time steps later, the population-level survival rate is 0.79 (versus 0.75 initially). The proportion of individuals with lower survival is now (12/81 = 0.148). Survival increased at the population level, but did not increase at the individual level.
MORE ON "HETEROGENEITY'S RUSES"
As illustrated above with multistate models, our results provided evidence of a positive relationship between reproduction and survival. Many studies have provided evidence that life-history traits are correlated. Whenever a correlation between reproduction and survival exits, regardless of the sign of that correlation, within-generation selection can also mask genuine effects of age on reproduction. Population-level reproductive patterns may not reflect the influence of age on reproduction within individuals.
Here is an example where reproduction and survival are positively correlated. Assume that we are assessing the influence of age on reproductive performance (“perf”; e.g., number of chicks fledged), and we find an age-related increase at the population level (see below). Such a pattern at the population level can be obtained in different ways. Reproductive performance may improve with age within each individual. In this case the population-level pattern reflects the genuine influence of age within individuals. However, a second possible scenario relies on the idea that the population includes a group of individuals with higher survival and higher probability of breeding successfully, and a group with lower survival and lower probability of success. Individuals with lower survival die younger, and the “average” probability of breeding successfully increases. There is an apparent “positive” influence of age on reproductive performance perceived at the population level, but such an increase is not valid at the individual level.
We addressed this issue using data from kittiwakes (Cam et al. 2002, Link et al. 2002ab), and compared the influence of age on fitness components at the individual level and the population level. Selection (including within-generation phenotypic selection) operates at the individual level. However, previous approaches to investigating variation in fitness relied on grouping individuals according to trait values and estimating group-specific fitness components, assuming homogeneity within groups. Such approaches may permit detection of genuine effects of age on fitness components, provided that the criteria used to classify individuals adequately reflect individual quality. A difficulty unveiled in studies of humans is that measurable individual characteristics seldom capture heterogeneity in a satisfying manner. This led demographers to develop statistical inference methods incorporating individual heterogeneity in survival. In these “frailty” models, each individual has “its own mortality risk”. As at the individual level “survival” is either “0” or “1”: it is impossible to measure an individual mortality risk using data from a single individual . Conceptually though, it is natural to think of a distribution of underlying survival probabilities in populations. For example, in biomedical studies it is often assumed that individuals have different a priori sensitivity to treatment. Assessment of the distribution of these individual values requires probabilistic approaches and a model . The models we specified permitted us to assess individual heterogeneity in survival and breeding probabilities, respectively, and the correlation between them at the individual level. The approach to estimation we used was based on Markov chain Monte Carlo to fit flat prior Bayesian models.
RELEVANT PAPERS
- Cam, E., W.A. link, Cooch, E.G. Monnat, J.Y, Danchin, E. 2002. Individual covariation between life-history traits: seeing the trees despite the forest. American Naturalist 159: 96-105
- Link, W.A., Cooch. E.G., Cam. E. 2002a. Model-based estimation of individual fitness. Journal of Applied Statistics 29: 207-224
- Link, W.A., Cam, E., Nichols, J.D., and Cooch, E.G. Of BUGS and birds. 2002b. Markov chain Monte Carlo for hierarchical modeling in wildlife research. Journal of Wildlife Management 66: 277-291
