Estimating fitness consequences of dispersal: a road to ‘know-where’? Non-random dispersal and the underestimation of dispersers’ fitness
CNRS – Université de Lyon, Université Lyon 1, CNRS UMR 5558, Department of Biometry and Evolutionary Biology, 43 Boulevard du 11 novembre 1918, Bâtiment Gregor Mendel, F-69622 Villeurbanne cedex, France; and
1Many studies investigating fitness correlates of dispersal in vertebrates report dispersers to have lower fitness than philopatric individuals. However, if dispersers are more likely to produce dispersing young or are more likely to disperse again in the next year(s) than philopatric individuals, there is a risk that fitness estimates based on local adult survival and local recruitment will be underestimated for dispersers.
2We review the available empirical evidence on parent–offspring resemblance and individual lifelong consistency in dispersal behaviour, and relate these studies to recent studies of fitness correlates of dispersal in vertebrates.
3Of the 12 studies testing directly for parent–offspring resemblance in dispersal propensity, five report a significant resemblance. The average effect size (r) of parent–offspring resemblance in dispersal was 0·15 [95% confidence interval (CI) = 0·07–0·22], with no difference between the sexes (average weighted effect size of 0·12 (0·08–0·16) and 0·16 (0·11–0·20) for females and males, respectively). Only three studies report data on within-individual consistency in dispersal propensity, of which two suggest dispersers to be more likely to disperse again.
4To assess the magnitude of fitness underestimation expected for dispersing individuals depending on the heritability of dispersal distance and study area size, we used a simulation approach. Even when study area size is 10 times the mean dispersal distance, local recruitment per breeding event may be underestimated by 4–10%, generating a potential difference of 4–60% in average lifetime production of recruits between dispersing and philopatric individuals, with larger differences in long-lived species.
5Estimates of both fitness correlates of dispersal and parent–offspring resemblance or within-individual consistency in dispersal behaviour have been reported for 11 species. Although some comparisons suggest genuine differences in fitness components between philopatric and dispersing individuals, others, based on adult and juvenile survival, are open to the alternative explanation of biased fitness estimates.
6We list three potential ways of reducing the risk of making wrong inferences on biased fitness estimates due to such non-random dispersal behaviour between dispersing and philopatric individuals: (a) diagnosing effects of non-random dispersal, (b) reducing the effects of spatially limited study area and (c) performing controlled experiments.
Most studies investigating fitness consequences of dispersal use comparisons of fitness estimates between philopatric and dispersing individuals (e.g. Hansson, Bensch & Hasselquist 2004; Pasinelli, Schiegg & Walters 2004 for recent studies; see Bélichon et al. 1996 and Supplementary material, Appendix S1 for a review). However, comparisons of fitness traits between philopatric and dispersing individuals are not free from problems. Here we highlight a major problem that could potentially affect the interpretations of such comparisons. Lifetime reproductive success (LRS) estimates of dispersing and philopatric individuals may be subject to systematic biases due to within-individual consistency and parent–offspring resemblance in dispersal behaviour. Although these problems are known (e.g. Greenwood, Harvey & Perrins 1979; Clobert et al. 1988; McCleery & Clobert 1990; Verhulst & van Eck 1996; van der Casteele 2002; see in particular Bélichon et al. 1996), they are frequently ignored. The aim of the present study is therefore to shed light on the potential importance of these problems in assessing individual dispersal fitness consequences.
We first review some of the most recent studies investigating fitness consequences of dispersal. As data on LRS and dispersal derives mainly from studies on birds and mammals, we restrict our review to vertebrates. Secondly, to assess the importance of biased fitness estimates with respect to dispersal status of individuals, we review all vertebrate studies investigating within-individual consistency and parent–offspring resemblance in dispersal behaviour. Thirdly, we use a simulation approach to quantify possible biases in fitness estimates with respect to individual dispersal status depending on the magnitude of within-individual consistency or parent–offspring resemblance in dispersal distance and species longevity. Fourthly, we review all cases for which both fitness correlates of dispersal and within-individual consistency or parent–offspring resemblance in dispersal have been investigated. We wish to point out that our main aim is to illustrate the potential impact of the problems highlighted, and not to criticize individual studies. Last, we review and list some methods to (i) evaluate whether fitness estimates may be biased; (ii) reduce the risk of making erroneous inferences on observed patterns; and (iii) possibly circumvent the problem of biased fitness estimates.
Fitness of dispersing and philopatric individuals
Because survival consequences of the movement phase of dispersal are notoriously difficult to study, especially in highly mobile species, most studies have focused upon the consequences at and following the establishment phase (Bélichon et al. 1996). Here, we review some of the most recent studies comparing fitness of dispersing and philopatric individuals following settlement (see Appendix S1 in Supplementary material; for earlier studies, see Bélichon et al. 1996). About half the comparisons of separate fitness components (survival and fecundity) show no difference between philopatric and dispersing individuals (68 comparisons of 133, in studies reviewed both in Bélichon et al. 1996 and this paper), and an equal number of studies report lower and higher fitness components for dispersing and philopatric individuals (30 and 35 comparisons, respectively).
Less than half (35 of 76) of the studies reviewed in Bélichon et al. (1996) and this paper (Supplementary material, Appendix S1) have analysed reproduction and survival simultaneously, and many studies focus upon a few independent fitness traits reflecting a particular process (e.g. mate or nest site acquisition, clutch size, recruitment). However, because dispersers and residents may adopt different lifetime strategies (e.g. Julliard, Perret & Blondel 1996; Spear, Pyle & Nur 1998; Marr et al. 2002), and compensation may occur between different fitness components (Bélichon et al. 1996; Lemel et al. 1997; Clobert et al. 2001), comparisons of separate fitness traits may be misleading. Preferably, one should compare lifetime number of recruits produced (i.e. including adult survival prospects and recruit production for each breeding attempt) but such data are not always obtainable. We investigated specifically studies comparing LRS. Actually, only a small fraction (11, i.e. 14%) of the 76 studies are based on estimates of LRS (Bélichon et al. 1996; Supplementary material, Appendix S1) and most (10) are recent ones (Supplementary material, Appendix S1). Of these 11 studies, eight (i.e. 73%) conclude that dispersers have lower LRS estimates than philopatric individuals at least for one sex, and only two conclude the reverse. Males and females were equally likely to show lower LRS estimates for dispersers (five studies for each sex) or no difference (three and four studies for males and females, respectively). However, these comparisons rarely allow a discrimination of direct consequences of dispersal from consequences of other processes to which dispersal is correlated (e.g. social status or phenotypic quality). In fact, all but five studies reviewed in Bélichon et al. (1996) and this paper are correlative. This confusion between correlates and consequences of dispersal has been noted earlier by several authors (Greenwood et al. 1979; Clobert et al. 1988; McCleery & Clobert 1990), but is still ignored frequently.
Individual consistency and parent offspring resemblance in dispersal propensity
Theoretical models of dispersal evolution commonly assume a genetic determination of dispersal propensity (Johnson & Gaines 1990; McPeek & Holt 1992), a certain level of heritability of dispersal being required for dispersal strategies to respond to selection (Roff & Fairbairn 2001). In vertebrates, however, dispersal is considered commonly as a highly plastic, state-dependent behaviour, determined by multiple internal, social and environmental factors (intraspecific competition, habitat variability, etc.) with low heritability (Ims & Hjermann 2001). The complex nature of dispersal as a trait and of its determinism, however, does not preclude phenotypic resemblance in dispersal behaviour between individuals within families, and between dispersal events within individuals. Both a genetic basis of dispersal and common parental or environmental effects on dispersal may cause some individuals and families to be more dispersive than others.
Individual consistency and parent–offspring resemblance in dispersal propensity are two key issues for a reliable estimation of fitness with respect to dispersal status. First, if individuals display a lifetime consistency in dispersal propensity, dispersers are more likely to disperse again, and thus leave the study area and be considered dead, than philopatric individuals. Such differential dispersal propensity will lead to the underestimation of local adult survival of dispersing compared to philopatric individuals (Fig. 1). Secondly, if offspring resemble parents in terms of dispersal propensity, dispersing parents produce young that are more likely to disperse, and thus leave the study area and be considered dead, than philopatric parents. As a result, local recruitment rate of dispersers’ offspring will be underestimated in comparison to that of philopatric individuals (Fig. 1). In both cases, differences are only apparent, i.e. linked to the spatial scale under consideration and the impossibility of distinguishing emigration out of the study area from mortality, and depend upon the degree of within-individual consistency and parent–offspring resemblance in dispersal propensity. The question is then how frequent and how strong are within-individual consistency and within-family similarities in dispersal propensity?
We searched the literature for studies investigating within-individual consistency and parent–offspring resemblance in dispersal, movement or exploratory behaviour using the Web of Science search engine (for key-words searched, see Supplementary material, Appendix S2). We found only three studies investigating within-individual consistency (references 1–3 in Table 1a) and 12 studies on 11 species investigating parent–offspring resemblance (references 6–17 in Table 1b) in dispersal distance or probability. Most studies are recent and were conducted on birds, reflecting a recent increased interest in the question of dispersal heritability in this taxon (Table 1). In addition, indirect support for within-family resemblance and within-individual consistency in dispersal behaviour is provided by evidence for a genetic basis of traits linked to dispersal (Trefilov et al. 2000; Sinervo & Clobert 2003; see also Sinervo et al. 2006; Myers & Krebs 1971) and correlation between dispersal and heritable personality traits (e.g. Dingemanse et al. 2003; Drent, van Oers & van Noordwijk 2003).
Table 1. Empirical evidence for non-random dispersal behaviour: (a) intraindividual consistency and (b) parent–offspring resemblance. We consider as a single study all publications on the same species in the same population within a subtable (a or b). Taxa: B: bird, M: mammal; h2 refers to heritability (for within-family resemblance) and R to repeatability (for within-individual consistency). Studies where evidence for or against non-random dispersal was indirect are in italics. When several relations were tested in the study (e.g. by parent or offspring sex), values are detailed: p: parent; o: offspring; f: father; m: mother; s: son; d: daughter. Effect sizes (Pearson's r) were calculated for parent–offspring relationships based on statistical information. For studies reporting log-likelihood or Fisher's exact tests, we used χ2 values as an approximation before calculating r
Individual consistency in dispersal behaviour has apparently been investigated rarely, as we found only three studies on bird species testing for within-individual consistency in dispersal propensity (Table 1a). In great tits Parus major L. and collared flycatchers Ficedula albicollis T., dispersing adults are more likely to disperse again in the next years compared to philopatric ones and such breeding dispersal occurred at the scale of the study area, while no such relationship was found in black kites Milvus migrans Boddaert. Two additional studies suggest indirectly no within-individual consistency in dispersal distance (see Table 1a). More data are needed before any conclusion can be drawn on the frequency and importance of within-individual consistency in dispersal behaviour and its effects on biasing fitness estimates of dispersers.
Of the 12 studies testing directly for parent–offspring resemblance in dispersal propensity, five report a significant resemblance (Table 1b). The remaining seven studies report no resemblance, but in many cases sample size was low (see Table 1b). Furthermore, parent–offspring resemblance in dispersal behaviour has been suggested in three other studies (McCleery & Clobert 1990; Julliard 1996; Orell et al. 1999). Based on the statistics reported, we calculated average effect sizes r and weighted r through Z-transformations (Cooper & Hedges 1994) for parent–offspring relationships for each study (Table 1b). At the level of study (n = 12), average effect size of parent–offspring resemblance was 0·15 [95% confidence interval (CI) = 0·07–0·22; based on mean effect size per study]. Because sex-biased dispersal is common in vertebrates (Greenwood 1980), one could expect effect sizes of parent–offspring resemblance to differ between sexes. The effect size r of parent–offspring relationship was, however, similar for both sexes: 0·12 (CI = 0·07–0·21) and 0·17 (CI = 0·04–0·32) for females and males, respectively, using one observation per study (with priority given to truly sex-specific relationships, e.g. mother–daughter over parent–daughter). The sex-specific weighted effect sizes were similar (females: mean = 0·12, CI = 0·08–0·16; males: mean = 0·16, CI = 0·11–0·20). Clearly, our results suggest that moderate parent–offspring resemblance in dispersal propensity may exist in many species and therefore we should be aware of its potential existence when estimating fitness consequences of dispersal, even in cases of apparent equal or higher fitness estimates for dispersing compared to philopatric individuals (Fig. 1). How frequent and how strong this parent–offspring resemblance is among species, however, remains an open question, in particular because of the limited number of studies reviewed here and the possibility of a publication bias in favour of significant relationships. However, the results show that it would be useful to ascertain whether the absence of parent–offspring resemblance is genuine or results from a small sample size or study area.
Whether parent–offspring resemblance and within-individual lifetime consistency in dispersal propensity arise from a genetic component of dispersal, early parental effects or a shared environment is not known in most cases (Massot & Clobert 2000; Massot et al. 2003; Clobert, Ims & Rousset 2004). However, the main issue here is not the causes of such phenotypic associations, but whether they exist and may generate biases in fitness estimates with respect to dispersal status of individuals. Apparent parent–offspring resemblance (or within-individual consistency) in dispersal may be created artificially by a simple ‘common environment’ effect, for example breeding site location within a limited study area, which constrains the array of distances observable from each site (van Noordwijk 1984). In such situations of no true parent–offspring resemblance, however, offspring of dispersers may still have a higher probability to leave the study area compared to those of philopatric individuals, for instance if dispersers are more likely to breed on the edge of the study area (see below). Parent–offspring resemblance in dispersal propensity may therefore not always be a prerequisite of biased offspring dispersal depending on parental dispersal. Another potential example of this is sex-biased dispersal, which could also create biases in fitness estimates based on number of local recruits produced when dispersing and philopatric individuals adjust offspring sex-ratio differently depending, e.g. on individual condition or environment. If dispersers overproduce the most dispersive sex, their fitness estimate is at risk of being underestimated. In this case, while no sex-specific parent–offspring (i.e. parent–son or parent–daughter) resemblance in dispersal propensity exists, an overall resemblance (i.e. mid-parent–mid-offspring) should, however, be suggested due to offspring sex-ratio bias.
Simulating the effect of parent–offspring resemblance on recruitment rates and LRS
To assess the magnitude of the effect of parent–offspring resemblance on estimates of recruitment and LRS of philopatric and dispersing individuals, we built a simple simulation model with parameter ranges mimicking those reported in empirical studies (e.g. Baker et al. 1995; Pasinelli et al. 2004 for study area size). We estimated the probability that recruits settle within the study area according to parental natal dispersal distance, degree of parent–offspring resemblance (varying between 0·0 and 0·6) and size of the study area (circular area of radius r varying between mean dispersal distance and mean distance × 6) (for more details on methods, see Supplementary material, Appendix S3). The model can also be applied to within-individual consistency in dispersal distance by computing the individual's subsequent dispersal distance rather than offspring distance (i.e. modelling the similarity of distances between dispersing events within individuals), and obtaining estimates of biases in local adult survival rates.
When there is no resemblance between parents and offspring in natal dispersal distance, no difference is expected in the probability of observing recruits (i.e. recruits staying within the study area) between philopatric and dispersing parents. However, as parent–offspring resemblance increases, the probability of observing offspring of philopatric parents is expected to increase, while the probability of observing offspring of dispersing parents is expected to decrease. Thus the difference in the probability of observing offspring between philopatric and dispersing parents should increase with increasing parent–offspring resemblance in dispersal. The simulations confirm this expected pattern for all scenarios of study area size (Fig. 2, from bottom to top of graphs). This difference in the probability of observing recruits can reach up to 17% (Fig. 2a for h2 > 0·40), not only because dispersers produce more dispersing, thus less observable, recruits (see upper right corners of each graph in Fig. 2), but also because philopatric individuals produce more philopatric, thus more observable, recruits (see upper left corners). Although the difference in observing recruits of philopatric and dispersing parents decreases when study area size increases (Fig. 2a–d), even for large study areas (10 times the mean dispersal distance: Fig. 2c) and moderate heritability levels (i.e. 0·3), this difference remains close to 6% (cf. class of distance 0–1 vs. classes 2–3 and above).
When calculating the difference in lifetime production of recruits between philopatric and dispersing individuals, the differences observed in yearly production of local recruits (i.e. recruits observed within the study area: Fig. 2) may multiply over the number of individual breeding events, i.e. the number of occasions when local juvenile survival rates can be underestimated, in relation with local adult survival. Clearly, the biased estimates of recruitment rate due to parent–offspring resemblance in dispersal in finite study areas will be amplified in estimates of lifetime production of recruits when within-individual consistency in dispersal is high (i.e. biased local adult survival rate), with the degree of amplification increasing with species longevity, as lifespan is a major important component of LRS (Stearns 1992). Using our simulations as an example, we obtained a level of underestimation of LRS for dispersing compared to philopatric individuals ranging from 2 to 40% depending on the species longevity, and estimated differences in yearly production of recruits and yearly breeding dispersal out of the study area (Table 2). The underestimation is two to three times higher in a long-lived compared to a short-lived vertebrate species (adult survival = 0·85 and 0·55, respectively: Table 2). Thus, although our simulation model relies on simplifying hypotheses (study area homogeneity and random habitat choice; no temporal emigration from the study area, etc.) and the quantitative predictions should thus be taken with care, it shows that non-random dispersal can possibly generate a large underestimation of LRS of dispersing compared to philopatric individuals when these are based on estimates of local survival rates.
Table 2. Magnitude of the underestimation of lifetime reproductive success (LRS) (i.e. total number of recruits produced over an individual's lifetime) for dispersing individuals compared to philopatric ones when parent–offspring resemblance or within-individual consistency in dispersal behaviour occurs (% underest. LRS). Five realistic types of life-history and associated average life-history trait estimates are presented from a very short- to a very long-lived species. Parent–offspring resemblance or within-individual consistency in dispersal behaviour lead to differences dj and da in the probability to stay within the study area for offspring and parents, respectively. We set dj here to 0·02, 0·05 or 0·10 (low to moderate values, see Fig. 2), and da to different fractions of each value of dj (0, dj/10, and dj/2) to model shorter breeding dispersal compared to natal dispersal. The detail of the computation of LRS estimates for philopatric and dispersing individuals, and the percentage of difference between both, is given in Supplementary material, Appendix S4
Reinterpreting comparisons of fitness estimates between dispersing and philopatric individuals
The consequences of within-individual consistency and parent–offspring resemblance in dispersal on a comparison of dispersing vs. philopatric individuals depend on the fitness components investigated. In general, all fitness components based upon adult or juvenile survival, such as lifespan, number of breeding attempts or recruitment rate, which are frequently used fitness parameters, are at risk of being underestimated in dispersers. We reviewed all studies with estimates on both fitness correlates of dispersal and within-individual consistency or parent–offspring resemblance in dispersal in the same species (Table 3). We found such data from 11 species; all but one were birds, with a total of 59 parameters compared between philopatric and dispersing individuals.
Table 3. Species for which data on both non-random dispersal behaviour and fitness comparisons between dispersing and philopatric individuals are available, and associated studies. Types of non-random dispersal behaviour: PO, parent–offspring similarity; WI, within-individual consistency in dispersal. Details of fitness differences are found in Bélichon et al. (1996) and Supplementary material, Appendix S1. Comparisons of fitness components that are open to the alternative explanation of biased fitness estimates due to non-random dispersal are indicated as ‘Yes’. Ambiguous data or interpretations, in particular due to small sample sizes, are indicated with ‘?’. The last species (pied flycatcher) is indicated in italics as the comparisons between philopatric and dispersing individuals in this species do not include fitness parameters at risk of biased estimates (i.e. involving local survival or recruitment rates). CS: clutch size, LRS: lifetime reproductive success
Type of non- random dispersal
Significant differences observed in fitness components
Possible bias in estimate?
(1) Sample sizes were low (see text). (2) Parent–offspring resemblance in dispersal behaviour was suggested (see Table 1a). (3) This study was based on radio-tracking data, and found no difference in survival between dispersing and philopatric individuals.
Studies on four species (great tit, collared flycatcher, red-cockaded woodpecker Picoides borealis Viellot and great reed warbler Acrocephalus arundinaceus L.) report both lower fitness estimates of dispersers vs. philopatric individuals and significant within-individual consistency and/or parent–offspring similarity in dispersal behaviour (species 1–4 in Table 3). These studies are particularly illuminating, as all are based on highly robust data. For all four species, fitness components based on adult and juvenile survival were lower for dispersing than philopatric individuals (Table 3; Supplementary material, Appendix S1 and Bélichon et al. 1996). Two of these species also displayed a significant within-individual consistency in dispersal propensity. Clearly, in these species comparisons based on juvenile or adult survival are open to the alternative explanation of biased fitness estimates due to parent–offspring resemblance and within-individual consistency in dispersal (see details in Table 3). Another five studies could potentially display a similar pattern, but results are more ambiguous as they are based on less robust (i.e. smaller sample size) non-significant parent–offspring and/or within-individual data (species 5–9 in Table 3). Finally, two studies seem to provide robust, unbiased comparisons, thus potential evidence for fitness costs linked to dispersal (species 10–11 in Table 3). Furthermore, in two species open to the alternative explanation of biased estimates (collared flycatcher: Pärt 1994; great reed warbler: Bensch et al. 1998; Hansson et al. 2004), dispersing males showed a lower ability to attract females and thus lower mating success than philopatric ones, and such differences cannot be explained by within-individual consistency or parent–offspring resemblance in dispersal. Therefore, for a given population, some parameters may be at risk of being underestimated in dispersers, while others may reveal genuine costs of dispersal. In total, 39 parameter comparisons suggest genuine differences in fitness components between philopatric and dispersing individuals while the remaining 20 are open to the alternative explanation of biased fitness estimates.
Can we diagnose and avoid biased fitness estimates?
Our review suggests within-individual consistency and parent–offspring similarity in dispersal behaviour to be a frequent phenomenon, and thus a potential problem when estimating fitness correlates of dispersal (Tables 1 and 3). We list three potential ways of reducing the risk of making wrong inferences on biased fitness estimates due to such non-random dispersal: (a) diagnosing effects of non-random dispersal; (b) reducing the effects of spatially limited study area; and (c) performing experiments.
A critical analysis of demographic data can help elucidating whether within-individual consistency or parent–offspring resemblance in dispersal may deflate fitness estimates of dispersers. Demographic relationships that may serve as an alarm are: (i) a difference in lifetime number of recruits between pairs breeding in the central vs. the peripheral parts of the study area; (ii) a difference between dispersing and philopatric individuals in lifetime number of recruits but not in number or quality of young produced (e.g. Wheelwright & Mauck 1998); and (iii) differences in adult survival, number of breeding attempts or lifespan and corresponding differences in LRS estimates (e.g. Pärt & Gustafsson 1989; Pärt 1990; Bensch et al. 1998; Hansson et al. 2004). In the first case, if dispersers are more likely to settle in the peripheral parts, there is a risk of biased fitness estimates with respect to individual dispersal history. Therefore, breeding site location (e.g. distance to the nearest edge of the study area) could be included together with dispersal status in statistical models when investigating fitness correlates of dispersal. In the third case, it is important to note that even slight non-significant differences in adult survival may inflate any differences in LRS estimates because, e.g. lifetime number of young is partly a pseudoreplicate of lifespan (see in Clutton-Brock 1988). Whether observed differences in LRS are driven by small differences in adult survival rates or annual production of recruits could be investigated by path analysis. If adult survival is a major component explaining the observed variation, the next step would be to include breeding site location in the same way as mentioned above. Biases in traits of recruits that are known to be associated with dispersal (e.g. sex, fledging date, fledging condition) may also be an indicator of a potential bias in offspring recruitment rate related to differential dispersal out of the study area.
Another approach may be to use indirect analytical methods to explore the potential biases in fitness estimates with respect to individual dispersal status. For instance, it is possible to compute dispersal probabilities needed to set equal LRS measures among individuals of different dispersal status, and then test whether resulting dispersal probabilities are plausible (see Box S1, Supplementary material). In a different line, several studies have used an indirect method to estimate dispersal and recruitment rates of unnoticed individuals dispersing out of the finite study area (e.g. Barrowclough 1978; Bowen et al. 1989; Sandell et al. 1990; Baker et al. 1995). The general idea developed in these studies is to correct the proportion of individuals observed at a given distance (or distance class) by accounting for the probability to detect (or recapture) individuals at that distance, depending on the non-sampled suitable habitat around the study area (see Box S1, Supplementary material). These correcting models can be used while accounting for individual (or parental) dispersal status and accordingly detect biases in dispersal probability or distance, and thus survival or recruitment rate. For example, if within-individual consistency or parent–offspring resemblance in dispersal behaviour exists, corrected estimates should reveal strong underestimation of numbers of individuals for large dispersal distances for dispersing compared to philopatric individuals. Many of the assumptions underlying these correcting methods (see Box S1) can only be fully tested in very well-studied systems. Nevertheless, simulations can be used to obtain confidence intervals and test the robustness of results by investigating the effect of assumptions (e.g. Baker et al. 1995). Although such analytical methods have to be scrutinized empirically in more detail before any firm recommendation can be given, they may prove useful in providing corrections of fitness estimates and diagnosing potential biases in observed LRS estimates due to non-random dispersal behaviour.
When non-random dispersal with respect to previous dispersal history is suspected, the sampling design has to be modified or adjusted to allow the detection of non-random dispersal and to reduce its influence on fitness components estimates. One obvious way to tackle the problem is to survey sites outside the core study area to assess the dispersing fraction of the population and its characteristics (Koenig et al. 2000; Winkler et al. 2005). Such effort should be sufficiently high and cover large areas with respect to potential dispersal distances outside the study area (e.g. Winkler et al. 2005). Given a standardized protocol, surveys outside the study area would allow the use of capture–mark–recapture methods to account for individual differences in resighting rates depending on dispersal status (Lebreton et al. 1992; Nichols & Kendall 1995).
Several other complementary ways can be used to deal with the problem of finite study area and dispersal. (1) Analyses could be restricted to data from the central part of a study area and use the remaining parts as resighting/detection area only (see e.g. Tinbergen 2005; Arlt & Pärt 2007). More generally, one could vary the size of the study area, either by ‘decreasing’ or increasing it (Koenig et al. 2000; Lambrechts et al. 2000; Zimmerman et al. 2007). In both cases, comparing fitness component estimates between dispersing and philopatric individuals in the restricted vs. extended area tests the robustness of differences with respect to potential biases due to non-random dispersal (see van der Casteele 2002; Tinbergen 2005). (2) Another method is to assess survival from independent data compared to data collected in the study area, for instance by uncoupling observations during the breeding season (defining dispersal events) from observations outside the breeding season and study area. Surviving individuals that are undetected in the study area must have dispersed to other breeding populations. This method can, however, be applied only to limited, specific systems where the breeding populations of a large area share a common, relatively small area outside the breeding season, such as some geese on their wintering grounds (e.g. van der Jeugd 2001). (3) The recent development of radio-tracking devices and automation of data recording now allows such tracking to also be used on small species (Naef-Daenzer, Widmer & Nuber 2001; Naef-Daenzer et al. 2005). Although the spatial range of radio-transmitters is currently limited, this technique may allow monitoring a sufficient sample of long-distance dispersers in the near future, provided signals are tracked over large areas. Satellite tracking of large species may also prove a suitable method in the future, given a large sample of individuals can be tracked at a moderate economic cost. The use of tracking methods in combination with experiments is promising. (4) Finally, other potential ways to estimate the effects of dispersal on fitness estimates include recording dispersal events and identify dispersers at the borders of the study area (Walpole et al. 2001) or using experimental enclosures interconnected by corridors (e.g. Boudjemadi, Lecomte & Clobert 1999; Gundersen, Andreassen & Ims 2002).
Even if fitness estimates are unbiased with respect to individual dispersal history, the problem of confounding factors (e.g. individual phenotypic quality) still remains. Although detailed knowledge of the study system, in particular factors affecting adult and juvenile survival may be helpful in identifying such confounding factors, only experimental manipulations testing the causality of relationships can provide unambiguous results. Experimental approaches, however, are often difficult to perform in the context of dispersal, in particular because of the tremendous effort needed. Artificial experiments creating patchy populations connected by corridors could potentially illuminate the problems of non-random dispersal, as dispersal can be controlled directly and possibly induced by modifying local conditions, individual or population characteristics (e.g. Boudjemadi et al. 1999; Gundersen et al. 2002; Cote & Clobert 2007). Carefully performed translocation experiments in the field followed by the monitoring of the forced dispersers can also be useful in testing for differential dispersal behaviour according to previous dispersal status and its subsequent potential fitness consequences. Although individual motivations to disperse will differ from natural situations, such translocations can provide means to test for the costs of movement itself (Bettinger & Bettoli 2002), to investigate the behavioural changes induced by a novel environment (Cowan 2001), and to compare costs (and/or capacities) of returning to the known area (Pärt 1995) between individuals of different dispersal status. It is worth noting, however, that manipulating dispersal rate or status of part of the population may induce direct consequences on the unmanipulated part of the population (e.g. when dispersing, an individual may affect the fitness of individuals in both its population of origin and of settlement; see also Boudjemadi et al. 1999).
Estimating individual fitness consequences of dispersal proves a difficult question to investigate as within-individual consistency and parent–offspring resemblance in dispersal seems to be a reality that has to be taken into account seriously when gathering fitness data. Future studies should always include an analysis on the risk of such biased fitness estimates before concluding on potential fitness consequences of dispersal, especially in long-lived species where the number of potential occasions for underestimating local adult survival and offspring recruitment is high, and even when the study area is large compared to observed dispersal distances. Actually, biases in fitness estimates may arise due to any behaviour affecting recapture probability differently between dispersing and philopatric individuals and their offspring. To avoid such biases, high detection/recapture effort over large spatial and temporal scales is necessary. Studies cannot hope to answer this question without devoting sufficient effort into the gathering of high quality fitness data: there is no shortcut into the issue of estimating fitness consequences of dispersal.
This study has been supported financially by the French National Centre of Scientific Research, CNRS PICS no. 3054 (to B. D. and T. P.), the French National Research Agency, contract ANR-06-JCJC-0082 (to B. D.) and the Swedish National Research Council (VR) (to T. P.). We thank P. Warin for invaluable help with the simulation and J. Clobert and four anonymous referees for comments on a previous version of the manuscript.