Professor K.J. Gaston, Biodiversity and Macroecology Group, Department of Animal and Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK, Tel.: +44(0)114 2220030; Fax: +44(0)114 2220002; e-mail: K.J.Gaston@sheffield.ac.uk
1A number of generalizations have been made as to the effects of the area of occupancy, population size, dispersal ability and body size of species on their relative rates of local colonization and extinction.
2Here, data on the breeding bird assemblage of Britain are used to test these generalizations. The complete geographical ranges of British birds have been censused twice, in the periods 1968–72 and 1988–91, allowing rates of colonization and extinction between these periods to be estimated.
3The local colonization dynamics of species are influenced independently by their range sizes and the dispersal abilities of adult birds: species with smaller range sizes and larger dispersal distances were more likely to have colonized new areas between the two census periods.
4The local extinction dynamics of species are influenced independently by their population sizes and body masses: species with smaller population sizes and body sizes were more likely to have gone extinct from areas inhabited in the first census period.
5These results remain when controlling for the effects of phylogenetic relatedness.
6These analyses uphold many commonly held generalizations about the correlates of local colonization and extinction, and suggest that the long-term evolutionary history of these bird species has influenced their potential to respond to current ecological conditions.
The spatial distributions of species are dynamic. Occupied sites become vacant as a consequence of local extinction, and vacant sites become occupied as a result of local colonization. The rates of colonization and extinction will determine the instantaneous overall area of occupancy of a species, and variation in these rates the temporal trends in this area. The spatial pattern of colonization and extinction events will determine how the distribution of a species moves through time. Both of these processes will occur across the distributions of most species, albeit that colonizations will predominate for species expanding their range, and extinctions for species whose range is contracting.
A number of general assertions are commonly made about patterns of local colonization and extinction across regions. In the case of extinction, for example, it is generally held that (i) narrowly distributed species and species with low local abundance tend to experience high levels of local extinction (Hanski 1982; Diamond 1984; Pimm 1991; Burgman, Ferson & Akçakaya 1993; Gaston 1994); (ii) species with high dispersal ability will tend to exhibit low levels of local extinction, because susceptible local populations will be rescued by immigration (a ‘rescue effect’; Brown & Kodric-Brown 1977; but see below); and (iii) large-bodied species will tend to have a disproportionate likelihood of local extinction (although the converse argument has also been made; for discussion see Pimm 1991; Gaston 1994; Brown 1995; Gaston & Blackburn 1995a, 1995b).
In this paper, we analyse interspecific relationships between local rates of colonization and extinction, and geographical range size (area of occupancy, sensuGaston 1991), population size, dispersal ability and body size, for breeding birds in Britain. This is perhaps the only assemblage for which appropriate data are available at a broad regional scale for a high proportion of the species, and for which information is also available with which to control for the possible effects of phylogenetic relatedness. It is also an assemblage and spatial scale at which strict metapopulation dynamics, as opposed to patchy, island-mainland and other forms of spatial population structure, are unlikely to be predominant (Gaston, Blackburn & Gregory 1997).
The terms ‘colonization’ and ‘extinction’ are frequently taken by ecologists to refer to the gain or loss, respectively, of spatially discrete populations. Here, we use them in a broader sense to refer to the gain or loss of a species from a defined point in space (in this case, a 10 × 10 km2 square of the British National Grid; see below). Our definition implies a view of the geographical range more as a continuous distribution than a set of discrete populations. However, the difference between these views is one of degree only: even discrete populations of a given species must usually be linked, while the loss or gain of defined areas to a geographical range must result from the loss or gain of individuals from these areas, and hence extinction or colonization in a broad sense.
Data for rates of colonization and extinction for breeding bird species in Britain were taken from Gibbons, Reid & Chapman (1993). The breeding distributions of all bird species in Britain have been censused twice at the national scale, in the periods 1968–72 (Sharrock 1976) and 1988–91 (Gibbons et al. 1993). The data obtained were used to map, for each census period separately, the presence or absence in the 10 × 10 km2 squares of the British National Grid of all bird species recorded breeding in Britain by each census (Sharrock 1976; Gibbons et al. 1993). Gibbons et al. (1993) present ‘Change’ maps that show those 10 × 10 km2 squares from which each species was recorded during the first but not the second census, and those squares from which each species was recorded during the second but not the first census. It is assumed that squares in the former set are examples of extinctions of a species from previously occupied areas, and that squares in the latter set are examples of colonizations of previously unoccupied areas (see also Thomas & Lennon 1999). Thus, summing the squares in each set gives measures of the rates of extinction and colonization, respectively, for each species between the two census periods. Although the two atlases were not generated using identical methodologies (see Gibbons et al. 1993 for discussion), the broad brush approach employed here should be robust to consequences of the differences (see also Thomas & Lennon 1999).
To obtain estimates of numbers of extinctions and colonizations for each species, we counted the number of each type of square on the ‘Change’ map for that species in Gibbons et al. (1993). Where the combined number of colonizations and extinctions was large (more than 100 or so), direct counts were performed only on the smaller set. The number of squares in the larger set was then calculated as (S + D) – F, where S is the species’ range size (number of occupied 10 × 10 km2 squares) in the second census, F is its range size in the first census, and D is the number of squares in the smaller set counted directly from the ‘Change’ map. If the breeding distribution of the species had decreased between the censuses, then D is the number of colonizations, while if the breeding distribution of the species had increased, D is the number of extinctions. Where the total number of colonizations and extinctions was small (less than 100 or so), we performed direct counts on both sets. However, Gibbons et al. (1993) deliberately omitted squares from the ‘Change’ maps of two species with small numbers of extinctions and colonizations, bittern Botaurus stellaris (L.) and red-necked phalarope Phalaropus fulicarius (L.). Thus, the totals for these species could only be calculated using the first method.
The degree of error in these counts was assessed in two ways. First, we performed replicate counts on the numbers of colonizations or extinctions for a sample of those species with large numbers (e.g. >100) of both. When replicate counts differed, they did so generally by about 1%. Where they did differ, the modal value was used. Second, for species where the total numbers of colonizations or extinctions were small (e.g. <100), total numbers of each were calculated using both the direct count and calculation methods described in the previous paragraph. Any differences in the results were then checked.
Numbers of colonizations and extinctions were calculated for 209 species of British breeding bird. This total represents all species mapped in Gibbons et al. (1993) except red-crested pochard Netta rufina (Pallas), scaup Aythya marila (L.), Temminck’s stint Calidris temminckii (Leisler), carrion/hooded crow Corvus corone corone/cornix (L.), brambling Fringilla montifringilla L., serin Serinus serinus (L.), common crossbill Loxia curvirostra L. and Scottish crossbill Loxia scotica Hartert. For all bar carrion/hooded crow and the crossbill species, this was because Gibbons et al. (1993) provide no ‘Change’ map. Number of colonizations and extinctions could not be calculated for the crossbill species because Gibbons et al. (1993) present a single ‘Change’ map for both species (they were not distinguished as separate species for the first census), while they could not readily be calculated for the carrion/hooded crow as Gibbons et al. (1993) present separate ‘Change’ maps for the two forms although they are considered to be conspecific.
The British breeding range sizes (number of 10 × 10 km2 squares of the British National Grid occupied) of each species in the period 1968–72 and 1988–91 were taken from Gibbons et al. (1993), and estimates of the current British breeding population sizes (number of individuals) of the species from Stone et al. (1997). Body masses (g) were taken from the sources identified in Gaston & Blackburn (2000, Appendix III, where all the range size, population size and body mass data are listed). Data on geometric mean natal dispersal distances (km) for 74 of the 209 species, and on geometric mean breeding dispersal distances (km) for 66 of the 209 species were taken from Paradis et al. (1998), where definitions of these measures are given and the methodology used to calculate them is described. The measures are based on data from the British Trust for Ornithology ringing scheme, and so are entirely independent of the atlas data.
In what follows, range size refers to that recorded in the first breeding bird atlas (Sharrock 1976), unless otherwise stated. Although there are 209 species in the data set, n = 201 for some analyses because seven species had range size = 0 in the first atlas, while an additional species showed no extinctions. Range size was logit(n + 1) transformed for analysis (Williamson & Gaston 1999), with the addition of 1 to account for zero values. Population size, body mass and measures of dispersal distance were log-transformed in all analyses.
An important issue here concerns the null hypothesis against which relationships should be judged. If the colonization or extinction of a species from a 10 × 10 km2 square is a random event, we expect species with larger ranges to exhibit greater absolute numbers of these events by chance alone. Therefore, the null form of the relationship between range size and these events is a positive linear relationship on untransformed axes (or r-shaped if number of extinctions is log-transformed), rather than zero. However, random colonizations and extinctions will give no relationship between range size and the percentage of this range represented by colonizations or extinctions between the two census periods, and so a null relationship of zero slope. [We verified that this is indeed the case using a randomization simulation in which the total number of colonization and extinction events observed for all species were assigned at random across species, with the probability of being assigned an event proportional to the range size of the species in the first atlas, and the number of extinctions assigned to a species constrained to be no greater than this range size.] Therefore, analyses of extinctions use the proportion of the first atlas range that went extinct between the two atlas censuses, while analyses of colonizations use the proportion of squares in the second atlas range that represent colonization events. Both proportions were arcsine square root-transformed, unless otherwise stated.
The phylogenetic relatedness of species means that they do not comprise independent data points for interspecific comparative analyses (reviewed by Harvey & Pagel 1991). Hence, where appropriate analyses were performed across species and within taxa using a method designed to control for phylogenetic association. One way to control for the effects of phylogenetic relatedness is to examine relations within each pair of taxa below a node in a bifurcating phylogeny. The relation between the variables is then unaffected by phylogeny, since the taxa in each comparison are equally related to each other. This method requires that the true phylogeny be known (Felsenstein 1985). Here, we use a model (Comparative Analysis by Independent Contrasts (CAIC, version 2·6·7; Purvis & Rambaut 1995)) which applies Felsenstein’s approach to data sets for which only approximate phylogenies are available. This method calculates a single value (‘contrast’) for each variable within each taxon (i.e. below each node in the incompletely resolved phylogeny) representing its magnitude and direction of change. Each contrast is then scaled using information on the length of the branches leading from that node (or an assumption about branch lengths is made if no such information is available; Pagel & Harvey 1989; Harvey & Pagel 1991). The independent contrasts calculated for two variables will show similar changes within each taxon if the variables are correlated. The set of within-taxon contrasts can be analysed using standard regression techniques, although regressions must pass through the origin (Garland, Harvey & Ives 1992).
Bird species were classified using the phylogeny of Sibley & Ahlquist (1990), with classification below the level of tribes based on Sibley & Monroe (1990, 1993), and the assumption that all branches in the phylogeny are of equal length. CAIC version 2·6·7 calculates diagnostic statistics to assess whether either the evolutionary assumption that evolution proceeds according to a random walk process, or the statistical assumption of homogeneity of variance (Purvis & Rambaut 1995), is violated in each univariate analysis. Violations of the statistical assumptions and evolutionary assumptions occurred in all analyses, and could not be removed by data transformations. Therefore, we ensured that within-taxon analyses were conservative by using the BRUNCH option in CAIC, and sign tests of evolutionary hypotheses of association (Purvis & Rambaut 1995).
Between the dates of the censuses for the two distribution atlases (Sharrock 1976; Gibbons et al. 1993), all British bird species changed their geographical range to some degree (Fig. 1a). Most range changes were small (fewer than 50 10 × 10 km2 squares gained or lost over a 20-year period), but a few were dramatic (more than 500 squares). However, these changes in range size alone obscure the fact that most contracting species also colonized some areas, while most expanding species also went extinct from some grid squares. The frequency distribution across species of the total number of grid squares that represent colonizations or extinctions reveals that the changes in grid square occupancy are frequently much greater than indicated by the change in range size alone (Fig. 1b): 36 species changed their occupancy of more than 500 squares (more than one-sixth of all British National Grid available squares), compared to 7 that changed their range size by that amount (Fig. 1a). Moreover, the number of grid squares that represent colonizations or extinctions will be an underestimate of the total number that have occurred over the 20-year period, as some cells whose occupancy appears unchanged will have experienced both colonization and extinction.
Most species show few colonizations or extinctions relative to their range size, although the proportion of colonizations has a second mode in the highest proportion class (Fig. 2). These frequency distributions of colonizations and extinctions are related to range size, with most species with high proportions having small range sizes. There are significant linear negative relationships between range size and the proportion of that range that went extinct between the two atlas censuses (r2 = 0·47, n = 201, P < 0·0001), and between range size and the proportion of squares in the second atlas range that represent colonization events (r2 = 0·78, n = 209, P < 0·0001), albeit that a squared term explains a significant additional 1% of the variance in the latter case. Thus, colonizations and extinctions contribute the highest proportions to the smallest ranges. Similar relationships pertain in all cases if log population size is substituted for range size (proportion colonizations: r2 = 0·74, n = 209, P < 0·0001; proportion extinctions: r2 = 0·67, n = 201, P < 0·0001). Colonizations and extinctions constitute the highest proportion of the ranges of species with the smallest population sizes. It follows from these results that species that colonized many squares between the periods 1968–72 and 1988–91 should also have gone extinct from many, and vice versa. Proportions of colonizations and extinctions are indeed positively correlated across species (r = 0·68, n = 201, P < 0·0001).
Although the relationships between range size and proportions of colonizations or extinctions differ from null expectations, these differences could arise because there is a positive correlation between range size and population size in these data (Blackburn et al. 1997), and because species with larger population sizes have proportionally fewer extinction and colonization events than expected (see above). Therefore, we performed multiple regressions to test for independent effects of range size and population size on colonization and extinction rates. Both logit range size and log population size explain independent interspecific variation in proportion of colonizations (overall r2 = 0·82, n = 209, P < 0·0001; Table 1). For a given range size, species with small population size colonized proportionately more new squares, and for a given population size, species with small range size colonized proportionately more new squares. However, only log population size explains variation in proportion of extinctions (overall r2 = 0·67, n = 201, P < 0·0001; Table 2). Thus, for a given range size, species with small population size went extinct from proportionately more squares, but for a given population size, proportion of extinctions was unaffected by range size.
Table 1. Multiple regression model of the effects of log population size and logit range size on arcsine square root-transformed proportion of squares in the second atlas range that represent colonization events for British bird species. Coefficient = regression coefficient, SE = standard error of the coefficient estimate. t = t-test of the null hypothesis that the coefficient does not differ from zero, and P = two-tailed probability that the null hypothesis is correct
Log population size
Logit range size
Table 2. Multiple regression model of the effects of log population size and logit range size on arcsine square root-transformed proportion of the range of British bird species that went extinct between the two atlas censuses. Coefficient = regression coefficient, SE = standard error of the coefficient estimate. t = t-test of the null hypothesis that the coefficient does not differ from zero, and P = two-tailed probability that the null hypothesis is correct
Log population size
Logit range size
Variations in proportion of colonizations and extinctions among British bird species are both positively associated with variation in both breeding and natal dispersal distances (Table 3): species that colonized or went extinct from proportionately more squares between the periods 1968–72 and 1988–91 tended to disperse further as both adults and juveniles. Both measures of dispersal are also negatively correlated with species range size and abundance (Table 3: see also Paradis et al. 1998). These relationships seem to explain the associations between extinction and dispersal, since neither measure of dispersal is a significant predictor of extinction rate in a multiple regression model with range size and population size, although both other independent variables contribute significantly to the overall proportion of variance explained in both cases. The situation for colonization rate is more complicated, as range size and breeding dispersal distance explain significant variation while population size does not when all three are entered into a multiple regression, while only range size explains significant variation in colonization rate in a multiple regression with population size and natal dispersal distance.
Table 3. Correlation coefficients (r) for the relationships between the measures of dispersal listed in the first column and arcsine square root-transformed proportion of colonizations (Colonizations), arcsine square root-transformed proportion of extinctions (Extinctions), logit range size, and log population size (Pop. size). n= sample size (number of species)
There is no relationship between proportion of extinctions and log body mass across British bird species (r2 = 0·01, n = 209, P = 0·091). A weak positive association is observed between proportion of colonizations and log body mass (r2 = 0·071, n = 209, P < 0·001). There is no significant curvilinear component to either relationship.
To derive best fit models for proportion of colonizations and extinctions in terms of the other variables, a series of stepwise regressions was performed using both forward addition and backward elimination approaches. The variables in the final models were then used to calculate a multiple regression, and those excluded added sequentially to this model to check that they did not explain significant additional variance in the dependent variable. The resulting models are given in Table 4 and Table 5. Across species, 80·3% of the variance in proportion of colonizations is explained by a model with range size and geometric mean breeding dispersal distance. The regression coefficient is positive for dispersal distance, but negative for range size. Of the variance in proportion of extinctions, 68·4% is explained by population size and body mass (Table 5). Proportion of extinctions increases as body mass and population size decrease.
Table 4. Multiple regression model of arcsine square root-transformed proportion of colonizations in terms of the factors in the first column. Coefficient = regression coefficient, SE = standard error of the coefficient estimate. t = t-test of the null hypothesis that the coefficient does not differ from zero, and P = two-tailed probability that the null hypothesis is correct. Overall r2 = 0·803, n = 66, P < 0·0001
Logit range size
Log geometric mean breeding dispersal distance
Table 5. Multiple regression model of arcsine square root-transformed proportion of extinctions in terms of the factors in the first column. Coefficient = regression coefficient, SE = standard error of the coefficient estimate. t = t-test of the null hypothesis that the coefficient does not differ from zero, and P = two-tailed probability that the null hypothesis is correct. Overall r2 = 0·684, n = 201, P < 0·0001
Log population size
Log body mass
Using CAIC to control for the effects of phylogenetic relatedness shows that both proportion of colonizations and proportion of extinctions are negatively related to both range size (colonizations: 65/74 contrasts negative, P < 0·001; extinctions: 65/72 contrasts negative, P < 0·001) and population size (colonizations: 61/74 contrasts negative, P < 0·001; extinctions: 63/72 contrasts negative, P < 0·001). Proportion of extinctions and colonizations are positively correlated (59/72 contrasts positive, P < 0·001). Natal dispersal distance is positively correlated with proportion of colonizations (22/31 contrasts positive, P = 0·029), while breeding dispersal distance is positively correlated with proportion of extinctions (21/29 contrasts positive, P = 0·024). Breeding dispersal distance is also negatively correlated with population size (20/28 contrasts negative, P = 0·036). These results concur with the interspecific analyses.
Within-taxon and interspecific analyses differ for body mass and the remaining correlations for dispersal distance. Body mass is negatively related to proportion of extinctions within taxa (48/72 contrasts negative, P = 0·0063), but shows no relationship to proportion of colonizations (41/74 negative, P = 0·42). These relationships were not significant and positive, respectively, across species. Aside from the two results in the previous paragraph, no other correlations of dispersal distance to proportion of colonizations, proportion of extinctions, range size or population size were significant (cf. Table 3).
Multiple regressions concur with the interspecific analyses in finding no significant effect of range size on proportion of extinctions independent of population size (P = 0·61), but a significant negative relationship of population size independent of range size (P < 0·001). Both population size and range size explain variation in proportion of colonizations in a multiple regression (overall r2 = 0·79, n = 74, P < 0·001; logit range size, P < 0·001, log population size, P < 0·001). Within-taxon analyses also confirm the significant independent effects of range size (negative) and geometric mean breeding dispersal distance (positive) on proportion of colonizations shown in Table 4 (overall r2 = 0·80, n = 29, P < 0·001; logit range size, P < 0·001, log geometric mean breeding dispersal distance, P < 0·001), and of body mass and population size (both negative) on proportion of extinctions shown in Table 5 (overall r2 = 0·69, n = 72, P < 0·001; log body mass, P = 0·038, log population size, P < 0·001). However, the results of these multiple regressions should be interpreted with caution given the problems with violations of statistical and evolutionary assumptions in these within-taxon analyses (see Methods).
The absolute number of sites or areas from which a restricted breeding bird species in Britain can become extinct is by definition limited, as is its potential to colonize unoccupied areas (unless colonizers arrive from outside Britain, which may occur at least at a low rate in many cases), but these few colonizations and extinctions form a relatively high proportion of the species’ overall range size. Widespread species also exhibit low levels of extinction and colonization between 1968–72 and 1988–91, but these few events consequently affect a small proportion of the overall range. Thus, the range size of a breeding bird species in Britain is strongly negatively related to the proportion of occupied areas that will subsequently undergo extinction, and of fresh areas that will be colonized. Although the opportunities for colonization and extinction are restricted for species with restricted geographical range in Britain, our results are not simply a consequence of the influence of such species on the overall patterns. Repeating the analyses presented here excluding bird species of restricted British range (defined as species in the lowest range size quartile; Gaston 1994) left most results unchanged (but see below).
The relationship of range size to proportion of colonizations seems likely to result simply because there are few unoccupied areas for widespread species. The relationship to extinctions could potentially arise because widespread species tend to have higher local densities than narrowly distributed ones (a difference manifested also in their total population sizes), thus rendering them less vulnerable to local extinctions. A positive interspecific density–occupancy relationship is a well-documented pattern for birds in Britain (reviewed by Gaston & Blackburn 2000). A role for abundance in determining levels of colonization and extinction, over and above that of range size, is confirmed by the finding that population size explains significant amounts of variance in colonization and extinction in addition to those explained by range size (Tables 1 and 2). Moreover, the effect of range size on extinctions is not significant independent of population size, suggesting that small population size is the key driver of extinction rates in British birds, and that the range size effect is a consequence of the colinearity of these two variables (although range size was significantly negatively related to extinction rate independently of population size when restricted range species were excluded).
Somewhat counter-intuitively, population size is negatively related to proportion of colonizations, even when controlling for range size: species with smaller population sizes have proportionately high rates of colonizations. However, the effect of population size on colonization rate seems to be a consequence of its negative correlation with dispersal distance (Table 3), as the population size effect disappears once dispersal distance is accounted for (Table 4). As expected, the proportion of colonizations exhibited by a species is positively correlated with its dispersal distance (Table 3): better dispersers appear to be better colonists. The relationship for breeding dispersal distance is lost when controlling for phylogeny, but the test used is conservative, and the correlation is close to significance (P = 0·061). The positive effect of dispersal ability is independent of other variables influencing the level of colonization, both across species (Table 4) and within taxa. The geometric mean dispersal distances of species in the analyses are 2·83 km for natal dispersal, and 0·84 km for breeding dispersal. The strong right skew in intraspecific dispersal distances of British birds (Paradis et al. 1998) indicates that individuals regularly move much greater distances. However, the moderate correlation coefficients between measures of dispersal and proportions of colonizations suggest that the mean distances capture important differences in the abilities of species to discover unoccupied but occupiable sites and, if necessary, to arrive in sufficient numbers to colonize them (bearing in mind that the ‘Change’ maps in Gibbons et al. (1993) include grid squares in which the species was present in 1988–91 but with no necessary evidence of breeding).
A high dispersal distance might also have been expected to reduce the level of local extinction exhibited by a species. However, the relationships between dispersal distance and number of extinctions (Table 3) are all positive, while dispersal distance does not explain any significant residual variation in number of extinctions when other variables are controlled for, either across species (Table 5) or within taxa. This suggests that the positive relationship between dispersal distance and extinction rate may be a simple consequence of the correlation of both variables with population size.
Alternatively, dispersal distance might increase with both colonization and extinction rates if there were some directional shift in the positions of the ranges of species, such that some areas were becoming available for fresh colonization, perhaps as a product of climate change, and others were being vacated and were not available for recolonization. There is certainly evidence for a systematic directional (northward) expansion in the ranges of some birds in Britain (Thomas & Lennon 1999). The distances concerned (after accounting for changes in the numbers of grid squares occupied, the northern margins of southerly species were estimated to have moved north by an average of 18·9 km – the width of less than two 10 × 10 km2 squares – in 20 years) seem insufficient to explain the results documented above, when compared with the distances over which the species commonly disperse, but the likely effect is difficult to judge.
Even accounting for net changes in range size, species whose ranges are expanding along one margin need not necessarily be contracting along another, and in the case of birds breeding in Britain in general they probably are not (see Gibbons et al. 1993). Many of the local extinctions that are occurring result from the direct loss of habitat, or reduction in its quality (which may amount to the same thing; many species of lowland farmland birds, for example, have declined in numbers, range size or both as a product of agricultural intensification; Gates et al. 1994; Fuller et al. 1995; Gillings & Fuller 1998; Siriwardena et al. 1998). High dispersal distances are unlikely to decrease the numbers of such extinctions because the probability of extinction of species from local sites cannot be systematically reduced by the recruitment of fresh individuals to the local population. Thus, there may be no reason to expect a negative relationship between dispersal distance and extinction in this system.
The body mass of breeding bird species in Britain is positively correlated with the proportion of colonizations exhibited by species between the periods 1968–72 and 1988–91. Thus, there is some support for the contention that body mass provides an indicator of likelihood of colonization, albeit that the relationship is weak. If body mass is related to dispersal ability in a simple fashion, it might be anticipated that it would be related also to the likelihood that species will colonize new sites or regions. For birds in Britain, Paradis et al. (1998) found that both breeding and natal dispersal distances were positively correlated with body mass across species. The relationship between colonizations and mass may be a consequence of this, and indeed mass fails to enter the stepwise multiple regression models of colonization rate, while dispersal distance does (Table 4). There is also no effect of body mass on colonization rate within taxa.
Greater body size has frequently been argued to increase the likelihood of local extinction, but there is no relationship between body mass and proportion of local extinctions experienced by bird species in Britain between 1968–72 and 1988–91. However, there is a significant negative relationship with body mass within taxa, and mass does enter significantly into the interspecific (Table 5) and within-taxon multiple regression models for number of extinctions. Thus, there is evidence that large body mass is actually advantageous in preventing extinctions in this system. The impression that large body mass should increase extinction probability may derive from the tendency for body mass and population size to be negatively correlated, as they are in British birds (Nee et al. 1991; Gregory & Blackburn 1995). In that regard, it is interesting that the body size effect here is only apparent once population size is controlled for. Nevertheless, large body mass has a number of potential advantages in ameliorating extinction rates (e.g. Gaston & Blackburn 1995a), and the precise cause of the effect here will require further study to clarify.
The results presented here identify proximate answers to the questions of what determines current variation in the rates of colonizations and extinctions in British birds: colonizations increase as the dispersal ability of the species increases and its range size decreases, while extinctions are higher for species of smaller body mass and smaller population size. These answers raise two further questions: what are the ultimate mechanisms that generate these proximate associations, and to what aspects of the environment are British bird species responding via the processes of colonization and extinction? A variety of answers have been proposed to the first question (e.g. for the relationship between body mass and extinction likelihood, see Pimm 1991; Brown 1995; Gaston & Blackburn 1995a; Owens & Bennett 2000; Bennett, Owens & Baillie 2001), but the ultimate mechanisms that generate these proximate associations remain to be elucidated.
Answers to the second question seem likely to involve the influence of anthropogenic changes to the environment. For example, increasing habitat fragmentation may explain why species with better dispersal ability have colonized more locations, while the lower extinction rate for large-bodied species may be a consequence of the greater ease with which conservation measures can be applied to populations of such birds. Firm links between range dynamics and environmental features will be difficult to establish though, because the environment is currently undergoing a variety of changes from a range of drivers (Moore, Chaloner & Stott 1996), because different species can show a variety of responses to environmental changes (Spicer & Gaston 1999), and because extinction and colonization can potentially occur even in constant environments through stochastic processes of immigration and emigration (Hanski 1999). Nevertheless, whatever is driving the relationships with colonization and extinction rates produces patterns that are consistent across species and within taxa, and hence that generates similar responses in different avian taxa. This suggests that the long-term evolutionary history of these species has influenced their potential to respond to current ecological conditions.
In sum, the local colonization dynamics of breeding birds in Britain are influenced independently by range sizes and dispersal abilities. Local extinction dynamics are influenced independently by population sizes and differences in body mass. These outcomes are the result of the first comprehensive set of such analyses for a single animal assemblage, and are robust to the phylogenetic relatedness of the species concerned. Thus, most of the commonly held generalizations about the correlates of colonization and extinction are sustained.
K.J.G. is a Royal Society University Research Fellow. T.M.B. was partly funded by a Leverhulme Special Research Fellowship. We thank S. Gaston and two anonymous referees for helpful comments on the manuscript.