Effects of experimental habitat fragmentation and connectivity on root vole demography


Dr R.A. Ims, Division of Zoology, Department of Biology, University of Oslo, PO Box 1050 Blindern, N-0316 Oslo, Norway. Tel. + 47 2285 7034, Fax: + 47 2285 4605. E-mail: r.a.ims@bio.uio.no


1. We used a factorial experimental design to test whether habitat fragmentation (two different fragment sizes) and connectivity (presence and absence of corridors between small fragments) affected population growth rate and two conventional measures of the demographic structure (proportion of reproductive adults and sex ratio) in 12 enclosed populations of root voles.

2. Because the matriline has been suggested to be a functional entity affecting the demography of Microtus populations, we employed a study protocol (including laboratory-raised founder animals and intensive live trapping combined with fluorescent powder techniques), which enabled us to track the matrilineal genealogy in the populations. Thus, the effect of habitat structure on the matrilineal structure of the populations, quantified by the Shannon–Wiener diversity index, could be tested for the first time.

3. Population growth rate was density dependent, but neither this population parameter nor sex ratio and proportion of reproductive adults differed between the habitat treatments. This was unexpected since the presumed determinants of Microtus demography, such as individual space use, dispersal distances and spatio-social organization had earlier been found to differ between the treatments.

4. The matrilineal structure (diversity) of the populations changed in response to the experimental habitat manipulations. In corridor-connected systems, some matrilines became numerically dominant, which lead to reduced matrilineal diversity compared to systems with isolated habitat fragments. We hypothesize that some matrilines were able to colonize and exploit corridor-connected fragments better than other matrilines. Matrilineal diversity was not related to any other demographic attribute (growth rate, sex ratio and functional stage structure) at the population level.

5. Our results suggest that fragmentation-induced changes of individual and matrilineal level attributes do not necessarily propagate into population level differences in vital rates. The notion that matrilines may be a functional entity in Microtus populations could not be supported.

6. Our experimental results suggest that population genetics, more than population demography, may be sensitive to habitat fragmentation and connectivity, within the specific temporal and spatial extent explored. In particular, the effect of corridors on matrilineal diversity imply that population genetic studies based on mtDNA markers should consider habitat connectivity when analysing genetic diversity.


Fragmentation of natural habitats often takes place at a scale where individual space use (e.g. Rolstad & Wegge 1987; Haila 1990; Rolstad 1991; Wauters, Casale & Dhondt 1994; Redpath 1995; Shepard & Swihart 1995) and the social structure of local demes (Lovejoy et al. 1986; Carey, Horton & Biswell 1992; Ostfeld 1992) change. Since the ultimate question concerning the effects of habitat fragmentation relate to the genetic and demographic performance of fragmented populations (Rolstad 1991; Fahrig & Merriam 1994; Dunning et al. 1995; Wiens 1995), it is of interest to know whether fragmentation-induced changes in space use and the social relationship between individuals translate into population level effects (Lima & Zollner 1996; Goss-Custard & Sutherland 1996). Scenarios for predicting population level responses to habitat fragmentation based on responses in individual space use have been suggested by several authors (Rolstad & Wegge 1987; Ostfeld 1992; Ims, Rolstad & Wegge 1993; Wiens et al. 1993; Sutherland & Dolman 1994). However, experimental tests employing replicates at the population level have been few.

Here, we analyse the demography of experimental populations of the root vole Microtus oeconomus (Pallas 1776), which had been manipulated with respect to the degree of habitat fragmentation (fragment size) and connectivity (presence/absence of corridors between fragments). The manipulations were carried out at a spatial scale expected to induce responses in individual space use and spatio-social organization according to the predictions of Ims et al. (1993). Indeed, our previous studies on the space use of reproductive individuals based on radio-telemetry (Andreassen, Hertzberg & Ims 1998), and an analysis of dispersal and aggregation pattern of all individuals in the populations based on capture-recapture (Bjørnstad, Andreassen & Ims 1998) supported the predicted responses. Long dispersal distances was observed in the most fragmented systems with and without corridors (a fission response; sensuIms et al. 1993), while individuals frequently included more than one habitat fragment in their home range when small fragments were connected by corridors (an expansion response). The most pronounced and consistent effect regarding spatio-social organization was the tendency in matrilineal-related individuals to aggregate on small fragments (a selective fusion response; sensuIms et al. 1993).

In particular, matrilineal aggregation in space has a particular bearing on our present analysis as it has been suggested that the matrilineal structure of Microtus populations may be a main determinant of population demography (Lambin & Krebs 1991). Specifically, it has been suggested that tight matrilineal structuring of the populations would enhance population growth rate due to co-operative behaviour between matriline members (Lambin 1994a,b; Lambin & Yoccoz 1998). If this also applies to matrilineal aggregation induced by habitat fragmentation as demonstrated by Bjørnstad et al. (1998) and Andreassen et al. (1998) for our experimental populations of the root vole, we would expect the most fragmented populations to show enhanced population growth. To explore the more general idea that matrilines are functional entities in Microtus populations (Boonstra & Rodd 1983; Boonstra et al. 1987; Lambin, Krebs & Scott 1992), that may or may not interact depending on the environmental setting (Lambin & Krebs 1991), we test whether the matrilineal composition of the populations change in response to habitat fragmentation and connectivity.

Materials and methods

Experimental area

The experiments took place on six rectangular 0·5-ha plots, situated on flat farmland at Evenstad Research Station in Hedmark county, Norway in 1990 and 1991. Each plot was fenced with a galvanized steel sheet fence extending 0·6 m above and 0·4 m below ground. The whole area (incorporating the six plots) was surrounded by a 1·5-m high chicken wire fence with electric wire running along the top, to keep mammalian predators out.

Root voles require meadow or mire vegetation which provide sufficient protective cover and food (Tast 1966). For a pilot study in 1989 (Ims et al. 1993), the plots were seeded with a mixture of 65% Phleum pratense Linné, 15% Festuca pratensis Hudson, 10% Agrostis tenuis Linné and 10% Trifolium pratense Linné, which developed into a dense, homogeneous meadow vegetation. The habitat structure was manipulated by mowing the vegetation outside designated habitat patches and corridors every week. Experiences from the 1989 pilot season showed that the animals avoided the open, mowed areas (i.e. the matrix) using them only for crossing and never spending time there (Ims et al. 1993; see also Andreassen et al. 1998). Herbicide was applied to a 2·5-m wide zone along the steel sheet fences to prevent any growth of vegetation.

Three habitat fragment configurations were employed (Fig. 1). Two configurations with six small fragments (15 × 15 m = 225 m2) were created to test the effect of connectivity; one with isolated fragments and the other with corridors (0·5-m wide strips of vegetation) connecting 3 and 3 fragments (Fig. 1). To test the effect of the degree of fragmentation the plots with small isolated fragments were contrasted with the third and last configuration which consisted of two large fragments (26 × 26 m = 675 m2). The total area of habitat was kept constant at 1350 m2 and the distance to the closest neighbouring fragment was 15 m in all plots. The nearest distance to the fence was 2·5 m. The extent of the area available to voles in each plot is comparable to what is commonly available to natural local populations of root voles, normally inhabiting small sedge or moist meadow patches along water-ways in boreal forest and tundra areas (Tast 1966, 1982; Lambin et al. 1992; Viitala 1994). The size of the small fragments was comparable to home range sizes of adult female root voles, while the large fragments were several times larger. Experimental habitat manipulation was identical in 1990 and 1991 with two spatial replicates of each of the three habitat configurations per year. However, the habitat configurations (as well as the factor strain; see below) were changed between the enclosures to ensure maximum spatial interspersion of treatments (Hurlbert 1984), i.e. no plot received the same habitat configuration and strain more than once (Fig. 1).

Figure 1.

Spatial layout of the different treatments (habitat configurations and strains) at Evenstad research station during the 2 years of the study. S: Southern strain. N: Northern strain.

Experimental animals

To test for possible interaction between habitat structure and innate characteristics of the individuals, two different strains of root voles were used in the experiments. The animals were from laboratory colonies kept at the Animal Department at the University of Oslo and had been exposed to the same conditions prior to the start of the experiment. One of the colonies was founded by wild-caught animals from Valdres in Oppland county (60°45′N, 9°30′E), south Norway (hereafter termed ‘Southern’), the other from Pasvik in Finnmark county (69°15′N, 29°25′E), north Norway (hereafter termed ‘Northern’; see Ims 1997).

Both colonies have been studied closely in the laboratory for several generations, and show genetically-based differences in body growth rate and behaviour. The Northern voles are larger, and have significantly higher body growth rate than Southern voles (Ims 1997). There are also indications that Northern animals are more aggressive, have different litter sex ratios (Ims 1994; Aars, Andreassen & Ims 1995) and show less paternal care towards infants in the laboratory than Southern animals (Ims 1997). Both strains are equally sexually dimorphic with respect to body size, males being ≈ 50% larger than females (Bondrup-Nielsen & Ims 1990).

Colonization of experimental plots

Laboratory animals were introduced to the plots on 1 July 1990 and 4 July 1991 (week 27). In 1990, three unrelated, adult females with litters colonized each plot, whereas in 1991 four females with litters colonized the plots, thus giving a somewhat higher initial (founder) density the latter year (Table 1). Both years the litters were allocated so as to standardize the initial sex and age structure, and population density among the plots (Table 1). In the analysis, all descendants from the three or four founder females per plot were treated as separate matrilines. All animals were individually marked and transported to the experimental plots in wire mesh cages. Cages, each containing one mother and her litter, were brought from the laboratory and placed at fixed locations in each plot 1–2 days before release. The relative locations (grid line intersections) were the same in each plot, and were chosen to maximize the initial distance between the matrilines within each plot. All litters were 16-day-old (weaning age) at time of release.

Table 1.  The number of founder animals in the 12 experimental populations. Litter size are averages for three (1990) and four (1991) founder matrilines per population. Females and males include all litter members and mothers
YearStrainConfigurationLitter size
 ± SE
1990SouthernLarge4·0 ± 1·2105
  Corridor3·7 ± 2·2104
  Small3·7 ± 0·995
 NorthernLarge6·3 ± 0·9148
  Corridor5·3 ± 0·3118
  Small5·0 ± 0·0117
1991SouthernLarge3·5 ± 0·397
  Corridor3·5 ± 0·3107
  Small3·3 ± 0·388
 NorthernLarge4·5 ± 0·3138
  Corridor5·0 ± 0·91211
  Small4·5 ± 0·7147

The timing of the release corresponds to the weaning of the first year-born cohort in sub-arctic root vole populations (Tast 1966). Reproductive females colonize habitat patches at the onset of the breeding season (Tast 1966) and local demes typically consist of matrilineally-related individuals (Lambin et al. 1992).

Three populations per strain were established per year (one in each habitat configuration). Thus, the experimental design comprised in total 12 population replicates crossed on the experimental factors strain and habitat configuration as well as the cofactor year in a 3 (configuration) × 2 (strain) × 2 (year) factorial design.

Population census

The populations were censused in the habitat area every week (16 weeks each year) after colonization. We used a grid of ‘Ugglan special’ live traps (Granab, Sweden) and pitfalls at 5- and 10-m intervals in the fragments, and in the corridors, so that every second trap station had both an Ugglan trap and a pitfall. Small fragments thus had eight Ugglan traps and four pitfalls, large fragments had 19 Ugglan and 10 pitfalls, while three Ugglan and one pitfall were placed in the corridors. The combination of the two trap types was chosen to ensure high trappability of all age categories (Beacham & Krebs 1980; Boonstra & Rodd 1984).

In 1990, regular live trapping was carried out once a week from Monday to Wednesday. Traps baited with whole-grained oats and a piece of apple were activated on Monday morning, and checked every evening and morning until Wednesday morning. In 1991 trapping was carried out during two separate days (Monday and Thursday) each week. Traps were then activated at 04·00 h and checked every 4 h until 20·00 h. The change in trapping schedule from 1990 to 1991 was done to shorten periods of trap confinements. Both trapping schedules provided high minimum trappability estimates (Krebs & Boonstra 1984); the 1990 estimates (91·4 ± 4·2%) being somewhat higher than the 1991 estimates (80·0 ± 3·0%). In both years, the population census terminated in week 42. We then trapped out all individuals from the plots. During this removal trapping snap traps (in addition to the live traps) were placed at all trapping stations and in selected runways between the trapping stations to ensure that all animals were removed. Removal trapping continued until all traps had been empty for at least three trap checks. In general, live trapping based estimates of population parameters from week 40–41 was very similar to estimates based on the removal trapping in week 42 (Fig. 2), confirming that live-trapping based enumeration yielded reliable estimates. In total, this study was based on 11 592 captures of 1179 individuals.

Figure 2.

Population size trajectories for the 12 population replicates included in the study.

In addition to the traps inside fragments and the corridors (Ugglan live-traps and pitfalls), Ugglan traps were placed close to the steel sheet fence in all corners and at the mid-point of the long sides of each plot (six in all). These were checked every day and edge trappings were used to identify dispersers that were subsequently removed from the plots to avoid fence effects (Boonstra & Krebs 1977). Animals caught in the edge traps at least five times, including at least two different sites during a period of less than 9 days were classified as dispersers and removed from the populations. The number of dispersers removed per population were generally low (4·9 ± 2·8 individuals) and did not make up more than 8·6% of the total number of individuals marked in any population.

Young animals were marked at the time of first capture. The following data were collected for each animal when trapped: identity, weight, sexual status and position of trap. For females, sexual status was determined by recording pregnancy (abdominal swelling), and whether they were lactating or not. Males were considered sexually mature if the position of the testes was scrotal.

We were able to assign all litters born to individual mothers and thus to one of the three or four matrilines in the populations, by a combination of different techniques; nests finding by radio-tracking lactating mothers, the staining of lactating mothers with fluorescent powder and trapping records (for details, see Aars, Andreassen & Ims 1994, 1995). The dates of parturition of individual mothers were estimated from sudden weight changes of pregnant females, onset of lactation and the size of young in nest or upon capture.


We analysed weekly (population specific) measures of population size and three measures of the demographic structure employing mixed linear models in SAS (PROC Mixed; Littell et al. 1996). The experimental factors (i.e. habitat configuration and strain) and year was specified as fixed effects, while population identity (n = 12) was included as random effect in these models. The three measures of demographic structure was (1) proportion reproductive adults (defined as scrotal males and primiparous females) in the population; (2) operational sex ratio (the proportion of reproductive females of all adults in the populations); and (3) matrilineal population structure (the proportionate abundance of the different matrilines in the populations). We used the Shannon–Wiener diversity index (Krebs 1989) to quantify matrilineal population structure (hereafter termed matrilineal diversity). The matrilineal diversity index (MD) is thus given by


where pi,j is the proportion of individuals belonging to the ith matriline in the jth week of the total number of individuals in the population in that week (j = 33–42, i = 1–3 in 1990 and 1–4 in 1991). Due to the different number of matrilines per population in the 2 years, the maximum value of MD (given the same number of animals in each matriline) was scaled to unity. For the linear model used to test for the effect of experimental factors, the MDj values were logit-transformed and weighted by the square root of the size of the population to approximate the decreasing sampling variance with increasing population size.

Hypotheses predicting demographic effects of habitat fragmentation often assumes that density-dependent mechanisms are involved (e.g. Ims et al. 1993). Hence, we tested for density dependence in population growth rate employing the statistical model


where Nt,i is the terminal size (week 42) of a given population i and Nt – d,i the size d weeks earlier. The exponent of this model includes the strength of the density dependence (β) and the effects (γ) of the j experimental factors Z and the constant α. We used a logarithmic link, i.e. log [E (Nt)], and log Nt – d was used as an offset term as advocated by Lebreton (1991). The model had to be fit to the data with a quasi-Poisson error due to a slight over-dispersion from a pure Poisson error. Note, that this error term measures the random variation between the 12 population replicates. We tried d in the range of 3–7 weeks.


Population size

Population trajectories for the 12 populations based on the number of animals known to be alive per week are presented in Fig. 2. The populations failed to increase and even declined slightly during the first 6 weeks (until week 32). This time lag was due to the time required for young animals to mature sexually and recruit new litters to the populations. This demographic ‘latency period’ (from start of the experiment and until week 32), which also was qualitatively very similar among the different populations (Fig. 2), will not be included in the analysis. As litters started to be recruited to the populations (after week 33), the populations increased for a rather short time period and, thereafter, reached a plateau or decreased somewhat towards the end of the experimental periods (Fig. 2).

The number of animals per plot from week 33 to week 42 was analysed with respect to the experimental factors (habitat structure, strain and year) using a mixed linear model (PROC Mixed; Littell et al. 1996). The log-transformed population trajectories were modelled as a first-order auto-regressive process. Tests for the effect of the experimental factors were based on the random variation between these trajectories using population as a random effect. There was no effect of habitat configuration, neither alone or in interaction with other factors (all P > 0·10). Much of the heterogeneity in the size of the populations was accounted for by year and strain (Table 2). The Northern populations grew to larger sizes in 1991 than in 1990, whereas the Southern populations reached very similar numbers both years (Fig. 2).

Table 2.  Statistical significant sources of variation (main effects of strain, year and week, and their interactions) in demographic population parameters. F-tests from the mixed linear model described in the main text
ParameterSource of variationd.f.F-ratioP
Sex ratioYear*Week1, 904·43<0·001
 diversityHabitat configuration2,86·300·0270

The shape of the population trajectories towards the end of the experimental season suggested density-dependent population growth (Fig. 2) and this was confirmed by our statistical test for density dependence. All time lags (d) in the range from 3 to 7 weeks before the termination of the experiment all gave significant (P < 0·05) negative density dependence (β < 0). However, the best model was obtained for d = 3 weeks [scaling parameter: 1·16, Wald χ2 = 8·59, P = 0·003, β = – 0·30, 95% CI = (–0·50, – 0·10); see Fig. 3]. A time lag of 3 weeks roughly corresponds to the time it takes a newborn animal to enter the trappable part of the population. None of the experimental factors or year had a significant effect on population growth rate.

Figure 3.

Population growth rate over the three last weeks for the 12 experimental populations (see Fig. 2) given as r = log Nweek 42 – log Nweek 39 plotted against log Nweek 39. The fitted line is based on the estimates obtained from the model described in the main text.

Proportion of reproductive adults

The proportion of reproductive animals decreased from week 33 (Fig. 4). In the mixed linear model based on the weekly logit-transformed proportions weighted according to the reciprocal of their binomial error variances, none of the effects including habitat configuration were significant (P > 0·10). The only significant source of variation was the interaction between strain and year (Table 2). Figure 4 demonstrates the nature of this interaction, which was mainly caused by a low proportion of adults in the relatively large Northern populations in 1991. Indeed, there was a significant negative correlation between population size and the proportion of reproductive adults at the end of the experimental period (r = – 0·75, n = 12, P = 0·005) suggesting density-dependent sexual maturation.

Figure 4.

Functional stage structure (proportion reproductive adults of all animals) in the Southern and the Northern populations, and operational sex ratio (proportion adult females of all adults) in the 2 years of study.

Sex ratio

Operational sex ratio stabilized at a level between 50 and 80% females after the initial ‘latency period’ (weeks 27–33). Habitat configuration did not affect sex ratio (P > 0·10; linear modelling of logit-transformed proportions weighted according to their binomial error variances). The strain * week interaction was the only significant source of variation (Table 2, Fig. 4).

Matrilineal diversity

Effect of experimental factors. The test for the effect of habitat configuration on the population specific MD-indices from weeks 33–42 was significant (Table 2). Planned contrast to test for connectivity and fragmentation effects showed that only the connectivity contrast (F1,8 = 9·97, P = 0·013) was significant (fragmentation contrast: F1,8 = 0·02, P = 0·887). The matrilineal diversity was lower in the corridor populations than in the small fragment populations (Fig. 5). Strain was a source of variability independent of habitat configuration (F1,8 = 9·97, P = 0·013) due to a consistently lower matrilineal diversity in the Southern than in the Northern populations (Fig. 5). The variation in matrilineal diversity among the populations was unrelated to population size (r = 0·03, n = 2, P = 0·92), age structure (r = 0·20, n = 12, P = 0·54) and sex ratio (r = – 0·29, n = 12, P = 0·36). The lack of any relationship between matrilineal diversity and population size suggests that the lack of success of some matrilines was compensated for by the high success of other matrilines.

Figure 5.

Indices of matrilineal diversity by strain and habitat configuration.

Sources of variation in success of matrilines. Plotting the development of individual matrilines showed that the matrilines in populations with corridors tended to be more or less successful than matrilines in populations with small isolated fragments (Fig. 6).

Figure 6.

Developments of individual matrilines as exemplified by the corridor and the small fragment population of the Northern strain in 1991.

To explore the demography of individual matrilines we constructed a ‘matriline tree’ for each of the 42 matrilines in this study. A matriline tree describes the development of an individual matriline in terms of longevity and the production of litters of female matriline members from the colonization of the plots, until the termination of the study (Fig. 7). To facilitate numerical analyses of the information contained in these trees they were described by a set of variables (see Table 3) representing aspects of survival and reproduction for what appeared to be the two most influential cohorts for the development of the matrilines: the founder mothers and their daughters. A principal component analysis ran on these demographic variables showed that the success of mothers and daughters had little inter-correlation as their variation could be attributed to two separate principal components (Table 3). The reproduction and survival variables appeared to be equally loaded to the principle components, in particular for daughters as these variables were highly inter-correlated. Plots of the scores on the component axes indicated that some corridor matrilines appeared to be separated from the matrilines in small fragment populations, especially in the quadrant with negative scores for the success of both mothers and daughters (Fig. 8).

Figure 7.

Examples of three matrilineal trees of the 42 constructed in this study. Upper left: matriline from Southern corridor population. Upper right: matriline from Northern large fragment population. Lower: matriline from Northern corridor population. The thick stem represents the laboratory-born founder mothers, and the thin stems her daughters, their daughters and grand-daughters. Nodes in the tree represent birth of non-successful litters not surviving the nestling stage (open circles) and successful litters (closed circles) entering the trappable population.

Table 3.  Loadings of the two principal components on the variables describing survival and reproduction in founder mothers (three variables) and their daughters (five variables) in each matriline. These variables were obtained from the matrilineal trees exemplified in Fig. 7. The original variable received appropriate transformations before analysis
DescriptionPC 1PC 2
Longevity mother0·090·62
Number of litters delivered by mother0·210·60
Number of founder daughters0·210·41
Average longevity daughters0·44–0·12
Cumulative longevity of all daughters0·42–0·18
Proportion of daughters producing a litter0·44–0·09
Number of litters born by daughters0·44–0·13
Average number of daughter in litters0·37–0·19
 born by daughter
Variance explained55%22%
Figure 8.

Scores of each matriline on the two principle component axes describing the variation in the matrilines with respect to mother (PC 2) and daughter (PC1) fitness variables.

The scores on the daughter axis (PC 1) were highly correlated (r = 0·75, P < 0·001) with the successes of the matrilines (i.e. matriline size at the end of the field season), while scores on the mother axis (PC 2) were not (r = 0·18, P = 0·3751).


We have earlier shown that habitat fragmentation, and connectivity affected individual space use and spatio-social organization in these experimental root vole populations (Andreassen et al. 1998; Bjørnstad et al. 1998). These responses were in accordance with the scenarios predicted by Ims et al. (1993). However, in contrast to our expectations we have shown here that these responses were not accompanied by changes in demographic parameters, such as growth rate, functional stage structure (proportion reproductive adults) or sex ratio at the population level. In particular, the spatial matrilineal structure, which appeared to be more aggregated on small than on large fragments (Bjørnstad et al. 1998) was not associated with higher population growth rate. Such an association between spatio-social structure and population level demography has been predicted, based on speculative reasoning about kin-selection and adaptive behaviour (Charnov & Finerty 1980; Hestbeck 1982) or suggested based on empirical results (Kawata 1987; Lambin & Krebs 1991, 1993; 1994b; Lambin 1994a; Mappes, Ylönen & Viitala 1995; Lambin & Yoccoz 1998).

What appear to be conflicting results may hinge on the fact that the present study employed habitat fragmentation as a factor structuring the spatio-social organization, while most previous studies experimentally manipulated the spatial kin-structure by removals (e.g. Lambin & Krebs 1993) and introductions (Boonstra & Hogg 1988; Mappes, Ylönen & Viitala 1995). Yet other studies employed an observational approach examining populations with a natural spatial and temporal variation in the degree of relatedness between individuals (Kawata 1987; Lambin 1994a,b). It is possible that the benefits associated with living in tight matrilineal clusters were offset by the cost of space or resource limitations on small patches in our small fragmented populations. Accordingly, there is evidence that resource competition may alter the reproductive strategy of Microtus females under space limitations (Lambin 1994c; Aars et al. 1995).

Another possibility for the apparent discrepancy between our study and previous studies may be different level of analysis. Most studies, both concerning the effect of habitat structure and social factors on vital rates, have used individuals as the analytical unit. This is often the case even when population level replicates are available and inferences about population level phenomena are to be made (Hurlbert 1984). Large demographic variation at the level of individuals and litters have been shown to be at least partly due to the influence of habitat structure and social factors both in root voles (Aars et al. 1995; Gundersen & Andreassen 1998), as well as in other species (Kawata 1987; Lambin 1994a; Redpath 1995; Bowers et al. 1996; Collins & Barrett 1997). In the present study, the influence of habitat structure on intra-population demographic variance was captured by our measure of matrilineal diversity. Our analysis of matrilineal diversity showed that corridors between small habitat fragments acted to increase demographic variation at the level of matrilines. However, this habitat-related effect did not translate into differences in population demography, i.e. there was no correlation between matrilineal diversity and the other population parameters (e.g. final population size). Theoretical studies have shown that the relationship between demographic variation at different levels (e.g. individuals and populations) may be very complicated (Lomnicki 1988; Murdoch et al. 1992; Bjørnstad & Hansen 1993; Judson 1994; Sutherland 1996). For instance, structured demographic variation at one level (e.g. individuals or matrilines) may not necessarily propagate into structured variation at a higher level (e.g. population). Also other experiments probing the effects of habitat fragmentation and destruction on vole populations have failed to find population level effects on vital rates (e.g. Harper, Bollinger & Barrett 1993; Johannesen & Ims 1996; Wolff, Schrauber & Edge 1997).

We found that the variable success of matrilines was approximately determined by survival and reproductive rates within the founder litters and that this intra-matriline covariance was largest in populations with corridor-connected patches. Ultimate mechanisms underlying this increased demographic variance between matrilines is difficult to identify a posteriori. However, it is likely that the presence of corridors somehow acted to release the effects of intrinsic, subtle qualitative difference between matrilines. In the absence of information about such relevant innate characteristics, in particular spacing behaviour (sensuKrebs 1979, 1985), we can only use our previous analysis of space use and dispersal to speculate on what they may have been.

With respect to individual space use, the corridor plots could only be distinguished from the small fragment plots based on a larger proportion of animals that moved between fragments (Andreassen et al. 1998). The southern strain which had generally lower matrilineal diversity than the Northern strain (shown here) had also generally longer dispersal distances (Bjørnstad et al. 1998). Mean female dispersal distance was negatively correlated with our measure of matrilineal diversity (r = – 0·62, n = 12, P = 0·03). Although it is difficult to disentangle cause and effect on the basis of our data, we suggest that differences between matrilines in their ability to colonize and exploit space in a density-dependent setting may have been magnified in the presence of corridors between fragments. The apparently compensatory growth rate of successful matrilines does not imply that competition among matrilines was operating as compensatory growth could stem from density dependence at the population level. We have no evidence supporting the suggestion that matrilines are functional entities in root vole populations with properties important for population regulation as have been suggested for other Microtus species (Boonstra & Rodd 1983; Boonstra et al. 1987; Lambin et al. 1992).

Concerns about the effect of habitat fragmentation on population viability involves both demographic and genetic effects (Fahrig & Merriam 1994; Meffe & Carroll 1994). Although the results of the present experiment must be interpreted with some cautions regarding its limited spatial and temporal extent (cf. Wiens 1989), we suggest that the alterations of the spatial configuration of habitat patches (including the degree of fragmentation and connectivity) at the spatial scale of patchy populations (cf. Harrison 1991) may be more influential to population genetics than to demography (see also Aars & Ims 1999). It has become fashionable to use mtDNA markers to study genetic variation in natural populations of microtine rodents and other mammals (e.g. Plante, Boag & White 1989; Taberlet et al. 1995; Ishibashi et al. 1996; Jaarola & Tägelström 1996; Stacy et al. 1997; Aars et al. 1998). Such markers reflect the matrilineal structure of the populations (Avise 1995). As our study has indicated that corridors may facilitate spatial expansion and numerical dominance of some lineages at the expense of others, population genetic studies employing mtDNA techniques need to consider habitat connectivity when attempting to explain genetic diversity.


This work was a part of the project ‘Habitat fragmentation: Implications for the dynamics of populations’ funded by the Research Council of Norway (NFR). We thank the following persons for assistance in the field: Jon Aars, Tolli Agustsson, Kari W. Berg, Ottar Bjørnstad, Sue Evans, Jorun Fauske, Barbara and Stefan Halle, Karine Hertzberg, Thomas Hansteen and Kjell Isaksen. Jon Aars and Edda Johannesen are thanked for providing comments to a draft of the manuscript.