Multiple ecological processes underpin the eruptive dynamics of small mammals: House mice in a semi‐arid agricultural environment

Abstract Mouse plagues are a regular feature of grain‐growing regions, particularly in southern and eastern Australia, yet it is not clear what role various ecological processes play in the eruptive dynamics generating these outbreaks. This research was designed to assess the impact of adding food, water, and cover in all combinations on breeding performance, abundance, and survival of mouse populations on a typical cereal growing farm in northwestern Victoria. Supplementary food, water, and cover were applied in a 2 × 2 × 2 factorial design to 240 m sections of internal fence lines between wheat or barley crops and stubble/pasture fields over an 11‐month period to assess the impact on mouse populations. We confirmed that mice were eating the additional food and were accessing the water provided. We did not generate an outbreak of mice, but there were some significant effects from the experimental treatments. Additional food increased population size twofold and improved apparent survival. Both water and cover improved breeding performance. Food and cover increased apparent survival. Our findings confirm that access to food, water, and cover are necessary for outbreaks, but are not sufficient. There remain additional factors that are important in generating mouse plagues, particularly in a climatically variable agricultural environment.

One possibility is that rainfall increases the availability of high-quality food which promotes reproduction (Bomford, 1987a(Bomford, , 1987bSingleton, Krebs, Davis, Chambers, & Brown, 2001;White, 2002). However, the results of food addition experiments have been conflicting. In a recent meta-analysis of the effects of food supply and predation during 148 experiments, Prevedello, Dickman, Vieira, and Vieira (2013) found that food supplementation increased small mammal population densities 1.5-fold and that immigration was the major reason for this in open populations. There were no effects on survival, although increases in reproductive rate were detected, but were minor compared to immigration (Prevedello et al., 2013). In specific studies effects on reproduction have been variable. In Australia, supplementary food resulted in an increase in the proportion of females breeding and an extension of the breeding season (Bomford & Redhead, 1987), but supplementary food had no effect on breeding performance during a year when populations increased to high numbers even in the absence of supplementary food (Jacob, Hinds, Singleton, Sutherland, & Ylönen, 2007;Ylönen, Jacob, Runcie, & Singleton, 2003). In the latter study, the failure to increase breeding may have occurred because insufficient free water was available to exploit the high protein content of the dry supplementary food (Ylönen et al., 2003). In a replicated experiment preceding the one reported in this paper, the addition of supplementary food and water had no biologically meaningful effect on population size or breeding performance during a year when populations density remained low (Brown, Arthur, Jones, & Davies, 2008).
Predators of mice in agricultural areas in southeastern Australia include foxes (Vulpes vulpes), feral cats (Felis catus), raptors such as brown falcons (Falco berigora), black-shouldered kites (Elanus axillaris) and Australian kestrels (Falco cenchroides), barn owls (Tyto alba), and various species of snakes. Under seminatural conditions protection from predation using artificial cover has been shown to reduce both the lethal and nonlethal impacts of predators on house mice (Arthur, Pech, & Dickman, 2004. In that experiment mice in both predator exclusion enclosures and those provided with artificial cover began breeding earlier in spring than those that had limited protection from predators (Arthur, Pech, & Dickman, 2004). An early onset of breeding is characteristic of mouse population outbreaks in southeastern Australia, with breeding in an outbreak year usually commencing earlier in southern hemisphere spring, in the middle of August, compared with late September-early October in other years (Singleton et al., 2001).
In addition to a rapid population increase in spring from low numbers, outbreaks in this study area follow one of two patterns (Singleton et al., 2001. In some cases, numbers decline rapidly in autumn and remain low over the following spring/summer period, while in others numbers decline less rapidly and populations increase again the following spring. These two-year outbreaks cause considerable damage because they result in very high densities of mice from the early stages of crop development right through crop maturation (Brown, Huth, Banks, & Singleton, 2007;Caughley, Monamy, & Heiden, 1994). It is currently not known why these different patterns occur.
This study followed on from that of Brown et al. (2008) using as experimental units many of the same plots placed along fence lines in cereal production areas in the Victorian Mallee. From midwinter (July) 2004 to late winter (August) 2004, the same supplementary high-quality food and water treatments as the preceding study were still in place. In August 2004, we added artificial cover to half the plots to provide mice with protection from avian and mammalian predators. The cover remained in place until the completion of the study in winter (June) 2005, giving us a 2 × 2 × 2 factorial design.
We assessed treatment effects on the proportion of adult females breeding, overall population numbers, and the apparent survival rates of mice. We tested the following predictions.
1. If the absence of high-quality food or water, or high predation pressure, either alone or in combination was restricting the onset of breeding, then the appropriate combination of treatments (addition of food and/or water and/or cover) would result in an earlier onset of breeding by mice, or a higher proportion of adult females breeding earlier in the breeding season.
2. If the absence of high-quality food or water or high predation pressure, either alone or in combination was limiting mouse F I G U R E 1 Feral house mice (Mus musculus domesticus) can reach very high densities during outbreaks and cause significant damage to grain crops in Australia population size, then the appropriate combination of treatments would result in higher population sizes being reached.
3. If the absence of high-quality food or water or high predation pressure, either alone or in combination was limiting apparent survival of mice, then the appropriate addition of these factors would result in higher apparent survival rates.

| Study site and experimental design
The study was conducted on a grain and sheep property 3 km west of the Mallee Research Station at Walpeup, Victoria, Australia (35°08′S, 142°02′E) between winter (July) 2004 and winter (June) 2005. The area has an average yearly rainfall of 335 mm. More rain falls in winter when rainfall is less variable (May-August, mean = 130 mm CV = 39%) than in summer (January-April, mean = 89 mm CV = 63%).
The other four months (September-December) have on average 118 mm falls with a CV of 50%. Rainfall records for during and preceding the study were obtained from the Mallee Research Station and used to predict the likely background numbers of mice based on current models (Kenney et al., 2003). The farm was approximately 2,000 ha in size, and each year about 50% of the fields are planted with wheat or barley. The remaining 50% of the fields are pastures grazed by sheep. Typical pasture consists of the wheat and barley stubble remaining from the previous year's harvest.
Plots (experimental units) were 240 m sections of internal boundary fences with winter cereal on one side (barley or wheat) and pasture (normally grazed) on the other (Figure 2). Twelve of our 16 plots were the same as those used in Brown et al., (2008), but four plots had to be moved to maintain cereal crop on one side and pasture on the other, because of crop rotation at our study site. For the first period of the study (July 2004-August 2004; 42 days), our treatments were additional food and additional water in a 2 × 2 factorial design, that is, four replicates of each combination. In spring (August) 2004, we randomly added cover to two of the four of each combination to give a 2 × 2 × 2 factorial design (Figure 2), that is, two plots were provided with additional food, water, and cover; two with additional food and cover; two with additional water and cover; two with additional cover only; two with additional food and water; two with additional food only; two with additional water only; and two were experimental control plots. These treatments then remained in place until the end of the experiment in June 2005. Each plot was separated from others by at least 300 m.

| Experimental treatments
The basic layout of each experimental treatment is shown in Figure 3.
The additional food was pelletized rat/mouse food (Rat and Mouse Breeder Pellets, Gordon's Specialty Stockfeeds, Yanderra, Australia) with a minimum crude protein content of 23%, minimum crude fat 6%, and maximum crude fiber 5%. At least one kilogram of food was added to each of ten five-liter plastic containers. Up to 2.5 kg of food was added to frequently visited food stations. Containers were spaced evenly along the 240 m length of each plot (one every 20 m) and tied to the bottom of the fence using tie wire. At each trapping session, the amount of food removed was recorded (based on change in weight from previous session) and additional food was added when necessary.
Additional water was provided in 20 L plastic containers. A laboratory water nozzle (AHS 25, 65 mm, CF Maddock and Company) was fitted to a rubber stopper that was placed in a hole cut approximately 50 mm from the bottom of the container. Wild mice under laboratory conditions are known to learn the technique of using the nozzles quickly, and we had evidence from the preceding experiment that mice in the field would use the water (Brown et al., 2008). Water containers were spaced evenly every 20 m midway between the food stations (Figure 3). At each trapping session, the nozzles on all water containers were tested by hand for proper function to ensure they had not become clogged and the water levels were checked and maintained to at least 10 L (half capacity). From October 2004 to the completion of the experiment, three additional 20-L drums of water were placed in fixed locations along water treatment plots, approximately 40 m from each end of the plot and one in the middle. These were added to provide animals with another source of water, supplementing the existing method of water provision. These water stations were buried so the tops of drums were at ground level. There was a 60 mm opening on the top of each drum. Secured to the opening was a 60 cm length of 15 mm nylon rope that was placed into the drum to provide climbing access to and from the available water. These stations were covered with a 1 m 2 sheet of corrugated iron to reduce sand and dust contaminating water. All additional water drums were filled to at least two-third full. The water level of additional drums was monitored during every session and replenished when necessary. Additional water stations were always covered by the existing wire mesh (described below) if they were on cover treatment plots.
In order to test whether mice were drinking from the water stations, a fluorescent nontoxic, flavorless, and odorless xanthene dye (Rhodamine B) was added to the water (0.2 g/L) on several plots where mouse densities were highest. This included either or both the above ground and or the buried water stations. If mice were drinking the water, Rhodamine B would appear in the blood up to three days after drinking or in the whiskers for up to seven weeks after drinking (Jacob, Jones, & Singleton, 2002;Fisher, 1999). On these plots, blood samples and whiskers were collected and analyzed using the methods described in Jacob et al., (2002).
Empty food and water containers were placed at all plots where food or water was not provided to control for any effect of the containers, for example, provision of cover. We anticipated mice would use the containers as shelter to build burrows beneath and that these burrows might lead to some mice monopolizing the shelter, the food, or the water. To avoid this, at each trapping session, containers that had burrows under them were moved up to 3 m along the fence.
Additional cover was provided using Waratah rabbit proof mesh (1.2 m wide, 105/4/1.4 gauge) that was cut into 40 m lengths ( Figure 3). Three lengths were placed at each cover plot and fixed length ways along the fence. One length was positioned in the middle of the fence line, and the other two lengths were positioned at either end of the fence line. The wire mesh was secured to the fence approximately 30 cm above ground, and the outer edge was crimped every 2 m by hand so the mesh would remain above ground level and provide adequate cover. On sites without additional cover, no wire netting (cover) was provided.

| Live trapping
Mice were live-trapped using Longworth small mammal traps (24 × 7 × 9 cm, Longworth Scientific) on all plots for three consec-

| Effect of treatments on reproduction
We assessed the effect of treatments on the proportion of adult females >71 mm in length that were in breeding condition (evidence of lactation or pregnant as determined by palpation). This length threshold is based on previous studies of house mice, which show that females can become sexually mature at this length (Singleton, 1983). We used generalized linear modeling with binomial errors. Adjusted Akaike information criterion (AIC c ) was calculated from the minimized negative log-likelihood using standard F I G U R E 2 Layout of study site between Walpeup and Toritta, northwestern Victoria. There were 16 experimental sites established, each 240 m in length. There were two replicates of each treatment (untreated control, food, water, cover, food + water, food + cover, water + cover, and food + water+cover), which were assigned randomly. Sites were established along internal fence lines between a wheat crop (shaded) and pasture for sheep grazing (nonshaded) ToriƩa 500 m N = Wheat crop = Trees = Houses = Untreated Control = Food = Water = Cover = Food + Water = Food + Cover = Water + Cover = Food + Water + Cover

Study site
formulas (Burnham & Anderson, 2002). The weight of support for each model conditional on both the data and all models in the set was calculated as described in Burnham and Anderson (2002).
Analyses were carried out in program R (R Development Core Team, 2007).

| Effect of treatments on population size
Population size was estimated using the Jackknife estimator in Program Capture (Otis, Burnham, White, & Anderson, 1978). The Jackknife estimator, which allows for individual heterogeneity in capture probability, may be biased low for estimating abundance of house mice, but still performs well as an index of abundance and hence is appropriate for comparing changes in population size under the treatments (Davis, Akison, Farroway, Singleton, & Leslie, 2003).
There was no evidence of correlation structure between successive measurements and results clearly indicated no treatment effects in the early stages of the experiment, so the effect of treatment on mouse abundance was analyzed separately for each trapping session. We used generalized linear modeling with normal errors.
Residual plots indicated that these were appropriate and data transformation was not required. Adjusted Akaike information criterion (AIC c ) was calculated from the minimized negative log-likelihood, and the weight of support for each model conditional on both the data and all models in the set was used for inference (Burnham & Anderson, 2002). Akaike weights were used to calculate modelaveraged parameter estimates and model average standard errors.
These estimates incorporate uncertainty from both individual models and the relative support of each model in the set (Burnham & Anderson, 2002).

| Effect of treatments on apparent survival
Mark-recapture modeling of individually marked mice was used to assess treatment effects on apparent survival (Φ), which includes actual survival and permanent emigration. The winter (August) 2004 session was used as the starting point for survival analyses because this was the time when all treatments including the cover treatment were in place. Program U-CARE was used to test the goodness of fit (GOF) of a simplified starting model with two "age-classes" for Φ and with time dependence in both age-classes (Choquet, Reboulet, Lebreton, Gimenez, & Pradel, 2005). The "age-classes" divide mice into those previously marked and those marked for the first time at a particular trapping session. The former tended to be adult animals, while the latter was a mix of juveniles, transient adults, and resident adults that escaped capture on previous occasions (Arthur, Pech, & Dickman, 2005). Insufficient data were available to define ageclasses based on size.
Combining tests 3.sm, 2.ct, and 2.cl in U-CARE indicated that a model with 2 "age-classes" for Φ and time-varying capture probability (p) was a good starting point for model selection. This model produced a 2 8 value of 7.8 (p = .45) indicating there was no need to adjust for overdispersion (Choquet et al., 2005). Markrecapture modeling was then conducted using the program R (R Core Development Team, 2007) package RMark (Laake & Rexstad, 2008) as an interface for Program MARK (White & Burnham, 1999).
Adjusted Akaike criteria were used to calculate Akaike weights. These combined with model-averaged parameters estimates were used for inference (Burnham & Anderson, 2002). The best model F I G U R E 3 Schematic representation of a 240 m experimental site showing the approximate location of the three trap lines, the food and water containers and the wire netting (three sections of 40 m) along a fence line between a wheat crop and a pasture crop used for sheep grazing. Three additional water stations (20 L drums) were provided (not shown). For sites without supplementary food or water, empty food and water containers were provided. On sites without supplementary cover, no wire netting (cover) was provided

| Biomass and grain samples
To measure availability of natural food and cover, biomass and grain samples were collected using the same techniques as described by Brown et al., (2008). Biomass samples were estimated from 0.25 m 2 quadrats (n = 20 per site) using a modified comparative yield technique (Friedel & Bastin, 1988;Haydock & Shaw, 1975). Reference

| Rainfall
Based on the logistic regression model of Kenney et al. (2003), the probability of an outbreak from the rainfall that spanned our experiment ( Figure 4) was 0.45 ("moderate" to "high") given the relatively good rainfall through summer (November 2004, December 2004, and January 2005. Using the "early predictor" model of Kenney et al. (2003), the maximum autumn abundance was predicted to be an adjusted trap success (ATS) of 61 which would be considered a small outbreak . Using the "late predictor" model of Kenney et al. (2003), the maximum autumn abundance was predicted to be ATS of 15. These predictions suggest relatively good conditions for mice during our experiment, at least in the early stages of the experiment, that is, spring 2004.

| Use of food and water
On plots provided with supplementary food throughout the experiment consumption of food between consecutive sessions ranged from a minimum of 96 g station −1 plot −1 to a maximum of 1,359 g sta-

| Predator activity
Signs of fox (Vulpes vulpes) activity including tracks in the sandy soil, scats, and digging were observed around all sites during all trapping sessions. On a few occasions foxes disturbed traps. Raptors including brown falcon (Falco berigora), Australian kestrel (Falco cenchroides), and black-shouldered kites (Elanus axillaris) were observed in the study area during all trapping sessions. There were also signs of snake activity including tracks and visual observations.
Based on ~30 years experience working on mice in the area, predator levels were typically low as found in nonoutbreak years.

| Effect of treatments on breeding
Pregnant and/or lactating females were caught in August 2004 across all treatments, and hence, breeding continued through winter (Brown et al., 2008). Up to 6 adult females were caught on plots in this trapping session (n = 43 in total), representing 37% (±7) (Table 1), 67% (±7, n = 46) of adult females were breeding on plots with added cover and 30% (±8, n = 33) of adult females were breeding on plots without artificial cover. Based on the second-ranked model, 75% (±7) of females were breeding on plots with water and added cover, 41% (±11) were breeding on plots with water but no added cover, 56% (±10) were breeding on plots with cover but not water and 23% (±8) were breeding on plots without water or cover treatments. At this time, trapped adult females were larger on plots with added cover and water than on other plots (mixed-effects model water:cover interaction term t 12 = 2.4, p = .03; cover and water length = 86.5 ± 1.1 mm, water only = 79.6 ± 1.6 mm, cover only = 82.8 ± 1.4 mm; neither treatment = 82.5 ± 1.4 mm).
There was no strong support for any treatment effects on the proportion of adult females breeding in midsummer ( Abbreviations: AIC c , Akaike information criterion adjusted for sample size; ΔAIC c , the difference between the model and the best model; N, number of parameters; ω i , Akaike weight.  (Table 2). Populations were larger on plots with both the food and water treatment in combination compared with plots where only one or the other was provided (Figure 6). There was weak evidence that food or water in isolation resulted in higher numbers compared with plots where neither was present ( Figure 6).

| Effects of treatments on population size
In midwinter (June) 2005, the top seven models, with a combined weight of 0.96, included an effect of food (Table 2).
Populations were approximately two times larger on plots with added food compared to plots without added food (Figure 7).
There was weak evidence for a small additive effect of the cover treatment; plots with added cover appeared to have slightly more

| Effects of treatments on apparent survival
The top 18 models (collective weight = 0.924) and 23 of the top 25 models included an effect of food on apparent survival, with a collective model weight of 0.958 of a possible 0.972 (Table 3).
There was no support of models where capture probability varied by plot. In the top 25 models, the collective model weight for models including added cover was 0.654 and for models including water was 0.371 (Table 3)

| D ISCUSS I ON
We were unable to generate an outbreak through provision of food, water, cover, or any combination of these. Our experiment commenced in winter, and during the early stages, background conditions were such that the early predictor model of Kenney et al. (2003) indicated the potential for a small outbreak. We did see a twofold increase in population density late in the experiment on sites where additional food was provided, which was slightly higher than the 1.5fold increase in population densities found by Prevedello et al. (2013) from a global meta-analysis of food addition experiments. While we did not generate an outbreak, some of the responses to our treatments were consistent with demographic patterns associated with outbreaks which included treatment effects on both reproduction and apparent survival.
Mouse population eruptions in southeastern Australia are characterized by an early onset of breeding  Note: The null model is an intercept-only model where the proportion breeding is the same in all plots.
Abbreviations: AIC c , Akaike information criterion adjusted for sample size; ΔAIC c , the difference between the model and the best model; N, number of parameters; ω i , Akaike weight.

F I G U R E 6
Model-averaged estimates (±SE) of population size in response to treatments in autumn (April) 2005, based on the top 5 models (combined Akaike weight 0.85) shown in Table 2, that is, the 5th ranked null model is included in the estimates  Abbreviations: AIC c , Akaike information criterion adjusted for sample size; ΔAIC c , the difference between the model and the best model; N, number of parameters; ω i , Akaike weight; a2, two "ageclasses"-newly marked and previously marked; p(.), constant capture probability.

TA B L E 3 Model selection table for the effects of treatments on apparent survival (Φ)
2001), and we found that breeding continued over winter and spring in 2004 on many of our plots regardless of treatment. It has been suggested that early breeding occurs when pasture growth and seed set promotes it (Singleton, 1989;Mutze, Veitch, & Miller, 1990;Pech et al., 1999) and our measurements were consistent with this; we found high levels of biomass of grasses and weeds along the area between the crop and fence lines early in the experiment. However, despite the apparent importance of food, Jacob et al. (2007) found that food addition did not alter ovulation rates or litter size. We also found no effect of food addition alone on early breeding, both in this experiment and in our preceding one (Brown et al., 2008). Adding water alone also had no effect on early breeding. However, our cover treatment appeared to further increase the proportion of females breeding in spring.
In a seminatural setting, protection from predation, either through predator exclusion fencing or by providing cover (in the same way it was provided in this experiment), has been shown to decrease the nonlethal impacts of predators on house mice (Arthur et al., 2004). In that study, protected mice had higher growth rates and began breeding earlier in spring when protection from predation made them more willing to access high-quality supplementary food. Female mice on the water plus cover treatment in our current experiment were larger and also had the highest proportion breeding in spring, consistent with this. The results suggest that even in the presence of natural cover, artificial cover may have reduced the nonlethal impacts of predation, but that this was not sufficient to produce population level increases at the scale of our plots.
As the experiment progressed, background conditions probably became less favorable for mice, based on the "late predictor" model of Kenney et al. (2003). There was evidence that additional food and cover increased the apparent survival rate of newly marked and previously marked animals, with food having the larger effect. Prevedello et al. (2013) found that population effects were larger when predation was reduced and populations were open to immigration, and that immigration was more important than survival.
There was evidence that mice were eating the food (reduction in weight of food containers) and drinking the water (evidence of Rhodamine B in samples from February-June) at treatment plots, but was this enough? There were up to 40 mice on each plot, and there was roughly 400 g eaten per station (10 stations per plot), so up to 4 kg food consumed/plot. This is equivalent of up to 100 g per mouse per monitoring interval (roughly every 2 months), so equivalent of about 1.67 g day −1 mouse −1 of additional food consumed. Bomford (1985) estimated that an adult mouse eats about 2.5 g of food each day (roughly 10% of body weight), so the additional food provided roughly two-thirds of the food required for a typical mouse to meet its daily energy requirements. There was also a large amount of food available as the crop matured (grain yield ~ 2,000-3,000 kg/ha), some of which was spilt at harvest time (722 kg/ha), potentially swamping any food addition at the treatment plots.
House mice (Mus musculus domesticus) are physiologically well adapted to semi-arid environments and can survive without freely available water by obtaining water from their food (Fertig & Edmonds,

F I G U R E 8
Model-averaged estimates of apparent survival from the top 10 models in Table 3. The water treatment made no biologically significant difference to apparent survival rates, so it has been removed from the figure to aid clarity. For example, the model-averaged predictions for the survival rates under the food and cover and water treatment were essentially the same as those for the food and cover treatment 1969; Moro & Bradshaw, 1999;Mutze, Green, & Newgrain, 1991;Prakash & Ghosh, 1975). During dry summers, breeding can be limited by low water availability (Newsome, 1969). The level of moisture stress can depend on a range of factors including rainfall, availability of food or shelter and the amount of weeds present (Mutze et al., 1991). Providing additional water for a field population of mice in California USA increased population size by about 35% compared with the experimental control site (Newsome, Stendell, & Myers, 1976), suggesting that water was a key limiting factor. Models with time-varying effects of treatments did not fall in the top 25, possibly because sample sizes were too small to detect time-varying effects.
There was no effect of treatment on population size up until late Mice are responding to multiple ecological processes, and by trying to make favorable conditions (through provision of food, water, and cover), we were unable to generate an outbreak or a mouse plague. There are a range of other intrinsic factors (e.g., social structure, infanticide, sexual maturation, aggression, dispersal, density dependence) and extrinsic factors (e.g., habitat, cover, food, predation, disease, weather) that are likely to be influencing mouse populations. The rapid decline observed in mouse populations (Brown, 2006) is thought to be driven by extrinsic factors (food depletion and disease; Brown, Singleton, Pech, Hinds, & Krebs, 2010) but also via infanticide (see Sutherland et al., 2005).
Food, water, and cover are important, but not sufficient to generate a mouse outbreak. There has been lots of conjecture about the effects of spatial heterogeneity in these ecosystems (Chambers, Singleton, & Wensveen, 1996;Krebs et al., 1995;Redhead, 1982), but these effects have not been quantified sufficiently to understand what is going on with movements of mice. Low recapture rates are typical (Krebs, Singleton, & Kenney, 1994), which suggests movements at local scales could be quite important. In vertebrate populations with multi-annual fluctuations, changes in social behavior and kin structure have been proposed as a causal mechanism for changes in spacing behavior which results in density fluctuations (Krebs et al., 1995;Sutherland et al., 2005). The maintenance of female kin groups through the preceding winter significantly improved recruitment during the subsequent breeding season and therefore is necessary for generating mouse outbreaks (Sutherland et al., 2005).
Stress-related genes may also play a role in tempering mouse population dynamics because stress genes are known to flow through to granddaughters especially after peak population densities have been reached (Boonstra, 1994).
Ultimately, we want to be able to confidently forecast when mouse populations are likely to cause economic damage and to encourage farmers and other land managers to take appropriate precautions to manage the risks. Our current models are still adequate for this purpose (Kenney et al., 2003;Pech et al., 1999), but improving our understanding of the mechanisms by which mice increase or decrease would add-value to these models and further improve our forecasting capability.

ACK N OWLED G M ENTS
We sincerely thank Mr Stan Stone and family for allowing us to conduct this project on his property. We are indebted to the ongoing support and rigorous discussions with Drs Grant Singleton, Lyn Hinds and Charles Krebs. We are grateful for comments from Wendy Ruscoe and Elliott Luck that improved the manuscript.
Funding for this project partly came from an ACIAR funded project "Improved management of small mammals in Tibetan grasslands" (AS1/2002/108). All experiments and procedures were approved by the CSIRO Sustainable Ecosystems Ethics Committee and conform to the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (Approval No was 03/04-09).

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R S CO NTR I B UTI O N S
All authors designed the study and interpreted the results. PRB and ADA conducted the analyses and led the writing of the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used for this study are available through the CSIRO Data