The importance of variation in the breeding performance of seed-eating birds in determining their population trends on farmland

Authors

  • Gavin M. Siriwardena,

    1. British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK; and
    2. Ecology and Behaviour Group, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK
    Search for more papers by this author
  • Stephen R. Baillie,

    1. British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK; and
    Search for more papers by this author
  • Humphrey Q.P. Crick,

    1. British Trust for Ornithology, The Nunnery, Thetford, Norfolk IP24 2PU, UK; and
    Search for more papers by this author
  • Jeremy D. Wilson

    1. Ecology and Behaviour Group, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK
    Search for more papers by this author
    • *

      Present address: Royal Society for the Protection of Birds, The Lodge, Sandy, Bedfordshire SG19 2DL, UK


Gavin M. Siriwardena (fax 01842 750030; e-mail gavin.siriwardena@bto.org).

Summary

1. Changes in agriculture have been linked to widespread declines in farmland bird populations. One approach to the identification of the causes of observed population changes is to investigate historical changes in national demographic rates.

2. We analysed the British Trust for Ornithology’s nest records database to investigate whether long-term farmland population trends could have been driven by changes in several components of the annual breeding performance of 12 granivorous bird species. Clutch size, brood size, chick : egg ratio and daily nest failure rates were analysed with respect to blocks of years during which abundance (as measured by the Common Birds Census) was increasing, stable or declining. The individual components of breeding performance were combined to provide estimates of the production of fledglings per breeding attempt.

3. Most species’ population declines were not associated with poor breeding performance per attempt. Effects of environmental change on this parameter therefore cannot be a general mechanism behind the major population declines seen.

4. A fall in linnet Carduelis cannabina fledgling production per attempt, driven primarily by increased nest failure rates during the egg period, represented the strongest evidence for an important effect of breeding performance on abundance. This change could have driven the principal population decline (1975–86) for this species.

5. Conversely, at least for the declining turtle dove Streptopelia turtur, skylark Alauda arvensis, tree sparrow Passer montanus, yellowhammer Emberiza citrinella and corn bunting Miliaria calandra, breeding performance per attempt was higher while populations declined.

6. Variation in annual survival and fledgling production per breeding attempt alone could not explain changes in abundance for at least seven species. This may suggest that changes in post-fledging survival rates and/or the number of breeding attempts per year could have been important.

7. Management to improve over-winter survival may be critical in reversing the population trends of most declining species, but such management might still best be directed at the breeding season. Post-fledging survival rates and the number of breeding attempts made within a season are little-studied demographic variables that are high priorities for future research and long-term monitoring.

Introduction

Declines in the range and abundance of many species of farmland bird since the early 1970s have caused great conservation concern across Europe (Marchant et al. 1990; Gibbons, Reid & Chapman 1993; Tucker & Heath 1994; Fuller et al. 1995; Gibbons et al. 1996; Baillie, Gregory & Siriwardena 1997). Many aspects of the widespread agricultural intensification that began or accelerated in the 1970s have been implicated as the causes of these declines, through their effects on the availability of nest sites and food sources for birds (Fuller et al. 1995; Campbell et al. 1997; Baillie, Gregory & Siriwardena 1997). These changes include increased agrochemical usage, the simplification of crop rotations, the intensification of grassland management (through re-seeding, fertilization and drainage) and the loss of seed-rich winter foraging habitats such as over-winter stubbles and open grain stores. Such changes could reduce the carrying capacity of farmland in a variety of ways, acting through survival or breeding success, according to when in the year and how a species makes use of affected farmland habitats.

While intensive studies have shown that certain aspects of agricultural intensification have caused the declines of species such as grey partridge Perdix perdix and corncrake Crex crex (Potts 1986; Stowe et al. 1993; Green, Rocamora & Schäffer 1997), such long-term research cannot be conducted economically for all the species affected. As a result, the mechanisms underlying most of the declines remain unknown (Baillie, Gregory & Siriwardena 1997), as do the reasons for the occurrence of contrasting trends for a range of other species on farmland. As population change must result from shifts in the balance between fecundity, mortality, immigration and emigration, historical demographic data spanning the periods of population change can be used to investigate the mechanisms behind population changes. Such historical data can be combined with intensive studies, both to point to new hypotheses to be tested in the field and to conduct tests of ideas generated by intensive work. Identifying how demographic rates have changed relative to observed population trends should help to distinguish between competing hypotheses as to the causes of species’ declines, suggest where intensive field tests of such hypotheses might best be focused and indicate the stage of the life cycle at which conservation measures should be targeted.

Changes in farmland bird abundance in Britain have been monitored by the British Trust for Ornithology’s (BTO) Common Birds Census (CBC) since 1962. Results from this census have alerted the conservation community to declines in many species. Methods have recently been developed to identify objectively periods of consistent increase, decline and stability within CBC time series, which are likely to reflect long-term changes in the underlying demographic rates (Siriwardena et al. 1998a). Other BTO data sets allow the integration of information on demographic rates with abundance data (Baillie 1990; Greenwood et al. 1993). For example, analyses of long-term ring-recovery data from the BTO’s ringing scheme have shown that the observed variations in survival are sufficient to have caused the recent changes in abundance of goldfinch Carduelis carduelis L., house sparrow Passer domesticus L., song thrush Turdus philomelos L. and reed bunting Emberiza schoeniclus L. (Thomson, Baillie & Peach 1997; Peach, Siriwardena & Gregory 1999; Siriwardena, Baillie & Wilson 1999), and are at least consistent with a causal relationship with abundance for a further nine species (Siriwardena, Baillie & Wilson 1998). Such relationships between abundance and survival do not preclude an important additional influence of breeding success, however, and variations in survival alone do not explain the population trends observed for species such as bullfinch Pyrrhula pyrrhula L. and linnet Carduelis cannabina L.

The BTO’s Nest Record Scheme (NRS; Crick & Baillie 1996) provides historical information on a range of components of breeding performance. In this paper, the relationships between variations in breeding performance and population trend are investigated for 12 predominantly granivorous species that are typical of British farmland: stock dove Columba oenas L., turtle dove Streptopelia turtur L., tree sparrow Passer montanus L., skylark Alauda arvensis L., chaffinch Fringilla coelebs L., greenfinch Carduelis chloris L., goldfinch, linnet, bullfinch, corn bunting Miliaria calandra L., reed bunting and yellowhammer Emberiza citrinella L.

We used nest record data from 1962 onwards to relate breeding performance variables to the periods of increase, decline and stability identified from the species’ CBC indices (Siriwardena et al. 1998). Higher breeding performance in a period with a more positive population trend (e.g. increasing) than in an adjacent period with a more negative trend (e.g. decreasing) would be consistent with the differences having driven the observed changes in abundance. The converse pattern might (for example) suggest a density-dependent response of breeding performance to population changes caused by another demographic mechanism. We used simple demographic models to estimate overall annual productivity from survival and abundance data, and used a comparison between this quantity and the numbers of fledglings produced per breeding attempt (derived from our nest records analyses) to estimate the importance of variation in the unmeasured components of annual productivity.

Our approach assumes that there is no net immigration to or emigration from farmland at the national scale. Large-scale exchanges of individuals between farmland and other habitats, as might occur in spatially structured populations exhibiting source–sink dynamics (Días 1996), could, in principle, mask relationships between breeding performance and abundance within farmland. However, in Britain at least, farmland is the single most important habitat for all these species except bullfinch, an equal proportion of whose population occurs in woodland (Gibbons, Reid & Chapman 1993; Gregory & Baillie 1998; Gregory 1999) and none of the species has undergone a markedly different population trend in woodland and farmland (Marchant et al. 1990; Crick et al. 1998). Our analyses are thus representative of a very high proportion of the UK ranges and populations of the species considered and are unlikely to be influenced by the demographic effects of local fluxes of individuals between habitats.

Methods

Cbc data and periods of consistent trend direction

The CBC is a volunteer-based survey in which counts of breeding territories are made over multiple recording visits to defined survey plots. Numbers of territories are determined using a mapping method and the subsequent interpretation of map registrations by trained analysts. Census plots are mostly concentrated in southern and eastern Britain, in which area plots defined as farmland are representative of agricultural land use (Fuller, Marchant & Morgan 1985). There has been a considerable turnover of CBC plots over time, but with no appreciable change in habitat coverage (Marchant et al. 1990). Further details of the CBC are published elsewhere (Marchant et al. 1990; Greenwood et al. 1993; Siriwardena et al. 1998). Siriwardena et al. (1998) developed a method for the identification of years where a species’ CBC abundance index (calculated by the Mountford method: Mountford 1982, 1985; Peach & Baillie 1994) was turning significantly. This used an estimate of the second derivative of the CBC index series (after it has been smoothed non-parametrically) and 95% confidence intervals estimated by bootstrapping. Periods between such significant turning points can be considered to be consistent in trend direction [although it is possible for consecutive periods to have the same trend directions (up or down) but different slopes; Siriwardena et al. 1998]. We used the turning points in the CBC index series for farmland plots identified by Siriwardena et al. (1998) to divide each of the above species’ trends into periods with trends classifiable as increasing, declining or stable. Where significant curvature in a trend was found over 2 or more successive years, we used the year in which the second derivative was most significantly different from zero to delimit trend periods on either side of the turn in the series. Each delimiting year was considered to fall into the trend period following the turning point: breeding performance in the delimiting year will be an influence in the subsequent interannual change in abundance. Each period between turning points was classified according to trend direction. Stable periods were defined as those where the mean of the annual estimates of the first derivative of a smoothed index series (after Siriwardena et al. 1998) fell between + 0·01 and − 0·01 (i.e. slopes of less than 1% per year: the estimates of the first derivative were first divided by the appropriate annual abundance index value so that they were interpretable as percentage changes); steeper slopes were classified as increasing or declining. The year-blocks thus identified were used in the nest record analyses described below.

The above procedure failed to identify turning points in the index series for three species, as sparse count data at the beginning (stock dove) or the end (turtle dove and corn bunting) led to wide confidence intervals around the species’ CBC trends and their second derivatives (Siriwardena et al. 1998). We therefore estimated the position of a single turning point in the index series of each of these species by eye, reflecting the single major change in trend direction in each case (inspection of the smoothed CBC trends for each of these species in fact suggests that only one important change has occurred since 1968 in each case; Siriwardena et al. 1998). While this procedure was not ideal, it allowed us to define year-blocks with broadly consistent trends for these species, thus making possible analyses of the breeding performance of the species that have undergone some of the largest declines of all. We noticed that changing the position of the estimated turning points by 1 or 2 years had little effect on the results when the periods compared were long. Table 1 lists the year-blocks used for each species.

Table 1.  Time periods with consistent trends as identified from turning points (Siriwardena et al. 1998a) in the smoothed CBC trend for each species. Estimates of block-specific population trend gradients were obtained by estimating the first derivatives of the smoothed CBC trend (after Siriwardena et al. 1998) and taking means by year-block after each annual estimate had been divided by the appropriate annual index value (to allow interpretation as percentage changes). Periods with mean first derivatives (shown in parentheses) in the range − 0·01 to 0·01 were classified as ‘stable’, mean first derivatives below this range indicated ‘declining’ trends and means above it ‘increasing’ trends. Note that successive periods with the same trend direction can occur when the two periods differ in the gradient of the trend
SpeciesNumber of year-blocksTime periods (mean first derivatives)
  • *

    No significant turning points were identifiable in the index series for corn bunting, stock dove and turtle dove because the confidence intervals around the second derivative curve were too large. The position of a single turning point was therefore estimated by eye to allow the comparison of different sections of the time series (see text).

Bullfinch51962–65 (0·155), 1966–74 (0·025), 1975–78 (–0·107), 1979–82 (–0·089), 1983–95 (–0·008)
Chaffinch31962–64 (0·077), 1965–87 (0·012), 1988–95 (–0·002)
Corn bunting*21962–73 (0·044), 1974–95 (–0·057)
Goldfinch51962–66 (–0·048), 1967–70 (–0·016), 1971–74 (0·114), 1975–86 (–0·041), 1987–95 (0·066)
Greenfinch51962–66 (0·130), 1967–74 (0·009), 1975–77 (–0·036), 1978–84 (–0·015), 1985–95 (0·016)
Linnet61962–66 (0·057), 1967–69 (–0·054), 1970–74 (0·019), 1975–78 (–0·069), 1979–86 (–0·061), 1987–95 (0·028)
Reed bunting41962–64 (0·006), 1965–75 (0·069), 1976–78 (–0·126), 1979–95 (–0·008)
Skylark41962–66 (0·022), 1967–77 (0), 1978–82 (–0·069), 1983–95 (–0·027)
Stock dove*21962–85 (0·079), 1986–95 (–0·002)
Tree sparrow21962–76 (0·009), 1977–95 (–0·102)
Turtle dove*21962–78 (–0·008), 1979–95 (–0·063)
Yellowhammer61962–65 (0·047), 1966–77 (–0·002), 1978–81 (0·013), 1982–87 (–0·006), 1988–91 (–0·050), 1992–95 (–0·027)

Nest record data

The BTO’s NRS (reviewed in detail by Crick & Baillie 1996) consists of a national network of volunteer observers who submit standardized records of nest contents, location, habitat and evidence of success or failure derived from one or more visits to a nest site. The NRS has run since 1939, and at least 100 nest record cards (NRCs) are received each year for each of 50 plus species from a range of habitats and from throughout Britain. Depending on the quality of the data on a given NRC (such as the number of nest visits on which the record is based), some or all of first egg dates, clutch and brood sizes, chick : egg ratio and daily nest failure rates can be estimated. For five of the species considered, the number of NRCs received by BTO in some years has exceeded the number for which resources have permitted computerization of the data, so some NRCs were unavailable for analysis. For chaffinch, greenfinch, linnet and reed bunting, the majority of the NRCs received have been computerized, and exclusions were chosen at random (except that single visit cards, which supply little information, were given low priority). For tree sparrow, annual samples of NRCs were selected for computerization in order to maximize the number of observers contributing to the sample (an equal number of NRCs was drawn randomly from those submitted by each observer).

For this study, the habitat data recorded on NRCs were used to select cards only from habitats matching those found on farmland CBC plots (Fuller, Marchant & Morgan 1985). Nest record habitat data consist of a hierarchical scheme in which a nest is assigned first to a general category (woodland, grassland, farmland, etc.) and then to a combination of three further levels of habitat detail (Crick 1992; Crick, Dudley & Glue 1994). Prior to 1990, habitat data were recorded differently (using a single code representing one of around 100 habitat types together with additional information about key components of the habitat). These data were translated into their equivalents under the current system. We used all NRCs classified as ‘farmland’, as well as those from dry grassland, water meadows and downland. The habitat selections ensured that the populations ‘sampled’ by the NRC and CBC databases were as similar as possible. We analysed nest record data in terms of clutch size (the maximum number of eggs found in a nest), brood size (the maximum number of young found in a nest), chick : egg ratio (the ratio of brood size to clutch size where the whole nest did not fail) and daily nest failure rates before and after hatching (see below), in order to investigate the variation in breeding performance per nesting attempt between the periods shown in Table 1. Clutch size data were rejected if egg laying could have continued after the last visit of the recorder. Our measure of brood size is likely to overestimate the brood size at fledging, but will approach it if mortality early in nestling life (when chicks are most vulnerable) is the most significant form of partial brood loss. The chick : egg ratio measure used here will therefore incorporate these early losses, as well as hatching success (the proportion of the eggs in the clutch that hatch successfully). The number and timing of the visits (relative to nest progress) recorded on each NRC determines which of the above variables can be calculated, so the sample sizes for the analyses differ between variables. The sample sizes used are shown in Table 2. Note that these sample sizes were too small to allow us to estimate reliable separate annual values for the breeding performance parameters for all species except chaffinch and linnet, but that the use of periods of consistent CBC trend avoids this problem.

Table 2.  Numbers of nests used in the analyses of each nest record variable
Sample sizes for analyses of
Brood sizeChick : egg ratioEgg period failure rateNestling period failure rateSpecies
Clutch size
Bullfinch445520234773639
Chaffinch1386227399336413383
Corn bunting255333175436425
Goldfinch267466164602547
Greenfinch926116065515051250
Linnet21022390138435403172
Reed bunting63885240710131007
Skylark758122047814771489
Stock dove1023147074314411389
Tree sparrow1121132682716471609
Turtle dove177276116343297
Yellowhammer1028163767721121856

Analyses of variation with respect to year-block

All the variables taken from NRC data were analysed using generalized linear models in the genmod procedure of SAS (SAS Institute Inc. 1996), with year-block (time period) as a categorical independent variable. Daily nest failure rates were estimated using a formulation of Mayfield’s (1961, 1975) method as a logit-linear model with a binomial error term, in which success or failure over a given number of days (as a binary variable) was modelled, with the number of days over which the nest was exposed during the egg and nestling periods as the binomial denominator (Crawley 1993; Etheridge, Summers & Green 1997; Aebischer 1999). Numbers of exposure days during the egg and nestling periods were calculated as the mid-points between the maxima and minima possible given the timing of nest visits recorded on each NRC (note that exposure days refer only to the time span for which data were recorded for each nest and do not represent the full length of the egg or nestling periods). Chick : egg ratio was also modelled with a logit-link and binomial errors, brood size forming the numerator and clutch size the binomial denominator. Individually, clutch and brood sizes were modelled with identity links and normal errors. The significance of the variation between year-blocks was tested by comparing the fit of each model incorporating the year-blocks with that of an intercept-only (constant) model using a likelihood-ratio test (SAS Institute Inc. 1996). Correlation coefficients were then calculated between the block-specific breeding performance parameter estimates and population trend gradients (Table 1) to assess whether the variation in breeding performance was consistent with a role in driving population trends.

Simulations run by Green (1999) have shown that a fall in a demographic rate that is density-dependent could induce a population decline but be transient, quickly being reversed as density falls. This means that the temporal resolution of the demographic analysis would have to be high or that statistical controls for the effects of density would be needed to detect the change in survival in such a situation. Although we did not control for density, the turning point/year-block approach was sufficiently flexible to detect short periods of decline or increase when the change in trend direction was large enough to be important, so our method is likely to detect changes in breeding performance under this scenario.

The NRS does not follow a formal sampling protocol. The submission of cards in any year depends entirely on where recorders are active and on which species they concentrate, so there could be geographical biases in the data. We investigated whether such biases influenced our analyses by testing the significance of the influence of latitude and longitude on breeding performance parameters. Latitude and longitude values were taken from the grid references on the NRCs, where they were available. For older NRCs, the computerized data on nest record location tended to be restricted to a code for the county from which the record came; we therefore assigned a latitude and longitude specific to the centre of each county to all the cards from that county. To test for effects of location, each variable was modelled as a linear function of each of latitude and longitude, and also as a function of both simultaneously, with an interaction term. The link functions and error distributions used were as described above for the temporal analyses, and significance was again tested using likelihood-ratio tests. Where significant effects of latitude, longitude and/or the interaction between them were found, the year-block models for the appropriate species and variables were re-run, now incorporating the relevant location effect(s). The significance of the year-block effect in each case was then re-assessed using a likelihood-ratio test (LRT) between a model incorporating block-specific variation and the location effect and one allowing only the location effect.

Analyses of nest records will also be subject to bias if the seasonal distribution of the data does not reflect that of a species’ nesting attempts and if breeding performance varies according to the timing of nesting within the breeding season. Studies of several of the species considered here have shown that breeding performance often improves as the breeding season progresses (Newton 1964, 1972; Snow & Mayer-Gross 1967; Murton 1968; Jönsson 1992), although in yellowhammers, at least, this relationship is not simple, with fewer but higher quality young being produced in early season nesting attempts (Kyrkos 1997). We also know that nest records tend to under-sample late season nests (probably second and subsequent broods in multi-brooded species; Crick & Baillie 1996). Biases due to this under-sampling will be constant if relationships between breeding performance and laying date do not change with time, and will not therefore affect conclusions based on changes in breeding performance. We investigated whether such biases existed and controlled for their possible effects within the range of laying dates available in the data. Statistical comparisons between year-blocks of the distribution of first egg dates through the season showed little sign of variation in these distributions for any species. We nevertheless repeated the analyses of breeding performance with respect to year-block, controlling for a quadratic relationship between first egg date and breeding performance, wherever preliminary analyses indicated that the latter relationship was significant. These analyses should show where changes in breeding performance between year-blocks have been due to changes in the timing of breeding in the nest record sample (such changes could be real or could reflect sampling bias). Where an interaction between a quadratic trend in first egg date and year-block was significantly related to a breeding performance parameter, we ran further check analyses controlling for this effect. These analyses showed where apparent differences in breeding performance between trend periods might be due to differences in the dependence of breeding performance on the timing of breeding.

Estimating productivity

The critical measure of the success of a nesting attempt is the overall production of fledged offspring. Variations in clutch size or a daily failure rate, for example, could only be important at the population level if they affected the number of fledglings produced. To confirm whether differences in individual nest record variables are likely to have had net effects on breeding performance, we combined the block-specific estimates of clutch size, chick : egg ratio and daily nest failure rates to estimate the number of fledglings produced per breeding attempt (after Hensler 1985) as follows:

FPAt =  CS  ×  CER  ×  (1  -  EFR)EP ×  (1  -  NFR)NP(eqn 1)

where FPAt is the number of fledglings produced per breeding attempt, CS is clutch size, CER is chick : egg ratio, EFR and NFR are the egg and nestling period daily nest failure rates, respectively, and EP and NP are the lengths of the egg and nestling periods in days. EP and NP were taken to be the mid-points of the ranges given in Cramp (1985, 1988) and Cramp & Perrins (1994a,b) for populations in or near the British Isles (Table 3). Correlations with block-specific population trend gradients (Table 1) showed whether the variations in FPAt were consistent with their having had causal influences on population trends. Period-specific FPAt values could be compared using pairwise z-tests (Hensler & Nichols 1981; Hensler 1985), but to avoid the problems with multiple testing that such a strategy would engender, we estimated 95% confidence intervals for each estimate instead. Confidence intervals for the estimates of fledgling production were calculated using an extrapolation of the formula provided by Hensler (1985) for an expression incorporating four (rather than three) components. The variances required for this formula were taken from the genmod output for each variable. To obtain estimates of the variances of the parameters modelled using logit-links that were on the same scale of measurement as the parameters themselves when the latter were back-transformed, the relevant models were re-run using identity link functions (D.L. Thomson & S.N. Freeman, personal communication).

Table 3.  Estimates of the length of the egg (EP) and nestling (NP) periods used to generate estimates of productivity (drawn from Cramp 1985, 1988; Cramp & Perrins 1994a,b). Lengths of egg periods comprise geographically relevant values for the incubation period plus 1–3 days (the modal clutch size minus one) to account for egg laying
SpeciesEgg period (days)Nestling period (days)
Bullfinch1615
Chaffinch1614
Corn bunting1610
Goldfinch1515
Greenfinch1615
Linnet1512
Reed bunting1611
Skylark139
Stock dove1825
Tree sparrow1518
Turtle dove1520
Yellowhammer1511

Nest record data provide no information on either the number of nesting attempts made by pairs during the breeding season or post-fledging mortality, so we could not estimate annual productivity directly from FPAt. We could, however, obtain estimates of annual productivity per individual from abundance and survival rate data, assuming that emigration and immigration have no net effect on breeding abundance at the national scale, using the following formula for abundance:

Nt+1 = (Nt ×  Pt ×  StFY) + (Nt ×  StAD)(eqn 2)

where Pt is annual productivity per individual in year t, Nt is abundance in year t, Nt+1 that in the following year, and StFY and StAD are the annual survival rates of first-years and adults, respectively, between year t and year t+  1. Therefore:

Pt  =  ((Nt+1/Nt)  -  StAD)/StFY(eqn 3)

We used this approach to estimate annual productivity for each year-block and, subsequently, the importance of variation in post-fledging survival and/or the number of breeding attempts made. Abundance was taken as the unsmoothed Mountford CBC index in each year, and the survival rates used were derived from recoveries of ringed dead birds under the BTO’s ringing scheme, as follows. Age- and annually time-specific survival rates taken from Siriwardena, Baillie & Wilson (1999) were used for bullfinch, chaffinch, greenfinch, goldfinch and linnet. Ring-recovery sample sizes did not allow different survival rates to be estimated for each year for the other species, so age- and block-specific estimates were calculated (where possible) following the methods used in Siriwardena, Baillie & Wilson (1998) (see the Appendix). Annual estimates of Pt were calculated by combining these survival rates with the annual CBC indices according to the above formula and then averaged by year-block. The ring-recovery data were too sparse to allow survival rates incorporating any temporal variation to be estimated for corn bunting, skylark and stock, dove so we were unable to estimate block-specific Pt for these species.

The first-year survival rates used here were calculated using birds ringed as fledged independent juveniles, but excluding those ringed in the nest. The estimates therefore excluded both nestling and post-fledging (pre-independence) mortality. As a result, Pt refers to the productivity per individual of fledged independent juveniles, taking account of both post-fledging survival rates and multiple breeding attempts. To assess the (combined) temporal variation in these parameters, we compared the year-block-specific variation in Pt with that in the fledglings produced per breeding attempt, FPAt. To do this, we calculated the ratio Rt=  (2 ×  Pt)/FPAt for each block. Given monogamy and a 1 : 1 sex ratio, Rt will be equal to the product of the mean post-fledging survival rate and the mean number of breeding attempts undertaken (but will also incorporate any sampling error in the measured demographic parameters). The assumptions of a 1 : 1 sex ratio and monogamy are unlikely to be tenable for any of the species considered (except perhaps the doves), so the value of Rt will incorporate an amount of bias relating to deviations from these assumptions. The variability of Rt will not, however, have been affected by this bias unless sex ratios and/or mating systems have changed over time. Thus, we examined the variability of Rt and Pt between year-blocks, and also the correlations between each of Rt and Pt and the block-specific population trend gradients (Table 1) to find where components of Rt appeared to explain the population trends observed.

Results

Analyses of variation in individual breeding performance parameters by year-block

The results of the tests of the variation in individual breeding performance variables with respect to time period are summarized in Tables 4–15. In the absence of controls for first egg date, at least one variable gave rise to a significant or near-significant result for every species except stock dove and goldfinch, and there were several instances where correlations with population trend gradient showed that the variation in breeding performance could have contributed to the major population changes seen. Egg period failure rates were higher during the principal decline of the linnet (1975–86). They were also higher in the recent period of stability in chaffinch abundance, compared with the preceding increase, but high failure rates during the rapid increase between 1962 and 1964 meant that the overall correlation with block-specific population trend gradients was positive. The recovery in linnet abundance from 1987 onwards were associated with a higher chick : egg ratio, leading to higher brood sizes despite a fall in clutch size. In addition, declining population trends for tree sparrow, reed bunting and greenfinch were associated with lower chick : egg ratios, suggesting possible roles in driving population trends. Note, however, that a converse relationship between breeding performance and population trend was indicated by the nestling period daily failure rate results for reed bunting and by three other components of breeding performance for tree sparrow (Tables 10 and 13).

Table 4.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for bullfinch. The breeding performance variable names are abbreviated as follows: egg period failure rate, EFR; nestling period failure rate, NFR; chick : egg ratio, CER; clutch size, CS; brood size, BS. The means estimated by the genmod procedure of SAS (SAS Institute Inc. 1996) are presented for each variable and time period, taken from models omitting controls for latitude and longitude. Likelihood-ratio test (LRT) results refer to comparisons of models allowing variation between time periods and models allowing no temporal variation [d.f. =  (number of trend periods) – 1]. Superscripts by variable names indicate where the models from which the LRT statistics were derived incorporated latitude and/or longitude effects, as follows: (1) latitude only; (2) longitude only; (3) latitude and longitude; (4) latitude, longitude and the interaction between them. Asterisks (*) in the LRT column show where a control for a quadratic trend in first egg date (FED) was conducted: an additional superscript indicates that the test result was different after the control and the amended result is shown below the table. Daggers (†) and superscripts show the same for a control for the interaction between a quadratic trend in FED and year-block. Correlations (weighted by year-block duration) with block-specific population trend gradients (see Table 1) show whether the significant variations in each breeding performance parameter are consistent with their having contributed to the observed population trends. Positive correlations are evidence of such consistency for CER, CS and BS, whereas negative correlations represent similar evidence for EFR and NFR: the coefficients meeting these criteria are shown in bold
Time period (trend direction)
Variable62–65 (+)66–74 (+)75–78 (–)79–82 (–)83–95 (=)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=8·72, P=0·07; correlation with trend slope =–0·54.

EFR0·0210·0330·0280·0180·01513·23, 0·01*0·08
NFR10·0280·0270·0190·0410·0204·32, 0·37†a
CER40·930·910·950·890·9513·97, 0·01–0·08
CS4·684·564·634·824·725·21, 0·27*
BS24·154·064·104·194·223·38, 0·50*
Table 5.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for chaffinch. See Table 4 for further explanation
Time period (trend direction)
Variable62–64 (+)65–87 (+)88–95 (=)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=10·57, P<0·01; correlation with trend slope =0·73.

  2. b After control for FED2 , χ2=5·23, P=0·07; correlation with trend slope =0·55.

EFR40·0320·0250·0316·91, 0·03*0·25
NFR0·0270·0260·0240·36, 0·83
CER40·880·860·853·63, 0·16†a
CS44·374·354·301·4, 0·50*b
BS43·853·653·703·7, 0·16
Table 6.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for corn bunting. See Table 4 for further explanation
Time period (trend direction)
Variable62–73 (+)74–95 (–)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=2·10, P=0·15.

EFR40·0430·0263·81, 0·051·0
NFR40·0420·0158·94, <0·01†a1·0
CER40·830·9311·36, <0·01*–1·0
CS43·954·295·14, 0·02–1·0
BS33·163·9419·29, <0·01*–1·0
Table 7.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for goldfinch. See Table 4 for further explanation
Time period (trend direction)
Variable62–66 (–)67–70 (–)71–74 (+)75–86 (–)87–95 (+)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=11·03, P=0·03; correlation with trend slope =0·96.

EFR0·0200·0130·0240·0180·0296·1, 0·19†a
NFR40·0360·0140·0180·0200·0117·5, 0·11*
CER40·920·900·930·910·851·65, 0·80
CS4·884·844·794·834·741·67, 0·80
BS34·294·234·294·334·183·45, 0·48
Table 8.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for greenfinch. See Table 4 for further explanation
Time period (trend direction)
Variable62–66 (+)67–74 (=)75–77 (–)78–84 (–)85–95 (+)LRT
χ2, P
Correlation with trend slope
EFR0·0220·0200·0200·0150·0232·92, 0·57
NFR0·0160·0210·0170·0190·0172·22, 0·70*
CER0·880·880·840·850·9011·8, 0·02†0·39
CS4·794·804·764·844·790·57, 0·97
BS4·264·133·964·094·156·26, 0·18
Table 9.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for linnet. See Table 4 for further explanation
Time period (trend direction)
Variable62–66 (+)67–69 (–)70–74 (+)75–78 (–)79–86 (–)87–95 (+)LRT
χ2, P
Correlation with trend slope
EFR0·0240·0150·0140·0230·0260·02121·67, 0·01†0·25
NFR0·0140·0120·0110·0170·0170·0175·93, 0·31
CER40·860·860·900·890·880·9350·77, <0·01*0·31
CS24·784·744·714·724·814·6619·99, <0·01†–0·48
BS44·084·014·134·154·184·3518·69, <0·01*0·24
Table 10.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for reed bunting. See Table 4 for further explanation
Time period (trend direction)
Variable62–64 (=)65–75 (+)76–78 (–)79–95 (=)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=13·39, P<0·01; correlation with trend slope =–0·30.

EFR10·0070·0110·0110·0164·39, 0·22†a
NFR40·0070·0250·0190·0246·29, 0·09*†0·30
CER40·920·940·890·8914·43, 0·020·80
CS44·584·444·404·534·04, 0·26*
BS4·144·023·954·164·89, 0·18*
Table 11.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for skylark. See Table 4 for further explanation
Time period (trend direction)
Variable62–66 (+)67–77 (=)78–82 (–)83–95 (–)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=10·23, P=0·02; correlation with trend slope =–0·48.

EFR10·0490·0350·0320·0343·81, 0·28
NFR40·0300·0410·0320·0330·12, 0·73
CER40·890·920·920·945·92, 0·12†a
CS43·383·333·573·6222·98, <0·01*–0·73
BS33·003·063·403·3334·8, <0·01–0·92
Table 12.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for stock dove. See Table 4 for further explanation
Time period (trend direction)
Variable62–85 (+)86–95 (=)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 ×year-block, χ2=5·73, P=0·02; correlation with trend slope =–1·0.

EFR0·0130·0120·13, 0·72
NFR0·0100·0081·18, 0·28
CER0·930·952·54, 0·11
CS2·042·020·76, 0·38*†a
BS1·861·860·1, 0·76
Table 13.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for tree sparrow. See Table 4 for further explanation
Time period (trend direction)
Variable62–76 (+)77–95 (–)LRT
χ2, P
Correlation with trend slope
EFR30·0110·00418·83, <0·01*1·0
NFR0·0130·0111·25, 0·62
CER40·840·818·28, <0·01*†1·0
CS34·855·2427·25, <0·01*–1·0
BS43·954·246·62, 0·01*–1·0
Table 14.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for turtle dove. See Table 4 for further explanation
Time period (trend direction)
Variable62–78 (=)79–95 (–)LRT
χ2, P
Correlation with trend slope
  1. a After control for FED2 , χ2=2·34, P=0·13.

  2. b After control for FED2 ×year-block, χ2=3·40, P=0·07; correlation with trend slope =–1·0.

EFR10·0320·0280·04, 0·84
NFR0·0210·0102·91, 0·09*a1·0
CER0·940·970·66, 0·42
CS2·001·980·3, 0·58†b
BS1·891·890·02, 0·90*
Table 15.  Results of tests of variation of individual breeding performance parameters with respect to time periods with consistent trend directions for yellowhammer. See Table 4 for further explanation
Time period (trend direction)
Variable62–65 (+)66–77 (=)78–81 (+)82–87 (=)88–91 (–)92–95 (–)LRT
χ2, P
Correlation with trend slope
EFR40·0560·0470·0240·0250·0250·02652·58, <0·01†0·62
NFR40·0660·0380·0310·0280·0430·02023·13, <0·010·57
CER0·870·880·850·910·950·9214·49, <0·01–0·82
CS33·483·423·533·513·593·5912·37, 0·03*–0·51
BS13·093·043·083·233·333·3021·9, <0·01*†–0·68

Significant results where breeding performance had improved when populations were in decline were at least as common as patterns indicating that changes in breeding performance could have contributed to changes in abundance. Such patterns were found in at least one variable for each of bullfinch, corn bunting, linnet, reed bunting, skylark, tree sparrow, turtle dove and yellowhammer (but were countered by other relationships for linnet and reed bunting; Tables 4, 6, 9–11, 13–15).

Across the 12 species, we found significant dependence of a component of breeding performance on a quadratic trend in first egg date in 24 cases, and therefore conducted further control analyses (shown by asterisks in Tables 4–15). In only two cases did the supplementary analyses suggest that the conclusions of our original analyses might be affected by changes in the timing of the annual nest record sample, and both involved marginal levels of significance. The test of the variation in chaffinch clutch size between year-blocks was more significant after the control term was added (suggesting that a change in clutch size might have contributed to the population trend), and the test for the turtle dove nestling period failure rate was less significant. There were 15 cases in which a component of breeding performance was significantly related to the interaction between the square of first egg date and year-block (shown by daggers in Tables 4–15). Adding a term controlling for this interaction failed to change the results of the tests against year-block in seven cases, made seven statistically significant at or near the 5% level when they had not been so previously, and made only one previously significant result non-significant (Tables 4–15). The latter involved the nestling period failure rate for corn bunting. The results where year-block effects had not previously been identified as significant showed that, although there were no strong effects of year-block on the variables concerned and the effects of the control term were more important, year-block did affect the residual variation from the effect of the control. These patterns included effects that could have contributed to population change (i.e. indicating positive associations between breeding performance and population trend gradient) for bullfinch (nestling period failure rate), chaffinch (chick : egg ratio) and reed bunting (egg period failure rate) and four additional effects (for goldfinch, skylark, stock dove and turtle dove) that were not consistent with roles driving population trends.

Variation in numbers of fledglings per breeding attempt

The combinations of breeding performance parameters into estimates of observed number of fledglings produced per breeding attempt, FPAt, are shown in Fig. 1. Fledgling number was not closely related to population trend direction for most species. Linnet FPAt was clearly low during the principal period of population decline (1975–86), although a similar drop did not occur during the short 1967–69 decline (Fig. 1f). Chaffinch FPAt may also have been slightly lower during the recent period of stability that followed the species’ principal increase, but the difference was marginal (Fig. 1b). The recent stable period for bullfinch and the 1962–66 decline in goldfinch abundance represent the only other instances where there was variation in FPAt consistent with a causal link to abundance. For bullfinch, FPAt was higher from 1983 to 1995 than it had been during the decline of 1975–82 (Fig. 1a), and the 1962–66 goldfinch decline may have been associated with low FPAt (although this effect was again marginal; Fig. 1d).

Figure 1.

Figure 1.

    Block-specific estimates of number of fledglings per breeding attempt (FPAt), derived from nest records. In each graph, periods with declining population trends are indicated by black bars, stable periods by grey bars, and increasing periods by white bars. The error bars show the upper half of 95% confidence intervals (which are symmetrical) calculated according to the method presented by Hensler (1985), further details of which are given in the text. FPAt was correlated with the block-specific population trend gradients (given in Table 1) as follows: (a) bullfinch, − 0·06; (b) chaffinch, − 0·07; (c) corn bunting, − 1·0; (d) goldfinch, − 0·13; (e) greenfinch, − 0·04; (f) linnet 0·33; (g) reed bunting 0·11; (h) skylark, − 0·85; (i) stock dove, − 1·0; (j) tree sparrow, − 1·0; (k) turtle dove, − 1·0; (l) yellowhammer, − 0·65. Fig. 1 continued.

    Figure 1.

    Figure 1.

      Block-specific estimates of number of fledglings per breeding attempt (FPAt), derived from nest records. In each graph, periods with declining population trends are indicated by black bars, stable periods by grey bars, and increasing periods by white bars. The error bars show the upper half of 95% confidence intervals (which are symmetrical) calculated according to the method presented by Hensler (1985), further details of which are given in the text. FPAt was correlated with the block-specific population trend gradients (given in Table 1) as follows: (a) bullfinch, − 0·06; (b) chaffinch, − 0·07; (c) corn bunting, − 1·0; (d) goldfinch, − 0·13; (e) greenfinch, − 0·04; (f) linnet 0·33; (g) reed bunting 0·11; (h) skylark, − 0·85; (i) stock dove, − 1·0; (j) tree sparrow, − 1·0; (k) turtle dove, − 1·0; (l) yellowhammer, − 0·65. Fig. 1 continued.

      There was little or no variation in FPAt between periods for greenfinch and stock dove (Fig. 1e,i), despite the variation found in some components of breeding performance. Reed bunting FPAt was highest during the 1962–64 stable period, but was constant (at a lower level) thereafter while abundance rose, fell and then became stable again. The values of FPAt for corn bunting, skylark, tree sparrow, turtle dove and yellowhammer varied markedly, but all tended to be higher when abundance was falling than when trends were stable or increasing (. 1c,h,j,k,l).

      The patterns apparent in Fig. 1 were supported by the correlations between FPAt and the block-specific population trend gradients (Fig. 1). Positive correlations were only found for linnet and reed bunting (and the latter was very low), highly negative correlations being more common. (None of the correlations was statistically significant, but they were intended simply as measures of association: any correlative tests with such small sample sizes have low statistical power such that significant results are unlikely a priori.)

      Table 16 summarizes the variation in and associations with population trend gradient of annual productivity (Pt) and the derived estimate of the product of the post-fledging survival rate and the number of breeding attempts made (Rt) for each species. Mean values (across year-blocks), coefficients of variation and correlations with population trend gradient are shown. There was considerable variation between blocks in each parameter for most of the species considered, but with some notable exceptions. For most species, relatively large coefficients of variation for Rt showed that the variation in annual productivity, Pt, was not well correlated with the variation in FPAt and suggested that variation in components of Pt that were not also components of FPAt had been important. Although only two species showed positive correlations between FPAt and population trend gradient (Fig. 1), there were six positive correlations between Pt and trend slope and eight between Rt and trend slope (Table 16). In each case, positive correlations with Pt and/or Rt showed that temporal variation in one or more of the demographic or error components of Rt (which included the number of breeding attempts and post-fledging survival rates) was needed to explain the observed changes in abundance. Positive correlations with Rt alone suggested that the variation in the variables that comprised Rt had been in the right direction for the variation to have contributed to changes in abundance, but that any effects had been small in magnitude. The coefficient of variation of Rt was lowest for tree sparrow and turtle dove (Table 16), indicating that the variation in Pt was closely related to that in FPAt. As FPAt was relatively high during the declines of both species, it is unlikely that declines have occurred in the numbers of breeding attempts made or in post-fledging survival.

      Table 16.  Variation between year-blocks in estimated annual productivity and the product of post-fledging survival rates and numbers of breeding attempts undertaken. Block-specific estimates are shown of the annual productivity of fledged independent offspring per individual (Pt, estimated from survival and abundance data) and of an estimate of the product of the post-fledging survival rate and the number of breeding attempts made [Rt=(2×Pt)/FPAt, where FPAt is the number of fledglings produced per breeding attempt as estimated from nest record data]. Means (weighted by year-block duration) and coefficients of variation (CV) across year-blocks are shown for each of Pt and Rt, as are correlation coefficients (also weighted by year-block duration) showing their associations with block-specific population trend gradients (given in Table 1 along with the year-block definitions). The coefficients of variation give some indication of the variability of each parameter, but should be treated with caution because they will incorporate a substantial amount of sampling error. Similarly, the correlations with trend gradients are intended only as an aid to interpretation: the slopes, Pt and Rt are all derived from the same (CBC) data and are thus not independent. A high correlation between slope and Rt indicates that there has been variation in abundance that is not explained by our data on survival and breeding performance per nesting attempt: the source of this variation could be unmeasured demographic parameters or sampling error. The numbers of breeding attempts (excluding repeats after failure) typically made by each species (taken from Cramp 1985, 1988; Cramp & Perrins 1994a,b) are shown to aid interpretation further. Note, however, that Rt divided by a number of breeding attempts will only be an unbiased estimate of the magnitude of post-fledging survival rates if the species concerned is monogamous and has a 1 : 1 sex ratio: this assumption is unlikely to hold for any of the species considered. Pt and Rt could not be estimated for corn bunting, skylark and stock dove because a paucity of ringing data prevented the estimation of survival rates
      Species Block-specific estimates of Pt and RtMean (CV) of Pt and RtCorrelation with trend slopeNumber of breeding attempts
      BullfinchPt3·44, 2·50, 1·28, 1·61, 2·272·28 (75·0)0·932–3
      Rt3·40, 3·11, 1·22, 1·90, 1·752·26 (99·1)0·81 
      ChaffinchPt1·03, 0·83, 0·840·85 (26·5)0·02Usually 1
      Rt1·30, 0·97, 1·091·02 (40·0)–0·41 
      Corn buntingPt1–2 (more often 2)
      Rt 
      GoldfinchPt1·79, 1·65, 1·85, 2·27, 3·112·30 (68·0)0·522–3
      Rt1·86, 1·14, 1·56, 1·86, 2·872·01 (82·5)0·56 
      GreenfinchPt1·78, 1·24, 1·07, 1·44, 1·981·59 (61·1)0·54Usually 2
      Rt1·54, 1·12, 0·96, 1·19, 1·711·37 (59·3)0·57 
      LinnetPt2·56, 1·55, 1·99, 1·27, 1·69, 2·492·01 (59·6)0·922–3
      Rt2·11, 1·11, 1·33, 1·07, 1·45, 1·941·58 (61·4)0·79 
      Reed buntingPt0·86, 0·97, 0·60, 1·100·99 (49·2)–0·301–2
      Rt0·50, 0·73, 0·46, 0·930·78 (71·7)–0·30 
      SkylarkPt2–4
      Rt 
      Stock dovePt2–4
      Rt 
      Tree sparrowPt1·24, 1·391·32 (32·1)–1·03
      Rt0·91, 0·860·88 (15·4)1·0 
      Turtle dovePt1·70, 2·302·00 (88·6)–1·02–3 (more often 2)
      Rt4·46, 4·494·47 (1·2)–1·0 
      YellowhammerPt1·17, 0·83, 0·71, 0·89, 1·24, 0·900·92 (47·1)–0·272–3 (more often 2)
      Rt3·88, 1·74, 0·97, 1·12, 1·73, 1·021·70 (131·2)0·29 

      Discussion

      General patterns in the components of breeding performance

      The results of this study show that variations with respect to changes in abundance have occurred in components of the breeding performance of each of the species that were considered. Several of these patterns could reflect demographic mechanisms through which changes in the agricultural environment have influenced populations. Many other statistically significant relationships showed improved breeding performance during periods of decline. It is clear, in general, that population declines have not been associated, for example, with reduced nest survival or falling clutch sizes. Moreover, our supplementary analyses suggest that biases caused by an under-sampling of late season nests by the NRS or by changes in the relationship between breeding performance and the timing of nesting have not seriously affected these results. While it is possible that some of the ‘significant’ results detected might have arisen by chance due to the large number of tests conducted, the occurrence of any such apparent relationships would not be biased with respect to their direction. Widespread simple effects of the environment on numbers of fledglings produced per breeding attempt across granivorous farmland bird species cannot therefore be implicated as mechanisms behind population changes observed at the national scale. Further, controlling for changes in the relationship between breeding performance and the timing of nesting revealed more significant effects of year-block, most of which (five of seven) indicated additional inverse relationships between breeding performance and population trend. In general, these control analyses suggest that environmental factors may have influenced both breeding performance and the overall timing of breeding simultaneously.

      Apparent increases in breeding performance during population declines

      For bullfinch, corn bunting, skylark, tree sparrow, turtle dove and yellowhammer, breeding performance tended to be lower during periods with more positive population trends (generally early in the CBC time series) than during decline periods (generally late in the time series). These results show that variation in the components of breeding performance that were investigated cannot have been the mechanism that has driven the declines in these species, but they also raise a question as to what process underlies the pattern found.

      One possible explanation of the increases in breeding performance is that density-dependent restrictions were imposed by competition when densities were high, and that these were relaxed when declines in abundance began to occur and densities fell. Other responses that can broadly be termed density-dependent could also explain these patterns. When populations decline, it is likely that birds of lower quality and those in marginal habitats will disappear first. Breeding performance is likely to be affected by bird and habitat quality. Therefore, if poorer quality birds and those from lower quality habitat had contributed to the nest record sample before the decline began, average productivity would increase during the decline. Such an increase would not require any improvement in the performance of the birds still contributing to the nest record sample, and could still occur even if their performance had declined. This scenario is most likely to have occurred where declines have not been caused by changes in breeding performance (Green 1999). Tests for a mechanism of this kind for corn bunting have led to the rejection of the hypothesis (Crick 1997), but it is possible that habitat quality within farmland is patchy at a scale too small to be revealed by regional or county-based comparisons or as trends with respect to latitude and longitude, and in too subtle a manner to be detectable in the habitat information in nest records, at least before the introduction of systematic habitat coding in 1990 (Crick 1992). Broadly density-dependent processes are likely to operate in the regulation of the populations of most small bird species found in temperate regions (Greenwood & Baillie 1991; Holyoak & Baillie 1996a,b). The operation of density-dependence is, however, difficult to demonstrate conclusively (Sinclair 1989) and there are other plausible explanations for the patterns we have found in nest record data.

      Seed-eater breeding performance may have been depressed in the 1960s by the effects of the widespread use of organochlorine seed dressings, and so might subsequently have increased as the use of these pesticides was discontinued progressively from 1962 onwards (Newton 1979). It has been suggested that the decline in organochlorine seed-dressing use explains the historical trends in skylark (Chamberlain & Crick 1999) and corn bunting (Crick 1997) breeding performance. Such pesticides are known to have had serious effects on the productivity of raptors through the thinning of egg shells (Newton 1979), although only laboratory evidence is available regarding the sensitivity of passerines to such effects (Jefferies 1973; Crick 1997). The population declines of raptors are better understood (Newton 1979), but there are indications that the declines among farmland seed-eaters were more severe in regions where organochlorines were used more (Newton 1972; Parslow 1973). Our results show that skylark and yellowhammer breeding performance remained low into the 1970s (Tables 11 and 15 and Fig. 1), when large-scale organochlorine use had diminished, thereby potentially contradicting the organochlorine hypothesis. However, it is known that organochlorine residues can persist for years or even decades in temperate soils (Edwards 1973), so they are likely to continue to affect birds for some time after direct applications have ceased. Further, the temporal resolution of the analyses for corn bunting, tree sparrow and turtle dove was insufficiently fine to allow inferences regarding the timing of changes in pesticide use to be made firmly. It would therefore be unwise to draw final conclusions without detailed information on the historical spatial patterns of pesticide application that could be related to breeding performance. It is notable in this context that skylark, yellowhammer, corn bunting, tree sparrow and turtle dove are probably the five species in this study that are most likely to feed on unripe grain or invertebrates away from the edges of cereal fields. They would therefore be the species most likely to be affected by the use of pesticides on such fields.

      Finally, it is also possible that the principal effect of agricultural change on productivity has been to increase the parental effort required to fledge a brood of a given size under deteriorating environmental conditions. If this has occurred, then the main influence on overall breeding productivity may have been to reduce the number of breeding attempts that birds are able to make in one season (Kyrkos 1997; Wilson et al. 1997; see below), effectively altering a species’ life-history strategy. Behavioural studies of birds undertaking strategies involving different numbers of breeding attempts in a season would be needed to investigate this possibility.

      Implications of variation in numbers of fledglings per breeding attempt

      Our estimates of the numbers of fledglings produced per breeding attempt, FPAt (Fig. 1), show where the net effect of variation in the individual variables tested is likely to have been large enough to have had an impact on abundance. For example, the results for greenfinch reveal that the observed variation in chick : egg ratio, which was consistent with a causal influence on the declining phases in the population trend (perhaps reflecting higher mortality early in chick life), had little effect on FPAt when all the variables were combined (Fig. 1e). The clear change in the egg period failure rate of linnet nests after 1975 (Table 9) was detectable in the total fledglings produced (Fig. 1f), however, suggesting that this deterioration in breeding performance could have had an important role in driving the species’ concurrent decline in abundance. The recent increase in egg period failure rate for chaffinch (Table 5) may have contributed to the recent stabilization of the CBC trend, but the magnitude of its effect suggests that it had a lower biological significance than the linnet result (Fig. 1b). Overall, the range of negative relationships between breeding performance and trend gradient discussed above were also generally clearly detectable in the estimates of FPAt and were much more common than patterns suggesting that changes in FPAt had caused trends in abundance (Fig. 1).

      The mechanisms underlying the population declines seen in skylark, bullfinch, yellowhammer, corn bunting, tree sparrow and turtle dove remain unknown. Decreases in annual survival have been proposed as candidate mechanisms for some of these species (Crick 1997; Donald 1997; Chamberlain & Crick 1999), but insufficient ring-recovery data are available to investigate past changes in survival in Britain directly for species such as corn bunting and skylark. Recovery data for yellowhammer and tree sparrow are also sparse, and this may explain the lack of statistical significance in the reductions in survival over time found by Siriwardena, Baillie & Wilson (1998). However, the best estimates of the variation in adult and first-year yellowhammer survival (from ring-recoveries) suggest that it has been sufficient to explain the species’ decline (Kyrkos 1997), and our comparisons of fledgling production per breeding attempt, FPAt, and annual productivity, Pt, suggest that this is also true for tree sparrow and turtle dove (Table 16). An alternative explanation has been put forward for the decline of the skylark. Cropping patterns on intensive farms mean that pairs are unlikely to be able to complete enough nesting attempts during the season to sustain their populations (Wilson et al. 1997).

      Post-fledging survival and the number of breeding attempts undertaken

      We could not measure directly post-fledging survival rates (here, survival over the period of days between fledging and the mean age at which first summer birds were ringed, i.e. the period between those covered by the nest record and ringing schemes) or the number of breeding attempts undertaken by pairs in a season. However, the ratios Rt derived from FPAt and Pt (Table 16) gave some indication as to their influence. Rt could be calculated for nine species and was positively correlated with block-specific population trend gradients for eight of them: bullfinch, chaffinch, goldfinch, greenfinch, linnet, reed bunting, tree sparrow and yellowhammer (Table 16). In addition, the coefficients of variation of Rt across year-blocks were high for all but two species (tree sparrow and turtle dove; Table 16). These results indicate that causes of variation in abundance additional to those represented by our estimates of breeding performance and survival rates are required to explain the population trends of most species. Together with sources of error, post-fledging survival and numbers of breeding attempts are key components of Rt and could therefore have been important wherever changes in Rt are needed to explain population changes. More specific conclusions cannot be reached from our results, however, because the unknown variables are confounded. In addition, by our method, any changes in the number of birds recorded by the CBC but which then fail to nest at all will also be incorporated in the variation in Pt and therefore Rt. Note, however, that the CBC only records territorial birds, so the failure of censused birds to breed is unlikely to be a major source of variation in Pt. The sources of error that will also have contributed to the variation in Rt include sampling and measurement errors in the estimates of abundance, survival and each component of breeding performance, as well as deviations from assumptions such as the constancy of the lengths of the egg and nestling periods and of demographic parameters within year-blocks. Errors are likely to have increased the apparent differences between FPAt and Pt, leading to over-estimation of the variation in Rt and therefore of the apparent importance of variation in post-fledging survival and the number of breeding attempts made. However, the difficulties inherent in measuring each of these parameters empirically mean, especially for scarcer species, that better estimates of their historical importance are unlikely to be forthcoming.

      There are few estimates of post-fledging survival rates for any species in the literature, but those available suggest that it can have a large impact on overall productivity. Studies using ring-recovery data for song thrush Turdus philomelos and radio-tracking for wood thrush Hylocichla mustelina show that survival rates over the first 2 months after fledging are approximately 38% and 42%, respectively (Anders et al. 1997; Thomson, Baillie & Peach 1999), while mark–resighting or mark–recapture studies of yellow-eyed junco Junco phaeonotus, great tit Parus major and zebra finch Taeniopygia guttata estimated survival over the first 30–40 days after fledging to be 49%, 43% and 28%, respectively (Dhondt 1979; Sullivan 1989; Zann & Runciman 1994). Survival over the post-fledging period for starlings Sturnus vulgaris varies between 39% and 62% according to fledging date and weight (Krementz, Nichols & Hines 1989). The number of breeding attempts made by individual pairs is also likely to be an important variable in the demography of many passerines (Kyrkos 1997; Wilson et al. 1997). Both of these parameters remain difficult to measure, but must be a priority for future studies of the demography of population changes.

      Linnet breeding performance and abundance

      The graph of block-specific FPAt for linnet shows a clear decline after 1975 (Fig. 1f); the same decline was apparent when we examined the variation in annual productivity Pt, and both variables were positively correlated with block-specific population trend gradients. This suggests that variation in the components of annual productivity that are not accounted for in nest record data did not compensate for the reduction in nest survival during the egg period that we detected (Table 9). The latter could have been caused by direct effects such as increased nest predation or indirect ones such as disturbance or nutritional stress to parents leading to desertion. Further, increased predation rates may ultimately reflect, say, habitat changes diminishing nest concealment or changes in parental behaviour induced by food shortages. Variations in linnet breeding performance after the end of the major decline cast some light on the important demographic mechanisms, however, as the egg period failure rate fell in this period (Table 9) while egg predator populations continued to increase (Gregory & Marchant 1996; Siriwardena et al. 1998). Changes in linnet abundance on individual CBC plots have also been found not to be significantly correlated with the presence of magpies (Thomson et al. 1998). Together with an increase in brood size and chick : egg ratio, the improvement in egg period nest survival has led to a slight increase in breeding performance since 1986 (Table 9 and Fig. 1f), but greater numbers of breeding attempts or post-fledging survival may also have contributed to the population increase (Table 16). Increases in the crop area of oilseed rape Brassica napus oleifera could have improved linnet breeding success, compensating for the herbicide-mediated decline in many farmland weeds (such as charlock Sinapis arvensis) that traditionally formed the bulk of the linnet’s summer diet (Moorcroft, Bradbury & Wilson 1997). Both the number of breeding attempts possible in a season and post-fledging survival could have increased in response to this improvement in food supplies, as could chick : egg ratio, through decreased mortality of very young chicks.

      The importance of variation in survival

      While the results suggest that one or more components of annual productivity may have played important roles in driving several species’ population trends, other evidence suggests that another demographic rate, survival, has been at least as important in many cases. Survival rates have varied in a manner consistent with their having had some causal influence on the abundance of goldfinch and chaffinch (Siriwardena, Baillie & Wilson 1999). For goldfinch, much of the CBC trend can be explained by variation in survival alone, but the comparatively low breeding performance between 1962 and 1966 found here (Fig. 1d) may explain a part of the CBC trend that survival did not explain (Siriwardena, Baillie & Wilson 1999). In addition, Siriwardena, Baillie & Wilson (1999) concluded that a small increase in goldfinch productivity must have occurred in recent years, in addition to the observed changes in survival, to explain the CBC trend in full. Comparison of the Pt and FPAt values for the period in question suggests that this increase is likely to have been due to improved post-fledging survival or an increase in the number of breeding attempts made. A parallel change in breeding success and survival appears to have occurred for chaffinch: as the species’ population trend ceased to increase in the late 1980s, both the number of fledglings produced per breeding attempt (Fig. 1b) and annual survival rates (Siriwardena, Baillie & Wilson 1999) may have fallen. In general, current evidence suggests that there have been changes in annual survival coincident with population changes for many farmland passerines, and that failures to detect such patterns may often only be due to the constraints of sample size (Siriwardena, Baillie & Wilson 1998).

      The evidence that annual survival has been the demographic rate most often related to population changes suggests that the key times of year that environmental changes have affected farmland bird populations have been those during which most mortality occurs. Such periods are likely to occur outside the breeding season and, at least for resident species, in harsh late winter conditions when metabolic demands are high and food supplies have been depleted. It is important to consider, however, that passerines (at least) are rather short-lived (adult annual survival rates typically below 60% and, for some species, below 40%: Siriwardena, Baillie & Wilson 1998). The likelihood that future reproductive success will compensate for current failure is therefore low. There may be a strong selection pressure for adult birds to compensate behaviourally for deteriorating habitat quality, such that breeding success is maintained. Effectively, breeding birds may simply work harder, absorbing the effects of a deteriorating environment in the breeding season, but perhaps paying for this through reduced numbers of breeding attempts or subsequent survival rates. Therefore, even where changes in over-winter survival have driven population changes, the ultimate environmental cause could still be conditions in the breeding season. Only detailed, long-term field studies of individual life-history variation in populations in differing agricultural environments will show whether such complex interactions between environmental change and demography are occurring.

      Recommendations for research and management

      Future research should focus on investigating the importance of variation in post-fledging mortality and the number of breeding attempts made as demographic parameters, both through field studies and, where possible, historical data sets. Efforts should also be made to improve the monitoring of these little-known parameters for the future (but note that constant-effort ringing schemes such as the one already operating in Britain can be used to generate estimates of annual productivity in which these parameters are integrated; Peach, Buckland & Baillie 1996). Thomson, Baillie & Peach (1999) were able to estimate post-fledging survival rates using ringing data for the mid-1970s onwards for one common species, but such modelling is unlikely to be feasible for many scarcer birds.

      Specific experimental manipulations of the main proximate influences on breeding success, such as food availability and predation risk, are likely to represent the only avenues available to approach a complete understanding of how environmental change has affected breeding performance and, subsequently, abundance. Our results show where changes in breeding performance might have affected abundance, and therefore where such more intensive studies might be focused.

      For the majority of declining granivorous farmland birds, measures to improve breeding performance per nesting attempt are unlikely to reverse the immediate demographic causes of the declines. In general, the results of the present study and of our complementary work on the variations in annual survival rates (Siriwardena, Baillie & Wilson 1998, 1999) suggest that measures to improve over-winter survival may be more likely to succeed in reversing the immediate causes of decline for many species. However, ameliorative management during the breeding season (e.g. reducing pesticide inputs to increase invertebrate availability or promoting floral diversity through reductions in herbicide use and the maintenance of uncropped margins) may increase survival rates by reducing the stresses on breeding adults or contribute to population recoveries by boosting productivity. Our studies of survival and breeding performance also suggest that the causes and mechanisms behind population change are likely to have been highly species-specific, even within the ecologically and/or phylogenetically similar group of species investigated here. As a result, the most effective environmental changes for the reversal of the decline of any given species are also likely to be species-specific. We can suggest, however, that breeding season management to increase the possible number of breeding attempts during the season (such as by providing diverse vegetation structures to produce suitable nesting habitat for skylarks throughout the spring and summer; Wilson et al. 1997) or to increase post-fledging survival rates should be given much greater consideration as a means of reversing population declines.

      Acknowledgements

      We would like to thank all the volunteers who have submitted nest record cards, conducted CBC surveys and ringed and recovered birds over the years. We are also grateful to BTO staff who have processed nest records data, especially Caroline Dudley, David Glue, Peter Beaven and Dawn Balmer. Rhys Green gave valuable advice on the analysis of nest failure rates, David Thomson helped with implementing our analyses in SAS and Stephen Freeman gave useful statistical advice. Mario Díaz Esteban and Rhys Green provided helpful comments on previous drafts. The project under which this work was conducted is funded by the UK Ministry of Agriculture, Fisheries and Food as contract BD0906. The Nest Record Scheme, the Ringing Scheme and the Common Birds Census are funded by a partnership of the British Trust for Ornithology and the Joint Nature Conservation Committee (on behalf of English Nature, Scottish Natural Heritage and the Countryside Council for Wales, and also on behalf of the Environment and Heritage Service in Northern Ireland). The Common Birds Census has also been supported by a contract from the Department for the Environment, Transport and the Regions.

      Received 13 February 1999; revision received 26 October 1999

      Appendix

      Appendix: note on the calculation of survival rates

      As stated in the text, estimated annual survival rates for bullfinch, chaffinch, goldfinch, greenfinch and linnet were taken from Siriwardena, Baillie & Wilson (1999). For the other species for which enough recovery data were available, we used block-specific estimates where a different survival rate was estimated for each period of consistent trend in a species’ CBC index series (Siriwardena et al. 1998), following the methods used by Siriwardena, Baillie & Wilson (1998). The survival rates used here, however, differ from those presented in Siriwardena, Baillie & Wilson (1998): in the latter, one survival rate was estimated for each trend direction (averaging across all declining blocks, etc.) rather than for each trend period (year-block). Thus, for the current analysis, a separate survival estimate was calculated for every time period defined in Table 1 within each species’ CBC time series; in contrast, a single survival rate was calculated across all periods with the same trend direction in Siriwardena, Baillie & Wilson (1998).

      Ancillary