The tradition of ringing birds as a means of studying avian populations is more than a century old (Greenwood 2009). In the first decades of bird ringing, the method was applied almost exclusively to answer questions about avian migration; how far do birds fly and which routes do they take? Beginning in the late 1960s, ornithologists began to monitor breeding population sizes using standardised mist netting (Robinson et al. 2009). Later the focus shifted to the study of processes underlying population change, namely the determination of demographic parameters such as birth and death rates, leading to the foundation of the first nation-wide constant effort scheme in Great Britain and Ireland in 1986 (Peach et al. 1996). The Dutch constant effort site (CES) programme followed in 1994 and since then has been continuously operated by hundreds of enthusiastic citizen scientists under the direction of the Vogeltrekstation — Dutch Centre for Avian Migration and Demography (van der Jeugd et al. 2007; Schekkerman et al., unpubl.). Currently, the CES programme encompasses 40 active mist-netting sites (on average one CES per 1000 km² of Dutch territory). In total, 78 sites have participated in the programme. The database accumulated from 1994 to 2009 comprises more than 250,000 records and is used to evaluate abundance, survival and productivity of the captured species.

For the understanding of the dynamics of passerine populations it is important to evaluate their reproductive success and identify temporal trends of productivity change. The CES programme is well-suited to provide such data since standardised mist netting is regarded an important and valuable technique for assessing species demography (Dunn & Ralph 2004). Furthermore, methodological adequacy of mist netting has been thoroughly scrutinized (see various contributions in Ralph & Dunn 2004).

Here we ask whether the effort invested in the CES programme is appropriate to detect temporal trends of productivity of passerine bird populations across The Netherlands. In the context of mist netting, productivity refers to the proportion of juveniles of a species captured at a site in a given year. It integrates fecundity (number of eggs laid), losses in the nest and mortality during the immediate post-fledging period (Robinson et al. 2007). We pose the following questions: Is the number of CESs sufficient to obtain a reliable estimate of the temporal trend of juvenile production? Would less effort produce the same results or should the number of CESs be kept the same or be augmented? Would lower numbers of CESs increase the risk of detecting a (negative or positive) trend although productivity did not change over time (Type I error) or not detecting a trend although productivity increased or decreased (Type II error)?

METHODS

Statistical analysis

The expected proportion of juveniles of a given species in year i at site j, p_ij, was estimated by the generalized linear model (GLM) logit(p_ij) = β_iY_i + β_jS_j, where Y and S are the categorical explanatory variables year and site, the β_i and β_j are the regression coefficients (one for each factor level) and the logit link function ensures that estimated probabilities lie within the interval (0–1) (Robinson et al. 2007). The inverse logarithm of the regression coefficients of the years, β_i, yields an index of productivity of the species across all CESs. The probabilities of catching juvenile and adult birds differ among species. The productivity index can thus be used to characterise changes of productivity of a given species over time, but should not be used for a direct comparison between species (Ballard et al. 2004).

The presence of transient birds (i.e. migrants and occasional visitors from elsewhere) can potentially bias measures of productivity by inflating the number of adults, juveniles, or both (Schekkerman et al., unpubl.). The effect of transient birds can be reduced by applying species-specific ‘date limits’, whereby trapping days that fall within the species' main migration periods are excluded from analyses. However, the application of such date limits did not improve breeding productivity and abundance estimates derived from constant effort ringing in The Netherlands (Schekkerman et al., unpubl.) and, therefore, was not applied.

The resulting time series with productivity indices from 1994 to 2009 are too short to allow for standard analytical approaches, for example, time series decomposition (Cowpertwait & Metcalfe 2009). We regarded the time series as long enough for detecting linear trends, but too short for the detection of higher order relationships. We thus assess linear trends of the productivity index over time by least-square regression. First, a linear regression of the productivity index on year was calculated using the complete records for each species. Then, the database was subsampled to assess the effect of number of CESs on the estimation of annual indices and slope and significance of their regression on year. The sizes of the subsamples were dependent on the number of CESs in which a given species was recorded (n_sites, Table 1). The largest subsample size, n_max, was the largest number divisible by 5 with n_sites— n_max ≥3. Subsequent subsample sizes were established in decreasing steps of five. For example, the Sedge Warbler Acrocephalus schoenobaenus was recorded in 46 CESs, thus the number of CESs in the subsamples was determined as n = 40, 35, 30, …, 5. For each n, 1000 randomly chosen subsets of the 46 CESs were drawn without replacement, their reproduction indices determined and their regression lines calculated. Subsampling was terminated for a given species when model fitting problems occurred in more than 10% of the cases. For each n we determined the proportion of samples that yielded positive or negative slopes at error levels of P > 0.1, < 0.1, < 0.05, < 0.01 and < 0.001. Slopes with P < 0.05 were regarded as statistically significant. All statistical analysis was done with the R software environment for statistical computing (R Development Core Team 2009).

Table 1.

The twenty most abundant species in the Dutch CES programme. n_ind = number of recorded individuals from 1994 to 2009; n_sites = number of sites in which a species was recorded; n_min = minimum subsample size for which in ≥90% of the iterations a GLM model could be fitted to attain a productivity index (see text for details); b = slope of the linear model: productivity index = a + b × year using complete records; P = error probability. Significant slopes (P < 0.05) are in bold. Productivity indices cannot be directly compared among species. To facilitate comparison of the slopes, the annual indices for each species were divided by their mean, thus centring all index values around 1, before calculating the linear regression.

RESULTS

The analysis using complete records yielded statistically significant linear temporal trends of the productivity index for two species of the 20 species (Table 1); the Garden Warbler (increasing productivity, P < 0.01) and the Reed Warbler (increasing productivity, P < 0.05). The minimum size of CES subsamples, n_min, that allowed for model fitting in at least 90% of the 1000 resampling iterations varied considerably among species (Table 1) and ranged from five (Winter Wren, Blue Tit, Blackbird, Dunnock and Bearded Reedling) to 35 (Bluethroat) and 55 (Robin). For the latter two species, the relationship between sample size and proportion of significant results could therefore not be established.

For six species (Marsh Warbler, Sedge Warbler, Reed Bunting, Great Tit, Lesser Whitethroat, Bluethroat) the probability of committing a Type I error (determining a significant slope although productivity did not change over time when the complete records were used) was never larger than 10% (Fig. 1). For seven other species (Blue Tit, Chiffchaff, Blackcap, Dunnock, Whitethroat, Winter Wren, Blackbird) the proportion of iterations yielding a significant linear relationship between time and productivity reached 20% (Fig. 2) although no temporal trend was detectable when using the complete records. In the case of the Chiffchaff, for example, a CES programme based on only 25 sites would have produced an erroneous positive productivity trend with a probability of 23% and an erroneous negative trend with a probability of 21% (Fig. 2B). In all cases, the Type I error probability diminished with increasing subsample size to levels of less than 10%. Four species (Willow Warbler, Willow Tit, Bearded Reedling, Song Thrush) showed proportions of significant slopes larger than 30% (Figs 3A–D). Type I error probability tended to increase with sample size for these species.

Figure 1.

Proportion of significant linear relationships between time and productivity index of species with a low probability of a Type I error (determining a significant slope although productivity did not change over time). For each number of sites the complete database was resampled 1000 times. The positive and ‘negative’ y-axes refer to the proportions of positive and negative slopes, respectively.

For the Garden Warbler, the proportion of iterations showing a significant temporal trend of increasing productivity increased steadily with increasing sample size and reached 94% at a sample size of 55 CESs. The probability of committing a Type II error (not detecting an existing trend) was >20% from a sample size of <45 (Fig. 3E). For the Reed Warbler, the proportion of significant results never exceeded 55% (Fig. 3F), indicating a high probability of not detecting a significant positive trend of the species at lower sample sizes. The proportion of results indicating a trend at P < 0.1 increased continuously with increasing sample size and reached 0.91 at a sample size of 50 sites (Fig. 3F).

Figure 2.

Similar to Fig. 1 for species with a moderate probability of a Type I error.

DISCUSSION

For the majority of the abundant passerine species, the number of sites in the Dutch CES programme is sufficient to prevent the detection of non-existing trends. The probability of finding false significant trends becomes of concern when 30 or fewer sites are used to estimate the productivity index (Figs 1, 2). The number of sites is also sufficient to detect real trends at a significance level of P < 0.01 as has been shown for the Garden Warbler. However, a slightly lower number of sites would have considerably decreased the chance of detecting this trend (Fig. 3E).

There are indications that an increased number of CESs would allow for the detection of a greater number of significant temporal productivity trends. Although the analysis using all available data of the Willow Warbler did not yield a significant slope, resampling showed that the proportion of significant results increases distinctly with increasing number of sites included (Fig. 3C). If an even larger number of sites with Willow Warbler records were available, a significant productivity decrease would probably be detected. This would be consistent with the nation-wide decrease of this species as determined by the Dutch breeding bird monitoring programme (van Dijk et al. 2009). A similar observation can be made for the Willow Tit, which shows a steep increase of the proportion of significant slopes with increasing number of sites (Fig. 3D). In this case a larger number of sites would increase the probability of finding evidence for a productivity increase. Such a result would contradict the abundance trend of this species, which is characterised by a steady decline over the past 20 years (van Dijk et al. 2009).

The Reed Warbler is the most abundant species caught in the CES programme (Table 1). Nevertheless, a larger number of sites with records of this bird would be desirable, since the resampling results are not consistent with the observed significant positive productivity trend in the complete dataset. The proportion of slopes with P < 0.05 oscillates around 0.5 and exhibits no distinct relationship with the number of CESs included in the analysis while the proportion of slopes with P < 0.1 increases steeply (Fig. 3F). This raises the suspicion that the detected productivity increase based on the complete data might be due to a Type I error and that the “true” error probability of the time-productivity relationship lies between 0.1 and 0.05. The long-term abundance trend of the Reed Warbler shows a considerable increase, especially in the 1970s and early 1980s. In the past 20 years, however, abundance has oscillated without any distinct trend (van Dijk et al. 2009, van Turnhout et al. 2010).

Figure 3.

Similar to Fig. 1 for species with a considerable probability of a Type I error.

Some species are relatively abundant and widespread, yet a CES programme with numerous sites is necessary to evaluate their productivity. For example, no GLM could be fitted to the data of the Long-tailed Tit (ranked 21st in abundance, 1471 caught individuals at 43 sites, not included in this study) and thus its productivity index not be estimated. Similarly, no GLM could be fitted to the data of the Robin (ranked 13th in abundance, recorded from 59 sites; Table 1) when fewer than 55 sites were used. On the other hand, the Bearded Reedling was recorded from only 21 sites, but yielded robust modelling results. The reasons for these differences among species are not yet clear. For example, all of them display data characteristics which might impair modelling success: Robin and Bearded Reedling both show skewed distributions among sites (many sites with few records, few sites with many records) and a considerable proportion of site-year combinations in which only juveniles were captured and no adults (Robin 8%, Bearded Reedling 14%). For these species, a high but variable number of transient birds in the captures is a possible source of bias. For example, the Robin was the only species in which the application of date limits to reduce the transient effect led to a closer agreement between annual indices of abundance derived from CE ringing and from the Dutch breeding bird monitoring programme (Schekkerman et al., unpubl.). However, high numbers of transient birds are also likely in Bearded Reedling due to its high mobility, loose territorial boundaries, and a high degree of postbreeding dispersal (Cramp 1988; Cramp & Perrins 1993).

We conclude that the Dutch CES programme works with a sufficient number of sites in order to detect reliable temporal trends for most of the more abundant passerine species. Increasing the number of sites would probably guarantee a more secure determination of productivity trends for those species which currently show ambiguous results (Reed Warbler, Willow Warbler, Willow Tit). In addition, it would render the whole CES programme more immune against site closedown since CESs are operated by volunteers and consequently are prone to abandonment due to retirement or manpower shortage.