focal species and study area
The study occurred in 25 deciduous riparian patches in Montana, which is the primary breeding habitat for both species in this region, along approximately 150 km of the Madison and Upper Missouri Rivers, May–July, 2003–05. Riparian patches were dominated by narrowleaf cottonwood Populus angustifolia (James), and a variety of willow species (e.g. Salix amygdaloides Anderss., S. exigua Nuttall). Other tree and shrub species included water birch Betula occidentalis (Hook.), mountain alder Alnus incana (Linnaeus), rose Rosa spp., and snowberry Symphoricarpos albus (L.) (Blake).
From 2003 to 2005, I used a song playback experiment to test the avoidance and attraction hypotheses for using social cues in habitat selection. Manipulating songs of each species to change social cues within plots is a powerful and appropriate method because both species sing often, breed in forests with substantial subcanopy structure (thus limiting availability of visual cues), use vocalizations in competitive interactions, and use song in mate attraction (Briskie 1994; Martin et al. 1996; Sherry & Holmes 1997).
The experimental design consisted of comparing habitat use in pre-treatment years to treatment years over two paired year combinations, 2003–04 and 2004–05. This design allows for the control and testing the effect of pre-treatment densities within sites and the control of natural annual variation in population density. The design focuses on the use of social information when settling into areas after spring migration. While some experimental evidence suggests birds use information gained in the previous breeding season for settling (e.g. Doligez, Danchin & Clobert 2002), most evidence to date on the use of social cues suggests that cues are often used during the spring settlement period (Alatalo, Lundberg & Bjorkland 1982; Mönkkönen et al. 1990; Thomson et al. 2003; Ward & Schlossberg 2004; Hahn & Silverman 2006). I considered individual point count locations (50-m fixed radius; 0·79 ha) in different riparian patches (plots hereafter) as the sampling unit (see also Nocera et al. 2006). Plot size was large enough to accommodate multiple males within the plot boundaries, based on the known small territory sizes of both species (typically 0·07–0·18 ha for flycatchers, 0·1–0·66 ha for redstarts; Briskie 1994; Sherry & Holmes 1997; Hahn & Silverman 2006). In 2003, six plots were chosen to span a gradient of subcanopy structure, as part of a larger study on riparian bird productivity. All six plots were used in 2003–04. To select plots used for the 2004–05 paired combination, I identified plots with suitable vegetation to reduce any confounding of vegetation quality, based on the occurrence of each species from a larger random sample of 75 riparian patches (166 plots) covering over 500 km of the Missouri and Madison Rivers, 36 (72 plots) of which fell within the study area used here. Based on suitable vegetation characteristics, 19 plots within the study area were randomly selected. All plots were > 500 m apart.
The experiment consisted of two treatments – redstart playback plots (n = 7) and flycatcher playback plots (n = 7) – and control plots (n = 11), which were similar to playback areas except no bird songs were manipulated. Control plots estimated natural annual variation in population densities. As in all other playback experiments testing for the use of social cues in habitat selection (Ward & Schlossberg 2004; Hahn & Silverman 2006; Nocera et al. 2006), control playbacks, such as songs from another species unlikely to be competitors with the focal species, were not used because of logistical constraints. However, another behavioural playback experiment on these two species showed no general effect of playbacks (Martin et al. 1996).
Treatments were randomly assigned, but flycatcher treatments were stratified by pre-treatment year flycatcher abundance (mean number of detections/point) for unoccupied (n = 3), low (1–2; n = 2) and high (> 2 detections/point; n = 2) abundance plots to ensure that treatments were applied across the natural range of densities and to assess if density influences the use of social cues. I chose this approach because for migratory birds, density effects on habitat selection can be assessed by estimating settlement as a function of density in the previous year (Newton 1998; Hahn & Silverman 2006), when birds show consistent patterns of habitat use (and/or site fidelity) among years. For least flycatchers, densities in one year were positively correlated with densities in the following year in control plots (F1,9 = 7·64, P = 0·022, r = 0·68; when excluding one outlier, P = 0·002, r = 0·85), suggesting that densities in the previous year may be a useful indicator of density in the current year (see also Perry & Andersen 2003). Redstart treatments were not stratified by pre-treatment redstart abundance because redstarts occurred only in 33% of the plots at the start of the study (i.e. two of the six plots in 2003).
Each treatment consisted of two playback stations, 100–120 m apart, with each station pointed toward the centre of the study plot (where point counts were conducted; see below). This cue addition increased ‘apparent’ male density in each plot by 2·56 males ha−1; natural density of males in occupied control and pre-treatment plots ranged from 0·66 to 3·4 males ha−1 for flycatchers and 0·64–2·56 males ha−1 for redstarts. Playbacks commenced on 5 May in 2004 and 1 May in 2005 and continued daily until 7–10 July. I continued playbacks throughout the breeding season because males often sing throughout this period and some migratory songbirds may move within the breeding season after nest failure (Briskie 1994; Sherry & Holmes 1997; Fletcher, Koford & Seaman 2006). Although social cues, such as song, may provide different types of information early vs. late in the breeding season, density estimates from early (1 June−15 June) and later in the breeding season (16 June−10 July) were not different in treatments plots (paired t-test: t = −1·05, P = 0·32). Each playback station consisted of a portable stereo mounted 1–2 m up a tree or shrub, wired to a timer and a deep-cycle marine battery. Stereos broadcasted songs at full volume (approximately 90 dB) from a CD from 04.00 to 10.00 h each day, mimicking the primary vocalization period for each species (see also Ward & Schlossberg 2004). Each CD contained 60 min of song tracks and 5 min of silent tracks in a random order. Timers played stereos for 65 min, shutdown the system for 5 min, and then repeated broadcasting the CD. For both species, local dialects recorded in Montana were broadcasted.
To estimate bird densities during the breeding season, my assistants and I surveyed each plot once during two periods in each year: 1–15 June and 16 June−10 July. I used 10-min, 50-m fixed-radius point counts for surveying birds (cf. Nocera et al. 2006), with one point count location centred in each plot (50–60 m from each stereo). Both species are thought to be highly conspicuous (Sherry & Holmes 1997; Perry & Andersen 2003), such that this sampling effort should be adequate for understanding habitat use. Surveys were conducted between sunrise and 5 h later, when birds are most active. Surveys were not conducted during high wind velocities (≥ 25 km h−1) or during precipitation. Before surveys, observers turned off portable stereos. During surveys, observers recorded all birds seen or heard, including how individuals were detected (by song, visual, or call), sex of individuals, time interval of detection (four equal intervals: 0–2·5, 2·5–5 min, etc.), and distances of birds from the centre point. To ensure accurate delineation of birds within or outside plots, distances to birds were estimated using a laser range finder. Note that these density estimates do not provide information on individual movements, which would require marking individuals, or explicit data on breeding status. Instead, densities only explicitly provide information on how habitat use is influenced by the addition of social cues.
While estimating densities provides crucial information on habitat selection behaviours, arrival dates provide another valuable measure of habitat preference (Newton 1998). In 2005, I estimated arrival dates on a subset of plots (four flycatcher, four redstart, six controls) by conducting point counts at each plot every 1–3 days (mean = 2·8 days; n = 137 surveys) between 3 and 31 May. At least one plot of each treatment was surveyed daily during this period. A species was considered to have arrived in a plot if it was detected on at least two consecutive visits.
Finally, at each point-count station I measured vegetation after the bird surveys to determine if there was any bias in treatment applications such that differences in vegetation structure could explain patterns of habitat use. To do so, I estimated metrics related to canopy, shrub and ground cover (see Appendix 1). Overall, there was no evidence for differences in vegetation structure among plot types (Table S1; Wilks’λ = 0·56, F16,30 = 0·62, P = 0·84).
Based on point counts during the breeding season surveys, I estimated bird density (birds per hectare) by correcting raw counts with estimated detection probabilities, which were estimated using a removal model in Program mark (see Appendix 2). I was unable to estimate detection probabilities for American redstarts due to their rarity in the study area (see below). For redstarts, I focus on changes in occurrence among treatments. Results for redstarts should be treated cautiously due to their rarity, but I provide these results for comparison among other relevant investigations (Sherry & Holmes 1988; Hahn & Silverman 2006).
To test for the attraction and avoidance hypotheses during the breeding season, I used a two-step approach. I first developed models to test hypotheses for using social cues while controlling for pre-treatment population densities and natural annual variation in population density. I tested for the effects of treatments on flycatcher density during the treatment year using ancova, with flycatcher density during the pre-treatment year as a covariate, which controls for effects of pre-treatment density in responses to treatments. For redstarts, I focused on changes in redstart occurrence (−1, local extinction; 0, no change; 1, local colonization) using ordinal logistic regression (i.e. cumulative logit analysis; Allison 1999). I used this approach in lieu of using logistic regression for treatment year occurrence data, because that approach is less sensitive to detecting avoidance and would require more parameters for estimation. The hypotheses for using social cues in habitat selection were analysed using three a priori contrasts within the ancova and ordinal logistic regression: (1) conspecific responses (conspecific – control); (2) heterospecific responses (heterospecific – control); and (3) differences in cue use (conspecific – heterospecific). Subtracting responses in controls in the first two contrasts accounts for natural annual variation in population densities.
My second step included testing for the effects of population density on the use of social cues by flycatchers by comparing models that reflect different hypotheses for density effects to a density-independent model. If population density influences the use of social cues in this experiment, density in the treatment year should vary based on pre-treatment densities for treatments relative to controls (a significant treatment × density interaction), whereas for density-independent responses, the treatment year densities should be independent of pre-treatment densities. I compared a treatment-only model (density-independent) to a model that incorporated pre-treatment density and its interaction with treatment as covariates (linear density-dependent), and a model that included pre-treatment density2 and its interaction with treatment as covariates (nonlinear density-dependent model; see Forsman et al. 2002; Fletcher 2006) using Akaike's Information Criterion (adjusted for sample size; AICc) and AICc model weights (Burnham & Anderson 1998). A model selection approach was used here because it allows for simultaneously comparing different models that reflect biological hypotheses (Johnson & Omland 2004).
Patterns of arrival were analysed using a proportional hazards model (Allison 1995), with treatment as a fixed effect and pre-treatment year density (birds per hectare) as a covariate. I included pre-treatment density as a covariate because for some species older animals may arrive earlier to the breeding grounds and are more likely to show site fidelity (Switzer 1993; Lemon et al. 1996). A proportional hazards model is appropriate for time-to-event data, such as arrival dates, and it allows for the censoring of data (Allison 1995), which is particularly useful for arrival data in situations where no animals may arrive to some sites during the sampling period. Arrival date was considered the midpoint between the last unoccupied and first occupied visit.