More passing vehicles lower stop-level BBS counts in 318 of 382 species (3.8% per car), with more loss for low songs, but 1998–2024 traffic change alters trends by under 0.1%/yr for 375

Question: Do vehicles that pass a North American Breeding Bird Survey (BBS) stop during its 3-minute count lower the bird count? Is the loss larger for low-pitched songs? Did the change in traffic from 1998 to 2024 bias species trends? Data: BBS 2025 release (doi:10.5066/P14SNUV4), 59,600 standard runs from 1998 to 2024 with vehicle counts at all 50 stops (4,692 routes), and 382 species. Method: for each species, a Poisson model of stop counts with stop and run fixed effects, which remove habitat and all effects common to one survey (observer, date, weather, year). The covariate log(1+cars) has two parts: the deviation from the stop's own linear traffic trend (primary) and the trend. A noise flag is also a covariate. We pooled with random effects and Holm tests. We wrote the plan before the species fits. Results: one car (compared with none) lowered counts by 3.8% (95% CI 3.4 to 4.1). The effect was negative for 318 species: 91 were Holm-significant below zero and none above zero. The loss increased with traffic: 2.0% at 1 car and 12.4% at 10 or more cars. Non-vehicle noise lowered counts by 17.5%. In 225 passerines, species with lower songs lost more: +0.031 per octave in the log effect (family-bootstrap CI 0.010 to 0.049; +0.025 after body mass, abundance and habitat density). Non-passerines lost 6.1% per car and passerines 2.7%. Traffic at stops increased by only 0.024 log units per decade. When we added the traffic effect to the trend models, species trends changed by a median of -0.004% per year (90% of species: -0.053 to +0.013). This change was smaller than the trend standard error for all 382 species. Limitations: the BBS records only vehicles on the route road during the count, not chronic noise from nearby roads. The primary estimate assumes that year-to-year traffic deviations at a stop are not related to bird behaviour there. The trend models are simple log-linear models, not the USGS hierarchical models.

Breeding Bird Survey · traffic noise · detectability · roadside survey bias · song frequency · acoustic masking · population trends · fixed-effects Poisson · point counts