Trends in Abundance and Phenology: Because list length analysis is a metric of relative abundance, average trends in abundance are expected to be centered around 0 across species; this expectation held (mean = 0.003 , sd = 0.045, greatest decline = -0.137 (Nymphalis vaualbum Compton tortoiseshell; Fig 1a), greatest increase = 0.190 (Poanes zabulon Zabulon skipper; Fig 1b)). Trends estimated from count models were also broadly centered around 0 (mean = 0.012, sd = 0.057, greatest decline = -0.159 (Nymphalis vaualbum Compton tortoiseshell), greatest increase = 0.213 (Poanes zabulon Zabulon skipper)). Trends estimated from list length were strongly congruent with trends estimated from counts (Pearson’s r = 0.89, P < 0.001). Therefore, we focus the rest of our analyses on trends estimated from list length analyses (see Appendix S2 ).
Across 26 years of continuous observation, species were, on average, advancing their onset of flight approximately 0.20 days per year (sd = 0.42, range = (advancing) -1.75 days per year for Battus philenor (Pipevine swallowtail; Fig 1c), to (delaying) +1.22 days per year for Vanessa cardui (Painted lady; Fig 1d)) with an estimated 75% of species advancing their 0.1 quantile observations. Species were, on average, delaying their 0.9 quantile day of year observation approximately 0.04 days per year (sd = 0.50, range = -1.76 days per year (Nymphalis vaualbum Compton Tortoiseshell), +1.29 days per year (Erynnis horatius Horace Duskywing)). Approximately 58% of species delayed their 0.9 quantile observations. Species were, on average, increasing their flight period (measured as the difference between the 0.9 quantile trend and 0.1 quantile trend) by 0.16 days per year (sd = 0.55 days per year, range = -1.96 days per year for Nymphalis vaualbum (Compton tortoiseshell), +2.05 days per year Battus philenor (Pipevine swallowtail). Approximately an estimated 63% of species increasing their observed flight period.
Trends in the onset of flight activity (0.1 quantile) were negatively correlated with changes in relative abundance, i.e., advancing the start of the flight period was associated with increasing abundance (r = -0.27, t = -2.7 , df = 87, p = 0.009, 98.03% of bootstrapped correlations < 0; Fig 2a). Trends in the end of flight activity (0.9 quantile) were positively correlated with changes in relative abundance, i.e., species that were extending their activity later into the year also tended to be increasing in abundance (r = 0.36, t = 3.59, df = 87, p < 0.001; 99.07% of bootstrapped correlations > 0; Fig 2b). The strongest association in our data was a positive correlation between the annual change in the flight period and the annual change in the relative abundance (r = 0.54, t = 5.9, df = 87, p < 0.001; 99.88% of bootstrapped correlations > 0; Fig 2c) such that species elongating their total flight time were increasing in relative abundance.
Mean flight dates were significantly changing in univoltine species (slope of mean day of year observation vs time ± SE: -0.27 ± 0.16) but not for multivoltine species (slope of mean day of year observation vs. time ± SE: -0.03 ± 0.26. Overall, there was no significant association between trend in mean date and trend in abundance for all species combined (r = 0.061, t = 0.574, df = 87, p = 0.568; 93.5% of bootstrapped correlations > 0). Grouping by voltinism and using abundance trends estimated using counts for direct comparison with MacGregor et al. (2019), mean day of year trends were not significantly associated with abundance trends for multivoltine species (r = -0.06, t = -0.43, df = 50, p = 0.672; 74.18% of bootstrapped correlations < 0 ), but were marginally significantly associated with abundance trends for univoltine species (r = 0.32, t = 2.00, df = 35, p = 0.053; 88.75% of bootstrapped correlations > 0).
Structural equation models: Our a priori structural equation model (Fig. 3) showed significant direct effects of flight period trends (scaled regression coefficient ± SE, β = 0.472 ± 0.087) and species’ range type (β = 0.308 ± 0.101) on trends in abundance, which presumably reflects other traits associated with northern vs. southern species. The effect of flight period trends on abundance trends was larger than the direct effects of range type. Differences in flight period were significantly associated with differences in voltinism (β = 0.430 ± 096). Compared to univoltine species, multivoltine species more often increased their flight period (days/year ± SE: -0.13 ± 0.09 and 0.35 ± 0.0 for univoltine and multivoltine, respectively) and increased in relative abundance (trend ± SE: -0.01 ± 0.01 and 0.02 ± 0.01 for univoltine and multivoltine, respectively). Differences in voltinism were weaker but significantly associated with range type (proportion multivoltine ± SE: 0.69 ± 0.06 and 0.32 ± .10 for southern and northern species, respectively). This a priori model demonstrated adequate, but not exceptional fit (χ2 = 3.076, df = 2, p = 0.215; recalling that, in SEMs p < 0.05 represents significant inconsistency with data).
Post hoc inspection of modification indices revealed one additional factor that was not present in our a priori model. This missing factor was a relationship between range type and flight period trends; species with southerly ranges were increasing flight periods more than species with northerly ranges (days/year ± SE: 0.35 ± 0.07 and -0.13 ± 0.09, for southern and northern species, respectively), regardless of voltinism. Including this relationship improved model fit (χ2 = 0.614, df = 1, p = 0.433), and slightly decreased the strength of the association between voltinism and flight period (updated β = 0.381 ± 0.099).