Females preferred the Solver song in all stimulus sets
The third Bayesian model (Fig. 3 ) confirmed a general preference for Solver songs and showed that in each stimulus set, Solver song was preferred over Non-Solver song. Aggregate preference for Solver song was 0.608 (95% confidence intervals: 0.592, 0.625). Aggregate preference for Non-Solver song was 0.392 (95% confidence intervals: 0.374, 0.408). In the stimulus set that was furthest apart in preference, Solver song had an estimated preference of 0.682 (95% confidence intervals: 0.669, 0.695) while Non-Solver song had an estimated preference of 0.318 (95% confidence intervals: 0.305, 0.331). In the stimulus set that was closest in preference, Solver song had an estimated preference of 0.523 (95% confidence intervals: 0.551, 0.535), while Non-Solver song had an estimated preference of 0.477 (95% confidence intervals: 0.564, 0.489). Even in the closest stimulus set, 95% confidence intervals did not overlap (thus making it a 5% or less chance that the Non-Solver song was preferred over the Solver song). We can therefore say that Solver song was preferred in all stimulus pairs and that population preference for Solver song (see Fig. 2 ) was not driven by one or two particularly attractive Solver songs, but by a consistent pattern in preference across all six stimulus sets.
These patterns remained consistent when the 6thstimulus set was dropped (Solver preference by stimulus set: 0.597, 95% credible interval: 0.583, 0.613; Non-Solver preference by stimulus set: 0.403, 95% credible interval: 0.387, 0.417).