Table 2. Quantile regression results summary for the roll-snap performance distribution of golden-collared and white-collared manakins, where the display length is evaluated as a predictor of display speed at different quantiles (τ) of the distribution.

A significantly negative slope (p<0.05; values adjusted to control the false discovery rate) suggests that there is a trade-off between speed and endurance at a given quantile. Here we present representative models from every 10th quantile (τ interval = 0.1), but we also characterized the performance distribution at a finer-grained scale (see Figure 3—figure supplement 1) at every whole quantile (τ interval = 0.01) between 0.1 and 0.9. We did not extend the analysis below the 10th or above the 90th quantile because quantile regression performs poorly at extreme portions of the distribution for smaller datasets (n < 1000).

Quantile (τ)	Golden-collared manakin				White-collared manakin
	Slope	s.e.	t-value	p-value	Slope	s.e.	t-value	p-value
0.1	−0.13	0.11	−1.18	.239	−0.02	0.58	−0.03	.976
0.2	−0.20	0.07	−2.95	.004*	0.03	0.51	0.06	.976
0.3	−0.31	0.07	−4.13	.0001*	−0.04	0.31	−0.14	.976
0.4	−0.44	0.08	−5.27	<0.0001*	−0.29	0.24	−1.24	.305
0.5	−0.53	0.08	−6.51	<0.0001*	−0.20	0.23	−0.85	.512
0.6	−0.44	0.07	−6.39	<0.0001*	−0.19	0.15	−1.27	.305
0.7	−0.58	0.03	−21.69	<0.0001*	−0.11	0.16	−0.67	.606
0.8	−0.60	0.08	−7.91	<0.0001*	−0.31	0.13	−2.33	.028*
0.9	−0.56	0.14	−4.03	.0001*	−0.41	0.15	−2.76	.020*