Table 4a:

Candidate models that were most supported (∆AIC_C = 0.00) for predicting the age class (immature or adult) of 83 female brown bears. Each model was the outcome of one of four individual analyses^a. For the first three analyses, we used different age thresholds (≥3, ≥4 or ≥5 years) for adulthood while retaining all records in the data set. For the final analysis, we re-analysed the data set that was used for the most supported model (∆AIC_C = 0.00), but excluded the records for bears at the threshold age (in this case, 3 year olds)

Adult age (y)	N	Model	Most supported model structureb	K	AICC	∆AICC	wi	Accuracy (%)
≥3	83	Hormone + study area	(test × cort3) + test2 + cort + (test × sa)	7	80.10	0.00	0.91	77.1
≥5	83	Hormone + study area	(cort × sa) + (test2 × sa) + (cort3 × sa)	8	86.84	6.74	0.99	74.7
≥4	83	Hormone + study area	(test2 × sa) + est3	5	93.28	13.18	0.90	77.1
Model based on reduced data set excluding records for 3-year-old bears
≥3	77	Hormone only (with interactions)	test2 + prog3 + (prog × cort) + est + (prog × cort3)	9	84.97	0.00	0.49	79.2

^aStatistics are number of estimable parameters in model (K), sample-size-adjusted Akaike information criterion (AIC_C), difference in AIC_C between top model and model i (∆AIC_C), Akaike weight (w_i) and accuracy (%).

^bVariables are testosterone (test), progesterone (prog), estradiol (est), cortisol (cort), ordinal day (d) and study area (sa).