Comparison of three sets (1 = log step-length, 2 = use of terrain ruggedness, 3 = use of forest by female elk as response variables, respectively) of Generalized Linear Mixed Models.

The structure of the fixed component of the model was constant across models (see Table 1 footnotes) with the only exception of age (not included, included) and age interacted with human-activity proxies (time of the day, distance from road, hunting season, and time of the week). All models had a random slope for true age and a random intercept for individual elk, as well as a random intercept for year–i.e., the best random effect structure selected in Table 1 –and were fit with Maximum-Likelihood estimation. Models indicated by an asterisk accounted for more than 0.90 of the Akaike weights and were further inspected for model averaging (S3 Table).

各模型的固定效应部分结构保持一致（详见表1注释），仅年龄变量（部分模型未纳入、部分模型纳入）以及年龄与人类活动代理变量（每日时段、距道路距离、狩猎季、每周时段）的交互项存在例外。所有模型均设置了真实年龄的随机斜率、个体麋鹿的随机截距，以及年份的随机截距——即表1中筛选出的最优随机效应结构——并采用极大似然估计（Maximum-Likelihood estimation）进行拟合。带有星号标记的模型其赤池权重（Akaike weights）占比超过0.90，将进一步用于模型平均分析（详见附表3）。