GEE model results. Predicted call rates from the GEE model as a function of caller depth (a), pitch (b), ODBA (c), roll (d), group composition change (e), surface movement (f), and group size (g). Solid lines show predictions from the best-fitting model, with shaded areas indicating 95% confidence intervals from a parametric bootstrap. To make these predictions, other predictors were fixed at the values most common in the data: group size 1, no group composition change, behavior Mill, depth 5, ODBA 0.1, roll 0, and pitch −29. All results shown are for calls detected on both accelerometers and acoustic record. (Note, we did not have enough data to make precise predictions for higher ODBA or extreme pitch and roll values, which were rarely or never observed in the dataset. Rug plots along the top of the tag data figure panels show locations where data points existed.).