Analyses. We fitted some of our foraging honey bees with copper tags as a control for any baseline disturbance caused by the insominator and for any environmental factors specific to the day of testing. We tested bees within the hive to maximize normalcy, increase the likelihood of bees dancing, and avoid confounders associated with testing isolated bees, which sleep less over consecutive days and die prematurely (16, 20, 22). Treatment effects were compared with results after a DD period to test for differences between sleep deprivation and general disturbance. The results are products of linear mixed-effects models, programmed in R (45), with bee population (treatment vs. control) and treatment day as fixed effects and bee identity as a random factor (i.e., observations were nested within a bee). Linear mixed-effects models were fitted using the lmer function in the lme4 package (46). The multcomp package was used to perform likelihood ratio tests to distinguish between competing models (47); the complicated (e.g., missing data, correlated covariates) and unbalanced nature of the data precluded the use of standard likelihood ratio tests (48). The resulting continuous, linear predictors are reported as mean ± SEM. To visualize the data (Figs. 3 and 4) and confirm the mixed-effects model results, we also calculated average values (one per bee) using JMP version 8 (SAS Institute). We set α = 0.05 for all tests. The sleep deprivation literature overwhelmingly documents decrements in precision and efficiency; because sleep deprivation was not expected to induce greater precision, we report one-tailed P values when testing our predictions relating to the impact of sleep deprivation on dances on the day after nocturnal perturbations. We report two-tailed P values under all other conditions.