Statistical Analysis

Exploratory visits (i.e., visits without nose poke or drinking) were aggregated for each day of the experiment by subjects. Generalized linear models (GLM) with log link using negative binomial distribution were constructed. Differences in initial exploratory activity were also compared between groups. To show how exploratory activity changed during the day, visits were aggregated for each 4-h period. For nose poke learning, a cosinor analysis was conducted to see how daytime exploratory activity patterns differed between groups (Cornelissen, 2014). Two main estimates were considered: mesor (mean activity) and amplitude (difference of peak and midline activity). Mean values for groups were compared using the t-test. The daytime mean activity was calculated for the first 72 h, groups were compared using bootstrapped Watson’s test.

Drinking volume was estimated by assessing the number of licks and lick duration. For acclimation, the number of licks for each animal was calculated for 4-h periods which also revealed how drinking behavior changed over the course of the day. When competing, animals were restricted both in their access to water bottles, and maximum lick duration for each visit. Therefore, the mean lick duration was calculated for each day of the experiment. In addition, linear mixed-effects models were used to compare groups with subjects as random factors (Bolker et al., 2009). Acclimation was treated separately from the other phases because of the difference in underlying data distribution due to the time limit introduced after acclimation.

In, side preference learning, the proportion of correct nose pokes to all nose pokes was calculated. This response variable was put in a binomial generalized mixed-effects model (GLMM) with treatment as the fixed effect. Binomial distribution was used because the number of correct responses out of all trials was measured. The proportion of correct nose-pokes is expected to change during time due to the initial period and the reversals. Therefore, the drink session was included as a random factor within the model. As dispersion was high, cumulative lick duration was included as a random factor. Differences in drinking volume changed the proportion of visits to the correct corner that was not related to learning. This addition to the model dropped the dispersion to near 1, while the mean of all random factors was close to 0 (−0.02). Type II Wald chi-square test was used to assess the treatment effect. The hierarchy was estimated by differences in lick duration within groups. Total lick duration was calculated for each day of the experiment and evenness of the values was used as a community measurement. Evenness is most often used to describe the distribution of individuals within a community using Pielou’s evenness index ranging from 0 to 1. This measure was adapted to reflect evenness in drinking among individuals, because, the lower the evenness, the stronger the hierarchy that is expected in the community. Hierarchy could be best observed during competition; therefore, rank abundance curves were fitted to the cumulative lick number per hour values of each subject of the groups. Models were compared using Akaike information criterion (AIC) to find the shape of the best fitting model.

Reentering visits (“guarding”) were defined as visits after which the same subject entered the corner. The number of reentering visits was calculated for each subject and day of the experiment. The maximum divided by mean values for groups was calculated to express the distribution of reentering visits within the groups. Experimental phases were merged based on whether water access was unlimited (acclimation, nose-poke learning) or limited (side preference learning, competition) during the given phase. Population-level values were compared by using a linear model. Calculations were carried out by using R².

Statistical evaluation was performed by unpaired t-tests to analyze spontaneous locomotor activity (ambulation and rearing), von Frey test, ultrasonic vocalization, and juvenile social play (pinning) results in GraphPad Prism version 7.04 for Windows (GraphPad Software, La Jolla, CA, USA).