Using pre-specified inclusion and exclusion criteria we identified all publications reporting relevant experiments (see below) by searching three electronic databases (PubMed, ISI Web of Science, and EMBASE) using the search strategy “(anxiety OR arousal OR learned helplessness OR explor^* OR choice OR learn^* OR cognition OR preference OR motor OR pain OR maze) AND (‘social status' OR ‘social rank' OR ‘social dominance' OR ‘dominance hierarchy' OR ‘social hierarchy' OR submiss^*) AND (mouse OR mus OR mice),” with search results limited to title and abstracts. The cut-off date for our search was on September 20, 2019. Further details on the search strategy can be found in the supplement (Supplementary Text 1). This study is in accordance with PRISMA guidelines and the Systematic Review Center for Laboratory Animal Experimentation (SYRCLE) (Hooijmans et al., 2014); the checklist can be found in the Supplementary Information section.

One investigator (JAV) retrieved and reviewed all publications. First the titles and abstracts of 30 randomly selected papers from the 696 (~4.3%) were screened to develop the key exclusion criteria. From this screening process we set the following exclusion criteria: studies were excluded if they: (i) did not include mice; (ii) used mice housed singly; (iii) did not report measures of social interactions between cage-mates; (iv) were part of a symposium/conference proceedings or review; or (v) were written in a language other than English. In the next screening step only studies were included that housed mice in static groups/pairs for 2 weeks or more, reported methods for measuring social dominance behavior, and reported behavioral tests from the following domains: anxiety, arousal, learned helplessness, exploration, preference, learning/cognition, motor, pain, or social, outlined and described in, “Mouse Behavioral Testing: How to use mice in behavioral neuroscience” (Wahlsten, 2011), page 40. Studies or data were excluded if: (i) dominance relationships were measured between non-cage-mates; (ii) treatments were administered in addition to behavioral phenotyping (only control groups from these studies were used); or (iii) data from the same study were published more than once (no studies met this criterion). A detailed study protocol and flow diagram of the search and exclusion process can be found in the supplementary (Supplementary Text 1 and Supplementary Figure 1).

Assessment of risk of bias and study quality were conducted independently by two reviewers (JAV and AJ) using a modified SYRCLE Risk of Bias Tool (Hooijmans et al., 2014) with the inclusion of sample size calculation (see Supplementary Table 1 for more details) for each of the 20 studies that met the prespecified inclusion/exclusion criteria. Any disagreements were resolved by consensus—these were low (<6%).

After compiling a final list of the 20 included studies, two reviewers (JAV and AJ) independently extracted the sample sizes, means, and standard deviations for each dominant and subordinate comparison (i.e., experiment) made within each study. For example, if a study compared dominant and subordinate mice on exploration in the open field; the sample size, mean value of exploration, and standard deviation of exploration was extracted for dominant male mice and the respective values for subordinate male mice were also extracted. This was done for each metric of each experiment within a study (a total of 99 metrics, across 55 experiments, across 20 studies). When studies housed more than two mice per cage, only the most dominant and most subordinate rankings were considered—intermediate rank assigned mice were excluded. Data were either copied directly from tables, calculated from data provided by the respective corresponding author (Larrieu et al., 2017; Varholick et al., 2018, 2019) or extracted using “Web Plot Digitizer” (Rohatgi, 2018). If data were not reported because there were null effects, the unreported data with a null effect was noted but not included in the meta-analyses (specifics provided in results). Once all data were collected by the two reviewers (JAV and AJ), a mean value for each mean and each standard deviation extracted was calculated and rounded to the nearest hundredths place. These mean values calculated between the two reviewers were used in the meta-analyses and reported tables. Both reviewers agreed on all sample sizes across the experiments.

All meta-analyses were calculated using jamovi (Jamovi. jamovi, 2020) and the MAJOR module (Hamilton, 2018). Jamovi is a Graphical User Interface (GUI) version of R, and MAJOR is based on the commonly used R package, Metafor (Viechtbauer, 2010). Separate meta-analyses were run for behavioral tests that had been used by 5 or more studies—the open field, elevated-plus maze, and Porsolt forced swim test. Because various outcome measures were often reported across studies for the same behavioral test (e.g., time spent in open arms, total distance traveled, number of open arm entries), we used the most frequently reported measure across studies for the meta-analysis. If a study did not use the most frequently reported measure, then we used the second most frequently used measure, and so forth. For example, the most frequently used measure for the open field was total distance traveled, followed by number of crossings on a grid, then velocity (cm traveled per second). More specifics can be found in each respective results subsection and all data can be found in the supplement.

Due to the high degree of heterogeneity between studies, meta-analyses were run by fitting a random-effects model using the standardized mean difference between dominant and subordinate mice for each respective outcome measure for each study. The sample sizes for each respective dominance status group were used for calculating each standardized mean difference for the meta-analysis. Sample sizes were determined by the number of animals per each dominance status group, the number of dominants and the number of subordinates, separately (intermediate rank assigned mice were excluded). A restricted maximum-likelihood (REML) estimation was used for calculating the heterogeneity statistic Tau². No moderator was used. Behavioral tests (i.e., experiments) for which fewer than 5 studies were available, were categorized within their respective domain described in, “Mouse Behavioral Testing: How to use mice in behavioral neuroscience” (Wahlsten, 2011), page 40. Hedge's g was then calculated for the metric with the largest effect size for each experiment. General comparisons (e.g., smaller vs. larger effect sizes) were made between studies within their domain foregoing any further statistical testing.

By electronic search we identified 20 studies (i.e., published manuscripts) (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992; Ferrari et al., 1998; Vekovishcheva and Sukhotina, 2000; Bartolomucci et al., 2001, 2004; Palanza et al., 2001; Fitchett et al., 2005a, 2009; Sá-Rocha et al., 2006; Saldívar-González et al., 2007; Wang et al., 2011; Colas-Zelin et al., 2012; Horii et al., 2017; Larrieu et al., 2017; Zhou et al., 2017; Kunkel and Wang, 2018; Varholick et al., 2018, 2019; Pallé et al., 2019) divided into 55 separate experiments that met our pre-specified inclusion criteria (Methods and Supplementary Text 1). These studies varied concerning strain, supplier, group-size, and whether littermates were housed together (Table 1). Almost half of the studies used outbred mice of various strains (n = 9/20), one study used inbred Balb/cJ mice, while the remainder used sub-strains of the C57BL/6 mouse (n = 9/20). Group-sizes varied from 2 to 9 mice per cage and only two studies housed litter-mates together.

The risk of bias evaluation of the 20 included articles in this review is reported in Figure 1 and Supplementary Table 3. All 20 studies had a low risk of bias concerning three indices: baseline characteristics, incomplete outcome data, and random housing. A total of 19 studies had a low risk of bias for sequence generation, the remaining article had an unclear risk of bias because it did not explicitly describe the method for determining dominance. Another combination of 19 articles had low risk of bias from allocation concealment, meaning that the social dominance status of the animal was concealed during the dominance assessment method. Only one study reported a sample size calculation. The other indices had more varied distributions of low, unclear, or high risk of bias. For example, seven articles expressly stated random outcome assessment reflecting a low risk of bias, while the other 13 had unclear risk of bias. Notably, 16 and 11 of the studies had a low risk of bias from investigator blinding and outcome assessor blinding, respectively; many of the other articles had unclear risk of bias from blinding as they did not expressly describe the blinding process. An example of high risk of bias for both investigator blinding and outcome assessor blinding would be that the subordinate ranked animals always had bite-marks while dominant ranked animals had none. Importantly, seven articles had a high risk of bias from excluding cage-groups that had an unstable dominance organization or ranking, this was consistently found as “other sources of bias.” In summary, all but one study (Varholick et al., 2019) had at least one unclear (19 out of 20) and/or at least one high risk of bias (12 out of 20) (Supplementary Table 3).

The collected studies differed in their method of dominance assessment, the frequency of measuring dominance, age of animals at grouping, and the age of animals when assessing dominance—not all studies reported these metrics (Table 2). Several studies used multiple methods to determine social dominance status (Vekovishcheva and Sukhotina, 2000; Wang et al., 2011; Larrieu et al., 2017; Zhou et al., 2017).

The most common method used in the assessment of social dominance status was home-cage behavior observation and scoring for three or more consecutive days before behavioral testing (12 out of 20) (Ferrari et al., 1998; Vekovishcheva and Sukhotina, 2000; Bartolomucci et al., 2001, 2004; Palanza et al., 2001; Fitchett et al., 2005a, 2009; Sá-Rocha et al., 2006; Saldívar-González et al., 2007; Wang et al., 2011; Horii et al., 2017; Larrieu et al., 2017). Mice that engaged in more offensive behavior (e.g., attack, chase, mount, bite) compared to defensive behavior (e.g., flee, freeze, supine posture) were rated dominant, while those that showed more defensive behavior than offensive behavior were ranked subordinate. A total of four studies of these 12 also considered bite-wounds as a sign of dominance where the subordinate incurred bite-wounds and the dominant had none (Ferrari et al., 1998; Bartolomucci et al., 2001, 2004; Colas-Zelin et al., 2012). This method of identifying dominance was used in two other studies without the provision of home-cage behavior (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992). Notably, one study measured home-cage behavior but only for a single day before testing general behavior (Colas-Zelin et al., 2012)—while all other home-cage behavior studies measured dominance for at least three consecutive days. The studies using home-cage observation had very inconsistent methods regarding the frequency of measuring dominance (Range = 1–14 consecutive days, Media n = 3 days), time spent observing dominance for each test day (5–60 min, see Table 2), the age of the animals at grouping (Range = 0–100 days of age, Median = 68.5 days of age), and the time the animals were housed together before recording dominance (Range = 0–90 days, Median = 10.5 days). Of the studies that evaluated home-cage behavior, three of them discarded groups where cage-mates did not have unique dominance ranks (Vekovishcheva and Sukhotina, 2000; Bartolomucci et al., 2001; Wang et al., 2011). Notably, the study by Vekovishcheva and Sukhotina (2000) experimentally formed groups with linear hierarchies by identifying the primary aggressor in a group of eight, then identifying the secondary aggressors by consecutively removing the group aggressor every 3 days until a final submissive animal that showed no aggressive behavior was left. The groups of eight were then reduced to groups of three composed of a primary aggressor, secondary aggressor, and the final submissive animal.

The next most common method for the assessment of dominance behavior involved the use of the tube-test (or competitive exclusion test) (seven out of 20) (Wang et al., 2011; Larrieu et al., 2017; Zhou et al., 2017; Kunkel and Wang, 2018; Varholick et al., 2018, 2019; Pallé et al., 2019). For this task, cage-mates are simultaneously placed on opposite ends of a long-narrow tube to impose a face-to-face conflict terminating when one cage-mate retreats backwards to their starting point. The retreating cage-mate is assigned a “loss” and its partner is assigned a “win.” These dyadic encounters are usually organized in a predetermined and random round-robin tournament. The total number of “losses” with the respective pairing compose the dominance hierarchy for each cage. Studies indicate dominance in the tube-test significantly correlates with home-cage observation (Wang et al., 2014), although some have questioned the utility of the test; namely that it doesn't always correlate with home-cage observation, animals adapt to the test over time, and it only measures a single facet of social dominance behavior (Wilson, 1968; Syme, 1974; Miczek and Barry, 1975; Benton et al., 1980; Curley, 2011; Varholick, 2019). Again, methodologies greatly varied between studies regarding the frequency of measuring dominance, the age of the animals, and the time the animals spent together before testing (see Table 2). For example, some studies measured dominance for 1 day every week for 3 weeks (Varholick et al., 2018, 2019), while others measured dominance across four or more consecutive days (Wang et al., 2011; Larrieu et al., 2017; Zhou et al., 2017; Kunkel and Wang, 2018), or every day until a group of cage-mates attained stable ranks (Pallé et al., 2019). Several studies discarded groups that had unstable hierarchies across their predetermined number of study days (Wang et al., 2011; Zhou et al., 2017).

Other methods of assessing dominance like the urine marking assay, ultrasonic vocalization, and warm-spot competition were always used in conjunction with either home-cage behavior or the tube-test but were rarely used in general [i.e., urine marking assay two out of 20 (Wang et al., 2011; Larrieu et al., 2017), ultrasonic vocalization one out of 20 (Wang et al., 2011), and warm spot competition one out of 20 (Zhou et al., 2017)]. In all cases dominance measured in these tests correlated with assessments in the home-cage or tube-test for the respective study. The urine marking assay was used for two studies (Wang et al., 2011; Larrieu et al., 2017), and involves placing two cage-mates in a novel empty cage separated by a mesh barrier with filter paper flooring. The dominant cage-mate will leave urine marks throughout their partitioned area while the subordinate will leave a pool of urine in a corner. The ultrasonic vocalization test was used once (Wang et al., 2011) in conjunction with the tube-test and home-cage observation. Here, separated males are presented with a female and 70 kHz vocalizations are recorded where the most dominant vocalizes for the longest and the subordinate often does not vocalize at all (Nyby et al., 1976). The warm-spot competition test was also used once (Zhou et al., 2017) with the tube-test, and involved presenting a group of cage-mates with a cold floored cage with a single warm-spot in the corner. Cage-mates that spent the longest time on the warm-spot were considered most dominant, with subordinate cage-mates spending the least amount of time on the warm-spot. Additional explanations of each dominance assessment method can be found in the Supplementary Text 2.

The 55 individual experiments from the 20 studies were categorized into several behavioral outcome assessment domains according to Wahlsten (2011) and the studies' method sections; Exploration (n = 17) (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992; Vekovishcheva and Sukhotina, 2000; Bartolomucci et al., 2001, 2004; Palanza et al., 2001; Sá-Rocha et al., 2006; Saldívar-González et al., 2007; Wang et al., 2011; Colas-Zelin et al., 2012; Horii et al., 2017; Larrieu et al., 2017; Zhou et al., 2017; Varholick et al., 2018, 2019), Anxiety (n = 17) (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992; Ferrari et al., 1998; Vekovishcheva and Sukhotina, 2000; Bartolomucci et al., 2004; Sá-Rocha et al., 2006; Saldívar-González et al., 2007; Colas-Zelin et al., 2012; Horii et al., 2017; Larrieu et al., 2017; Varholick et al., 2018, 2019; Pallé et al., 2019), Learned Helplessness (n = 5) (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992; Saldívar-González et al., 2007; Horii et al., 2017), Cognitive (n = 8) (Fitchett et al., 2005a, 2009; Colas-Zelin et al., 2012; Varholick et al., 2018, 2019), Social (n = 2) (Zhou et al., 2017; Kunkel and Wang, 2018), and Sensory (n = 6) (Colas-Zelin et al., 2012; Zhou et al., 2017). Because tests measuring exploration or anxiety were most frequent, they were further sub-divided into a specific test and other category: for exploration, Open Field (n = 11) and Other Exploration tests (n = 6); and for anxiety, Elevated Plus-Maze (n = 13), and Other Anxiety tests (n = 4). Several studies met all inclusion and exclusion criteria, yet reported null effects without the provision of data: Exploration (n = 3/16) (Wang et al., 2011; Colas-Zelin et al., 2012; Zhou et al., 2017), Anxiety (n = 4/17) (Bartolomucci et al., 2004; Colas-Zelin et al., 2012) Cognitive (n = 4/8) (Fitchett et al., 2009; Colas-Zelin et al., 2012), Social (n = 1/2) (Zhou et al., 2017), Sensory (n = 5/6) (Colas-Zelin et al., 2012; Zhou et al., 2017). These studies are included in the tables but are excluded from the meta-analyses (Bartolomucci et al., 2004; Wang et al., 2011; Zhou et al., 2017). Only 2 studies used males and females (Varholick et al., 2018, 2019), while the remaining 18 studies exclusively studied males.

Meta-analyses for exploration behavior were divided into two separate analyses for behavior in the open field and other exploration tests. The category designated as “other” in exploration behavior was not used in meta-analyses due to high heterogeneity in paradigm methodologies between studies (e.g., novel object exploration, hole-board crossings, and activity meter). For the meta-analysis that was run, open field behavior had heterogeneous metrics reported between studies. Thus, we prioritized common metrics for the data analysis. The most common metric for the open-field was total distance traveled, followed by number of crossings, then velocity; and for general exploration the number of crossings was most common followed by total distance.

The meta-analysis on exploration in the open field estimated a medium effect size of 0.484 (k = 9, se = 0.273) that was not statistically significant (p = 0.077, 95% CI = −0.05, 1.02, Figure 2A), with experiments generally finding a small and statistically non-significant effect with dominant mice exploring more than subordinates (7/9). The other two experiments found a large statistically significant effect of dominants exploring more than subordinates (Saldívar-González et al., 2007) or a small non-significant effect in the opposite direction (Larrieu et al., 2017). There was significant between-study heterogeneity (Tau² = 0.466, se = 0.332, df = 8, p = 0.002). For the experiments categorized as “other” within the exploration domain, the mean differences between dominant and subordinate mice, the pooled standard deviations, and Hedge's g values were calculated for each study (Table 3).

As in the exploration domain, analyses in the anxiety behavior domain were sub-divided into elevated plus-maze and other anxiety tests. A meta-analysis was performed for the elevated plus-maze, while tests in the “other” category were not used in meta-analyses due to high heterogeneity in paradigm methodologies between studies (e.g., light/dark box, defensive burying, shuttlebox). The elevated plus-maze had heterogeneous metrics reported between studies, thus we prioritized common metrics; percent entries in open arms was most common, followed by percent duration in open arms, and finally number of open arm entries. Notably, not all studies directly reported “percent” but provided enough information to calculate the percentage (e.g., number of open arm entries divided by total entries), allowing more studies to have more comparable metrics. The meta-analysis on elevated plus-maze behavior yielded a small effect size of 0.132 (k = 10, se = 0.206) that was not statistically significant (p = 0.522, 95% CI= −0.27, 0.54, Figure 2B). With about an equal number of experiments finding a small and non-significant effect in contradicting directions (4 indicating increased anxiety for dominants and three increased anxiety for subordinates), another two with significant effects in contradicting directions (Horii et al., 2017; Larrieu et al., 2017), and one study finding virtually no effect (Varholick et al., 2018). There was, again, significant between-study heterogeneity (Tau² = 0.256, se = 0.197, df = 9, p = 0.003). For the experiments categorized as “other” within the anxiety domain, the mean differences between dominant and subordinate mice, the pooled standard deviations, and Hedge's g values were calculated for each study (Table 3).

The only tests that measured learned helplessness in this review were those using the Porsolt forced swim test. Five studies conducted this test in relation to dominance and thus satisfied our criteria for inclusion. All studies reported the same metric, duration immobile, thus the meta-analysis was limited to comparing duration immobile between studies. The meta-analysis on the Porsolt forced swim test estimated a small effect size of 0.0480 (k = 5, se = 0.872) that was not statistically significant (p = 0.956, 95% CI = −1.66, 1.76, Figure 2C). This was attributable to contradictory statistically significant findings across experiments with three reporting subordinates spend more time immobile (Hilakivi et al., 1989; Hilakivi-Clarke and Lister, 1992; Horii et al., 2017) and two reporting dominants spend more time immobile—albeit the latter experiments were from the same study (Saldívar-González et al., 2007). There was again substantial between-study heterogeneity (Tau² = 3.574, se = 2.691, p = 0.001).

A secondary set of forest-plots for open field, elevated plus-maze, and Porsolt forced swim test focusing on relevant study characteristics (i.e., strain, group-size, and dominance assessment method) can be found in the supplement (Supplementary Figure 2).

Given a high degree of heterogeneity across behavioral outcome assessments and inconsistent reporting between studies within the separate domains; cognitive, social, and sensory, the data were insufficient for meta-analyses. However, for comparison to the previous meta-analyses and further discussion, the mean differences, pooled standard variations, and Hedge's g values are reported in Table 4. Experiments marked with a “=” sign in the “Dom vs. Sub” column did not report values in their study, they just reported that there was no significant difference between groups.

Similar to the meta-analyses for exploration and anxiety behavior, experiments generally reported no large effects between dominant and subordinate mice for cognitive, social, or sensory behavior. Regarding cognitive behavior, most experiments reported small effect sizes with no statistical significance (n = 6/8), albeit five of those experiments did not report the data (Fitchett et al., 2009; Colas-Zelin et al., 2012). Only two studies measured social behavior, beyond dominance, with opposing results; one with no effect and no reported data (Zhou et al., 2017) and another with a large effect indicating dominant mice had increased social memory compared to subordinate mice (Kunkel and Wang, 2018). Finally, regarding sensory behavior all experiments found virtually no effect between dominant mice and subordinate mice, with 5 out of 6 not reporting the data.