The present study has been authorized by the Committee on Bioethics of the University of Pisa (Review No. 5/2020; AOO “CLE”—Prot.: 0036356/2020 of 10/04/2020). The study was purely observational and data were entered in an anonymous form (an alphanumerical code has been uniquely assigned to each subject). People have been observed in their natural social setting without any modification of their ordinary and daily activities.
Data collection and subjects
The data were collected in Italy across 5 months (May–September 2020) compatibly with the d.l. n.33 “further urgent measures to contrast the epidemiological emergency from COVID-19” issued by the Italian Government on May 16th, 2020.
The observations were temporally distributed across morning (from 07:00 am to 01:00 pm), afternoon (from 01:00 pm to 07:00 pm) and night (from 07:00 pm to 03:00 am). Experimenters observed subjects in their natural social settings during their daily activities (at work, restaurants, cinemas, gyms, waiting rooms, social parties, social meals, public parks, family environments, etc.). The subjects, who were unaware to be observed (blind data collection), were people known (family members, friends, acquaintances and co-workers) and unknown (strangers) to the experimenters. The observed persons could know each other or not. A total of 184 persons (88 women, 96 men) were observed and included in the dataset.
To be included in the analysis, the sequence of actions had to fulfil several criteria both during the Experimental (EC) and Control condition (CC). The CC had to be identical to the EC except for the presence of the behaviour “looking at the screen”. During the EC, we considered as trigger the person who took, kept in hands, and manipulated his/her smartphone (e.g. fiddling and swiping) and looked at the screen for at least 5 s. During the CC, we considered as trigger the person who took, kept in hands, and manipulated his/her smartphone (e.g. fiddling and swiping) for at least 5 s without looking at the screen. In both conditions, the screen of the device had to be visible and not covered by any cover. Only those events in which the device automatically illuminated by touching were included in the dataset, since the light had to be present both in EC and CC. The two different conditions were randomly distributed, and the different observation bouts were separated by at least 10 min. The main triggers were M.G.R (male) and V.M. (female) who were the experimenters as well. We opportunistically gathered data also when other people (unconscious male and female triggers), not aware of the ongoing study, spontaneously manipulated/looked at their own smartphones for at least 5 sec.
The observer was defined as the person who visually perceived the triggers’ action and had his/her smartphone within reach. In short, the observer should have the opportunity to engage in the same action of the trigger in both EC and CC. The experimenters had to be able to see the gaze of the observers during both EC and CC. Immediately after the trigger took the device (t0), all individuals visually perceiving the triggers’ action were observed for 3 min by the experimenters who checked for the presence/absence of a congruent mimicry response in the observers.
The latency in the response (when present) was scored on six levels made of 30-s blocks with the aid of a wristwatch (Casio F-91W-1YER-P), a device that allows checking time without any kind of manipulation and light production (no illuminated screen).
Before starting systematic data collection, reliability between the experimenters (V.M.; M.G.R.) was tested. During 15 observational sessions, both experimenters gathered data concurrently on the same observers. At the end of the training period, the Cohen’s kappa values (k) were calculated for (i) the opportunity to be seen by the potential observers, (ii) the occurrence of the mimicry event, and (iii) the time latency. For all these conditions, the k values were always higher than 0.85 (Kaufmann and Rosenthal 2009). At regular intervals, to check for reliability, both experimenters collected data concurrently on the same group of people always obtaining k values higher than 0.85.
After 3 min, to make the data registration possible and unnoticed, the experimenters moved away from the observed subjects and took notes of their behaviour on smartphones or paper. The identity of the observed subjects was stored under alphanumerical codes.
We excluded from the database all the cases in which people, while using their smartphone, actively solicited the observers’ attention by indicating/showing the device (nonverbal solicitation) and/or verbally inviting to use it (e.g. “look at that video” and “look at that post on…”).
Both in EC and CC, we recorded the behaviour of the observer (presence of mimicry response/absence of mimicry response) during a 3-min time slot after seeing the trigger’s action. The occurrence of mimicking was coded as 1 (presence) or 0 (absence).
The response latency was measured as the time delay between the first touching of the smartphone by the trigger (t0) and the first touching of the smartphone by the observer (tx). The time latency was scored on six levels: 0 < tx ≤ 30 s = 1; 30 s < tx ≤ 1 min = 2; 1 min < tx ≤ 1.5 min = 3; 1.5 min < tx ≤ 2 min = 4; 2 min < tx ≤ 2.5 min = 5; 2.5 min < tx ≤ 3 min = 6.
The timing of observations was clustered as follows: morning (07:00 am–01:00 pm) = 0; afternoon (01:00 pm–07:00 pm) = 1; night (07:00 pm–03:00 am) = 2.
We recorded and categorized the sex (men = 0; women = 1) and age of the trigger and observer (18–25 years = 0; 26–40 years = 1; 41–60 years = 2).
The relationship between the trigger and the observer was clustered on four categories: strangers (people who had never met before = 0), acquaintances (people who exclusively shared an indirect relationship based on a third external factor—work duty, colleagues, friends in common, friends-of-friends = 1), friends (not kin subjects sharing a direct friendship relationship = 2), regular engaged partners and kin (family members and cohabitants = 3). In most cases, the relationship between the observed people was known to the experimenters. When the trigger was different from and unknown to M.G.R. and V.M., the experimenters collected personal information (e.g. age, relationship between the observed subjects) by engaging in a friendly conversation. When it was not possible to gather information on the age of the observed subjects or on their relationship, we excluded the record from the dataset.
Since food is a strong factor of affiliation in humans, we also categorized the social context in which the data have been recorded as a function of the absence = 0 or presence of food = 1. The context “presence of food” began when the subjects sat down at the table and ended when they left the table. In addition, during meals, subjects had the opportunity to manipulate their devices if they wanted. Social breakfasts, lunches, dinners and happy hours were included in the cluster ‘presence of food’. All the other social contexts such as working, travelling, relaxing time, board gaming, card gaming, studying in libraries and waiting in a sitting room (e.g. hair dressing salons, dentist studios) were clustered as ‘absence of food’.
Experimental and control conditions were randomly distributed across all the possible contexts and the periods of the day.
Data analysis and statistics
From a total of 820 events (NEC = 472; NCC = 348) involving 184 subjects (women = 88; men = 96), we extracted for the analysis 721 events (NEC = 386; NCC = 335) involving 103 subjects (women = 50; men = 53) that were tested for both conditions (EC and CC). To investigate the factors affecting the mimicry response in the use of smartphones, we ran a Generalized Linear Mixed Model (GLMM) with a binomial error distribution by means of the R-package glmmTMB 1.2.5042 package (Brooks et al. 2017), using absence/presence of mimicry as response variable. We included only the subjects who had at least one observation in the EC and one in CC (N = 721 cases). The fixed effects were the condition (Control condition, CC; Experimental condition, EC), the age of the trigger and the receiver (18–25 years; 26–40 years; 41–60 years), the sex of the trigger and the receiver, the level of familiarity between the trigger and the receiver (strangers; acquaintances; friends; kin), the period of the day (morning, afternoon and night), and the context (presence of food; absence of food). The identities of the trigger and the receiver were entered as random factors.
The overall significance of the full model was tested by comparing this model with the model including only the random effects (Forstmeier and Schielzeth 2011) by means of the Likelihood Ratio Test (LRT; Dobson 2002). The LR test was used also to test the significance of the fixed factors using the function Anova in the R-package car 3.0–10 (Fox and Weisberg 2019). To exclude the occurrence of collinearity among predictors, we examined the variance inflation factors (VIF; Fox 2016) by means of the R-package performance 0.4.4 (Lüdecke et al. 2020). No collinearity has been found between the fixed factors (range VIFmin = 1.06; VIFmax = 1.57). Model fit and overdispersion were checked using the R-package DHARMa 0.3.3.0 (Hartig 2020). The marginal R2, which represents the variance explained by fixed factors only, and the conditional R2, which represents the variance explained by the entire model including both fixed and random effects (Nakagawa et al. 2017), were calculated using the R-package MuMIn 1.43.17 (Bartoń, 2020). Then, we used the “confint(x)” function to interpret the estimated effects as relative odds ratios. Relative odds ratio (i.e. the expected odds change for one unit increase in the explanatory variable when the remaining variables are set to their reference category) were used to evaluate the magnitude of the estimated effects. We performed all pairwise comparisons for the levels of the multilevel factor with the Tukey test (Bretz et al. 2010) using the R package emmeans (Length et al. 2020).
Lastly, to test whether the distribution of the mimicry response was homogenous across the six 30-s time windows, we applied the Chi-square test. From the original dataset (N = 820 events), we included in this analysis only the mimicry events occurred in the six different 30-s time window slots during the EC (N = 249 cases; women = 54, men = 59). All calculations were performed using R 4.0.3 (R Core Team 2020).