Keeping One’s Distance: Mask Wearing is Implicitly Associated With Psychological Distance

Participants

We recruited a total of 354 German-speaking individuals for four independent experiments with identical procedures and materials (except for the IAT, which differed according to the psychological distance dimension in question). ¹ Of these, five (0.01%) individuals were excluded based on our preregistered exclusion criterion (i.e., IAT response latencies of 10% of all trails <300 ms). The inclusion of these participants did not impact the results reported below. Participants were recruited online to participate in a brief (10 min) study on “Attitudes and Thoughts During the COVID-19 Pandemic.” To this end, we used 216 university email distribution lists and made sure that each experiment was advertised using a distinct, nonoverlapping set of email lists. As an incentive, participants could enter a draw for a gift card for an online retailer worth 25€ or obtain course credit. After consenting to participate, participants completed the IAT. JavaScript was used to achieve a millisecond-accurate recording of response latencies (for further details on response-timing accuracy in JavaScript online studies, see Reimers & Stewart, 2015). Table 2 details the final sample characteristics for each experiment.

OPEN IN VIEWER

Table 2.

Sample Characteristics in Experiments 1–4.

	Sample Characteristics
Experiment	Sample Size	Gender Composition (% of Female/Male)	% Students	M age	SD age	Age Range
Experiment 1	92	81.5%, 18.5%	22.8	28.07	7.73	18–56
Experiment 2	83	84.3%, 14.5% a	69.9	26.59	7.52	19–54
Experiment 3	78	64.1%, 35.9%	28.2	28.74	9.69	18–58
Experiment 4	96	68.8%, 31.2%	43.8	24.48	5.54	18–50

Note. N = 349.

^a Gender information was missing for one individual (1.2%).

Power Considerations

The required minimum sample size per experiment was predetermined on the basis of an a priori power analysis using G*Power (Faul et al., 2009). On the basis of similar CLT-research using the IAT (Bar-Anan et al., 2006), we expected to find a medium effect of .41 (α = .05, power: .80) and calculated a sample size of at least N = 49 per experiment. Given the novelty of the studied effect, we deliberately oversampled to ensure high power (i.e., N range: 78–96, for Experiments 1–4).

Procedure

In all four experiments, participants completed an IAT. Given that the IATs were identically constructed, we only describe the IAT from Experiment 1 examining social distance. Figure 1 shows example stimuli used in the IAT.

OPEN IN VIEWER

Figure 1.

Implicit Association Test sample screens and stimuli for Experiment 1 (social distance).

Pictures of masked or unmasked faces and words representing socially close or distant targets were presented in the center of the screen. Individuals were asked to classify these stimuli as fast as possible into four different categories using two different keys; two target categories (“masked faces” vs. “unmasked faces”; four pictures each) and two attribute categories (“near people” vs. ”distant people”; four words each). The speed to associate masked faces + distant people and unmasked faces + near people (Figure 1A and B) relative to the opposite pairing (Figure 1C and D) was taken to reflect an implicit mental association between masked faces and social distance. ²

Block 1 (20 trials) introduced the target category (masked faces vs. unmasked faces), and Block 2 (20 trials) introduced the attribute category (near people vs. distant people). Both were combined in critical Blocks 3 (20 trials) and 4 (40 trials). The target categories switched sides in Block 5 (20 trials), resulting in switched pairings in critical Blocks 6 (20 trials) and 7 (40 trials). As counterbalancing the order of compatible and incompatible blocks was found to distort effect sizes and, hence, bias effects-size estimates, the order of the combined blocks was held constant (for a discussion of this procedural detail, see Banse et al., 2001).

For each participant, IAT scores were calculated using the improved scoring algorithm (Greenwald et al., 2003). This score divides the difference between critical block means by the individual-respondent standard deviation of all the latencies in the two critical blocks. Higher scores reflect higher implicit associations between masked faces and psychological distance relative to psychological proximity. We applied the improved scoring algorithm separately to two mutually exclusive subsets of the critical trials of the IAT to estimate its reliability (Schmukle & Egloff, 2006). The reliability of the IAT was high (Spearman–Brown corrected split-half reliability based on parallel test halves: .87–.94, for Experiments 1–4).

Stimuli Used in the IATs

For the target categories (i.e., masked vs. unmasked faces), four male and four female pictures with a neutral facial expression were taken from the FACES database (Ebner et al., 2010). An independent pilot study (N = 200; M _age = 23.77 years, SD = 3.69; 100 male, 100 female) confirmed that the four pictures were of equal attractiveness and trustworthiness. ³ A professional graphic designer modified the pictures by individually applying face masks to cover the same part of each face. All images were converted to gray scale and adjusted for size, luminance, and contrast. Altogether, there were eight pictures (i.e., two versions of a masked male, masked female, unmasked male, and unmasked female person). For the attribute categories (i.e., psychological distance vs. proximity), words were adopted from the study by Bar-Anan et al. (2006), which similarly examined all four psychological distance dimensions using IATs. Lexical analyses revealed that that the words belonging to either attribute category did not differ in terms of word length (ps ≥ .182) or word frequency (ps ≥ .302).

Data Analyses

To test our main hypothesis that masked faces are associated with psychological distance relative to psychological proximity, we analyzed mean RTs in milliseconds for congruent (masked faces/distance + unmasked faces/proximity) and incongruent (unmasked faces/distance + masked faces/proximity) conditions using paired samples t tests. Although not preregistered, we aimed to provide a methodological replication of our results by analyzing the IAT scores using one-sample t tests (test value = 0, as this would indicate a neutral association between the target and attribute categories). This analysis allowed us to take the individual standard deviation of response latencies into account (i.e., individual variability of the latencies), thereby providing a more stringent test of our hypothesis. This analysis yielded similar results to those obtained below (see Supplementary Table S1).

Public Health and Social Implications

How might the effect we documented affect physical distancing, which represents another primary means of pandemic control? Past research has shown that health behaviors are often implicitly influenced by cues in the environment (Hollands et al., 2016; Sheeran et al., 2013). Face masks may similarly serve as a psychological distance cue and, hence, increase distancing behaviors without participants realizing the impact of the cue. Face masks may thus “nudge” individuals toward the social norm to keep more distance. Consistent with this, field experiments have shown that people keep a greater distance to masked versus unmasked individuals (Marchiori, 2020).

Alternatively, it is also possible that face masks may decrease, rather than increase, physical distancing. If people perceive masked individuals as more distant, as the present findings show, they may be inclined to keep a lesser distance to masked individuals because all else being equal a person would seem farther away if masked than not. This notion is consistent with the risk compensation hypothesis (e.g., Hedlund, 2000), according to which people may compensate for the risk reduction provided by masks by adjusting related behaviors (e.g., approaching others). Future research is needed to determine whether an implicit association between masked faces and psychological distance increases or decreases compliance with physical-distancing recommendations.

The effects of mask wearing may have other consequential social implications beyond health-protective behaviors. A recent behavioral economics study, for example, revealed that mask wearing alters people’s behavior in a decision-making task (Fatfouta & Oganian, 2020). Specifically, individuals accepted more monetary offers (including more unfair offers) from a masked than from an unmasked proposer. Mask wearing was interpreted as increasing social distance and, hence, prompting more rational bargaining behavior due to a focus on more abstract considerations. Our research adds credence to this notion and suggests that incidental cues can affect behavioral processes outside people’s awareness. On the other hand, mask wearing might have adverse social consequences. Indeed, prior research has shown that mask wearing may impair smooth social interactions (Carbon, 2020). One might even speculate that the abstract and distanced representation of mask wearers might deindividuate them and lead to treating them as an undifferentiated collective (LeBon, 1895/2002; Zimbardo, 1969)

Potential Limitations and Directions for Future Research

Despite their robustness, the generalizability of the current findings and their implications need to be considered with caution. First, the studies were conducted in Germany at a time when mask ordinances were in place and most people in this country have already grown accustomed to wearing and seeing others wearing. Compared to early in the pandemic, mandatory face mask use may have affected the implicit associations that have emerged. Indeed, implicit measures are known to be influenced by contextual factors including the time period in which the data are collected (Gawronski, 2019). Future research would benefit from examining the malleability of implicit associations across the different waves of the COVID-19 pandemic (and beyond). For example, as more people get vaccinated and mask ordinances are lifted, the implicit associations between masked faces and psychological distance may dissipate over time.

Second, an important open question pertains to the formation of implicit evaluations of masked (vs. unmasked) faces. At present, we do not know whether the results are reflecting newly formed associations during the pandemic or, alternatively, whether the same results would have been obtained prior to the pandemic and whether they may hold long after the pandemic ends. Assuming that implicit measures capture overlearned associations produced by long-term socialization (e.g., Wilson et al., 2000), one might speculate that the implicit associations we have documented would depend on the frequency of exposure to masked faces (e.g., as part of one’s profession). For example, mask wearing has been traditionally required by some professions (e.g., surgeons, dentists) and is common in some cultures (e.g., East Asian). Longitudinal research on such groups might shed light on the trajectory of the implicit associations we observed between masked faces and psychological distance.

Third, further research is needed to elucidate the mechanism that might underlie this association. We argued that masked faces are associated with psychological distance because they are represented more schematically. This needs to be tested more directly by examining potential mediating perceptions (e.g., perceived abstractness).