Abstract
This paper reviews research published from 1998 to 2022 that involved collection of original data related to behavioral compliance with safety signs and labels. This review provides a needed update to previous similar reviews using a published taxonomical approach. The approach was used here to categorize study methodology and was extended to further categorize studies by whether they addressed behavioral compliance or intention/predicted likelihood of compliance. Nearly all of the studies reviewed assessed behavior in laboratories or via surveys of hypothetical scenarios, and relatively few used real-world experimental or real-world observational methods. Results revealed a lack of empirical demonstrations that intention and predicted likelihood of compliance are reliable predictors of real-world behavioral compliance. There was also a notable lack of evidence that warnings design aspects impact such real-world compliance.
Keywords
Introduction
A number of reviews pertaining to research on behavioral compliance with safety signs and labels have been conducted (e.g., McCarthy, et al., 1984; DeJoy, 1989; Cox et al., 1997; Ayres, et al., 1998; Silver & Braun, 1999; Rogers, Lamson, & Rousseau, 2000; Kalsher & Williams, 2006; Hancock, et al., 2020). The current paper provides a descriptive analysis of trends in applied research on safety signs and labels, rather than a formal meta-analysis or theoretical framework. As such, the current paper provides a much-needed literature review update of applied behavioral compliance research published from 1998 to 2022, continuing the use of the methodological taxonomy used by Ayres, et al. (1992, 1998) to categorize the studies reviewed. The current paper also expands the taxonomy by examining both behavioral compliance and intention/predicted likelihood of compliance measures.
Method
Search Methods and Inclusion/Exclusion Criteria
Various journals and online publication databases (e.g., Google Scholar, Human Factors and Ergonomics Society journals, Journal of Safety Research, Safety Science) were searched using the terms “warnings,” “warnings effectiveness,” and “warnings compliance.” Publications were also gathered from existing compilations and citations found in other publications. The review set was limited to publications that:
were dated 1998-2022;
described collection of original data;
studied responses to static, printed, safety-related on-product labels or safety signs; and
examined behavioral compliance and/or intention/predicted likelihood of compliance.
Excluded were studies that used other dependent variables (e.g., studies of symbol comprehension) and studies of non-safety-related warnings (e.g., warnings related to product performance, property damage, or security threats in information technology settings). Also excluded were studies regarding safety-related information encountered while driving (e.g., road signs), auditory or dynamically displayed warnings, weather-related evacuation orders (e.g., hurricanes, tornadoes), and warnings about consumption of alcohol, tobacco, unhealthy foods, or cannabis.
For publications reporting results from multiple studies, each study was considered separately. No evaluation of study quality was made in including or excluding publications.
Categorization
Each included study was categorized according to the methodological taxonomy developed by Ayres, et al. (1992). Categories distinguished between studies that were conducted in a laboratory (or other contrived environment) versus experiments or observations conducted in real-world situations with unsuspecting participants. Studies involving laboratory settings were further categorized by whether (1) the product of interest was obvious to the participants or (2) the product of interest was disguised such that participants were led to believe that it was incidental to the purpose of the experiment. Studies involving real-world settings were further categorized by whether (3) the researchers manipulated independent variables as part of an experiment or (4) the researchers made observations but did not manipulate any variables. The resulting four categories were: (1) Product-Obvious Contrived-Environment Experiments, (2) Product-Disguised Contrived-Environment Experiments, (3) Real-World Experiments, and (4) Real-World Observations.
As a step beyond the taxonomy developed by Ayres et al. (1992), each study was further categorized by its dependent measure based on whether the study measured behavioral compliance and/or intention/predicted likelihood of compliance. The latter (intention/predicted likelihood of compliance) was further subdivided by whether the study measured participants’ predictions of their own compliance (i.e., intentions) or participants’ predictions of the likelihood of compliance of others. See Table 1 for a summary of the resulting extended taxonomy.
Extended taxonomy.
Results
A total of 68 studies were included in the current review set (see Table 1). Of these 68 studies, 49 were noted as Product-Obvious Contrived-Environment Experiments, 12 were noted as Product-Disguised Contrived-Environment Experiments, six were noted as Real-World Experiments, and three were noted as Real-World Observations. (Two studies included both Product-Obvious and Product-Disguised elements.) A total of 43 studies measured behavioral compliance, and 28 measured intention/predicted likelihood of compliance. (One study measured both behavioral compliance and intention/predicted likelihood of compliance.)
Discussion
Research Trends
The pace of research on behavioral compliance with safety signs and labels appears to have slowed in recent years relative to the late 1990s and early 2000s. This observation may be an artifact of the exclusion criteria used in the current paper; for example, there may have been a shift towards studying dynamic warnings or warnings in vehicles. It may also reflect a shift toward studying other dependent variables, such as comprehension and information recall.
A variety of independent variables were manipulated in the reviewed studies, including various warnings design aspects and characteristics of the message receivers and contexts. Warnings design aspects included variables such as ANSI or ISO formatting features, signal words, pictograms, colors, source attribution, and explicitness. Receiver/context characteristics included variables such as product familiarity, training, cognitive task loading, and social influence. Of particular interest in the current paper is whether and how manipulations of warnings design aspects affect behavioral compliance in real-world settings.
Study Designs
Consistent with reviews of earlier time periods (e.g., Ayres et al., 1998), laboratory experimentation remained the numerically predominant method for investigations of compliance with safety signs and labels published from 1998 to 2022. About 87% (59/68) of the studies in the current review set used the Contrived-Environment Experiment method, in which respondents knew they were participating in a research study. Of these studies, the majority (83% [49/59]) used the Product-Obvious method in which participants also knew which product or warning was being evaluated. These studies included both laboratory experiments involving physical interactions with a product and survey research involving entirely hypothetical scenarios. The Product-Disguised Contrived-Environment Experiments (20% [12/59]) differed only in that participants were unaware of what product or warning was being evaluated. (Two studies included both Product-Obvious and Product-Disguised elements.)
Real-World studies in which participants were unaware that they were involved in research constituted the remaining 13% (9/68) of the studies in our review set. The majority of these studies (67% [6/9]) used the Real-World Experiment method in which the experimenter manipulated conditions in real-world settings. The remaining studies (33% [3/9]) used the Real-World Observations method in which the researcher(s) made observations but did not manipulate independent variables.
Behavioral Compliance
While behavioral compliance with warnings has been relatively frequently studied under controlled conditions in Product-Obvious Contrived-Environment Experiments, including in laboratory settings, studies of behavioral compliance in real-world settings are still uncommon. Behavioral compliance was measured in less than half (45% [22/49]) of the Product-Obvious Contrived-Environment Experiments in the current review set. The remaining studies (57% [28/49]) examined intention and/or predicted likelihood of compliance. Table 2 (One study measured both behavioral compliance and intention/predicted likelihood of compliance.) Of these studies, 18 measured intention (predicted likelihood of participants themselves complying), and 12 measured participants’ predictions of the likelihood of others complying; two studies measured both.
Publications reviewed.
In contrast, behavioral compliance was measured in all studies using the Product-Disguised Contrived-Environment Environment, Real-World Experiment, and Real-World Observation methods. Overall, behavioral compliance was measured in about two-thirds (63% [43/68]) of the studies included in the current review set. Notably, none of the Real-World Experiment or Real-World Observation studies concluded that warnings design aspects had a significant impact on behavioral compliance.
Conclusions
The studies reviewed here demonstrate that: (1) measures of intention/predicted likelihood of compliance have often been substituted for measures of behavioral compliance in Product-Obvious Contrived-Environment Experiments, including both laboratory and survey research; (2) behavioral compliance has been studied in laboratory settings, such as in Product-Obvious Contrived-Environment Experiments, more often than in Real-World Experiments or Real-World Observations; (3) there has been a lack of empirical demonstrations that intention and predicted likelihood of compliance are reliable predictors of real-world behavioral compliance; and (4) there has also been a lack of evidence that warnings design aspects impact such real-world compliance.
Some researchers have suggested that intentions can reliably predict behavior (e.g., Silver & Braun, 1999). However, in research specific to safety signs and labels, others have shown that intentions/predictions of compliance are not consistent with behavioral compliance (e.g., Frantz et al., 2005a). The current review shows that, despite this documented discrepancy, studies persist in measuring intention/predicted likelihood of compliance instead of actual behavioral compliance. Given this discrepancy and other findings reviewed above, it appears that there is still no scientific basis for accurately or reliably predicting the effects of various manipulations of warnings design aspects on behavioral compliance in real-world contexts.
