Abstract
There is often a discrepancy between people’s intentions to engage in protective behaviours and their objective actions, known as the intention-behaviour gap. This study investigated the relationships between behavioural correlates (attitudes, habits, beliefs, norms, intentions, risk-perception, worries, and perceived health) and self-reported behaviour in predicting objectively determined behaviour, using physical distancing during COVID-19 as a case study. Data were collected from 565 visitors of a Dutch art fair in August 2020, including pre-event questionnaires, objectively determined behaviour during the event, and post-event self-reports. Regression and Gaussian Graphical Model analyses revealed that self-reports were poor predictors of objectively determined behaviour. Among behavioural correlates, only perceived health showed a weak association with objective behaviour. These results demonstrate a marked disconnect between intentions, self-reports, and actual behaviour, underscoring the necessity of incorporating objective measures in public health policy design and evaluation.
Keywords
Introduction
Successfully navigating public health crises requires substantial shifts in human behaviour. Individual and collective actions are crucial in preventing, mitigating, and responding to such threats. Public health regulations, therefore, become essential for guiding individual behaviours to protect public safety. The importance of these regulations was exemplified during the recent COVID-19 pandemic, where the spread of the airborne virus could only be controlled through behavioural interventions such as physical distancing and wearing face masks before vaccines became available (Haug et al., 2020). These behavioural interventions, particularly physical distancing, were central to global efforts to reduce interpersonal contact and limit virus transmission (Sunjaya and Jenkins, 2020).
Understanding the factors that influence adherence to public health regulations is essential for designing and evaluating the effectiveness of such interventions. Yet, common approaches measuring behavioural responses often have their limitations. To illustrate, during the COVID-19 pandemic, adherence to public health regulations was primarily monitored by changes in the number of infections and hospitalisations (Glogowsky et al., 2021; Jarvis et al., 2020). Such measures, however, have a delayed effect due to the incubation time of the virus and cannot directly reflect the effectiveness and uptake of public health regulations (Borsboom et al., 2022).
Alternatively, the uptake of public health regulations can be monitored using behavioural correlates such as attitudes, habits, beliefs, norms, intentions, risk perception, worries, and perceived health, that can help assess psychological and social drivers of behaviour and are believed to be associated with adherence to public health regulations (Bish and Michie, 2010; Czeisler et al., 2020). Even though intentions and attitudes predict health-related behaviour, research shows a significant gap between what people intend to do and their actual behaviour (Armitage and Conner, 2001; Sheeran, 2002). This ‘intention-behaviour gap’ has been documented across various health-related behaviours, from physical activity to adherence to COVID-19 guidelines (Gibson et al., 2021; Rhodes and de Bruijn, 2013).
Finally, uptake of regulations can be monitored by directly asking about the behaviour of interest itself. Unfortunately, a similar problem arises as self-reported behaviour may not reliably reflect actual behaviour either. For example, Kormos and Gifford (2014) found that while self-reports of pro-environmental behaviour show a positive connexion with actual behaviour, 79% of the variance in behaviour remains unexplained. We find similar patterns in public health contexts. For example, individuals with obesity systematically underreport caloric intake compared to objective biomarkers (Archer et al., 2018; Lichtman et al., 1992), and smokers have been shown to underreport cigarette use when validated via biochemical markers like cotinine (Benowitz et al., 2020; Gorber et al., 2009). These misalignments highlight limitations of self-reported data, including recall bias, social desirability bias, and a mismatch between perceived and actual behaviour. Participants may unintentionally misremember their actions or report behaviours they believe are expected or socially approved, rather than what they actually did.
Taken together, these observed discrepancies highlight the limitations of self-reported data and raise questions about their reliability in evaluating adherence to health regulations during crises. Accurately assessing adherence to public health regulations, therefore, requires an understanding of how correlates of behaviour (e.g. intentions and attitudes), self-reports, and actual behaviour relate to each other. In the current study, we aim to address this gap and evaluate an integrated framework mapping relationships between psychological correlates, self-reported adherence, and objectively observed behaviour. Using physical distancing during COVID-19 as a case study, we analyse unique data collected during an art fair in the Netherlands in August 2020 (Blanken et al., 2021; Tanis et al., 2021), to explore whether gaps between self-reported and objectively determined physical distancing behaviour emerge, and how behavioural correlates relate to adherence to health regulations. We aim to discover the pathways and disconnects between intention, perception, and action, advocating for objective measurement in public health evaluation. A more comprehensive understanding of how these factors interact can inform the design and monitoring of more effective public health interventions, with the ultimate goal of improving compliance with public health interventions in future crises.
Methods
This study is based on previously published data, and no new data collection involving human participants was conducted. Ethical approval for the original study was obtained from the ethics review board of the University of Amsterdam (2020-CP-12488). The dataset was originally collected as part of a large-scale field experiment investigating behavioural interventions promoting physical distancing during COVID-19 (Tanis et al., 2021). The experiment varied walking directions (bidirectional, unidirectional, and no directions) and supplementary interventions (face masks and buzzer feedback) across time slots. Blanken et al. (2021) used these data to show that unidirectional walking directions and immediate buzzer feedback reduced visitor contacts, while face masks had no significant effect on distancing behaviour.
Participants
Participants were 839 visitors at an art fair held in August 2020 in Amsterdam, The Netherlands. For the current study, we included all participants who completed a questionnaire on behavioural correlates prior to visiting the art fair and wore a so-called ‘Social Distancing Sensor’ (SDS; Sentech BV, 2020) from which objective physical distancing measured could be derived, resulting in a final sample of 565 (67.3% of the total sample) participants. This sample consisted of 329 (58.2%) women, 233 (41.2%) men, and 3 participants (0.5%) who did not indicate their sex. No participants selected the ‘other’ sex category. The sample had an average age of 42.7 ± 15.8 years. Of the included participants, 209 (37.0%) completed a questionnaire on their self-reported behaviour upon leaving the art fair. This subgroup consisted of 108 (51.7%) women and had a mean age of 45.7 ± 14.8 years. Compared to the full sample (M = 42.6 ± 15.8 years), these participants were significantly older, t(326.48) = −2.40, p = 0.017 and consisted of fewer women, χ2(6) = 114.78, p < 0.001.
The original data is described in detail in Tanis et al. (2021), and all data is openly available on Figshare and a relational MySQL database.
Materials
Correlates of behavior
We assessed eight behavioural domains from the Smart Distance Lab questionnaires (Tanis et al., 2021): risk perception, COVID-19-related worries, norms, beliefs, attitudes, intentions, habits, and health perception. Each domain was measured through study-specific items developed for this project, rated on seven-point Likert scales (see Table 1 for full item wording and scale endpoints). For the risk perception domain, one item used a 0–100 scale and was rescaled to a 1–7 range to align with the other items in the domain before computing the average score. Domain scores were calculated by averaging relevant items in each domain.
Questionnaire domains, items, and response scales from the smart distance lab study (Tanis et al., 2021), translated from Dutch to English.
Objective behaviour
To access directly observed behaviour and compliance with physical distancing regulations, participants wore an SDS attached to a lanyard around their neck during the art fair. The SDS uses ultra-wideband technology to measure distances between devices and registers the number of times a sensor is within a specified distance from other devices. In our case, a contact was registered when another SDS was less than 1.5 m away, in line with physical distancing regulations in the Netherlands at that time. A higher number of contacts indicates less physical distancing. The device operates at an accuracy of up to 10 cm, registering contacts at a frequency of 1 Hz (Sentech BV, 2020). While no direct reliability estimates of the SDS are available, ultra-wideband technology has been shown to be among the most precise indoor localisation technologies with high accuracy (Elsanhoury et al., 2022). To accommodate visitors from the same household, up to four sensors were linked by simultaneously placing them on different base stations at the time of activation. Contacts between paired sensors were not registered, and no feedback was given (Blanken et al., 2021; Tanis et al., 2021).
Self-reported behaviour
After the art fair participants reported their adherence to physical distancing regulations on a seven-point Likert scale, see Table 1.
Procedure
Participants were recruited through social media promotion of the art fair. During recruitment, participants were informed that attending the art fair would involve participation in a naturalistic study. They were asked to complete an online pre-art fair questionnaire, which included an informed consent form and provided a code to purchase a ticket for the art fair. This questionnaire collected demographic information and assessed the correlates of behaviour as described in the Materials section.
Upon entering the art fair, it was checked whether visitors completed the pre-art fair questionnaire that included informed consent. Visitors who had not filled out the questionnaire were only asked to sign an informed consent form. Participants then received a lanyard with the SDS, which was activated at an activation desk. After activation of the SDS, participants entered the art fair, where they could stay as long as desired.
When participants exited and returned their SDS, they were asked to scan a QR code that gave access to the post-event fair questionnaire, in which participants were asked to report their adherence to physical distancing regulations.
Analyses
Self-reported and objectively measured behaviour
First, we evaluated the extent to which self-reported and objectively measured physical distancing behaviours aligned. We predicted the number of objectively determined contacts during the art fair (‘objective behaviour’) by the self-reported adherence to physical distancing regulations (‘self-reported behaviour’) using regression analysis. We expected self-reported adherence to negatively predict the number of contacts made, reflecting that people who reported adhering better to the physical distancing regulations would make fewer contacts while visiting the art fair.
Network analysis
Second, we aimed to explore how behavioural correlates related to each other and to objectively measured physical distancing. Our primary goal was to identify which pre-event correlates uniquely predicted distancing behaviour during the fair. This temporal separation assessing correlates before the event and behaviour during it, tests a key assumption of behavioural monitoring: that questionnaire assessments can generalise across settings and predict future actions (Sniehotta et al., 2014). We estimated a Gaussian Graphical Model (GGM) including all pre-event correlates (risk perception, attitudes, etc.) and the number of contacts (objective behaviour) as continuous variables.
Given our focus on links between correlates and behaviour, we used ggmModSelect (hypertuning = 0) to avoid bias from model overfitting (Borsboom et al., 2022; Isvoranu and Epskamp, 2023). The network visualised conditional associations between variables after controlling for all others, with edge thickness/colour saturation (blue = positive, red = negative) indicating association strength (Epskamp, 2016; Epskamp et al., 2018). Network stability was assessed via 1000 nonparametric bootstrap samples (Epskamp et al., 2018).
All analyses in this study were performed using R in RStudio version 4.1.3, with bootnet package version 1.5.
Results
Alignment between self-reported and objective behaviour
A total of 209 participants completed the post-questionnaire assessing their adherence to physical distancing regulations during the art fair. Most visitors rated themselves as highly adherent, with self-reported distancing scores ranging from 1 (not at all) to 7 (all the time), and a mean of 4.63 ± 1.32. For these same participants, the number of physical contacts objectively measured using the SDS ranged from 0 to 40, with a mean of 7.93 ± 7.36.
To examine whether participants’ subjective reports aligned with their real-world behaviour, we tested whether self-reported adherence predicted the objectively measured number of contacts during the event. We found that their self-reported adherence to physical distancing did not predict their objectively determined number of contacts during the event (F(1, 207) = 3.13, p = 0.08; B = −0.68, SE = 0.38, t = −1.77, p = 0.079; R2 = 0.015). Only 1.5% of the variation in objectively determined number of contacts could be explained through their self-reported adherence, suggesting a limited correspondence between self-reported and objectively determined behaviour.
Network analysis of correlates of behaviour
Next, we investigated the interplay of behavioural correlates and their relationship to objective physical distancing by estimating a Gaussian Graphical Model. As shown in Figure 1. The number of contacts made during the event related to only one behavioural correlate: perceived overall health. Perceived overall health demonstrated a weak negative association with the number of physical contacts (r = −0.12), suggesting that individuals with worse perceived health were slightly less likely to adhere to physical distancing guidelines.
Network analysis of behavioural correlates and objectively determined distancing.
Bootstrap analysis indicated that this edge was included in approximately 65% of bootstrap samples, suggesting moderate stability. While traditional predictors of health-related behaviours, such as intentions, attitudes, habits, beliefs, and social norms, did not exhibit direct links to physical distancing, most correlates were interrelated. This suggests that these behavioural correlates may influence one another but operate largely independently of objectively determined physical distancing behaviour.
Discussion
The primary aim of this study was to explore the extent to which behavioural intentions, perceptions, and actions align in the context of public health interventions. We investigated their alignment using a dataset collected during Covid-19 where physical distancing regulations were in place during an art fair. The unique dataset included measures on several behavioural correlates of physical distancing behaviour, such as intentions, habits, and attitudes, assessed prior to the event; objective assessments of the number of contacts made during the event, and self-reported adherence to physical distancing regulations upon leaving the event. Our study revealed a fundamental disconnect between psychological drivers, self-reported adherence, and objectively determined behaviour, demonstrating that neither intentions nor self-reports alone reliably predicted actual compliance with protective measures.
Our findings closely align with the well-documented intention-behaviour gap (Sheeran, 2002) and extend it by empirically contrasting three assessment layers (correlates, self-report, observation) within a real-world setting. While theories like the Theory of Planned Behaviour (TPB; Ajzen, 1991; McEachan et al., 2011) posit that intentions predict actions, our network analysis found no direct links between TPB-aligned correlates (attitudes, norms, intentions) and objective distancing. Instead, these correlates were linked among themselves, suggesting they may influence each other more than objectively determined behaviour.
Notably, perceived health emerged as the only correlate directly associated with distancing. While the association was small in magnitude (r = −0.12), its direction was unexpected when viewed through the lens of the Health Belief Model (HBM; Rosenstock, 1974). The HBM generally predicts that individuals perceiving themselves as having poorer health would view themselves as more susceptible to severe outcomes, thereby increasing their adherence to protective measures. However, our results indicated that individuals with better perceived health were slightly more likely to adhere to distancing guidelines (i.e. had fewer contacts). This suggests that in this specific naturalistic setting, general health perception may not have functioned as a proxy for COVID-19-specific vulnerability. This may be a limitation of using a single-item general health measure, which may not have captured the specific ‘perceived susceptibility’ required to trigger the HBM’s predictive pathways. Alternatively, a selection effect should also be considered. People in poor health probably chose to avoid a crowded indoor event like this during a pandemic, meaning our sample primarily represents the ‘healthy’ population. Our data support this: health ratings were highly skewed towards the top of the scale (mean ± SD = 5.87 ± 0.99, median = 6.0; on a 1–7 scale, where 1 = very bad and 7 = very good, with 75% of participants rating their health as good or very good (⩾5)). Together, these findings suggest that the relationship between health perceptions and distancing behaviour may be more complex than existing models anticipate. Moreover, self-reports did not predict objective distancing in this dataset, reinforcing evidence that self-reports may diverge from observed actions across domains (Klesges et al., 1995; Kormos and Gifford, 2014). Notably, a subset of participants were assigned to buzzer conditions in which the SDS actively alerted them in real time upon coming within 1.5 m of another visitor (Blanken et al., 2021; Tanis et al., 2021). Such participants had concrete, immediate feedback about their distancing behaviour during the event, and might therefore be expected to produce more accurate retrospective self-reports. However, the correlation between self-report and objective behaviour in the buzzer group alone remained non-significant (t(93) = −1.18, p = 0.24; r = −0.12), and a formal interaction test confirmed that the buzzer condition did not moderate the self-report/behaviour relationship (B = −0.09, SE = 0.80, t = −0.11, p = 0.91). This suggests the disconnect between self-reported and objectively measured behaviour remains even when participants had direct, real-time feedback about their own distancing.
These findings question the reliance on subjective measures such as self-reports and behavioural correlates for designing and evaluating public health interventions. Using such measures as proxies for actual adherence may lead to misguided policy decisions. For example, our data revealed no consistent inverse relationship between self-reported adherence and objectively measured physical distancing. Many individuals who rated themselves as highly compliant still made a high number of close contacts, suggesting a tendency to overestimate or misperceive their own behaviour. Similarly, emphasising changing attitudes or intentions may not reliably translate into observable behaviour. To ensure accurate assessment and effective policy, it is essential to incorporate objective behavioural measures alongside subjective reports.
The study contains some limitations. First, participants who completed the post-questionnaire differed significantly in age and gender from the full sample. Previous research has found that older adults report higher adherence to COVID-19 guidelines (Wolfe et al., 2021), suggesting that the post-questionnaire completers, who were significantly older than the full sample (M = 45.7 vs M = 42.7 years), may have been somewhat more prone to reporting high compliance. As a result, findings related to self-reported behaviour should be interpreted with some caution, as this subsample may not be fully representative of all attendees. Second, one reason for the lack of direct links between behavioural correlates and objectively determined behaviour in our network analysis may be that the assessments are taken from different time points and contexts: the behavioural correlates were assessed upon purchasing the tickets, which often took place days before the event itself and in different contexts. At the same time, if we lack sensitivity to pick up determinants of behaviours when assessed at different times or contexts, this would still call for caution in interpreting behavioural correlates as indicators of behavioural uptake of public health regulations. Future research should aim to replicate this framework in diverse, real-world settings and use repeated or real-time measures (e.g. smartphone tracking) to better align psychological assessments with behavioural outcomes. Incorporating longitudinal designs could further clarify how psychological correlates evolve over time and influence behaviour. Such work will enhance understanding of how health beliefs influence behaviour and improve the identification of effective points for public health interventions.
Our results demonstrate the necessity of including objective behavioural measures alongside self-reported data when evaluating public health interventions (Tanis et al., 2022). In current literature, interventions are commonly evaluated using correlates of behaviour and self-report measures alone (Armitage, 2007; Webb and Sheeran, 2006). Our findings suggest this is insufficient: self-reports and behavioural correlates both showed limited correspondence with objectively measured distancing, meaning evaluations that rely on them risk missing the full picture. This has implications beyond pandemic contexts, for behaviours like medication adherence, smoking cessation, and diet management. The gap between what people report and what they do remains a persistent challenge for public health.
Footnotes
Ethical considerations
Consent to participate
Consent for publication
Participants in the original study provided informed consent for the publication of their anonymised data. No additional consent was required for the use of de-identified data in this study.
Author contributions
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.