Background
Understanding relationships between psychological state and protective behaviors during respiratory infectious disease epidemics (RIDEs) can inform risk communication and interventions addressing behavior change [
1,
2]. Studies of behavioral change during RIDEs usually assess risk perception as an affect-neutral cognitive (“cognitive”) process, commonly using measures such as perceived personal probability of infection or perceived severity of the illness [
3‐
5], or as a more affect-active process, by assessing worry and anxiety [
6‐
10]. The latter are frequently referred to as “affective” dimensions of risk, though worry is often considered a cognitive dimension of anxiety [
11]. The dual-process theory proposes that responses to external stimuli involve two different processing systems, one being deliberate, slow and rule-based, the other being experiential, quick and intuitive [
12]. These two systems may reflect distinct response pathways to risk: risk-as-analysis (cognitive estimates) and risk-as-feeling (affective estimates) [
13]. The affect heuristics and risk-as-feeling hypotheses imply that affect quickly and more efficiently guides cognitive risk analysis and behavior [
13‐
15]. Previous studies found that in the RIDE situation when personal threat is highly uncertain, affective measures of risk more powerfully predict protective behavior uptake than do cognitive measures [
6,
10]. Therefore, both cognitive and affective components of risk appear to be relevant to understanding RIDEs-related population behavior [
1].
In the early epidemic stage when uncertainty about the epidemic characteristics, treatment and prevention is higher, affective responses may be optimal for guiding behavioral change [
6,
9,
10] but cognitive risk responses should increasingly drive behavior as the epidemic evolves. We term these “psycho-behavioural” associations. Given this, the question arises: should studies or assessments done early in the epidemic emphasize affect-based assessments of risk, whereas those performed later in the epidemic emphasize cognitive-based measures, in order to optimally predict behaviors? Otherwise, it is possible that research conducted in different stages of an epidemic may observe different strengths for the same psycho-behavioral association and misattribute these. Observed variation in the strength of specific psycho-behavioral associations across an epidemic introduces avoidable measurement error in the target cognitive/affective measure which will subsequently influence its association with behavioral change, reducing the apparent reliability of risk assessments as predictors of behavior change during RIDEs. This raises questions about whether the same or similar associations would be repeatedly identified in surveys conducted in different epidemic periods within the same population. However, very few studies appear to have examined the consistency of these psycho-behavioral associations across different RIDE stages [
6]. We therefore performed secondary data analyses for data collected in a series of ten consecutive cross-sectional surveys spanning the epidemic wave of 2009 pandemic influenza A/H1N1 in Hong Kong [
16]. The objectives of this study were to compare the strength and stability of associations between affective and cognitive measures of risk and the adoption of RIDE-related health protective behaviors. This was assessed by comparing the associations between health protective behaviors against A/H1N1 and different cognitive/affective measures of risk used for each of the ten cross-sectional surveys.
Most psycho-behavioral studies of new communicable respiratory disease outbreaks were rapidly implemented [
2]. Consequently, many used unrefined questionnaires, with several suffering from minimal theoretical support for the inclusion of specific psychological variables, items of limited utility in understanding behavioral change or items that may have posed task difficulty for respondents [
1,
2], and multiple items, which increase interview load, thereby reducing interview efficiency and the accuracy of results. To inform future item selection, we therefore also sought to assess the difficulty respondents faced in answering different question measures of risk perception. This was done by examining proportions of missing data for different psychological measures as an indirect reflection of task difficulty.
Our null hypotheses were:
1.
Cognitive and affective measures of risk will not differ in terms of stability of association with adoption of protective behaviors across the ten cross-sectional surveys;
2.
For the same associations measured at different epidemic periods, strength of associations between affective/cognitive measures and adoption of health protective behaviors will not decline/increase across epidemic stage;
3.
There will be no difference in proportions of missing data for cognitive estimation items such as estimates of the likelihood of contracting influenza infection than other risk assessment formats reflecting no differences in the difficulties posed to respondents by such items [
17,
18].
Discussion and conclusions
Our findings were mostly consistent with those hypothesized and the null hypotheses were largely rejected. The main finding is that affective measures of risk perception generally had stronger associations with reported adoption of health protective behaviors during the A/H1N1 pandemic than did cognitive measures. This finding is consistent with those from other studies conducted during both SARS [
6] and pandemic A/H1N1 [
10,
24], suggesting that affective components contribute significantly to adoption of protective behaviors in response to the threat during epidemics over and above simpler cognitive risk estimates. While previous studies were mainly conducted in early epidemic periods [
10,
24], this study examined affective-behavioral associations across the entire epidemic wave of A/H1N1 in Hong Kong and found that the association between affect-loaded risk measures and adoption of protective behaviors were consistently strong and positive across different epidemic periods.
Studies of the anxiety-behavior association throughout the SARS epidemic found consistently significant and positive associations during the early epidemic phase surveys but mostly non-significant associations in late epidemic phase surveys [
6]. The present study did not duplicate this pattern for any of the four affective measures. Reported anxiety level was inconsistently associated with adoption of health protective behaviors in these 10 surveys. One possible reason could be that the measure we used assessed general anxiety only rather than anxiety specific for A/H1N1. Furthermore, overall reported state anxiety levels remained quite stable and consistently low throughout the A/H1N1 epidemic [
16], indicating a floor effect, suggesting that a low level of anxiety has little effect on these behaviors. Other affective measures including anticipated worry, experienced worry and current worry generally involve less intense affective components compared with anxiety and thereby are more likely to covary with behavioral change. In particular, our study found that experienced worry and current worry seemed to have stronger associations with adoption of protective behaviors than did anticipated worry. One possible reason could be that the actual affective experience or associated processing may be more strongly associated with behavioral change than its anticipation, which may be subject to forecasting errors [
23].
Cognitive risk assessments, in particular perceived susceptibility to A/H1N1 (either absolute or relative susceptibility) had weak associations with adoption of protective behaviors. This suggests that cognitive-behavioral models such as the Health Belief Model [
25,
26] that rely primarily on purely cognitive estimates of risk to predict behavioral change should perform relatively more poorly at predicting the adoption of protective behaviors during RIDEs. Cognitive-behavioral models generally assume rational processing of external information to inform action. However, during RIDEs particularly in the early stages, uncertainty is usually widespread and poses high [
9] or ambiguous personal threat. Consequently, people may face difficulties when attempting to quantify the probabilities of their risk of acquiring the infection and the severity of associated disease. Whether it is threat ambiguity, task difficulty in determining risk magnitude or a primary affective response that modifies cognition, that leads to affect-related measures dominating remains unclear. This study found that the proportions of missing data for purer cognitive risk perception measures, particularly perceived absolute/relative susceptibility to A/H1N1 were greater than for affect-loaded measures, suggesting that respondents may face greater task difficulties in comprehension and/or responses to such questions under epidemic circumstances. Further study is needed to confirm the extent of this effect.
Perceived relative susceptibility seemed to have stronger associations with adoption of protective behaviors than perceived absolute susceptibility. Perceived susceptibility measured in this relative way involves social comparison and accommodates the influences of optimistic bias [
27] and therefore probably involves more cognitive processing. More cognitive processing is associated with greater risk estimates and psychological distress [
28]. This might account for the more substantial associations with behavioral change than seen for simple personal risk estimates.
Associations between cognitive risk perception measures and protective action were quite inconsistent across the ten selected surveys in this study. Previous reviews concluded associations between cognitive risk perception and adoption of protective behaviors during RIDEs were inconsistent [
29]. Our evidence suggests a major reason for this inconsistency lies in these studies being conducted in different epidemic stages [
6,
30,
31]. Our hypothesis was that cognitive factors were more important in changing human behaviors in the later epidemic stage when people had more knowledge and less uncertainty about the threat. This study found that the associations between perceived A/H1N1 severity relative to SARS and adoption of each of the three protective behaviors became significantly and consistently positive starting from survey 10 after the A/H1N1 case confirmations had peaked, consistent with our hypothesis. However, this pattern of associations was not found for perceived susceptibility.
Perceived A/H1N1 infectivity relative to seasonal flu, though not measured before survey S9 had weaker associations with adoption of health protective behaviors, than did perceived A/H1N1 severity relative to SARS in each survey and overall. However, these two measures assessed different aspects of A/H1N1 severity with the former focused on the infectivity rate of A/H1N1 while the latter may primarily focus on the fatality rate of A/H1N1. Further study is needed to confirm which aspects of disease severity could be more important in motivating behavior change.
Study limitations include the serial cross-sectional design and thereby reverse-causality remains a possible explanation. Nonetheless, it is difficult to think of plausible mechanisms whereby, for example, disinfecting one’s home will lead to greater worry regarding infection. Alternatively, the associations could be spurious but this is unlikely given the consistent pattern of the associations in 10 separate samples. It therefore seems most likely that the protective behaviors are consequential on the risk perceptions, and not vice versa. Examining psycho-behavioral associations using longitudinal data during RIDEs is difficult due to their often-rapid evolution and the short lead-time compared to the need to obtain and retain large cohorts for follow-up surveys. Conducting a series of consecutive cross-sectional surveys to investigate the psycho-behavioral associations is a better option than using a single cross-sectional survey. There may be concerns about the generalizability of our findings to more severe RIDEs. For example, during the initial phase of the SARS epidemic, population state anxiety regarding the epidemic was much higher and thereby had strong association with protective behavioral change [
6]. However, SARS was the first of the new wave of RIDEs, and a degree of risk fatigue may have subsequently set in. Considering the common situation during RIDEs, we believe that most of the findings in this study could be applicable in other RIDEs. Finally, because all data were self-reported the results may reflect social desirability bias.
This study raises important implications for future respiratory communicable disease-related psycho-behavioral research and public health interventions. First, affective responses improve understanding of behavioral responses throughout different RIDE periods and must form part of measures in relevant studies. However, intense but nonspecific affect such as generalized state anxiety is probably less useful for understanding public behavioral responses during most epidemics where perceived milder threat fails to arouse such affect. Less intense, specific affective responses to a identifiable, if uncertain threat that currently activates or has in the past activated worry may be more likely to show strong and consistent effects on behavioral change across different epidemic periods. Second, cognitive risk estimates during the early epidemic stage may be poor at predicting human behavioral change and present task difficulties to respondents. However, cognitive risk estimates may inform individual behavioral change later in the RIDE epidemic trajectory and should be included in studies conducted during these phases. Relative measures of perceived susceptibility appear superior to perceived absolute susceptibility in predicting behavioral change and thereby are preferable where questionnaire brevity is an issue. From a public health perspective, recognizing that the public may not show expected “rational” behaviors during RIDEs is important. Therefore, risk probabilities alone are unlikely to be sufficient to motivate protective behaviors. What affective strategies to use to best motivate behavioral change awaits clarification.
Acknowledgment
This work received financial support from the Research Fund for the Control of Infectious Disease, Food and Health Bureau, Government of the Hong Kong Special Administrative Region (SAR) (grant number PHE-01), the Harvard Center for Communicable Disease Dynamics from the United States National Institutes of Health (NIH) Models of Infectious Disease Agent Study program (grant number 1 U54 GM088558), and the Area of Excellence Scheme of the Hong Kong University Grants Committee (grant number AoE/M-12/06). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We thank Public Opinion Poll, The University of Hong Kong, for conducting the telephone interview.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
QL participated in the study design, analyzed the data, interpreted the data and drafted the manuscript. BJC supervised the research, contributed to study design, data interpretation and amended the manuscript. WWTL contributed to study design, data interpretation and amended the manuscript. DWMN contributed to questionnaire design, coordinated data collection and amended the manuscript. RF conceived of the study, designed the questionnaire, interpreted data and amended the manuscript. All authors read and approved the final manuscript.