Abstract
Objectives:
The identification of factors which influence peoples’ preparation for health safety risks posed by natural and man-made disasters is a central concern in health education. Prior studies have generally approached this issue from either a cognitive or a social influence perspective, and have failed to recognise the increased importance of terrorism-related concerns in motivating health safety preparedness behaviour. The purpose of this study was to develop a unified social cognitive framework for understanding peoples’ preparations for health safety risks, focusing on terrorism-related cognitive and social influences.
Design/Method/Approach:
Participants in the National Survey of Disaster Experiences and Preparedness reported preparedness actions they had taken since 2001, their appraisals of terrorism-related threat and coping, and whether they knew others who had taken preparedness actions because of terrorism. Using a logistic binomial hurdle statistical model, number of actions taken was regressed on terrorism-related vulnerability, severity, response efficacy, self-efficacy, and informational social influence. Simultaneous models both of taking any action and of the number of actions taken were tested.
Findings:
After controlling for demographic variables, both taking any action and the number of preparedness actions taken were positively related to terrorism-related informational social influence, response efficacy, and self-efficacy; effects of terrorism-related vulnerability and severity appraisals were much smaller. Compared to cognitive factors, terrorism-related informational social influence had a substantially larger effect on taking any action, and a moderately larger effect on the number of actions taken.
Conclusion:
Terrorist-related informational norms were more salient than cognitive factors in influencing peoples’ preparation for health safety risks. Participants who knew someone who had taken one or more emergency preparedness actions because of terrorism were significantly more likely to take any preparedness action, and to take more preparedness actions, themselves. These findings are consequential in developing future educational initiatives.
Introduction
Recent natural and man-made events demonstrate how emergencies occur with little or no warning and devastating consequences, including the 2010 Haitian earthquake (318,000 lives lost, 1.3 million injuries, 97,294 destroyed houses and 188,383 damaged houses in Port-au-Prince: United States Geological Survey [USGS], 2010); the Joplin, Missouri, 2011 tornado (161 casualties, 1,000 injuries and US$2.8 billion damages: USGS, 2012); hurricane Katrina (986 casualties: Brunkard, 2008); and the 11 September 2001 terrorist attacks (nearly 3,000 casualties, and damages exceeding US$10 billion: Institute for the Analysis of Global Security, 2016). The impact of such events underscores the need to mitigate the harmful effects of unforeseen emergency events.
Taking a small number of simple preparatory actions (including developing an emergency plan, stockpiling food and water supplies, purchasing emergency-related items and duplicating important documents) has been linked to lives saved versus lost (Ablah et al., 2007; Balluz et al., 2000; Gheytanchi et al., 2007). Yet, despite intensive campaigns to motivate individuals to prepare for emergencies, only a minority of the US general population (30%–40%) is emergency prepared (Ablah et al., 2009; Eisenman et al., 2009a; Hamann, 2016; Kohn et al., 2012; Paton, 2003). One recent study found only 25% of the respondents reported that they were well prepared (DeBastiani et al., 2015). A study of differences in individual-level post-9/11 terrorism preparedness in Los Angeles County revealed that only 28% of the respondents had emergency supplies (Eisenman et al., 2006). Less than half (43%) of the respondents to a Federal Emergency Management Administration (FEMA) (2013) survey reported having a household emergency plan.
The persistent limited effectiveness of public campaigns to motivate people to prepare for emergencies has prompted calls for developing better models identifying the key factors underlying peoples’ emergency preparation motivations (Eisenman et al., 2009a). The current study drew on two predominant human behaviour paradigms – Social Influence Theory and Protection Motivation Theory (PMT) – to build and test a comprehensive model of influences on emergency preparedness, particularly in the context of heightened concerns about terrorism since 11 September 2001. PMT, by itself, has been found to be limited in explaining such differences (see the meta-analyses by Floyd et al., 2000; Milne et al., 2000). We hypothesised that terrorism-related cognitive and social influence factors would independently relate to preparedness behaviours. The validation of such a model would help to fill a gap in understanding why people do – and do not – prepare for health-related risks and to inform campaigns aimed at improving emergency preparedness.
Social Influence Theory
People are influenced by what they observe others to be doing (Asch, 1956; Berkowitz, 1972; Cialdini and Goldstein, 2004; Darley and Latane, 1968; Deutsch and Gerard, 1955; Milgram, 1963; Sherif, 1936; for reviews, see Wood, 2000), especially in novel, ambiguous or uncertain situations (Bearden and Etzel, 1982; Griskevicius et al., 2006; Hochbaum, 1954; Park and Lessig, 1977; see the meta-analysis by Rivis and Sheeran, 2003). Deutsch and Gerard (1955) distinguished between accepting information from others as evidence of reality (‘informational social influence’) and conforming to the expectations of others (‘normative social influence’). Although some contemporary theoretical models emphasise the precepts of normative social influence (notably, the Theory of Planned Behavior: Ajzen, 1991), informational social influence would seem more relevant to the study of preparedness for health safety risks associated with disasters where the situational uncertainty causes individuals to draw on others’ actions to determine their own behaviour (FEMA, 2007).
Informational social influence has been linked to individuals’ need to identify effective behaviour (Cialdini et al., 1990) in multiple areas, including water conservation (Onyenankeya et al., 2015; Richetin et al., 2016), transgression tolerance (Wang et al., 2015), shopping behaviour (Demarque et al., 2015), muscular endurance (Priebe and Spink, 2014) and speeding/dangerous driving (Forward, 2009). No study to date has systematically investigated the independent effect of informational social influence on disaster preparedness, particularly terrorist-related preparedness actions:
H1. Terrorism-related informational social influence will be independently related both to taking any emergency preparedness action and to the number of preparedness actions taken.
PMT
Cognitive appraisal models have been extensively employed to explain individual differences in emergency preparedness (Floyd et al., 2000). Based on expectancy-value theory (Edwards, 1954; Hovland et al., 1953) and derived as an elaboration of Leventhal’s (1971) Parallel Response Model, PMT (Rogers, 1975, 1983) was originally introduced to study the influence of fear messages on changes in attitudes, intentions and behaviour, but has since been adopted as a more general model of decision making in relation to emergency-related threats (Eisenman et al., 2009b; Maddux, 1993). PMT posits that individual motivation to prepare for the effects of a health safety risk, such as a hurricane, flood, earthquake, or act of terrorism, is based on the results of two cognitive appraisal processes: the individual’s appraised threat level (as represented by the joint effects of vulnerability and severity) and the individual’s appraised coping level (as represented by the joint effects of response efficacy and self-efficacy). Under this model, adoption of protective behaviours is more likely when both threat appraisal and coping appraisal are high.
In a further refinement of PMT, an extensive study by Bourque et al. (2013) of the vulnerability component suggests that risk perception may operate on preparedness indirectly, through knowledge, appraised efficacy, and milling behaviour. While PMT and related cognitive models such as the Extended Parallel Process Model (Witte and Allen, 2000) have generally found empirical support (Boer and Seydel, 1996; Floyd et al., 2000; Milne et al., 2000; Rogers and Prentice-Dunn, 1997), they generally omit the direct effects of the individual’s social environment on motivation to engage in preparedness behaviour:
H2. Terrorism-related appraisals of vulnerability, severity, response efficacy and self-efficacy will be independently related both to taking any emergency preparedness action and to the number of preparedness actions taken.
Hurdle models
Because emergency preparedness is often low, studies commonly operationalise preparedness as ‘taking no action’ versus ‘taking any action’, thereby losing valuable information inherent in count measures. Conversely, studies operationalising preparedness as a count of actions taken generally assume a linear relationship (thereby permitting infeasible expected counts) and normally distributed errors (unlikely given low averages, resulting in standard error underestimation and inflated Type I error).
To avoid these problems, we employed a hurdle model (Mullahy, 1986), which explicitly distinguishes between a process which focuses on whether or not any preparedness action was taken (‘preparedness initiation’) and a process which focuses on the number of actions taken, given that any action was taken (‘preparedness intensity’). It is plausible to conjecture that these two processes are distinct and that different weights are attached to social influence and cognitive appraisal components of each. Hurdle models permit the factors involved in the two processes to be simultaneously (but independently) estimated, and permit a more precise standard error estimation. To the best of our knowledge, hurdle models have not been employed in prior emergency preparedness research despite the potential of such models to distinguish between the two cognitive processes.
The present study
Drawing on Social Influence Theory and PMT, this study developed and tested an integrated hurdle model of individual preparedness for natural and man-made emergencies. We hypothesised that terrorism-related informational social influence, threat appraisal and coping appraisal would each reflect a positive and independent relationship with both preparedness initiation and preparedness intensity. Of particular interest is whether the respective weights associated with social influence, threat appraisal and coping appraisal factors differ, both for preparedness initiation and preparedness intensity.
This study contributes to health education in several ways. The study tests an integrated model of terrorism-related informational social influence and cognitive appraisal effects on emergency preparedness, and consequently compares the relative salience of social and cognitive appraisal factors. The novel adoption of a hurdle model permits the disaggregation of preparedness initiation (taking any action) and preparedness intensity (number of actions taken, given that one was taken). The study results are intended to inform the design of health risk preparedness educational campaigns.
Method
Participants
The data for this study were drawn from the National Survey of Disaster Experiences and Preparedness (NSDEP: Bourque, 2014), a national survey focused on individuals’ perceptions related to, and preparedness/mitigation actions for, natural and man-made disasters. The present study focused on the n = 2,317 respondents who provided data for all study variables. On average, respondents were 46 years old (standard deviation [SD] = 15.6). Women constituted 60% of the sample; 71% self-identified as White, 15% as Hispanic/Latino, 10% as Black or African American, and 4% as belonging to other racial/ethnic groups; 65% were married; and 49% reported having at least one child under the age of 18 living in the household.
Measures
Preparedness
Preparedness was measured by counting the number of ‘yes’ answers to four questions: Since September 11th, 2001 … have you developed emergency plans [evacuation, meeting places]; stockpiled supplies [food, water, antibiotics, etc.]; purchased things to make you safer [gas masks, duct tape, things to make your house safer, etc.]; and duplicated important documents [birth certificate, medication prescriptions, and passports]?)
The resulting scale ranged from 0 to 4.
Social influence
For each of the four preparedness items, respondents were asked whether they knew anyone who had taken that action ‘because of terrorism since September 11th, 2001’. The number of ‘yes’ responses to those questions formed the informational social measure.
Threat appraisal
Vulnerability
Perceived risk of terrorism was measured by averaging the responses to two questions: ‘How likely is it that a terrorism event like an explosion, biological, chemical, or radiological agents being released (in your community / that affects your home) will occur in the next 6 months?’ Answers were given on 5-point bipolar scales ranging from 0 = not at all likely to 4 = definitely will occur. Cronbach’s alpha was .78.
Severity
Perceived severity of terrorism was measured by averaging the responses to two questions: ‘If a terrorism event like an explosion, biological, chemical, or radiological agents being released (in your community / that affects your home) were to occur, how serious do you think the impacts would be?’ Answers were given on bipolar scales ranging from 0 = not at all serious to 4 = extremely serious. Cronbach’s alpha was .76.
Coping appraisal
Self-efficacy
Self-efficacy was measured by averaging the responses to three questions: ‘How sure are you that you could (effectively protect yourself from a future terrorist attack / respond quickly to a terrorist attack / recover effectively from a terrorist attack over the long-term)?’ Answers were given on 5-point bipolar scales ranging from 0 = not at all sure to 1 = extremely sure. Cronbach’s alpha was .63.
Response efficacy
Response efficacy was measured by averaging the answers to four questions: ‘How effective do you think (developing emergency plans / stockpiling supplies / purchasing things to make them safer / duplicating important documents) is for people dealing with terrorism?’ Answers were given on 5-point bipolar scales ranging from 0 = not at all effective to 4 = extremely effective. Cronbach’s alpha was .76.
Control variables
In view of their recognised relationship to emergency preparedness in prior studies (see Bourque et al., 2012), all multivariable analyses controlled for age, education, ethnicity, gender, income, home ownership and number of children under the age of 18 living in the home.
Procedure
Telephone interviews were completed on a statistically representative sample of 3,300 households between 13 April 2007 and 13 February 2008. Major metropolitan areas considered to be at higher risk of terrorism (viz. Los Angeles, New York and Washington, DC) were oversampled to allow comparisons with the rest of the continental USA, which were considered to be at lower risk of terrorism. The interviews were offered in English and Spanish, and a US$20 incentive was offered to encourage participation in the study.
Analytic procedure
As previously noted, assuming a normally distributed error term for emergency preparedness is probably unwarranted and can lead to seriously incorrect inferences (McCullagh and Nelder, 1989). Additionally, the often-made assumption that the zero counts and non-zero counts come from the same data-generating process is dubious and remains untested. The hurdle modelling approach introduced by Mullahy (1986) allows for separation of two processes (behaviour initiation and behaviour intensity), leading to a potentially more accurate model of emergency preparedness.
Hurdle models assume that a binomial probability governs the binary outcome of whether or not a count variable (such as the number of preparatory activities undertaken) has a 0 or positive realisation. If the realisation is positive, the ‘hurdle’ is crossed, and the conditional distribution of the non-zero observations is governed by a truncated-at-zero count data model. For unbounded count data, a truncated Poisson distribution or (in the case of overdispersion) truncated negative binomial distribution is assumed. Hurdle models are distinct from zero-inflated models, where zeros are assumed to be generated by both processes. For the current study, we employed a truncated binomial distribution in view of the relatively small maximum number of actions investigated in this study. We compared the standardised regression estimates of an identical set of cognitive appraisal and social psychological variables for both processes.
All analyses in this study were unweighted, given Korn and Graubard’s (1999) cautions regarding the use of weights in inferential analyses and in order to supplement the results published by others who conducted unweighted analyses on the same dataset (e.g. Bourque et al., 2012). All analyses used SAS Version 9.3 (SAS Institute Inc., 2011), with Proc NLMIXED to test the hurdle model.
Results
Model fit
To assess the hurdle model utility, the observed distribution of number of actions taken was compared to a standard binomial distribution. A far higher observed percentage (35%) of respondents engaged in none of the four possible preparedness actions than would be expected under a standard binomial model (19%) and large discrepancies were found for one and two actions as well (see Figure 1). Clearly, a binomial distribution does not describe the error structure well. In contrast, the discrepancies were greatly reduced by the hurdle model.
Observed, binomial and hurdle probabilities of number of actions taken to prepare for terrorism (binomial and hurdle probabilities are based on maximum likelihood parameter estimates of an intercept-only model).
Descriptive statistics
Preparedness
Average level of preparedness was low (Table 1). A total of 35% reported taking none of the four basic preparedness actions, 26% reported taking only one, 17% reported taking two, 13% reported taking three, and 9% reported taking all four, resulting in an average of 1.35 (SD = 1.31) actions taken. In order, 39% reported duplicating important documents, 38% reported stockpiling supplies, 34% reported developing emergency plans, and 24% reported purchasing things to become safer. Co-occurrence rates were higher than would be expected if the actions were independently taken (which is a basic tenet of binomially distributed random variables). Among those who reported duplicating important documents, 67% reported also stockpiling, compared to 23% reporting stockpiling among those who did not duplicate important documents (phi = .43, p < .0001). Consistent with the expectations under a hurdle model, this prevalence difference drops markedly among those who took at least one action (phi = .19, p < .0001).
Descriptive statistics and correlations for study variables.
Main diagonal contains reliabilities.
p < .05; **p < .01; ***p < .001.
Social influence
On average, 1.36 (SD = 1.31) preparedness actions were thought to have been taken due to terrorism by individuals known to respondents. A total of 41% of the sample reported knowing someone who had stockpiled supplies, 38% reported knowing someone who had developed (or was developing) emergency plans, 29% reported knowing someone who had duplicated important documents, and 27% reported knowing someone who had purchased items to make them safer.
Threat appraisal
Most respondents appraised as low the likelihood of a home- or community-affecting terrorist event within the subsequent 6 months (vulnerability: M = 0.58, SD = 0.86). The likelihood of a terrorist event affecting their home (community) in the subsequent 6 months was appraised as ‘extremely unlikely’ by 68% (61%). Less than 3% appraised their vulnerability as ‘extremely likely’.
In contrast, most respondents appraised the severity of a home-affecting or community-affecting terrorist event should one occur (severity) as high (M = 3.13, SD = 1.05). The impact of such an event affecting their home (community) was appraised as ‘extremely serious’ by 57% (52%). Less than 6% of the respondents appraised the severity as ‘not at all serious’.
Coping appraisal
On average, respondents rated themselves as somewhat unsure that they could protect themselves/respond quickly/recover long-term from a future terrorist attack (i.e. their self-efficacy), with a mean of 1.59 (SD = 0.96). A total of 40% responded that they were ‘not at all sure’ that they could protect themselves from a future terrorist attack, 22% reported being not at all sure that they could respond quickly, and 22% reported that they were not at all sure that they could recover effectively from a terrorist attack over the long term. Only 3% of the sample reported being ‘extremely sure’ that they could protect themselves in the event of a terrorist attack; 15% reported being extremely sure that they could quickly respond or recover long-term from a terrorist attack.
Average level of response efficacy was somewhat higher (M = 2.33, SD = 1.05). Duplicating documents was reported as ‘extremely effective’ for dealing with terrorism by 35%, developing emergency plans by 33%, stockpiling by 29%, and purchasing other items by 23%. A total of 20% viewed at least one of these actions as ‘not at all effective’.
Zero-order correlations
Preparedness was moderately correlated with informational social influence (r = .47, p < .001), and less strongly with the cognitive appraisal measures (ranging from r = .04 for severity to r = .32 for response efficacy, all p < .05). Small intercorrelations among the four cognitive appraisal components were found, ranging between r = .02 and r = .23, and between the informational social influence and cognitive appraisal components, ranging between r = .05 and r = .18.
Social cognitive model
Table 2 presents the results of the multivariable social cognitive binomial hurdle model. In order to compare the relative strengths of the effects, all social and cognitive variables were standardised to a variance of 1.0.
Parameter estimates for the social cognitive hurdle model.
D/S/W: Divorced/Separated/Widowed.
p < .05; **p < .01; ***p < .001.
For both regression analyses (preparedness initiation and preparedness intensity), near-zero effects were found for both vulnerability (respectively, b = −.043, p > .05; b = .075, p < .05) and severity (respectively, b = −.049, p > .05; b = −.008, p > .05). In contrast, considerably stronger statistically significant preparedness initiation effects were found for self-efficacy (b = .276, p < .001), response efficacy (b = .505, p < .001) and informational social influence (b = .937, p < .001). Similarly, statistically significant preparedness intensity effects were found for self-efficacy (b = .216, p < .001), response efficacy (b = .385, p < .001) and informational social influence (b = .571, p < .001).
In view of the consistently stronger relationships exhibited by informational social influence compared to the cognitive appraisal factors, we tested a series of additional models for each of the two outcome variables (initiation and intensity), where the parameter estimate of each cognitive appraisal variable was, in turn, constrained to equality with the informational social influence parameter estimate. For all tests associated with each outcome variable, the social influence factor was significantly stronger than the cognitive appraisal factor. Further probing revealed the magnitude of informational social influence across the two regressions: a one standard deviation difference in informational social influence was associated with a 155% increase in the odds of preparedness initiation and a 77% increase in the odds of engaging in all four preparedness behaviours. These results suggest that, compared to cognitive appraisal effects, informational social influence may be significantly more salient both for engagement in any preparedness activity and for the number of preparedness activities undertaken, and may be more salient for preparedness initiation than it is for preparedness intensity.
Discussion
No study to date has systematically investigated the collective effects of terrorist-related cognitive appraisal factors and informational social influence on preparedness health safety risks such as floods, earthquakes and terrorist attacks. A primary objective of this research was to test a unified social cognitive model of individual preparedness for health safety risks based on the informational social influence component of Social Influence Theory and the cognitive appraisal components of PMT. We hypothesised that terrorist-related social norms and cognitive appraisal would independently influence two outcomes associated with preparedness behaviour: (1) taking any of a set of four possible actions in preparation for a health safety risk (preparedness initiation) and (2) the number of precautionary actions actually taken (preparedness intensity). Of particular interest was whether either cognitive appraisal or social factors carry greater salience with regard to either of the two outcomes.
For both outcome variables, terrorist-related informational social influence exhibited a significantly stronger independent relationship with preparedness for a disaster than any of the terrorist-related cognitive appraisal factors, after controlling for potentially confounding effects. Also for both outcome variables, the two coping appraisal factors (self-efficacy and response efficacy) exhibited significantly stronger independent relationships with preparedness than either of the two threat appraisal factors (vulnerability and severity), which were found to exhibit much weaker independent effects. Additionally, the informational social influence effect was found to be more salient for preparedness initiation than for preparedness intensity.
The independent effects of social influence and coping appraisal underscore the importance of developing emergency preparedness models that draw from multiple theoretical perspectives. These results comport with (respectively) the ‘social cues’ and ‘protection action perception’ components of Lindell and Perry’s (2012) comprehensive Protective Action Decision Model. The significantly stronger independent effects of informational social influence are of particular importance, given how little attention this factor has received in past studies of preparedness in general, and preparedness for terrorism in particular.
The weak relationships between preparedness and both vulnerability and severity, at both the bivariate and multivariable levels, suggest that these factors may not be strongly salient in explaining differences in emergency preparedness. While the weak bivariate relationships might otherwise be attributed to suppressor effects, the weak multivariable relationships do not lend themselves to an easy alternative explanation. Subject to replication by others, these results suggest that, ceteris paribus, threat appraisal factors exert minimal independent influence on disaster preparedness compared to social influence effects and coping appraisal effects.
The rationale for employing a hurdle model was twofold: (1) obtaining more refined standard error estimates for use in the statistical tests and (2) differentiation of the preparedness initiation process from the preparedness intensity process. Both rationales were supported in the current study: the error term conformed much more properly to a hurdle model specification than to a binomial model specification and social influence was linked more strongly to preparedness initiation than to preparedness intensity. For both preparedness initiation and preparedness intensity, informational social influence was found to have a significantly greater independent effect than coping appraisal, which (in turn) was found to have a significantly greater independent effect than threat appraisal.
These results come with several caveats. The results are based on self-reported measures from a cross-sectional survey, limiting causality attributions: it is possible, for example, that appraisal of self-efficacy occurs after an individual has engaged in (and learned from) a particular preparedness behaviour. A complete implementation of PMT requires measuring costs and rewards. The study focused on only informational social influence and not normative social influence. The four preparedness behaviours examined here represent a subset of possible actions taken to prepare for health safety risks. Additionally, this study focused on terrorism-related cognitive and social influences, which may not generalise to other types of influences.
Conclusion
The results of this study suggest several ways in which contemporary models of preparedness for health safety risks might be improved. First, the independent effects of terrorism-related social influence and cognitive factors found in this study underscore the need to build and test models of health safety risk mitigation that draw from multiple perspectives (e.g. social influence and cognitive appraisal perspectives, comporting respectively with the ‘social cues’ and ‘protection action perceptions’ components of Lindell and Perry’s 2012 Protective Action Decision Model). Models that focus exclusively on cognitive appraisal factors or exclusively on social factors fail to adequately address the complexity of human behaviour. Second, despite message manipulation research spanning multiple decades which demonstrates the potential utility of threat appraisal processes to motivate individuals (e.g. Rogers, 1983), the results of the present field study suggest that naturally occurring differences in perceived threat may not be as salient in preparedness choices as differences in the perceived behaviour of others and perceived efficacy. Third, these results suggest the importance of distinguishing between processes involved in taking any action and processes involved in choosing the number of actions to take. This study’s findings that social influence may be more salient than cognitive appraisal factors in predicting preparedness for disasters, and particularly salient in preparedness initiation, are of potential value both to practitioners developing effective intervention models and to theorists designing comprehensive models of individual differences in health safety preparedness.
Footnotes
Acknowledgements
We would like to thank Philip J Moore for his invaluable comments on an earlier draft of this article.
Funding
The authors received no financial support for the research, authorship and/or publication of this article.