How Do Perceptions of Risk Communicator Attributes Affect Emergency Response? An Examination of a Water Contamination Emergency in Boston, USA

A water main break that contaminated the Boston area's water distribution system prompted a four‐day “boil water” order. To understand risk communication during this incident, 600 randomly sampled residents were mailed questionnaires, yielding 110 valid responses. This article describes how perceptions of different social stakeholders influenced whether respondents complied with the Protective Action Recommendation—PAR (i.e., drank boiled water), took alternative protective actions (i.e., drank bottled water or/and self‐chlorinated water), or ignored the threat (i.e., continued to drink untreated tap water). Respondents perceived technical authorities (i.e., water utility, public health, and emergency management) to be higher on three social influence attributes (hazard expertize, trustworthiness, and protection responsibility) than public (i.e., news media, elected officials) and private (i.e., self/family, peers, and personal physicians) intermediate sources. Furthermore, respondents were most likely to comply with the PAR if they perceived authorities and public intermediates to be high on all three attributes and if they had larger households and lower income. Contrarily, they were more likely to take alternative actions if they were younger and had higher levels of income, risk perception, and emergency preparedness. These results underscore the need for technical authorities to develop credibility with their potential audiences before a crisis occurs.


Justification for This Study
In recent years, traditional hydrology studies have been criticized for their overly narrow focus on natural processes such as water quality, and failure to integrate social, cultural, political and economic values and processes, that shape water governance issues (Sivapalan et al., 2014). Hence, new socio-hydrological frameworks like the integrated Structure, Actors, and Water framework (Haeffner et al., 2018, pg. 665) have been developed and used to study perceptions of city leaders and the public at large (from Utah constituencies) on key water issues. Findings suggest these two groups differed in their views dramatically. While constituents were concerned about future water supply and price, leaders were concerned with deteriorating water infrastructure. They suggested that 10.1029/2021WR030669 3 of 23 these differences in the perceptions, information, and experiences of individuals and organizational actors need to be understood in light of how they create impediments to a more sustainable water management system (Pg. 665).
In their research on the relationship between consumers' risk perceptions of arsenic exposure in tap water and the purchase of bottled water, Jakus et al., (2009) found that people systematically underestimated the "true risk" which was based on scientific estimates as a benchmark. They concluded that their population was not purchasing enough bottled water and suggested that this is a key finding. Policy makers need to decide if "consumer choice based on existing perceived risks is acceptable from a public perspective or if it is in the public interest to provide more information on the risks of tap water consumption and the choices available to customers" (pg.7). Their findings also revealed that more easily recognizable water quality characteristics like taste, smell had greater influence than the perceived risk in causing people to buy bottled water, However, all else being equal, those with greater risk perceptions were willing to spend more money on bottled water than those with lower perceived risk. Price et al. (2015) tested attributes of water message structure and content (i.e., for potable recycled water) and found that complex messages and those that communicated about risk were most effective in positively affecting risk perceptions but not necessarily greater support for recycled water use. Risk information only influenced the risk perception of people residing in the area where the issue was more relevant. They highlighted the importance of understanding people's motivations to process information and suggested that repeated exposure to specific types of information would be useful. However, they called for finding ways "to inoculate people against counter claims of opposition groups" (pg. 2185).
In the past, the ultimate receivers of threat information (i.e., those in the risk area) were limited to one-to-one communications such as telephone and face-to-face communication to engage in the collective sensemaking process known as milling (Wood et al., 2018). Now, when people receive information from various public and private sector entities, they have access to social media such as Twitter that allow a single person to broadcast simultaneously to many others. This makes it possible for uninformed or malicious actors to have a much greater influence on the responses of the risk area population (Gao et al., 2020;National Research Council, 1989). Hence, scholars call for distinguishing the roles and functions between public and private intermediaries in the risk communication process (Kousky & Kunreuther, 2017;Steinberg et al., 2016).
In summary, the current study of water contamination incidents in Boston is unique as it leverages theories and findings from disaster sciences, specifically, the social-psychological theory of Protective Action Decision Model (PADM- Lindell, 2018;Lindell & Perry, 2004, 2012 to understand how individuals' perceptions of messages from community stakeholders (public and private influencers) affect their risk perceptions and thereby their decisions to comply with official PARs, or take an alternative protective actions, or take no action at all. The findings can guide policies to mitigate conflicts in messaging and reduce risks from future water contamination incidents, as well as to understand what water utility and emergency management officials can do differently to increase compliance with official PARs. It will also illustrate how individuals' demographic characteristics influence their preferences for bottled water over boiled water (the PAR) and why policy makers and urban hydrologists must consider a socio-hydrological perspective (Sivapalan et al., 2012) while making investments in water infrastructure and innovative designs for ensuring water quantity and quality, respectively.
Against this background, this article examines what attributes of information sources influenced the actions that residents took after receiving advisories regarding the water contamination and PAR. Specifically, it identifies eight types of stakeholders who served as risk communicators and classifies them into three categories, namely authorities (water utility, public health, emergency management, elected officials), public intermediate sources (news media), and private intermediate sources (risk area residents and their families, peers, and personal physicians). It also examines how these stakeholders' three key attributes-hazard expertize, trustworthiness, and protection responsibility-affected people's decisions to comply with the PAR (i.e., boil water), take alternative protective actions (i.e., drink bottled or self-chlorinated water), or ignore the threat. Additionally, the relationships of risk perception, preparedness, and demographic characteristics are explored as other predictors of households' responses to the water contamination threat.
The remainder of this article is divided into five sections. Section 2 discusses the study's theoretical foundation-the Protective Action Decision Model (PADM) and Communication Network Model (CNM)-and reviews research on the influence of community stakeholders' attributes on protective actions. The section concludes with a list of research objectives along with research hypotheses and research questions that guide this study. Section 3 provides a description of the questionnaire items, sampling procedure, and data collection procedure. Section 4 presents the survey results and Section 5 discusses their theoretical and practical implications, as well as the study's limitations. Finally, Section 6 presents the study's conclusions.

Theories Framing Risk Communication
The Classical Persuasion Model proposed by Lasswell (1948) identifies five principal components of risk communication, namely, who (source), says what (message), in what medium (channel), to whom (receiver), and with what effect (effect). Further, the Shannon-Weaver model (Shannon & Weaver, 1949) focused attention on the linear relationship between message framing and transmission, from an information source to a receiver through a transmitter or a channel (Al-Fedaghi, 2012). Riley and Riley (1965) modified the Shannon-Weaver model by positing that mass communication occurs within a social system, between communicators and receivers, both of which are part of larger primary groups and are influenced by those groups. Thus, they viewed communication as influenced by multiple entities, with communication flowing between and within those social groups. Consistent with this framework, Katz and Lazarsfeld (1955) proposed the Two-Step Flow of Communication Model that highlights the importance of intermediate sources such as opinion leaders in disseminating a message from a communicator to receivers. Lindell and Perry (2004) integrated these perspectives into the PADM, which describes the way that people process threat information and choose disaster responses. One important aspect of the PADM involves people's perceptions of information sources in terms of hazard expertize, trustworthiness in providing accurate information, and responsibility for protecting those at risk. In addition, as indicated in Figure 1, the CNM posits that an original source such as a WDS operator, can transmit messages directly to those at risk (Channel A) and to intermediate sources such as the news media (Channel B) who relay the messages to those at risk (Channel C) using a one-to-many broadcast process. In addition, there is a one-to-one contagion process in which message recipients exchange information with each other (Channels D and E), leaving very few isolates who fail to receive a warning (Lindell, 2018;Rogers & Sorensen, 1988).
The message diffusion process relies on social connections in which ultimate receivers-including oneself and one's family, friends, relatives, neighbors, and coworkers-communicate information to each other about hazards and protective actions. Despite extensive research on the role of informal warning sources (e.g., Lindell et al., 2019), few studies based on the PADM and CNM have addressed the characteristics of these sources that influence people's warning responses.

Influence of Stakeholder Attributes
The impacts of communicator attributes in persuasion have a long history of study (Gass & Seiter, 2014) and, specifically, have been the subject of research on the effects of risk communicators' attributes on PAR compliance (Heath et al., 2018;Martin-Shields, 2019;Wang et al., 2018). Consistent with Petty and Cacioppo's (1986) Elaboration Likelihood Model, scholars have found that communicator attributes can have direct or indirect effects on an individual's decision to take protective actions. A direct effect occurs if perceptions of the communicator's attributes directly influence the adoption of protective actions, whereas an indirect effect occurs if perceptions of the communicator's attributes alter how people interpret the communicator's message (i.e., perceive the risk), which in turn affects their decision to take protective actions (Arlikatti et al., 2007(Arlikatti et al., , 2014Gladwin et al., 2001). This causal relationship may vary depending on the hazard, the information sources, and the situation. During high-stress situations, for example, people may rely on a heuristic process and focus more on an information source's characteristics than the message content itself (Kahlor et al., 2003;Reynolds, 2011).
Following French and Raven (1959), perceptions of stakeholders' expertize can be understood as beliefs about their possession of essential information about a situation (e.g., the concentration of a contaminant in parts per million) and about cause-and-effect relationships relevant to that situation (e.g., the probability of adverse health effects, given that contaminant concentration). People generally attribute higher levels of expertize to authorities and news media due to the belief that these stakeholders have relevant educational credentials and experience (Arlikatti et al., 2007;Lindell & Perry, 1992;Murphy et al., 2018;Perry & Lindell, 1990;Sager, 1994;Taibah, et al., 2017). Other studies have found that optimistic bias causes people to rate themselves as having higher expertize than their peers (Hatfield & Job, 2001;Klar & Ayal, 2004;Weinstein, 1989). Nevertheless, people tend to rate their own expertize somewhat lower than authorities and the news media (Arlikatti et al., 2007).
Perceptions of trustworthiness, a source's willingness to provide accurate information, are built on personal admiration (Eagly & Chaiken, 1998;French & Raven, 1959;Raven, 2008), as well as familiarity (Perry & Lindell, 1990), so, according to the Onion Theory (see, for example, Wu et al., 2020), people tend to trust those who are closer to them (Godschalk et al., 1994). Among all stakeholders, peers often receive the highest ratings of trustworthiness due to shared life experiences (Arlikatti et al., 2007;McGuire, 1985;Quarantelli, 1960;Taibah et al., 2017). Even though people rate their peers as less knowledgeable than themselves about a hazard, their high ratings of trustworthiness lead people consult those peers to confirm a warning (Wood et al., 2018) and sometimes heed peers' recommendations rather than those of authorities (Arlikatti et al., 2014).
Ratings of expertize and trustworthiness have been found to be strongly related (Arlikatti et al., 2007). Indeed, some studies have noted that a stakeholder's perceived expertize and trustworthiness combine to produce an overall perception of credibility (McCallum et al., 1991;Wei et al., 2018). Stakeholders perceived as credible can influence information acceptance and shape people's protective action decisions (Gauntlett et al., 2019;Lindell & Perry, 2012;Mileti & Peek, 2000). Conversely, studies highlighting the failed communication during Hurricane Katrina in 2005 found that messages received from non-credible sources were ineffective (Cole & Fellows, 2008). Hence it is important for risk communicators to develop credibility with their audiences during the continuing hazard (Lindell & Perry, 2004) or pre-crisis (Seeger, 2006) phase. Understanding how individuals evaluate stakeholder credibility can also assist risk communicators in tailoring messages and improving their perceived credibility among all population segments (Taibah & Arlikatti, 2015;Taibah et al., 2017).
Responsibility is a consequence of the rights and duties of a position within a social network (Eagly & Chaiken, 1998;French & Raven, 1959;Raven, 2008). Some studies have found that people believe in personal responsibility when it comes to protective actions (Garcia, 1989;Grothmann & Reusswig, 2006;Mulilis & Duval, 1997). However, other studies have found that people often believe authorities are responsible for protecting the public during an emergency (Arlikatti, et al., 2007;Giroux et al., 2009;Terpstra & Gutteling, 2008) because they are expected to plan and prepare for such events (Basolo et al., 2009). An explanation for the apparent inconsistency in these results is that people are more likely to attribute protection responsibility to government if they do not know any protective actions to take, if they consider the available protective actions to be insufficiently effective, or if those protective actions require resources that they lack (Lindell & Perry, 2000a, 2000b. Ultimately, people who believe preparedness is an individual's responsibility are more likely to take protective actions (Garcia, 1989;Lindell & Whitney, 2000).

Influence of Receiver Attributes
Some scholars have found that consumers' demographic characteristics can be linked to the purchase of bottled water (Merkle et al., 2012;Triplett et al., 2019). Affluent households with young children, and greater levels of education, those on a public water system, and those having concerns related to taste, smell and clarity were more likely to purchase bottled water while older adults were less likely than younger to consume bottled water (Jakus et al., 2009). In further trying to understand the willingness to pay for improvements to water systems, Genius and Tsagrakis (2006), found that both experiences with water shortages and drinking water from sources other than the tap were important determinants of Greek city residents' willingness to pay for a fully reliable water supply. Those not affected by water scarcity and already drinking tap water had a smaller willingness to pay, while those relying on bottled water had a higher willingness to pay. Willingness to pay increased with age up to a certain point (50 yr) and decreased, possibly because of the level of earnings going down and no young children in the household (pg. 8).
However, Tanellari et al. (2015) found that Washington DC suburban consumers' willingness to pay for water utility improvement programs was negatively affected by the cost of the proposed improvement. When asked which of three programs-water quality improvement, pinhole leak damage insurance, or public infrastructure upgrade, 44% respondents were not willing to pay into any program, but the highest support was for public infrastructure improvements.

Research Objectives, Questions, and Hypotheses
Objective 1: To examine how respondents rate each stakeholder's social influence in terms of expertize, trustworthiness, and protection responsibility.
1. RH1: There will be significant differences among the mean ratings of the stakeholders on the three social influence attributes (expertize, trustworthiness, and protection responsibility) 2. RH2: Stakeholders' attribute profiles on expertize and trustworthiness will be much more similar to each other than either one is to protection responsibility 3. RH3: Mean ratings and interrater agreement on hazard expertize will be highest for authorities (i.e., water utility, public health, emergency management, and elected officials), next highest for public intermediate sources (i.e., news media), and lowest for private intermediate sources (i.e., self/family, personal physician, and peers) 4. RH4: Mean ratings and interrater agreement on trustworthiness will be highest for private intermediate sources (i.e., family, personal physicians, and peers), next highest for public intermediate sources (i.e., news media), and lowest for authorities (i.e., water utility, public health, emergency management, and elected officials) 5. RH5: Mean ratings and interrater agreement on protection responsibility will be highest for self/family, next highest for authorities (i.e., water utility, public health, emergency management, and elected officials), and lowest for public (i.e., news media) and other private (i.e., peers and personal physicians) intermediate sources Objective 2: To explore the mechanism of how stakeholders' social influence affects respondents' adoption of protective actions.
1. RH6: Stakeholders' overall social influence (the average of all three stakeholder attributes) will have positive correlations with risk perception and PAR compliance (i.e., drinking boiled water) Finally, responses to three broader questions are sought. Namely, 1. RQ1: Do stakeholder perceptions have a direct effect on response actions or an indirect effect via their effects on risk perception? 2. RQ2: Do demographic characteristics, preparedness, experience, or risk perceptions affect the adoption of protective actions to water contamination as strongly as stakeholder perceptions? 3. RQ3: Are there differences in the predictors of the PAR compliance, the alternative protective actions, and ignoring the threat?

Data Collection
The data reported here is derived from a survey conducted by the Texas A&M University Hazard Reduction & Recovery Center (HRRC) six months after the May 1-4, 2010, Boston water contamination incident. The team randomly selected 600 households from the affected communities and, following Dillman's (1999) survey procedure, mailed the first wave of survey packets containing a cover letter, an informed consent form, a questionnaire, 7 of 23 and a stamped return envelope to the selected households. This was followed by a reminder postcard and two more waves of survey packets at two-week intervals to those who had not returned a completed questionnaire. Of the 600 selected addresses, 102 were undeliverable. Of the remaining 498, 117 respondents returned questionnaires. Of these questionnaires, seven had over 25% missing items and were excluded from the data set. This yielded a final response rate of 22.4%, which is lower than contemporaneous HRRC surveys using the same procedure-39.9% from the Hurricane Katrina evacuation survey and 41.8% from the Hurricane Rita evacuation survey , 42.8% from the Christchurch earthquake response survey, and 55.3% from the Tōhoku earthquake response survey (Lindell et al., 2016).
The lower response rate might be the result of these other disasters causing substantial deaths, injuries, and economic losses, whereas the water contamination incident produced only minor disruption and, quite possibly, limited interest to most residents. By comparison, general population survey response rates currently average less than 10% (Leeper, 2019) and some hazards surveys have response rates this low (8% in Jiang et al., 2021) or lower (2% in Martin et al., 2020), so this water contamination survey's response rate is substantially above average. Of the valid responses, 46 respondents were from Boston, 23 from Brookline, and 41 from Somerville. Moreover, 61% of the respondents were female, 75% Caucasian, 39% married, and 48% were homeowners. The respondents had an average age of 48 yr, 16 yr of education, an annual average household income of US $67,057, and two members per household. Despite an over-representation of females, the sample was generally consistent with the 2000 Boston census data.

Questionnaire
The survey comprised multiple measures used to examine residents' PAR compliance, some of which were reported by Lindell, Huang, and Prater (2017) and Lindell, Mumpower, et al. (2017). This article focuses on portions of the questionnaire not previously analyzed in those studies. First, water contamination response was measured by three variables-PAR compliance, alternative protective actions, and ignoring threat-measured on a 1-5 scale (from Not at all = 1 to Very great extent = 5) of the extent to which they used boiled water, bottled or self-chlorinated water, and untreated tap water as their drinking water source, respectively. Each respondent's risk perception was measured by averaging the ratings of the likelihood of getting sick from untreated tap water through seven different exposure paths (have a glass of water to drink, rinse fresh vegetables such as lettuce, cook some spaghetti noodles, brew a pot of coffee, rinse their mouths after brushing their teeth, take a shower, and wash clothes) with the same 5-category extent scale, which yielded a measure with high internal consistency reliability (Cronbach's α = 0.83). Measures of the eight stakeholder types on the three stakeholder attributes comprised ratings of WDS personnel, public health personnel, emergency management personnel, and elected officials; news media, personal physician; and peers, and self/family on hazard expertize, trustworthiness (only family was the referent on this attribute), and responsibility with the 5-category extent scale. This generated 24 perceived stakeholder attribute items. An overall social influence score was created for each of the three stakeholders by averaging the three attribute ratings for each stakeholder.
Missing data analysis revealed that the highest rate was 28.2% and a test of missing completely at random revealed a non-significant result (χ 2 1,434 = 1,444.9, p > 0.05), indicating that the missing data occurred completely at random rather than a result of any specific variables. Hence, missing values were replaced by the Expectation-Maximization algorithm in SPSS 17.0.

Tests for Pseudo-Attitudes
Quantitative researchers face the problem of pseudo-attitudes when asking research participants to rate unfamiliar objects or concepts (Converse, 1970;Schuman & Kalton, 1985). Specifically, participants who want to avoid appearing ignorant might provide responses that are created in reaction to the questionnaire rather than ones that tap stable attitudes. One indication of pseudo-attitudes is that respondents check the scale midpoint, rather than leaving the answer blank, to indicate an opinion on topics to which they have given little or no thought. This leads to central tendency bias if this is the case for many respondents (Cascio & Aguinis, 2004). To test whether responses are due to central tendency bias, variable means can be tested to determine if they differ significantly from the scale midpoint (Cascio & Aguinis, 2004). A series of t tests revealed that, of the three behavioral and 25 psychological variables, 25% (7/28) of them have ratings that are not significantly different from the mid-point (3) of the 1-5 rating scale. However, a mean rating M = 3.0 could be the result of response distributions as dissimilar as, at one extreme, all respondents providing a rating of "3" and, at the other extreme, half providing a rating of "1" and the other half providing a rating of "5" (Lindell & Brandt, 2000). Since all respondents providing a rating of "3" is what would be expected with central tendency bias, it is also important to determine if there is a high level of interrater agreement, which can be measured by r WG -an index that ranges −1.0 ≤ r WG ≤ +1.0 and has a value of zero when the ratings have a uniform random distribution (LeBreton & Senter, 2008). None of the seven items whose means were nonsignificantly different from the midpoint had interrater agreement higher than r WG = 0.70, a reasonable threshold for concluding the presence of pervasive central tendency bias. Hence, it is reasonable to conclude that the data are not significantly affected by pseudo-attitudes.

Analyses
The first objective (Examine how respondents would rate each stakeholder's attributes of expertize, trustworthiness, and protection responsibility) was examined using descriptive statistics and multivariate analysis of variance (MANOVA). Interrater agreement was tested using the Dunlap et al. (2003) table of statistical significance for r WG . Differences among the three attribute profiles were calculated by computing the root-mean-squared (RMS) differences between each pair of attributes over all stakeholders. The second objective (Explore the mechanism of how stakeholders' social influence affects people's adoption of protective actions) involving RH6 and RQ1-RQ3 was tested using correlation and regression analysis.
In the analyses, that follow, there are (8 × 7)/2 = 28 paired t tests for comparisons of the eight stakeholders on each of the three attributes for a total of 84 statistical tests. In addition, there are 199 tests on correlation and regression coefficients, so the total number of 283 statistical tests makes experiment-wise error rate a concern (Ott & Longnecker, 2015). Specifically, the expected number of false positive tests is FP = α × n, where FP is the number of false positive test results, α is the Type I error rate, and n is the number of statistical tests. If α = 0.05 and n = 283, then FP = 14. Benjamini and Hochberg (1995), see, for a more recent discussion, Glickman et al., 2014 advocated that researchers', (a) specify a false discovery rate (d) for the entire study, (b) sort the p i significance values for the individual tests in ascending order 1 ≤ i ≤ n, and 3 classify each p i ≤ d × i/n as statistically significant. In the present study, the exact critical value of p i = 0.019, which we rounded down to p = 0.01 for that only p-values less than this are classified as statistically significant.

Profile and Cluster Analysis
The hypothesized classification of stakeholders (i.e., risk communicators) is mostly, but not completely, supported by the data. Specifically, the profiles in Figure 2 suggest that the hypothesized grouping of stakeholders into authorities, public intermediate sources, and private intermediate sources is generally supported, but elected officials tend to be rated more like news media rather than other authorities, whereas personal physicians tended to be rated differently from other private intermediate sources.
To further examine the hypothesized stakeholder groups, the profiles of the eight stakeholders were submitted to a hierarchical cluster analysis using squared Euclidean distances as the proximity measure and Ward's method as the clustering method. This analysis produced the dendrogram in Figure 3 that reveals three primary clusters, the first of which is defined by three of the authorities-water utility, public health, and emergency management. The second primary cluster is defined by peers and self/family, whereas the third primary cluster is defined by elected officials and news media. The second and third clusters merged with each other and then, much later, with personal physicians, after which all clusters merged. Based on these results, the categorization of stakeholders was revised to technical authorities (i.e., combining water utility, public health, and emergency management), public intermediate sources (i.e., combining elected officials and news media), and private intermediate sources (i.e., combining self/family, peers, and personal physicians).

Tests of RH1-RH5: Perceived Stakeholders' Social Influence Attributes
RH1 (There will be significant differences among the mean ratings of the stakeholders on the three social influence attributes-expertize, trustworthiness, and protection responsibility) is supported by a MANOVA that reveals significant effects for stakeholder (Wilks Λ = 0.32, F 7,103 = 30.75, p < 0.001), and interaction (Wilks Λ = 0.50, F 14,96 = 6.88, p < 0.001), but not attributes (Wilks Λ = 0.94, F 2,108 = 3.18, ns). As indicated in Figure 2, the significant stakeholder effect is due to differences between the highest and lowest-rated stakeholders on each of the three attributes. These were largest for protection responsibility (M 1 -M 2 = 4.15-2.13 = 2.02, which is 50.5% of the 1-5 rating scale) followed by trustworthiness (M 1 -M 2 = 3.95-3.08 = 0.87-21.8% of the rating scale), and expertize (M 1 -M 2 = 3.87-3.03 = 0.84-21.0% of the rating scale). The interaction is due to differences among stakeholders in the differences among their ratings across attributes. Specifically, peers,  personal physicians, and news media have their highest ratings on trustworthiness, followed by expertize and protection responsibility. By contrast, the ratings of the water utility differed slightly on the three attributes but in the opposite direction-highest on protection responsibility, followed by expertize and trustworthiness. Finally, the ratings for self/family, elected officials, emergency management, and public health are all equally high on all three attributes.
Consistent with RH2 (Stakeholders' attribute profiles on expertize and trustworthiness will be much more like each other than either one is to protection responsibility), the difference between the mean rating profiles of expertize and trustworthiness is RMS = 0.23, whereas the differences of the mean rating profiles of those variables with protection responsibility are RMS = 0.39 and RMS = 0.55, respectively.
Partly consistent with RH3 (Mean ratings and interrater agreement of hazard expertize will be highest for authorities, next highest for public intermediate sources, and lowest for private intermediate sources), a MANOVA reveals significant differences in expertize ratings among stakeholders (Wilks Λ = 0.05, F 8,102 = 251.60, p < 0.001). As indicated in Table 1, technical authorities received the highest mean ratings (Public health M = 3.87, Water utility M = 3.77, and Emergency management M = 3.73). However, the lowest technical authority (emergency managers) has a nonsignificantly higher rating than news media or self/family (M = 3.58 and 3.40, respectively). In turn, these stakeholders received higher ratings than elected officials, peers, and personal physicians (M = 3.15, 3.05, and 3.03, respectively). Contrary to the hypothesis, there are no meaningful differences in interrater agreement on the ratings for most of the stakeholders. Specifically, respondents have moderately high agreement on the ratings of self/family (r WG = 0.47, p < 0.001), followed by technical authorities ( WG = 0.44, p < 0.001), public intermediate sources ( WG = 0.44, p < 0.001), and peers (r WG = 0.40, p < 0.001). However, there is virtually no agreement on personal physicians (r WG = 0.12, ns).
Mostly contrary to RH4 (Mean ratings and interrater agreement on trustworthiness will be highest for private intermediate sources, next highest for public intermediate sources, and lowest for authorities), a MANOVA indicates significant differences in trustworthiness ratings between stakeholders (Wilks Λ = 0.05, F 8,102 = 251.24, p < 0.001). As Table 1 indicates, news media (M = 3.95), a public intermediate source, received nonsignificantly higher ratings of trustworthiness than two of the technical authorities-public health and emergency management (M = 3.83 and 3.76, respectively), but the latter had nonsignificantly higher ratings than water utility, family, and peers (M = 3.59, 3.61, and 3.48, respectively). This latter group has significantly higher ratings than elected officials (M = 3.33), who have higher ratings than personal physicians (M = 3.08). However, partly consistent with the hypothesis, interrater agreement on trustworthiness is moderately high for news media (r WG = 0.45, p < 0.001) and technical authorities ( WG = 0.41, p < 0.001), but is a bit lower for the other two intermediate sources-elected officials and peers (r WG = 0.31 and 0.22, ns, respectively), and very low for self/family (r WG = 0.10, ns) and personal physicians (r WG = 0.09, ns).
Partially consistent with RH5 (Mean ratings and interrater agreement on protection responsibility will be highest for self/family, next highest for authorities, and lowest for public and private intermediate sources), a MANO-VA revealed significant differences in protection responsibility ratings among stakeholders (Wilks Λ = 0.04, F 8,102 = 318.33, p < 0.001). Table 1 shows that two technical authorities-water utility and public health (M = 4.15 and 3.96, respectively)-have significantly higher ratings than emergency management and self/family (M = 3.75 and 3.45, respectively), who have higher ratings than news media and elected officials (M = 3.21). In turn, these have higher ratings than peers and personal physicians (M = 2.69 and 2.13, respectively). Also, partly consistent with the hypothesis, technical authorities generally have the highest interrater agreement on protection

Tests of RH6, RQ1-RQ3: Effects of Stakeholders' Social Influence on Protective Actions
Table 2 displays the means, standard deviations, and intercorrelations among the variables in RH6 (Stakeholders' overall social influence will have positive correlations with risk perception and PAR compliance). Contrary to the hypothesis, risk perception has nonsignificant correlations with the overall social influence of all stakeholders. However, the overall social influence of authorities (r = 0.25) and public intermediate sources (r = 0.28) is positively correlated with PAR compliance, but none of the stakeholders' overall social influence variables has a significant correlation with taking the alternative protective actions or ignoring the threat.

RQ1 (Do stakeholder perceptions have a direct effect on response actions or an indirect effect via their effects on risk perception?) is first answered by the nonsignificant correlation of risk perception with PAR compliance.
Specifically, in the absence of a significant correlation of risk perception with PAR compliance, stakeholder attributes cannot have an indirect effect on PAR compliance via their effects on risk perception. In addition, risk perception has nonsignificant correlations with the alternative protective actions and ignoring the threat. Table 2, which show that age has a negative correlation (r = −0.41) and income has a positive correlation with taking an alternative protective action (r = 0.23), whereas those having a higher preparedness level are more likely to ignore the threat (r = 0.23). Next, regression analyses for PAR compliance were conducted in the three stages displayed in Table 3. In Model I, PAR compliance was regressed onto the demographic variables, preparedness, and experience, whereas in Model II, compliance was regressed onto each stakeholder's overall social influence. After first entering all relevant variables into the regression model, backward deletion was used to discard nonsignificant predictors. Model I identified one statistically significant predictor, income (β = −0.27), with an adjusted R 2 = 0.03. Model II retained public intermediate sources (β = 0.28) and personal physician (β = −0.34) as the significant predictors with a significant adjusted R 2 = 0.15. It is noteworthy that Table 2 indicates that technical authorities and public intermediate sources' ratings were highly correlated (r = 0.67) and had approximately equal correlations with PAR compliance (r = 0.25 and 0.28, respectively), yet had been identified as distinct stakeholders. Thus, a re-estimated equation with both variables entering into Models II and III yielded regression coefficients of β = 0.28. Moreover, the results in Model III produced a statistically significant adjusted R 2 = 0.28 with significant coefficients for income (β = −0.25) and household size (β = 0.26), as well as for authorities and public intermediate sources (β = 0.40), and personal physician (β = −0.47).

RQ2 (Do demographic characteristics, preparedness, experience, or risk perceptions affect the adoption of protective actions to water contamination as strongly as stakeholder perceptions?) was first examined by the correlations in
The test of RQ3 (Are there differences in the predictors of the PAR compliance, the alternative protective action, and ignoring the threat?) regressed alterative protective action and threat-ignoring behavior onto all predictor variables followed by backward deletion of the nonsignificant predictors. Table 4 indicates that the analysis of alternative protective actions produced a statistically significant adjusted R 2 = 0.31 with significant coefficients for age (β = −0.50), income (β = 0.24), preparedness (β = 0.27), private intermediate sources (β = −0.20), and risk perception (β = 0.26). Analysis of ignoring the threat produced a model having a smaller but statistically significant adjusted R 2 = 0.12 with significant coefficient for preparedness (β = 0.22).

Discussion
The findings of the cluster analysis generally support the PADM and CNM proposition that stakeholders can be meaningfully divided into authorities, public intermediate sources, and private intermediate sources. However, the original classification requires some modification; respondents viewed elected officials as a public intermediate source like the news media rather than as one of the technical authorities. This suggests that elected officials are viewed simply as conduits for information from water utility, public health, and emergency management personnel rather than experts in their own right. In addition, despite frequently being mentioned in warnings as a supplemental source of health information, respondents viewed personal physicians as quite different from the   other stakeholders, especially because of their low ratings on protection responsibility. More broadly, however, there was a nonsignificant level of agreement on the ratings of physicians on all three stakeholder attributes. This suggests that many people consider personal physicians to be largely irrelevant in a water contamination incident so, although there is no harm in identifying them as an information source, few people are likely to contact them for information.

Tests of RH1-RH5: Perceived Stakeholders' Social Influence Attributes
The findings in support of RH1 (There will be significant differences among the mean ratings of the stakeholders on the three social influence attributes-expertize, trustworthiness, and protection responsibility) are noteworthy because they indicate that respondents differentiate among water contamination stakeholders on these attributes. In turn, this underscores the importance of identifying the origins of these perceptions and the effects of those perceptions on PAR compliance, consumption of bottled or self-chlorinated water, and ignoring the threat. Possible origins of each of these perceptions are addressed below.
The findings in support of RH2 (Stakeholders' attribute profiles on expertize and trustworthiness will be much more like each other than either one is to protection responsibility) are important because they replicate findings from Arlikatti et al. (2007) and Wei et al. (2018), which suggest that these social influence attributes are not independent. Nonetheless, it is unclear if trustworthiness is inferred from expertize, expertize inferred from trustworthiness, or if both are inferred from other sources. The finding that authorities are viewed as having high trustworthiness aligns with other studies and is likely due to belief that they are more knowledgeable about hazards (Arlikatti et al., 2007;Lindell & Perry, 1992;Sager, 1994;Taibah et al., 2017 (Hoorens & Buunk, 1992), which is people's tendency to regard themselves as being above the average and then estimate others in accordance with this anchor point (Alicke & Govorun, 2005;Goethals et al., 1991). However, this explanation only accounts for comparison to other private intermediates because self/family received lower expertize ratings than technical authorities and news media, a similar pattern to the one found for volcano (Perry & Lindell, 1990) and earthquake (Lindell & Whitney, 2000) hazards.
There is some evidence that people's familiarity with a hazard reduces the differences in perceived expertize among stakeholders because Lindell and Perry (1992) reported that respondents near the Mount St. Helens volcano rated themselves as more similar to authorities in hazard expertize (12 yr after the volcano erupted) than for two less familiar hazards-toxic chemicals transported along a nearby rail line and radiological hazard from a nearby nuclear power plant, a finding seconded by Wu et al. (2017) study of the Oklahoma earthquake. This suggests that Boston-area respondents considered water contamination to be a more familiar, and perhaps much more personally controllable, hazard than these other environmental hazards. Otherwise, news media but not DV = Alternative action DV = Threat-ignoring behavior  elected officials are viewed as having high expertize and therefore an important channel from which to receive information. These results are consistent with previous studies on perceived stakeholder expertize, which suggest that technical authorities are thought to have high expertize due to their educational credentials, whereas news media are thought to have high expertize due to their close contact with scientists and other experts (Arlikatti et al., 2007;Latré et al., 2018).
RH4 (Mean ratings and interrater agreement on trustworthiness will be highest for private intermediate sources, next highest for public intermediate sources, and lowest for authorities) was only partially supported by the finding that news media (a public intermediate source) was rated highest of all the stakeholders, which can be explained by a parasocial relationship that develops between the local media and their audiences that can increase trust (Sherman-Morris et al., 2020). Contrary to the hypothesis, however, all private intermediate stakeholders were rated lower than technical authorities and news media. However, after excluding personal physicians, the differences among private intermediate stakeholders were not significant. One possible explanation for the differences among stakeholders with respect to trustworthiness is that respondents infer this attribute from a variety of sources. For example, Perry and Lindell (1990) reported that residents of areas near Mount St. Helens regarded the county Department of Emergency Services and County Sheriff as the most credible information sources because of their special skills (expertize) and past reliability (trustworthiness), which were attributable to relevant educational credentials, acceptance by currently trusted sources, and past job performance. Accordingly, the high mean ratings and levels of agreement regarding the trustworthiness of technical authorities and news media in the present study could be a result of their salient public image and trusting relationships with respondents.
Conversely, even though the ratings of family's and peers' trustworthiness are unexpectedly low, this finding is consistent with a study by Arlikatti et al. (2007). These relatively low trustworthiness ratings may be due to differential exposure to these stakeholders. Specifically, people generally see authorities and public intermediates on their best behavior, whereas they see their peers and their families along the entire range from their best to their worst behavior. Since negative instances, especially emotionally charged ones, are particularly memorable (Kensinger & Ford, 2020), this might account for the relatively low ratings of these two types of stakeholders.
The lack of complete support for RH5 (Mean ratings and interrater agreement on protection responsibility will be highest for self/family, next highest for authorities, and lowest for public and private intermediate sources) is also somewhat surprising because technical authorities, rather than self/family, received the highest ratings for protection responsibility. This might be due to differences among hazards because Arlikatti et al. (2007) and Lindell and Whitney (2000) found that self/family had higher ratings than authorities for earthquake protection responsibility, whereas Wei et al. (2018) and Wu et al. (2017) reported that self/family had lower ratings than authorities for Oklahoma human-induced earthquake and seasonal influenza protection responsibilities, respectively. One possibility is that authorities are perceived to have substantially more control over water contamination than earthquakes, whereas another possibility is that people attribute protection responsibility to authorities when they themselves lack knowledge about effective protective actions which, in turn, arises from their lack of disaster experience or hazard education. For example, Krasovskaia et al. (2007) found that respondents who attributed responsibility for flood prevention to authorities also had a passive attitude toward flood risk due to a false sense of security that came from never having experienced a flood.
The finding that the public intermediate sources were rated next highest on protection responsibility, but with a nearly uniform distribution of protection responsibility ratings, can be attributed to disagreements about their roles as information sources. For example, the state of emergency declared by the Boston Mayor was simply a repetition of the message given by the State Governor, retransmitting incident information and PARs originated by the Massachusetts Water Resources Authority. Hence, the respondents may rate the protection responsibility of public intermediates in accordance with perceptions of these stakeholders' social functions.

Tests of RH6, RQ1-RQ3: Effects of Stakeholders' Social Influence on Protective Actions
Regarding RH6 (Stakeholders' overall social influence [the average of all three stakeholder attributes] will have positive correlations with risk perception and PAR compliance), Models II and III in Table 3 reveal that the social influence of authorities and public intermediate sources (both β = 0.40) together with personal physicians (β = −0.47) has direct effects on PAR compliance. These findings are consistent with some findings of direct effects of stakeholder attributes on protective actions (Heath et al., 2018;Lindell & Whitney, 2000) but not Arlikatti et al. (2007), who found evidence of both direct and indirect effects. One explanation is that the PAR in the water contamination incident was for a protective action (boiling tap water) that was perceived to be no more effective than the alternative protective actions (bottled water and self-chlorinated water) but required more time and effort.
Moreover, the equal weights for authorities and public intermediate sources in Models II and III imply that these sources could substitute for each other in communications with the public. However, PAR compliance is more likely if they are communicating the same message and thus have additive effects. Conversely, authorities and public intermediate sources will tend to cancel each other if their messages conflict. Thus, the consistency of messaging by authorities and public intermediate sources can be expected to have a major effect on PAR compliance in future water contamination incidents.
There was a negative answer to RQ1 (Do stakeholder perceptions have a direct effect on PAR compliance or an indirect effect via their effects on risk perception?) because a mediation effect for stakeholders' social influence on response actions via risk perceptions was precluded by the finding that risk perception itself was not significantly correlated with any of the response actions. This negative result is not completely contrary to Lindell and Perry's (2004) assertion that stakeholders' social influence could elicit direct compliance via Petty and Cacioppo's (1986) peripheral route, rather than via their central route because Lindell and Perry (2004) acknowledged the possibility of both routes. Thus, given that only a direct effect was found in this study, it remains to be determined which personal characteristics and incident conditions favor a direct effect and which favor an indirect effect.
Regarding RQ2 (Do demographic characteristics, preparedness, experience, or risk perceptions affect the adoption of protective actions to water contamination as strongly as stakeholder perceptions?), the effect size changes of household size and social influence of personal physician, from a nonsignificant correlation to a significant regression coefficient require an explanation. One possibility is that the significant effect of household size on PAR compliance is due to concern about children's health. Specifically, whereas single people or childless couples might be willing to take chances with untreated tap water, parents are unlikely to take similar chances with their children's health. On the other hand, the significant negative effect of personal physician social influence can be explained as an artifact of collinearity among the stakeholder ratings because Table 2 indicates that these variables (Variables 10-13) have an average intercorrelation of r = 0.46. Consequently, the standardized regression coefficient for authorities increases from its correlation (from r = 0.25 to β = 0.40), public intermediates decreases from its correlation (from r = 0.28 to β = 0.13), and personal physician becomes more negative (from r = −0.15 to β = −0.47).
The results for RQ3 (Are there differences in the predictors of the PAR compliance, the alternative protective action, and ignoring the threat?) indicate that there are distinctly different predictors for these three dependent variables. One possible explanation for the significant effects of age (β = −0.50), income (β = 0.24), and preparedness (β = 0.27) on the adoption of alternative protective actions is that these are proxies for respondents' routine drinking water sources, especially bottled water. As Lindell et al. (2017a) found in other data from this incident, people who routinely drank bottled water before the incident would be more likely to continue to drink it during the boil water order. The nonsignificant effect of stakeholders' overall social influence, together with the significant effect of risk perception on the adoption of alternative protective actions, is noteworthy. As one respondent indicated that "I was very sensitive about this water contamination because I was 7 months pregnant at that time. If I was not, I could have drunk boiled tap water more, but I did not." This comment implies that the reason why some people drank bottled or self-chlorinated water was not to reject compliance with authorities' PARs, but rather a personal risk perception that indicates an alternative protective action would yield the same level of protection (Lindell et al., 2017b).
The positive effect of preparedness on threat-ignoring behavior is somewhat puzzling because it suggests that optimistic bias misleads households into believing they are well-prepared, causing them to overlook their risk exposures (see, for example, Lo & Cheung, 2015). However, this finding needs to be tested further to see if it can be replicated and explained in future research.

A Conceptual Diagram Explaining Information Flow for a Water Contamination Incident
In summary, judgments of stakeholder attributes, especially protection responsibility can be explained by a process that integrates the CNM in Figure 1 with the Sociotechnical Systems Model in Ehsan Shafiee et al. (2018) and the Chain of Events Model from Lindell and Perry (1992). According to the chain of events at the top of Figure 4, contamination enters a WDS and disperses throughout the system, producing exposures if people drink the contaminated water, and adverse health effects depending on the contaminant's toxicity and the quantity consumed. The second chain of events involves the events in the social system that respond to the environmental chain of events. Specifically, people consider WDS operators responsible for detecting contaminant intrusion by carefully monitoring the system (indicated by dashed line) and taking corrective action to eliminate that contamination from the WDS (indicated by the solid lines to contamination and dispersion).
In addition, the WDS operator is responsible for promptly notifying public health and emergency management agencies, elected officials, and news media (also indicated by dashed lines), as well as transmitting warnings to those at risk indirectly via the news media (e.g., TV, radio, and newspapers) and directly via their Internet and social media sites (e.g., Facebook and Twitter). To the degree that there has been an absence of contamination incidents in the past, people are likely to consider WDS operators to be expert and trustworthy. However, if there is a contamination incident, people hold WDS operators responsible for conducting interventions that terminate the intrusion of contaminants into the system and preventing further dispersion by flushing the contaminants that have already entered.
Unlike the WDS operators, who have information about the system and physical control of it, public and private intermediates only have information about WDS contamination. In addition to information that the WDS operator provides about the state of the system, public health and emergency management agencies have specialized expertize about the effects of that contamination on public health and the appropriate PARs that should be issued. Self and family are ultimately responsible for deciding whether to comply with authorities' PARs but can consult with peers and personal physicians to confirm the warning and discuss the logistics of protective action-the process of milling (Wood et al., 2018). Elected officials and news media are perceived to have protection responsibility only to the extent that people consider it to be their role to disseminate prompt and accurate information about the incident to the public, whereas peers have protection responsibility only to the extent that they are expected to provide such information to their friends, relatives, neighbors, and coworkers. Finally, personal physicians are considered to have low levels protection responsibility because their role is to provide advice on their patients' personal health rather than public health. However, they can implement remedial actions to minimize the adverse health effects to those who get sick. In summary, the interpretation of protection responsibility is complex because the various stakeholders differ in the actions that they can take at successive stages in the environmental chain from contamination through dispersion and exposure to health effects.

Study Limitations
It is important to acknowledge that this study has its limitations. The response rate was only 22%, which raises concern that the respondents may not be truly representative of the population affected by this water contamination incident. However, a low response rate does not necessarily imply response bias because the latter occurs only if demographic characteristics are significantly correlated with questionnaire response, which they are not (Groves & Peytcheva, 2008;Tourangeau, 2017). Moreover, a low response rate does not seem to bias central tendency estimates such as means and proportions (Keeter et al., 2000). Finally, when testing path models, the issue of generalizability from the sample to the population most directly concerns whether the sample's correlation and regression coefficients for the psychological and behavioral variables-not their means and proportions-are representative of those in the population to which the results will be generalized. This generally means that the issue is whether there is adequate variation in the variables to avoid bias in those correlation and regression coefficients. Thus, even if there is bias in the estimated means and proportions on the psychological and behavioral variables, there will be little effect on correlation coefficients unless there are ceiling or floor effects that cause these coefficients to be systematically underestimated (Lindell & Perry, 2000a, 2000bNunnally & Bernstein, 1994).
It is also important to point out that, since the respondents' rating of trustworthiness were quite similar to their ratings of expertize, an anchor effect might have occurred in which respondents made judgments based on an initial value (Furnham & Boo, 2010). However, it cannot be determined for certain if an anchor effect occurred since expertize and trustworthiness ratings differed in their similarity to protection responsibility ratings. Moreover, as a cross-sectional design, this study is also limited in its ability to determine definite causal inferences (Lindell, 2008). A longitudinal study might reveal more informative findings on the cause-and-effect relationships among residents' ratings of stakeholders' attributes, especially if there is an event that changes those ratings.

Conclusions and Future Work
To understand risk communication and PAR compliance during the 2010 Boston water contamination incident, 600 randomly sampled residents were mailed questionnaires, yielding 110 valid responses. The findings from this study have some important implications for other water contamination incidents. First, water from alternate sources, although untreated, was later found to be safe and did not cause any detectable negative health effects. This may have been perceived by some people as excessive caution by government health authorities that could lead to a cry wolf effect (Breznitz, 1984). However, it usually takes repetitive false alarms warning the public to take protective actions, but the threat not materializing, to cause the public to lose faith in official warning systems (LeClerc & Joslyn, 2015;Ripberger et al., 2015;Simmons & Sutter, 2009). Although false alarms can lead to noncompliance when a real threat strikes (Atwood & Major, 1998;Jauernic & Van Den Broeke, 2016;Rigos et al., 2019;Sharma & Patt, 2012), the Boston water contamination was a single incident rather than a repetitive series, so it is unclear if any local residents considered the incident to be a false alarm that should be ignored in the future.
Second, the comparable levels of perceived expertize, trustworthiness, and protection responsibility of authorities from agencies such as water utilities, public health, and emergency management means that they need to communicate the same message-or at least compatible messages. Additionally, these findings suggest that using public intermediate sources to support warning message dissemination will also increase compliance. Since people are prone to seek additional information after receiving a message, especially if there is ambiguity (Lindell, 2018;Lindell et al., 2019;Wood et al., 2018), a larger number of sources disseminating the same message-or clearly compatible messages-will confirm the initial message and influence those at risk to comply with PARs more rapidly. Third, authorities need to plan, long before an incident occurs, how to warn people about water contamination through multiple channels to increase PAR compliance (Lindell & Perry, 2004). They are undoubtedly familiar with disseminating warnings through conventional channels such route alert, broadcast media, social media, and emergency notification systems (Arlikatti et al., 2014). However, another possibility is to inject food grade dye into the water main systems to alert the population to stop drinking tap water and trigger an appropriate response action (Rasekh et al., 2014). The advantage of food dye is that it would provide an immediately recognizable environmental cue that a household's tap water is unsafe to drink.
Fourth, water contamination incidents need to be taken seriously by the public as they can be caused by security breaches and vandalism as well as accidental pipe breaks. Even though a majority of the direct threats received by water distribution operators are hoaxes intended to receive media attention or settle a personal grudge, WDS operators must take each event seriously by adhering to the USEPA (2004) Response Protocol Tool Box. Specifically, Module 5: Public Response Guide outlines public health response measures that can potentially minimize public exposure to contaminants; and Module 6: Remediation and Recovery Guide, outlines the remedial and recovery process once the contamination incident is confirmed. Various organizations that are likely to be involved and their roles are also listed. The public needs to be made aware that these procedures have been established to increase their confidence that the technical authorities are executing a planned, rather than improvized, response.
In 2010, when this water contamination incident unfolded in Boston, the State of Massachusetts website did not have specific information related to drinking water health and safety. However, following this incident they added a section, titled "Drinking water boil orders and public-health orders", for people to learn how public health orders protect them from contaminated water supplies (Mass.gov, 2022). Detailed information about the following topics-water-borne illness, general precautions during a boil order, tips for water use during a boil order, what to do after the order is lifted-is followed by a short quiz at the end of each section. However, the link to access this information is a bit obscure and the contents on the webpage rather tedious to read without pictures or videos. Other state governments such as Michigan (https://www.youtube.com/watch?v=AkU6U8-5ztk), nonprofit organizations such as Boil Water Watch (https://www.youtube.com/watch?v=3zoojollIBA), and private entities are posting instructional videos with water experts. These YouTube videos present facts and animations (https:// www.youtube.com/watch?v=REiMJ5iLZRs) related to water contamination and boil water orders. Other states may want to provide similar content on their websites, especially if they make these resources more accessible and understandable to consumers of different demographic segments.
Finally, water safety management needs to be integrated with the rest of a community's comprehensive emergency operations planning (Lindell & Perry, 2007). To ensure an effective and timely water incident response, training on crisis communications should be provided to water utility personnel, as well as other technical authorities, hospitals/clinics, regional poison control centers, and news media. These can be in the form of charettes at town hall meetings or tabletop exercises, drills, and full-scale exercises. The USEPA has developed an SRS Exercise Development Toolbox to support the design and development, implementation, and evaluation of exercises for water contamination scenarios. The roles and responsibilities of all parties should be understood as promulgated by the whole community approach described in the National Response Framework (FEMA, 2019). In this way, water contamination events will be taken seriously and the public health and safety protected.