1. Background
Effective health communication is essential for disease prevention, health promotion and improved quality of life [
1]. During public health emergencies, the general public can only protect themselves if messages from the response agencies are received and understood, preferably from a trustworthy source. When public health is threatened by contaminated mains water, water companies have the responsibility of issuing one of three water notices to their consumers: “Do Not Use”, “Do Not Drink” and “Boil Water”. In England and Wales, “Boil Water” notices are quite frequently issued, while “Do Not Drink” notices are very rare [
2,
3]. This paper explores the effects that the cause of the water incident and the circumstances under which the water notice was issued have on consumers’ behaviour. We will contrast results from a routine “Boil Water” notice study (
i.e., an incident that triggered a routine response) with results from a previous study of a natural disaster that involved a “Do Not Drink” notice immediately followed by a “Boil Water” notice [
4].
Typically, risk communication is broken down into four elements: message, source, transmitter and receiver [
5]. It is well known that a source’s perceived credibility will influence the public’s perceptions and behaviour, but it has proved challenging to identify which factors influence trust and credibility, especially at a local level. Interestingly, flavour and odour have been found to have the strongest relationship with more or less trust in the water companies [
6]. An added difficulty is the potential impact that a transmitter can add to the mix, especially since government generated health advice is typically communicated through various media outlets, in particular, during emergencies [
7].
Different information channel preferences have been found for natural
versus routine incidents [
8] and different channels may affect accuracy of risk knowledge [
9]. During natural disasters, many agencies nowadays rely on transmission via television [
10,
11], radio [
12], email or mobile phone [
13]. Of crucial import is the timeliness of the advice, the loss of which will result in a lack of public comprehension of the advice, which in turn will increase confusion and anxiety, and reduce compliance levels [
14]. Brodie and colleagues [
15] found that about one-third did not get the message about the impending arrival of Hurricane Katrina at all, while a further one-third did not understand how to evacuate. Similarly, during Hurricane Rita, as few as 31% of people issued with a “Boil Water” notice were aware of it [
16]. Some studies of transmitter effects during routine water incidents report that transmitter use has no effect on behaviour [
17,
18], while others report positive effects from interpersonal contacts [
19] and negative effects from mass media [
20]. Studies of transmitter effects during natural disaster water incidents report that consumers may depend upon word of mouth more than media sources [
16].
Typically, non-compliance with ‘Boil Water’ advice ranges between 9% and 20% [
18,
21], whereas after Hurricane Rita, two-thirds did not boil their water for drinking [
16]. If we include other ingestion actions, e.g., brushing teeth or preparing/cooking food, non-compliance increases dramatically to 57% and 77% for human error and natural disasters, respectively [
16,
21]. Recently, theories of public non-compliance have digressed from ideas of ‘irrational’ behaviour towards the idea that risky actions are choices resulting from individual and societal factors. At a personal level, these factors may include demographics [
22], knowledge and experience of the situation [
23] and general health beliefs [
24]. Perceptions of the risks and recommended actions also influences compliance [
25]; for example, high perceived risk has been found for involuntary rather than voluntary risks [
26], unfamiliar rather than familiar risks [
27] and risks that are not controllable by the individual [
28]. In particular, technological mishaps are often perceived as high risk events [
29]. Routine incidents that involve some degree of human error also tend to be regarded as actively imposed and unacceptable [
30], whereas natural disasters are regarded as imposed without human agency and thus more acceptable.
Returning to the issue of trust, citizen advisory groups, health professionals, safety professionals, scientists and educators are consistently considered trustworthy or credible sources of information on environmental risk, except that those associated with an industry or believed to have a monetary interest are suspect sources [
31,
32]. Overall, it has been found that men have a greater level of trust in scientific authority, while women tend to exhibit higher levels of concern [
33]. Older age is also consistently associated with greater trust in government, science and experts [
34]. Climate change risk perception studies have found that highly educated people, whether measured in terms of general education or science education, tend to defer to scientific authority [
35]. Due to the often inevitable interaction between demographics, as well as with other factors, such as trust and attitudes, it is often impossible to demonstrate the true impact of one single demographic variable upon water perceptions [
36].
While consumers normally put great trust in information from physicians and other health professionals [
19], in some studies, they are the source that consumers say they are least likely to consult about drinking water concerns [
32]. Personal dissemination networks have been shown to be particularly vital for vulnerable sub-populations [
11], and interpersonal information is often perceived as more credible and efficient than official information sources [
19,
37]. Thus, information source use and preferences might be a more relevant determinant in water safety communication, than demographics.
Of the four elements of risk communication, this paper primarily focuses on the effects of sources, transmitters and receivers upon consumer compliance to health advice. Through quantitative analysis, we show how different information sources and demographic characteristics affect consumer behaviour during the two water incidents. Undoubtedly, the message content also contributes to consumer understanding and behaviour. However, during the incidents in this study, the message content was a variable that we were unable to control for as both incidents involved a great number of sources and transmitters with a huge number of messages. In order to address issues such as how the linguistic framing of water advice might affect consumer compliance, we conducted a separate qualitative analysis of the water advice reported by the media during the natural disaster incident [
38].
No previous studies have directly compared behaviour during routine water notices with behaviour during natural disaster water notices. We aim to show how the causes and circumstances of routine incidents versus natural disaster incidents result in differences in consumers’ compliance with advice, use of information sources, recall of advice received and their satisfaction with the information.
2. Methods
2.1. Two Incidents
At the Pitsford water treatment works in Northamptonshire in June 2008, a routine incident occurred after a rabbit entered the works [
39]. Low levels of
cryptosporidium oocysts were detected and a routine, precautionary “Boil Water” notice was issued to 258,000 people for ten days. In contrast, in summer 2007, the UK experienced its worst ever floods [
40]. The Mythe water treatment works in Tewkesbury (Gloucestershire) was flooded, resulting in complete loss of water to 340,000 residents. When tap water was restored, consumers were issued a “Do Not Drink” notice for seven days, which was then replaced with a “Boil Water” notice for a further four days.
This section will outline the methods used to study the routine “Boil Water” incident at Pitsford. To ensure comparability, the same study design was employed here as in our previous study of the natural disaster incident at the Mythe water treatment work [
4]. It should be noted that the distance between the two locations is 68 miles, but the households affected by the routine incident in 2008 had not been affected by the natural disaster incident in 2007.
2.2. Study Design and Sample Selection
A postal questionnaire study was sent to 1000 households affected by the routine incident in February 2009 (8 months after the incident). We obtained postcodes for affected areas from the Drinking Water Inspectorate. The Royal Mail Postcode Address File was then used to provide full addresses within these postcodes, from which 1000 were selected using a random number generator. Any business or school addresses were substituted for a further randomised selection of residential addresses.
Ethical approval was granted by the King’s College London Social Sciences, Humanities and Law Research Ethics Sub Committee.
2.3. Questionnaire Design
The postal questionnaire surveyed respondents’ uses of unboiled and boiled tap water, the advice that they remember receiving and the information sources that they consulted. Mainly close-ended questions were employed, which were a combination of yes/no questions, ranking questions, and “tick only one” and “tick as many as apply” multiple choice questions. The questionnaire was piloted twice on undergraduate students from King’s College London (N = 50), and minor revisions were made to wordings. The final questionnaire was sent out with a detailed project description and a stamped, addressed return envelope. A reminder was sent four weeks later to those who had not yet replied.
2.4. Coding
For ranking questions, where participants ticked rather than ranked options, a single tick was coded as rank one, whereas multiple ticks were given the same rank (e.g., three ticks were ranked as 2). Where participants first ranked options but then ticked one further option, the tick was coded as their lowest rank. On “tick only one” questions, multiple ticks were coded as inconclusive, with the exception of the water advice recollection question where multiple ticks were coded as “believed more than one advice was in place” so that uncertainty could be accounted for. For information source questions where respondents ticked “other’ or “website” but then provided additional information, answers were re-coded so that e.g., “television” includes listening, phoning and visiting websites of television channels/programmes whereas “website” includes internet-only sources. Open-ended questions were quantified where possible; e.g., home ownership was translated into the binary categories “yes, home owner” and “no, not home owner”. For all questions, non-responses were coded as missing data and inconclusive replies were largely excluded from analysis.
2.5. Analyses and Hypotheses
Data were entered into Microsoft Access 2007 and then cross-checked against the original responses. For statistical analysis, data were transferred into SPSS version 16. Once the data had been transferred into SPSS, they were validated a second time. As some respondents did not fully answer some questions, the sample size varies between questions.
As the routine incident only affected the drinking water, that is, other resources such as traffic, communications and electricity were unaffected, we hypothesised apriori that non-compliance with water advice would be higher for the natural disaster event compared to the routine incident. In addition, we predicted that demographic factors (such as age, gender, home ownership and employment, which were coded and explored identically for both studies), drinking water preferences, and use of information sources could have had an effect on participants’ perceptions and behaviours; however, as no formal hypotheses were defined apriori for the impact of demographics, drinking water preferences and use of information sources, statistical outcomes for these variables should be interpreted solely as indicators of the potential strength of association. Quantitative analysis is mainly descriptive. Inferential analysis was carried out using Chi-Square, Mann-Whitney, ANOVA, and Linear Regression. For all analyses with multiple predictor variables, only those variables that were significant at the p < 0.2 level in single predictor models were included in the multiple predictor models. The least significant variable was then removed from the model until all predictor variables were significant at the p < 0.2 level. The value of the model in predicting each dependent variable was then derived from the tests of between subjects’ effects in the corrected model. Throughout, the level of significance was set at 5% and only responses with at least 10 responses were included as dependent variables.