Promoting voluntary risk detection behaviour: an intervention study based on trending social norms combined with emotional appeals

Abstract

Well-owners are exposed to substantial health risks but rarely test their water and infrequently maintain their wells. We designed and tested an intervention aimed to promote voluntairy risk detection behavioiur. The intervention was based on a trending social norm message (suggesting an increasing trend of well-owners testing their wells) with emotional appeals. Two emotional appeals were used: fear frames (an indication of the severity of the risk) and worry appeals (indication of how many well-owners worried about contamination risks). In an online experiment, 262 Irish private well-owners were randomly allocated to six different risk communication messages that followed the design described above. The results showed a trending social norm message combined with a strong worry appraisal primarily raised levels of worry which, in turn, influenced attitudes towards water testing and behavioural intentions to maintain wells. Our results illustrate the potential of stressing a trending social norm and worry appraisal to promote voluntary risk detection behaviours.

Keywords:

• Risk detection behaviour
• trending social norms
• fear frame
• worry appraisal
• worry

Introduction

Many health policies rely on promoting voluntary behaviour such as risk detection behaviour and screening, as these are cost-effective, and they promote commitment and responsibility for one’s own health. However, interventions aimed at changing voluntary behaviour often have limited effects if targeted at a broad audience (Keller and Lehmann 2008; Snyder et al. 2004). In many cases, changing voluntary risk detection behaviour is especially difficult when only a minority engages in this behaviour, implying that there is no social norm to guide their behaviour (Schultz et al. 2018; Stok et al. 2012). The contribution of this paper is the development and test of an intervention strategy that could promote voluntary risk detection behaviour in a weak social norm context. As a case in point, the focus is on private well-owners’ detection behaviour of the potential contamination of their drinking water supply. This is a known health risk area where the uptake of water testing and management is low (indicating a weak social norm), despite the existence of real health risks (Flanagan, Marvinney, and Zheng 2015; Geiger and Eshet 2011; Hooks, Schuitema, and McDermott 2019; Hynds, Misstear, and Gill 2013).

In rural areas across the world people rely on private wells to provide them with drinking water. Private wells are known to be susceptible to bacterial contamination from the land surface, for example due to anthropogenic activities such as the run-off from processing, agricultural land use, and waste-water systems. Well water can also pose serious health risk from natural or geogenic contamination (e.g., arsenic, uranium, nickel, chromium), which arises from the interaction of groundwater with rocks and minerals in the sub-surface (Ayotte, Gronberg, and Apodaca 2011; Rao and Mamatha 2004). The estimates of private wells that contain at least one hazardous element at a concentration that exceeds health regulations and guidelines vary from 13% in the US (Ayotte, Gronberg, and Apodaca 2011) to 30% in Bangladesh (Ahmad, Khan, and Haque 2018) and 30% in Ireland (EPA 2020). Drinking contaminated water can have acute health effects, such as vomiting, abdominal pain and diarrhoea, however, consistent exposure to toxic chemicals via groundwater can have more long-term effects too such as diabetes, pulmonary disease, and cardiovascular diseases (WHO 2018). Also, several types of cancer have been associated with long-term exposure to arsenic, one of the most prominent natural contaminants in groundwater-derived drinking water (Smith, Lingas, and Rahman 2000).

Unlike public water suppliers, private well-owners are usually not legally obliged to test their drinking water quality to ensure compliance with drinking water regulations (US EPA 2019). Thus, well-owners are largely responsible for the overall management of their own drinking water systems including well drilling, maintenance, water quality management, and water testing. Despite recommendations to test their well water on a regular basis for contaminants (EPA & WHO 2002; US EPA 2019), private well-owners rarely do so. They tend to test their water on an ad-hoc basis, that is, as a one-off or if issues arise (Flanagan, Marvinney, and Zheng 2015; Hooks, Schuitema, and McDermott 2019; Hynds, Misstear, and Gill 2013; Imgrund, Kreutzwiser, and de Loë 2011; Kreutzwiser et al. 2011).

Most campaigns to encourage well-owners to test their water and maintain their wells on a regular basis focus on information provision, that is, they inform people about the risks and what may be done to minimise these (e.g., Islam et al. 2011). While this may stimulate people to search for more information and increase their knowledge, it often does not change their behaviour in the desired direction (Huurne and Gutteling 2008). Some have argued this is because private well-owners generally do not perceive large risks (Flanagan, Marvinney, and Zheng 2015; Hanchett et., 2002; Jones et al. 2006), whilst others argue that a lack of worry explains why messages are not very effective (Morris, Wilson, and Kelly 2016; Schuitema, Hooks, and McDermott 2020; Syme and Williams 1993).

In this paper, we present an intervention study that was designed to test if enhancing risk perceptions and/or worry would be most effective in affecting well-owners’ attitudes to well water testing and well maintenance intentions. This intervention includes the use of a trending social norm (TSN) message, which was included to take the weak social norm around well water testing and maintance into consideration..

Intervention design

Trending social norms

One noticeable feature among well-owners is that relatively few well-owners test their water (Flanagan, Marvinney, and Zheng 2015; Hynds, Misstear, and Gill 2013; Imgrund, Kreutzwiser, and de Loë 2011; Kreutzwiser et al. 2011), which indicates that there is weak social norm around water testing (Hooks, Schuitema, and McDermott 2019). Social norms are defined as ‘rules that are understood by members of a group, and that guide/constrain human behaviour’ (Cialdini and Trost 1998, 152). They essentially act as cues for people in social situations surrounding what actions are (dis)approved or expected within society (Cialdini and Goldstein 2004). People tend to comply with social norms when existing social norms are strong (Schultz et al. 2018). However, when social norms are weak, it is difficult to encourage behaviour change.

An emerging line of research suggests that in case of a weak existing social norm, the use of trending (also referred to as dynamic) social norms may be an effective way to change people’s behaviours (De Groot 2022; De Groot, Bondy, and Schuitema 2021; Loschelder et al. 2019; Mortensen et al. 2017; Sparkman et al. 2021; Sparkman and Walton 2017). TSN refer to information about how other people’s behaviour changes over time. As such, despite a weak social norm, one may indicate that the trend of a social norm is gradually changing in the desired direction. Mortensen et al. (2017) found that while brushing teeth more people reduced their water consumption after reading a TSN message, compared to a descriptive social norms message (i.e., emphasising that a minority engages in certain behaviour). Similar results were found in studies on reducing meat consumption (Sparkman et al. 2021; Sparkman and Walton 2017). As there is a weak social norm to test and frequently maintain private well, we used a TSN message as a starting point for our intervention design and combined this with emotional appeals.

Emotional appeals

Emotional appeals have long been used in health communication campaigns, which aim to arouse specific emotional responses, such as fear, guilt, or pleasure, which in turn may lead to changes in health behaviours (for reviews see Cameron and Chan 2008; Nabi 2015; Yousef, Rundle-Thiele, and Dietrich 2023). However, there are various reasons why emotional appeals fail at doing so. For example, if the audience’s involvement is low, an emotional response is not likely to occur (Turner 2012) or they feel in control over risks (Hooks, Schuitema, and McDermott 2019; Nordgren, Van der Pligt, and Van Harreveld 2007). This is likely to be the case for well-owners, as they are found to have low risk perceptions (Flanagan, Marvinney, and Zheng 2015; Hanchett et al. 2002; Hooks, Schuitema, and McDermott 2019; Jones et al. 2006) and are not very worried (Morris, Wilson, and Kelly 2016; Schuitema, Hooks, and McDermott 2020; Syme and Williams 1993).

Conceptually, perceived risks and worry are related, but distinct concepts. The perception of risk is conceptualised as an assessment of the likelihood that there will be harm (Brewer et al. 2007; Loewenstein et al. 2001; Sjöberg 1998; Weinstein 1999). In this definition, two dimensions underpin the perceived risks, that is, an assessment of the likelihood that some sort of harm will occur (‘likelihood of harm’) and an assessment of the consequences of risks (‘consequences of harm’). Thus, risk perceptions are a cognitive process in which the likelihood of a risk (e.g., how likely is it that my drinking water is contaminated?) and its consequences (e.g., how ill would I get if I drank contaminated water) are assessed.

Worry is originally also perceived as a cognitive process, that is, as a process that describes a gradual involvement with external threats due to its relevance to some anxiety-inducing content (Breznitz 1971). Tallis and Eysenck (1994), for example, propose that worry is the relatively automatic response to the evaluation of a thread in terms of its imminence, likelihood, and cost set against ones perceived self-efficacy. Here, worry is seen as a cognitive process that influences and informs situations that are perceived as a risk or a threat. Partly in response to this, Loewenstein et al. (2001) propose that worry can also be seen as an emotional response to a risk, which diverges from cognitive evaluations and hasa different impact on risk-taking behaviour (see also Schwartz, Sagiv, and Boehnke 2000; Sjöberg 1998, 2007; Slovic et al. 2002). Building on this, worry is increasingly conceputalised as an active emotional state that is linked to an adaptive behavioural response which aims to reduce a particular thread (Van der Linden, 2017; Smith and Leiserowitz 2014). In line with this, we consider worry as an emotional state that is the response to a perceived risk.

Research has shown that both risk perceptions and worry can transfer via social networks (Parkinson 2011; Parkinson and Simons 2009, 2012). For example, Hendriks, de Bruijn, and van den Putte (2012) found that the exposure to health messages on binge drinking led to more conversations about the negative effects of alcohol consumption, which in turn increased intentions to refrain from binge drinking. The suggestion that health messages can trigger social conversations about related health risks is based on the idea that emotional appeals can spread through social networks, typically referred to as emotional contagion, and implies that people’s emotions can be influenced by the way others feel (Parkinson 2011; Parkinson and Simons 2009). As risk perceptions and worry have been pointed out as emotions that can transfer via social networks (Parkinson 2011; Parkinson and Simons 2009, 2012), we designed an intervention that combines a TSN message with emotional appeals that could arouse risk perceptions and worry. The emotional appeals we focus on are fear framing and worry appraisal.

Fear framing (FF)

An overwhelming part of risk communication research focuses on how evoking fear can be used as emotional appeals (Nabi 2015). Fear appeals are persuasive messages that attempt to arouse fear by emphasizing the potential danger and harm that will befall individuals if they do not adopt the recommendations (Maddux and Rogers 1983). FF is the manipulation of the amount of fear that is aroused, thus the manipulation of the fear levels. Typically, a thread, the size of potential losses, or severity of risks are stressed in a high-fear frame (high-FF) message.

The general assumption is that FF messages cause arousal, which is why risk perceptions (Keller and Lehmann 2008; Lu, Xie, and Zhang 2013; Slovic 1987) and worry (Hine and Gifford 1991; Sobkow et al. 2020; Tanner et al. 2008) will increase. This is because arousal increases people’s motivation to elaborate on the message, and therefore people are more likely to pay attention to the message and ignore other cues and act on it. Consequently, we expect that high-FF messages will increase well-owners’ worry and risk perceptions about contamination levels of their drinking water.

H₁: A high-FF message leads to higher levels of worry (a) and risk perceptions (b) compared to a low-FF message.

Worry appraisals (WA)

The term emotional contagion is used to describe how emotions can move from one person to another. It means that the way someone feels can be transferred to other people (Parkinson 2011; Parkinson and Simons 2012). Worry is an emotional state that was shown to spread in social contexts (Parkinson 2011; Parkinson and Simons 2012; Schwab, Cullum, and Harton 2016), which suggests that feelings of worry can be transferred to other people. Parkinson and Simons (2012) suggested that the transfer of worry across social groups serves as a warning signal to other people and signals that there may be a risk or threat present. Consequently, a health message that stresses that others are worried about a risk (high-WA) may amplify worry and risk perceptions. Hence, we hypothesise that strong-WA messages can raise levels of worry and risk perceptions as follows:

H₂: A strong-WA message increases worry (a) and risk perceptions (b) more than a weak-WA or no-WA message

Interaction FF and WA

It can be argued that the intensity of the message depends on the combination of messages: a multiple message, for example sending a TSN message in combination with a high-FF message and strong-WA message would combine all intensifying features. We expect that such a combination would increase the level of arousal, and therefore increase worry levels and risk perceptions:

H₃: Worry (a) and risk perceptions (b) increase most under the condition of a high-FF and strong-WA message in comparison to conditions whereby either FF or WA are low/weak.

Worry and risk perceptions as mediators

As described above, we designed our intervention strategy to directly target worry levels and risk perceptions among well-owners. We expect that an increase in worry and risk perceptions, in turn, affect well-owners’ attitudes and behavioural intentions. In other words, we expect that our intervention indirectly influences well-owners’ attitudes and behavioural intentions, with worry and risk perceptions acting as mediators.

H₄: Worry (a) and risk perceptions (b) mediate the influence of the WA intervention on well-owners’ attitude to water testing.

H₅: Worry (a) and risk perceptions (b) mediate the influence of the FF intervention on well-owners’ attitude to water testing.

H₆: Worry (a) and risk perceptions (b) mediate the influence of the WA intervention on well-owners’ behavioural intentions to maintain their wells.

H₇: Worry (a) and risk perception (b) mediate the influence of the FF intervention on well-owners’ behavioural intentions to maintain their wells.

Moderated mediation effect: perceived control as moderator

Increasing risk perceptions and worry are only likely to lead to changes in attitudes and behavioural intentions if various preconditions are in place (Milne, Sheeran, and Orbell 2000). According to the Protection Motivation Theory (Rogers 1975), one important pre-condition is self-efficacy, which refers to the belief in one’s own ability to influence events (Bandura 1977). Self-efficacy is very closely related to the concept of perceived control, i.e., one’s perception of how well they are able to perform a given behaviour (Meyerowitz and Chaiken 1987; Nordgren, Van der Pligt, and Van Harreveld 2007). Previous studies show that well-owners who feel strong control, compared to those who do not, believe that that their water quality is higher and maintain their wells more frequently (Schuitema, Hooks, and McDermott 2020). Therefore, for those who feel in control, increasing worry and risk perceptions may lead to more positive attitudes to water testing and well maintenance. Those who perceive low control over contamination risks may not feel capable to act and thus one may expect that when people perceive low levels of control, increased worry levels and risk perceptions will not change their attitudes and behavioural intentions.

As we assume that worry and risk perception are mediators (H₄–H₇), this implies that there could be a moderated mediation effect (Hayes 2015). A moderated mediation effect implies a conditional indirect effect, depending on the value of the moderator. Concretely, we propose that worry and risk perceptions only mediate the influence of the interventions on attitudes and behavioural intentions under the condition that perceived control is high (as opposed to low):

H₈: Worry (a) and risk perceptions (b) increase well-owners’ attitude to water testing when perceived control is high as opposed to low.

H₉: Worry (a) and risk perceptions (b) do not increase well-owners’ behavioural intentions when perceived control is high as opposed to low.

Method

Procedure and sample

An online panel was used to collect data among 262 private well-owners in Ireland. As there is no registry of the exact number or location of private well-owners in Ireland, a panel study allowed us to pre-screen for private well-owners among the panel. All participants provided consent before filling out the survey and the study was accepted as a low-risk study by the Research Ethics committee of University College Dublin on the grounds that the research was done through anonymous surveys that do not involve the collection of identifiable data or health data. As a result, the research was exempt from seeking full ethics approval.

All respondents lived in rural areas, as that is where private wells are typically used. The mean age category was 45–54 years old. Educational and income levels were lower than the national average, which is common in rural areas. Females were largely overrepresented in our sample (n = 209), which is probably because in rural areas, women are more often involved in panel studies. Attempts to increase the number of male private well-owners largely failed because it was hard to identify male well-owners in the online panel. To investigate whether the overrepresentation of women could have influenced our results, we tested if men and women differed on the variables used in our analyses, and no significant differences were found. We also asked respondents whether they identified as the main householder (202 respondents did (74%); of which 153 were women). We assumed that people who identify as the main householder also had influence on the household decision making process.

Experimental design

There were six experimental conditions following a 2 (FF) by 3 (WA) between-subjects design (sample size per condition ranged between 42 and 46). In all six conditions, respondents saw the same TSN message, which read ‘Research found that 16% of private well-owners tested their water. This has increased from 11% in 2016’.

The manipulation of FF was done by stressing the severity of the health risks. In the high-FF condition, the text read ‘drinking contaminated water can pose serious health risks’ and was accompanied by a red triangle with a with exclamation mark, to stress danger. The low-FF condition was presented as ‘drinking contaminated water can pose some health risks’ and there was no exclamation mark presented.

There were three levels of WA. In the first level WA was not mentioned to create a benchmark. The second level presented a weak-WA message indicated that a minority of well-owners worried about contamination of their wells: ‘Only 20% of private well-owners worry about contamination of their wells’. The last level presented a strong-WA message, indicating that 16% of private well-owners worry about contamination of their wells’.

All messages began with the FF, followed by a WA and TSN message. After this, we provided some advice, which was the same in every condition: ‘You can test your water by contacting your local authority, the health service executive or a number of private water testing facilities’. The message was printed in front of a picture where a hand is pouring water in a glass from a jar. See Table 1 for an overview of the conditions, the exact order in which the manipulated texts were presented, and the visual presentation of the messages.

Table 1. Overview of experimental design and scenarios.

	Low-FF	High-FF
TSN Trending social norm
TSN + weak-WA Trending social norm + weak worry appraisal
TSN + strong-WA Trending social norm + strong worry appraisal

Variables

For information to be effective in changing people’s perceptions, attitudes, and behaviours, it is important that they believe the information is realistic and true. To check whether this was the case, we asked respondents to indicate how credible, realistic, and convincing they thought the information was after reading it. We used 5-point binary scales (not very credible/very credible; unreliable/reliable; unconvincing; convincing) to measure this. Next, respondents filled out a survey in which we measured the following variables.

Risk perceptions were measured with 3 items: ‘My water is a health risk’, ‘My water is not suitable for drinking purposes’, and ‘My water is generally free of contaminants [reversed]’. The items formed a reliable scale (α = .78; M = 2.14, SD = .92).

Worry was measured with 4 items, that is, ‘I am not concerned about my water quality [reversed]’, ‘I worry about contamination of my drinking water’, ‘I worry about contamination of my water harming the health of my family’, and ‘I worry about harmful bacteria in my drinking water’ (α = .79; M = 2.09, SD = 1.04).

Perceived control was measured with the items ‘I feel in control of my water quality’, ‘I have the ability to detect my contamination within my water supply’, ‘I am best placed to manage my water quality’, and ‘Contamination of my water is not easy to detect or contain [reversed]’ (α = .74; M = 3.11, SD = 0.83).

Attitudes to water testing was also measured with 4 items: ‘Testing my water is a good way to manage my water quality’, ‘Water testing is a waste of time [reversed]’, ‘Water testing is unnecessary [reversed]’, and ‘Water testing should be done every year’ (α = .78; M = 3.46, SD =0.39).

Behavioural intentions to well management was measured by asking respondent about the likelihood that they would carry out six types of behaviour. Going forward, how likely are you to carry out the following behaviours? ‘I will test my water on a regular basis’, ‘I will seek information on water testing in my area’, ‘I will seek information on drinking water contamination in general’, ‘I will seek more information on the maintenance of my well’, ‘I will seek information on water testing in general’, ‘I will not change my current behaviour at all’ [reversed]. These 6 items formed a reliable scale (α = .71 (M = 3.56, SD = 0.77).

All items were scored on 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree), apart from behavioural intentions, which ranged from 1 (very unlikely) to 5 (very likely).

Results

Check information quality

All six messages were perceived to be credible (across all 6 messages: M = 3.96; SD = 1.13), realistic (across all 6 messages: M = 3.98; SD = 1.10), and convincing (across all 6 messages: M = 3.84; SD = 1.16). Analyses of Variance (ANOVAs) showed no significant differences between the six messages in how credible (F (5, 256) = 0.68, ns), realistic (F (5, 256) = 0.73, ns), or convincing (F (5, 256) = 1.77, ns) the information was. Therefore, we concluded that the messages were generally considered as realistic and true.

Direct and indirect effects of interventions on attitudes and behavioural intentions

We tested the moderation-mediation model using the PROCESS macro developed by Hayes (2013) version three (template 14) to test our hypotheses (see also Figure 1). We ran this macro twice with attitudes (Y₁) and behavioural intentions (Y₂) as dependent variables separately. FF (X₃) and WA (X₁ and X₂) were included as independent variables. FF was used as moderator (X_1*X₃ and X_2*X₃) to test the interaction between WA and FF on worry and perceived risks. Worry (M₁) and perceived risks (M₂) were included as mediators and perceived control as moderator (W) of the effect of worry (W*M₁) and perceived risks (W*M₂) on the two dependent variables attitudes and behavioural intentions. All results are presented in Table 2.

Figure 1. Overall tested model.

Table 2. Influence of intervention strategies on attitudes and behavioural intentions with risk and worry as mediators and perceived control as moderator.

Direct effects
a-paths: X > M	Worry (M1)	Risk perc. (M2)	b-paths: M > Y	Attitude (Y1)	Intention (Y2)
a1X1: TSN + weak-WA	.14	−0.06	b1M1: Worry	.22*	.69***
a2X2: TSN + strong-WA	.51*	.43*	b2M2: Risk perceptions	.02	−0.56**
a3X3: FF	.50*	.42	b3W: Perceved control	.21*	.20
a4X1*X3: TSN + weak-WA*FF	−0.32	−0.21	b4W*M1: Perc. control*Worry	−0.03	−0.11
a5 X2*X3: TSN + strong-WA*FF	−0.41	−0.53	b5W*M2: Perc. control*Risk perc.	−0.04	.15**
Indirect effects
ab-paths: X > M > Y1 (Attitudes)				Coefficient	95% CI lower; upper
ab1X1: TSN + weak-WA > M1: Worry > Y1: Attitude				−0.00	−0.044; .040
ab2X2: TSN + strong-WA > M1: Worry > Y1: Attitude				.04	.002; .087
ab3X3: FF > M1: Worry > Y1: Attitude				.03	.000; .072
ab4X1: TSN + weak-WA > M2: Risk perceptions > Y1: Attitude				.02	−0.010; .053
ab5X2: TSN + strong-WA > M2: Risk perceptions > Y1: Attitude				−0.02	−0.053; .012
ab6X3: FF > M2: Risk perceptions > Y1: Attitude				−0.02	−0.051; .006
ab-paths: X > M > Y2 (Intentions)
ab7X1: TSN + weak-WA > M1: Worry > Y2: Intentions				−0.00	−0.108; .099
ab8X2: TSN + strong-WA > M1: Worry > Y2: Intentions				.10	.002; .212
ab9X3: FF > M1: Worry > Y2: Intentions				.08	−0.002; .171
ab10X1: TSN + weak-WA > M2: Risk perceptions > Y2: Intentions				.03	−0.020; .090
ab11X2: TSN + strong-WA > M2: Risk perceptions > Y2: Intentions				.03	−0.101; .018
ab12X3: FF > M2: Risk perceptions > Y2: Intentions				−0.02	−0.083; .011
Mediation moderation effects: conditional indirect effect of perceived control on X > M > Y
Moderation mediation: WA > M1: Worry > Y1: Attitudes				−0.00	−0.022; .008
Moderation mediation: WA > M2: Risk perceptions > Y1: Attitudes				−0.00	−0.017; .005
Moderation mediation: FF > M1: Worry > Y1: Attitudes				−0.01	−0.034; .012
Moderation mediation: FF > M2: Risk perceptions > Y1: Attitudes				−0.01	−0.029; .011
Moderation mediation: WA > M1: Worry > Y2: Intentions				.03	−0.017; .102
Moderation mediation: WA > M2: Risk perceptions > Y2: Intentions				−0.01	−0.090; .023
Moderation mediation: FF > M1: Worry > Y2: Intentions				−0.03	−0.084; .019
Moderation mediation: FF > M2: Risk perceptions > Y2: Intentions				.03	−0.015; .078

Note: Significant effects were made bold *p≤.05; **p≤.001; 95% CI = 95 Confidence Interval; Dummy coding: Norm: X₁ (1= TSN + weak-WA; 0 = remaining two norm conditions) and X₂ (1= TSN + strong-WA; 0 remaining two norm conditions); Frame (1 = high-fear; 0 = low-fear).

Effects of interventions on worry and perceived risks (a-paths)

Firstly, we investigated how the FF and WA interventions affected participants worry levels and perceived risks (the a-paths in Table 2; see also Figure 1). We found that the high-FF message resulted in increased worry levels compared to the low-FF message (a_{3_worry} = .50, p ≤.05), but no significant effect on perceived risks were found. Hence, H_1b was confirmed, whilst H_1a was rejected. For the WA intervention, the results showed that both worry (a_{2_worry}= .51, p ≤.05) and risk perceptions (a_{2_risk}= .43, p ≤.05) were higher in the strong-WA condition compared to the average of the other two WA messages. Thus H₂ was confirmed.

Next, we investigated if there was an interaction effect of the FF and WA interventions on worry and perceived risks. Specifically, it was hypothesised in H₃ that worry and perceived risks would be highest when a TSN message was combined with both the high-FF and strong-WA message as opposed to the other messages. The results did not confirm such an interaction effect of FF and WA on worry or risk perceptions, and thus H₃ was rejected.

In summary, the analyses indicated that worry was significantly higher for the strong-WA and high-FF message; risk perceptions were higher for the strong-WA message; and the expected interaction effect of the WA and FF interventions on worry and risk perceptions was not confirmed.

Effects of worry and perceived risks on attitudes and behavioural intentions (b-paths)

H_4, H_5, H₆, and H₇ proposed that the effect of the WA and FF interventions on well-owners’ attitudes and intentions were mediated by worry and risk perceptions. To assess if there were such mediation effects, we first looked at how the proposed mediators, worry and perceived risks, influenced attitudes and behavioural intentions (indicated by the b-paths in Table 2; See also Figure 1). The results showed that the more well-owners worried about the contamination risks of their wells, the more positive their attitude towards well water testing was (b_{1_attitude}= .22, p ≤.05) and the stronger their behavioural intentions to maintain their wells (b_{1_intention}= .69, p ≤.001). Contrary to the expectations, risk perceptions did not significantly predict attitudes and, unexpectedly, a negative relationship between risk perceptions and behavioural intentions was found (b_{2_intention}= −0.56, p ≤.001). The latter meant that well-owners had weaker intentions to maintain ther wells when they perceived high contamination risks. We believe that this may be explained by the moderating role of perceived control, which we will go into in the next section.

Worry and risk perceptions as mediators (ab-paths)

The next step is to investigate the mediating role of worry and risk perceptions on the influence of the interventions on attitudes and behavioural intentions (indicated by the ab-paths in Table 2; see also Figure 1). As expected, worry was found to mediate the effect of the WA message on attitudes to water testing and intentions to well maintenance. Specifically, a strong-WA message led to higher levels of worry, which in turn, resulted in more positive attitudes to water testing (ab₂ = .04 [95% CI: .002; .087]; confirming H_4a) and intentions to maintain their wells (ab₈=.10 [95% CI: .002;.212]; confirming H_6a). Worry did not significantly mediate the effect of the FF message on well-owners’ attitudes and behavioural intentions (thus rejecting H_5a and H_7a), despite the positive effect that the high-FF message had on worry levels. This suggested that the strong-WA message, but not the FF message, increased attitudes and behavioural intentions because it raised the extent to which well-owners’ worried about the quality of their drinking water.

Risk perceptions did not significantly mediate the effect of the WA and FF interventions on attitudes or behavioural intentions and thus H_4b, H_5b, H_6b, and H_7b must be rejected. In conclusion, worry, but not risk perceptions was a mediator for the effect of the WA interventions on attitudes and intentions.

Perceived control as moderator (mediation moderation effect)

We expected a moderated mediation effect (H₈–H₉), meaning that we expected the mediating role of worry and risk perception to depend on the level of perceived control over contamination risks.¹. The analysis did not demonstrate such a mediation moderation effect. When looking at the perceived control as a moderator (indicated by the b-paths in Table 2; see also Figure 1), we found that perceived control was a significant moderator in the mediation effect of perceived risks on behavioural intentions (b_{5_intention}= .15, p ≤.001; confirming H_9b)². However, as risk perception was not a significant mediator, it is no surprise that there was no significant moderation mediation effect found. The remainding mediation moderation effects were not statistically significant either, and therefore we conclude that perceived control did not moderate the mediation effect of worry and risk perceptions on well-owners’ attitudes (rejecting H₉) behaivoural intentions (rejecting H₈).

Conclusion and discussion

We designed and tested an intervention that aimed to encourage voluntary risk detection behaviours by tailoring it to its social context. Specifically, we aimed to encourage positive attitudes and well maintenance intentions among private well-owners, for which there is generally a weak social norm. We found that a TSN message (i.e., stressing that an increasing number of well-owners tested their water) combined with a strong-WA message (i.e., stressing that a majority of well-owners worried about contamination risks) indirectly influenced attitudes and behavioural intentions because it raised worry levels among well-owners. Thus, worry, but not risk perceptions, mediated the effect of the WA intervention on attitudes and behavioural intentions. This is line with other studies who found that worry was the main predictor of voluntary health risk detection or prevention behaviours. This was for example found for genetic testing for health risks (Cameron and Reeve 2006), cancer screening (Cameron 2008; Janssen et al. 2013), heart disease risk prevention (Lee et al. 2011), smoking (McCaul et al. 2007), and COVID-19 prevention behaviours (Sobkow et al. 2020). It has been reasoned that worry triggers negative arousal, which is important to stimulate positive attitudes and behavioural intentions (Leventhal 1970; Leventhal, Phillips, and Burns 2016) as it can enhance information processing (Meijnders, Midden, and Wilke 2001), feelings of responsibility to act (Bouman et al. 2020), and reduce uncertainty (Smith et al. 2007). We believe that our WA intervention may serve as a confirmation of one’s own (latent) worries, which may increase their confidence to act on them and overrule the effects of a weak social norm. Further research should confirm this hypothesis.

The result that worry can encourage voluntary health risk detection behaviour does not mean that we should unrestrictedly aim to raise worry levels. In fact, interventions should only raise worry levels moderately to be effective (Rothman et al. 2006). Increasing the extent to which people worry too much may be counterproductive as well as a risky or unethical intervention strategy (see Duke et al. 1993 for a comprehensive discussion about the ethical implications). In addition, it is also important that people can act on their worry (Sobkow et al. 2020; Tannenbaum et al. 2015; Witte and Allen 2000) and, for example, have access to knowledge, test facilities and practical advice on which actions can address these concerns.

Our findings further suggest that the use of FF messages is not the most effective way to change attitudes and behaviours of health risk detection behaviours. Although we did see that worry increased when well-owners read the high-FF message, there was no indirect effect on well-owners’ attitudes and behavioural intentions. Our findings contrast with research on health risks that suggests increasing perception of risks as a tool to change people’s attitudes or behaviours towards those risks (Hanchett et al. 2002; Hynds, Misstear, and Gill 2013; Lund and Helgason 2005). Contrary to these studies, we found that stronger risk perceptions reduced the intention to maintain wells.

In sum, our results suggest that interventions based on trending social norms and worry appraisals can raise feelings of worry, which is more effective in promoting voluntary health risk detection than increasing risk perceptions. When designing this type of intervention, caution is needed with its specific wording as boomerang effects can occur (Schultz et al. 2007, 2018). For example, because we used a TSN message as a baseline in all interventions and combined this with emotional appeals messages, we cannot separate the effect of both messages.

We note that the effect sizes of the interventions were rather small, which is common for interventions based on social influence techniques (Abrahamse and Steg 2013). Therefore, we believe that health communication and social influence interventions like ours should be combined with other (policy) measures and changes. In the case of well water management, one may think of facilitating well-owners in managing their wells, for example by providing a (free) water testing service. In addition, when new wells are drilled and constructed there are requirements to adopt current best practices in relation to well-head design that provides protection from surface-derived contaminants and, more broadly, good groundwater governance and management is crucial (Di Pelino et al. 2019; Megdal 2018).

One caveat of our results is that we had a large overrepresentation of women. Women are known to worry more and perceive risks differently than men (McQueen et al. 2008; Sjöberg 1998) and be more sensitive to emotional appeals (Dube and Morgan 1996). Therefore, our results need to be interpreted with care and we cannot generalise our results to all (male) well-owners. However, gender differences in worry and risk perceptions are typically found in research that focuses on individual risks, like specific illnesses. The risk of well water contamination applies to a household and typically require joint decisions made by household members. Men and women play a different role in such intra-household decisions, but both contribute to them (De Palma, Picard, and Ziegelmeyer 2011). As 74% of our sample identify as the main householder, we assume they have influence on household-based decisions around well management, but we have no data support this assumption.

In conclusion, we have shown the potential of an intervention approach to encourage voluntary risk detection behaviours that moderately raises the extent to which people worry about risks. This approach relies on the social context of the behaviour at stake. Specifically, when a social norm to act is weak, stressing a trending social norm in combination with a strong worry appraisal could play a role in the promotion of voluntary risk detection behaviours.

Acknowledgements

The authors are grateful for the input of Prof Frank McDermott on earlier drafts of the paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data available on request from the corresponding author.

Additional information

Funding

Funding for this project was received from the Irish Research Council fund for Policy and Society, co-funded by the Geological Survey of Ireland (GSI) and the Irish Research Council (IRC), Grant no. RfPS/2016/61. The also project benefitted from the support of iCRAG (SFI Research Centre in Applied Geosciences) under grant number 13/RC/2092_P2 and co-funded under the European Regional Development Fund.

Notes

1 Note there was a significant main effect of perceived control on attitudes to water testing, implying that well-owners who felt more in control had more positive water testing attitudes. As we did not have specific hypothesis about main effects for perceived control, we do not go further into this effect.

2 Asimple slope analysis revealed that for those who perceived low levels of control, behavioural intentions decreased when risk perceptions were high (β = 0.22, p ≤.01), which was opposite to our expectations. This may be because high risk perceptions in combination with feeling incapable to act (indicated by low perceived control) could result in a type of in-action or paralysis to act. For those who perceived high control, risk perceptions did not influence their intentions to maintain their wells (β = 0.03, ns). Perhaps they felt that they already maintained their wells sufficiently or overly confident in their abilities to detect, monitor and prevent contamination risks.