Communicating Evidence About the Causes of Obesity and Support for Obesity Policies in British and US populations: Two Population-Based Survey Experiments

Background: Public support for numerous obesity policies is low which is one barrier to their implementation. One reason for this low support is the tendency to ascribe obesity to failings of willpower as opposed to the environment. Correlational evidence supports this position: beliefs about the causes of obesity are associated with support for policies that aim to reduce obesity. However, the experimental evidence for the causal nature of this association is mixed. Methods: In two experimental studies participants were randomised to receive no message, messages about the environment’s inuence on obesity (Study 1 & 2), or messages about the environment’s inuence on human behaviour (Study 1). We investigated whether communicating these messages changed support for policies to tackle obesity and beliefs about the causes of obesity. Participants were recruited from nationally representative samples in Great Britain (Study 1 & 2) and the USA (Study 2) (total N = 4391). Study 2 was designed to replicate two previously published studies. Results: While the belief that obesity is caused by the environment was associated with support for obesity policies, neither study found evidence that communicating the messages increased support for obesity policies or strengthened beliefs about the environment’s role in obesity. Conclusions: The current study replicates earlier studies reporting an association between beliefs about obesity and the environment but does not replicate two earlier experimental studies that suggested the association is causal. The evidence reported here suggests that people’s beliefs about the causes of obesity are resistant to change and therefore not a promising avenue to increase support for obesity policies. cheap, unhealthy foods around”. The belief that human behaviour is inuenced by the environment (Behavioural Causal Beliefs: Environment) was measured with two items (r = .56) that were adapted from the Obesity Attributions items described above: “People’s behaviour is strongly inuenced by their environment and surroundings” and “The cost and availability of products inuence what people buy and choose”. The causal belief items were rated on a seven-point scale (1 = Strongly disagree; 7 = Strongly agree) and were presented in counter-balanced order. These questions were given to all participants, regardless of intervention group. marketing made two current was taking

"If a person is aggressively competitive in his behavior, is he this kind of person, or is he reacting to situational pressures?". The formation of causal beliefs is often based on implicit assumptions and incomplete data, and thus prone to bias (Ross, 1977). One such bias is the correspondence bias -or fundamental attribution error -in which people overestimate the role of the self in the causes of behaviour and discount situational in uences (Gilbert & Malone, 1995). There is also some evidence for a self-serving bias whereby people are more likely to attribute a negative behaviour to personal characteristics when they are observing others engage in this behaviour and less likely to attribute this same behaviour to personal characteristics when they themselves are engaged in this behaviour (Malle, 2006). Given the self-protective effects of causal attributions, it is unsurprising that they may at best only be a rough approximation of reality.
Research based on attribution theory has led to the well replicated result that causal beliefs are associated with attitudes. For example, the belief that homosexuality is a choice is associated with negative attitudes toward people who are homosexual and with less support for equal rights and samesex marriage (Haider-Markel & Joslyn, 2008;Sakalli, 2002;Whitehead, 2014). The belief that people who are poor are responsible for their own misfortune is associated with negative attitudes and less support for policy measures for those who are poor (Linos & West, 2003;Zucker & Weiner, 1993). Similar results are reported for other groups: people who are transgender, people with a mental illness, people with criminal records, and -most relevant to this paper -people who are obese (Crandall et  Most people believe that being overweight or obese is due to a lack of personal responsibility, with fewer people acknowledging the in uence of the environment, such as the widespread availability of unhealthy foods and lack of public space for physical activity ( in uences may explain -in part -the relatively low public support for government intervention to tackle obesity by changing environments (Beeken & Wardle, 2013;Hilbert et al., 2007;Mazzocchi et al., 2015). As this past research is correlational, it is di cult to determine if these beliefs are consequential -i.e., whether believing that the environment causes obesity leads to more favourable attitudes toward policies that help to reduce obesity.
Several studies have attempted to increase support for obesity policies by drawing on attribution theory and communicating information about the environment's in uence on obesity. For example, Pearl and Lebowitz (2014) communicated a message that included statements highlighting how the high availability and low price of unhealthy foods contribute toward obesity. Participants who read these messages reported greater support for policies to reduce obesity compared to participants in the control group who did not read any message. Subsequent studies have either found smaller effect sizes than those reported by Pearl (Garbarino, Henry, & Kerfoot, 2018). It is unlikely that differences in intervention content explain these mixed effects. Although the interventions varied across studies, they shared several key messages about the in uences on obesity, including: portion size (Garbarino et  Given the mixed results from the experimental literature -which have become apparent amid concerns about the reproducibility of existing research (Open Science Collaboration, 2015) -robust studies are needed to reduce the existing uncertainty. In addition to the speci c approach of changing obesity attributions to in uence support for policies to reduce obesity, we also investigated whether people's broader attributions about human behaviour may in uence support for obesity policies that aim to change behaviour.

Study 1
The aim of Study 1 was to identify messages that would be most effective at changing causal beliefs for use in Study 2. A further aim was to investigate whether and to what extent these messages changed support for obesity policies.

Method
This study was pre-registered with the Open Science Framework (DOI: https://osf.io/tzy86/? view_only=510388557da84b95b7ed1a346cce1fae). There was one deviation from the registered protocol: we increased the sample size from 375 to 1681 to increase statistical power, in line with a recent study with similar methods (Ortiz et al., 2016). This decision was made after the protocol was registered, but before commencement of data collection. Supporting data can also be found on the OSF (https://osf.io/qkgdw/?view_only=322f2eab83a945bc99a487aeec1841ff).

Participants
A nationally representative sample from Great Britain (n = 1681) was recruited via YouGov's online panel (www.yougov.co.uk). The recruitment method used quotas for age, gender, social grade, education, region, political attention, and voting in the 2017 General Election and 2016 EU referendum. This sample size ensured similar group sizes from a comparable study (Ortiz et al., 2016). The exact effect size from Ortiz et al. (2016) could not be calculated; however, a power calculation suggested that this sample size would provide 80% power to detect small effects between two groups (d = .26) after a Bonferroni adjustment (α = .0125). Data were collected between 29 th and 30 th August 2018. After applying weighting, mean age = 48.33 (SD = 16.87) and 51.6% were female. See Supplemental Table S1 for the full demographic characteristics of the sample.

Design
The study was an online, between-subjects experiment in which participants were randomly allocated to one of ve groups differing in the messages that they received about (i) the environment's in uence on obesity, or (ii) the environment's in uence on human behaviour in general: After viewing the randomly assigned messages, participants completed a short questionnaire. The randomisation and questionnaire were programmed in YouGov's software.

The interventions
The interventions comprised four messages (see supplement). We hypothesised that the two obesity messages (Groups 2 & 3) would strengthen the belief that the physical environment causes obesity when compared to the control group (Group 1); and that the two behaviour messages (Groups 4 & 5) would strengthen the belief that the physical environment in uences human behaviour more generally when compared to the control group (Group 1). No directional predictions were made over which message would be most effective at changing the target belief or public support. It was planned that one message from each of the two sets -obesity and human behaviour -would be selected for use in Study 2 based on their effectiveness at instilling or strengthening the target causal belief and on the self-reported subjective comprehension of the message's content.
The message used in Group 2 was adapted from Pearl and Lebowitz (2014), with images of the obesogenic environment added below the text as the addition of images to text has been shown to increase attention, comprehension, and recall of information (Houts, Doak, Doak, & Loscalzo, 2006). These images included examples of the message content, such as the high availability of less healthy foods and aggressive food advertising. The obesity message presented in Groups 3 was developed speci cally for the present study based on aspects of environment that have been linked with obesity: cost, availability, and marketing (Kirk, Penney, & McHugh, 2010;Swinburn et al., 2011). References to "evidence" and "research" were added to increase the persuasiveness of the message (Allen & Preiss, 1997; Pornpitakpan, 2004; Reynolds, Stautz, Pilling, van der Linden, & Marteau, 2020). The messages in Groups 4 and 5 were designed to mimic the structure of the two obesity messages, however the focus was changed from obesity to human behaviour in general, in which two examples are given: which mode of transport people use and whether people purchase items in single-use plastics. These examples were chosen as they are daily behaviours that are in uenced by the same environmental factors of cost, availability, and marketing.

Measures
Policy support. Support for three obesity prevention policies was assessed using a single item adapted from earlier research (Reynolds et al., 2018): "Do you support or oppose the new policy?" rated on a seven-point scale (1 = Strongly oppose; 7 = Strongly support). The three policies were: a 20% tax on confectionary, reduction in the size of unhealthy ready meals, and banning advertising for unhealthy foods during children's television. The three policies were presented as a package, so participants rated whether they supported or opposed the implementation of all three. These three policies were chosen as they have not been implemented in the UK and research suggests that they would be effective ( The low cost, widespread availability and marketing of unhealthy foods are to blame for the high rates of obesity" and "People are obese because there are so many cheap, unhealthy foods around". The belief that human behaviour is in uenced by the environment (Behavioural Causal Beliefs: Environment) was measured with two items (r = .56) that were adapted from the Obesity Attributions items described above: "People's behaviour is strongly in uenced by their environment and surroundings" and "The cost and availability of products in uence what people buy and choose". The causal belief items were rated on a seven-point scale (1 = Strongly disagree; 7 = Strongly agree) and were presented in counter-balanced order. These questions were given to all participants, regardless of intervention group.
Comprehension. Participants rated the clarity and comprehension of the intervention that they were randomised to read (Subjective Comprehension) with two items (r = .78): "I found the information in the summary I just read clear" and "I found the information in the summary I just read easy to understand". The items were rated on a seven-point scale (1 = Strongly disagree; 7 = Strongly agree) and were presented in counter-balanced order.
Other variables. The research agency provided demographic data including age, gender, socio-economic status (Lambert & Moy, 2013), education (adapted from: Clarke, Sanders, Stewart, & Whiteley, 2003), and region. Educational achievement was recoded into three categories: low education (no education, GCSEs (General Certi cate of Secondary Education) or similar); medium education (A-levels, non-degree teaching quali cations, or similar); and, high education (degree awards or higher). Socio-economic status was also recoded into three categories: low (DE), medium (C1C2), and high (AB). The recoding was done in accordance with previous research (Reynolds et al., 2019).

Analyses
YouGov provided sampling weights to improve the representativeness of the sample, which were applied for all analyses in this study. Potential confounding variables (gender, age, SES, education, and region) were compared across groups using a percentage method to assess chance imbalances following randomisation (Moher et al., 2010). Several chance imbalances above 5% points were identi ed for all ve variables across the groups. Thus, the main analyses used ordinary least squares regression, controlling for these ve demographic characteristics, to test the main effects of experimental group on support for policies to tackle obesity and beliefs about the causes of obesity. Sensitivity analyses were conducted in which covariates were not included, to determine whether the main pattern of results would change (see supplement). Model diagnostics were examined and were acceptable.
The criteria for signi cance was set at α = .0125 for all four outcomes (α = .05/4 = .0125), after applying a Bonferroni adjustment. Outliers (±3SDs from the mean) on continuous variables were removed. 30 outliers were removed (2%) from the Behavioural Causal Beliefs: Environment variable, and 31 were removed from the Subjective Comprehension variable (2%). There were no other outliers. Sensitivity analyses were conducted in which outliers were not excluded, to determine whether the main pattern of results changed. Cohen's d statistics are covariate adjusted.

Policy support
None of the intervention messages increased support for the obesity prevention policies when compared to the control group (all ps > .0125, see Table 1 for full results).
[INSERT TABLE 1 HERE] Causal beliefs (manipulation checks) There was a statistically signi cant effect of the message presented in Group 2 pertaining to the environment's role in obesity on beliefs about the environment's in uence on human behaviour. Participants who received this intervention message believed that the environment had a greater in uence on human behaviour than those in the control group, B = 0.18, 95% CIs [0.04, 0.31], p = .009, d = .20, representing a small increase on the 1-7 rating scale (see Tables 1 for full results).
There was also a statistically signi cant effect of Group 3's message about the environment's in uence on obesity on beliefs about the environment's in uence on obesity. Participants who received this intervention believed that the environment had less in uence on obesity than those in the control group, B = -0.27, 95% CIs [-0.47, -0.07], p = .007, d = . 21. This effect was in the opposite direction to that which was predicted.
There were no other statistically signi cant effects of any intervention, compared to the control group, on beliefs about the environment's in uence in obesity or in human behaviour in general (all ps > .0125).

Subjective Comprehension
There were no statistically signi cant differences between the two obesity messages in terms of subjective comprehension, B = -0.04, 95% CIs [-0.19, 0.11], p = .623, d = .04. However, the human behaviour message communicated in Group 5 was rated as signi cantly clearer and easier to understand when compared to the human behaviour message in Group 4, B = 0.21, 95% CIs [0.07, 0.36], p = .005, d = .22, a small increase on the 1-7 rating scale (see Table S2 for descriptive statistics).

Discussion
The primary aim of this study was to select messages that would be most effective at changing causal beliefs for use in Study 2. The results showed that none of the interventions changed the target belief in the hypothesised direction. In keeping with this, there were no changes in support for obesity-related policies. However, we found that participants who read one of the messages containing information about the environment's in uence on obesity (Group 2) were more likely to believe that the environment in uenced human behaviour than those in the control group. Also, those who read the second obesity message (Group 3) were less likely to believe that the environment in uenced obesity than those in the control group. Despite changes in causal beliefs from those who read these two messages, there was no subsequent change in support for policies among these participants. These ndings suggest that messages designed to induce the belief about the environment's in uence on both obesity-related behaviours and human behaviour in general do not directly in uence attitudes toward obesity policies.
The lack of evidence for an effect of any of the messages on policy support was unexpected as the message used in the Group 2 had previously been found to increase public support (Pearl & Lebowitz, 2014). Other studies using similar interventions also showed changes in attitudes toward obesity policies and beliefs about the causes of obesity (Ortiz et al., 2016). We offer two possible explanations about the con icting results between our study and those of two others that reported statistically signi cant effects (Ortiz et al., 2016; Pearl & Lebowitz, 2014). The rst explanation concerns cultural differences. Studies conducted by (Ortiz et al., 2016;Pearl & Lebowitz, 2014) were both conducted in the USA. The current study was conducted in Great Britain. There is some evidence that US populations are less supportive of regulation to change health-related behaviour than are those in Great Britain which may affect the sensitivity of the two populations to messages targeting these attitudes (Ipsos MORI, 2012; Petrescu, Hollands, Couturier, Ng, & Marteau, 2016). Second, there may be other differences in sample characteristics that affected responses to the interventions. The sample used in Pearl and Lebowitz (2014) consisted solely of people who were overweight or obese. It is possible that people who are already overweight or obese may be more likely to revise their beliefs about the causes of obesity when presented with information about the environmental in uence on obesity-related behaviours. However, this hypothesis remains untested.

Study 2
The results of Study 1 were originally intended to inform which interventions to use in Study 2. However, as none of the four interventions changed the target belief in the hypothesised direction and none changed public support, we decided to re-examine the central claim of this research, which we assumed to be true prior to conducting Study 1. The revised aim of Study 2 was to determine if we could replicate previously published effects on the communication of obesity attribution messages and support for obesity policies. To do this we decided to use two interventions used in two previously published studies reporting effects of communicating information about the environment's in uence on support for obesity policies (Ortiz et al., 2016;Pearl & Lebowitz, 2014). We decided to reuse the intervention from Pearl and Lebowitz (2014) that we tested in Study 1, but we removed the images to ensure that the presentation of the message was identical to the original study. The second message that we selected to test also successfully changed beliefs and attitudes in its original study (Ortiz et al., 2016). A further goal was to test two explanations for the con icting results between Study 1 and previous research: differences in nationality and BMI. To do this we tested for interactions between intervention, country (England vs USA), and BMI. Based on prior research, Study 2 was designed to test three pre-registered hypotheses:

I
Communicating messages that attribute obesity to the environment will (a) increase support for obesity prevention policies and (b) strengthen the belief that the environment causes obesity II These effects will be greater amongst: 1. participants from the USA 2. participants who are obese or overweight III Participants from England will report greater levels of support for obesity prevention policies and will be more likely to believe that the environment causes obesity.

Method
This study was pre-registered with the Open Science Framework (DOI: https://osf.io/juemn/? view_only=6f26ee717e254c168172c60de5e912e0). There was one deviation from the registered protocol. The criteria for signi cance was changed as a principal components analysis (PCA) suggested a two-factor solution for our primary outcome. This change is described in the analyses section below.

Participants
Two nationally representative samples from England (n = 1397) and from the USA (n = 1315) were recruited via YouGov's existing online panels. The recruitment method used quotas including age, gender, and education (for both countries); social grade, region, political attention, voting in the 2017 General Election and 2016 EU referendum race (for England only); and race, voter registration, and voting in the 2016 Presidential Election (for the USA only). This sample size ensured similar group size from a comparable study (Ortiz et al., 2016) and approximately 30 times greater group size from another (Pearl & Lebowitz, 2014). A power calculation suggested that the current sample size would provide 80% power to detect small effects between two groups (d = .14) after a Bonferroni adjustment (α = .025) and when combining the two samples. Data were collected between 10 th and 13 th December 2018. After applying weights, the mean age = 48.36 (SD = 17.01) and 51.5% were female for the English sample and mean age = 47.31 (SD = 17.69) and 51.4% were female for the USA sample. See Supplemental Table S3 and S4 for the full demographic characteristics of the sample.

Design
We conducted an online between-subjects experiment, in which participants were randomly allocated to one of three groups (see Box 1) differing in their exposure to messages about the environment's in uence on obesity.
Group 1: Control group: received no message.  The randomisation was conducted using the research agency's software. Participants completed a short questionnaire after receiving the interventions. The study was conducted simultaneously in England and the USA.

The interventions
Two messages were taken from previous studies. These messages below highlight several aspects of the environment that have been shown to in uence obesity: cost, availability, portion size, placement, and marketing (Ejlerskov et  There are also parts of [America/England] in which there is limited access to grocery stores and fresh foods, and high availability of fast food restaurants and convenience stores that sell less healthy food. Restaurant portion sizes have increased in recent years, leading people to eat more food overall, and research has indicated that food advertisements and marketing increases consumption of unhealthy foods. Therefore, aspects of the food environment play a role in causing obesity." For example, food advertising can lead to the selection of unhealthy food and beverages. Certain food additives, such as extra salt, sugar, and caffeine, can also increase the desire for unhealthy food. And the placement of snack food and sugary beverages at checkout counters, especially in non-grocery retail stores, can often result in unintended food purchases and overeating. Consumers should be able to make their own dietary choices. But they also need to be free from the in uence of heavy advertising, exposures to habit forming food ingredients, and invasive food product placement."

Measures
Primary outcome(s). Acceptability of seven policies, randomly ordered, was assessed using one response item for each (Reynolds et al., 2018): "Do you support or oppose the new policy?" rated on a seven-point scale (1 = Strongly oppose; 7 = Strongly support). These seven policies were: a 20% tax on confectionary; reduction in the size of unhealthy snack foods; banning advertising for unhealthy foods during children's television; a policy to increase the availability of healthy foods in worksites, schools, and hospitals; a limit on the maximum size of sugar-sweetened beverages in fast food restaurants; calorie labels on restaurant menus; and a ban on unhealthy snack foods in schools. We used a more comprehensive set of policies in Study 2 to match the policies assessed in the studies from which we sourced the interventions (Ortiz et al., 2016;Pearl & Lebowitz, 2014). These seven items were converted into two outcomes: support for encouraging policies and support for discouraging policies (see Analyses section).
Causal beliefs (manipulation checks). The belief that obesity is caused by the food environment, genetics, and a lack of willpower were each measured with two response items (r = .73; .70; .77, respectively) (Reynolds et al., 2018): "[Cause] is to blame for Obesity" and "People are obese because of [Cause]". Each was rated on a seven-point scale (1 = Strongly disagree; 7 = Strongly agree). These items were presented in counter-balanced order.
Other variables. BMI was calculated from self-reported height and weight. The research agency provided demographic data including age, gender, socio-economic status (Lambert & Moy, 2013), education (Adapted from: Clarke et al., 2003), and region. For the English sample, educational achievement was recoded into three categories: low education (no education, GCSEs or similar); medium education (Alevels, non-degree teaching quali cations, or similar); and, high education (degree awards or higher). Socio-economic status was also recoded into three categories: low (DE), medium (C1C2), and high (AB). For these transformations see Methods section reported in Study 1.
For the USA sample, educational achievement was recoded into four categories: low education (no high school, high school graduates), medium-low education (some college, 2 year college), medium-high education (4 year college graduate), and high education (post-graduate degree).

Analyses
A PCA suggested a two-factor solution for the policy support items that explained 67% of the variance: (1) support for policies to discourage consumption of unhealthy foods and drinks (Discouraging policies); and (2) support for policies to encourage consumption of healthy foods (Encouraging policies). Support for Discouraging policies ranged from -2.07 (strongly oppose) to 2.13 (strongly support) and support for Encouraging policies ranged from -2.68 (strongly oppose) to 1.68 (strongly support). See Table S6 for factor loadings.
YouGov provide sampling weights to improve the representativeness of the sample, which were applied for all analyses in this study. The main analyses used hierarchical OLS regressions to test the main effects and interactions between country, intervention group, and BMI on support for policies to tackle obesity and beliefs about the causes of obesity. The pre-registered criteria for signi cance was set at α = .05 for the primary outcome (policy support), and α = .05/4 = .0125 for the three secondary outcomes, after applying a Bonferonni multiplicity adjustment. However, as the PCA suggested a two-factor solution for policy support, this was changed to α = .025 for the co-primary outcomes and α = .01 for the three secondary outcomes.
Potential confounding variables (SES, education, gender, age and region) were compared across groups using a percentage method to assess chance imbalances following randomisation (Moher et al., 2010). Several chance imbalances above 5% points were identi ed for all ve of these variables across the groups and therefore gender and age were included as covariates in the models as a sensitivity analysis.
It did not make sense to control for SES, region, and education as these were measured with different items across the English and USA samples. Sensitivity analyses were conducted in which covariates were not included into the models, to determine whether the main pattern of results would change (see supplement). Model diagnostics were examined and were acceptable.
Outliers (±3SDs from the mean) on continuous variables were removed. 47 outliers were removed (2%) from the Encouraging policies variable, and 50 were removed from the BMI variable (2%). There were no other outliers. Sensitivity analyses were conducted in which outliers were not excluded, to determine whether the main pattern of results would change. Data in Figure 1 and 2 were dichotomised (1-4 = 0, 4.01-7 =1) to indicate the proportions of participants that found each policy acceptable (i.e., those rating above the scale midpoint). These dichotomised data are provided to aid interpretation, and are not used in any inferential analyses. Cohen's d statistics are covariate adjusted.
In an exploratory analysis, we used the two one-sided tests (TOST) procedure (Lakens, Scheel, & Isager, 2018), to evaluate whether our results were equivalent to those reported in the two studies that we were aiming to replicate. Equivalence bounds were set as Δ L = -.10 and Δ U = .10 given the size of effects in similar elds. This provided two p-values by using t-tests of both below the lower bound (Δ<Δ L , the lower tail p-value) and above the upper bound (Δ U >Δ, the upper tail p-value) with adjusted degrees of freedom using the Sattherwaite method (Lakens, 2017). Equivalence is shown if the largest p-value is signi cant (i.e. data is consistent with being within the two boundaries), and therefore only one p-value requires reporting. The inference criterion was set at α = .025 in line with the co-primary analyses.

Results
Policy support There were no statistically signi cant effects of the interventions on the primary outcomes of support for policies. There was no effect of the Group 2 message when compared to the control group on support for encouraging policies, B = 0.05, 95% CIs [-0.03, 0.14], p = .215, d = .06, or for discouraging policies, B = -0.01, 95% CIs [-0.10, 0.08], p = .823, d = -.01 (see Table 2). The equivalence tests were non-signi cant for both encouraging policies, t(31.22) = 4.35, p = 1.00, and discouraging policies, t(30.91) = 3.94, p = 1.00. This suggests that neither of these two analyses was statistically equivalent to the results of the original study based on equivalence bounds of Δ L = -.10 and Δ U = .10 (Pearl & Lebowitz, 2014) where the effect size was larger d = .94.
There were also no statistically signi cant interaction effects on either of these two policy support outcomes (see Table S7). This includes two-way and three-way interactions between intervention group, country, and/or BMI.
There was a statistically signi cant main effect of country on policy support. Participants from the USA reported less support for both sets of obesity prevention policies compared to English participants: Encouraging policies, B = -0. 15 Beliefs about the causes of obesity (manipulation checks) There was no statistically signi cant effect of the interventions on the belief that the environment in uences obesity, the belief that genetics in uence obesity, or the belief that a lack of willpower in uences obesity (see Table 3). There were also no statistically signi cant interaction effects on any of these three causal belief outcomes (see Table S8). This includes two-way and three-way interactions between intervention group, country, and/or BMI.
There was a statistically signi cant effect of country on two out of three causal beliefs. American participants were more likely than English participants to believe that genetics in uences obesity, B = 0.61, 95% CIs [0.51, 0.71], p < .001, d = .46, whereas English participants were more likely than USA participants to believe that a lack of willpower in uences obesity, B = -0.22, 95% CIs [-0.33, -0.11], p < .001, d = .15.

Discussion
The aim of Study 2 was to replicate effects of messages previously found in two separate studies to change beliefs and/or attitudes about obesity and obesity policies (Ortiz et al., 2016;Pearl & Lebowitz, 2014). The results of Study 2 did not replicate these ndings. The results of the current study did, however, support the main conclusion reached in Study 1: there is no evidence that communicating information about the environmental causes of obesity changes support for policies to reduce obesity.
We tested two hypotheses to explain why the interventions used in Study 1 did not change support for policies in a manner consistent with previous studies. The rst hypothesis was that participants from the USA may be more sensitive to the messages and thus more likely to change their beliefs and attitudes. Study 2 provided no evidence that country of residence (USA vs. England) moderated the effect of messages designed to communicate the impact of environmental causes on obesity on policy support.
The second hypothesis, that overweight or obese participants are more likely to change their beliefs and attitudes when presented with information about the environment's in uence on obesity, was also unsupported. There was also no evidence of a three-way interaction effect between BMI, country, and intervention. In summary, there was no evidence that two of the notable differences between Study 1 and previous studies -BMI and country of residence -accounted for the lack of replication. One possible factor explanation for the lack of replication is time; it may be that more people are now aware of the environment's in uence on obesity and therefore ceiling effects are observed.
The results of Study 2 also showed differences in beliefs and attitudes between the USA and English Samples. Participants from the USA, relative to participants from England, were more likely to believe that genetics and less likely to believe that the environment and a lack of willpower caused obesity. Support for policies to reduce obesity was higher in England than in the USA in keeping with existing evidence

General discussion
The results of two studies suggest that communicating information about the environment's in uence on obesity does not change attitudes towards policies that aim to reduce obesity. This does not replicate earlier studies that reported statistically signi cant effects using the same messages as those used in the current study (Ortiz et  advertising, and the work environment in obesity (Ortiz et al., 2016). The message of interest only changed one of these: it strengthened the belief that advertising foods in uences obesity. This therefore provides some evidence that changing causal beliefs can increase policy support although the current study did not support these conclusions.
As the current evidence suggests that beliefs about the causes of obesity are resistant to change, more innovative methods will be needed that can combat the repeated exposure to media which regularly emphasises that obesity is just a symptom of poor self-control ( (2014) used a sample solely of people who were overweight or obese. We addressed this by testing whether BMI moderated the effect of intervention, which it did not. Second, our primary and secondary outcomes differed from the outcomes used in Ortiz et al. (2016) and Pearl and Lebowitz (2014). This was done as several of the policies used in these studies have already been implemented in the UK. The differences in outcomes was most notable for Study 1, whereas in Study 2 we used 5/5 of the policies used by (Pearl & Lebowitz, 2014) and 2/4 of the policies used by (Ortiz et al., 2016). Differences in outcome such as these are common in this literature, and while unlikely to account for the lack of replication, may affect the estimates of effect size. A further limitation is that the images which were added to the messages in Study 1 were not compared against a text & no image group, so it is unclear what effect the addition of these images may have had on the textual messages used in prior studies.

Conclusions
Across two pre-registered studies -comprising samples from the USA and Great Britain -we found no evidence that communicating information on the environment's in uence on obesity-related behaviours increased support for policies designed to reduce obesity. These results do not replicate earlier research that used the same interventions to address this question. Exploratory correlations reported in the supplement are consistent with previous studies that identify a relationship between the belief that the environment causes obesity and support for obesity polices. However, the current results suggest that that people's beliefs about the causes of obesity are strongly held and resistant to change. If these beliefs Department of Health and Social Care or its arm's length bodies, the National Institutes of Health, and other Government Departments. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Authors' contributions TM and JR conceived the idea for the study. TM, JR, and MV contributed to the design. JR drafted the manuscript. TM, MV, MP, MH, and KR provided comments on earlier drafts. All authors read and approved the nal version of this manuscript.