1. Introduction

According to the International data collated by the World Health Organisation (World Obesity Federation, 2022), the prevalence of overweight and obesity amongst children aged 5 to 19 years increased nearly threefold in boys, and more than doubled in girls between 1975 and 2016. Similarly, adults living with obesity rose 138% along that period, with a 21% rise between 2006 and 2016. The severity of the problem is such that leading international experts predict that in the long run, in developed countries like the UK and the USA, it will overtake tobacco as the primary cause of preventable death (Maddock, 2004). In the case of Spain, in 1990 the proportion of adults living with obesity was 11.5%, and in 2014 it was 22.8%; regarding overweight or obesity, the figures rose to 58.9% in 1990 and 69.3% in 2014. The recent figures are very similar. In fact, one of the principal causes of death in Spain is obesity, despite the well-established benefits of the so-called ‘Mediterranean Diet’.

Health problems such as respiratory difficulties, chronic musculoskeletal problems and depression are clearly linked to obesity, converting obesity into a public policy problem (Ruhm, 2007). What is more, it is a risk factor for diseases such as heart malfunctioning, hypertension, cancer and diabetes (Lobstein et al., 2003). The British Department of Health suggests that, in addition to health issues, obese people tend to suffer from social and psychological problems as well (Department of Health, 2007). For example, overweight children are often targets for bullies, and these emotional traumas spill into other aspects of their lives and continue to harm the individual throughout adulthood. Thus, this global phenomenon has huge implications in terms of health care expenditure (García-Goñi & Hernández-Quevedo, 2012). Regarding health costs, the World Obesity Federation Report (World Obesity Federation, 2022) estimates that the economic impact of overweight and obesity in Spain represented by 2019, 2.09% of GDP1.

One explanation for rising obesity levels is that welfare-improving technological change has both lowered the cost of food and increased the cost of participating in physical activity (Lakdawalla & Philipson, 2009) and, at the same time, sedentary habits (such as watching TV or playing video games) have become more widespread. On the other hand, geographic variation in obesity levels may reveal the impact of the built environment on obesity (Reidpath et al., 2002). If changes to the built environment do in fact affect the behaviour of individuals as regards health and eating patterns, then this could be an important tool for policymakers to curb the rise in obesity (Gutiérrez-González et al., 2023). This is particularly relevant to the extent that health decisions are based on the subjective perception of health. In relation to being overweight, adults are supposed to make health care decisions when they notice they have a problem with excess weight above normal. In this regard, it is important to study how health problems perceptions are set up. Consequently, we investigate how self-reported weight perception (SWP) and derived health decisions are relative to the context/environment in which a person interacts.

Using data from all the waves of the Spanish National Health Survey (ENS, Encuesta Nacional de Salud, Spanish abbreviation) with information on SWP, weight-loss medication intake and dieting, –1987 to 2011– we show that actual body mass index of the adult population has increased considerably in the corresponding 24-year period, while SWP and weight control behaviour have not changed accordingly. We also show people in the most overweight regions do not seem to perceive their weight or behave differently from people living in regions at the bottom of the BMI ranking. One plausible explanation for this pattern could be found in the environment or context of individuals. For example, experiments suggest that judgments of the severity of health conditions are context-dependent (Melrose et al., 2013). Body width perception, probably the most significant variable affecting weight perception, is affected by context effects: those individuals who are exposed to mostly wide body forms are more likely to judge a body width as thinner than those individuals who are shown narrow forms (Wedell et al., 2005). Overweight young males tend to be perceived as more normal, healthier, and less needed to lose weight when subjects are exposed to photographs of males with obesity (Robinson & Kirkham, 2014). Even body weight preferences and attraction seem to be affected by the context people experience (Boothroyd et al., 2012; Robinson & Christiansen, 2015).

In a similar way, people may evaluate their own weight in relation to a population norm that may change over time, across regions and for males vs. females. If individual BMI changes alongside population norms, we could expect no statistical relationship between BMI and weight perception. Health decisions based on one's own weight perception may follow a similar pattern. To investigate this question, we analyse how SWP, and weight control behaviours vary with individual BMI and average BMI of the population at each region-year-sex group as two independent factors. The results indicate that weight perception, intake of weight-loss medication, and dieting are highly correlated with individual body mass index. However, we find evidence consistent with weight perception being a relative concept: population norm – average BMI in the region-year-sex reference group – harms SWP and weight control decisions. These results apply to both males and females separately; although statistical significance changes, to some extent, for the two subgroups.

We could only find one previous study (Blanchflower et al., 2009) that analysed the effect of relative weight on both feeling overweight (reporting current weight to be “too high”) and weight-control behaviour (having dieted in the last 12 months). Using cross-sectional data from the Eurobarometer (including 26 European countries) they estimated the effect of individual BMI and relative BMI (compared to average in the same region, age and gender group). Their findings showed that relative BMI affects positively (negatively) the feeling of being overweight for females (males) after controlling for individual BMI. They also estimated a positive effect on the probability of dieting, although statistical significance is not achieved for males.

Oswald & Powdthavee (2007), exploiting one cross-section from the British Household Panel Surveys, also pointed to the idea of weight status perception influenced by relative weight (e.g. the difference between individual BMI and average BMI of a reference group) in addition to individual weight. Even though they did not define a population norm, they used one of their findings, that the sign of the impact of education level on weight perception changes from negative to positive after controlling for individual BMI, as evidence for weight perception to be affected by the BMI norm of the reference group of each education subgroups, i.e. highly educated individuals tend to have a higher perception of overweight than low educated ones with the same BMI because they differ in their reference group weight norm.

We believe our study contributes to this literature in at least three aspects: 1) we focus on a large dataset from the ENS in Spain; 2) we analyse a period of 24 years over which the average BMI increased from normal weight to overweight, and; 3) we estimate the effect of relative weight on three dependent variables: SWP, being currently on a diet, and weight-loss medication intake. Noticeably, the analysis of the intake of weight-loss medication is novel; neither of the two abovementioned studies considered this weight control strategy. Also, our study, compared to Blanchflower et al.’s research, implies some worth mentioning methodological differences. First, weight perception and dieting are defined differently. Our SWP variable comes from a question in the ENS where respondents are asked to respond if their weight, relative to their height, is below, equal or above normal. This question could be interpreted as BMI status perception since it considers that weight has to be evaluated relative to height and compared to a normal value. On the other hand, the perception question in Blanchflower et al. (2009) is ‘Would you say that your current weight is: Too low; About right; Too high?’. It does not mention weight as a concept in relation to height, and it uses the word ‘right’ instead of ‘normal’. Nonetheless, one can anticipate that respondents with different social norms will have a different concept of ‘normal’ or ‘right’. Also, in our analysis, we analyse whether participants are currently on a diet, while Blanchflower et al. (2009) consider dieting in the last 12 months. Finally, although a causal inference method is not applied in our analysis, we perform robustness checks that help understand the nature of the relationships estimated by: a) controlling for survey year unobserved heterogeneity and region fixed effects; b) considering alternative specifications of the effect of individual BMI; c) using different proxies for BMI norm; d) correcting BMI for weight and height reporting bias as found in a previous Spanish study.

This paper is structured as follows: Section 2 introduces a rationale for the effect of population norms on weight control behaviour and how people compare their weight to this norm. In Section 3 we describe the data employed, in section 4 the empirical methodology is commented, section 5 shows the main results of the analysis and section 6 concludes with some discussion.

2. Weight control behaviour and population weight norms

We assume individuals have some control over their body weight b. For example, they could change their food intake and their energy consumption. Increasing weight status could have some benefits and costs. For example, sugary and highly calorific food could be tasty. Also, stop visiting the gym could mean more free time for enjoyable activities. On the other hand, being overweight and obese comes at a cost in terms of health status and financially. We anticipate that people consider all costs and benefits when choosing their weight control strategies. In this analysis, we proposed that individuals care about how their own body weight compares with the weight norm of their reference group m. For example, a person may want to lose weight if people in their comparison group lost weight recently.

A utility function proposed in the literature (Blanchflower et al., 2009; Oswald & Powdthavee, 2007) can account for direct utility (u(b)) and costs (cb) of body weight and the social comparison element (μ(m — b)):

Assuming there are no further constraints, a subject would want to set their weight so that the first derivative with respect to b equals zero:

in such a way that people would want to gain (lose) weight whenever W_b > 0 (W_b < 0) is satisfied.

In order to anticipate how people are affected by the weight norm, we must look at the cross-partial derivative:

that implies that if μ''(m — b) < 0, i.e. the comparison element of W is strictly concave with respect to the argument (m — b), it will lead to W_bm > 0; hence the marginal utility from body weight (W_b) will increase (decrease) with the population weight norm. Therefore, subjects that compare themselves with heavier (slimmer) reference groups will be less (more) prone to engage in weight control strategies (e.g. dieting, exercising, or taking medications).

In an analysis of the patterns in obesity in the US, a previous work (Burke & Heiland, 2007) proposes that people prefer to be as close as possible to a weight norm, such that the comparison utility is given by:

where coefficient J represents the strength of the social interactions, i.e. the extent to which the squared distance between the actual weight and the norm matters. In this case, μ(.) is concave, and the maximum of it is achieved when b = m. There seems to be empirical and theoretical justification for adopting this functional form. Experimental evidence shows that, in the context in which subjects are exposed, appealing or perceived-as-healthy body weights lie in the middle (not at the extremes) of the body weight distribution (Robinson & Christiansen, 2015; Robinson & Kirkham, 2014). Also, it seems intuitive that people want to be similar to the population norm, i.e. they would not want to be too heavy or too slim compared to their reference group. Nonetheless, it could also be argued that, for certain social groups, there might be a status element of being the thinnest body; so that μ(m — b) is an increasing function and individuals would want to maximise m — b. Notice that this status-based utility component has been previously acknowledged (Blanchflower et al., 2009) and is consistent with the theoretical result of a positive effect of the norm on the individual's weight.

3. Data and descriptive analysis

3.4. Descriptive analysis

Table 1 shows that the Spanish adult population has become older and more educated. Changes in work status indicate how the labour market was affected by the crises of the early 1990s and of the Great Recession, with the working adult population decreasing in 1995 and 2011 with respect to the corresponding previous periods available. The regional distribution of the sample has been very stable over time. Regarding BMI, the average has increased from normal weight in 1987 (24.23) to overweight in 2011 (25.47). Interestingly, the dependent variables have not changed in the same way: for example, the Spanish population, represented in the sample, does not seem to perceive themselves as more overweight (above normal) over the same time interval. Nor has weight control behaviour changed as it would be expected: weight loss medication intake does not show any time trend, and a higher proportion of subjects reported to be on a diet in 1987 than in 2006 and 2011.

For a more detailed analysis, the mean and 95% confidence intervals of BMI, proportion of overweight/obesity, and the three dependent variables are depicted by survey year in Figure 1. It is noticeable that BMI (panel a) and the proportion of overweight/obese subjects (panel b) have a statistically significant positive trend over the period considered. However, SWP (panel c) and weight control behaviour variables (panels d and e) do not follow the same pattern. It rather seems that SWP, weight loss medication intake and dieting are insensitive to SWP; i.e. people living in different periods do not seem to have perceptions or adopt weight control strategies in line with their actual weight status. A similar conclusion can be obtained if we consider the mean and 95% confidence interval of the analysed variables across regions in Figure 2, where regions are ordered according to proportion of overweight/obesity in the horizontal axis from left to right. There are statistically significant differences in BMI and overweight/obesity between the least overweight regions (like Madrid, the Basque Country, La Rioja and the Balearic Islands) and those with the highest proportion of overweight or obese subjects (Andalusia, Galicia, Castile-La Mancha and Extremadura). However, SWP, intake of medication and dieting in each region do not seem to have any relationship with the corresponding aggregate measures of weight status. In Figure 3, the sex differences in BMI/overweight (panels a and b) and the dependent variables (panels c, d and e) show an interesting pattern. While males are more overweight, females perceived themselves as having a weight above normal more frequently, and they are more likely to take weight loss medication or be on a diet.

The econometric analysis proposed in this study is aimed at testing some hypotheses to explain the puzzling results presented in Figures 1-3. In particular, we have analysed whether individual perceptions and behaviour are insensitive to individual weight status, giving place to no (or counterintuitive) statistical relationship between aggregate SWP, medication intake and dieting, on one side, and aggregate BMI, on the other side, as shown in the descriptive analysis. Alternatively, we can test whether subjects adapt their perceptions and behaviour to the average BMI of their reference groups so that, even if individuals are sensitive to their individual BMI, no variation is expected in SWP, intake of medication and dieting across units of populations (regions or years) with different weight social norms.

Table 1Sample size, relative frequencies, means and standard deviations for the analysis sample. 

	Year of survey
	1987	1993	1995	1997	2001	2003	2006	2011	Total
N	21,422	16,550	5,253	5,363	18,006	20,406	24,943	17,883	129,826
Region
Andalusia	0.157	0.158	0.169	0.164	0.169	0.169	0.172	0.173	0.167
Aragon	0.034	0.032	0.037	0.033	0.031	0.031	0.031	0.026	0.031
Asturias	0.033	0.026	0.025	0.028	0.028	0.026	0.026	0.025	0.027
Balearic Islands	0.019	0.020	0.019	0.017	0.018	0.022	0.022	0.022	0.020
Canary Islands	0.033	0.036	0.038	0.037	0.038	0.044	0.047	0.046	0.041
Cantabria	0.015	0.014	0.012	0.012	0.014	0.014	0.013	0.011	0.013
Castile La Mancha	0.042	0.041	0.038	0.045	0.043	0.043	0.042	0.042	0.042
Castile Leon	0.077	0.071	0.071	0.069	0.068	0.060	0.059	0.058	0.065
Catalonia	0.161	0.162	0.159	0.165	0.159	0.161	0.149	0.159	0.159
Valencian Community	0.096	0.102	0.102	0.098	0.102	0.103	0.114	0.112	0.104
Extremadura	0.028	0.029	0.032	0.028	0.027	0.025	0.021	0.023	0.026
Galicia	0.067	0.075	0.069	0.068	0.068	0.068	0.061	0.056	0.066
Madrid	0.137	0.134	0.131	0.133	0.134	0.131	0.141	0.143	0.136
Murcia	0.017	0.022	0.018	0.021	0.026	0.029	0.029	0.034	0.025
Navarre	0.014	0.014	0.014	0.014	0.012	0.014	0.014	0.015	0.014
Basque Country	0.062	0.059	0.060	0.060	0.056	0.053	0.052	0.049	0.056
La Rioja	0.007	0.007	0.008	0.007	0.006	0.007	0.008	0.007	0.007
BMI	24.23 (3.7)	24.75 (3.7)	24.87 (3.8)	25.03 (3.9)	25.14 (4.0)	25.14 (4.0)	25.40 (4.0)	25.47 (4.1)	25.06 (3.9)
Age (10 years)	3.97 (1.7)	4.22 (1.9)	4.16 (1.8)	4.25 (1.8)	4.37 (1.8)	4.54 (1.9)	4.50 (1.8)	4.64 (1.8)	4.37 (1.8)
Sex
Male	0.529	0.511	0.515	0.514	0.498	0.480	0.497	0.489	0.502
Female	0.471	0.489	0.485	0.486	0.502	0.520	0.503	0.511	0.498
Marital status
Single	0.335	0.318	0.331	0.332	0.322	0.331	0.326	0.321	0.327
Married	0.609	0.608	0.593	0.589	0.580	0.571	0.575	0.566	0.584
Separated/Divorced	0.011	0.019	0.020	0.023	0.035	0.026	0.041	0.050	0.030
Widowed	0.045	0.055	0.056	0.056	0.064	0.072	0.058	0.063	0.060
Work status
Working	0.455	0.436	0.400	0.410	0.465	0.473	0.524	0.449	0.455
Unemployed	0.078	0.099	0.115	0.117	0.077	0.059	0.066	0.144	0.094
Retired	0.103	0.150	0.166	0.176	0.187	0.187	0.181	0.192	0.172
Student/Housework/other	0.363	0.315	0.320	0.298	0.271	0.282	0.229	0.215	0.280
Education level
Primary	0.582	0.621	0.580	0.580	0.592	0.434	0.393	0.199	0.478
Secondary	0.346	0.201	0.229	0.268	0.266	0.355	0.357	0.566	0.335
Tertiary	0.071	0.178	0.191	0.152	0.142	0.210	0.250	0.235	0.186
Self-reported weight perception
Below normal	0.094	0.082	0.073	0.071	0.064	0.059	0.063	0.059	0.069
Normal	0.501	0.499	0.483	0.489	0.509	0.574	0.507	0.589	0.524
Slightly above normal	0.332	0.339	0.346	0.353	0.349	0.327	0.360	0.298	0.337
Much above normal	0.073	0.080	0.099	0.087	0.078	0.040	0.070	0.054	0.070
Weight loss medication
No	0.997	0.997	0.999	0.999	0.998	0.998	0.997	0.998	0.998
Yes	0.003	0.003	0.001	0.001	0.002	0.002	0.003	0.002	0.002
Dieting
No	0.813	.	.	.	.	.	0.879	0.909	0.876
Yes	0.187	.	.	.	.	.	0.121	0.091	0.124

Note: “N” is the number of observations included in the econometric analysis. Individuals with missing value in (at least) one of the variables were excluded. In addition, 37 observations were removed with extremely high BMI (>100). Also, 42 adults with BMI<12 were removed from the analysis. Relative frequencies reported for categorical variables. Mean and standard deviation (in parenthesis) reported for continuous variables.

Figure 1Average and 95% confidence intervals of body mass index and dependent variables by survey year. 

Figure 2Average and 95% confidence intervals of body mass index and dependent variables by region. 

Figure 3Average and 95% confidence intervals of body mass index and dependent variables by sex category. 

4. Empirical strategy

Each subject j reported their weight perception SWP_j, an ordinal variable that could take four values: 1 if subjects report their weight to be Lower than normal, 2 if Normal, 3 if Slightly higher than normal, and 4 if Much higher than normal. The latent variable is given by:

Where x_j is the vector of the values of the explanatory variables for subject j, β is the vector of the coefficients for each variable, and u_j is a random error that follows a standard logistic distribution.

The ordered logit assumes that the cumulative probability of reporting a specific value i for SWP_j is given by:

Where c_i (for i = 1,2,3) are constant parameters that have to be estimated along with the vector β by maximum likelihood, and c₄ = ∞. In the result section, we show the average marginal effect of covariates on the probability of each value (i = 1,2,3,4). In addition, we report standard errors clustered at the level of region-year-sex group, i.e. at the level at which values of BMI norms change in our study.

Different specifications were considered by including a particular set of covariates. First, we estimated a model where only individual BMI was included as an explanatory variable. Secondly, we added BMI norm to estimate the role of norms on SWP after controlling for the BMI of the subject. Thirdly, we controlled for age and sex. Finally, we additionally controlled for marital status, work status and education level. The same model specifications were estimated for males and females separately to explore if there are gender differences in the impact of BMI population norms.

Robustness checks were included in the supplementary material by exploring model variants. First, survey year fixed effects were included as controls to estimate the effect of BMI norm after accounting for unobserved heterogeneity across survey years. Alternatively, we also controlled for region fixed effects to show the results after conditioning for unobserved regional factors. Second, the functional forms for the effect of individual BMI were changed from linear, in the base case analysis, to logarithmic and parabolic, respectively. Third, instead of using average BMI in each region-year-sex group, as a proxy for the BMI norm, we explored the use of the median value and the percentile rank of the subjects within their reference group. Finally, we re-estimate the main econometric models after correcting participant-reported BMI for reporting bias using estimates from a previous study in Spain (Gil & Mora, 2011).

The same model specifications were used to estimate the intake of weight loss medication and dieting, separately. Given the binary nature of those two dependent variables, a logit model was estimated where only two ordered values (i = 1,2) applied: No/Yes intake and No/Yes dieting, respectively.

5. Results

6. Discussion and conclusions

The associations estimated between the explanatory and dependent variables in this study are consistent with BMI being significant for explaining SWP in a relative way. More precisely, weight perception seems to be adapted to the BMI norm in each region-year-sex reference group. The magnitude of the estimated relationship is important; the results imply that one additional point in the BMI norm would reduce the perception of having a weight above normal (slightly higher or much higher than normal) by 10.64 percentage points on average. These results can be interpreted in line with recent findings on weight perception (Boothroyd et al., 2012; Robinson & Christiansen, 2015; Robinson & Kirkham, 2014). We rely on the relative nature of weight perception to explain the null correlation between aggregate BMI and weight perception over time or across regions in Spain. We also believe that our results could partially explain why females see themselves as more overweight than males when the opposite is actually true. More importantly, we have shown evidence that weight control behaviour is sensitive to individual BMI relative to the population norm, in such a way that two identical individuals (with the same weight status) could be carrying out very different weight control strategies if they belong to contexts where BMI norms are not the same. In the base case model, we have estimated that an additional point in the BMI norm is associated with a decrease in 0.159 and 6.44 percentage points in the proportion of individuals who would take medications or be on a diet, respectively.

The relative judgment of relevant aspects of life has been applied in the economic literature. For example, similar explanations are considered to account for the Easterlin Paradox, the empirical fact that life satisfaction seems to not change in the long term despite large increases of income (Easterlin, 2001; Easterlin et al., 2010), i.e. income and weight status seem to be relative concepts that are judged in comparison to norms; if the weight status norm (income norm, in Easterlin paradox) is increased, I am less likely to perceive myself as overweight/obese (wealthy). In another area of research, closer to our study, the context given by weight or physical fitness of peers and neighbours has also been analysed by economists finding significant positive effects on weight or fitness of university students, US Air Force Academy cadets or beneficiaries of housing vouchers; this evidence is consistent with an adaptation to the norm given by the context (Carrell et al., 2011; Kling et al., 2007; Ludwig et al., 2011; Yakusheva et al., 2014).

The pattern found in our analysis implies a vicious cycle: the more overweight the population gets, the less aware its members are of their high BMI and the less likely they are to engage in weight control behaviour. Our results imply that as the BMI norm increases, the less likely the population is to be on a diet which prevents self-control or weight corrections. Therefore, we believe that the behavioural pattern shown in our results could be one plausible explanation of why developed societies are getting more overweight/obese; see an analysis for the US (Burke & Heiland, 2007). In turn, the relative nature of weight perception and self-control behaviour could be an additional factor to those pointed out in the literature (Cawley, 2015).

The current study could have implications for health policy. Public interventions based on a good understanding of the weight perception process could imply better health policies. Thus, manipulations of the environment or contextual BMI could be used to encourage behavioural changes or to sustain healthy lifestyles for adults who, subsequently, pass these behaviours on to young children (Popkin et al., 2005). If people adapt their weight perception and weight control behaviour to social norms, one could expect that manipulation of the social norms, i.e. promoting the idea of a healthy weight, will have an impact on people’s weight status.

Alternatively, our results could support public interventions on weight control based on economic principles. From an efficiency point of view, interventions affecting overweight or obesity levels of the population could be acceptable in cases in which individual decisions impose externalities on society, when markets are not perfectly competitive, if information about the consequences of obesity is not accurate or available, and if individuals are not perfectly rational (Cawley, 2004, 2015). Brunello et al. (2009) argue that the fact that many people fail to describe themselves as obese or overweight is an important problem supporting public intervention. In this paper, our findings suggest that people integrate social norms when they evaluate their own weight status and make weight control decisions; this may raise concerns about the adequacy of decisions taken in contexts where social norms are dramatically changed (e.g. when the reference group is characterised by either large levels of obesity or extreme slimness) or even distorted. The large increase in BMI, experienced in the world over the last few decades, could have changed social norms to make individuals prone to a relaxation in weight control behaviours. Public health and medical interventions to make people more aware of their weight status could lead to a more efficient situation.

From a microeconomic perspective, an individual could perfectly adapt their optimal weight to the weight of their contemporaries; for example, individual utility could be affected by the relative position of an individual compared to a population norm (Barbieri, 2022; Blanchflower et al., 2009; Oswald & Powdthavee, 2007; Rosin, 2012). Indeed, there is evidence of life satisfaction decreasing for individuals that has a high BMI rank in a reference group (Kim, 2022). The estimated negative effect of BMI norm on weight loss medication and dieting could be interpreted as the behaviour of rational decision makers adapting their weight control strategies to the population norm. However, this could be considered rational behaviour only if people are fully aware of their objective weight status and the derived consequences, including health implications.

The results presented here imply that there is an externality created by changes in the weight status of members of a population. It has been pointed out that an individual with increased BMI may incur higher healthcare costs due to obesity-related diseases or disabilities (Bhattacharya & Sood, 2006, 2011; Biener et al., 2017; Cawley, 2015; Cawley & Meyerhoefer, 2012; Hecker et al., 2022; Lehnert et al., 2015). However, the results here exposed imply a new externality because an increased BMI of some individuals in a society will impact the social norms of other contemporaries, having an impact on their weight control strategies and, therefore, on their BMI level. In turn, if I change my weight status, I will impact my health status; at the same time, I could be affecting the weight status (via changes in social norms) of other people and hence their health status.

We acknowledge that our model could be affected by some econometric problems. For example, measurement errors, due to variables reported by the ENS participants, and endogeneity (e.g. unobserved factor that may be related to social norms) given that we do not use natural experiment data. In this sense, we tested the effect of social norms under alternative model assumptions. The robustness checks show that the sign of the estimated effect of the BMI population norm is unchanged after controlling for survey year unobserved heterogeneity. In addition, conclusions are unchanged when region fixed effects are included in the econometric models. The main results are also consistent to changes in the functional form of individual BMI (logarithmic or parabolic) and the variable used as a proxy for BMI norm (median value or percentile rank within the reference group). Also, the same conclusions are achieved regarding the effect of BMI norm after correcting for reporting biases in weight and height using previous estimations in a Spanish study. Notably, correction of systematic bias associated to self-reported weight, height and social norms is especially important since this type of non-classical measurement errors could imply biased estimators with the wrong signs; while classical measurement errors (e.g. errors in self-reported weight and height that are uncorrelated with true value of the variables) only attenuates the effect of the variable with reporting errors (Bound et al., 2001).

Even after careful robustness checks, the empirical approach presented here prevents us from interpreting the estimated relationships between social norms and weight control behaviour as causal links; they rather show statistical associations that could potentially be driven by the associations of omitted variables and the explained variables. One omitted variable is the immigration status of the ENS participants: only available for the year 2011. This variable could be an important factor given the increase of non-native population during the study period (mainly during the 90s and 2000s). In this regard, we have run exploratory estimations, including immigrants (those born abroad) or excluding them from the 2011 sample, obtaining very similar results for both groups and not affecting the conclusions about the effect of social norms.