Evaluating and establishing national norms for mental wellbeing using the short Warwick–Edinburgh Mental Well-being Scale (SWEMWBS): findings from the Health Survey for England

There has been growing interest in measuring mental wellbeing, recognising that mental health is more than the absence of mental illness, and the desire for policy makers to assess progress outside of the usual economic and material indicators [1‐3]. Mental wellbeing has been defined by different authorities as various combinations of optimum functioning and feeling [4]. Mental wellbeing has been found to have a U-shaped relationship with age [5, 6]. It is linked with good physical health and with longevity among older adults [7]. Its relationship with social and economic circumstances is complex [6, 8].

The Warwick–Edinburgh Mental Wellbeing Scale (WEMWBS) was developed in 2007 [9] to support the development of an evidence base relating to public mental health. Public mental health encompasses the promotion of mental wellbeing, the prevention of mental illness and recovery from mental illness. The 14 items of the WEMWBS scale are all positively worded and relate to the main components of mental wellbeing, defined as ‘feeling good and functioning well’ [4]. Its strengths include the ability to capture both the eudaimonic (people’s functioning, social relationships, sense of purpose) and the hedonic perspectives on wellbeing (e.g. feelings of happiness). In most validation studies, scores resolve to a single component.

In 2009, a short version (seven items) of the scale (SWEMWBS) was resolved using the Rasch measurement model [10]. Five items were removed from the 14-item WEMWBS to improve the overall fit of the data to the Rasch model; and two items were removed due to local item dependency (i.e. residual associations in the data after the Rasch-based trait score had been removed). The remaining seven-item scale fitted the expectations of the Rasch model, and a linear transformation of the score was then obtained, to facilitate the use of valid parametric statistical analyses. The items in SWEMWBS present a picture of mental wellbeing in which psychological functioning dominates subjective feeling states, but the superior scaling properties and reduced participant burden have made it the instrument of choice in some studies. Both scales have proved very popular with practitioners and researchers in the UK and further afield. There were 1841 registrations for use between October 2012 to March 2016, and the numbers are increasing annually [11].

Although both scales have been used to evaluate interventions and to examine the epidemiology of mental wellbeing, more research has been published on the full 14-item scale, including population norms for European countries [12‐15]. A recent study in England [8] showed surprising findings relating to the social distribution of mental wellbeing. The expected social inequalities distribution was demonstrated for those at the lower end of the mental wellbeing scale—a group at high risk of mental health problems—but not for those at the high end of the mental wellbeing scale. Differences between predictors of the low end of the wellbeing scale with the high end of the wellbeing scale were also found with health behaviours. Obesity and being a non-drinker of alcohol were associated with the low end of the mental wellbeing scale but not with the high end, while smoking and low fruit and vegetable intake were associated with both increased odds of the low and decreased odds of the high end of the scale [16]. Whether the short seven-item scale exhibits similar properties to the full 14-item scale in a nationally representative sample in England has yet to be explored. SWEMWBS may have lower face validity than the full scale, focusing on items relating to functioning and excluding items relating to feeling aspects [17]. Since the short scale is being used widely in England, it is important to establish national norms for the short scale and evaluate how it performs against the full scale, so that practitioners and researchers using SWEMWBS, for example those conducting small-scale studies on local areas, have a meaningful benchmark with which to compare their results. This study therefore aimed to compare the performance of SWEMWBS and WEMWBS in the English population.

Around 80% of the original sample (N = 34,155) answered all seven SWEMWBS items (N = 27,169), which was around 2% higher than the number of participants who answered the 14 item WEMWBS (N = 26,617). The response rate by year corresponded to 85% in 2010, 84% in 2011 and 61% in 2012 (information collected during the nurse visit), and 88% in 2013 within co-operating households.

Tables 1 (men) and 2 (women) present national norms for SWEMWBS across social and demographic variables, with p values for the joint hypothesis test (i.e. whether the coefficients for the difference in mean scores across categories were simultaneously equal to zero) and p-values for the null hypothesis of zero change in SWEMWBS per unit change in the continuous predictor. The same analyses were carried out for WEMWBS (presented in supplementary Table 1). In addition, norms for socio-economic, demographic and health sub-categories by age group are presented for SWEMWBS in supplementary Tables 2 (men) and 3 (women).

Mean SWEMWBS scores for men and women were 23.7 and 23.6, respectively (ES = 0.03, 95% CI: 0.01–0.06), and were not statistically different (p = 0.100). The largest differences across mean scores of SWEMWBS were observed across the categories of self-rated health, ranging from 19.3 for men reporting ‘very bad’ health to 24.7 for men reporting ‘very good’ health (ES = −1.52), and 19.6–24.9 for women (ES = −1.42). Effect sizes for limiting longstanding illness (versus none) were moderate in magnitude (ES = −0.54 and −0.52 for men and women, respectively). Mean scores for SWEMWBS varied significantly across the categories of income, education and Index of Multiple Deprivation (p < 0.05). Effect sizes for the lowest income quintile (versus highest) ranged from small to moderate. With regard to age, the largest effect size was observed for the 65–74 group versus the 16–24 group (ES = 0.25 and 0.29 for men and women, respectively). Differences in mean SWEMWBS scores across the nine Government Office Regions were statistically different to zero (p < 0.001), but the effect sizes were small in magnitude (ES < |0.20|). Differences in SWEMWBS scores across ethnic groups were statistically significant for men but not for women; the effect size for Black men (vs. White men) was moderate in magnitude (ES = 0.37).

Variation in scores on WEMWBS by age and across subgroups followed a similar pattern to SWEMWBS (supplementary Table 1), including the magnitude of effect sizes. However, in contrast to SWEMWBS, gender differences in wellbeing scores using the 14-item scale were statistically significant (p = 0.009), but the estimated change in wellbeing score for a one-unit change in age group (fitted as a continuous term) was not significantly different from zero (p = 0.749).

Table 3 presents Spearman correlations between mental and physical health variables and both SWEMWBS and WEMWBS. Statistically significant but moderate correlations between SWEMWBS and the happiness index (ρ = 0.53, p < 0.001), GHQ12 (ρ = −0.52, p < 0.001) and EQ-VAS (ρ = 0.40, p < 0.001) were found. There were weaker correlations with self-rated health (ρ = −0.33, p < 0.001) and limiting longstanding illness (ρ = −0.21, p < 0.001).

Correlation coefficients were very similar between SWEMWBS and WEMWBS; where they differed, WEMWBS had slightly higher correlations (up to 0.03 difference). Correlations with the mental and physical health variables were of a similar magnitude for SWEMWBS and WEMWBS even when comparing across the two different, randomly generated samples (Supplementary table S4). With regard to the internal reliability of the scales, Cronbach’s alpha for SWEMWBS and WEMWBS was 0.84 and 0.92, respectively, both exceeding the acceptable conventional level of internal agreement (0.70).

Table 4 presents results from multinomial logistic regressions for SWEMWBS categorised into low (15%), medium (71%) and high (14%) wellbeing (proportions were the same for WEMWBS to zero decimal points). Focusing on SWEMWBS scores only, and the low versus the medium wellbeing categories, participants aged 25–54 were more likely to have low than medium wellbeing compared with 16- to 24-year-olds. Participants with worse self-rated health were more likely to have low than medium wellbeing (e.g. bad/very bad health: odds ratio (OR) = 9.51 (95 % confidence interval 8.05–11.23)). Similar gradients were demonstrated for education (e.g. no qualifications OR = 1.42 (1.23–1.65)) and income (e.g. lowest quintile 1.48 (1.26–1.73)). Eating less than one portion of fruit and vegetables a day compared with five or more was associated with an increased odds of low versus medium wellbeing (1.42 (1.22–1.67)). Current smokers were more likely than non-smokers to have low versus medium wellbeing (1.28 (1.15–1.41)). Moderate drinkers were less likely than non-drinkers to have low wellbeing (0.87 (0.79–0.97)). Associations with low versus medium wellbeing as measured by WEMWBS showed a similar overall pattern to SWEMWBS, although there were a few differences in which comparisons attained statistical significance, e.g. participants in the Black ethnic group having the lowest odds of low wellbeing (vs. White participants) on SWEMWBS (0.68 (0.51–0.92)), and participants in the mixed ethnic group having the lowest odds of low wellbeing on WEMWBS (0.56 (0.36–0.87)).

For the high versus the medium wellbeing categories for SWEMWBS, older age groups (aged 55+) were more likely to have high wellbeing than 16–24-year-olds. This is in contrast to the finding of the younger age groups having higher odds of low wellbeing and demonstrates the well-known U-shaped association between wellbeing and age. Those with worse self-rated health were also the least likely to have high wellbeing (e.g. bad/very bad health OR = 0.21 (0.16–0.28)). However, the categories of income and educational status showed no association with the odds of high wellbeing, unlike the findings for the odds of low versus medium wellbeing. Participants in the Asian (OR = 1.56 (1.28–1.91) and Black ethnic groups (OR = 2.25 (1.77–2.87) were more likely to have high wellbeing than participants in the White ethnic group. There were gradients in the associations with lower fruit and vegetable consumption, and with higher alcohol consumption, with lower odds of high versus medium wellbeing found for participants in these groups (e.g. <1 portion a day (0.76 (0.63–0.93); >8 units alcohol 0.81 (0.72–0.93)). Obese (1.22 (1.09–1.37)) and morbidly obese (1.66 (1.29–2.13)) participants were more likely to have high wellbeing than those with normal weight; although overweight and obese participants had higher odds of having low versus medium wellbeing when adjustment excluded self-reported health (see Discussion). Again, analyses using WEMWBS showed a similar overall pattern, but some categories differed in whether they attained statistical significance: for example, associations for marital status and morbid obesity were not statistically significant for WEMWBS. Models using the subset of participants with complete data (supplementary table S5) showed no substantial differences between SWEMWBS and WEMWBS, nor with the main models.

The Bland–Altman plot for the comparison of each instrument is depicted in Fig. 1. The average discrepancy between the SWEMWBS and WEMWBS scores was 2.1 (95% CI: −0.80–5.01). The difference in scores demonstrated proportional error, with a slight tendency for this to increase with larger mean scores. The line of equality fell within the 95% CI of the mean difference meaning no absolute bias. Table 5 presents Spearman correlations between SWEMWBS and WEMWBS, and weighted kappa statistics between SWEMWBS and WEMWBS grouped into low, medium and high categories, within different subgroups. Correlations between SWEMWBS and WEMWBS were very high and statistically significant (0.95–0.96, p < 0.001) within subgroups of sex, education, income and the Index of Multiple Deprivation. For self-rated health, correlations were high and statistically significant, although slightly lower in magnitude (0.80–0.85, p < 0.001). Coefficients were also high, albeit lower in magnitude, for the comparisons of SWEMWBS with the seven redundant items in WEMWBS (0.84–0.87, p < 0.001). Weighted kappa coefficients showed substantial to almost perfect relative agreement between the two classifications across subgroups (0.79–0.85).

SWEMWBS performed comparably to WEMWBS in these analyses, demonstrating the expected population distributions and correlations with social variables for low wellbeing, and mimicking recent findings demonstrated with WEMWBS for high wellbeing. There was proportional disagreement presented in the Bland–Altman plot, reflecting the difference in scaling for SWEMWBS transformed to a metric scale, while no such transformation was required for WEMWBS. This small difference between the scales could also have affected differences found between SWEMWBS and WEMWBS in other analyses. However, despite this, SWEMWBS behaved very similarly to WEMWBS. The well-documented income and education gradients for low versus medium wellbeing were not found for high versus medium wellbeing. Similar moderate correlations were found between SWEMWBS and GHQ12 and EQ-VAS, as had been previously demonstrated for WEMWBS [9]. In men, SWEMWBS also followed the well-known U-shaped distribution by age for wellbeing, with its nadir between 35 and 55 years for low wellbeing [5, 9]. In women, we observed a slight difference in the norms for the two scales as its nadir was in the 16–24 age group, making the U-shaped distribution by age less clear.

The main difference between the performances of the two measures related to gender. Norms for WEMWBS were slightly higher for men, whereas for SWEMWBS norms did not vary significantly by gender. This is consistent with a study that found SWEMWBS to be gender neutral in a Swedish and Norweigan population [27]. The items common to both instruments include feeling useful, dealing with problems well, thinking clearly and autonomy. The majority of the seven WEMWBS items that are not present in the SWEMWBS relate more to the affective or feelings components of wellbeing (feeling good about self, confident, cheerful, loved, having energy to spare): each of which varied significantly by gender (p < 0.001, data not shown). It is therefore not surprising that WEMWBS detects more gender differences than SWEMWBS. The other two WEMWBS items not present in SWEMWBS relate to functioning (interest in new things; feeling interested in other people), which did not vary significantly by gender (p = 0.126 and p = 0.776, respectively, data not shown) [12]. However, it is important to note that average scores on WEMWBS may not vary much by gender, given by results in other contexts [12] and the small effect sizes found in this study. Surprising results relating to high versus medium wellbeing included the increased odds in Black and Asian ethnic groups, and in those who were obese, found with both instruments. Increased odds of high wellbeing among ethnic minority groups have been found before [6, 8], in particular among the Black minority ethnic group, which was suggested to be driven largely by high mean scores for wellbeing among Black African groups [6]. Black African groups were also found to have better self-reported health than White British groups after extensive adjustment for health behaviour and SEP confounders [28]; this may be attributable to a ‘healthy migrant effect’. Further studies wishing to look at ethnic differences in mental wellbeing should differentiate between Black Africans and Black Caribbeans and other heterogenous groups, where numbers allow. The minor differences between WEMWBS and SWEMWBS in the magnitude of the correlations observed for different ethnic groups are likely to be due to small sample sizes in some minority groups.

The higher odds of high versus medium mental wellbeing among overweight or obese participants were more marked with SWEMWBS than with WEMWBS and remain largely unexplained. It is important to recognise that these are only seen after adjustment for general health. Adjustment for general health in our models explains the different findings between health behaviours and wellbeing from those of Stranges et al. [16], including the non-significant associations between decreased odds of high wellbeing among non-smokers, and increased odds of low wellbeing among obese individuals.

The strong association between both low versus medium and high versus medium wellbeing and fruit and vegetable intake, even after adjustment for a number of socio-economic factors, suggests fruit and vegetable consumption as a possible causal factor in mental wellbeing. However, this present study was conducted using cross-sectional data and so we cannot rule out the possibility of reverse causality. Our findings must also be interpreted with caution due to the inevitable problem of residual confounding. Nevertheless, the associations between fruit and vegetable consumption and wellbeing deserve further investigation using longitudinal data. Since our primary aim was to evaluate how SWEMWBS performed against WEMWBS, further investigation was beyond the scope of the present study. It is also important that future studies examine the sensitivity to change of SWEMWBS compared with WEMWBS. Given the larger number of items in total, and the greater contribution of ‘feelings’ items, it remains possible that WEMWBS is more sensitive than SWEMWBS to change in intervention studies. This difference may prove important in small-scale evaluations of community-based mental wellbeing interventions [29].

The participants who answered SWEMWBS in our sample were given the full WEMWBS questionnaire. Participants may respond differently if asked only the SWEMWBS subset of questions, due to different question ordering, the shorter length and the absence of any influence that the omitted questions in SWEMWBS may have on the full WEMWBS responses. Around 80% of the sample answered the SWEMWBS questionnaire. Among non-responders there was a higher proportion of males, those living in the most deprived quintile, and low qualifications than responders (p < 0.001, data not shown). It is likely that these people may have lower mental wellbeing; therefore, the norms for SWEMWBS shown in the present study may be overestimated. However, we feel that the use of a nationally representative survey, and the use of non-response weighting, offset this limitation. The consistency of our findings with other studies suggests that our results do not have large biases although we accept this limitation as a caveat to our findings. Our analysis has largely focused on a comparison of SWEMWBS with WEMWBS to evaluate SWEMWBS as a tool to measure mental wellbeing; however, we acknowledge that SWEMWBS is subject to the same limitations as WEMWBS; for instance, we found minimal effect sizes across certain subgroups such as region with both instruments, despite significant p values, which is likely to be an artefact of the large sample size. Measuring mental wellbeing as a single construct may mask its multidimensionality [30].

SWEMWBS’s performance is very similar to that of WEMWBS. In this context, the 2% higher response rate observed for the SWEMWBS items within the Health Survey for England WEMWBS questionnaire, and its lower participant burden, will continue to make it a popular choice for both large-scale social surveys and intervention studies. However, those particularly interested in gender differences in mental wellbeing may prefer to use the full 14-item instrument. Further studies are needed to ensure that SWEMWBS performs as well as WEMWBS in terms of responsiveness to change in intervention studies.

The datasets supporting the conclusions of this article are available via the UK Data Service repository, subject to their end user license agreement:

NatCen Social Research, Royal Free and University College Medical School. Department of Epidemiology and Public Health. (2015). Health Survey for England, 2010. [data collection]. 3rd Edition. UK Data Service. SN: 6986, http://​dx.​doi.​org/​10.​5255/​UKDA-SN-6986-3.

NatCen Social Research, University College London. Department of Epidemiology and Public Health. (2013). Health Survey for England, 2011. [data collection]. UK Data Service. SN: 7260, http://​dx.​doi.​org/​10.​5255/​UKDA-SN-7260-1.

NatCen Social Research, University College London. Department of Epidemiology and Public Health. (2014). Health Survey for England, 2012. [data collection]. UK Data Service. SN: 7480, http://​dx.​doi.​org/​10.​5255/​UKDA-SN-7480-1.

NatCen Social Research, University College London. Department of Epidemiology and Public Health. (2015). Health Survey for England, 2013. [data collection]. UK Data Service. SN: 7649, http://​dx.​doi.​org/​10.​5255/​UKDA-SN-7649-1.