Aims of the study
We aimed to answer the following research questions:
-
1.
What are the national norms for SWEMWBS in the general population in England and across socio-demographic subgroups? How do subgroup differences in scores on SWEMWBS compare with those on WEMWBS?
-
2.
How well does SWEMWBS correlate with GHQ-12, EQ-VAS, happiness index, and self-reported health and limiting longstanding illness, as compared to correlation of WEMWBS with such instruments?
-
3.
Does SWEMWBS reproduce associations with social and health behaviour variables similar to the full version?
-
4.
How closely does the measurement of mental wellbeing with SWEMWBS approximate to the measurement by WEMWBS, and within different subgroups? In addition, how well does SWEMWBS capture those at the low and high ends of the mental wellbeing scale compared with WEMWBS?
Study participants
The study uses data from the Health Survey for England 2010–2013 (N = 27,169), the first survey years which included the 14-item WEMWBS [18–21]. The Health Survey for England interviews each year a new, random, nationally representative sample of the population living in private households in England [22]. Participants were selected using a multi-stage, stratified, probability design, with postcode sectors used as primary sampling units, randomly selected using the Postcode Address file. Data included spoken answers to questions, written answers in self-completion booklets, and biomedical information, which was collected via face-to-face interviews followed by a nurse visit. WEMWBS was self-completed confidentially as part of the interviewer visit, with the exception of 2012 when this was done during the nurse visit. NHS Research Ethics Committee approval was obtained prior to each survey commencing from the Oxford B (2010) and Oxford A (2011–13) Research Ethics Committees. Participants gave informed verbal consent prior to the interview.
Data
WEMWBS and SWEMWBS
Answers to each item on WEMWBS (and SWEMWBS) are provided using a five-point Likert scale (none of the time, rarely, some of the time, often, all of the time), and scored from 1 to 5 respectively, with all items being scored positively. Scores on all items are then summed to give a WEMWBS score (range 14–70) (see Box 1).
SWEMWBS uses seven items from the full 14-item WEMWBS (items in bold in Box 1). As with WEMWBS, scores on SWEMWBS are summed (range 7–35). As described earlier, SWEMWBS scores were transformed (set out in a conversion table published in a previous study [10, 23]) to facilitate the use of parametric statistical analyses. SWEMWBS was embedded within the full scale, so each HSE participant had scores on both scales (with the exception of 512 participants who completed the seven SWEMWBS items but did not complete the full 14-item scale).
To examine whether SWEMWBS was able to capture those with lower wellbeing scores as well as WEMWBS, three-category versions of SWEMWBS and WEMWBS scores were derived. Low and high categories were based on scores that were at least one standard deviation below and above the mean, respectively [16]. Categories for SWEMWBS were: ‘low’: 7–19.3; ‘medium’: 20.0–27.0; and ‘high’: 28.1–35. For WEMWBS, scores were ‘low’: 14–42; ‘medium’: 43–60; and ‘high’: 61–70.
Demographic, socio-economic, health and health behaviour data
Data on sex, age group, marital status, ethnicity, highest educational qualification, quintiles of equivalised household income, economic status, self-rated health and limiting longstanding illness were reported in the face-to-face interview. Region and area-deprivation (derived from the Index of Multiple Deprivation) were based on the participant’s address.
Instruments measuring mental and overall health in the HSE included the General Health Questionnaire (GHQ-12), an instrument comprising scores from 12 questions measuring psychological morbidity (2010 and 2012 only). For each of the 12 questions, participants were given a four-point response scale, ranging from ‘not at all present’ to ‘present much more than usual’. The first two responses were coded as zero, and the third and fourth responses were coded as one, providing a maximum score of 12. In addition the EQ-VAS score, a visual analogue scale where participants rate their health from ‘worst imaginable health state’ (0) to ‘best imaginable health state’ (100) (2010–2012 only), and a happiness index (2010 and 2011 only) were included in the analysis. Within the happiness index, participants were asked to rate how happy they were from 0 (unhappy) to 10 (happy). These measures were collected via the same self-completion booklet that contained WEMWBS.
Health behaviours included current smoking status; alcohol consumption; and fruit and vegetable portions per day (not asked in 2012), which were self-reported. Body mass index categories were derived from height and weight measurements carried out by trained interviewers. Categorisation of alcohol consumption on the heaviest drinking day in the last 7 days was based on daily limits of alcohol consumption as recommended at the time of the survey (≤4 units a day for men, ≤3 units a day for women). These were as follows: non-drinker, moderate drinker (within daily limits), excess drinker (exceeding daily limits but less than twice the recommended limits) and heavy episodic drinker (over twice the recommended limits). Categorisation of fruit and vegetable consumption was as follows: 5 or more portions of fruit and vegetables a day, 3 to <5, 1 to <3, and <1 portion a day. BMI groups were defined as underweight (<18.5 kg/m2), normal weight (18.5 to <25 kg/m2), overweight (25 to <30 kg/m), obese (30 to <40 kg/m2) and morbidly obese (≥40 kg/m2). Physical activity was covered only in 2012, so numbers did not allow its inclusion in this study.
Statistical analysis
Establishing Norms (research question 1)
Sex-stratified national norms for SWEMWBS were calculated, including the mean, 10, 15, 50, 85 and 90th centile across the key demographic variables. The same norms stratified by age group are shown in supplementary tables. Norms for the present study can be read along age, sex and one other dimension only.
First, we used univariable linear regression to estimate the difference in mean SWEMWBS scores fitting variables such as age group and income as categorical variables. Statistical significance was examined using a joint hypothesis test (i.e. whether the coefficients for the difference in mean scores were simultaneously equal to zero). Second, categorical variables such as income were fitted as continuous terms to estimate the change in SWEMWBS per unit change in the predictor. Third, the magnitude of the association was estimated with the effect size (ES), computed as the difference between the mean wellbeing scores of two subgroups, divided by the pooled standard deviation. Uncertainty in ES was estimated using bootstrap confidence intervals based on the noncentral t distribution. The cut-offs and the interpretation of ES were: low (|0.20| ≥ ES ≤ |0.50|), moderate (|0.50| > ES ≤ |0.80|) and high (ES > |0.80|). The same analyses were repeated for WEMWBS. We hypothesised that SWEMWBS would show similar variation across subgroups as WEMWBS.
Criterion validity (research question 2)
Spearman correlation coefficients (ρ) were estimated between SWEMWBS and the five variables of physical and mental health including GHQ-12 score, EQ-VAS, happiness index, self-rated health and limiting longstanding illness. To account for the complex survey design (including non-response weighting), the rank of the variable was regressed on the rank of SWEMWBS. Since the Spearman correlation coefficient is equal to the slope of the regression between the ranked values of the two measures, its value was estimated by regressing the rank of participants on SWEMWBS on the rank of the physical and mental health variable [24]. In the present study, SWEMWBS was embedded in WEMWBS, and to avoid the issue of overlap, we also randomly split the data into two halves (N
1 = 13,584, N
2 = 13,311) and carried out the same analyses on the two independent samples for SWEMWBS (N1) and WEMWBS (N2), respectively. This is presented in the supplementary tables. In addition, to examine the internal consistency of the shorter scale as compared with WEMWBS, we calculated Cronbach’s alpha for each scale, with a value of over 0.70 considered to be indicative of acceptable internal reliability [25].
We expected correlations between physical and mental health variables and SWEMWBS to be of a similar magnitude to correlations with WEMWBS. In line with the literature on WEMWBS, we hypothesised that SWEMWBS would have statistically significant but moderate correlations with GHQ-12 [9] and lower correlations with variables that measure overall health, such as EQ-VAS, the former measuring mental ill health and the latter measuring overall health, which are different from mental wellbeing.
Similarities in association with social and health variables (research question 3)
To address research aim three, the three-category versions of SWEMWBS and WEMWBS were used as outcome variables in separate multinomial logistic regression models, comparing low with medium wellbeing and high with medium wellbeing. The decision to model SWEMWBS as a categorical variable rather than continuous was based on the different associations at the low and high end of the spectrum found in a previous study [8]. Modelling SWEMWBS as a continuous variable therefore would mean that some of these differing properties may be masked. Variables in single, fully adjusted models included sex, age group, marital status, ethnic group, highest educational qualification, economic status, equivalised income quintiles, self-rated general health, body mass index, fruit and vegetable intake, alcohol consumption, smoking status and survey year. To maximise all available cases on each variable, missing data were recoded into a ‘missing’ category, including missing 2012 data on fruit and vegetable consumption. However, we also repeated the analysis using listwise deletion which is presented in the supplementary tables. We prefer to present the former as the main model as it maximised all available information, including data from 2012.
Relative validity (research question 4)
To assess the extent of agreement between the two scales, we used the Bland–Altman method to plot the difference in scores for each respondent (WEMWBS–SWEMWBS) against the mean of the two scores. WEMWBS score was first divided by two to make the scale comparable to SWEMWBS, which ranges from 7 to 35. The Bland–Altman plot enables a visual inspection of the association between the differences in scores and the magnitude of wellbeing. Spearman correlation coefficients were calculated between SWEMWBS and WEMWBS, both overall and within subgroups, to explore similarities in the consistency of rankings. Since SWEMWBS was embedded in WEMWBS, potentially leading to upward bias in the estimates of correlation, we also present Spearman correlation coefficients between SWEMWBS and the seven items from the 14-item WEMWBS that were not included in the shorter scale. To explore the classification accuracy of SWEMWBS relative to WEMWBS, weighted kappa statistics were calculated between the three-category version of SWEMWBS and WEMWBS, and repeated within population subgroups. To assess the strength of agreement, we used the Landis and Koch classification [26]: slight: 0–0.20; fair: 0.21–0.40; moderate: 0.41–0.60; substantial: 0.61–0.80; and almost perfect: 0.81–1.00. Percentage agreement in the classification was also assessed.
Non-response weighting (which accounts for non-response by households, individuals within co-operating households and, for HSE 2012, non-response to the nurse visit) was applied to all analyses. Data management was performed using SPSS version 20.0 (SPSS Inc., Chicago, Illinois, US) and analysis was conducted using Stata version 14 (StataCorp LP, College Station, Texas, US) accounting for the complex sample design.