Background
Mental health is essential for individuals’ overall well-being and relates to their emotional, psychological and social functioning [
1]. The World Health Organisation (WHO) defines mental health as a state of well-being beyond the mere absence of disease and emphasized the need for promoting mental well-being, creating space for health-oriented approaches to mental health over illness-oriented services. Several successful research and scientific efforts have since enriched our knowledge about mental health and resulted in the development of a wide array of measurement tools to assess different aspects of mental well-being and evaluate the impact of mental health interventions. The Positive and Negative Affect Scale [
2], Satisfaction With Life Scale [
3], Psychological Well-being Scales [
4], Affectometer [
5], Short Form (SF)-36 [
6], WHO-Five Well-being Scale [
7], Warwick Edinburgh Mental Well-being Scale (WEMWBS) [
8], Positive Mental Health Instrument (PMHI) [
9] and the Positive Mental Health Questionnaire [
10] are some of the measures used globally in the assessment of mental well-being.
Of these, the PMHI has been widely used in Singapore's population and validated among the general population samples and mental health service users [
9,
11,
12]. The multidimensional scale has 47 items representing six domains of positive mental health - general coping, emotional support, spirituality, interpersonal skills, personal growth and autonomy, and global affect. The body of work around the PMHI has so far provided important information on sociodemographic variations in the level of positive mental health in terms of gender, ethnicity and marital status in the Singapore population [
13]. In general, women scored higher on emotional support and lower on personal growth and autonomy domains, non-Chinese ethnicities had higher levels for all positive mental health domains and being married (vs being separated/ divorced) was associated with higher spirituality. Studies in clinical populations have found gender differences in positive mental health of patients with schizophrenia [
14] and positive association of positive mental health with life satisfaction and general functioning, and inverse relation with depressive symptom severity among patients with mental disorders [
15]. A study conducted among mental health professionals found that employee positive mental health was associated with their life satisfaction and profession - compared to allied health employees, psychiatrists had lower scores on spirituality and nurses had higher scores on personal growth and autonomy [
16]. More recently, population norms were estimated in a representative national sample in Singapore [
12]. Current studies with the PMHI involve associations with negative aspects of mental health, lifestyle and behavioral factors (Unpublished). These findings have important practice and policy implications for mental health promotion in the general population, clinical settings and workplace.
An important limitation of the PMHI is that although it is relatively quick to administer, it can still impose considerable respondent burden in studies involving multiple assessments or in frail populations. In addition, the PMHI has a multidimensional structure and items belonging to four of the subscales are not presented in a sequential order, instead they are spread across the tool. The subscale score calculations thus involve using an algorithm making it challenging for quick, routine use in clinical practice. A short 19-item multidimensional version was previously developed in Singapore that provided advantages by halving the administration time and showing low ceiling effect [
17]. However, it still does not adequately address the PMHI limitations. The availability of a much shorter unidimensional version would tackle these limitations more efficiently and enhance its applications in routine use.
This study aimed to: (i) develop a short unidimensional scale that could offer brevity over the 47-item PMHI while retaining its construct properties using a Rasch model approach on pooled data from three earlier studies, (ii) establish its reliability and internal validity in the cross validation sample, and (iii) confirm its psychometric properties by external validation in a large representative general population sample.
Discussion
This study used a Rasch model approach to develop a short unidimensional measure from the 47-item PMHI that can be used to efficiently assess the level of positive mental health among the adult multi-ethnic population in Singapore. The new measure not only showed high agreement with the original measure but it was also valid and reliable, contained acceptable levels of DIF and produced significant correlations in the expected directions with convergent and discriminant measures.
Rasch analysis has been extensively applied in instrument development and improvement in order to ensure that they provide unidimensional measurement of a construct [
22]. It is a psychometric technique that increases the precision with which researchers construct instruments taking into account respondents’ performances. The technique provides several advantages in the assessment of health outcomes that are not directly measurable by offering a framework for item selection and the benefit of allowing comparisons across diseases and subgroups [
28]. A number of outcome measures including the Short WEMWBS [
8], Patient-Reported Outcomes Measurement Information System (PROMIS) [
29], EUROHIS-QOL 8-item index, a short version of the 26-item WHO Quality of Life questionnaire (WHOQOL-BREF) [
30] and the Multiple Sclerosis Quality of Life (MSQOL)-54 [
31] were developed using Rasch analyses. Such patient reported and preference based outcome measures usually have ordinal scales and are prone to high floor or ceiling effects [
32]; Rasch analysis has been particularly useful in reducing these. The R-PMHI had very low floor effect (0.3%) and a ceiling effect of 6%. This is high compared with 0.94% ceiling effect seen for total PMH measured by the original measure but much lower than the 10–14% ranges seen for its six subscales [
12]. It is possible that items with higher ceiling effect were retained in the 6-item R-PMHI despite the Rasch framework. Since ceiling effect can reduce scale responsiveness or its capacity to detect changes over time [
29], it is important to assess this for the R-PMHI in prospective and experimental studies.
Item response theory models (IRT) systematically evaluate item bias using DIF which refers to the situation where members from different groups (e.g., age, gender) who have the same level of positive mental health have a different probability of selecting a certain response to a particular item. IRT-DIF analysis on R-PMHI identified two items with significant DIF (Table
4). The original item selection and development of the 47-item PMHI were based on IRT analyses whereby items with significant IRT-DIF were removed from the item pool [
9]. None of the six items retained in the R-PMHI had demonstrated DIF in the general population previously. However, a subsequent investigation among mental health service users had identified one of these items, ‘How often in the last 2 weeks have you felt ..Calm’ to have DIF by age [
11]. Similar result was obtained in the current analysis. In addition, the item ‘I make friends easily’ exhibited DIF by gender and ethnicity which was not observed in the earlier studies. The impact of DIF observed in the R-PMHI should be considered while comparing these groups. However, a study in the Singapore general population found that although 20 of the SF-36 questionnaire items had DIF by the history of chronic conditions, its impact on the assessment of health related quality of life was minimal [
33]. Scott and colleagues [
34] also caution against employing DIF analysis on its own and highlight the importance of assessing other statistical and psychometric results “when deciding whether a particular DIF effect is of sufficient practical importance to require modification of an item or scale”.
The R-PMHI also had four items with infit and outfit statistics over the threshold (− 2 to 2, Table
3). The presence of under fitting items in instruments can potentially impact its validity, whereas overfitting items tend to overestimate differences in raw scores [
35]. The former can lead to under-detection of health problems (e.g. false negatives on screening measures), while the latter interferes in comparisons within and between individuals. Although the clinical impact of erroneously removing misfitting items has not been investigated, research indicates that retaining misfitting items has little or no impact on the efficacy of measures [
36] and therefore, inclusion of these items in the R-PMHI is unlikely to impact its application. Moreover, while threshold scores below − 2 suggest statistically significant over or under fitting for four items, their mean squares were higher than 0.70 (ranging from 0.74–0.88), suggesting that the actual size of the misfit is not large enough to be concerning.
Given the satisfactory fit to the Rasch model for the 6-item R-PMHI, confirmation of its unidimensionality and the validity of the measure were established in two samples - an internal random split-half sample embedded within the data and an external dataset (Table
5). The R-PMHI also demonstrated high internal consistency reliability in these samples (Cronbach’s alpha of 0.764 and 0.806, respectively). The original PMHI has consistently generated Cronbach’s alphas around 0.9 in previous studies [
9,
11,
17]. Although Cronbach’s alphas within 0.70–0.95 indicate superior reliability, high values could indicate item or subscale redundancy [
37]. In that, R-PMHI could serve useful by managing the redundancy in the original measure and increasing efficiency of measurements while being psychometrically sound.
The study also estimated preliminary positive mental health values as established by the R-PMHI and explored socio-demographic correlates of positive mental health using the measure. The 47-item PMHI generated values in the range of 3.93 to 4.61 for total positive mental health among populations comprising mental health service users [
38] and the general population [
12]. The R-PMHI score of 4.86 seen among the general population in this study is much higher than the earlier estimates. While it is possible that participants in the general population sample had good overall mental health, it is also likely that the R-PMHI includes items with higher ceiling effects resulting in a higher composite score. These should be considered while interpreting and comparing R-PMHI scores. Future development of the R-PMHI should involve studies in clinical or vulnerable populations and possible expansion of the 6-point Likert scale to 7 to 10 response options which has been shown to reduce ceiling and floor effects and provide better normalization in the data [
39].
R-PMHI scores were associated with ethnicity in this study (Table
7). This finding is consistent with earlier studies [
13,
15,
38]. However, R-PMHI scores did not show a significant relationship with marital status as seen in all previous studies using the PMHI where those who were married had significantly higher total PMH as compared to those who were never married and/or were divorced /separated/ widowed [
12,
13,
38]. While similar findings were observed with the SWEMWBS in UK [
40], the association of marital status with mental well-being components has not been consistent [
41,
42]. This highlights the importance of evaluating the living arrangement such as co-habiting over just being married as well as quality of marital life to be considered while investigating this association. This was, however, not captured in our data and remains a future research goal while understanding the R-PMHI score and its correlates.
A challenge, as with any unidimensional scale, is that a single score reflects all the components of positive mental health [
43]. Thus, unidimensional rating scales provide relatively less information concerning the comprehensive nature of a construct. Positive mental health being a multidimensional construct ideally requires a multidimensional measure for its assessment. However, the length of the 47-item PMHI poses several challenges for its routine clinical use. It is crucial to evaluate if any construct properties were lost during item reduction. Given that the current short measure did not comprise any items belonging to general coping and spirituality domains (Table
3), it will be of value to evaluate the effect of domain exclusion in future studies.
The six items that remained after the item-reduction partly represent the construct of ‘emotional intelligence’ [
44] which relates to understanding one’s emotions as well as others, and believed to comprise one or more of five components - ‘self-awareness, self-regulation, motivation, empathy and social skills’ [
45]. Emotional intelligence has been linked to productivity, happiness and psychological well-being and forms an integral part of positive psychology [
46,
47]. There are clearly several overlaps and correlations between positive mental health or well-being and emotional intelligence – both can be acquired, are used in targeted interventions in psychotherapy, have shown links with health behaviors such as exercise and linked to self-efficacy, functioning and reduction in depressive symptoms and suicidality [
47,
48]. The difference probably lies in their application to mental health promotion; while emotional intelligence has been largely used in reference to students, working adults and specifically healthcare professionals, ‘positive mental health and well-being’ are often applied to population level health promotion. Future research on the R-PMHI would benefit by understanding in what way these constructs differ from or contribute to one another.
Limitations
Some limitations of this work should be considered while interpreting the results. Although the pooled samples used for this study were large, represented varied settings and included participants from a wide range of socio-cultural backgrounds, majority were recruited via convenience rather than probabilistic sampling. The generalizability of these findings thus warrants further investigation. The development and validation studies were also restricted in the number of variables and measures that were used to assess concurrent validity. Further research will be needed to examine the validity of the R-PMHI in relation to domain-specific positive mental health measures that were employed in the development of the original PMHI. This could provide important information on the construct properties of the short tool vis-a-vis deletion of items belonging to specific subscales. Additionally, the R-PMHI development was guided by a statistical construct and psychometric analyses, where data from previous studies were analyzed without administering the 6-item scale to a new study population. Hence, further investigations and external validation of the measure should be considered. Putting these limitations aside and the need for further research that accompany the development of any new measure, the development and validation of the R-PMHI is adequate whereby it can be offered to adult Asian populations as a brief measure that is reliable, valid, and efficient in the assessment of positive mental health.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.