This pilot study used a sample from SA workplaces (N = 755). All participants were considered skilled workers, had a minimum 10 years of schooling and identified themselves as proficient in English, although only about 20% reported English as their first language. The educational inclusion criterion was partially to ensure a level of English proficiency sufficient to complete the LOT-R and other measures. As a result of a technical error, language data were only available for 82% of the sample, with the distribution presented in Table 1. Participants were recruited to complete the measures anonymously during visits to their workplaces, which comprised a wide range of occupational backgrounds (see Table 1).

The LOT-R was administered in its standard version in English. The original normative study (Scheier et al., 1994) reported a single factor accounting for 48% of variance, with α = 0.78. Test–retest reliability ranged from 0.68 over 4 months to 0.79 over 28 months. International studies reported a range of alpha coefficients, from 0.58 to 0.80 (see Table 2), whilst a local study (Koen et al., 2011) reported α = 0.59. Psychometric properties of the LOT-R from various cross-national studies are summarised in Table 2 for comparison with figures from the present sample.

Study participants also completed a selection of other measures. Not all participants completed all scales, and the total N for each scale will be indicated in the applicable tables. The following clinical measures of mental health were included in the study.

The Patient Health Questionnaire for depression (PHQ-9; Gilbody, Richards, & Barkham, 2007) is a nine-item measure that is scored on a four-point Likert scale (range 0–27), with higher scores indicating higher levels of depression. Moderate correlations have previously been reported for the LOT-R and PHQ-9 (Glaesmer et al., 2012), and other scales of depression (Zenger et al., 2013).

The Generalised Anxiety Disorder questionnaire (GAD-7; Löwe et al., 2008) is a seven-item measure that is scored on a four-point Likert scale (range 0–21), with higher scores indicating higher levels of anxiety. Moderate correlations have also been reported for the LOT-R and GAD-7 (Glaesmer et al., 2012), and other scales of anxiety (Zenger et al., 2013).

The CAGE questionnaire for problematic alcohol use (Dhalla & Kopec, 2007) is a four-item measure, scored as YES/NO (range 0–4), with higher scores indicating more problematic alcohol use.

The following measures of general psychological well-being were also included in the study:

Finally, two scales of psychological resilience were included to examine associations between the LOT-R and other measures from positive psychology:

Participants also completed a brief health questionnaire and were asked to indicate their health status with regard to debilitating acute or chronic diseases. Its purpose was to exclude severe medical conditions that could unduly influence responses to the psychological scales.

All statistical analyses were conducted by Statistical Package for Social Sciences (SPSS version 27) and analysis of moment structures (AMOS). Internal consistency was examined with Cronbach’s alpha, inter-item correlations and corrected item-total correlations. Against the ongoing debate on dimensionality, the lack of previous factor analytic studies from SA and the poor alpha coefficients found in the current sample, a confirmatory factor analysis (CFA) was conducted. Confirmatory factor analysis is a special form of factor analysis used to test whether data fit a hypothesised measurement model (Marker, 2002). The Maximum Likelihood estimator was used to explore a 1- and 2-factor model fit. For a CFA, the global fit χ² would be preferred to be small and not significant. This is rarely achieved, and the following indices with cut points were also taken into consideration: the root mean square error of approximation (RMSEA) should be < 0.06 to < 0.08 for continuous data, whilst both the comparative fit index (CFI) and the Tucker–Lewis index (TLI) should be > 0.95 (Schreiber, Nora, Stage, Barlow, & King, 2006).

The effects of socio-demographic variables were explored using Pearson’s correlation coefficients (for age effects) and t-tests for independent samples (for gender and language effects). For this analysis, language was coded into two groups, namely, English first language (21.3%) and non-English first language (78.7%).

As mentioned earlier, cross-national comparisons indicated variable scores between countries (Gallagher et al., 2013; Schou-Bredala et al., 2017), requiring individual LOT-R scores to be interpreted using local norms (Glaesmer et al., 2012). In line with best practice for LOT-R reporting (Glaesmer et al., 2012), SA normative data will be presented using standardised scores.

Construct validity was explored by calculating the associations of LOT-R scores and markers of clinical mental health (PHQ-9, GAD-7, CAGE) and general psychological well-being and resilience (SOS, STPI, DRS-15, MTQ-18) using correlation with Correction for Attenuation.

This study was a voluntary, anonymous, survey. The approval to conduct the study was received from the Stellenbosch University Health Research Ethics Committee (No. N20-07-078).

The 6-item LOT-R was subjected to CFA, and the results are presented in Table 3. All model fit indices of the CFA indicated that the assumption of a bidimensional structure of the LOT-R fits the data much better than the unidimensional structure. Although the two-factor model did not obtain a non-significant χ², the value was not excessively high. The RMSEA (0.053) was sufficiently small (< 0.06), and the CFI (0.93) was close enough to 0.95, although the TLI (0.83) was the exception (Table 3). The two subscales did not significantly correlate with each other (r = –0.007, p = 0.838), further suggesting two independent constructs, rather than a bipolar scale.

There were no significant age effects for the total score (r = 0.056, p = 0.125) or optimism subscale score (r = –0.031, p = 0.394), with a significant but very small effect for the pessimism subscale score (r = 0.096, p = 0.008). No significant gender differences were observed (Table 4a). For the sub-sample where language data were available, there was a significant difference in the mean scores between the English first language and non-English first language subgroups (Table 4b) although the mean difference was < 1, which may not be practically meaningful. Differences across language groups for the optimism subscale were non-significant, but significant for the pessimism subscale (mean difference = 1).

Construct validity indicators are reported in Table 5. Dispositional optimism correlated with clinical measures of depression and anxiety, and perceived stress overload, with moderate effect sizes. Correlations for the three clinical scales, as well as the SOS, were stronger for the pessimism than for the optimism subscale.

For measures of general psychological well-being, correlations with large effect sizes were observed for dispositional anxiety, curiosity and depression. Furthermore, correlations with large effect sizes were found for the DRS-15 and the MTQ-18. Again, in some cases (e.g. MTQ-18), stronger correlations were observed for the pessimism than the optimism subscale.

In the absence of significant age and gender effects, normative reference data were developed for the full sample (Table 6).

The total LOT-R mean score of 16.4 was comparable with most international samples (with the notable exception of Brazil; Bastianello et al., 2014), as were the subscale means. The significant differences in the mean scores from local studies may emphasise differences within the SA society – the current sample mean fell in-between the two previous reported local means (which represented discrete and highly individualised samples), and the internationally comparable mean score could possibly be attributed to the wide range of occupational domains included in the present sample, as opposed to the previous SA samples. This may speak to the need for adequately diversified sampling when doing any general health psychology research in SA.

Despite similar LOT-R mean scores, the standardised scores distribution for the SA sample (Table 6) differed in its nuanced spread to normative data from comparable international studies (cf. Glaesmer et al., 2012; Hinz et al., 2017; Schou-Bredala et al., 2017; Zenger et al., 2013), emphasising the requirement for local reference norms to enable meaningful interpretation of individuals’ scores.

Evidence of a two-factor scale structure was found in the results of the CFA, which suggested the hypothesised bi-dimensional model as best fit for this SA sample. The two factors displayed no significant correlation with each other and further appeared to display different patterns of correlation with other measures. In this regard, the three clinical scales, as well as the SOS and MTQ-18, showed stronger associations with the pessimism than with the optimism factor. Although the findings around dimensionality appear contrary to some recent reports, which suggested that the LOT-R taps a single construct (Cano-Garcia et al., 2015), it does follow the pattern found with European, South American and Asian population samples (Glaesmer et al., 2012; Lai & Yue, 2000; Zenger et al., 2013). More problematic is the poor internal consistency. The weak alpha stands in contrast with other reports and cautions against an uncritical use of the LOT-R in the African context. Language diversity, particularly in responding to negatively valanced items, may have contributed to the poor internal consistency.

No significant age or gender effects were observed, and it is consistent with previous studies. Home language offered a more complex outcome: whilst there was a significant difference in mean scores between English first language and non-English first language speakers, the difference was very small, and any practical meaning is not yet clear. Further research studies would be required to enhance confidence when using the English version test across SA language groups (at least in cases where appropriate English proficiency can be demonstrated). Interestingly, there was no significant difference across the two language groups for optimism mean scores, but a significant and larger mean difference for pessimism scores. In terms of direct language effect, the use of negatively valanced items – such as the three items of the pessimism subscale – has previously been implied as problematic in non-English first language-speaking SA samples (Arendse et al., 2020), where the negative wording may require a higher level of English proficiency to interpret accurately. A similar split between positively and negatively worded items have also been observed in Chinese samples (Lai & Yue, 2000). In terms of actual optimism, South Africa’s political history resulted in individuals raised with different levels of access to resources and ensuing beliefs regarding future opportunities, which could conceivably have influenced the development of dispositional optimism across different subgroups (which historically were often associated with language). This, however, remains speculative, and further research would be required to investigate these issues formally.

Evidence of construct validity was observed in the meaningful correlations with markers of clinical mental health, general psychological well-being and resilience, in this sample of healthy South Africans. In general, correlations with mental health markers were similar or slightly higher than what have been reported in previous studies. As expected, LOT-R scores were associated with depressiveness and anxiety, as well as problematic alcohol use. Emotional disposition, as a measure of general psychological well-being, and quantified by the STPI, showed the highest correlations with dispositional optimism, which closely reflected the original conceptualisation and reported correlations of Scheier et al. (1994). Furthermore, contrary to some previous reports (cf. Glaesmer et al., 2012; Zenger et al., 2013), a general pattern appeared where mental health constructs were better characterised by the presence of pessimism than the absence of optimism. This observation supports previous reports that pessimism, but not optimism, was a better predictor of longer term psychological and physical health outcomes (Robinson-Whelen, Kim, MacCallum, & Kiecolt-Glaser, 1997).

The association with perceived stress, whilst in the expected direction, was not as strong as previous reports (Chang, 1998; Yew et al., 2015), although this may be partly because of different measures used (i.e. PSS vs. SOS). The association with resilience measures followed the expected direction. The comparatively weak correlation with the DRS-15 may be instrument, rather than construct, related, as a previous study recommended caution when using the DRS-15 for measuring resilience in the SA context (Arendse et al., 2020). The strong correlations with the MTQ-18 suggest that both the LOT-R and the MTQ-18 may be useful to measure constructs of positive psychology in SA. Across the various instruments, full-scale correlations were stronger, and until further research is carried out, the use of total scores rather than subscale scores would be recommended for future SA health psychology studies. It was noteworthy that the pattern of correlations was consistent across measures of psychological distress (e.g. PHQ-9, GAD-7), as well as measures of psychological well-being (e.g. STPI, MPQ-18).

The evidence of construct validity – in its association with measures of mental health and psychological well-being – provide support for the use of the LOT-R in local health psychology research.

A number of limitations to this study need to be mentioned. It was a pilot study, with concomitant limited size, and the sample cannot necessarily be considered representative of a general population of proficient English speakers. Furthermore, English proficiency was assumed. The assumption was based partly on self-evaluated proficiency, and partly on reported educational attainment, and it is recognised that education may not be a good proxy for language proficiency in SA. Future SA studies will need to expand sampling to clarify language effects, as well as repeat factor analysis and internal consistency calculations with larger samples. Expanding studies to include other samples of sub-Saharan Africa would further elucidate the influence of localised environments.

When further validation for the use of the LOT-R in African contexts has been obtained, it can be productively applied to local health research. The LOT-R was originally conceptualised to express relationships between dispositional optimism and long-term psychological and physiological health outcomes, and could be used for the same purpose in longitudinal studies to explore relationships between dispositional factors and health in local contexts.