A Cross-Cultural Evaluation of the Personal Wellbeing Index – School Children in Samples of Australian and Portuguese Adolescents

Cross-cultural comparison studies on the SWB variable are valuable because they enable governments and policymakers to better understand how well their citizens are faring relative to other countries. Such knowledge has the potential, for example, to enable decision makers from low scoring countries to seek policy direction from countries who rank high according to average levels of SWB (UNICEF 2013). According to Cummins et al. (2009), enhancing the SWB of populations not only enhances the functioning of people, but of the population as a whole. Thus, the implications of cross-national comparison studies lie in their ability to stimulate discussion on matters that concern population SWB as well as the development of policies and practices that offer citizens the best possible opportunity for living happy and satisfying lives. While the SWB of adult populations has been explored extensively, there is a relative paucity of research concerning the SWB of children and adolescents (Huebner et al. 2004). Recently, as part of a global effort to better understand the various influences on SWB in young people, researchers have begun measuring and reporting SWB across cultures (e.g., UNICEF 2013). At the heart of such endeavours are SWB measures that function equivalently across different cultural groups. However, there is limited evidence for the cross-cultural equivalence of such measures (Casas et al. 2013). The current study addresses this important psychometric issue by evaluating the measurement invariance of a commonly used adolescent SWB measure. This introduction will begin by presenting SWB Homeostasis Theory (Cummins 2010) as the underlying theoretical paradigm that describes stability and change in the SWB construct. Methodological issues associated with comparing adolescent SWB across cultures will then be discussed. Finally, measurement invariance will be outlined as the necessary precondition for valid and unambiguous cross-cultural comparisons on the SWB variable.

SWB and the Theory

Subjective wellbeing (SWB) can be defined as a normally positive state of mind that involves the whole life experience (Cummins 2010). While there have been a number of different theories proposed over the years that attempt to explain the normally positive and stable nature of SWB frequently observed within the literature (e.g., Brickman and Campbell 1971; Eid and Diener 2004; Headey and Wearing 1989; Michalos 1985), one theory that offers the most comprehensive and empirically validated description of this construct is SWB Homeostasis Theory (Cummins 2010). Homeostasis theory proposes that, in a manner analogous to the maintenance of body temperature, each person has a biologically determined level of SWB that is actively maintained and controlled within a narrow, positive range of values around a ‘set-point’ (Cummins 2010). According to recent empirical evidence by Cummins et al. (2014), individual SWB set-points normally range between 70 and 90 points on a standard 0–100 point scale, with an average of 80 points. This range corroborates data collected as part of the Australian Unity Wellbeing Index over the years 2001–2013, with the normal range for SWB for groups in the Australian population calculated between 73.7 and 76.7 points (Cummins et al. 2013). According to Cummins (2010), humans have evolved to experience a level of SWB that is positive and stable because feeling good about oneself serves an adaptive function and provides the motivation for living.

At the heart of SWB is a construct named Homeostatically Protected Mood (HPMood; Cummins 2010). Initially termed ‘Core Affect’ by Russell (2003), HPMood can be conceptualised as the affective core of SWB and the positive mood that homeostasis seeks to defend (Blore et al. 2011; Davern et al. 2007; Tomyn and Cummins 2011). According to Homeostasis Theory, under normal, unthreatening conditions, SWB is maintained by the homeostatic system at a steady level within the normal range (Cummins et al. 2014). However, when a person is faced with a sufficiently adverse level of challenge (e.g., divorce, severe illness, job loss) and the demands placed on a person exceed his or her coping resources, homeostasis can be defeated (Cummins 2010). When homeostasis fails, HPMood loses its association with SWB as negative affects tied to the source(s) of challenge assume control (Cummins 2010), resulting in what we refer to as depression. Return to the set-point is then believed to be determined by a person’s access to resources, such as social support and psychological services.

Homeostasis Theory, in addition to population normative data for SWB, provides a theoretical and empirical basis that helps facilitate the identification of people who may be experiencing homeostatic challenge or defeat and in need of supportive services. However, the normative data currently available are derived from adult samples, with a dearth of comparative adolescent data available. Thus, our understanding of adolescent SWB is again argued to be limited (Casas et al. 2011; Huebner et al. 2004; Park 2004).

Cross Cultural Response Bias and Measurement Invariance

While revealing differences in adolescent SWB across countries is a desirable outcome, these emerging efforts are problematic for a number of reasons. A potentially serious threat to the validity of these large-scale comparison studies is that the SWB measures employed have not been validated for use cross-culturally. Determination of an instrument’s cross-cultural utility is crucial because people in different cultures may employ varying response styles when answering survey questions, threatening the comparability of scores (Diener et al. 1995; Oishi 2006; Van de Vijver and Poortinga 1997). For example, in collectivist cultures where value is placed on modest self-presentation (e.g., in Taiwan, Japan, and China), people are less likely to rate themselves at the extreme ends of a Likert scale, when compared to people from individualistic cultures (e.g., North Americans; Lee et al. 2002; Stening and Everett 1984). These results have been replicated using adolescent samples, with the same response-bias emerging in the data (Chen et al. 1995). Therefore, it can be difficult to determine whether cross-cultural differences reflect true variations on the underlying SWB construct, or whether these differences reflect measurement bias.

Tests of measurement invariance can be used to determine whether a scale measures the same underlying construct in various groups under investigation (Meredith 1993). Measurement invariance tests are typically undertaken using multiple-group confirmatory factor analysis (CFA; Gregorich 2006). This procedure compares a hierarchical set of measurement models to evaluate different levels of invariance, namely: Configural, Weak, Strong, and Strict (Gregorich 2006). Each level of measurement invariance provides additional information regarding the cross-cultural utility of the instrument under investigation. For example, evidence of strict invariance (the most stringent level of invariance) indicates that differences between groups on mean SWB scores reflect substantive differences in the underlying construct and are not due to measurement artefacts, such as response biases that may be inherent within particular cultural groups (Vandenberg and Lance 2000). Thus, if a research objective is to compare mean SWB scores across cultures for the purpose of rank-ordering nations from highest to lowest (e.g., UNICEF 2013), then strict invariance between samples on the same or parallel versions of the measure must first be established. Without the establishment of measurement invariance, confidence is limited in statements such as “The Netherlands heads the league table of children’s subjective well-being with 95 % of its children reporting a high level of life satisfaction” and “Only in Poland and Romania does the ‘high life satisfaction’ rate fall below 80 %” (UNICEF 2013, p. 39).

According to Tomyn et al. (2013), measurement invariance is a commonly neglected psychometric property of frequently used adolescent SWB measures. This lack of measurement invariance testing is not exclusive to adolescent SWB research, but rather a pervasive issue in the social sciences (Gregorich 2006). Gregorich proposes two key explanations for this concerning trend in the social sciences - the general lack of awareness in the scientific community regarding the importance of invariance testing; and because many researchers lack the technical skills required to conduct such tests. However, while measurement invariance testing of adolescent SWB measures is scant, a small number of researchers have recently directed their attention towards this important area of cross-group investigation (e.g., Casas et al. 2013; Tomyn et al. 2013). The only adolescent SWB instrument that appears to have undergone measurement invariance testing, is the Personal Wellbeing Index-School Children (PWI-SC; Cummins and Lau 2005), the same instrument adopted by the Children’s Society (2014) as one of their cross-cultural wellbeing outcome measures. In one such study, Casas et al. (2013) investigated the cross-group equivalence of the PWI-SC for use with adolescent samples from Algeria and Spain using multiple-group CFA. Strict measurement invariance was not established, indicating that quantitative comparisons of SWB composite scores between the countries investigated are not scientifically defensible. The authors concluded that PWI-SC scores appear to be only partially comparable across cultures. However, it is important to note that Casas et al. used a modified version of the PWI-SC by including additional items not part of the original scale for exploratory purposes.

In another study, Tomyn et al. (2013) investigated the cross-group equivalence of the original unmodified 7-item PWI-SC in samples of Indigenous and non-Indigenous Australian adolescents using multiple-group CFA in AMOS. Results demonstrated strict measurement invariance for the PWI-SC in both samples, which supports quantitative comparisons between groups as valid. The researchers concluded that the PWI-SC appears to measure the same underlying SWB construct in each sample. A major implication of this study is that there now exists evidence that a measure of SWB is suitable for use with both Indigenous and non-Indigenous Australian adolescents—a measure which could be used in the future as a means to help monitor progress toward ‘closing the gap’ between these two groups.

Summary and Study Aims

It is clear from the limited number of studies that measurement invariance testing of adolescent SWB instruments is in its infancy. Thus, it remains unclear whether commonly used adolescent SWB instruments, such as the PWI-SC measure the SWB construct equivalently in adolescents from different cultural groups. Measurement invariance is a necessary precondition for valid and unambiguous cross-group comparisons. To this end, the aim of the present study is to test whether the PWI-SC functions equivalently within adolescent samples from two different cultures using multiple-group CFA. In light of recent evidence supporting the cross-group utility of the PWI-SC in Australia (Tomyn et al. 2013), and given that the PWI-SC is designed as a cross-cultural instrument for measuring adolescent SWB, it is hypothesised that the Australian and Portuguese versions of the PWI-SC will function equivalently.

Participants

Participants comprising the Australian adolescent group were a convenience sample of 1104 Victorian high-school students. Participants ranged in age from 12 to19 years (M = 14.42, SD = 1.63). There were 682 males (61.8 %) and 422 females (38.2 %). Participants comprising the Portuguese adolescent group were a convenience sample of 573 Portuguese high-school students. Participants ranged in age from 12 to 18 years (M = 14.32, SD = 1.72). There were 268 males (46.8 %) and 290 females (50.6 %). Fifteen respondents did not indicate their gender (2.6 %).

Materials

SWB was measured using the PWI-SC (Cummins and Lau 2005). The PWI-SC is a 7-item measure of SWB that asks respondents to indicate their level of ‘happiness’ with seven life domains using an 11-point end-defined rating scale ranging from 0 (Very sad) to 10 (Very happy). The seven are Standard of Living, Health, Achieving in Life, Relationships, Safety, Community Connection, and Future Security. For example, the Future Security item asks respondents, “How happy are you about what may happen to you later in your life?” Previous research reports good internal reliability (Cronbach’s α = 0.82) and construct validity for this scale in Australia, with a one-factor structure that explains approximately half of the variance (Tomyn and Cummins 2011). In the present study, the corresponding Cronbach’s alpha was 0.81. The Portuguese students completed a translated version of the PWI-SC in Portuguese, with adequate psychometric properties, in terms of reliability (Cronbach’s α = 0.84) and the single-factor structure explaining approximately half of the variance (Dias and Bastos 2014).

Procedure

The Australian data was collected from three independent high schools across two studies conducted in the Melbourne metropolitan region in 2010 and 2011. Ethics approval was obtained from the Victorian State Department of Education to enter schools. Of the seven schools invited, three agreed to take part. Students were then approached by the lead researcher in allocated class time and invited to participate. If consent forms were returned signed by parents or guardians, students then received the pencil and paper questionnaire, which they completed in regular class time.

The Portuguese data were obtained following an email that was sent to members of the International Wellbeing Group (an international collaboration of SWB researchers) by the Lead Researcher from Australia asking if any researchers were willing to share adolescent SWB data from their respective countries for the purpose of participating in a cross-cultural research project. Researchers from India, the United States, and Portugal replied. However, only the Portuguese data was deemed adequate for use in the proposed study following an initial integrity check of the data (e.g., ensuring the correct 11-point end-defined scale was employed). Following ethics approval, the Portuguese data were collected in Portugal from two public schools in suburban, middle-class areas in 2012. After obtaining parental consent, students completed a translated pencil and paper version of the PWI-SC in their classrooms. A secondary, de-identified data set was sent via email from the Portuguese research partner to the student researcher for data cleaning and analysis in SPSS and AMOS.

Data Analytic Strategy

Measurement invariance testing was conducted in AMOS version 17 (Arbuckle 2006) using multiple-group CFA. The first step involved establishing a baseline model fit for each sample respectively (model 1 for Australian sample; model 2 for Portuguese sample). The baseline model assessed whether, in both groups, the seven PWI-SC items loaded onto a single factor, confirming the one-factor structure proposed by the scale developers (International Wellbeing Group 2013). Following guidelines suggested by Byrne (2010) and Hu and Bentler (1999), model fit was examined using the following fit statistics: Comparative Fit Index (CFI; CFI ≥ 0.95 for good fit, CFI ≥ 0.90 for acceptable fit), Root Mean Square Error of Approximation (RMSEA; RMSEA < 0.05 for good fit, ≤ 0.08 for adequate fit), Standardized Root Mean Square Residual (SRMR; SRMR < 0.05 for good fit), Relative Chi-square (χ ²/df: χ ²/df < 5 for adequate fit), and chi-squared (χ ²; significant values reflect poor model-fit). It is important to note that χ ² and χ ²/df (unlike CFI, RMSEA, and SRMR) are both sensitive to sample size and should be interpreted in relation to other fit indexes in the determination of overall model fit (Cheung and Rensvold 2002; DiStefano and Hess 2005).

Once adequate model fit was established for the baseline models, four increasingly stringent measurement invariance assumptions were tested in sequence using multiple-group CFA (Gregorich 2006). The first and least restrictive model (model 3a) tested for configural invariance. Configural invariance requires that items load onto the same factors across data sets, but allows item parameters (factor loadings, residual variances, and intercepts) to vary across groups. The second model (model 3b) tested for weak invariance. To test for weak invariance, factor loadings are constrained to be equal between groups, and model fit re-evaluated. Evidence of adequate fit for this model indicates that similar, but not necessarily identical, latent constructs have been measured across groups (Widaman and Reise 1997). The third model (model 3c) tested for strong invariance. This involves constraining factor loadings and item intercepts to equality across groups to evaluate potential systematic bias in responses from one group to another. The fourth model (model 3d) tested for strict invariance by constraining residual variances to equality across groups. Evidence of adequate fit for this model indicates that differences on item composite scores (i.e., mean scores) can be attributed to substantive differences in the underlying construct and are unlikely to be due to measurement bias (Gregorich 2006).

Adequacy of model fit for each of the four measurement invariance models (3a-3d) was determined using the same criteria mentioned above in addition to the CFI change statistic (ΔCFI) which examines the change in CFI when additional cross-group constraints are imposed on a measurement model (ΔCFI ≤ 0.01 for acceptable degree of change; Cheung and Rensvold 2002).

Data Screening and Cleaning

To standardise data, all reported values for the PWI-SC have been converted to a Percentage of Scale Maximum (%SM) which converts data onto a 0–100 scale (Cummins and Lau 2005). There were no missing data in the Australian or Portuguese data sets. All cases were then examined for response set. This is deemed to occur when a respondent consistently scores at the scale minimum (0) or maximum (10) for all seven of the PWI-SC domains. Two response sets were found in the Australian sample (all 10’s). These cases are considered unreliable and were subsequently removed prior to the main analyses as suggested by the scale developers. Examination of z-scores revealed no univariate outliers on domain happiness variables with a z score > ± 2.88, a criteria recommended by Tabachnick and Fidell (2007). However, 49 cases (33 Australians and 16 Portuguese) were identified as multivariate outliers, with a Mahalanobis distance greater than the critical value of 24.32 for the corresponding degrees of freedom (7). These cases were deleted prior to the main analyses, as recommended by Tabachnick and Fidell (2007). Absolute skew and kurtosis values were within the acceptable ranges of ±2.0 and ±7.0 respectively, demonstrating that domain scores were normally distributed in both data sets (Curran et al. 1996). Investigation of histograms revealed normal distributions with a negative skew. However, a mild violation of multivariate kurtosis was observed in both the Portuguese (Mardia’s coefficient = 19.36, p < .001) and Australian (Mardia’s coefficient = 20.73, p < .01) samples. Accordingly, a Bollen-Stine bootstrapped corrected approach (on 2000 samples of the original data set) was used to correct for multivariate non-normality, and to allow for a more accurate estimation of model fit statistics in measurement invariance tests (Byrne 2013).

Descriptive Statistics and Correlations

Means, standard deviations, and correlations between the PWI-SC domains for the Australian and Portuguese samples are presented in Tables 1 and 2, respectively.

All within-group inter-domain correlations were significant and there were some clear differences between the patterns of these correlations between the two samples. For example, Portuguese and Australian respondents differed on correlations between Community Connection and other variables, with Portuguese showing a stronger correlation to all other domains, including with the SWB composite. Examination of Fisher’s Z-transformation revealed these were all significant differences (p < .01). In other words, Community Connection was more predictive of all other domains among Portuguese than among Australian adolescents. Moreover, careful examination of the Fisher’s Z-transformation of the two correlation matrices revealed that groups significantly differed on the relationship between Achieving in Life and Health (p < 0.05), Safety and Community Connection (p < .05), and between Safety and Future Security (p < .05).

Multi-Group CFA

As shown in Table 3, unidimensional modeling of the PWI-SC items provided adequate fit for both Portuguese and Australian adolescent sample groups.

Although χ ² was significant and χ ²/df was above the acceptable range (for the Australian sample; model 1), this likely reflects the moderate to large sample size rather than model misspecification given that other fit indices (i.e., CFI, RMSEA, and SRMR) were all at good, or at least acceptable levels. It is important to note that in the baseline models (i.e., models 1 and 2) and subsequent invariance models (i.e., models 3a-3d), theoretically acceptable co-variances were fitted between error terms to better model fit in both samples. For example, error terms were fitted between interrelated SWB domains with moderate to high correlations. These included: Achieving in life and Safety (r = 0.39 and 0.41), and Relationships and Future Security (r = 0.39 and r = 0.54).

Table 4 presents the evaluation of measurement invariance across both samples. At no point do these models exhibit a change in CFI that exceeds the acceptable minimum change cut-off criterion of 0.01. Furthermore, at each step, the fit statistics suggest that the models provide an acceptable representation of the data. It is clear that placing increasingly stringent invariance conditions on the data did not lead to a substantial reduction in model fit. Again, significant χ ² values (for models 3a–3d) likely reflect the moderate to large sample sizes employed in this study. Collectively, these findings suggest that the PWI-SC performs comparably across Australian and Portuguese adolescent groups.

Between Group Differences in SWB and Domain Happiness Scores

After establishing measurement invariance, an exploratory MANOVA was conducted to investigate differences in mean SWB and domain happiness scores between the Australian and Portuguese samples groups. This was significant, Pillai’s Trace = 0.105 F(7, 1620) = 27.206, p < .001, ηp2 = 0.105. Power to detect the effect was very high (1.00). Prior to investigating the Univariate ANOVA results, the homogeneity of variance assumption was tested for each of the eight dependent variables. Levene’s test of equality of error variances was not met for all comparisons (<0.05), however, examination of standard deviations (see Tables 1 and 2) revealed that none of the largest standard deviations were greater than four times the corresponding smallest, suggesting that the tests are robust to this violation (Howell 2010). Table 5 displays the means and standard deviations for each DV, and corresponding degrees of freedom, F-statistic, significance value and effects size (partial η2) for each Univariate ANOVA.

As shown in Table 5, four one-way ANOVA’s are statistically significant (p < .001). Dunnett’s C post-hoc comparisons were employed as a conservative test of the significance of differences between the two groups, as recommended by Tabachnick and Fidell. While groups did not differ significantly on the SWB composite variable, the Australian sample scored higher on ‘Standard of living’, ‘Safety’ and ‘Future security’; while the Portuguese sample scored higher on ‘Health’. These results will be discussed.

The aim of the current study was to use tests of measurement invariance to determine the extent to which the Australian / English and translated Portuguese versions of the PWI-SC function equivalently. Similar to Tomyn et al. (2013), we evaluated whether both versions of the PWI-SC have the same factor structure (configural invariance, underlying meaning (weak invariance), whether they exhibit similar levels of response bias (strong invariance), and are subject to similar levels of error in measurement (strict invariance). The establishment of these forms of measurement invariance supports defensible cross-group quantitative comparisons of SWB between Australian and Portuguese adolescent populations using this measure (Gregorich 2006).

Consistent with the major hypothesis, the results demonstrated strict invariance between both versions of the PWI-SC. This result not only lends additional support that the PWI is a uni-dimensional measure of subjective wellbeing (IWG 2013), but the imposition of increasingly stringent cross-group equivalence criteria (for factor loadings, item intercepts and item error variances) failed to worsen this model fit. Thus, the data support quantitative between-group comparisons between Australian and Portuguese SWB data as valid. This finding is consistent with those of Tomyn et al. (2013), who determined measurement invariance in samples of Indigenous and non-Indigenous Australian adolescents using the PWI-SC, and provides further evidence of the cross-group utility of the PWI-SC for cross-cultural research purposes (Casas et al. 2011; Lau et al. 2005). The finding of measurement invariance also suggests that the PWI-SC may be measuring the same underlying SWB construct in both sample groups, thus supporting SWB Homeostasis theory and the notion that individual differences in SWB for all people may be explained by individual differences in HPMood, the construct believed to be at the heart of SWB judgments and the construct that homeostasis seeks to defend (Cummins 2010). It is plausible that the two samples responded similarly to PWI-SC items, at least in part, because the same underlying HPMood factor described is the driving force behind SWB judgments, as claimed by Blore et al. (2011), Davern et al. (2007) and Tomyn and Cummins (2011). Indeed, mean SWB scores of 77.29 and 78.11 presented in Tables 1 and 2 in the Australian and Portuguese samples, respectively, are not different to the hypothesised normal range proposed by Cummins et al. (2014) for populations operating normally within their set-point range (70–80 points).

While the PWI-SC appears to be a suitable measure of SWB in both cultures, some differences were evident in inter-domain correlations. The most obvious difference was in the domain of Community. This domain correlated more strongly with the other domains in the Portuguese sample. In other words, community connection appears more predictive of the other six domains in the Portuguese sample compared to the Australian sample. In explanation of this finding, studies conducted in the Portuguese context highlight the importance of strong feelings of community belonging as central to feelings of wellbeing and that are associated with positive relations with peers (Elvas and Moniz 2010; Gaspar et al. 2006). Other studies (e.g., Ciochină and Faria 2009) suggest that strong connections to community in Portugal are associated with more a collectivist culture. Based on accumulated evidence, it appears that the concept of ‘community’ may play a stronger role in the construction of SWB in Portuguese adolescents. Interestingly, however, there was no significant difference in the means for this domain between the two samples. Thus, future research is needed to gain a better understanding of the relationship between community connection and SWB and the mechanisms by which it may play such an important role in supporting overall wellbeing in Portuguese culture and how this relates to Australian culture.

Other notable between group differences were observed on the domains of Standard of Living, Safety and Future Security, with Australian adolescents scoring significantly higher. It is possible that economic factors, such as differences in average family yearly earnings and employment opportunities, may explain differences on the ‘Standard of Living’ and ‘Future security’ domains, at least in part. These seem as plausible explanations in the wake of the 2008 Global Financial Crisis, with Portugal reported to have experienced higher rates of unemployment and general diminishing of family earnings as a consequence (OECD 2014). While both means were approximately normal, a higher mean for ‘Safety’ in the Australian sample was also an interesting finding. However, this finding may be cultural as Australia has a strong prevention focus with among the world's highest safety standards with the Australian Safety Standards regime, and has a culture surrounding safety: Victoria, Australia’s southern-most mainland state, was the first jurisdiction in the world to introduce mandatory seat-belt laws (Evans 1985), including anchorages for child seat harnesses in 1964, Australia founded the first child-safety focused charity in 1979. Moreover, a number of initiatives have been implemented to educate children and adolescents and to increase their awareness on safety and related issues (Akosile 2015; Elixhauser 1990; Dolan et al. 2005). Finally, the reason for the higher mean for the domain of ‘Health’ in the Portuguese sample is not clear, but we can speculate that this may be related to domain compensation (Best et al. 2000), a process which maintains subjective life quality by compensating a fall in satisfaction / happiness in one domain (e.g., Standard of Living) by a rise in another (e.g., Health).

A major implication of the collective findings and the establishment of measurement invariance is the facilitation of a future research study that may seek to offer a comprehensive explanation for the differences in the happiness domain profiles of Australian and Portuguese adolescents which may reveal insights into those factors that promote and hinder normal psychological wellbeing in each respective sample groups. The present study also highlights the importance of testing for measurement invariance in the scales employed in large-scale cross-cultural adolescent SWB studies (e.g., UNICEF 2013). While large-scale cross-cultural studies are important, in terms of their ability to stimulate discussion and the development of policies aimed at supporting positive adolescent SWB, measurement invariance testing is commonly neglected. In order to have great greater confidence in the results of these important cross-cultural studies, it is recommended that researchers assess the equivalence of employed SWB instruments, using stringent multiple-group CFA procedures, prior to the analysis and reporting (e.g., rank ordering) of means for participating countries. For example, considering the large number of countries (29) involved in UNICEF’s (2013) cross-cultural investigations, measurement invariance testing would help the researchers determine whether differences between nations reflect true differences on the SWB as opposed to measurement bias. This has major implications for the accurate identification of countries and cultures in which adolescents are experiencing low SWB and are in need of supportive services.

Finally, the lack of measurement invariance testing is not exclusive to adolescent SWB research, but rather a pervasive issue in the social science research (Gregorich 2006). This study highlights that measurement invariance testing should be extended to other self-report measures (e.g., self-esteem, optimism, resilience) to enable accurate comparisons among diverse cultural groups on important psychological outcome variables.

Study Limitations

A limitation of this study is that both samples were recruited via convenience sampling methods and therefore may not be representative of all the qualities and characteristics of adolescents in each respective population. Thus, it could be argued that further research that explores the psychometric properties of the PWI-SC in more nationally and socio-culturally representative samples is warranted. A second limitation of this study is that only two cultural groups were examined (due to limited availability of viable data sets). To be more confident in the measurement invariance of the PWI-SC, a future study could sample a broader range of adolescence cultural groups. In particular, measurement invariance research into the PWI-SC is required among cultural groups that evidence a self-report response bias. For example, research conducted in collectivist cultures such as Taiwan, Japan, and China, has found that both adult and adolescent respondents are less likely to rate themselves at the extreme ends of a Likert-type scale, when compared to individuals from individualistic cultures (e.g. North Americans; Lee et al. 2002; Stening and Everett 1984). Thus, more research is needed to gain further insight into these processes.

Summary and Conclusions

In order to make accurate assessments of differences in adolescent SWB across cultures, it must be empirically demonstrated that any proposed scale functions equivalently within the cultural groups under investigation. Using samples of Australian and Portuguese adolescents and the PWI-SC as the measure of SWB, the finding of equivalence suggests that the meaning of the SWB construct is the same for adolescents representing both cultural groups. This, in turn, supports the validity of quantitative comparisons between the Australian / English and adapted version of the PWI-SC to Portugal. This study has major implications for future studies that may wish to explore differences in SWB between these two nations; as well as offers researchers interested in conducting cross-cultural research involving adolescents or adults a means of enquiry with which to evaluate to the validity of measures employed.