Research based on the outlook on wellbeing as the fulfillment or realization of one’s daimon or true nature throughout full engagement and optimal performance in the existential challenges of life (Ryan and Deci, 2001) has focused on what makes people healthier and enables them to adjust psychologically to changes in the environment (e.g., life style habits, such as being physically active and avoiding smoking). This approach goes back to Jahoda (1958), who tried to conceptualize “psychological health” based on developmental psychology theories that described positive mental health, including Jung’s account of individuation (Jung, 1933), Erikson’s psychosocial development (Erikson, 1959), Allport’s formulation of maturity (Allport, 1961), Roger’s depiction of a fully functioning person (Rogers, 1961), and Maslow’s conception of self-actualization (Maslow, 1968). On this basis, Ryff (1989, 1995, 2014) developed a multidimensional model that can be measured with an instrument consisting of 120 items—20 items per each psychological wellbeing dimension: self-acceptance (i.e., the knowledge, acceptance, and awareness of personal limitations), positive relations with others (i.e., the ability to create deep and meaningful connections with others), environmental mastery (i.e., the sense of control over one’s life situation), personal growth (i.e., the tendency to see life as a growing experience and as an opportunity to develop one’s talents and potential), autonomy (i.e., the sense of living in accordance with one’s own convictions and free will), and purpose in life (i.e., the tendency to perceive meaning, purpose, and direction in one’s own life). For administration purposes (e.g., time- and cost-effectiveness), Ryff has developed different versions containing 54 items, 42 items, 39 items, and 18 items.

The 18-item version comprises three of the original 20 items to assess each dimension (Ryff and Keyes, 1995). This shortened version’s subscales have correlation coefficients ranging from 0.70 to 0.89 with their corresponding subscales in the original 120-item version (Ryff and Keyes, 1995). Moreover, the study that first examined the 18-item version using Confirmatory Factor Analysis (CFA), supported the six-factor original multidimensional model of psychological wellbeing in a cohort of elderly Canadians (Clarke et al., 2001). Further studies using samples from different countries, such as Iranian (Khanjani et al., 2014), Portuguese (Fernandes et al., 2010), and Italian (Sirigatti et al., 2009) have also confirmed, using CFA, the goodness of fit of the 18-item version of Ryff’s Psychological Wellbeing Scale (see Table 1 for a compilation of some of these studies). Finally, Lindfors et al. (2006) showed that the internal consistency coefficients of one (of two) Swedish 18-item version were higher than those of the original 120-item version, suggesting adequacy of the shortened version in a sample of white-collar Swedish workers. Hence, most studies suggest that the 18-item version is relatively valid and reliable for the measurement of Ryff’s multidimensional model of psychological wellbeing.

Nevertheless, there is only one study (Lindfors et al., 2006) confirming the psychometric properties of one of the Swedish 18-tem version. The other version, translated by Garcia (2006), has been used across many more studies and is yet to be validated. What is even more, to the best of our knowledge, all studies have exclusively used Classical Test Theory (CTT) to examine the psychometric properties of the 18-item version. Importantly, СTT has some limitations, such as the fact that item difficulty and item discrimination are group dependent, and therefore the results are dependent on the constitution of the sample (e.g., sample size, age, gender distribution, and etcetera). Additionally, scores obtained from CTT methods are completely test-dependent—since test difficulty directly influences the yielded test scores, this can influence the results. For instance, in their multi-group analyses, Sirigatti et al. (2013) showed that the factor structure of the 18-item version was the same across Italian and Belarusian samples. However, in a South African sample, the best solution was a two-factor model, in which all the positively worded items were clustered in the first factor and all the negatively worded items were grouped in the second factor (Henn et al., 2016). Furthermore, in their study, Hsu et al. (2017) showed that the discriminant validity of the 18-item version was questionable since five out of the six factors had considerable cross-loadings. Last but not least, the true-score model upon which much CTT is based on does not reflect participants’ responses to each specific item. As a result, it cannot be predicted how a participant will respond to a specific item (Hambleton and Jones, 1993).

More modern measurement theories, such as Item Response Theory (IRT) were developed to overcome the above-mentioned limitations. Specifically, IRT modeling provides a way of constructing psychological instruments and examining their measurement characteristics, including dimensionality of the instrument and the quality of response categories in Likert-type scales (Linacre, 2002). Thus, using IRT, researchers can improve measurement accuracy and reliability notably (An and Yung, 2014). Additionally, researchers can determine how many items are suitable to measure a specific construct with, and can therefore make significant reductions in assessment time and effort and, at the same time, increase the validity of the scale (e.g., Edelen and Reeve, 2007; Reise and Waller, 2009).

Given the importance of fast and accurate assessment of psychological wellbeing as an indicator of mental health, many studies have used the 18-item version of Ryff’s Psychological Wellbeing Scale. However, the majority of these studies, if not all, have examined the psychometric properties of the scale using CTT (i.e., some types of factor analysis), showing mixed results. Therefore, we argue that there is a further need to study this specific 18-item version of Ryff’s Psychological WellBeing Scale and to replicate the results in different cultures (cf. Cohen, 1990, p. 1311). Thus, we aimed to examine and describe the psychometric properties of the 18-item Swedish version of Ryff’s Psychological WellBeing Scale translated by Garcia (2006); (see Table 2) using both Classical Test Theory (CTT) and Item Response Theory (IRT). This version has been used in many different studies across a wide range of participants (e.g., Garcia and Siddiqui, 2009; Garcia, 2011; Garcia et al., 2012, 2015, 2016a, 2017, 2018; Garcia and Moradi, 2013). We examined, for the first time, its factor structure using Structural Equation Modelling (SEM) and its relationship with subjective wellbeing, sociodemographic factors (e.g., education level, gender, age), lifestyle habits (i.e., smoking, frequency of doing exercise, and exercise intensity), and health issues (i.e., pain and sleeping problems). Moreover, as an addition to the existing literature, we analyzed the properties of the 18 items using Graded Response Model (GRM)—a specific method within IRT for measures that use Likert scales. In this endeavor, we use previously published data (e.g., Garcia et al., 2016c).

The present data was previously published open access (Garcia et al., 2016c). In the original study, after consulting with the Network for Empowerment and Well-Being’s Review Board we arrived at the conclusion that the design of the present study (e.g., all participants’ data were anonymous and will not be used for commercial or other non-scientific purposes) required only informed consent from the participants.

In the original data collection (Garcia et al., 2016c), a total of 768 Swedish participants, including 279 women and 489 men, aged 13–76 years (M = 25.21; SD = 11.34) were selected using volunteer sampling. Participants reported a mean of 4.57 years of education (SD = 4.86) after ground school.

Firstly, as it is recommended for ordinal Likert-type scales, the internal consistency of the subscales of the 18-item version was examined using the equivalent of Cronbach’s alpha coefficient, which is based on polychoric correlation matrix rather than the Pearson correlation matrix (Zumbo et al., 2007; Gadermann et al., 2012). Here, a correlation coefficient of 0.70 or higher was considered as an acceptable level of internal consistency of the items (see Cicchetti, 1994).

Secondly, we applied CFA using Weighted Least Square (WLS) to examine a priori models of the internal structure of the scale—this type of analysis provides less bias and more accurate results to ordinal Likert-type scales (Mindrila, 2010; Li, 2016). We used the following statistical tests and indices to assess the “goodness-of-fit” of the models (acceptable values in parenthesis): the Goodness-of-Fit Index (GFI > 0.95), the Adjusted Goodness-of-Fit Index (AGFI > 0.95), the Non-Normal Fit Index (NNFI > 0.95), the Comparative Fit Index (CFI > 0.95), the Root Mean Square Residual (RMSR < 0.05), The normal chi-square (χ²/df < 5), the Root Mean Square Error of Approximation (RMSEA = 0.10: moderate fit, RMSEA = 0.08: reasonable fit, and RMSEA = 0.05: excellent fit.), and its 90% confidence interval (Bentler and Bonett, 1980; MacCallum et al., 1996; Loehlin, 2003; Miles and Shevlin, 2007). We used multiple indices since they provide different information about the model fit, including the absolute fit, fit adjusting for model parsimony, and fit relative to a null model. Used together, these indices provide a more conservative and reliable evaluation (Maruyama, 1998). Moreover, the multivariate skewness in our data and the fit indices (except for SRMR) of all models were corrected using the Satorra-Bentler scaled chi-square test statistic (Hu et al., 1992; Bentler, 1995).

Third, we examined the fundamental assumptions for conducting graded response model (GRM) and ran the model. The GRM is a flexible and widely used IRT model for analyzing polytomous item responses, such as those found in Likert-type scales (Forero and Maydeu-Olivares, 2009). Unidimensionality of the subscales was measured by estimating internal consistency reliability and conducting one-factor CFA using LISREL 8.8 (Jöreskog and Sörbom, 2006). This statistical analysis allowed us to examine how well the items within each subscale collectively measured the same underlying construct. By conducting the one-factor CFA, we aimed to ensure that the items within each subscale were highly related and contributed to measuring the intended psychological wellbeing construct. The internal consistency analysis provided valuable insights into the degree of coherence and homogeneity among the items within each subscale. Moreover, local independence was evaluated by examining residual correlations among items in the one-factor model. The GRM model assumes that the item parameters (e.g., item difficulty, discrimination) are invariant across different groups or administrations of the test. This assumption ensures that the same underlying trait is measured consistently across different populations or time points (Zumbo, 1999). To measure invariance, we conducted Differential Item Functioning (DIF) across gender group. Finally, the GRM assumes that the probability of endorsing each response category for an item increases monotonically as the level of the latent trait increases (Embretson and Reise, 2013). In the context of wellbeing, this means that individuals with higher wellbeing levels are more likely to choose higher response categories on the items compared to those with lower wellbeing levels. Monotonicity was measured by conducting GRM and checking item fit. GRM was fit to the data and model, and item fit was determined using R 3.5.3. In addition, item scores were used to calibrate item “difficulty” on a logit scale with a midpoint of 0. Difficulty parameters were also inspected to determine whether items supported the comprehensive measurement of the underlying latent construct with minimal gaps and redundancy.

Fourth, the concurrent validity was evaluated by examining the correlations between scores in the different psychological wellbeing dimensions (i.e., self-acceptance, positive relations with others, environmental mastery, personal growth, autonomy, and purpose in life) and subjective wellbeing (i.e., temporal satisfaction with life, positive affect, and negative affect), demographic variables (i.e., gender, age, and education), lifestyle habits (i.e., smoking, frequency of doing exercise, and exercise intensity), and health issues (i.e., pain and sleeping problems).

Fifth, since the items of the 18-item version were ordinal and scored on a 6-point Likert scale, GRM (GRM; Samejima, 1968), conducted in MIRT 1.3 (Chalmers, 2012), was used as the appropriate IRT model (Edelen and Reeve, 2007; Ravens-Sieberer et al., 2008). The items that were reverse scored were coded so that higher scores represent greater psychological wellbeing. In GRM, each item has two types of parameters, including “discrimination” and “threshold.” Discrimination or “slope” parameter specified by “ɑ” shows the extent to which an item is related to, in this case, psychological wellbeing and how well an item discriminates between people with different levels of that latent trait. Items with higher discrimination parameters provide more information about psychological wellbeing. Generally, item discrimination values between 0.01–0.34 are considered “very low,” values between 0.34–0.64 are interpreted as “low,” values between 0.65–1.34 are “moderate,” values between 1.35–1.69 are “high,” and values >1.70 are considered “very high” (Baker, 2001). The other parameter, threshold, specified by “b,” refers to the point at which a respondent with a given level of psychological wellbeing has an equal probability (50:50) of responding above the threshold j (j = 1... m_i, where m_i + 1 = K_i which is the number of response categories for item i). For each item, the number of threshold parameters is equal to the number of item response categories minus one (i.e., K-1).

Finally, DIF was used to examine the systematic errors (bias) due to gender (478 males vs. 276 females). Significant DIF values, evidenced by the logistic regression, indicate that one group of respondents has a higher or lower score on an item compared to another group after adjusting for the overall scores of the respondents. Having established the scale composition based on the results of the psychometric analyses, scale scores were calculated by averaging constituent items such that all scale scores ranged from 1 to 6, with higher scores indicating greater psychological wellbeing. Discriminative validity was also evaluated by testing expected gender- and grade-level differences (Langer et al., 2008). We also calculated between-group effect sizes (ES, d), which we interpreted as significant if they were higher than 0.20 (Cohen, 1988).

Table 3 shows descriptive statistics of the items in the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale. As shown, all items were homogeneous and there was very little missing data, ranging from one missing response (i.e., 0.1% for items 1, 3, 9, 10, 11, 12, and 14) to seven missing responses (i.e., 0.9% for item 4). Given that missing data were at random and accounted for less than 5% (Pallant, 2005), we used listwise deletion without data imputation in order to handle the missing data. Of the 18 items, a ceiling effect was found only for three items (i.e., item 8: “I think it is important to have new experiences that challenge how you think about yourself and the world.,” item 9: “I gave up trying to make big improvements or changes in my life a long time ago.,” and item 14: “I have not experienced many warm and trusting relationships with others.”). Importantly, items 9 and 14 are negatively scored. Moreover, many of the respondents endorsed the last three response options of the six-point Likert scale, which might indicate the (in)frequency of these behaviors in the general Swedish population.

A preliminary analysis showed that for all six subscales, all items met the univariate outlier criteria [−3.00 > Z _x > +3.00]. The decision about keeping or removing outliers was made based on a comparison between the original mean and a 5% trimmed mean. Since the presence or absence of outliers did not influence the main findings of the study, we included the outliers and used robust analyses for estimating relevant statistical parameters (Pallant, 2005; Tabachnick and Fidell, 2007). Testing the assumption of normality revealed a positive but non-substantial skewness in all sub-scales of the 18-item version—A value lower than 0.05 on the Kolmogorov–Smirnov test suggests that the normality assumption is violated.

As indicated in Table 3, the values of univariate skewness ranged from +0.01 for item 10 to −1.56 for item 2. As shown in Table 3, Cronbach’s alpha and ordinal theta coefficients for the total scale were in acceptable range, respectively. Furthermore, these statistics were 0.68 for environmental mastery, 0.76 for self-acceptance, 0.55 for personal growth, 0.52 for positive relations with others, and 0.48 for both autonomy, and 0.8 purpose in life (95% CI).

In addition, the intercorrelation matrix presented in Table 3 shows that almost all the items within five of the six subscales have a positive moderate relationship with each other—with values ranging from 0.14 to 0.63 (based on the corrected item-total correlation for subscale’s items) and from 0.25 to 0.77 (based on the corrected item-total correlation for total scale’s items). However, the items within the purpose in life subscale had non-significant to moderate relationships with each other, with coefficients ranging from −0.08 to 0.37 (based on the corrected item-total correlation for subscale’s items) and from 0.09 to 0.28 (based on the corrected item-total correlation for total scale’s items). Our results also indicated that removing the purpose in life item 7 (“I live life one day at a time and do not really think about the future”), slightly increases the internal consistency of the subscale. Finally, the intraclass correlation coefficients and mean of inter-item correlations were 0.79 (95% CI = 0.77 to 0.81) and 0.17 for the total scale; 0.53 (95% CI = 0.47 to 0.58) and 0.27 for the positive relations with others subscale; 0.69 (95% CI = 0.65 to 0.72) and 0.42 for the environmental mastery subscale; 0.77 (95% CI = 0.74 to 0.80) and 0.53 for the self-acceptance subscale; 0.48 (95% CI = 0.42 to 0.55) and 0.24 for the autonomy subscale; 0.56 (95% CI = 0.50 to 0.61) and 0.30 for the personal growth subscale; and 0.08 (95% CI = −0.035 to 0.19) and 0.03 (95% CI = −0.012 to 0.07) for the purpose in life subscale.

Categorical principal component analysis (Mair and de Leeuw, 2010) was used to assess the dimensionality of the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale. As the loadings plot in Figure 1 shows, the 18-item did not load efficiently on one general psychological wellbeing factor. Moreover, purpose in life item 7 (“I live life one day at a time and do not really think about the future”) pointed in a different direction compared to the other items within the subscale and the scale as a whole. Therefore, it is not surprising that the consistency of the items of this subscale was very low. Moreover, exploratory factor analysis based on polychoric correlations by parallel analysis (Revelle, 2017) indicated that there were five factors and four components. MAP test and Bic index also showed two and six factors, respectively. Hence, the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale cannot be considered as a unidimensional scale with 18 items. The unidimensionality of the subscales was also assessed by one-factor CFAs using LISREL version 8.8 (Jöreskog and Sörbom, 2006). Local independence was also examined within each scale using paired residual correlations among items in the one-factor CFA model (Reeve et al., 2007). Given that the residual correlations were < 0.30 for all other item pairs within scales, local independence, an important premise for conducting IRT, was considered as confirmed.

As mentioned before, the internal consistency of all subscales of the 18-item version, except for purpose in life (Cronbach’s alpha = 0.08), was satisfactory. According to the Root Mean Square Error of Approximation (RMSEA) and the Comparative Fit Index (CFI), the one-factor CFA model fits our data well for five out of the six subscales. Specifically, as with its poor internal consistency, the purpose in life subscale was also a poor fit for the one-factor CFA model [RMSEA = 0.09, CFI = 0.82, see Hu and Bentler, 1999 who recommend an acceptable model fit of RMSEA ≤ 0.06 and CFI ≥ 0.95]. Moreover, despite the removal of items, unidimensionality for the purpose in life subscale was not confirmed.

To further examine the six-factor structure of the 18-item version, as suggested by Ryff (1995), the CFA was conducted using LISREL 8.8 (Jöreskog and Sörbom, 2006), and the goodness of fit was tested for eight models: Model 1 (M₁) examined a one-factor model in which all the 18 items were loaded on a single psychological wellbeing factor; Model 2 (M₂) consisted of a two-factor orthogonal model in which all the positively worded items were clustered in the first factor and all the negatively worded items were grouped in the second factor (Henn et al., 2016); Model 3 (M₃) consisted of a six-factor orthogonal model; Model 4 (M₄) examined a six-factor oblique model resembling the exploratory factor analysis conducted by Ryff (1995); Model 5 (M₅) tested a six-factor first-order model and one-factor second-order that loaded by all six first-order factors; Model 6 (M₆) evaluated a six-factor and four-factor second-order model that loaded by a four-factor model based on the four most highly correlated dimensions: environmental mastery, personal growth, purpose in life, and self-acceptance (Ryff and Singer, 2006; Springer et al., 2006); and finally, model 7 (M₇) assessed a five-factor first-order oblique model after removing the purpose in life subscale. The oblique model was used because we expected the factors to be correlated with each other based on a theoretical standpoint. Moreover, the variance of each factor was set to 1.0 for all models.

As indicated in Table 4, the fit indices for none of the models met most of the specified fit criteria (i.e., RMSEA < 0.05 and χ²/df < 5). Although the six-factor oblique model, as the prior and theory-derived model (M₄), provided a better fit, it was not completely satisfactory (χ²/df = 5.55; CFI = 0.89; NNFI = 0.85; and RMSEA = 0.077; 90% CI = 0.071–0.083). However, after removing the purpose in life scale, a five-factor first-order oblique model (M₇) provided a better fit (χ²/df = 4.35; CFI = 0.92; NNFI = 0.90; and RMSEA = 0.066; 90% CI = 0.059–0.073), but not perfect goodness-of-fit. Path diagram for model 7 is shown in Figure 2. Then, the parsimonious principle (Bollen, 1989) was used to compare the fit indices of the M_1–7 as nested models with those of the M₀ as the baseline/null model. The comparison between the M₄ with the M₂ (Δχ² = 300.14, Δdf = 14, p < 0.001) and M₅ (Δχ² = 69.72, Δdf = 9, p < 0.001) as competitive models, indicated that the six-factor oblique model (M₄) was the optimal model. However, the comparison of the M₄ with the M₇ (Δχ² = 318, Δdf = 40, p < 0.0001) demonstrated that the five-factor first-order oblique model (M₇) is the final parsimonious model. Factor loadings for the five-factor oblique model ranged from 0.46 to 0.90, and each item showed adequate factor loading on the related factor.

After that, in order to examine whether or not the factor structure of the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale was equivalent across gender, multiple-group CFA (Meredith, 1993) was conducted using R package lavaan (Rosseel, 2012). The basic definitions of the five levels of factorial invariance were: configural invariance (the pattern of fixed and free parameters), weak factorial invariance (corresponding factor loadings), strong factorial invariance (corresponding indicator means), strict factorial invariance (corresponding indicator residuals), and finally latent variances and covariance (Byrne et al., 1989; Cheung and Rensvold, 2002). In the case of configural invariance, χ², RMSEA, CFI, NNFI, and other fit indices were used to examine whether or not the combined models have good model fit. Furthermore, for metric, scalar, residual, and latent variances and covariance, the RMSEA values and RMSEA confidence intervals of the hierarchical (nested) models were compared. For example, in the case of comparison of the weak and strong factorial invariance models, if the RMSEA values fall within one another’s confidence intervals, this shows strong factorial invariance. Then, the changes in the CFI of hierarchical (nested) models were examined. Also, the change in CFI for the weak and strong factorial invariance models was assessed. If the change is less than 0.01, this shows strong factorial invariance (Cheung and Rensvold, 2002). Satorra-Bentler χ² statistics was used to compare all constraint models presented in Table 5 with the staring model (Mo₁), in which no constraint was imposed on the estimation of parameters, except for the mean of latent variables that was constrained to zero. Given that changes in the model fit index were minimal, metric invariance was established for the five-factor oblique model across gender. As indicated in Table 5, the hypothesized factor structure of the 18-item version (i.e., five-factor oblique model) fits the data well across gender, showing that the same construct is being measured across gender. That is, it shows equal form (i.e., the number of factors and the pattern of factor-indicator relationships are the same), equal factor loadings, equal thresholds (i.e., when observed scores are regressed on each factor, the thresholds are equal), and equal residual variances across women and men.

Table 6 presents the Pearson correlation coefficients between the total psychological wellbeing score and each subscale’s score with the three subjective wellbeing components, demographics, lifestyle habits, and health. Almost all of the psychological wellbeing subscales were significantly associated with each other, with correlation coefficients ranging from 0.1 to 0.68. However, no significant relationship was found between purpose in life and autonomy (r = −0.02, ns). Moreover, as Table 6 shows, psychological wellbeing was positively associated with both temporal satisfaction with life (r = 0.65, p < 0.01) and positive affect (r = 0.58, p < 0.01), but negatively associated with negative affect (r = −0.52, p < 0.01). With regard to sociodemographic factors, lifestyle habits and health issues, psychological wellbeing was negatively correlated with a high frequency of feeling pain (r = −0.24, p < 0.01) and sleeping problems (r = −0.22, p < 0.01); it was positively associated with age (r = 0.13, p < 0.01), exercise frequency (r = 0.17, p < 0.01), and exercise intensity (r = 0.17, p < 0.01), but negative association with smoking (r = −0.14, p < 0.01), and it was not significantly associated with gender (r = 0.05), or education (r = 0.06).

Table 7 demonstrates the GRM-IRT parameter estimates of the 18 items in the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale. Discrimination or slope parameter (a) is proportional to the steep of the item characteristic curve. Compared to items with less steep slopes, those with steeper slopes are more useful for separating respondents into different levels of psychological wellbeing. Theoretically, a parameter can vary from -∞ to +∞; but typically range from about 0.5 to +2.5 (Edelen and Reeve, 2007). Items with negative, zero, and near-zero slopes show that the probability of answering the item with higher scores decreases for individuals with high levels of psychological wellbeing, which is counterintuitive and therefore indicates that such items should be modified or deleted (Baker, 2001). As indicated in Table 7, there is only one negative slope value, the purpose in life item 7 (“I live life one day at a time and do not really think about the future”; −0.218), which therefore needs to be revised in future studies. Moreover, the self-acceptance item 16 (“In many ways, I feel disappointed about my achievements in life”; 2.843), had the highest discrimination estimation, while the purpose in life item 2 (“I sometimes feel as if I’ve done all there is to do in life”; 0.185) had the lowest. In addition, the environmental mastery item 12 (“In general, I feel I am in charge of the situation in which I live”) and the self-acceptance items 16, 1, and 15 (“In many ways, I feel disappointed about my achievements in life”; “When I look at the story of my life, I am pleased with how things have turned out.”; and “I like most aspects of my personality”) had the highest discrimination estimations. In contrast, the lowest slope parameters belonged to the purpose in life items 7 and 2 (“I live life one day at a time and do not really think about the future”; “I sometimes feel as if I’ve done all there is to do in life”) and to the autonomy item 10 (“I tend to be influenced by people with strong opinions”). See Table 7 for the details.

Table 7 also shows the intercept parameters (d) for the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale. This parameter is the theta value that should have a probability of 0.5 for adjacent categories and represents the threshold required to move from point 1–2 (d1) in the Likert scale, point 2–3 (d2), and so on. The fit indices in Table 7 indicated low fit with the model for the positive relations with others items 18 and 3 (“Maintaining close relationships has been difficult and frustrating for me”; “People would describe me as a giving person, willing to share my time with others”); for the environmental mastery item 11 (“I am quite good at managing the responsibilities of my daily life”); for the self-acceptance item 15 (“I like most aspects of my personality”); for the autonomy item 6 (“I have confidence in my own opinions, even if they are contrary to the general consensus”); and for the purpose in life item 4 (“Some people wander aimlessly through life, but I am not one of them”). However, the model is considered poorly fitted.

Moreover, Table 7 shows DIF examined by the likelihood ratio-based significance test under the IRT framework (IRT-LR; Thissen et al., 1986). First, as recommended, we created a baseline model in which all items have the a and d parameters constrained across groups. Furthermore, the model contained freely estimated latent mean and variance in all but one group, namely the “reference” group. Such a model fixes the metric of the groups so that item parameter estimates do not contain latent distribution characteristics. The results are anchor items that are DIF-free and items that are suspected of DIF. Next, anchor items were used to study items that show DIF based on freeing a and d parameters, respectively. In this study, since all chi-square values were non-significant (p > 0.05), each subscale was analyzed separately. For each subscale, none of its three items showed DIF; thus, we regarded them as anchor items, and there was no need to proceed with the analysis. Also, as indicated in Table 7, the means of items for males and females were close to each other, confirming the results of the DIF analysis.

Furthermore, Figure 3 shows the scale’s and subscales’ test information. The minimum value of test information is 0, which indicates that the test provides no information about the latent trait. In this case, the test cannot distinguish between individuals with different trait levels effectively. The maximum value of test information is unbounded in theory. However, in practice, the maximum value is influenced by the number of items, their psychometric properties, and the range of latent trait levels covered by the test. As the number of items increases, the maximum value of test information also increases, allowing the test to provide more precise estimates of individuals’ trait levels (Baker, 2001; Embretson and Reise, 2013).

As indicated, except for the total psychological wellbeing score and the self-acceptance subscale, the subscales’ information is very low—especially for the purpose in life scale. Therefore, it is necessary to revise these subscales. Furthermore, with regard to the total psychological wellbeing score, only respondents with theta values between −2 to 1 were measured accurately. Additionally, as shown in Figure 4, the following items had low information and need revision: the positive relations with others item 3 (“People would describe me as a giving person, willing to share my time with others”); the environmental mastery items 17 and 11 (“The demands of everyday life often get me down”; “I am quite good at managing the responsibilities of my daily life”); the self-acceptance item 15 (“I like most aspects of my personality”); the autonomy items 10 and 13 (“I tend to be influenced by people with strong opinions”; “I judge myself by what I think is important, not by what others think”); the personal growth items 5 and 9 (“For me, life has been a continuous process of learning, changing, and growth”; “I gave up trying to make big improvements or changes in my life a long time ago”); and the purpose in life subscale items 4 and 7 (“Some people wander aimlessly through life, but I am not one of them”; “I live life one day at a time and do not really think about the future”). Finally, the expected item score for the subscales of the 18-item Swedish version of Ryff’s Psychological Wellbeing Scale is presented in Figure 5. As indicated, the purpose in life item 7 (“I live life one day at a time and do not really think about the future”) had a low relationship with the latent trait. Therefore, this item must be revised and either modified or deleted.

The present study had several limitations. Firstly, the sample was obtained using volunteer sampling method and it is therefore not a representative sample of the Swedish population. Secondly, we used solely self-report instruments; therefore, the association between the variables might have been distorted by shared method variance. Further research needs to be conducted to assess its psychometric properties in various populations. Future research should, for example, examine the cross-cultural replicability of the scale’s factor structure through assessing samples from the indigenous populations in various countries and indigenous-language translations of the scale. Also, future longitudinal research is needed to test reciprocal and bidirectional associations between psychological wellbeing and other variables, such as personality. Future research should also include multi-informant assessments to provide a more conservative and more accurate test of the psychometric features of the 18-item version. Finally, given the contradictory results regarding gender differences in several studies, more research is needed to assess the 18-item version’s dimensionality among men and women.