Introduction
Wisdom is a key aspect of positive mental health and is thought to develop over the course of the adult life span. It is a multifaceted concept that has proved difficult to define and measure (Glück et al., Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013). A main measurement challenge concerns which elements of wisdom can be captured with self-report instruments (Webster, Reference Webster2003, Reference Webster, Sternberg and Glück2019). Wisdom researchers theorize the construct can be indirectly gauged by assessing agreed-upon attributes of the wise, although these measures depend on some consensus of what the basic components of wisdom are (see Bangen et al., Reference Bangen, Meeks and Jeste2013; Grossmann and Kung, Reference Grossmann, Kung, Kitayama and Cohen2019; Meeks and Jeste, Reference Meeks and Jeste2009). With self-report measurement, measures generally fall into one of the two broad categories: “Personal wisdom,” wisdom gained from self-reflection and personal life experiences and “general wisdom,” which is concerned with problem-solving applied to general life experiences (Glück et al., Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013). Most self-report measures measure personal wisdom (Glück et al., Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013).
Given that researchers often want to measure a wide range of psychological constructs, they are frequently hampered by lengthy measures (Thomas et al., Reference Thomas, Bangen, Ardelt and Jeste2017). Shorter measures are easier and faster to administer and can be beneficial for researching vulnerable population groups who could fatigue easily due to medical conditions or groups with shorter attention spans (Thomas et al., Reference Thomas, Bangen, Ardelt and Jeste2017). In addition, longitudinal research typically involves large sets of variables so there is pressure to minimize the number of items per scale.
Although there is a scarcity of brief wisdom measures, some attempts have been made to generate such measures. For example, there is a brief version of the Three-Dimensional Wisdom Scale-12 (3D-WS-12) from Ardelt’s (Reference Ardelt2003) 39-item 3D-WS measure (Thomas et al., Reference Thomas, Bangen, Ardelt and Jeste2017). There is also a brief wisdom screening scale by Glück and colleagues (Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013) reflecting a broad conception of wisdom, drawing items from the 3D-WS, the Adult Self-Transcendence Inventory (ASTI; Levenson, Jennings, Aldwin, & Shiraishi, Reference Levenson, Jennings, Aldwin and Shiraishi2005), and the Self-Assessed Wisdom Scale (SAWS; Webster, Reference Webster2007). Webster (Reference Webster, Sternberg and Glück2019) notes that convergent validity between the 3D-WS and the SAWS is relatively weak, suggesting they may be measuring different facets of wisdom. The SAWS has stronger subscale reliability compared to other measures (Glück et al., Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013) and has a distinct theoretical background (described further below) and a brief English version of the SAWS could be beneficial in stimulating future wisdom research. Recently, Fung, Chow, and Cheung (Reference Fung, Chow and Cheung2020) constructed a brief 9-item Chinese wisdom measure from SAWS items for studying wisdom in older adults, however, their measure did not preserve the five-factor structure of the SAWS. Furthermore, since the scale is unidimensional, its utility might be limited as the measure cannot be used by other researchers to study different facets of the wisdom construct.
Development of the SAWS
Webster (Reference Webster2003) developed the SAWS after an extensive search of the literature, originally developing a 30-item tool which he later expanded into its current 40 items. Webster (Reference Webster2007) defined wisdom as, “the competence in, intention to, and application of critical life experiences to facilitate the optimal development of self and others” (p. 164; italics by the original author). The SAWS included five components that integrate to typify characteristics of those who are wise, that is: Reminiscence/Reflection, Humor, Emotional Regulation, Critical Life Experiences, and Openness.
During the SAWS development, Webster (Reference Webster2003, Reference Webster2007) used heterogeneous samples encompassing the adult life span including diverse ethnicities. The SAWS has reported some high-quality psychometric characteristics with very good reliabilities for some sub-scales. Furthermore, Glück et al. (Reference Glück, König, Naschenweng, Redzanowski, Dorner, Straßer and Wiedermann2013), demonstrated convergent validity between the SAWS and performance measures such as the Berlin Wisdom Paradigm and with other self-report wisdom tools (Taylor et al., Reference Taylor, Bates and Webster2011). Moreover, the SAWS predicts several theoretically relevant wisdom constructs such as ego integrity and meaning in life (Webster, Reference Webster2010; Webster et al., Reference Webster, Weststrate, Ferrari, Munroe and Pierce2018) as well as wisdom nominees (Krafcik, Reference Krafcik2015), and generally shows good to excellent psychometric properties with translated scales when used in cross-cultural research (e.g. Alquraan et al., Reference Alquraan, Alshraideh and Bsharah2010).
Finally, while self-report indicators have often been associated with social desirability, evidence from Taylor et al. (Reference Taylor, Bates and Webster2011), and Thomas et al. (Reference Thomas2019) indicated that the SAWS had a nonsignificant association with social desirability. When Brienza, Kung, Santos, Bobocel, and Grossmann (Reference Brienza, Kung, Santos, Bobocel and Grossmann2018) compared four self-report measures of wisdom, the SAWS displayed the least impression management bias (r = .22).
Thus, the SAWS is an appropriate base for the development of a short measure of wisdom. Smith, McCarthy, and Anderson (Reference Smith, McCarthy and Anderson2000) indicated using an insufficiently validated parent scale often lead to error in the development of its abbreviated form. Therefore, we chose to use the SAWS as the foundation of a short measure of wisdom, and accordingly, the overall objective of this research is to re-examine the SAWS five dimensions for possible construction of a brief measure using a sample that is inclusive of adolescents and adults.
Measurement model issues
The analyses used to develop the five-factor structure of the SAWS have been challenged in the literature. Webster (Reference Webster2007) conducted a principal components analysis (PCA) on the scale items setting the number of components to five a priori based on a search of the literature. The PCA findings supported the five factors with a good alpha reported for the total scale (α = .90). However, reliability estimates produced with PCA are inherently unstable (Flora and Flake, Reference Flora and Flake2017; Tabachnick and Fidell, Reference Tabachnick and Fidell2019).
We note that Webster (Reference Webster2007) replicated the PCA model in a confirmatory factor analysis (CFA) using the same sample. Nonetheless, a factor structure derived from PCA will almost always fit well in a CFA using the same sample as the technique capitalizes on chance factors in the data (Flora and Flake, Reference Flora and Flake2017). Of note is that in PCA, the components are based on shared, unique, and error variances whereas in exploratory factor analysis (EFA), the unique and error variances are estimated and factored out and are not used to create the factors (Tabachnick and Fidell, Reference Tabachnick and Fidell2019). PCA, therefore, benefits from the error variance in the matrix of loadings to estimate components, which EFA eliminates. Also, during the CFA, Webster entered the SAWS subscales as manifest variables rather than as latent indicators.
In CFA, the error terms in the model are specifically estimated. When factors (i.e. latent variables) are entered as manifest variables, the error unique to individual items is aggregated into a single error term, which may obscure problems in measurement related to error variance or that the items may not be functioning well together. Webster (Reference Webster2007, p. 175) noted his CFA methodology used factors as manifest variables rather than as latent indicators was not ideal, reminding us, “The CFA must be interpreted cautiously given the number of parameters to number of participants ratio and some weak and non-supportive fit indices results (e.g. significant χ2 value)”. Indeed, Webster’s CFA results of the SAWS prompted wisdom researchers Greene and Brown (Reference Greene and Brown2009) to suggest, “The instrument requires revision before being used as a measure of wisdom” (p. 292).
Potential age differences
Wisdom is generally thought to increase from adolescence through adulthood and on into older adulthood, although the trajectory is not consistent in the literature or across measures. The SAWS generally shows a weak positive relationship to age when samples only include adolescents to younger adults. This is in accord with studies by Pasupathi, Staudinger, and Baltes (Reference Pasupathi, Staudinger and Baltes2001) who found wisdom-related knowledge and judgment increased in a sample of adolescents and young adults aged 14–37 years, with no further increases after 25 years. However, the SAWS demonstrates a weak curvilinear relationship with the apex at midlife when middle aged and older adults are included (e.g. Webster et al., Reference Webster, Westerhof and Bohlmeijer2014). This latter finding is consistent with that of other contemporary scholars (e.g. Ardelt et al., Reference Ardelt, Pridgen and Nutter-Pridgen2018; Sternberg, Reference Sternberg2005; Thomas et al., Reference Thomas, Bangen, Ardelt and Jeste2017). The indication is that wisdom requires time, difficult life experiences, self-reflection, and learning from one’s experiences for wisdom to actualize (Ardelt et al., Reference Ardelt, Pridgen and Nutter-Pridgen2018; Webster et al., Reference Webster, Westerhof and Bohlmeijer2014). Yet, despite older individuals’ distinct advantage in greater life experiences, older age is often accompanied by increases in dogmatism and mental rigidity, which are known to hinder wisdom development (Meacham, Reference Meacham and Sternberg1990). Furthermore, compared to older persons, midlife individuals, show greater openness to experiences giving them an advantage in wisdom (Webster et al., Reference Webster, Westerhof and Bohlmeijer2014).
Summary
In summary, the primary goal of our research was to reevaluate the five-factor structure of the SAWS in order to examine whether an abbreviated set of items could be empirically derived. We started with a CFA instead of EFA because Webster’s (Reference Webster2003, Reference Webster2007) research suggests that the 40-item SAWS is a valid and reliable scale. Since our purpose was to generate a short form of the SAWS, an initial CFA implies that we are confirming a pre-established structure. However, a convergent EFA conducted with an independent sample would provide additional support for the existing structure and items that could be selected for a short version of the SAWS. In line with these objectives, the following research questions were proposed:
Question 1: Will the five-dimensional factor structure of the full SAWS replicate using a CFA?
Question 2: If the answer to Question 1 is that the five-factor structure does not replicate, will an EFA of the SAWS items in a separate sample produce a five-factor solution?
Question 3: Will a brief form of the SAWS will show acceptable fit and measurement invariance using a CFA and significant mean wisdom differences across age groups from adolescence into the older adult years? Measurement invariance by age is key to an accurate understanding of the trajectory of wisdom across the adult life span.
Method
Study design
This study used a cross-sectional survey design in an Australian setting. Items measuring wisdom were self-assessed. All the measurements were incorporated into an electronic survey format.
Participants
The total sample consisted of 709 participants (M age = 35.67 years; age range = 15–92 years) of whom 22% were male, and 78% female. According to Erikson’s (Reference Erikson1959) psychosocial stages, the sample included: Adolescents (15–18 years, n = 81), young adults (19–40 years, n = 396), middle age (41–65 years, n = 190), and older adults (66–92 years, n = 42). Of the respondents, 80.10% were White Australians, 3.40% Aboriginal and Pacific Islanders, 10.40% reported other ethnicities, while 6.10% did not specify their ethnicity. English as a first language was spoken by 93.80% of the sample. Participants were well educated (M = 14.79 years; range = 2–36 years), with good self-reported health (M = 7.53; range = 1–10). Data were randomly split into n = 356, and n = 353 subsamples for the CFA, and EFA. There were no significant differences in these demographic variables between the two groups (see Appendix A).
Measures
Background variables
Participants were asked their age, gender, educational level, ethnicity, occupation, the country in which they were born, whether they were retired or not, and the main language spoken at home. They were also asked to rate their health on a scale of 1–10 (1 = poor, 10 = excellent). Other measures were administered, but the SAWS was the relevant measure for this study.
Wisdom
Wisdom was assessed with Webster’s (Reference Webster2007) 40-item SAWS, with its five dimensions of: Reminiscence/Reflection, Humor, Emotional Regulation, Experience, and Openness. Each subscale is made up of eight positively phrased statements and responses are rated on a 6-point Likert-type scale (1 = strongly disagree to 6 = strongly agree). Raw scores are summed up to produce a total wisdom score. The SAWS is unrelated to age, or gender (Moberg, Reference Moberg2008; Webster, Reference Webster2007). PCA supported a five-factor structure (Taylor et al., Reference Taylor, Bates and Webster2011; Webster, Reference Webster2007). Webster reported high reliability for the total scale (α = .90), with a 2-week test–retest reliability of (α = .84).
Procedure
Participants were recruited via newspaper advertisements, community groups, and word of mouth at the School of Psychology courses at a demographically diverse Australian university. Data were collected according to the university’s Human Research Ethics Committee protocol for online surveys. Student participants received credits depending on the rules of the class in which they were enrolled. Nonstudent respondents were required to be 18 years or older, but psychology university students could be as young as 15. Participation was voluntary and anonymous. Participants were informed that the survey would take about 30–45 minutes to complete and could be accessed individually at home via the provided link. Informed consent was assumed when participants navigated to the survey after reading the information page which contained the purpose of the study including the rights of participants.
Statistical analyses
IBM SPSS version 26 software program was utilized for data screening and the EFA. Analysis of Moment Structures (AMOS) version 26 software, using the robust Maximum Likelihood (ML) parameter estimation method, was employed for the CFA. This method involves performing (1) Bootstrapping, (2) Bias-Corrected Confidence Intervals, and (3) Bootstrapped ML. We note that in large samples, variables with statistically significant skewness often do not deviate enough from normality to make a meaningful difference in the analysis (Tabachnick and Fidell, Reference Tabachnick and Fidell2019). Also, when variables are not expected to be normally distributed, Hoyle (Reference Hoyle and Hoyle1995) recommended using the ML estimation method when conducting CFAs, as ML has been found to be robust to violations of multivariate normality.
In accord with Byrne (Reference Byrne2016), identification was achieved through the specification of our model to account for a unique solution to the model parameters. Specifically, we fixed the initial loading on each latent variable and allowed the other parameters to be estimated. We then calculated the information available from the observed variance–covariance matrix and ensured this information was sufficient to estimate the unknown parameters, thus ensuring we had an over identified model. Our a priori calculations were then compared to the output, which confirmed the model was sufficiently identified when our CFA was conducted.
Hu and Bentler’s (Reference Hu and Bentler1999) criteria for acceptable model fit were followed. We reported several fit indices. The Comparative Fit Index (CFI), the Goodness of Fit Index (GFI), and the Tucker–Lewis Index (TLI), where values > .95 indicate well-fitting models. The Standardized Root Mean Square Residual (SRMR; < .08) and the Root Mean Square Error of Approximation (RMSEA; < .06 indicates a good fit, .06–.08 satisfactory fit, and >.10 poor models). The χ2 was reported, but it is not considered an adequate estimate of model fit, as it is sensitive to large samples (Kahn, Reference Kahn2006). The χ2/df ratio which considers sample size is reported; with values of < 3 considered acceptable (Byrne, Reference Byrne2016).
Data screening
Data screening revealed one out of range score which was corrected manually. Analysis of missing data indicated Little’s (Reference Little1988) Missing Completely At Random (MCAR) test was not significant (p = .12) supporting that missing data patterns were random. The missing values represented 2.54% of the data. Mean replacement was used to impute missing values, with item-level imputation in the SAWS, and the original sample of 709 was preserved. We created two subsamples of n = 356 for Sample 1 and n = 353 for Sample 2 based on random assignment of cases to samples. Both samples matched the a priori sample size estimate of 300 cases suggested as common “rules of thumb” for CFA and EFA procedures (Comfrey and Lee, Reference Comfrey and Lee1992; Worthington and Whittaker, Reference Worthington and Whittaker2006). The subsamples of n = 356 for Sample 1 and n = 353 for Sample 2 were, therefore, deemed adequate.
Results
CFA findings
A CFA was performed on the first subsample (n = 356) to attempt replication of the five-factor structure. The factors were allowed to covary, but we did not use covariation of the error terms. The model fit indices were χ2 (730) = 2133.83, p < .01, χ2/df = 2.92, CFI = .72, GFI = .74, TLI = .70, SRMR = .08, RMSEA = .07, 90% CI [0.07, 0.08]. Although the RMSEA was marginal, the hypothesized model was a poor fit given the very low values of CFI, GFI, and TLI. To answer Question 2, an EFA was conducted on Sample 2 (n = 353) to see what factor structure emerged empirically. We changed samples to reduce the potential influence of unique error variance within the first sample when extracting the EFA factors. Thus, we used the second sample for an EFA comparison against the CFA results from the first sample.
EFA findings
Different methods can be used to specify the number of factors to retain during factor analysis, although the five-factor model developed by Webster was not a good fit with the full 40 items, we wanted to maintain that conceptual structure with a smaller set of items if possible. Also, the scree plot suggested five factors that converged with Webster’s (Reference Webster2003, Reference Webster2007) theoretical concepts for the SAWS factors.
The EFA was performed utilizing the ML extraction method with Promax rotation (permitting correlated factors) on the 40-item SAWS. In accord with Webster’s factor structure of the SAWS (Webster, Reference Webster2003), we fixed the number of factors to be extracted at five. Correlations between factors all exceed .32 (see Appendix B), which suggests there is sufficient overlap in variance among the factors to support an oblique rather than orthogonal rotation (Tabachnick and Fidell, Reference Tabachnick and Fidell2019). Since the correlations are well below .70, the indication is that each facet is assessing a different aspect of the wisdom construct.
The Kaiser–Meyer–Olkin (KMO) value was .87, and all KMO values for individual items were greater than .74, exceeding the recommended value of .50 (Field, Reference Field2017). Bartlett’s test of sphericity reached statistical significance, supporting the factorability of the correlation matrix. Initial analysis was run to obtain eigenvalues for each factor in the dataset. Howard’s (Reference Howard2016) rule that variables should: (a) “load onto their primary factor above 0.40, (b) load onto alternate factors below 0.30, and (c) demonstrate a difference of 0.20 between their primary and alternative loading” (p. 55) was used. The five factors in combination explained 45.10% of the common variance.
Factor 1 was composed of eight items (3, 8, 13, 18, 23, 28, 33, 38) from the SAWS Reminiscence/Reflection factor. Factor 2 was composed of seven items (4, 9, 14, 24, 29, 34, 39) from the Humor factor. Factor 3 was composed of five items (12, 17, 22, 27, 37) from the Emotional Regulation factor. Factor 4 included five items from the Experience factor (1, 6, 16, 21, 26), while Factor 5 included four items from the Openness factor (5, 15, 25, 35), with one item (36) from the Experience factor. Table 1 shows the factor loadings after rotation.
EFA for 40 SAWS items
Factor loadings >.40 are in boldface.
There is no precise rule for the retention of items within an EFA (Meyers et al., Reference Meyers, Gamst and Guarino2006). In general, items should load as highly as possible with no significant loadings on multiple factors. However, the reliability of items within a factor structure seems to be significantly compromised when loadings are below .30, with items loading around .40 being marginal (Gorsuch, Reference Gorsuch1983; Stevens, Reference Stevens2002). Loadings above .50 are desirable and are considered good for the replicability and utility of scales (Comfrey and Lee, Reference Comfrey and Lee1992; Meyers et al., Reference Meyers, Gamst and Guarino2006). Based on this guideline, we noted that there were at least three items loading above .45 for all extracted factors. We observed that Webster (Reference Webster2003, Reference Webster2007) had an equal number of items for each factor. In addition, DeVellis (Reference DeVellis2017) posited that questionnaires should possess the most parsimonious, that is, the simplest structure.
Accordingly, to retain Webster’s (Reference Webster2007) scale structure of an equal number of items for each subscale, to not privilege any subscale in the total score by having an unequal number of items per subscale and to maximize the internal consistency and parsimony, we selected the three highest loading items on each of the five factors to comprise a 15-item version of the SAWS. Thus, we had an equal number of items per factor, with a loading of at least .50 for each item.
The Cronbach’s alpha for this 15-item five-factor SAWS were: Reminiscence/Reflection (α = .74), Humor (α = .72), Emotional Regulation (α = .85), Experience (α = .73), and Openness (α = .56), and α = .80 for the total SAWS-15 scale. Prior to combining the scores into a single wisdom score, we evaluated a higher order construct of wisdom (see Appendix C). The fit indices for this higher order measurement model were good (χ2 (85) =325.92, p < .001, χ2/df = 3.83, CFI = .92, GFI =.94, TLI = .91, SRMR = .05, RMSEA = .06, 90% CI [0.06, 0.07]).
SAWS-15 five-factor CFA
A CFA was performed on Sample 1 (n = 356) to assess fit for the alternative five-factor 15-item SAWS model derived from Sample 2. The items included in the CFA served as manifest indicators of Reminiscence/Reflection (items 8, 13, 23), Humor (items 4, 14, 24), Emotional Regulation (items 12, 17, 27), Experience (items 1, 6, 26), and Openness (items 5, 15, 35). We used the variance–covariance matrices where the factors covaried, but we did not covary the error terms. Results showed a good fit to the data in this independent sample χ2 (80) = 217.05, p < .001, χ2/df = 2.71, CFI = .91, GFI = .93, TLI = .89, SRMR = .06, RMSEA = .07, 90% CI [0.06, 0.08]. The model gave a good fit for the data and explained 67.76% of the common variance.
Multigroup measurement invariance
The hypothesis that the SAWS-15 is measurement invariant with regard to age groups interpret a measure in a conceptually similar manner; a necessary step before assessing mean differences (Byrne, Reference Byrne2016). The sample was subdivided using Erikson’s (Reference Erikson1959) psychosocial stages, as each stage accomplishes a different developmental task. The groups were: Adolescents (15–18 years, n = 81), young adults (19–40 years, n = 396), middle age (41–65 years, n = 190), and older adults (66–92 years, n = 42). We compared the fit of progressively restrictive models. Employing CFA modeling, we focused our tests of invariance across age groups as recommended by Byrne (Reference Byrne2016) with respect to (1) factor loadings, (2) factor loadings and intercepts, and (3) factor loadings, intercepts, and error variances.
Model 1 tested the overall fit of the baseline unconstrained (configural) model across the four age groups. The fit indices χ2 (320) = 651.33, p < .001, CFI = .90, SRMR = .07, RMSEA = .04, 90% CI [0.03, 0.04] supported configural invariance, indicating suitability for metric invariance testing (Byrne, Reference Byrne2016). Model 2 tested metric invariance by constraining the factor loadings across age groups to be equal. Findings indicated acceptable data model fit χ2 (350) = 681.19, p < .001, CFI = .90, SRMR = .08, RMSEA = .04, 90% CI [0.03, 0.04]. Due to the sensitivity of χ2 to sample size and non-normality, Cheung and Rensvold (Reference Cheung and Rensvold2002) recommend a ΔCFI ≤ .01 between two nested models would support measurement invariance. The nonsignificant Δχ2 (30) = 29.86, p = .473, and the ΔCFI < .01 both supported metric equivalence at path coefficients level. Model 3 tested scalar invariance by constraining item intercepts to be equivalent across age groups. Results indicated acceptable model fit χ2 (395) = 734.39, p < .001, CFI = .89, SRMR = .09, RMSEA = .04, 90% CI [0.03, 0.04]. The Δχ2 (75) = 83.06, p = .245 and the ΔCFI = < .01 between Models 2 and 3 indicated differences in factor variances and covariances are not due to age group-based differences.
Mean SAWS-15 differences
A one-way between-groups analysis of variance (ANOVA) examined the impact of the four age groups on wisdom, measured by SAWS-15. There was a statistically significant difference at the p < .05 level in wisdom scores for the four age groups, F(3, 705) = 11.25, p < .001 with medium effect size, η2 = .05 (Cohen, Reference Cohen1988). Post hoc group comparisons using Hochberg’s GT2 (homogeneity of variance with unequal sample sizes; Field, Reference Field2017) indicated the mean score for adolescents (M = 65.42, SD = 8.98) was significantly lower than all other groups. Young adults (M = 69.12, SD = 9.19) differed significantly from the middle age (M = 72.04, SD = 8.53), but both did not differ significantly from the older adults (M = 71.09, SD = 9.21).
Discussion
This study tested Webster’s (Reference Webster2007) SAWS five-factor structure of Reminiscence/Reflection, Humor, Emotional Regulation, Experience, and Openness. For a multidimensional construct like wisdom, subscales need to exhibit well defined, replicable factor structures that agrees with the underlying theory (Byrne, Reference Byrne2016).
The question of whether the 40-item five-factor SAWS is a reliable measure of wisdom that would replicate in the current research through a CFA was answered in the negative. A possible explanation for the current study’s lack of concordance with Webster’s (Reference Webster2007) CFA findings is the different analytic strategies used. In the current study, the SAWS five-dimensional structure with its 40 items was submitted to a CFA. Webster used the SAWS five factors or latent variables of Reminiscence/Reflection, Humor, Emotional Regulation, Experience, and Openness as manifest variables rather than as latent variables. Although Webster argued that such a methodology was justified to simplify model parameters, he conceded, “This is a suboptimal strategy” (p. 171).
The earlier cross-validation of the PCA findings with a CFA, using the same data (Webster, Reference Webster2007), is also an issue in testing the validity of a factor structure. A factor structure derived from a PCA or EFA will almost always fit very well in a CFA when using the same sample because this procedure capitalizes on any chance factors present in the dataset (Flora and Flake, Reference Flora and Flake2017). Having a large sample, we were able to split the data randomly into two independent samples so that the EFA could be conducted on an independent sample, and then validated with replication in the other sample using CFA. To our knowledge, this is the first report of such replication using SAWS items.
Exploratory factor analysis
Since the SAWS five-factor structure was not supported by the CFA, we used EFA to find where the 40 items of the SAWS were fitting. Results showed that the SAWS EFA did not map onto the theoretical model given 8 of the 40 items (20%) cross-loaded on other factors with loadings ≥ .30, and 11 items (27.50%) did not load on any factor. However, the analysis also found that a subset of items from the SAWS using the original five-factor structure fit the data and was a good fit in the replication sample.
The question of whether the derived brief scale (SAWS-15) would show measurement invariance and significant mean wisdom differences across age groups was answered in the affirmative. The relationship between wisdom and age showed that wisdom was greater in successive age groups from adolescents to the middle aged with older adults between and not different from young and middle-aged adults. Although Moberg (Reference Moberg2008) and Webster (Reference Webster2007) found no relationship between age and wisdom with the 40-item SAWS more recently, other scholars (e.g. Ardelt et al., Reference Ardelt, Pridgen and Nutter-Pridgen2018; Thomas et al., Reference Thomas, Bangen, Ardelt and Jeste2017; Webster et al., Reference Webster, Westerhof and Bohlmeijer2014), reported a curvilinear relationship between wisdom and age with the peak at midlife. Despite most societies associating wisdom with older age (Assmann, Reference Assmann, Featherman, Lerner and Perlmuter1994), the wisdom–age relationship is complex. This could be because growing old does not automatically confer individuals the ability to self-reflect and integrate one’s life experiences in a wisdom fostering manner; qualities which wisdom scholars and laypersons agree are necessary for wisdom to actualize (Glück and Bluck, Reference Glück, Bluck, Ferrari and Weststrate2013).
Limitations
While our sample has the advantage of being large enough to support these analyses, the respondents were mostly female, less than 45 years of age, and the large majority were White Australians. Generalizability is thus potentially limited. Nevertheless, the use of a sample with a wide age range has the potential to help in explicating the wisdom–age relationship. Future research with other populations and with population-based sampling rather than convenience samples would strengthen the case of the usability of this brief version of the SAWS.
Self-report measures can limit validity due to social desirability responses by respondents (Shaughnessy et al., Reference Shaughnessy, Zechmeister and Zechmeister2012). An inclusion of a social desirability measure in future research would clarify whether respondents are giving socially desirable responses rather than accurately reporting their attributes, thus avoiding a common pitfall of self-report measures. This is important considering that Taylor et al. (Reference Taylor, Bates and Webster2011) and Thomas et al. (Reference Thomas2019) reported that the SAWS measure is free of biased responding, while other researchers (e.g. Brienza et al., Reference Brienza, Kung, Santos, Bobocel and Grossmann2018) have reported otherwise.
Finally, the internal consistency of the Openness subscale at α = .56 could be viewed as a limitation, although a coefficient in this range is acceptable during measurement development and when using scales with a low number of items (DeVellis, Reference DeVellis2017). We suggest that because reliability estimates using the alpha coefficient is highly biased against scales with three items or less (Carmines and Zeller, Reference Carmines and Zeller1979), in future research other types of reliability estimates should be used, such as retest reliability (Carmines and Zeller, Reference Carmines and Zeller1979; DeVellis, Reference DeVellis2017).
Conclusion
The present research explored the validation of the five-factor structure of the SAWS and confirmed the five-factor structure for a shorter version of the scale. To the best of our knowledge, this is the first time a brief five-factor structure has been advanced and replicated with an independent sample.
A key finding was that a brief version of the SAWS with a five-factor structure was a good fit to the data and was replicable in an independent sample and was measurement invariant with respect to age, and so can be used to explore the trajectory of wisdom development across adulthood and into older ages. In respect to future research, we suggest that range content and criterion-related validation research be conducted as well as further reliability estimates using a range of methods. While clearly more research on this revised version is needed, this brief version can be seen as a step toward making one of the most commonly used measures easier to use in large-scale research programs and in clinical and educational settings.
Conflict of interest
None.
Description of author roles
T. Leeman: analysis of data and writing.
Bob Knight: study design and writing.
Erich Fein supervised data analyses and contributed to writing.
Sonya Winterbotham directed data collection and data cleaning.
Jeffrey Webster consulted on data analysis and writing.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1041610220004202.
