Personality psychology is built on the discovery that humans are not blank slates that can be programmed by reinforcement schedules; the strong, situationistic version of human nature that dominated psychology during the area of behaviorism and was maintained by experimental social psychologists.
Instead humans have personality that is reflected in everyday terms like kind, assertive, fearful, courageous, punctual, spontaneous, sociable, curious, or creative. Personality psychologists developed the Five Factor Model to describe the variation in people’s personalities at an abstract level. This model has been the dominant framework to study personality since the 1980s. Longitudinal and twin studies have shown that these personality traits are partially heritable, not much influenced by parenting, and highly stable in adulthood.
Not everybody is happy with the existence of stable personality traits, especially because some traits are considered to be undesirable. Many people would like to be less prone to anxiety and other negative thoughts and feelings. Others want to be more outgoing and sociable. Companies want their workers to be more achievement motivated and hard working. Teachers and professors would like their students to be more curious. This has led to research programs that make change of personality attainable.
One line of research examines voluntary personality change. Just like loosing weight is possible, changing personality is possible if you just want it and are willing to work on it. The evidence suggests that small changes of personality are possible, but even this evidence is not conclusive and limited to short follow-up periods.
This blog post is about the second line of research into personality change. The basic idea is that behaviors require skills and skills can be learned. Making eye contact with a stranger is just like riding a bike. The first time without training wheels, it can be scary, but then it gets easier. So people who are shy can just learn social skills to become more sociable. Some people are always late, but being on time is a skill that can be learned. Soto et al. (2022) call these social, emotional, and behavioral skills, and introduced the Behavioral, Emotional, and Social Skills Inventory (BESSI) to measure these skills.
The BESSI aims to measure 37 skills. The key difference between the measurement of skills and personality traits is the framing of self-report questions. Personality is typically measured by asking participants about their typical tendencies or behaviors. In contrast, the skills measure asks participants about their level of expertise.
“Its instructions asked participants to rate how well they could perform each behavior, reflecting their current level of expertise, on a scale ranging from 1 = not at all well
(beginner) to 5 = extremely well (expert).” (Soto et al., 2022).
The rest of the question is often similar for skills and personality traits. For example, the Big Five Inventory item “Stays optimistic after experiencing a setback” is nearly identical to the BESSI item “Stay optimistic when things go wrong.”
The high similarity between personality and skill items raises concerns about participants’ willingness or ability to distinguish between these two questions. To demonstrate that they do, we need evidence of discriminant validity. That is, participants’ answers to the two questions should differ in a predictable manner.
In addition, the idea of personality skills raises some theoretical questions. If some people have optimism skills, why do they have a pessimistic personality that lowers their wellbeing. Why are these people not use the skills that they apparently posses to make themselves happier?
Evidence of Construct Validity
I am focusing on wellbeing because I study wellbeing/happiness. Personality skills may have benefits for other outcomes (e.g., better time management skills may help with productivity), but I was intrigued by the idea that people can learn specific skills that increase their wellbeing because other attempts to do so have not been very successful.
Soto et al. (2022) examined the relationship between BESSI scales and wellbeing in a study with 492 high-school students. Wellbeing was measured with Diener’s Satisfaction With Life Scale, a valid measure of subjective wellbeing. The 37 skill scales used to create five broader (domain) scales. Simple correlations were positive for all five skill domains. However, these simple correlations can be inflated by response styles like socially desirable responding. More informative are regression results. The regression results in Table 15 show the strongest unique positive relationship for Emotional Resilience skills, b = .3. Self-management and social engagement showed weaker positive relationships, b ~ .15. Cooperation skills were unrelated and innovation skills were negative related to wellbeing, b = -.23.
Further analyses suggested that two specific skill scales account for most of the variation in life-satisfaction, namely confidence regulation, b = .30, and capacity for optimism, b = .23. Together, these two scales imply that people who are above average on these two skills have a 75% chance of being above average in life-satisfaction. This effect size is stronger than the effect size for extraversion or income. Taken at face value, the results suggest that learning emotional resilience skills could make people happier.
Own Analyses
Study 1
In the new spirit of open science, Soto and colleagues shared their data (https://osf.io/4zgyr/) to allow independent researchers to critically examine the evidence. In the same spirit, I used their data to fit a measurement model to their data. The difference between this analysis and Soto et al.’s published result is that correlations with scale scores assume that scales are perfectly valid measures of the construct that they aim to measure. For example, it is assumed that the BESSI “Capacity for Optimism” scale is a perfect measure of individuals’ skills to maintain an optimistic attitude even during difficult times. Personality psychologists use scale scores even though they are aware that scales are not perfect measures. For example, Soto et al. (2022) note that “This positive manifold [positive correlations among all scales] likely reflects, at least in part, measurement artifacts (e.g., evaluative bias, response style,
use of unipolar scales; Anusic et al., 2009; Goldberg, 1992; Rammstedt et al., 2013). However, it may also partly reflect a substantive individual difference (e.g., in people’s overall levels of
functioning; Musek, 2007). Future research can test these possibilities” (p. 217).
My own analyses take up this request for future research using their own data to demonstrate and control for the influence of desirability bias in skill ratings on correlations between skills and life-satisfaction.
I also included other outcome variables in my analysis. The purpose of including other outcome variables is to explore how skills might influence life-satisfaction. For example, Soto et al. also included measures about relationships with parents and GPA. It is possible that some skills help to raise GPA which in turn might increase students’ life-satisfaction. Other skills might help to have better relationship with parents, which could also increase life-satisfaction of high school students.
In short, I use advanced statistical methods that have been around for 60 years to provide a better understanding of the relationship between personality skills and wellbeing, using Soto et al.’s data to test Soto et al.’s (2022) claim that the BESSI scales are valid measures of personality skills that predict – and possibly cause – variation in important life outcomes like life-satisfaction.
The BESSI has 192 items. It is not possible to fit a measurement model for 192 items with just 492 participants. Fortunately, it is not necessary to use all items to create a measurement model. A minimum of two items per construct is sufficient to create a measurement model. Initial analysis confirmed Soto et al.’s findings that life-satisfaction is mainly related to emotion regulation skills. Thus, these constructs were measured with more than two items to get more stable parameter estimates. The full results are reported on OSF (https://osf.io/5dqzv/). The key finding was that confidence skills were the only direct predictor of life-satisfaction with a strong effect size, b = .40, SE = .06. Additional unique predictors were relationship satisfaction with father, b = .35, SE = .06, relationship satisfaction with mother, b = .21, SE = .06, GPA, b = .16, SE = .06, and relationship with peers, b = 15, SE = .06. The only skill predictor of these life-satisfaction predictors was a negative effect of warmth skills on GPA. This produced a weak indirect relationship with life-satisfaction. Thus, the key finding is that confidence skills are the only reliable predictor of life-satisfaction. Of course, these results are limited to Soto et al.’s high-school student sample, but the other datasets did not include life-satisfaction measures to examine the generalizability of this finding.
Study 2
Study 1 assumed that the desirability factor reflects a rating bias. However, it may also partially reflect some real differences in skills. Study 2 examines this possibility by modeling personality and skill items in a single model. The data are from Soto et al.’s (2022) Study 4 with N = 313 university students. Students completed the BESSI and the BFI-2, a Big Five personality questionnaire developed by Soto and John (2017). The BFI-2 uses three facets for each of the Big Five traits and each facet is measured with four items.
I developed a measurement model for the BFI 2 with a desirability bias and an acquiescence bias factor. In Study 2, I used this model to examine convergent validity between the desirability factors for the BESSI and the BFI-2. In addition, I examine discriminant validity of the BESSI scales by examining the unique variance in BESSI scales that is not explained by desirability bias or the personality traits measured with the BFI-2.
The detailed results of the model and the code to reproduce the results are posed on OSF (https://osf.io/5dqzv/). The key finding is that the desirability factor of the BFI-2 was correlated very highly with the desirability factor of the BESSI scales, r = .84. In addition, the BESSI desirability factor was related to the acquiescence factor of the BFI-2, r = .22.
The following results show the factor loadings of the BESSI scales on the desirability factor, the relationship to the strongest personality predictor from the BFI-2, and the amount of explained and unique variance in the BESSI scales.
| BESSI Scale | DESIRABILITY | PRED | Effect Size | EV | UV |
|---|---|---|---|---|---|
| SM-Time Management | 0.39 | RES | 0.62 | 0.75 | 0.25 |
| SM-Organizational Skills | 0.48 | ORG | 0.80 | 0.86 | 0.14 |
| SM-Capacity for Consistency | 0.48 | RES | 0.63 | 0.63 | 0.37 |
| SM-Task Management | 0.45 | PRO | 0.83 | 0.89 | 0.11 |
| SM-Detail Management | 0.57 | RES | 0.56 | 0.65 | 0.35 |
| SM-Rule Following Skill | 0.45 | RES | 0.96 | 0.74 | 0.26 |
| SM-Responsibility Management | 0.61 | RES | 0.58 | 0.72 | 0.28 |
| SM-Goal Regulation | 0.71 | RES | 0.29 | 0.77 | 0.23 |
| SM-Decision Making Skill | 0.68 | RES | 0.45 | 0.66 | 0.34 |
| SE-Leadership Skill | 0.53 | ASS | 0.76 | 0.87 | 0.13 |
| SE-Persuasive Skill | 0.49 | ASS | 0.39 | 0.68 | 0.32 |
| SE-Expressive Skill | 0.59 | SOC | 0.36 | 0.48 | 0.52 |
| SE-Conversational Skill | 0.56 | SOC | 0.65 | 0.74 | 0.26 |
| SE/SM-Energy Regulation | 0.60 | PRO | 0.50 | 0.73 | 0.27 |
| CO-Perspective Taking Skill | 0.71 | SYM | 0.52 | 0.82 | 0.18 |
| CO-Capacity for Trust | 0.54 | TRU | 0.71 | 0.79 | 0.21 |
| CO-Capacity for Social Warmth | 0.68 | EXT | 0.43 | 0.75 | 0.25 |
| CO-Teamwork Skill | 0.66 | POL | 0.33 | 0.58 | 0.42 |
| CO/SM-Ethical Competence | 0.69 | RES | 0.30 | 0.60 | 0.40 |
| ER-Stress Regulation | 0.52 | ANX | -0.62 | 0.84 | 0.16 |
| ER-Capacity for Optimism | 0.52 | DEP | -0.68 | 0.77 | 0.23 |
| ER-Anger Management | 0.56 | EMO | -0.51 | 0.70 | 0.30 |
| ER-Confidence Regulation | 0.52 | DEP | -0.64 | 0.68 | 0.32 |
| ER/SM-Impulse Regulation | 0.55 | CON | 0.40 | 0.60 | 0.40 |
| IN-Abstract Thinking Skill | 0.73 | INT | 0.53 | 0.82 | 0.18 |
| IN-Creative Skill | 0.59 | INV | 0.70 | 0.84 | 0.16 |
| IN-Artistic Skill | 0.42 | ART | 0.48 | 0.65 | 0.35 |
| IN-Cultural Competence | 0.63 | OPE | 0.22 | 0.45 | 0.55 |
| IN/SM Information Processing Skill | 0.62 | CON | 0.32 | 0.48 | 0.52 |
| XX-Self Reflection Skill | 0.69 | – | – | 0.48 | 0.52 |
| XX-Adaptability | 0.63 | ANX | -0.22 | 0.55 | 0.45 |
| XX-Capacity For Independence | 0.66 | CON | 0.33 | 0.58 | 0.42 |
The most important finding is that the confidence skill scale had a strong loading on the desirability factor, b = .52 and is strongly negatively related to the Depression facet of the BFI-2, r = -.64. Only 32% of the variance in this scale is unique variance that could add to the prediction of life-satisfaction above and beyond the variance explained by a depressive disposition. Previous studies have shown that a depressive personality is a strong predictor of life-satisfaction (Anglim et al., 2020; Røysamb, Nes Czajkowski, & Vassend, 2018; Schimmack, Oishi, Furr, & Funder, 2004). A latent variable analysis of Anglim et al.’s data showed an effect size of b = .6 for the depression facet of the IPIP-300. With effect sizes of r = -.64 between depressiveness and confidence skills and r = .6 for depressiveness and life-satisfaction, depressiveness accounts for most of the correlation between confidence skills and life-satisfaction in Study 1, r = .40. Thus, the existing data suggest that confidence skills do not make a strong unique contribution to life-satisfaction. However, it remains possible that confidence skills have an indirect effect on life-satisfaction under the assumption that confidence skills reduce the disposition to experience depressive affect. However, this is an unproven causal assumption and it is equally possible that people who are prone to depression rate themselves as low on confidence skill items.
In conclusion, Soto et al.’s article provides no evidence for the claim that personality skills measured with the BESSI influence life-satisfaction or that improving these skills would produce an increase in life-satisfaction.
Study 3: Multi-Method Study
The most important form of construct validity examines convergent validity across different methods. In personality psychology, the most common approach to provide this information is to complement self-ratings with informant ratings by knowledgeable others like parents, spouses, or close friends. Soto et al. (2022) did not provide information about convergent validity, but a large project by the OECD (Organization for Economic Cooperation and Development) obtained data on children’s personality skills, using self-ratings, ratings by a caregiver (mostly mothers), and a teacher. These data were used in a JPSP article by Guo, Tang, Marsh et al. (2022) to relate personality skills to life-satisfaction.
The abstract claims that the “inclusion of multi-informant ratings substantially enhanced the ability of social–emotional skills in predicting outcome variables, with parent- and self-rated skills playing important, unique roles” (p. 1079). The results section reports that personality skills explained 70% of the variance in life-satisfaction! This is an unbelievable result because the outcome measure was a single life-satisfaction rating that have at best 70% reliable variance. Thus, the authors are claiming that life-satisfaction is fully determined by personality skills. This is implausible because 40% of the reliable variance in life-satisfaction judgments is heritable and stable over long periods of time, whereas skills are by definition learned behaviors. The estimate is also vastly larger than the effect size estimate based on Soto et al.’s (2022) data.
Another major problem of their analysis was that they used ratings by all three raters as predictor variables. This decision implies that each skill measure measures a unique construct without measurement error. However, the authors’ do not explain how self-reported skills are conceptually different from informant rated skills. Theoretically, a skill is a skill is a skill and does not depend on the observer of a skill. Either I can ride a bike or I cannot ride a bike. Thus, it makes more sense to treat unique variance in ratings by a single rater as systematic measurement error and to use the shared variance among raters as a measure of the actual skill. This use of multi-method data is the most commonly used approach to separate construct variance from method variance. I therefore conducted a proper multi-method analyses of the openly accessible OECD data. The complete results are posted on OSF.
I used a multi-group model to distinguish between younger ((10y) and older (15y) cohorts. The measurement model assumed equal validity for parent ratings, but allowed for different validity of self-ratings, under the assumption that cognitive abilities to make self-ratings increase from age 10 to 15. Method variance was modeled with residual correlations among ratings by the same rater. Correlations showed the strongest simple correlation for the Optimism skill factor followed by the Energy skill factor. I followed up on this model with a regression model. The only statistically significant predictor was the Optimism skill factor. The effect size was smaller for the younger cohort, b = .27, than for the older cohort, b = .44. One possible explanation for this finding is that skills become more important as children become more autonomous. Another explanation could be that life-satisfaction ratings of younger children are less valid. However, even the strong effect size of b = .44 in the older cohort implies that skills explain only 20% of the variance in life-satisfaction, not 70% as claimed in Guo et al.’s article.
For some unknown reason, Guo et al. limited their analysis to the Finish sample. Table 2 reports the results for the Finish and the other samples. The results for the Finish sample produced somewhat stronger effect sizes with b = .42 in the younger cohort and b = .53 in the older cohort. Thus, while skills may play a bigger role in Finland, the authors failed to point out that data from other nations were available and produce weaker effect size estimates.
| Sample | Younger | Older |
|---|---|---|
| Canada (Ottowa) | 0.58 | 0.33 |
| USA (Huston) | 0.40 | 0.41 |
| Columbia (Bogota) | 0.26 | 0.49 |
| Columbia (Manizales) | 0.36 | 0.44 |
| Finland (Helsinki) | 0.43 | 0.53 |
| Russia (Moscow) | 0.32 | 0.48 |
| Turkey (Istanbul) | 0.34 | 0.54 |
| Sout Korea (Daegu) | 0.37 | 0.44 |
| China (Suzhou) | 0.20 | 0.21 |
Importantly, even the strong effect sizes of b > .5 for the younger cohort in Canada and the older cohorts in Finland and Turkey do not provide strong evidence that optimism skills can be learned and increase life-satisfaction. A plausible alternative explanation is that skill measures are confounded with inherited personality traits.
General Discussion
The scientific search for predictors of life-satisfaction is nearly 100-years old (Hartmann, 1936). If these predictors are causes of life-satisfaction, changes in the predictors would result in changes in life-satisfaction. Decades of research have identified some predictors of life-satisfaction that are stable and others that can change. Twin studies suggest that stable predictors like personality traits are partially inherited and difficult to change. Twin studies and longitudinal studies also show that other factors can change and predict changes in wellbeing. For example, marriage and divorce produce changes in life-satisfaction.
The concept of personality – socio-emotional -skills is relatively knew and aims to bridge stable and changing predictors of life-satisfaction. The key distinction between a personality trait and a personality skill is that personality skills are learned behaviors. It is assumed that they are “social–emotional skills are more malleable than cognitive skills through targeted interventions, programs, and policies” (p. 1080). This assumption implies that it is possible to teach children skills that can improve their life-satisfaction. Two studies suggest that the key skills that predict life-satisfaction are related to self-esteem, confidence, and optimism. This finding is consistent with evidence that personality traits related to self-esteem are strong predictors of life-satisfaction. However, the existing evidence makes it impossible to assess whether skill measures are valid measures of learned skills or whether these measures merely reflect differences in personality traits.
Future research needs to validate skill measures and demonstrate that interventions can actually change self-esteem and life-satisfaction. At present, the assumption that happiness is a skill that can be learned lacks empirical support, contrary to the sweeping and invalid claims in prominent publications that suggest skill measures are valid and that skills have a strong influence on life-satisfaction.