Age differences in personality traits and social desirability: A multi-rater multi-sample study

Highlights

• Age differences in personality traits are smaller from informants' perspective.

• Discrepancies in self- and informant-ratings are associated with social desirability.

• Socially desirable responding in older age may bias age differences in self-ratings.

• We recommend combining self- and informant-ratings for describing age trends.

Abstract

Little research has examined age differences by using more than one source of information. We compared age differences in Five-Factor Model (FFM) facets and nuances in self-reports and ratings by knowledgeable informants using samples from three countries (Estonia, Germany, and the Czech Republic; N = 5,624). We hypothesized that age differences would be larger in self- than informant-reports, because of greater social desirability in self-descriptions with advancing age. Indeed, we found that age differences were systematically smaller in informant-reports compared to self-reports; this trend was stronger for traits independently rated as socially desirable. As age differences may be best approximated by average trends of self- and informant-reports, we provide meta-analytic age trends for multi-rater composite scores of the FFM facets and nuances.

Introduction

Cross-sectional and longitudinal studies in various countries have shown moderately consistent age differences in personality trait scores across the life span (Costa et al., 2000, Costa and McCrae, 2002, Roberts et al., 2006, Srivastava et al., 2003), reflecting normative gains and/or losses in these traits (Roberts & Mroczek, 2008). Many mean-level age trends are nonlinear with steeper increases or decreases during the first three decades of life and even reverse trends in old age (Donnellan and Lucas, 2008, Kandler et al., 2015, Lucas and Donnellan, 2011, Mõttus et al., 2012). For example, Openness appears to have a curvilinear developmental trend across the lifespan: it first increases during adolescence, remains relatively constant during young and middle adulthood and declines in older age (Costa and McCrae, 2002, Roberts et al., 2006). Generally, the broad Five Factor Model (FFM) domains of Neuroticism, Extraversion, and Openness to Experience tend to decline, whereas Agreeableness and Conscientiousness show an upward trend between adolescence and later adulthood (Caspi et al., 2005, Costa and McCrae, 2002, Wortman et al., 2012). Because most of these normative changes appear to be towards greater social and psychological adjustment, they have been collectively referred to as the “maturity principle” of personality development (Caspi et al., 2005, Hogan and Roberts, 2004). There are, naturally, individual differences in these patterns of change (e.g., Schwaba & Bleidorn, 2018), with some people changing more or less than others (Roberts & Mroczek, 2008).

There are two main explanations for such normative maturation. First, the Five-Factor Theory (FFT) of personality holds personality traits as biologically anchored basic tendencies. The FFT thus attributes age-trends in personality traits to biologically based increases in maturity and emotional functioning (Costa and McCrae, 2002, McCrae and Costa, 2008). That is, the intrinsic maturation refers to the natural progression of personality development that occurs without much regard to cultural and historic context or individuals’ particular circumstances (McCrae et al., 2000; McCrae and Sutin, 2018). This view accounts for the finding that similar mean-level trends can be found in genetically related species, such as orangutans and chimpanzees (Weiss & King, 2015), for instance. However, an alternative explanation stems from the Social Investment Theory (SIT), which states that people become committed to age-graded social roles involving work, family and wider community, and these commitments create reward structures that promote becoming more emotionally stable, conscientious and agreeable, for example (Roberts et al., 2005). As most people go through the transitions into these social roles and associated commitments at roughly similar ages, this can account for the normative mean-level age-trends. In line with the SIT, it has been found that the culture-specific timing of normative transitions appears to be linked with earlier or later personality maturation (Bleidorn et al., 2013), although these findings were not replicated in a more recent study comparing participants from 23 cultures speaking 17 different languages (McCrae et al., 2021). The socioanalytic model of maturity (Hogan & Roberts, 2004) also emphasizes the importance of taking into account the relationship between the individual and his or her society, as maturation has both identity and reputational elements.

Any theorizing on the causes of the mean-level changes in personality traits presumes having robustly established these changes in the first place. Most of the findings have been based on either self-reports or, less commonly, reports by knowledgeable informants (e.g., McCrae et al., 2005) but, individually, each of these data sources provides only a limited amount of valid information. For example, McCrae (2015, 2018) and McCrae and Mõttus (2019) have shown that in many personality scales there is about as much systematic method-specific variance (e.g., variance specific to the self-report method) as there is variance in the trait of interest (variance common to source methods). In principle, then, age-differences observed in any single source of personality information could largely result from method effects (e.g., evaluative biases) that vary with age. If so, how can we know whether and to what extent the observed age differences are really in the traits per se as opposed to reflecting age-related method effects? To disentangle age-associated substantive and method variances, findings based on different methods should be compared.

For example, true developmental or age-related changes in personality ought to be consensually valid: different observers (including the self) should agree on the direction and size of changes (Costa et al., 2019). The use of multiple informants has become one of the most valuable tools in personality research (Allik et al., 2015, Kenny et al., 2006, Mõttus et al., 2020; Vazire, 2006; McCrae & Mõttus, 2019), supported by the finding that personality constructs have sufficient measurement invariance across the rating sources (Mõttus et al., 2020).

But the few studies that have tried to replicate self-report-based age differences in informant ratings have yielded inconsistent results (McCrae et al., 2004, McCrae et al., 2004). For example, McCrae, Terracciano, and colleagues (2005) asked over 11,000 students from 49 countries to complete a personality inventory about either a younger (aged 18–21) or an older person (over age 40) they knew well and found that although the overall pattern of age differences was similar to those in previous, self-report-based studies (e.g., Costa & McCrae, 2002), the effect sizes were notably smaller for Neuroticism and Agreeableness, with the latter not even being significant in most country-samples. Rohrer and colleagues (2018) used a data set of over 10,000 self- and informant-ratings of personality traits, with >200 observations per each year of age spanning from 14 to 29 years. They showed that age trends in informant-reports were consistent with the respective trends in self-reports for Extraversion and Openness, and that there were no self- or informant-reported increases in Agreeableness. Self- and informant-reported Conscientiousness were both positively correlated with age, but the increase in informant-reports seemed to “lag behind” the increase in self-reports. And whereas relatively older individuals reported increasingly higher levels of emotional stability, informant-reports of emotional stability remained mostly stable, meaning that informant-reports did not confirm the expected maturation between the ages of 14 and 29. In contrast, in a 17-year longitudinal study of German adolescents, parents reported a decline in adolescents’ Neuroticism that was not seen in self-reports, and increases in parent-reported Openness and Conscientiousness were significantly larger than as judged by children themselves (Luan et al., 2017). In addition, the few studies that have compared self- and other-rated personality development over time (Oltmanns et al., 2020; Lenhausen et al., 2021; and Watson & Humrichouse, 2006) have found that informant-rated personality trait trajectories indicate less maturation (e.g., decreases in Agreeableness) than self-rated trajectories.

These findings are consistent with the possibility that oft-observed age differences in personality scores may be to some, perhaps even substantial, extent be mixed with (single source) methodological artefacts. But what could possibly bias either self-reports, informant-ratings, or both?

Advancing age is associated with a stronger tendency to present oneself in a positive light, meaning that older people tend to respond to questionnaire items in more socially desirable ways (Dijkstra et al., 2001, Paulhus, 2002, Soubelet and Salthouse, 2011). Socially desirable responding is a general tendency – sometimes also a temporary motivation – to give desirable answers on self-reports to appear positive (Paulhus, 2017). This has been shown for ratings of well-being (Fastame & Penna, 2012) and substance abuse (Welte & Russell, 1993), as well as affect and personality (Soubelet & Salthouse, 2011). Because a) the observed mean-level trends in personality are generally towards greater maturity, which is socially desirable (e.g., Caspi et al., 2005), and b) the mean-level trends in personality appear stronger in self-reports than in informant-ratings (McCrae and Terracciano, 2005, Rohrer et al., 2018), it seems plausible that the findings about mean-level trends in personality are confounded with age-variant socially desirable responding. That is, socially desirable response biases might distort the true age trends in personality traits, leading to an overestimation of age differences in some (i.e., more socially desirable) personality traits. Earlier research combining a self-report personality inventory with a social desirability scale has indeed suggested that some of the age-related differences in self-reported affect and personality traits could be attributed to age-related increases in socially desirable responding (Soubelet & Salthouse, 2011).

How can we find out whether age-trends in personality traits are confounded with socially desirable responding and whether this can account for differences in the age trends between self-ratings and informant-reports? As was done by Soubelet and Salthouse (2011) we could residualize trait scores for an individual-level measure of socially desirable responding and re-estimate the associations of these residualized scores with age. If age-differences in personality traits reflect socially desirable responding, the relationships between personality traits and age should be weaker when adjusted for socially desirable responding. However, it is difficult to separate variance due to socially desirable responding from valid trait variance based on one source of information alone; what looks like “style” may in fact be “substance” (McCrae & Costa, 1983).

For this reason, it may be more useful to identify any systematic differences across different rating perspectives. Specifically, we can compare age trends across traits and rating methods and see whether the differences between rater perspectives (i.e., self- versus informant-ratings of personality traits) track with the traits’ social desirability levels. If age trends in self-reports are biased due to social desirability, then a) more socially desirable traits should show commensurately larger age differences, b) the trend of larger age differences in more socially desirable traits should be stronger in self-reports compared to informant-reports, and c) differences between self-reports and informant-ratings in the age trends should be proportional to traits’ social desirability (i.e., the more desirable the trait, the stronger the age trends in self-reports in comparison to informant-ratings).

But these hypotheses, based on comparing traits in their numeric properties such as social desirability and age differences, are hard to test with a sample of just a few traits such as the domains of the FFM.

Fortunately, there is more to personality traits than the Big Five. For example, Soto and John (2012) argued that understanding age trends in personality requires consideration of narrower traits such as lower-level personality facets because the broad domains do not capture all the meaningful information about how personality traits develop across adulthood. The existing facet-level studies have indeed revealed that these specific traits show distinctive patterns of age differences.

For example, some facets of Neuroticism, such as N5: Impulsiveness, are more strongly related to age, whereas others (e.g., N1: Anxiety) have weaker associations (Costa and McCrae, 2002, McCrae et al., 2004, McCrae et al., 2004) with age. In a large sample with longitudinal data, Terracciano and colleagues (2005) showed that although the age-trajectories of the self-reported NEO PI-R facets generally corresponded to those of the factors they defined, the strengths of the association with age varied substantially across facets. Their research showed the largest age effects for three facets of Extraversion (i.e., E4: Activity, E5: Excitement Seeking, and E6: Positive Emotions) and two facets of Openness (i.e., O1: Fantasy and O3: Feelings), which all were smaller in older individuals. The largest age-related positive slopes were found for three facets of Agreeableness (i.e., A1: Trust, A2: Straightforwardness, and A4: Compliance; Terracciano et al., 2005). Soto and John (2012) demonstrated using the CPI-Big Five that the Industriousness facet of the Conscientiousness – but not the Orderliness facet – showed a significant positive age trend, the Adventurousness facet of Openness to Experience – but not the Idealism and Intellectualism facets – showed a significant negative trend, and the Rumination facet of Neuroticism – but not the Anxiety and Irritability facets – showed a significant negative trend (Soto & John, 2012). That facets of the same domains vary in age differences has also been reported in numerous other studies (e.g., Bleidorn et al., 2009, Jackson et al., 2009, Mõttus et al., 2015, Roberts et al., 2006; Soto et al., 2011).

But even facets may not be the lowest level of the personality trait hierarchy as they may be split into hundreds of yet narrower characteristics, nuances (McCrae, 2015). The unique variance in nuances, after controlling for the facets and FFM domains, tends to demonstrate the essential properties of traits such as stability over years, agreement across assessment methods and heritability (Mõttus et al., 2017; Mõttus et al., 2020). As far as our current ability to measure personality goes, nuances can be operationalized with individual personality test items (McCrae & Mõttus, 2019) although they do not equate items (Condon et al., 2020). It is therefore not entirely surprising that items of the same facets tend to vary in age differences.

For example, Lucas and Donnellan (2009) conducted an item-level analysis using nationally representative Australian data and found that not all personality items (or more specifically, self-ratings of different adjectives) showed patterns similar to those exhibited by their respective domains or traits (Lucas & Donnellan, 2009). Mõttus and colleagues (2015) also reported that different items of the same facets often showed different, sometimes even opposing, relations with age. Mõttus and Rozgonjuk (2021) systematically compared the Big Five domains, facets, and 300 items used to measure them (capturing variance of nuances on top of facets and domains) in terms of how much age-relevant information each level contained. Controlling for model over-fitting by testing and validating models in separate samples, they found that nuances (i.e., items) contained over 40% more age-related variance than facets. Strikingly, residualizing the 300 items for all domains and facets had no impact on how much age-relevant variance the items collectively contained, suggesting that age-differences in the domains and facets could be mostly ascribed to nuances.

These findings suggest that our between-trait differences-based hypotheses that link age differences in self- and informant-ratings of traits with the social desirability levels of these traits should not be tested based on five broad domains alone. Instead, the hypotheses should also be tested at the level of facets. Moreover, harnessing the unique age-relevant information in items and capitalizing on their sheer number, our hypotheses can be tested at the level of nuances, or single items. In fact, there is a long history of using item-level analyses to test hypotheses based on between-trait differences (e.g., Funder & Dobroth, 1987).

The main aims of this study were to explore whether – and to what extent – age-differences in self-reports of personality traits are associated with, and possibly inflated by, socially desirable responding, and to report age differences in the FFM facets and items with reduced biases. Do people indeed become more pleasant, emotionally stable, and down to earth as they age or could such findings, at least partly, be explained by how older individuals choose to respond to certain personality questionnaire items? More specifically, we compared age-related mean differences in self- and informant-rated personality traits and explored the possible association of social desirability (as rated by panels of judges) with these differences. If the results showed larger age differences in self-rated personality traits than in informant-ratings, then some of the differences could be ascribed to measurement artefacts such as socially desirable responding. If age-related changes were found to be more pronounced for socially desirable traits but similar in self- and informant-rated personality traits, then this would reflect substantive age differences (e.g., due to social adaptation or maturation).

Our multi-rater research focused on different levels of the trait hierarchy, as distinctive developmental patterns have been found for traits below the broad personality domains such as the Big Five, facets (McCrae et al., 2000, Terracciano et al., 2005), and even for micro-traits or nuances below facets (Mõttus et al., 2017; Mõttus et al., 2019; Mõttus & Rozgonjuk, 2021). Moreover, we aimed to cross-validate the findings and therefore used samples from three different countries: Estonia, Germany, and the Czech Republic. Finally, we meta-analyzed the findings across the samples. Although a few studies have previously published some parts of the present data concerning age differences in personality traits¹, no previous study has carried out analyses on age differences in personality traits in these samples in connection with social desirability ratings of these traits. Moreover, to the best of our knowledge, age differences in personality traits have previously not been examined with a comparable level of comprehensiveness in any cross-sectional study, using both self- and informant-ratings at the level of facets as well as single items of the NEO Personality Inventories – Revised and 3 (NEO PI-R and NEO PI-3), and meta-analyzing the parameters based on samples from different countries across samples.