Whether and how personality characteristics change across the adult life span has been debated for decades (Costa et al., 2018; Roberts & DelVecchio, 2000; Roberts et al., 2006). Evidence has now accumulated that inter-individual differences in the way people act, think, and feel (Roberts, 2009)—mainly conceptualized as the Big Five (neuroticism, extraversion, openness, agreeableness, conscientiousness)—consists of both stable and malleable parts (Anusic & Schimmack, 2016; Wagner et al., 2019). Given the relevance of personality for important life outcomes (Ozer & Benet-Martinez, 2006; Roberts et al., 2007; Soto, 2019), researchers are striving to better understand how between-person differences in personality are changing with age or time. Extant research has distinguished between two types of stability and change: rank-ordering and mean levels. The relative ordering of people appears to be characterized by an inverted U-shape with a high point (though not at perfect stability) reached in midlife (Anusic & Schimmack, 2016; Ferguson, 2010; Milojev & Sibley, 2017; Wagner et al., 2019). At the same time, mean-levels of undesirable personality characteristics are known to decrease as people move from young adulthood to mid-adulthood (e.g., neuroticism), whereas mean levels of productive parts tend to increase (e.g., agreeableness, conscientiousness), thereby suggesting an age-related change towards greater maturity. Later in life, mean levels of several traits begin to decrease (e.g., conscientiousness, openness), others increase again (e.g., neuroticism; Graham et al., 2020). Besides the existing evidence on the very broad Big Five domains, less is known about change patterns on the more diversified facet level. Knowledge about facet-level change trajectories is, however, highly desired as it provides more detailed information on changes in the specific content of personality. For example, when people show increases in how conscientious they are, do people become both more orderly and more achievement-orientated or are trait-level increases primarily driven by increases in just one of the facets of the broader trait construct? Findings from the very few previous longitudinal studies on facet-level stability and change point to comparable rank-order stabilities of facets that are to those observed for the broader traits, at least among college samples (Klimstra et al., 2018) but to differential change mean-level trends of facets (Brandes et al., 2020; Roberts et al., 2006; Terracciano et al., 2005). Most of these studies, however, face methodological challenges that prevent clear interpretation and inference. For instance, the majority of studies used inventories designed for a time-efficient assessment of personality at the trait-level and thereby often assess facets with a single item only or simply do not assess some facets. Furthermore, age differences were always evaluated using artificially built age groups rather than examining age as a continuous variable. Further, because the implementation of the theory-based personality scales in a well-fitting confirmatory measurement model is hard to achieve, only a few of the previous studies of personality facets in adulthood have actually made use of latent variable measurement models that accounted for measurement error. With the current study, we seek to extend the existing knowledge on facet-level personality change by combining item sampling and person sampling procedures (Olaru et al., 2019; Hildebrandt et al., 2016) that allow for sound inferences about the relative ordering of people and mean-level changes in personality facets. In doing so, we use the NEO-PI-R, assessed in the Seattle Longitudinal Study, a large cohort-sequential study where 1,6667 people born between 1902 thru 1976 were followed for over 11 years. To accommodate previous methodological challenges, we will use ant-colony optimization (Leite et al., 2008; Gabriel Olaru et al., 2019) as a pre-analytical step and subsequently apply local structural equation modeling (LSEM; Hildebrandt et al., 2009, 2016) within second-order growth curve models (Collins & Sayer, 2001) to investigate personality stability and change along a continuously modeled age variable. Previous Knowledge on Stability and Change of Personality Facets Personality characteristics are subject to change across the whole life span both in terms of rank-orders and mean-levels (Anusic & Schimmack, 2016; Graham et al., 2020; Roberts & DelVecchio, 2000; Roberts et al., 2006; Wagner et al., 2019). Reasons for stability and change have been traced back to a multitude of different sources that can be roughly divided into genetic/personal and environmental/situational sources and their interplay (for an overview see Wagner et al., 2020). Knowledge on how and why personality change is highly desirable as it offers opportunities for intervention (Bleidorn et al., 2019). At the same time, different facets might be differently sensitive to interventions (Kennair et al., 2020) rendering it important to identify the specific change pattern of personality facets. Because of the paramount role personality characteristics and their changes play for a multitude of important life outcomes such as occupational success, health, or well-being (Brandt et al., 2021; Roberts et al., 2007; Soto, 2019), understanding personality stability and change can help to identify people at risk for detrimental life conditions. Particularly on facet level, differential change can pave the way for a better understanding of antecedents, drivers, and consequences of change, as they provide more detailed behavioral and motivational information about which particular aspect of personality has changed. Facet-level analyses are, however, quite demanding in terms of the longitudinal data. Specifically, such analyses require a psychometrically sound and comprehensive personality inventory. Given that most longitudinal studies on personality stability and change are based on short inventories, previous opportunities to estimate facet-level change are limited. In the following, we summarize existing evidence for personality stability and change on the facet level. Rank-Order Stabilities in Facets Findings from the Big Five trait level showed that the relative ordering of people appears to be characterized by an inverted U-shape becoming increasingly stable until about midlife without ever reaching perfect stability (Anusic & Schimmack, 2016; Ferguson, 2010; Milojev & Sibley, 2017; Roberts & DelVecchio, 2000, Wagner et al., 2019). This findings is interpreted as support for the cumulative continuity of personality development (Roberts & Caspi, 2003; Roberts & Nickel, 2017). On the facet level, no existing longitudinal study has investigated rank-order stabilities across larger age ranges of the adult life span. However, two studies from childhood and adolescence, one study using college students, and two studies in early adulthood might allow to specify first hypotheses: The two studies focusing on rank-order stabilities in children and adolescents used child-specific personality inventories that were rated by the children’s mothers (Brandes et al., 2020; de Haan et al., 2017). Looking at 1-year stability from age 9 to age 13, Brandes and colleagues (2020) found different amounts of stabilities for facets of extraversion with sociability (r = .72) being more stable than positive emotions (r = .62) and being considerate (r = .61). Looking across a time interval of 1.5 years in young children (2-4.5 years) and adolescents (6-17 years), De Haan et al. (2017) found comparable rank-order stabilities for facets and traits (r = >.50 -.87) with partly lower stabilities in irritability and egocentrism (agreeableness). Looking at 3- and 4-year stability in American and Belgian college students, Klimstra and colleagues (2018) also found comparable rank-order stabilities of traits and facets using the NEO-FFI. Roughly the same pattern was observed in young adulthood, with comparable rank-order stabilities of traits and facets (Deventer et al., 2018; Mund & Neyer, 2014). One exception was found for negative affect, with slightly lower stabilities (r = .41) across 15 years than neuroticism (r = .56). Given that there is no study examining rank-order stability of facets beyond young adult samples, this study will add substantially to the current literature. Mean-Level Changes in Facets In terms of mean-level changes, patterns were found to differ across different phases of the adult life span (Roberts et al., 2006). A recent study, synthesizing 16 longitudinal studies, showed particularly strong changes in mean levels of personality traits from young to middle adulthood that roughly points to greater maturity (Graham et al., 2020). Mean-levels of undesirable personality characteristics decrease (e.g., neuroticism) in early adulthood and productive parts tend to increase (e.g., conscientiousness, agreeableness). Later in life, the maturation pattern seems to reverse with mean-level of conscientiousness showing decreases and neuroticism showing increases again (Graham et al., 2020). Whereas mean-level of extraversion showed steady decreases across the life span, openness was found to be relatively stable in middle adulthood but decreased in later adulthood. Findings for agreeableness were more mixed across studies with mean-level increases in some studies but no mean-level change in others. So far, only two longitudinal studies examined mean-level changes in facets covering large age ranges across the adult lifespan. Using a facet-sensitive inventory—the NEO PI-R—Terracciano et al., (2005) analyzed mean-level changes across 11 waves from 1989 to 2004 in an age-diverse American adulthood sample, whereas most participants were older than 60 years. They applied multi-level modeling using manifest personality sum scores and found a heterogeneous pattern across the different facets of the Big Five traits. Interestingly, all six facets of neuroticism (anxiety, angry hostility, depression, self-conscientiousness, impulsiveness, vulnerability) showed the same mean-level change patterns as the overall neuroticism trait. For all other traits, mean-levels changes at the facet level in part substantially differed in size and direction from the mean-level change observed for the broader trait. For conscientiousness facets, for example, deliberation showed the strongest mean-level increases and, similar to all other facets (competence, order, dutifulness, achievement striving, self-discipline) declined in late adulthood (after age 60-70 years). For extraversion facets, activity showed most rapid mean-level declines in older ages, whereas excitement seeking declined most strongly in early adulthood. The other facets of extraversion (warmth, gregariousness, assertiveness, positive emotions) showed curvilinear gradients with peaks in mean level around the age of 60. In terms of openness facets, mean-level of openness for values declined relatively evenly across the adult lifespan whereas mean-level of openness to feelings and actions showed accelerated declines only in old age. Mean-levels of the other openness facets (aesthetics, ideas, fantasy) were highly stable. Finally, regarding the facets of agreeableness, both compliance and straightforwardness showed strongest and most consistent mean-level increases across adulthood, whereas trust evinced steady increases only until the age of 60. Mean-levels of the three other agreeableness facets (altruism, modesty, tender-mindedness) were again rather stable. Besides this first detailed longitudinal facet-level study, Soto and John (2012) investigated mean-level changes in facets across five assessment points in a sample of 125 women aged 21 to 61 using 16 facets of the CPI-Big Five. Potentially due to the differences in sample structures and used inventories, results between the two studies were quite inconsistent. Soto and John (2012) found age-related decreases in mean-level only for depression (neuroticism) and age-related increases in mean-level of self-discipline, but no changes in orderliness (conscientiousness). Furthermore, mean-level of gregariousness decreased whereas assertiveness (both extraversion) increased. In terms of agreeableness, compassion and humility increased. Openness and its facets showed no longitudinal trends in mean-level at all in contrast to findings from Terracciano et al. (2005). Two other longitudinal studies with a different study focus also provide information on mean-level changes in personality facets in young adulthood using three repeated personality assessments across a study interval of 15 years (Mund & Neyer, 2014) and 4 years (Deventer et al., 2019). The authors found changes in all traits and facets (except for activity; extraversion) that point to mostly the same direction of change in traits and facets (Deventer et al., 2019; Mund & Neyer, 2014). Only some exceptions were found for positive affect (extraversion), goal striving (conscientiousness), and unconventionality (openness) that evinced opposite mean-level change patterns as compared with the trait domain. Findings from cross-sectional studies also highlight that the size of age-related differences is not necessarily consistent across facets and domains (Jackson et al., 2009; Mõttus & Rozgonjuk, 2019; Soto et al., 2011). More longitudinal research is needed tackling the differential change patterns of facets using comprehensive personality assessments across broad age ranges in a life span sample. Methodological Challenges When Studying Stability and Change of Facets Although first studies hint at differential changes of facets and their related domains, evidence hinges on specific characteristics of the original studies that limit clear interpretations of observed trends. Besides the different age ranges covered across studies which most obviously influence the observed pattern of change, we identified at least five other reasons that challenge the opportunities to detect facet-level changes. First, although there is relative consensus between researchers regarding the higher-order structure of personality characteristics on domain level (mostly five or six factors), less agreement exists regarding the lower-order structure of facets. Researchers proposed different numbers and names of facets (John et al., 2008) and also the fathers of the five-factor model state that “the facets of the NEO Inventories thus constitute one possible and useful way to specify narrow traits within the five domains, but they are clearly not the only way” (Costa & McCrae, 2017, p.12). The different operationalizations of facets is of course reflected in the different inventories used in the previous work that all draw on different facet structures (John et al., 2008). Due to the many ways that facets can be characterized and described, interpretations of facet-level change strongly depend on their operationalization. The only study so far (Terracciano et al., 2005) to have investigated stability and change in personality facets using the NEO-PI-R, calling for replication of findings. Second, as longitudinal studies require substantial investments of time and money, researchers often use the opportunity to obtain repeated measurements of many different constructs and, thus, often need to use shorter personality inventories primarily designed to assess the global domain level and not the facets. So far, it remains unclear how valid conclusions are about facet-level changes that were derived from short domain-level inventories that are not able to capture the whole breadth of facets. More longitudinal research is needed based on inventories designed to assess personality facets. Third, when comparing personality scores across time or age, the items used need to reflect the underlying construct equally well (Borsboom, 2006). It is a well-known phenomenon in personality research, however, that the implementation of the theory-based personality scales in a well-fitting confirmatory measurement model that is able to test these assumptions is difficult to achieve (Brandt et al., 2020; Hopwood & Donnellan, 2010; Olaru et al., 2019; Vassend & Skrondal, 1997). Different strategies have been applied to overcome this problem. For instance, manifest scale scores were used instead of latent measurement models (e.g., Soto & John, 2012; Terracciano et al., 2005) that correct for measurement error. Particularly when studying psychological constructs which can hardly be observed directly, the probabilistic nature of associations between measured items and latent psychological constructs calls for the consideration of measurement error (Borsboom, 2008). Furthermore, strong assumptions of simple structure within the independent cluster model of confirmatory factor approaches were relaxed by allowing for cross-loadings (Asparouhov & Muthén, 2009; Marsh et al., 2010) or by using item parceling (Deventer et al., 2019; Mund & Neyer, 2014) in order to increase internal consistency of indicator variables. New item sampling procedures such as ant colony optimization (ACO; Leite et al., 2008; Olaru et al., 2019) offer opportunities to overcome these problems by selecting items sets that match scale-based predefined criteria, for instance, in terms of model fit and scale-based reliabilities. Fourth, in the application of longitudinal models, measurement invariance is considered a precondition to make sure that differences across age and time can be ascribed to actual personality changes and not to changing psychometric properties of the measured constructs (Guenole & Brown, 2014; Schmitt et al., 2011). Although testing for measurement invariance across distinct categories such as groups or assessment points is straightforward, testing for measurement invariance along a continuous variable such as age is much more complicated. At the same time, continuous age analyses would provide a more precise and realistic view of differences in measurement properties between people of different ages. New modeling procedures have been proposed that allow for a continuous treatment of variables in measurement invariance testing (Hildebrandt et al., 2009, 2016). Fifth, previous research mainly contrasted stability and change in personality by grouping people in artificial age brackets (e.g., Brandes et al., 2020; de Haan et al., 2017). Recently introduced person sampling procedures, however, open up promising avenues to study how differences or changes in age are related to differences or changes in rank-order stabilities and mean-levels (Hildebrandt et al., 2009, 2016; see also Olaru et al., 2019; Wagner et al., 2019).