In 2000, Costa, Herbst, McCrae and Siegler published the article “PERSONALITY AT MIDLIFE:
STABILITY, INTRINSIC MATURATION, AND RESPONSE TO LIFE EVENTS”. The article reported the biggest study of personality stability and change at that time.
Over 1000 participants (N = 1,779) took the NEO, a measure of the Big Five personality traits, 9 years apart. Participants were 39 to 45 years old at time 1. The main finding was that mean levels of personality hardly changed. If anything all scales, except for agreeableness showed a small decrease. This finding led to the conclusion that personality is largely stable in adulthood.

Six years later, Roberts, Walton, and Viechtbauer reported the results of a meta-analysis of personality change over the life course. The results of this meta-analysis were dramatically different. In particular, conscientiousness showed marked increases throughout adulthood. According to this meta-analysis, conscientiousness would still increase by about half a standard deviation from age 30 to age 75.

Sometimes, meta-analysis are considered superior to original studies because they incorporate all of the available evidence. However, meta-analyses are also problematic because they combine a heterogeneous set of studies. The main limitation of Roberts et al.’s (2006) meta-analysis was the lack of good data. Costa et al.’s (2000) article was by far the largest sample with adult (age > 30) participants. Other studies sometimes had samples of fewer than 100 participants or examined very brief time intervals that leave little time for changes in personality. For example, one study was based on 37 participants with a 2 year retest interval (Weinryb et al., 1992). Thus, the amount of (mean-level) change in personality in adulthood remains an open empirical question that can only be answered with better data.
Fortunately, longitudinal data from large samples are now available to shed new light on personality change in adulthood. A few days ago, I posted results based on three wavers spanning 8 years in the German Socio-Economic panel. The results showed mainly cohort effects and little evidence of personality change with age. The figure below shows the results for conscientiousness. Only the youngest cohort (on the right) shows some increases from 2005 to 2013.

Method
Here I present the results of an analysis of the MIDUS data. To examine age and cohort effects I fitted a measurement model (Schimmack, 2019) to the three waves of the MIDUS. I also divided the sample into three cohorts of 30-40 year olds (1965-75), 40-50 year olds (1955-65) and 50-60 year olds (1945-55) in 1995. The measurement model had metric and scalar invariance for all 9 groups (3 cohorts x 3 waves) and had acceptable fit to the data, CFI = .952, RMSEA = .027, SRMR = .055. The MPLUS syntax can be found on OSF (https://osf.io/23k8v/files/). The sample sizes for the three cohorts were N = 1,625, N = 1,674, and N = 1,279, although not all participants completed all three waves. Results were similar when data were analyzed with listwise deletion. The standardized means of the latent variables were centered so that all group means are deviations from the overall mean.
Results
Conscientiousness
The results for conscientiousness are difficult to interpret. Unlike the SOEP data, conscientiousness scores increase from Wave 1 to Wave 3 in all three cohorts. The effect size is modest for the 18 -year interval but would double for a longer period from age 30 to age 70. Thus, an exclusive focus on change over time would be consistent with Roberts et al.’s findings. However, the figure also shows that there are no cohort differences in conscientiousness. That is, 50-60 year olds in 1995 (cohort 1945-55) did not score higher than 40-50 year olds in 1995, although they are 20 years older. One possible explanation for this finding would be a cohort effect that offsets the age effect, but this cohort effect would imply that younger generations are more conscientious than older generations. The problem with this explanation is that there is no evidence or theory that would suggest such a cohort effect.

The alternative explanation would be period effects. Period effects would change conscientiousness scores of all cohorts in the same direction. However, there are also no theories or data to suggest that conscientiousness has increased from 1995 to 2009.
In conclusion, it remains unclear whether and how much conscientiousness levels increase with age. Although new and better data are available, the data are inconsistent and inconclusive.
Agreeableness
The results for agreeableness are similar to those for conscientiousness. A focus on the longitudinal trends suggests that agreeableness increases with age, which mirrors Roberts et al.’s (2006) meta-analysis. This time, the oldest cohort also shows a pattern that is consistent with an age effect. However, other interpretations are possible. The SOEP data suggested a small cohort effect with younger cohorts being less agreeable. Thus, the differences between cohorts may not be age effects. The effect sizes over an 18-year interval are small, but might add up to the d = .4 effect size from age 30 to 75 suggested by Roberts et al.’s (2006) meta-analysis.

Roberts et al.’s (2006) meta-analysis also suggested that neuroticism decreases with age, while the SOEP data didn’t show an age-trend for neuroticism. The MIDUS data also show little evidence that neuroticism decreases with age. Longitudinal trends were only notable for two cohorts and the effect size of d = .2 over an 18-year period is small.
Neuroticism

Openness
At least the results for openness are consistent with previous findings that openness is fairly stable during adulthood.

Extraversion
This is also the case for extraversion.

Conclusion
The bedrock of science are objective empirical observations that produce a consistent picture of a phenomenon. Obtaining such consistent evidence can be difficult. Studying personality change is difficult for many reasons. Following a large sample of participants over time is hard and costly. Even cross-sectional and longitudinal information in combination is sufficient to disentangle age effects from period effects or cohort effects. It doesn’t help when effect sizes are small. Even a moderate effect size of d = .5 over a period of 10 years, implies only a tiny effect size of d = .05 over a one-year period. Moreover, personality measures have only modest validity and are influenced by systematic measurement error that can produce spurious evidence of personality change.
The study of mean differences also has the problem that many causal factors can explain a time-trend in the data at the mean level, and that mean level changes are most likely the aggregated effects of several causal factors at the individual level (e.g., work experiences or health problems may have opposite effects on conscientiousness). Thus, progress is more likely to be made by focusing on individuals’ trajectories rather than mean levels.
The broader implications of these findings are that there is no evidence that personality changes in substantial ways throughout adulthood. This conclusion is limited to the Big Five, although Costa and McCrae also found little evidence for age effects at the level of more specific personality traits. Of course, 20-year olds behave differently than 40-year olds, or 60-year olds. However, these changes in actual behaviors are more likely the result of changing life-circumstances than changes in personality traits.
2 thoughts on “Personality Change in the MIDUS”