Stability and change of personality across the life course: the impact of age and major life events on mean-level and rank-order stability of the Big Five

Published in Journal of Personality and Social Psychology 101(4): 862–882 (2011). The published version was not read. Read as SOEPpapers 377 (DIW Berlin, May 2011): the in-press manuscript, marked "not the copy of record", © APA. Universities of Münster and Mainz. Provenance: papers/carry_on/specht2011_personality_life_events.provenance.json.

What was read

All 2,786 lines of pdftotext -layout output (70 pages): series front matter, abstract, introduction, method, results, discussion, footnotes, references, Tables 1–8 and the captions of Figures 1–7. Figures are images and were not inspected.

Question

Does adult personality change with age and with specific major life events?

• Mean levels and rank order, across all of adulthood.
• Does personality predict which events happen (selection)?
• Do events change personality (socialization)?
• Does grouping events by valence hide or distort effects?

Method

• Data. SOEP 2005 and 2009: N = 14,718 people aged 16–82 in 8,443 households (mean age 47.2).

• Personality. BFI-S, 15 items (3 per trait, 1–7 scale), with α .50–.65 and 6-week retest > .75. Modelled as latent factors with strict measurement invariance.

• Events. Twelve objective events reported 2006–2009, after the first personality measure:

  Event	N
  Marriage	664
  Moved in with partner	675
  Divorce	229
  Separation	690
  Death of spouse	228
  Left parental home	302
  Child left home	1,256
  Birth of child	993
  Death of parent	998
  Unemployment	860
  Retirement	693
  First job	456

• Models.

  • Latent change models: intercept = selection, slope = change against same-age non-experiencers.
  • Latent moderated regressions for rank-order stability.
  • Controls: sex, age, age² and age³. Household clustering was corrected.
  • Event effects were interpreted at p < .01, with binomial tests for the number of hits among 60 tests.

Results

• Over four years (Table 2):

  Trait	Mean change (d)	Rank-order stability
  Emotional stability	+.10	.73
  Extraversion	−.13	.74
  Openness	−.17	.72
  Agreeableness	−.16	.68
  Conscientiousness	−.10	.64

• Age.

  • Rank-order stability follows an inverted U for four traits, peaking around 40–60. Conscientiousness grows more stable throughout adulthood.
  • Change continues after 30 and is often stronger in old age, against the "hard plaster" hypothesis.
  • Even after sex and age, people differ widely in change: slope variance .33–.54 in first-wave SD units.

• Selection (3 of 60 tests at p < .01; binomial p = .02). Before the event:

  • more extraverted people were more likely to move in with a partner (.179);
  • more agreeable people were more likely to become unemployed (.120);
  • less conscientious people were more likely to start a first job (−.185);
  • less emotionally stable women were more likely to marry.

• Socialization (7 of 60 at p < .01; binomial p < .001), in SD units against the reference group:

  Event	Trait	Effect
  Marriage	Extraversion	−.126
  Marriage	Openness	−.168
  Separation	Agreeableness	+.143
  Divorce	Conscientiousness	+.249
  Birth of child	Conscientiousness	−.130
  Retirement	Conscientiousness	−.169
  First job	Conscientiousness	+.194

  • Sex-specific effects:
    • leaving the parental home: women gained emotional stability;
    • separation: men gained openness;
    • death of a spouse: women's conscientiousness fell and men's rose.
  • Unemployment and death of a parent changed no trait.

• Grouping events by valence misleads. Positive and negative clusters showed almost none of these effects. The cluster-level selection effects (emotional stability protects against negative events) held for no single negative event.

• Rank-order stability changed with only 2 events (not more than chance, p = .12):

  • death of a spouse lowered agreeableness stability (−.247);
  • a child leaving home raised openness stability (.093).

Limits

• Two waves, 4 years apart; no return-to-baseline could be seen.
• Short, low-reliability scales (α .50–.65), handled with latent models.
• Many tests (effects read at p < .01, with binomial checks). Attrition was small but selective (continuers were older, more often female and more conscientious).
• Event effects compare against same-age non-experiencers and cannot separate an event from what led to it.
• Inconsistencies found: none that affect results.
  • The abstract puts the stability peak "between the ages of 40 and 60"; the results say "in about the 40s and 60s". The discussion gives trait-specific peaks from 40–50 to 50–60.
• Checked:
  • 5 traits × 12 events = the 60 tests quoted.
  • The seven socialization and three selection effects listed in the text all appear in Table 4 with p < .01.
  • The simple-slope description of the death-of-spouse × sex effect is consistent with the interaction (.546, p = .002).

What it means for Kurisutina

• Q1: traits barely move with events, and the moves are event-specific. Over four years, only 7 of 60 event–trait combinations moved, by about 0.1–0.25 SD, and several were not the intuitive direction:

  • marriage: less extraverted and less open;
  • divorce: more conscientious;
  • job loss: no change.

  A replica should not "grow" or "break" in its personality after a big event. If its traits shift noticeably after a narrated job loss, that is the base model's story of what job loss does, not what people do (inferred).

• What varies is the person. After age and sex, people still differ markedly in how their traits changed. Events explain little of that. This agrees with Beck & Jackson, Kiley & Vaisey and Salganik: much individual change comes from things not in the record.

• Changeability depends on age. Rank-order stability peaks in midlife and falls later, except conscientiousness. A replica of a 70-year-old should be more changeable than one of a 50-year-old. A fixed global change rate is wrong.

• Don't pool events by valence. Pooled positive and negative events hide the effects. This supports the pilot's per-event strata and a per-event analysis in LISS T1.

• Selection is real. People who later lose their job differed in agreeableness beforehand.

  • In the pilot, event strata are selected groups; the "other" donor from the same stratum shares that selection, which is why the design uses it.
  • In LISS, pre-event levels must be controlled or matched, as the T1 plan does.

• Scale of the pilot's targets (inferred). The GSS pilot's items are attitudes and well-being, not traits. Traits moving this little over four years is consistent with Kiley & Vaisey's rare persistent attitude change. Event effects on the pilot's items should mostly show in satfin, finalter and happy, where Kiley's φ is highest.

Cross-references

• summaries/carry_on/kiley2020_personal_culture.md: attitudes on the GSS panels change little and persistently for few.
• summaries/carry_on/beck2018_idiographic_networks.md: consistency as a person-level trait.
• summaries/carry_on/kinzinger2026_soep_twins.md: SOEP-based digital twins.
• docs/research/liss_q1_design.md: the LISS T1 design.