Studying changes in life circumstances and personality: it's about time

European Journal of Personality 28: 256–266 (2014), DOI 10.1002/per.1951. A position paper (no new data), from a DFG network on adult personality development. Read as a PDF posted on Ulrich Orth's site; publisher layout, 11 pages. Provenance: papers/carry_on/luhmann2014_its_about_time.provenance.json.

What was read

All 783 lines of pdftotext -layout output: abstract, every section, footnotes 1–3, acknowledgements and all references. The two-column layout interleaves in places; every line was read. Figures 1 and 2 are drawings of artificial data; their captions were read, the drawings were not inspected.

Question

Theories say life events change personality, but little is known about when and at what rate. What does taking time seriously demand of theories and study designs?

Definitions

• Life event: a time-discrete transition that brings a major change in status (e.g. marital status) or social role. Distinct from short-term fluctuations (no status change) and from slow developmental transitions (not time-discrete).
• Because they are clearly timed, such events are "less susceptible to retrospective memory distortions" and easy to date.
• "Personality" is taken broadly: core traits (Big Five) and surface characteristics such as subjective well-being (SWB). Most examples come from SWB research.

Criticism of past work

• Most studies had two measurement occasions, with the event somewhere between them, at different times for different people. The shape of a trajectory cannot be seen from two points; studies with more occasions rarely have more than four.
• Lags range from months to a decade. Studies under a year are rare, partly because panels measure annually or less and partly from the belief that personality changes slowly. Roberts et al. (2006) excluded retest intervals under a year from their meta-analysis. Personality can change within months after a major transition (Bleidorn 2012) or an intervention, so event effects may have been underestimated or missed.

The six notions

1. People differ before the event (selection). Events are not random. High extraversion goes with more desirable events, high neuroticism with more undesirable ones; divorce with high neuroticism and low agreeableness and conscientiousness; moving in with a partner with extraversion. Higher life satisfaction predicts marriage and parenthood and fewer job losses, separations and moves over the next two years (Luhmann et al. 2013). So cross-sectional comparisons of people with and without an event are uninterpretable; prospective longitudinal designs are needed. Pre-existing differences can also moderate reactions (neuroticism and reactivity; "but see Yap, Anusic, & Lucas, 2012").
2. Change can be non-linear and discontinuous. SWB changes fastest in the first months after an event. Quadratic change needs at least three occasions. Occasions should be dense around the event and sparse later. What a short study sees cannot be extrapolated beyond its span. Theory is "conspicuously silent" on timing.
3. Change can be reversible. SWB returns, on average and after many events, to pre-event levels within a few years. Distal events (six months to five years earlier) predicted less change in neuroticism than recent ones (within six months; Riese et al. 2014). A set-point account would predict reversal. Testing it needs one pre-event and at least two post-event occasions, the first before the reversal is complete (Figure 2, Trajectory B: change that is fully reversed by the post-event measurement is never seen).
4. Change can start before the event. Job loss, retirement, marriage and childbirth can be anticipated; divorce is preceded by separation and marital problems. For divorce, life satisfaction falls for several years before; its lowest point is the year before the divorce, and the divorce year is the first year it rises (Lucas 2005). A study that starts shortly before the event will mistake anticipatory change for a stable pre-existing difference, and may conclude that divorce raises life satisfaction. Whether traits other than SWB show anticipation is unknown; identity-driven change should start before the event, role-demand-driven change after it.
5. The control-group problem. Within-person change around an event can be age or period change. Options: discordant monozygotic twins; matched controls; propensity-score matching, whose quality depends on the covariates chosen. Example: after childbirth, parents' life satisfaction declines, but not more than demographically matched controls (Yap et al. 2012).
6. Beyond single major events.
   • Non-normative events (off-time, e.g. widowed at 25) have stronger SWB effects (Luhmann et al. 2012). For traits the authors expect the opposite: normative events give clear role information, which favours change (Caspi & Moffitt 1993).
   • Non-events: a desired, normative transition that does not happen (no job, unwanted childlessness). Not timed, so not events in the strict sense; effects on personality unstudied.
   • Multiple events. Repeated events: losing a job a second time goes with lower life satisfaction than the first time; a second divorce with higher life satisfaction than the first (Luhmann & Eid 2009). No study has looked at repeated events and the Big Five. Co-occurring events (job loss and divorce): additive or multiplicative, same or different domain, causal or not: all open.
   • Minor events: daily hassles and uplifts; major events may act partly through the minor events they cause.

Conclusions

• The ideal study is prospective, intensive (many occasions before and after) and well-timed (dense when change is expected). Long lags miss shape and reversible change; short lags miss slow or delayed change.
• National panels (SOEP, HILDA) are indispensable for rare events and matched controls, but their personality measures are recent, short and less reliable, and taken every four years.
• Explaining change needs mediators measured at least as often as the outcome and before it. A strict test measures traits expected to change and traits expected to stay stable.
• Mechanisms likely differ by event and trait: Neyer & Asendorpf (2001) found that entering a partnership moderated changes in several traits, and dissolution did not.

Limits

• A perspective paper. All empirical claims are citations; the key ones for us (Lucas 2005 on divorce, Luhmann & Eid 2009 on repeated events, Riese et al. 2014) were not read here, so they are second-hand.
• Most evidence is SWB; the authors say generalisation to traits is open.
• No inconsistencies found. The Figure 1 example is described twice (event group higher at both occasions; both groups decline in parallel) and the two descriptions agree.

What it means for Kurisutina

• LISS T1 needs timing choices fixed before any outcome is loaded (step 3 of the order of work):
  • Baseline. If the pre-event answer is the last wave before the event, it can already contain anticipation (the divorce case: the low point precedes the event). The baseline should be a wave well before the event, or both should be declared with the far one primary. Kettlewell et al. (2020) used 2–3 years before as the reference.
  • Post-event timing. LISS core modules are annual and events are dated by month, so the lag from event to the next measurement varies from 0 to about 12 months, where SWB changes fastest. That lag should enter the population transition model, or be restricted. Two post-event waves are needed to see a reversal.
  • Controls. T1's non-event intervals and its population model conditioned on pre-event answers play the control group's role. Age should be in that model, because normative decline is much of some event effects (Anusic et al. 2014).
  • Repeated events. Luhmann & Eid (2009, second-hand) say the second occurrence of an event can differ in mean from the first (job loss worse, divorce better). A constant shift does not change T1's correlation of first and second reactions, but the population model should know the occurrence number; a T2 replica must too.
• The GSS pilot has the two-occasion design this paper criticises. Waves are two years apart and the event date within the interval is unknown, so a lost-job stratum mixes fresh and two-year-old events, and a reversed reaction is invisible. The pilot's contrasts are unaffected (all arms see the same records), but a null on person-specific change there cannot rule out reactions that fade within two years. Read the results with that in mind; the pilot stays frozen.
• For the replica itself. A person's change after an event is not a step: it can start before, can be fastest at first and can reverse. A replica that "carries on" needs time since an event, not only the event, in whatever it keeps about a change; an update that permanently shifts the slot would model Trajectory C only.

Cross-references

• summaries/carry_on/luhmann2012_adaptation.md: the meta-analysis behind the non-normative and reversibility points.
• summaries/carry_on/yap2012_personality_moderation.md, summaries/carry_on/anusic2014_hilda_personality.md: matched controls, normative decline.
• summaries/carry_on/kettlewell2020_life_events.md: anticipation, 2–3 year reference period, co-occurring events.
• summaries/carry_on/infurna2016_resilience.md: declines before the event, individual trajectories.
• docs/research/liss_q1_design.md: T1 and the order of work.