Does Personality Moderate Reaction and Adaptation to Major Life Events? Evidence from the British Household Panel Survey

Journal of Research in Personality 46(5): 477–488, DOI 10.1016/j.jrp.2012.05.005. Read as the NIH author manuscript (PMC3462459; accepted, not copyedited; © 2012 Elsevier; PMC text-mining permitted, not open access). Michigan State University.

Not to be confused with the follow-up by Anusic, Yap & Lucas (2014), which uses Australian HILDA data and includes affect: J Res Pers 51: 69–77, DOI 10.1016/j.jrp.2014.04.009, PMC4235663. That paper was identified from Crossref and PMC metadata and not read.

Provenance: papers/carry_on/yap2012_personality_moderation.provenance.json.

What was read

• Read: all 819 lines of the text extracted from the PMC XML. That covers the abstract, introduction, method, results, discussion, limitations, conclusion, the four footnotes, the acknowledgements, the full reference list (80 entries), Tables 1–5 and the highlights. Figure 1 is an image; its caption was read.
• Not read: the online supplement, a Word file that holds the model equations and R scripts (footnotes 2–3).
  • The PMC link serves a JavaScript download page to scripted requests.
  • Europe PMC does not serve supplements for this record, which is not open access.

Question

• Do the Big Five traits, measured before the event, predict how much a person's life satisfaction changes with marriage, the birth of a first child, widowhood and unemployment?
• As groundwork: how large are the reaction and the long-term change for these events, once normative (age-related) change is separated out with matched controls?

Method

• Data. British Household Panel Survey (BHPS), waves 1–18 (1991–2008).

  • Annual; a multistage clustered probability sample of households in Great Britain; everyone aged 16 or over.
  • More than 30,000 people took part in at least one wave.

• Outcome. One item, satisfaction with life overall, from 1 (not satisfied at all) to 7 (completely satisfied).

  • Asked from 1996 on, except in 2001. Its reliability in this sample is about .70 (Lucas & Donnellan).
  • An intercept-only multilevel model on all life-satisfaction data (N = 26,641) gives a mean of 5.22, a within-person SD of 0.91 and a between-person SD of 0.94.
  • Effects are expressed in units of the between-person SD.

• Event samples, each with at least one rating before the event and one after:

  Event	n	Selection	Age at event, mean (SD)	Data before / after
  Marriage	1,366	never married at the start, stayed married	29.8 (6.1)	3.9 / 4.6 waves
  First child	1,742	first child during the study	29.3 (6.4)	3.7 / 4.9 waves
  Widowhood	562	married at the start, not remarried; anyone separated or divorced since 1996 excluded	71.1 (11.4)	4.5 / 4.3 years
  Unemployment	1,458	not unemployed at the first wave; rated before, during and after a bout	35.4 (14.2)	3.2 years / 4.3 waves after the bout; bouts averaged 1.2 years

• Controls. People without the event: those who stayed single, stayed childless, stayed married or were never unemployed.

  • They were matched one to one on a propensity score from sex, age, age², log household income and education (R package arm).
  • Table 1 lists 1,331, 1,687, 556 and 1,364 matched pairs.

• Three kinds of mixed model (R, lme4):

  1. Basic nonlinear model, event group only. Parameters:
     • a baseline asymptote;
     • peak change, the event year minus baseline (the reaction);
     • asymptote change, the post-event minus the pre-event asymptote (a lack of adaptation);
     • two rates of change.
  2. With controls, seven parameters.
     • Added: a yearly change common to both groups (linear in years of participation), a group effect (0 = event, 1 = control) and a first-year level.
     • Peak change and the asymptote difference are now relative to where the event group would have been without the event.
     • First year, peak change and asymptote difference have random effects.
     • The matching covariates were left out; footnote 4 says simpler models that included them gave "identical conclusions".
  3. Personality moderation.
     • The Big Five were measured once, in 2005 (wave 15), with a 15-item BFI: three items per trait, rated 1–7. Cronbach's α: N .68, E .54, O .67, A .53, C .51.
     • Only personality measured before the event was used, because events change personality. That leaves 328 (marriage), 389 (first child), 130 (widowhood) and 197 (unemployment) people, with at most three post-event waves.
     • A simpler two-level linear model: a pre-event baseline, an event-year change (a dummy for the event year) and a post-event change (a dummy for the two years after). All three are random.
     • One trait at a time, grand-mean centred, moderates all three terms: five models per event.

Results

Reaction and long-term change, in points on the 1–7 scale. Between-person SDs are given in brackets where the text gives them; * p < .05 (Tables 2 and 3).

• Normative change. Life satisfaction fell by 0.01–0.02 points per year in all four analyses.
  • There were no pre-existing differences before marriage, first child or widowhood.
  • People who stayed employed were 0.20 points (0.21 SD) more satisfied at baseline than those who later became unemployed.
• The controls change the conclusions.
  • Married people end up no more satisfied than before marriage, but more satisfied than they would have been single, because the singles declined.
  • The lasting drop after unemployment disappears against controls: it is age-related decline.
  • Widowhood leaves a lasting deficit beyond normative change.
  • The widowed start high (baseline 5.66); the authors attribute this to older adults in the BHPS being more satisfied.
• Personality and baseline (Table 5; coefficients per trait point):
  • Neuroticism: lower baseline satisfaction in all four samples (−0.16 to −0.26);
  • Agreeableness: higher baseline for marriage, first child and widowhood (0.16–0.19);
  • Conscientiousness: higher for marriage and first child (0.21, 0.20);
  • Extraversion: higher for first child and unemployment (0.13, 0.15).
• Personality and change: 5 of the 40 trait × change coefficients reach p < .05 (our count from Table 5; SE in brackets):
  • marriage: Agreeableness → post-event change −0.14 (0.05). Agreeable people were less satisfied in the years after marriage, the opposite of the prediction.
  • first child: Openness → event-year change +0.11 (0.05); Neuroticism → post-event change +0.12 (0.04).
  • widowhood: Neuroticism → event-year change +0.18 (0.08). People high in Neuroticism dropped less. The authors attribute both Neuroticism effects to a lower baseline ("did not have as far to drop").
  • unemployment: Agreeableness → event-year change +0.18 (0.09): a milder first reaction and no long-run difference.
  • Size (our conversion, using the event-sample SDs of Table 4): 0.12–0.25 points per SD of the trait. The mean changes in the same models are 0.03–0.90 points.
• The predicted effects were absent.
  • Neuroticism did not make negative events worse.
  • Extraversion moderated nothing (estimates −0.08 to +0.01). The authors read this as a true null rather than low power.
  • Conscientiousness did not moderate unemployment, so Boyce et al. (2010, GSOEP) did not replicate.
  • Extraversion did not buffer widowhood, so Pai and Carr (2010) did not replicate.
  • The one partial replication is Openness and first child (Dyrdal & Lucas, GSOEP). It is not exact: here the effect is on the first reaction, there it was on long-term change and in men only.
• The authors' explanations for the failed replications:
  • differences between British and German respondents;
  • differences in method, although the Big Five measure is the same;
  • only three years of data after the 2005 personality assessment, although the GSOEP studies used 3–4 waves too;
  • different covariates: Boyce and Wood controlled for marital and employment status.
• Conclusions. The abstract says the traits "did not have consistent moderating effects". The conclusion goes further: "personality does not moderate the impact of life events on life satisfaction".

Limits

• The authors' limits.
  • Selective attrition: within-person effects may differ among dropouts.
  • The choice of control groups: some variables are not matched, and the conclusions on marriage and unemployment depend on the controls.
  • The nonlinear models cannot carry moderators such as age and gender.
• The moderation test is weak (our assessment).
  • 128–389 people per event, at most three post-event waves, one trait per model.
  • Three-item trait scales (α .51–.68) and a single-item outcome (reliability about .70).
  • For Extraversion, the 95% intervals on change span about ±0.12 to ±0.20 points per trait point in the widowhood and unemployment samples. So "true lack … rather than lack of power" holds at best for marriage and first child.
• Multiple tests. Forty tests on change, with no correction.
  • Five are significant where two are expected by chance. If the tests were independent, P(≥ 5) ≈ .05; they are not independent.
  • The pattern is scattered, but the conclusion's "does not moderate" is stronger than the abstract's "not consistent".
• Narrow outcome and predictors. Life satisfaction only, with no affect measure; the Big Five only, with no other stable characteristic.
• Inconsistencies in the paper.
  • Life-satisfaction years:
    • The series is dated "11 waves … (1996–2007, omitting 2001)".
    • The paper also uses waves up to 2008, and "3 years of available data following wave 15" (2005). That needs 2006–2008.
  • Matching:
    • The text says the propensity match "ensures" similar groups.
    • Table 1 shows event groups 3–4 years younger for marriage, first child and unemployment (25.4 vs 29.5; 25.2 vs 29.6; 32.8 vs 35.6).
    • "Intermediate general" education is about twice as common in the event groups (19 vs 8%, 23 vs 8%, 27 vs 12%).
    • The matched Ns in Table 1 are below the event samples in the text (1,331 vs 1,366; 1,687 vs 1,742; 556 vs 562; 1,364 vs 1,458), with no explanation.
    • Table 1's ages do not match the text's ages at the event. They are probably ages at entry; this is not stated.
  • "Yap et al. (2012)" is missing from the reference list. It is the Swiss Household Panel study that the discussion relies on for widowhood and unemployment.

What it means for Kurisutina

• Question 1: traits in the slot would not predict how this person reacts.
  • Stable traits did not predict, in any consistent way, how life satisfaction moved after the same event.
    • Five effects out of forty, scattered over events.
    • Both Neuroticism effects are put down to a lower starting level.
    • One runs against the prediction (Agreeableness after marriage).
    • Only one partly matches earlier work (Openness and first child).
    • What predicted change was the event itself.
  • On this evidence, trait-like contents do not let a replica react "as this person" rather than as the average person in the same state.
  • The individual part of a reaction would have to be observed: interviews before and after the person changes. That is where Eckstein and Peng also pointed.
  • The null is weak. It rules out large trait moderation, not small effects.
  • The open follow-up is whether a person's own reaction repeats across events. The candidate paper, luhmann2009_repeated_events, has not been fetched.
• The one person-specific influence with support is the starting level.
  • The authors explain the Neuroticism effects this way (not tested in the paper).
  • In the GSS pilot, every condition sees the earlier answers to the target items. The population table P(later | earlier, event) conditions on them too, so this part is already in the baseline the slot must beat (D1).
  • The paper is in line with declared expectations E4 (the rest of the record adds little) and E5 (D2 intervals include 0).
  • A clearly negative D2 on happiness would go beyond anything found here for traits. It should then be checked against the demographic distance between donor and person (X6).
• Sizes to expect per stratum. These are for global life satisfaction in British data; the pilot's happiness and financial satisfaction items were not measured here.
  • widowed: the largest reaction, and the only one clearly lasting against controls (−0.69, then −0.19);
  • lost job: −0.35 to −0.40 while unemployed;
  • found job: after the bout, no difference from controls, so expect a return towards the earlier level;
  • married: +0.31 to +0.48 in the year of the marriage, then back to the pre-marriage level but above single controls;
  • divorced or separated: not studied here (see infurna2016_resilience).
  • GSS waves are two years apart, with the event somewhere in between. The later interview therefore mixes event-year reaction with early adaptation, and the per-stratum results (X3) should be read that way.
• Credit the replica for the event effect only.
  • The conclusions about marriage and unemployment reversed once normative decline was modelled.
  • In the pilot the matching comparison is between the two tables, P(later | earlier) and P(later | earlier, event), together with the no-event pairs.
  • A replica's reaction to an event should be scored against the no-event course of similar people, not against the earlier answer.
• Most year-to-year movement is not reaction.
  • A person's rating varies from year to year about as much as people differ from each other: within-person SD 0.91, between-person SD 0.94.
  • The item's reliability is about .70; in the GSS, happy is 0.60 and satfin 0.73 (hout2016).
  • The person-specific part of a reaction is small against this noise.

Cross-references

• summaries/carry_on/luhmann2012_adaptation.md: meta-analytic average reactions to the same events; Lucas co-authored both papers. Luhmann infers individual differences; this paper tests a named source of them person by person and finds little.
• summaries/carry_on/infurna2016_resilience.md: similar events in the German panel. Individual trajectories are heterogeneous and continuous; there are no predictors.
• summaries/carry_on/eckstein2022_context.md: personal learning parameters do not transfer across tasks. Here, stable traits do not transfer to reactions to events.
• summaries/carry_on/peng2025_funhouse.md: generic psychometrics carry little outcome-specific signal, and twins do not reproduce their person's updating.
• summaries/carry_on/hout2016_gss_reliability.md: reliabilities of the pilot's targets (happy 0.60, satfin 0.73).
• docs/research/gss_pilot_design.md: expectations E4 and E5; extra analyses X3 and X6.