The differential impact of major life events on cognitive and affective wellbeing

SSM – Population Health 10: 100533 (April 2020; online 23 December 2019), DOI 10.1016/j.ssmph.2019.100533, PMC6939089, open access (CC BY-NC-ND 4.0). University of Technology Sydney, University of Sydney and others. Code: github.com/datarichard/shit-happens (not read). Provenance: papers/carry_on/kettlewell2020_life_events.provenance.json.

What was read

All 446 lines of the text derived from the PMC XML: abstract, highlights, introduction, methods, results, discussion, limitations, statements, all 22 references and Table 1. The converter checks that the text keeps every non-whitespace character of the XML.

Most results are only in Figures 1–5, which are images; they were not inspected. The effect sizes, time courses and AUCs by event are therefore not available here beyond what the text states. The supplement (Tables S2, S3, S6; Figures S2–S4; balanced and "uncontaminated" models) was not read.

Question

What is the independent effect of each of many major life events on wellbeing, allowing for other events that happen at the same time? And do events affect cognitive wellbeing (life satisfaction) and affective wellbeing differently, in size, duration and adaptation?

Method

• Data. HILDA (Australia) waves 2002–2016. The panel began in 2001 with 13,969 persons in 7,682 households.

• Outcomes.

  • Cognitive: life satisfaction, 0–10, from the interview.
  • Affective: a summed score of SF-36 item 9 (9a–9i: mental health and vitality items such as "been happy", "felt so down in the dumps…", "full of life", "worn out"), α = 0.894, from the self-completion questionnaire.
  • The text reports their correlation as "r² = 0.23".

• Events. 22 events, each asked as "in the last 12 months" with the quarter it happened (Table 1). The analysis singles out 18, defined in advance as 9 positive and 9 negative. Among them:

  • "widowed" (death of a spouse or a child);
  • separated;
  • money lost;
  • health shock;
  • fired (fired or made redundant);
  • hired (changed employer);
  • promoted;
  • retired;
  • money gained;
  • pregnant;
  • married;
  • childbirth;
  • moved.

  Life events are asked after the wellbeing items.

• Model. Linear fixed effects on within-person change.

  • Dummies for the time since the event: pre36, pre24, pre12 before (pre36, 2–3 years before, is the reference) and post00 to post36 after.
  • Unconditional: year dummies only.
  • Conditional: all other events, age, education and area socio-economic status added.
  • Total impact is the area under the curve (AUC) from −24 to +48 months.
  • Cluster-robust SEs. Figures show 90% intervals for time courses and 95% for AUCs.

Results (as stated in the text)

• Ordering holds with and without conditioning on other events:

  • worst: widowed, separated, money lost;
  • best: married, childbirth, money gained.

  Conditional effects are generally a little closer to zero, because events rarely co-occur. Widowed is unchanged by conditioning.

• Cognitive against affective.

  • Marriage, retirement and childbirth raised life satisfaction with "no overall effect" on affective wellbeing.
  • Separation hit life satisfaction harder than affect.
  • Moving lowered life satisfaction with no affective response.
  • Negative events otherwise hit both.
  • The average trends of the two outcomes diverge over 2001–2016: affect declined recently, life satisfaction did not.

• Little independent effect: promotion, being fired, and a friend's death, once other events are accounted for (abstract). Pregnancy's positive effect on life satisfaction is "all but reversed" once childbirth is included.

• Anticipation. Effects often start before the event (−24 or −12 months). For moving the effect is entirely anticipatory. Reconciliation shows a negative lead-up.

• Adaptation. "Affective hedonic adaptation to all positive events occurred by two years". Monetary gains and retirement kept benefiting life satisfaction. The effect of a spouse's death "seemed to diminish by 2 years".

Limits

• Averages only. Fixed effects give within-person average effects among people who had the event. The authors note the estimates "cannot be used to predict whether an event … will increase or decrease wellbeing in any particular case". Nothing here addresses individual differences in reaction.
• Measurement.
  • The affective measure is an SF-36 health scale, not experience sampling.
  • Unbalanced panel.
  • Years in which the event's quarter is missing are dropped (4.5–31.7% of cases).
  • Censoring at the ends of the window.
  • Sensitivity analyses are in the supplement only.
• Inconsistencies found:
  • The abstract says separation, injury or illness and monetary loss had negative impacts that "did not display hedonic adaptation". The discussion says both wellbeing components "show hedonic adaptation" and that the death of a spouse's impact diminishes by two years.
  • The discussion says health shocks, widowhood and separation "tended not to have as much negative impact on both" as financial loss. The results list widowed and separated among the three worst.
  • "18 major life events" (abstract, title of the analysis) against the "22 life events" of Table 1. The four extra are covariates only, or excluded from plots.
  • "Widowed" includes the death of a child.
  • Figure titles read "HLDA".

What it means for Kurisutina

• Question 1: no evidence on individual differences. It is a population-average study. Its use is on which events matter, and through which channel.
• Job loss may act through money.
  • "Fired" had little independent effect once concurrent events were in the model; a major financial loss had the largest total impact.
  • In the GSS feasibility check, lost job carried most information for financial satisfaction and change in finances, and less for happiness.
  • Together these suggest the pilot's lost-job stratum works through the financial channel. A replica that knows the event but not the financial consequence has to infer it.
  • The per-stratum analysis (X3) should be read with that in mind.
• Evaluations and affect again diverge. Positive events lift life satisfaction but not affect (as in Anusic et al. 2014). The replica's model of change must be item-specific. The LISS test should keep evaluation and affect outcomes separate.
• Co-occurrence and anticipation.
  • Events come bundled and are often anticipated. For LISS T1, co-occurring events should enter as covariates when a person's reaction to one event is estimated. The pre-event baseline should be taken far enough before the event (the paper uses 2–3 years) to avoid anticipation.
  • In the GSS pilot, the earlier wave may already contain anticipation, and the prompt names only the one event behind the stratum.

Cross-references

• summaries/carry_on/anusic2014_hilda_personality.md: the same panel; life satisfaction against SF-36 affect.
• summaries/carry_on/luhmann2012_adaptation.md: meta-analytic cognitive/affective split.
• summaries/carry_on/infurna2016_resilience.md: declines before the event.
• docs/research/gss_panel_feasibility.json, docs/research/gss_pilot_design.md: event-linked items, X3.
• docs/research/liss_q1_design.md: T1 design.