Haehner, Pfeifer, Fassbender & Luhmann (2022/2023). Are changes in the perception of major life events associated with changes in subjective well-being?

Journal of Research in Personality 102, article 104321 (issue of February 2023; DOI registered 30 November 2022 per Crossref), DOI 10.1016/j.jrp.2022.104321. The published version is closed access (OpenAlex) and was not read.

• Main text read as the PsyArXiv preprint (DOI 10.31234/osf.io/hzkq4, version 1, the only version), in the second and current revision of its file, uploaded on 14 June 2022 (OSF API). Revision 1, of 23 February 2022, was not read.
  • OpenAlex labels this download the accepted version. The PDF's own header says "Draft version June 14, 2022. This paper is submitted for publication and has not yet been peer reviewed". Blinding placeholders remain.
• Supplement read as the publisher's file for the published article (Elsevier mmc1.docx, created 22 November 2022, served openly).
  • Main text and supplement therefore come from different versions. The supplement numbers its tables up to two higher than the preprint cites them, so the published main text was revised after this draft (inferred).
• OSF lists the preprint as CC BY 4.0; the PDF asks readers "Please do not copy without author's permission".
• Ruhr University Bochum and University of Siegen. Provenance: papers/carry_on/haehner2023_perception_change_swb.provenance.json.

What was read

• Preprint: all 1,021 lines (my count) of pdftotext -layout output (30 pages): title page, abstract, author note, both studies, general discussion, footnote 1, references, Tables 1–6 and the caption of Figure 1.
• Supplement: all 1,323 lines (my count) of pandoc -t plain output: Tables S1–S29 and the note of Figure S1.
  • Figure S1, the supplement's only figure, was viewed as an image.
  • Pandoc drops bold type, which marks significant cells; the bold cells of Tables S23–S29 were recovered from the .docx XML.
• Not read or not inspected:
  • the drawings of Figure 1 (model diagrams; caption read);
  • the published article and preprint revision 1;
  • the preregistrations (osf.io/sg3nk, osf.io/x52bq, osf.io/3yw2a) and the appendix to the preregistration on OSF;
  • data, codebooks and R scripts (OSF cp7d6, components e23bc and b392y).
• In the preprint's Tables 2 and 5 the bold marking is lost in extraction. Significance there was read from the printed p-values.

Question

When a person's appraisal of the same major life event changes over three months, does their subjective well-being (SWB: life satisfaction, positive affect, negative affect) change with it? Which appraisal dimensions matter? An exploratory add-on asks which of the two changes first.

Method

• Design. Two short-term longitudinal online studies.
  • At the first occasion each person named or picked their most important recent event. They rated it on the Event Characteristics Questionnaire (ECQ; Luhmann et al. 2021), nine dimensions, items 1–5.
  • About three months later they were shown the same event and rated it again. SWB was measured both times.
• Study 1, exploratory. The What's NEXT? Study, 2018–19, T1 and T2 only. This is the same sample and the same T1 event as in haehner_perception_stability (Table S1).
  • 857 registered. N = 619 at T1 and 433 at T2, after quality checks and after dropping events older than 3 months. Mean age 21.48 (SD 4.05), 73% female.
  • The event was freely named from the last 3 months and coded into categories (κ = .87). Most frequent: vacation 52, begin college 48, relocation 47, school graduation 46 (Table S2).
  • Measures: ECQ 37-item version; SPANE 3 + 3 items, frequency in the last month; SWLS 3 items.
• Study 2, confirmatory and preregistered. The "One Year of Corona Pandemic" study, 2021. Occasions at 0, 1 and 12 weeks; T1 and T3 are used, 12 weeks apart.
  • 1,075 registered. N = 691 at T1 and 438 at the second rating. Mean age 34.18 (SD 12.02), 71% female.
  • The event was the most significant one ticked on a 32-item checklist for the past 12 months, with a free option. Most frequent: change in the work situation 61, death of a loved one 57, new job 55, relocation 49, own serious illness or injury 43 (Table S13).
  • Measures: final 38-item ECQ; SPANE 6 + 6; SWLS 5 items.
  • Power: rΔΔ = .20 detectable with 600 people at 84% power (α = .05).
• Model. Bivariate latent change score models (lavaan), one per SWB component × ECQ dimension, so 27 per study.
  • Each construct has a latent factor at both occasions and a latent change factor. Weeks since the event predicts the change. Missing data are handled by full-information maximum likelihood.
  • (Partial) scalar measurement invariance was established first.
  • The target statistic is rΔΔ, the correlation between the two latent changes.
  • Study 1 used α = .01. Study 2 used α = .05 for the 8 correlations significant in Study 1 and α = .01 for the other 19.
• Robustness checks (Study 2).
  • Only the items the two ECQ versions share.
  • Covid-19 stress as a covariate; age as a covariate.
  • Only event categories present in both studies (N = 533).
  • Only events from the last 3 months (N = 136).
• Temporal order (exploratory, not preregistered). Each model gained two paths: SWB at T1 → change in the appraisal (Path A), and appraisal at T1 → change in SWB (Path B). 27 models × 2 paths × 2 studies = 108 tests at α = .01.

Results

• How much appraisals change in three months. Manifest T1–T2 correlations (Tables S6, S18):

  Measure	Study 1	Study 2
  Valence	.90	.91
  Predictability	.84	.83
  Challenge	.78	.83
  Impact	.75	.78
  Emotional significance	.73	.74
  Change in world views	.69	.66
  Extraordinariness	.67	.65
  Social status change	.67	.77
  External control	.66	.72
  Life satisfaction	.78	.87
  Positive affect	.54	.70
  Negative affect	.54	.72

  • The supplement's descriptive tables (S5, S17) add an "ICC" column without defining it. Its values match these correlations to within .03 (my comparison).
  • Standardised mean latent change (Mz; Tables 1 and 4):
    • Study 1: impact −0.57, emotional significance −0.42, change in world views +0.36, life satisfaction +0.37.
    • Study 2: negative affect −0.44, impact −0.34, emotional significance −0.33, social status change +0.23, positive affect +0.20.

• Study 1: correlated change (Table 2). 8 of 27 correlations reach α = .01:

  • life satisfaction: valence .34, social status change −.28, external control −.25;
  • positive affect: challenge −.26, valence .19, external control −.19;
  • negative affect: challenge .31, valence −.23.

  The authors call these effects "at least of medium size".

• Study 2: replication (Table 5). 4 of the 8 replicate at α = .05:

  • life satisfaction with external control −.18 (p = .008) and with social status change −.17 (p = .014);
  • positive affect with challenge −.19 (p = .003);
  • negative affect with challenge .19 (p = .006).

  Not replicated: valence with life satisfaction .04 (p = .592), with positive affect .10 (p = .074) and with negative affect −.06 (p = .301); external control with positive affect −.02 (p = .697). New at α = .01: social status change with negative affect .17 (p = .009).

• Change is not the same as level (my comparison). The T1 correlations between scale means (Tables S6, S18, between persons) and the change correlations (Tables 2, 5), Study 1 / Study 2:

  Pair	Level at T1	Change (rΔΔ)
  External control – life satisfaction	−.04 / −.11	−.25 / −.18
  Social status change – life satisfaction	−.31 / −.31	−.28 / −.17
  Challenge – negative affect	.33 / .38	.31 / .19
  Valence – life satisfaction	.26 / .25	.34 / .04

  External control barely relates to the level of life satisfaction but does relate to its change.

• Robustness (Table 6; Tables S23–S27).

  • Covid-19 stress or age as covariates: correlations nearly identical.
  • Shared items only: larger standard errors. Of the four replicated pairs, only positive affect–challenge (−.19) and life satisfaction–external control (−.19) stay significant.
  • Shared event categories (N = 533): stronger. Life satisfaction with external control −.26 and with social status change −.26; valence with positive affect .15 (p = .024).
  • Events from the last 3 months (N = 136): the affect couplings are largest, none significant at .01.
    • Positive affect with emotional significance −.31, external control −.29 and valence .29.
    • Negative affect with social status change .31 and challenge .28.

• Time since the event matters for affect much more than for life satisfaction (Figure S1, viewed).

  • The mean |rΔΔ| over the nine dimensions falls as older events are included. For positive affect it goes from about .18 (events up to 3 months old) to about .09 (up to 12 months); for negative affect from about .20 to .08.
  • For life satisfaction it stays between about .11 and .07.
  • My computation from Tables S27 and 5 reproduces the end points: .183 and .087, .197 and .082, .107 and .068.

• Which changes first (Tables S28, S29). 6 of 108 paths are significant (bold), matching the text's "102 out of 108 associations were not significant".

  • SWB at T1 → later change in appraisal, 5 paths:
    • negative affect → challenge rises, in both studies (b = 0.12, p = .009; b = 0.14, p = .001);
    • Study 2 only: life satisfaction → extraordinariness falls (−0.06, p = .004) and social status change falls (−0.07, p = .002); negative affect → valence falls (−0.11, p = .010).
  • Appraisal at T1 → later change in SWB, 1 path: extraordinariness → positive affect falls (Study 2, −0.16, p = .001).
  • The authors "refrain from drawing any conclusions about the direction of causality".

• Other.

  • Weeks since the event predicted only the change in challenge, in Study 1 (b = −0.03, p = .003).
  • Covid-19 stress predicted changes in valence (p = .009) and in change in world views (p = .003).
  • Age predicted no change (all p > .010).

Limits

• A between-person correlation of within-person changes. Each person contributes one change per variable.
  • rΔΔ says that people whose appraisal changed more also changed more in SWB. It does not show that one person's appraisal and SWB move together over time, or that coupling differs between people (inferred from the model).
• One lag. Two ratings three months apart. The time-frame analyses suggest that the coupling with affect is strongest soon after the event (verified); other lags are not measured.
• Direction is not identified.
  • Correlated change fits several stories: re-appraisal driving SWB, mood colouring the memory of the event, or a third factor. The authors name mood-congruent memory and, as causes of re-appraisal, changed consequences of the event.
  • The temporal-order paths are exploratory. At α = .01, about 1.08 false positives are expected among 108 tests; 6 were found (my computation).
• Same-occasion self-report. Both constructs come from the same questionnaire. The authors rule out acquiescence (no link with the number of reverse-keyed items), not mood colouring both ratings at each occasion (inferred).
• Samples and events. German-speaking and 71–73% female. Study 1 is young with mostly positive events; Study 2 ran during the pandemic. Only each person's most important event was rated.
• Attrition was tested at α = .01. At .05, Study 1 completers and dropouts differed in age (d = 0.20), gender and emotional significance (Table S4).
• Effect size. rΔΔ between .17 and .34 means re-appraisal shares about 3–12% of the variance of SWB change (my computation, r²).
• Version. Preprint main text (not peer reviewed per its header), published supplement.
• Inconsistencies found:
  • Direction of impact change. The Study 1 text lists impact as "increased". Table 1 shows −0.08 (Mz −0.57), a decrease.
  • Labels. Table 4 is titled "for Study 1" but holds the Study 2 sample (N = 691). Study 2's second rating is T3 in the method but T2 in the sample text and the Table 4 header. Table S19, in the Study 2 part, is also titled "Study 1".
  • ECQ item list (Table S11).
    • "The event was stressful." is listed twice among the shared challenge items.
    • The main text's Study 1 examples "The event was surprising" (predictability) and "The event was extraordinary" (extraordinariness) are listed as Study 2 only.
    • The table gives Study 1 four extraordinariness items; the text says three.
    • These are verified by comparing the texts. That one item's two translations were listed as two items is a guess.
  • Table notes. The notes of Tables S23 and S27 describe the event-category restriction; their titles describe the shared-items and last-3-months restrictions.
  • Two p = .010 paths treated differently (Table S29). Negative affect → valence is bold; negative affect → impact is not. The count of 6 depends on this (verified from the .docx). Unrounded p-values on either side of .01 are a guess.
  • The explanation for valence. The authors write that "these differences can at least partly be explained by differences in the employed time frame of the considered events" and the event categories. That holds for affect. For life satisfaction the valence coupling stays at .05 under both restrictions (Tables S26, S27), against .34 in Study 1 (verified numbers).
  • Table references. The preprint cites Tables S26–S27 for the temporal-order paths and S5 for correlations. In the published supplement these are S28–S29 and S6.
  • Same sample, different N. Study 1 has N = 433 at T2 here and 430 in haehner_perception_stability, for the same sample and event (both verified; no reason given).
  • Checked by script:
    • Tables 1 and 4 equal S5 and S17 once the ICC column is removed.
    • p from z agrees within .005 in Tables 2 and 5.
    • All 24 Mz values equal M/SD within rounding.
    • The Figure S1 end points, the 8 Study 1 hits, the 4 replications and the 6 of 108 paths all check.

What it means for Kurisutina

• Q3: re-appraisal is a memory change, and it moves with well-being.
  • Over three months people re-rate the same event with manifest correlations of .65–.91 by dimension. So a real part of the appraisal changes (verified).
  • Changes in appraisal go with changes in SWB, dimension by dimension: challenge with affect, and external control and status threat with life satisfaction. These replicate in two studies (verified).
  • The temporal paths point mostly from current well-being to later re-appraisal. Higher negative affect predicted the event coming to seem more challenging, in both studies (verified numbers). Reading this as mood-driven re-appraisal is inferred, and the evidence is weak.
  • A replica that writes an appraisal once and never revises it cannot show this coupling. One that re-derives appraisals from its current state might, but could overshoot: in people the coupling is |r| ≈ .2–.3, not 1 (inferred).
  • Proposed test, not started. Change the replica's state, e.g. by a run of negative experiences, then have it re-rate earlier events. Compare with the human reference: a small positive path from negative affect to challenge (b ≈ 0.12–0.14).
  • This agrees with haehner_perception_stability. Corrected for the printed alphas, the three-month correlations of Study 1 are .81–.95, except impact at 1.01. The latent twelve-month coefficients there (.622–.793) imply three-month values of about .89–.94 (fourth roots). Both are my computations.
• Q1: what the slot must hold.
  • How a person reacts to an event depends on their current appraisal of it, and that appraisal is not a fixed property. It shifts over weeks and moves with their mood (verified as a correlation across people; per-person coupling untested).
  • So, per significant recent event, the slot should hold the current appraisal by dimension and the time since the event, together with the person's current affective state. The replica should update the appraisal as time passes and its state changes, not freeze it (inferred).
  • The dimensions that carry SWB change are not only valence. External control and status threat go with life satisfaction, and challenge with affect. Valence coupled with all three SWB components in Study 1 (young adults, events within 3 months) but with none in Study 2; with positive affect it returned only in the restricted analyses (verified).
  • Whether this person's SWB tracks their own re-appraisals is untested. That needs several events or occasions per person (inferred).
• Held-out appraisal test (goal doc proposal). The target moves: a person's own three-month re-rating agrees at only .65–.91. Impact and emotional significance fall on average and world-view change rose in Study 1 (verified). Give the replica the time since each event and score whether it predicts that drift. Its agreement ceiling should be the person's own re-rating agreement (inferred).
• LISS T1 and T2 (docs/research/liss_q1_design.md).
  • LISS records no appraisal. Part of each reaction is therefore driven by an unrecorded appraisal that is itself moving, which will dilute T1's reaction consistency (inferred).
  • Timing: the coupling with affect is strongest for events up to about 3 months old and halves by 12 months, while life satisfaction couples weakly throughout (verified, Figure S1). For affect outcomes, the measurement nearest after the event, within about 3–6 months, is where appraisal-driven differences should show (inferred).
  • SPANE here asks about the last month. In LISS, the Health module's items (PHQ: "over the last two weeks", per the LISS design doc) are closer to that frame than the Personality module's "right now" PANAS (inferred).
  • T2 (own-full against own-state): an earlier reaction carries how that occurrence was appraised. A model that sees the reaction but not the appraisal cannot separate the person's style from the occasion (inferred).
• GSS pilot (frozen; nothing to change).
  • happy is the only target where this paper's coupling applies. GSS records no appraisal, and its two-year interval lies well past the 3-month window where coupling is strongest (inferred). This is consistent with the declared expectation E4, that the rest of the record adds little (inferred).
  • The paper says nothing about satfin, finalter, health, trust, polviews, partyid, attend or confinan (verified: SWB only).
• Could the method be a condition or a measure?
  • Measure for a replica population. In a synthetic-persons run (goal doc part (c)), have several hundred replicas each rate an event and answer SWB items at two points. Fit the same bivariate change models, and compare their rΔΔ pattern with Tables 2 and 5. It needs hundreds of persons, so it describes a population of replicas, not one replica (proposal).
  • Measure for one person. The measurement-burst proposal (goal doc; haehner_perception_stability summary) re-rates reference events at each burst. Correlating one person's re-appraisal changes with their affect changes across bursts gives the within-person coupling this paper lacks (proposal).
  • Condition. A memory policy that re-rates stored events from the current state at each checkpoint, against one that freezes the first appraisal. Score whether appraisal and SWB changes couple as in people (proposal).

Cross-references

• summaries/carry_on/haehner_perception_stability.md: the same Study 1 sample and event; stability over a year.
• summaries/carry_on/luhmann2021_ecq_taxonomy.md: the ECQ; appraisals and prospective SWB change.
• summaries/carry_on/haehner2022_event_perception.md, summaries/carry_on/kritzler2022_event_perception_profiles.md, summaries/carry_on/rakhshani2021_traits_event_perception.md: the rest of the appraisal thread.
• summaries/carry_on/hirst2009_911_memory.md, summaries/carry_on/schacter2011_adaptive_distortion.md: how memories of events change.
• summaries/carry_on/sliwinski2009_stress_bursts.md: measurement bursts; stable reactivity against current state.
• docs/research/liss_q1_design.md, docs/research/gss_pilot_design.md.