Luhmann, Fassbender, Alcock & Haehner (2020/2021). A dimensional taxonomy of perceived characteristics of major life events

Journal of Personality and Social Psychology 121(3), 633–668 (2021; online 27 April 2020 according to OpenAlex), DOI 10.1037/pspp0000291. The published version is closed access and was not read. Read as the PsyArXiv preprint, version 1, the only version, whose file was uploaded on 6 March 2020 and never revised (OSF API). It states that it was accepted on 13 February 2020 and is "not the copy of record", so it is the accepted manuscript after peer review (DOI 10.31234/osf.io/yx5pk). OSF lists it as CC BY 4.0, but the PDF carries an APA copyright line and asks readers not to copy or cite it without permission. Ruhr University Bochum and University of Cologne. Provenance: papers/carry_on/luhmann2021_ecq_taxonomy.provenance.json.

What was read

All 4,368 lines of pdftotext -layout output (100 pages; my count with wc and pdfinfo): the title page with the acceptance statement, author note, abstract, Studies 1–5, general discussion, references, Tables 1–8, Appendix A (the 38 final items in German and English) and the captions of Figures 1–7. Table 1, the literature review, is a very wide table whose extracted layout is fragmented; it was read as extracted.

Not read or not inspected:

• the figures, which are images and hold the trajectories and simple slopes of Studies 3–5;
• the supplemental material: Study 1b, Tables S2–S22 (including the Study 5 simple slopes, S19, and robustness checks, S20–S22) and Figures S5–S7;
• the Study 5 preregistration (osf.io/kvf5g), the OSF data and scripts (osf.io/g4uyx), and the published version.

Questions

1. Which perceived characteristics distinguish major life events, and can they be measured reliably?
2. Do these perceptions explain individual differences in how subjective well-being (SWB) changes after an event, beyond personality and demographics? Studies 3–4 test this retrospectively, Study 5 prospectively.
3. How do they relate to taxonomies of everyday situations (DIAMONDS)?

The premise is "intracategorical variability": "MLEs that are objectively identical can result in widely different subjective experiences across individuals". A major life event (MLE) is one that is clearly timed, disrupts everyday routine, and is perceived as personally significant and memorable.

Method

• Study 1, literature review. PsycINFO, Psyndex and Google Scholar, plus backward and forward searches (Table 1). Seven broad dimensions came out: valence; controllability and attribution; anticipation; familiarity; intensity and frequency; adjustment and change; emotional significance.
• Study 2, item pool. 57 items. 449 German-speaking adults (81.5% female, mean age 26.1), 2015. Each rated one self-chosen event from the last 24 months. Exploratory factor analysis on polychoric correlations. Seven to nine factors were indicated; nine were kept (51 items with clean loadings).
• Studies 3 and 4, item selection and retrospective SWB.
  • 56 candidate items were cut by item response theory, classical item analysis and ant colony optimisation, then checked with confirmatory factor analysis (CFA).
  • Study 3: N = 226 (84.0% female, mean age 25.8), an event from the last 6 months.
  • Study 4: N = 373 (64.1% female, mean age 36.2), with the time frame randomised: any time, 6 months, or last week.
  • Retrospective SWB: mood and life satisfaction rated against the person's normal, from −3 ("much worse than normal") to +3 ("much better than normal"). Four time points: one month before the event, the event, one month after, and today. The two were averaged into one score.
  • Mixed models: time point × valence, then all ECQ subscales, age, gender, extraversion and neuroticism (BFI-2-XS). Study 3 tested 8 two-way and 8 three-way interactions at α = .0031; only those hits were retested in Study 4, at α = .05.
  • Study 4 also measured the DIAMONDS situation characteristics (S8*, 24 items).
• Study 5, prospective.
  • Sample: the What's Next? Study, the same study that Haehner et al. (2022) used (haehner2022 summary). Young adults after school or college graduation; five online waves at 0, 3, 6, 9 and 12 months; July 2018 to October 2019.
  • Event: the one reported at T2, i.e. one that occurred between T1 and T2. It was rated at T2 on an earlier 37-item ECQ version, whose items differ from the final ones on 7 of the 9 subscales.
  • N = 429 at T1 and T2 (75.5% female, mean age 21.7), then 333, 294 and 285; 235 took part in all five waves.
  • Outcomes: life satisfaction (the first three items of the Satisfaction With Life Scale) and mood (6-item SPANE). They were analysed separately, because they correlated only .62 to .68.
  • Model: piecewise growth, with a dummy for T1 (the only pre-event wave) and weeks since the event. The intercept is the level at the event. Outcomes are in T1-SD units. Valence was centred on the scale midpoint 3 (its mean was 3.96, SD 1.23), so most events were positive.
  • Slopes were fixed: "models with random slopes did not converge". Quadratic change did not improve fit. Three-way interactions were tested at α = .05 each, with no correction.

Results

• Nine dimensions make up the Event Characteristics Questionnaire (ECQ):
  • valence, impact, predictability, challenge, emotional significance, change in world views, social status change, external control and extraordinariness;
  • all valence items loaded on one bipolar factor. Positive and negative aspects could not be separated; the authors suspect that self-chosen events have unambiguous valence.
• The final ECQ has 38 items: six for valence and four for each other subscale.
  • Cronbach's α was .67–.97 in Study 3 and .68–.94 in Study 4. Extraordinariness was the lowest (.67, .68).
  • The challenge model fitted poorly in Study 4 (χ²(1) = 17.80, RMSEA .21).
  • The combined CFA fitted well (CFI .98 both, RMSEA .061 and .053) after one world-views item was dropped.
  • The four social-status items are all about loss: status threatened, standing hurt, status impaired, reputation damaged (verified, Appendix A). So "social status change" measures perceived status damage.
• Retrospective SWB (Studies 3–4).
  • Time × valence: χ²(3) = 331.76 and 246.81. SWB rose at positive events, fell at negative ones, then returned towards normal, more fully after negative events.
  • The other eight subscales improved fit beyond age, gender, extraversion and neuroticism (χ²(8) = 27.68 and 32.01).
  • Three-way interactions that replicated: impact, predictability, social status changes. In the authors' words, "Adaptation to negative events was slower for individuals who perceived the event as higher in social status change, and adaptation to positive events was slower for individuals who perceived the event as higher in impact and lower in predictability" (the figures were not inspected).
  • My computation from Table 5: the valence effect on remembered SWB was 1.61 and 1.42 SD per SD at the event, and 0.78 and 0.55 "today". The valence-linked difference falls by half or more between the event and the present.
• Overlap with situation taxonomies (Study 4, Table 7). DIAMONDS explained R² = .78 of valence and .75 of challenge, .37 of social status change and .29 of change in world views. It explained .20 or less of predictability, emotional significance, impact, external control and extraordinariness (.07).
• Time frame (Study 4). Events recalled from a longer window were rated higher on challenge, emotional significance, external control, extraordinariness, impact and change in world views.
• Study 5, prospective (Table 8).
  • Valence moderated change in both outcomes. Per SD of valence, the dummy × valence term was −0.07 for life satisfaction and −0.12 for mood (T1 to event), and the weekly slope term was −0.002 and −0.004 (after the event).
  • My computation from Model 1, assuming weeks-since-event is 0 at T1, as the paper's description implies. Per SD of valence, people were already 0.16 (life satisfaction) and 0.14 (mood) T1-SD apart before the event. They were 0.23 and 0.26 apart at the event, and 0.15 and 0.10 apart by 40 weeks after it. The part added at the event (0.07, 0.12) was gone within about 40 weeks; the rest predates the event.
  • Adding the other eight subscales improved fit (χ²(8) = 23.78 against the initial model, 17.72 beyond the covariates). These are main effects, i.e. levels, as the text itself says: "A significant fixed effect indicates that the respective predictor accounts for individual differences in the level of the trajectories".
  • Level effects in Model 2:
    • life satisfaction: valence +0.10, social status change −0.11;
    • mood: challenge −0.16, external control +0.07;
    • neuroticism (−0.37 and −0.45) was larger than any ECQ level effect; extraversion was +0.13 and +0.14.
  • Shape: for mood, no subscale beyond valence mattered. For life satisfaction, four three-way terms with weeks since the event: challenge (b = −0.003, p = .014), extraordinariness (−0.002, .023), impact (−0.002, .034) and social status changes (−0.002, .005). Negative events that were also challenging, unusual, impactful or status-damaging were followed by the steepest rises in life satisfaction. The impact term vanished with the reduced item set.
  • Retrospective and prospective data disagree. Impact and social status change went with slower adaptation in Studies 3–4 and with more pronounced change in Study 5. Predictability did not replicate, and challenge and extraordinariness were new. The authors offer hindsight re-evaluation of the event and "affective backcasting" (a misremembered adaptation) as explanations.
• Between events versus between people: not estimated. The authors: "we were not yet able to investigate whether event characteristics accounted for systematic differences among MLEs because the number of specific MLEs was not yet sufficient for these kinds of analyses". Separating "consensual ratings" from the "idiosyncratic perspective" is left to future work.
• Size against Haehner et al. (2022), my computation.
  • Here: per SD of perceived valence, the immediate SWB change is 0.07 (life satisfaction) and 0.12 (mood) T1-SD. The slope terms add about 0.08 and 0.16 over 40 weeks.
  • Haehner et al. report at most about 0.05 SD of Big Five change per SD of an ECQ dimension, in the same study and wave (haehner2022 summary).
  • So the SWB effects are about 1.5–3 times the largest trait effect, and larger for mood. The models differ (growth curve with fixed slopes against latent change scores), so the ratio is rough (inferred).

Limits

• Samples. German, young, mostly female, many students. Events were self-chosen and mostly positive.
• Retrospective data. SWB in Studies 3–4 is remembered, and Study 5 shows that it can disagree in direction.
• Study 5 cannot size individual differences in change. With fixed slopes, differences in change enter only through the measured moderators. The residual person-specific variance, and the share of it that the ECQ explains, are not reported (inferred from the model description).
• The claim of prediction "over and above" personality rests on levels. Traits entered only as main effects, not as moderators of change. There is no head-to-head test of traits against appraisals for change (inferred from the Methods and Table 8).
• Timing of the ratings. The ECQ was rated at T2, on the same occasion as a T2 SWB measurement and weeks after the event. The valence–SWB association after the event is partly concurrent. The pre-event gap (0.16 and 0.14, my computation) shows selection or anticipation, which the paper does not discuss (inferred).
• Multiple tests in Study 5. Eight subscales × two time terms × two outcomes gives 32 uncorrected tests; 4 reached p < .05, against about 1.6 expected by chance (my count). In Study 4, α = .05 was justified by retesting only Study-3 hits; in Study 5 every subscale was tested.
• Versions and language. The ECQ version changed between studies. The English items are "translated for the purpose of data presentation only and have not yet been validated empirically".
• Inconsistencies found:
  • Valence CFA. Table 4 gives df = 2 for valence in both studies. That is what a one-factor model with four indicators gives; six indicators give 9. The combined model's χ²(521) matches 35 items, i.e. four valence items plus the other subscales minus the dropped world-views item (my computation). The text says all six valence items were retained and does not say which were modelled.
  • Text against Table 5.
    • The text says "change in world views was positively associated with levels of retrospective SWB in both studies". Table 5 gives p = .171 and .078.
    • "Valence ... negatively" associated refers to the effect at the reference time point, one month before the event. At the event, the valence effect is +1.61 and +1.42 (my computation).
  • Summary of Studies 3–4. It names "valence, impact, and social status change" as shaping the trajectories. The Results and Table 6 name impact, predictability and social status changes.
  • Table 7. The text says R² for the remaining subscales was ".20 or less"; change in world views has .29.
  • DIAMONDS count. The Measures section speaks of "the nine DIAMONDS situation characteristics"; DIAMONDS has eight, and Table 7 regresses on "the 8 DIAMONDS subscales".
  • Study 5 start date. "July 2018" (Sample) against "August 2018" (Measures).
  • Footnote 3. It sizes Study 4 from "all significant three-way interaction effects in Study 4", presumably meaning Study 3.
  • Standardisation of valence. "T1 standard deviation" in the text, "total sample standard deviation" in the Table 8 note. Valence was rated only at T2.
  • Which outcome. The Study 5 model comparisons (23.78, 17.72) are reported once, without saying whether they are for life satisfaction or mood.
  • Appendix A. Item 14 is marked reverse-scored in German but not in English.
  • Table 1. The year column disagrees with the citation for Brown & Harris (1973 against 1978) and Gentzler et al. (2013 against 2014).
  • Arithmetic checked by script: all 18 p-values of Table 4, all 32 of Table 6, eight model comparisons, the total N of 1,477 (449 + 226 + 373 + 429), and the exclusions of Studies 3 and 4.

What it means for Kurisutina

• Question 1: appraisal is where this person enters, but this paper cannot say how much of it is the person.
  • Prospectively, the person's perceived valence set the direction and size of the SWB reaction to their event (verified, Table 8). This is a reason for the slot to hold how she appraises, not only what happened (inferred).
  • Each person rated one event, and between-event and between-person variance were not separated (verified). Whether there is a stable personal appraisal style remains open. The companion paper finds 63–94% of long-range variance between persons, but with person and event mixed (haehner_perception_stability summary).
• The shape a replica's change should have. In this sample, the part of the reaction tied to valence appeared at the event and was gone within about 40 weeks. The rest of the valence gap predates the event (my computation; young adults, linear slopes). A replica that carries on should show a transient reaction and a return, not a lasting shift of that size (inferred).
• Retrospective accounts are not measured reactions.
  • The interviews that fill the slot are retrospective. Here, remembered trajectories disagreed in direction with prospective ones for impact and social status change (verified).
  • So a slot's how she reacted to X from interviews is her current account, not her past change (inferred). That suits the held-out appraisal test, whose target is what she says now. It should not be used as ground truth for her way of changing. Prospective panel data (LISS) are the better target for that.
• Held-out appraisal test (goal doc, latest entry): design constraints.
  • Use one ECQ version for the person, the replica and the baseline. The items differ between versions, and one Study 5 effect (impact) vanished on the reduced item set, which the authors put down to a single item (verified).
  • The English wording is unvalidated (verified). If the participant answers in English, report that; if German, use the Appendix A originals (inferred).
  • Record time since each event. Events recalled from longer windows were rated as more challenging, impactful and extraordinary (verified, Study 4); the typical-profile baseline must be matched on it (inferred).
  • Situation ratings capture most of valence and challenge (R² .78, .75) but little of predictability, external control, extraordinariness and impact. Those four may be where a person-specific appraisal is least recoverable from a description of the event, and so the more informative targets (guess).
  • The slot also needs the person's baseline well-being and traits. Neuroticism shifted SWB levels (−0.37, −0.45) more than any appraisal did (verified, Table 8).
• Measurement-burst proposal. With five waves and one event per person, person-specific slopes could not be estimated (random slopes did not converge; verified). Recovering a person's own way of changing needs many more occasions per person, which is the case for the burst design (inferred).
• LISS T1 (docs/research/liss_q1_design.md).
  • LISS records no appraisal. A T1 reaction score will mix a stable reaction style with how differently the person appraised each occurrence (inferred).
  • The valence-linked reaction faded within about 40 weeks here, so the planned months-since-event term is essential. A post-event wave within about six months is where appraisal-driven differences should be largest (inferred).
  • Pre-event SWB already differed by later appraisal (0.16 and 0.14 SD per SD, my computation). T1's population model conditions on pre-event answers and so absorbs that part. What is left for a T1 reaction is the event-linked part, which was small here: 0.07–0.12 SD per SD of valence (inferred).
  • Possible addition before T1 is fixed, if the repeated-event counts allow: compare consistency between event types whose valence is more or less ambiguous (a breakup's valence varies most; kritzler2022 summary). Consistency should be lower where appraisal can differ between occurrences (guess).
• GSS pilot (docs/research/gss_pilot_design.md, frozen; nothing to change).
  • Its two-year intervals will often fall after a reaction that fades within a year (inferred).
  • The pre-event state that predicts appraisal is partly in the earlier answers that every condition sees (inferred).
  • An own-slot gain would have to come from record items that predict how she appraised the event, for example marital happiness before a divorce (guess).
  • Beyond valence, appraisals shaped life satisfaction here but not mood. This bears on X2 only indirectly: GSS happy is one global item, and no appraisal is recorded (verified from the design). It is not a reason to revise the declared expectation (inferred).

Cross-references

• summaries/carry_on/haehner_perception_stability.md: the same study; stability of the same person's appraisal over a year.
• summaries/carry_on/haehner2022_event_perception.md: ECQ and Big Five change in the same sample and wave.
• summaries/carry_on/kritzler2022_event_perception_profiles.md: typical profiles; within-event variability.
• summaries/carry_on/rakhshani2021_traits_event_perception.md: traits barely predict appraisals.
• summaries/carry_on/luhmann2012_adaptation.md, summaries/carry_on/luhmann2014_its_about_time.md: reaction and adaptation; timing.
• 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.