Paternal Relational Valuations and the Construal Asymmetry — Pre-registration of Stage 0

Stage 0 is a descriptive online survey of older U.S. parents about their retrospective reflections on past work–family decisions. The survey includes one between-subjects experimental component: respondents are randomly assigned to one of two versions of a hypothetical scenario involving a young parent facing a work–family trade-off, which differ only in the gender of the protagonist (male or female). The survey also includes a within-subjects sensitivity check on the order of rows in a price-list elicitation task. Specific details of these randomizations are reserved for the hidden registration to preserve the integrity of the measurement, and will be released in full after data collection is complete

2026-05-28

2027-05-29

H1. Asymmetric counterfactual regret. Difference between the share reporting B2a = "too little" (wished for more time with their child) and the share reporting B4a = "too much" (wished they had worked less). Test: McNemar's paired test within respondents.
H2. Continuous relational-regret magnitude. Median additional shared moments per week (b2b_moments_per_week) among respondents who report "too little" on B2a. Test: one-sample Wilcoxon signed-rank against a null median of zero.
H3. Retrospective willingness-to-accept for parental time. Bracket midpoint between the last "Yes" and first "No" in the dollar multiple price list, in dollars per year (mpl_wta), restricted to interior switchers. Primary covariate specification is Tobit right-censored at $200,000. Test: one-sample Wilcoxon signed-rank against a null median of $5,000.
H4. Strength of advice to reduce hours by protagonist gender. 1–7 Likert advice-strength item (D2), with the contrast E[D2 | Anna] > E[D2 | Alex]. Test: one-sided t-test.
H5. Test of the wording-artifact hypothesis on cross-gender relational regret share. The pre-registered prediction is null: the share reporting B2a = "too little" is statistically indistinguishable between male and female respondents. Test: HC3 logistic regression of I(B2a = "too little") on female_respondent. Rejection criterion: |marginal effect| < 9.2 percentage points (95% CI half-width at n = 225 per gender).

(1) Asymmetric counterfactual regret (H1). Constructed as the within-respondent difference between two binary indicators: r_too_little = I(B2a = "too little") and c_too_much = I(B4a = "too much"). McNemar's test on discordant pairs. Pilot 3 evidence: 11-to-1 discordant pairs, p = 0.003.
(2) Continuous relational-regret magnitude (H2). Constructed from b2b_moments_per_week, the open numeric follow-up shown conditionally to respondents who report B2a = "too little." Tested via one-sample Wilcoxon signed-rank against null median zero.
(3) Retrospective willingness-to-accept for parental time (H3). Operationalized as the midpoint between the last "Yes" and first "No" in the dollar MPL (mpl_wta), restricted to interior switchers (mpl_consistent = TRUE, mpl_always_yes = FALSE, mpl_always_no = FALSE). Primary covariate specification is Tobit right-censored at $200,000. Right-censoring share (mpl_wta_censored) and floor share reported as auxiliary descriptives.
(4) Strength of advice by protagonist gender (H4). The 1–7 Likert D2 item, with contrast E[D2 | Anna] > E[D2 | Alex]. Pilot 2 effect +0.57 (p = 0.014); Pilot 3 point estimate +0.34 (underpowered at n = 50; ~99% power at main-wave n).
(5) Cross-gender relational regret share (H5). Share reporting B2a = "too little" compared between male and female respondents via logistic regression. The pre-registered direction is null — the wording-artifact hypothesis predicts the v3 categorical item produces no cross-gender gap. 95% CI half-width at n = 225 per gender is 9.2 percentage points.

S1–S2. Continuous advice outcomes. Hours recommendation to the protagonist (D1, integer 20–45) and percent-of-income willingness-to-accept (mpl_pct_wta), both analyzed by protagonist gender and respondent gender.
S3. Hours-controlled relational regret intensity. Median regression (qreg) of b2b_moments_per_week on A3 (daily interaction hours), A4 (weekly work hours), and demographic controls, restricted to respondents with B2a = "too little." Primary specification updated to median regression per Pilot 3 outlier sensitivity (Change 3). HC3 OLS reported as robustness.
S4. Financial slack moderates WTA. Tobit regression of mpl_wta on financial_slack (E9) with demographic controls, right-censored at $200,000 (Change 4). Predicted positive coefficient on financial_slack.
S5–S7. Within-gender splits of primary outcomes H1, H2, and H3. McNemar, median regression, and Tobit specifications reported separately for male and female respondents.

(1) Distributions of counterfactual regret. Share selecting each response option ("Too little," "About the right amount," "Too much," "I'm not sure") in B2a and B4a, reported separately by domain and respondent gender. Descriptive support for the H1 asymmetry outcome.
(2) Retrospective satisfaction (B1, B3). 7-point Likert satisfaction items for the relational and career domains, reported by domain and respondent gender.
(3) Continuous advice and percent-of-income WTA (S1, S2). Hours recommendation (D1) and percent-of-income WTA (mpl_pct_wta), analyzed by protagonist gender and respondent gender.
(4) Determinants of regret magnitude and WTA (S3, S4). S3: median regression (qreg) of b2b_moments_per_week on A3, A4, and demographic controls, restricted to B2a = "too little" subset; HC3 OLS as robustness. S4: Tobit regression of mpl_wta on financial_slack (E9) with demographic controls, right-censored at $200,000.
(5) Within-gender splits of H1, H2, H3 (S5, S6, S7). McNemar, median regression, and Tobit specifications reported separately for male and female respondents.

Stage 0 is an online survey of older U.S. parents (age 55+) administered through the Prolific platform, running approximately 15 minutes. Target analytic sample is n = 450 (225 male, 225 female) after pre-registered exclusions, with 240 of each gender recruited to allow for approximately 6 percent Tier 1 plus Tier 2 attrition. Recruitment is stratified 50/50 by gender via two separate Prolific studies.
The survey collects retrospective reflections on past work–family decisions, including satisfaction with past choices, counterfactual reflections on time and work allocation, willingness-to-accept measures for additional time with one's child, and advice in a hypothetical scenario about a young parent. The survey includes one between-subjects randomization: the gender of the protagonist in the hypothetical advice scenario (male vs. female), 50/50. Specific details are reserved for the hidden registration and will be released in full after data collection is complete.
Stage 0 is the first wave of a multi-stage research project on parental valuations of time with children. Subsequent stages (Stage 1, Stage 2) examine forward-looking parental decisions and behavioral anchors, and will be pre-registered separately as their designs are finalized. Stage 0 results will inform the design of subsequent stages.

Not available

Computer-based randomization performed server-side at participant entry (oTree 6.x platform).

Individual respondent (Prolific worker). No clustering or higher-level randomization unit.

No

480 individuals recruited (240 male, 240 female); 450 analytic after pre-registered exclusions (225 male, 225 female). No clustering; randomization at the individual level.

480 individuals recruited (240 male, 240 female); 450 analytic after pre-registered exclusions (225 male, 225 female).

Row order of dollar MPL: ~240 ascending / ~240 descending
Protagonist gender: ~240 male (Alex) / ~240 female (Anna)
Total cells: 2 × 2 = 4 unique combinations, ~120 individuals per cell at recruitment (~112 analytic per cell after pre-registered exclusions). Marginal n per arm at recruitment ~240; ~225 analytic per arm after exclusions.

Stage 0 has two pre-registered arm comparisons among its primary outcomes. The asymmetric counterfactual regret outcome (H1) is descriptive and reported as a point estimate with 95% confidence interval rather than an MDE-detectable contrast. For the advice-strength contrast by protagonist gender (H4), with approximately 225 analytic respondents per arm and within-arm SD ≈ 1.5 on the 7-point scale, the MDE at 80% power is approximately 0.22 SD for a one-sided directional test (α = 0.05) — roughly 0.33 points on the 7-point scale, or 5.5% of the scale range. The Pilot 2 observed effect was +0.57 points (d ≈ 0.41); power to detect this effect at main-wave n is approximately 99.6%. For the cross-gender relational regret share (H5), the pre-registered prediction is null (wording-artifact equivalence test). At n = 225 per gender, the 95% CI half-width on the share difference is 9.2 percentage points. The rejection criterion is |marginal effect| < 9.2pp; no MDE applies. For the dollar WTA one-sample test (H3), with n = 450 interior switchers and within-sample SD ≈ $25,000, the MDE against the null median of $5,000 is approximately $3,700 at 80% power. Primary specification is Tobit right-censored at $200,000 (Change 4); z(0.025) + z(0.20) = 2.802.