Preferences as Memory States: Experience Effects at the Moment of Choice

• Two-wave crisis-cue panel: Participants from two birth cohorts (1983–1990 and 2001–2008) complete two sessions approximately one week apart. Main-panel participants read a passage about the 2008 financial crisis in one session and a length- and structure-matched neutral passage in the other, with passage order randomized. A calibration arm reads two different neutral passages in randomized order, providing a benchmark for week-to-week variation in the certainty equivalent.

• Account-history experiment: Participants earn a starting balance in a real-effort slider task and are randomized to monitor either a payoff-bearing own account or an otherwise matched study reference account. Every participant observes a five-round return history containing a focal −40% round, a focal +40% round, a zero-return round, and two filler rounds. A later randomized cue targets one episode by its serial position immediately before participants value the common risky lottery. A randomly assigned half of participants additionally rates the emotional intensity of each round. Among camera-qualified participants, facial-movement and webcam-gaze measures are recorded as secondary outcomes.

2026-08-30

2026-09-30

The primary outcome in both experiments is the incentivized certainty equivalent (CE) of a fixed 50–50 lottery paying $5.00 or $0.00, elicited through ten adaptive binary choices between the lottery and sure amounts on a $0.00–$5.00 grid. The primary CE measure is the posterior mean of each participant’s underlying valuation, estimated from the ten choices. The posterior standard deviation is retained as a diagnostic of elicitation precision.

The certainty equivalent is the sure amount of money that a participant values equally to the fixed lottery. Participants make ten binary choices between the lottery and a sure payment, adapting the staircase logic of Falk et al. (2023). Each offer depends on the participant's previous choices and is selected to be maximally informative about the participant's underlying point of indifference.
Because individual choices may contain response error, the primary CE is not read off any single choice. Following the DOSE approach of Chapman, Snowberg, Wang, and Camerer (2024), a logistic response-error model treats each choice as a noisy signal of the participant's underlying valuation. The primary CE is the posterior mean of that valuation after all ten choices. One of the ten choices is randomly selected and paid as chosen.
Three prespecified choice-based codings of the same ten choices are used as robustness outcomes: a model-free interval midpoint, a final-bracket coding, and a minimum-violation coding. Full definitions are provided in the pre-analysis plan.

References:
Chapman, J., E. Snowberg, S. W. Wang, and C. Camerer (2024). “Dynamically Optimized Sequential Experimentation (DOSE) for Estimating Economic Preference Parameters.” NBER Working Paper 33013.
Falk, A., A. Becker, T. Dohmen, D. Huffman, and U. Sunde (2023). “The Preference Survey Module: A Validated Instrument for Measuring Risk, Time, and Social Preferences.” Management Science 69(4), 1935–1950.

Primary hypotheses:
• H1: In the account-history experiment, within the own-account arm, cueing the focal −40% episode lowers the CE relative to cueing the focal +40% episode. Thus, the prespecified loss-minus-gain CE contrast is negative.

• H2: The focal loss-versus-gain cue contrast is attenuated in the reference-account arm relative to the own-account arm. The prespecified cue-by-account-arm interaction is positive, meaning that the loss-versus-gain contrast is less negative in the reference-account arm.

• H3: In the two-wave panel, reading the 2008 financial-crisis passage lowers the CE within respondent relative to the matched neutral passage. The prespecified effect is negative.

• H4: The crisis-cue effect is more negative for respondents born in 1983–1990 than for respondents born in 2001–2008. The prespecified cue-by-cohort interaction is negative.

In the account-history experiment, secondary outcomes cover the three-level loss/zero/gain cue benchmark; memory and internalization; negative affect and downside availability; background beliefs; open-text choice reasons; and camera-based facial-movement and gaze measures among camera-qualified participants. The camera-based measures include baseline-adjusted facial valence during encoding and the cue window, facial-pattern alignment between encoding and cue retrieval, first stable gaze orientation toward the lottery's $0 outcome, fixation-locked facial valence, and facial-pattern alignment between encoding and the fixation-locked response. In the crisis-cue panel, secondary outcomes cover recall source and quality; negative affect and downside availability; perceived financial vulnerability and recent financial context; economic beliefs; and open-text choice reasons.

These outcomes characterize the behavioral and memory processes accompanying the primary valuation effects. H5 is a prespecified secondary hypothesis in the randomly assigned rated half of the own-account loss/zero/gain sample: the association between signed cue content and the certainty equivalent is predicted to be stronger for episodes rated as unusually intense relative to the participant's other episodes. Because encoding intensity is measured rather than randomized, H5 lies outside the H1–H4 confirmatory family. The remaining tested secondary outcomes are organized into prespecified families with Benjamini–Hochberg adjustment within family, never pooled across experiments: five families in the account-history experiment (three-level benchmark; memory and internalization; emotions and downside availability; background beliefs; open-text reasons) and three in the crisis-cue panel (recall source and quality; one combined family covering emotions, downside availability, vulnerability, recent context, and economic beliefs; open-text reasons). The facial-movement and gaze family is adjusted using a design-consistent permutation step-down procedure and is analyzed separately from the confirmatory behavioral outcomes. Manipulation checks and diagnostics, including the encoding manipulation check and perceived final account balance, are reported descriptively. Full definitions and estimation details are provided in the pre-analysis plan.

• Crisis-cue panel: a two-wave within-person design with randomized passage order. Each main-panel respondent values the same lottery in both waves, roughly seven days apart, once after the 2008 financial-crisis passage and once after a length- and structure-matched neutral passage. A neutral–neutral calibration arm benchmarks week-to-week variation under matched content. The two birth cohorts contribute equally.

• Account-history experiment: a single-session design separating encoding from retrieval. Participants earn a starting balance, are randomized to an own or reference account, and observe a locked five-round percentage-return history. Within each focal matched set the percentage-return history is held fixed while account arm and cued episode vary, so the cue contrast holds the realized history fixed by design.

• The two experiments use independent samples. Participants may take part in only one experiment and can be randomized only once.

Not available

• Crisis-cue panel: Computerized individual-level randomization within birth cohort and recruitment batch. Participants are assigned to the main panel or calibration arm and then to passage sequence. The neutral passage used in each relevant design cell is randomized and balanced across cells.
• Account-history experiment: Computerized individual-level randomization within three slider-performance strata (0–16, 17–33, and 34–50 correct). Focal assignments are organized in four-person matched sets containing one participant in each account-by-cue cell; all four members receive the same percentage-return history. Nonfocal assignments are organized in matched own/reference pairs. Intensity-rating status is randomized at the matched-set or pair level.

The primary treatments in both experiments are assigned at the individual level. Passage sequence in the crisis-cue panel is randomized within cohort-by-recruitment-batch blocks. In the account-history experiment, focal assignments are organized in four-person matched sets and nonfocal assignments in two-person own/reference pairs. The secondary intensity-rating factor is assigned at the matched-set or pair level.

• Planned clusters for the intensity-rating factor: 600 randomization sets, comprising 400 focal matched sets and 200 nonfocal pairs.

Yes

600 randomization sets in the account-history experiment: 400 four-person focal matched sets and 200 two-person nonfocal pairs. The secondary intensity-rating factor is assigned at the set or pair level, while the primary account-by-cue and own/reference treatments are assigned at the individual level. The crisis-cue panel randomizes all participants individually.

3,550 unique completers: 1,550 in the two-wave crisis-cue panel and 2,000 in the account-history experiment. These participants yield 5,100 certainty-equivalent measurements in total. Realized counts may modestly exceed these targets because all participants from already opened recruitment batches or allocation sets are retained.

• Crisis-cue panel: 650 crisis-to-neutral, 650 neutral-to-crisis, and 250 neutral-to-neutral participants, equally divided across the two birth cohorts.

• Account-history experiment: 400 participants in each of the four focal cells (own-loss, own-gain, reference-loss, reference-gain), 100 per account arm in the zero-return category, and 50 per account arm and magnitude in the 10% and 30% filler categories. The intensity-rating factor is split evenly, with 1,000 participants assigned to provide round-level intensity ratings and 1,000 assigned to receive no rating prompt.

See the pre-analysis plan.