# Habit Task — Version 3 (Planned)

## Overview

Redesign motivated by the persistent flat null at v2 n=30. Advisor reasoning: the within-subjects design may be too noisy to detect the predicted social-habit effect, and the difference in outcome-delay timing between social and individual blocks (subjects wait for other players in social blocks but not in individual blocks) is a confound that could account for learning differences. Version 3 addresses both concerns.

## Two Variants Being Prepared

### 3a: Within-Subjects (Modified)
- Within-subjects, alternating social and individual blocks (as in v2)
- **Individual blocks still run as multiplayer sessions** — subjects genuinely wait for all 3 players to respond before seeing their own outcome, matching the temporal experience of social blocks
- **No social outcomes shown in individual blocks** — subjects see only their own reward, never the other players' choices or avatars
- **Cover story** explains the wait: "we're waiting for all 3 players to choose before showing your outcome"
- **Explicit block-transition instruction screens** clearly separate social vs. individual contexts (replacing the implicit context cue that was removed when background colors were dropped in v2)

### 3b: Between-Subjects
- Two groups, n=50 each, total n=100
- One group runs all 12 blocks in social condition (192 social trials)
- Other group runs all 12 blocks in individual condition (192 individual trials, still multiplayer with hidden social info)
- Same yoked-wait mechanism as 3a (real multiplayer sessions, social outcomes hidden in individual sessions)
- Same cover story explaining the wait

## Changes from Version 2

1. **Outcome delay matched between conditions.** v2 had structurally different choice-to-outcome delays in social vs. individual blocks (social waited for others, individual did not). This may have produced learning differences attributable to reinforcement timing rather than social information per se. v3 matches delays naturally by running all conditions as multiplayer sessions.
2. **Explicit block-transition cues** (within-subjects variant). Instruction screens at block boundaries clearly mark context shifts. Replaces the implicit context separation that was removed in v2 along with the background colors.
3. **Between-subjects variant added** to eliminate within-subjects noise from order effects, fatigue, and individual differences in habit susceptibility.

## Unchanged from Version 2

- Silent reward reversal at training-to-test transition (no instructions or notification)
- 192 trials, 12 blocks × 16 trials per block
- Training (blocks 1-6) followed by test (blocks 7-12)
- Probe DV: P(choose originally-high shape)
- Social blocks: avatars + reveal of others' choices on training probes; avatars only on test probes

## Methodological Note on the Individual Condition

In both 3a and 3b, the "individual" condition is technically run within a multiplayer session — three real subjects participate, the session waits for all three to respond before advancing, but the individual-condition subject never sees any social information. This preserves the temporal experience of the social condition while removing the informational manipulation.

Analogous to the private-response control in Asch's conformity experiments: the social context is preserved, but social information at the response phase is removed. Cleanly isolates the informational variable.

## Hypotheses

Same as v2: social-context S-R associations should persist more strongly through the silent reversal than individual-context S-R associations.

Prediction with matched outcome delays: any observed social-vs-individual habit difference can now be attributed to social information rather than to timing differences in reward delivery.

## Expected Outcome

If the prior null was driven by temporal confounds or within-subjects noise, the v3 design should detect the predicted social-habit enhancement. If the null persists at v3 with these corrections in place, that constrains the theoretical claim — habit-via-imitation may not be differentially stronger than habit-via-reward at the magnitudes the design can detect.