The science
Combined estimate
How we turn separate studies into one number, and why you should treat it as a model.
- Headline
- About 1.9×
- Range
- +55% to +125%
- Status
- Modeled, not measured
Method
- Take the effect size for each mechanism that has evidence.
- Do not add them: they overlap, so we treat them as diminishing returns.
- Shrink each for publication bias and real-world conditions.
- Land on a combined effect of about d = 0.7 to 1.0.
- Convert that to a success rate for a given baseline.
What we counted
Each mechanism, the best evidence we found, and how we treated it.
| Mechanism | Best evidence | Treatment |
|---|---|---|
| Roadmap and proofs | d ≈ 0.65 | Counted, discounted |
| Needle and monitoring | d = 0.40 | Counted |
| Compass | g ≈ 0.28–0.34 | Counted |
| Coaches | g = 0.43–0.74 | Counted, heavily discounted |
| Stake and pot | Large for takers, +3 points when offered | Counted, modest |
| Gamification | g ≈ 0.25–0.36 (education) | Counted, small |
| Peers | SMD ≈ 0.35 (health) | Counted, small |
| Witnesses | Inside monitoring | Not added again |
| Journal, identity, heatmap, share cards | No direct tests | Not counted |
Results
These effects are shifts in outcomes, not percentage points, so the conversion depends on the baseline.
| Baseline success | With Road Mates | Relative |
|---|---|---|
| 30% | about 57–68% | +90% to +125% |
| 20% | about 44–56% | about 2.2× to 2.8× |
The headline
About 1.9× as likely to reach your goal (+90%, range +55% to +125%), assuming a 30% baseline. We picked a number near the low end on purpose. The earlier estimate was 1.8×. Adding gamification and peers raised it a little, because most of what they do overlaps with mechanisms we already counted.
Caveats
- No study has tested this exact bundle. The number is built from studies of its parts.
- People who join choose to, so some of the lift may be self-selection.
- Staking effects are likely smaller than they look in the trials.
- Gamification and peer evidence come from education and health, not goal pursuit.
- Several linked sources are summaries of the original papers.
- We will replace this estimate with measured results from our own users.
Sources
- Gollwitzer & Sheeran 2006 ↗
Meta-analysis of 94 tests of if-then planning (secondary copy).
- Harkin et al. 2016 ↗
Meta-analysis of 138 randomized studies on monitoring goal progress.