Forecast
Write down a dated, falsifiable prediction now, so it can be scored honestly later.
Variants
- default
- Estimate how long each of today's tasks will take
Data fields
The fields a spec's page can set for this component (see switchback show forecast --json):
questionresolve_bytasks
Grounding
Helps: when a prediction will resolve before the next return; a dated written probability stops hindsight rewriting what you believed, a moderate and well-replicated bias, but learning calibration from a handful of scored forecasts is untested
Backfires: when forecasts are judged by how they turned out rather than by the probability given, a robust bias that affects even people who say outcomes should not count, when too few forecasts are scored to learn from, or when the base-rate line is written and then ignored as not applying to this case
- CMaking explicit probability forecasts and then receiving outcome feedback or scores improves calibration over successive rounds.Mellers et al. (2014), Aiyer et al. (2023)
- AAfter an outcome is known, people misremember their earlier probability estimates as closer to the outcome than they were. A dated, written prediction removes that memory distortion when the forecast is scored.Guilbault, Bryant, Brockway & Posavac (2004), Fischhoff (1975), Sanna, Schwarz & Stocker (2002)
- CStarting a prediction from the base rate for its reference class ("how often does this kind of thing happen?") improves forecast accuracy compared with starting from the specifics of the case.Buehler, Griffin & Ross (1994), Mellers et al. (2014), Flyvbjerg (2013), Kahneman & Tversky (1982)
- BA number written or seen earlier pulls later numeric judgements towards it, so a gut score written *before* criteria scoring cannot be anchored by the total.Klein et al. (2014), Jacowitz & Kahneman (1995), Epley & Gilovich (2001), Tversky & Kahneman (1974)