Score us
Bring a study you already ran.
Bring the questionnaire. Keep the result hidden. We field the same items and lock our predictions. You do the scoring.
- Bring
- Question stems and answer options
- Audience cells and field dates
- Your metric, frozen before we run
Bring a question, or a study you already ran. On the call we can field the same items on the panel and you can score the result yourself.
Put “past study” in the Calendly notes, and email the questions plus the human result to jeffrey@lewsearch.com before the call. Hide the answers if you want. We will not look until you score us.
Bring: Question stems and answer options. Audience definition or cells (age, gender, region, party, income). Field dates. The human result, kept on your side until we lock.
You leave with: A locked prediction file in your codebook and, once you score it, a per-question miss in percentage points that you computed.
What we see
Stems, options, cells, field dates.
What we don't see
Your human results, until our predictions are locked. Score locally and share nothing, or share aggregate deltas if you want them added, anonymized, to a future registry. Opt-in only.
≈10 pp
as shipped · 1,521 polls · 50 states
7.47 pp
calibrated benchmark · 404 non-electoral Q
9.97 pp
strict held-out
Calendly not loading? Email jeffrey@lewsearch.com or call 440-283-9270.
Questions
- What if some items don't simulate well?
- We tell you which ones we'd decline before we run. Election margins, for example, we don't claim.
- Will you publish my result?
- No. The misses we publish are on public polls. Customer replications stay private unless you opt in later.
- Is this the same as sample boosting?
- Same mechanism. Sample boosting adds rows to a study you fielded; scoring us first is how you decide whether to trust the rows.
- How long does it take?
- A single instrument, days. A tracker is Concierge or Enterprise work and gets scoped on the call.