Method — civic lottery methods
Updated: 2026-10-05. Stream C. Connects institutional choices to corpus evidence; no CA-R3 software selection.
Sampling frame
| Frame | Properties | Corpus instance |
|---|---|---|
| Electoral register | covers registered voters; misses non-registered (younger, mobile, lower-income skew) | Ireland 2016–18 initially; Paris electoral lists + volunteers (SRC-0017, SRC-0025) |
| Address/household database (postal) | near-universal dwelling coverage; person-level draw requires in-household selection | CA UK Royal Mail database, 30,000 invitations (SRC-0038); Ireland post-2019 postal lottery (SRC-0017) |
| National population register | person-level, high coverage | Ostbelgien, Brussels draws (SRC-0021, SRC-0031) |
| Membership/self-enrolled register | only members | party-internal (Your Party MOC, SRC-0066) |
| Open application | pure self-selection — not a lottery frame in the statistical sense | HotP open entry, weighted (SRC-0068) |
| Nomination | social-network reach, unclear population | XR/open models (SRC-0070) |
Method note: frame choice is the first representativeness decision; coverage gaps cannot be repaired downstream (SRC-0038, SRC-0120).
Stage-one invitation
- Volume/response: CA UK sent 30,000 for ~110 seats; response rates are low and skewed — volume is a design parameter (SRC-0038, SRC-0123).
- Weighting at invitation: deprivation-weighted oversampling (CA UK) compensates known non-response gradients; equal raw invitation probability vs corrected — a fairness choice (SRC-0038).
- Outreach: postal letters remain the least-excluding channel; digital-first invites compound self-selection (SRC-0123, SRC-0120).
- Pure random vs weighted: pure maximises equal probability; weighted improves achieved representativeness — the recurring tension (see
stratification-and-fairness.md).
Stage-two selection
- Stratified random sampling / quotas: define strata cells, draw to fill quotas — the corpus norm (SRC-0038, SRC-0048, SRC-0021).
- Optimisation (leximin/maximin): compute a distribution over quota-compliant panels maximising worst-case inclusion probabilities, then one simple draw (Panelot method, SRC-0078, SRC-0080; OpenDLP algorithms SRC-0079). The optimisation determines probabilities; the draw is still a lottery.
- Simple random within pool: rare for final panels; common for reserves.
- Attitudinal variables: usable in stage two (CA UK climate-attitude stratum) — see
stratification-and-fairness.mdfor the legitimacy debate (SRC-0038).
Replacement and attrition
- Reserves: selected alongside the panel, same stratum; replacement on dropout before/at start (OpenDLP replacement flow, SRC-0084ish — see
opendlp.md). - Post-selection balancing: late attrition forces re-stratification or accepting drift — documented as manual practice (SRC-0123).
- Reproducibility threat: replacement decisions made ad hoc break audit claims — reserves should be drawn in the same audited draw, not selected later (method requirement evidenced in practice guidance, SRC-0123).
Auditability
- Published algorithm (OpenDLP algorithms doc, sortition-algorithms library; CA UK published selection code) — SRC-0079, SRC-0038.
- Seed/witness model: deterministic draw + public seed + independent witness = reproducibility claim; OpenDLP’s seed/witness ceremony is not fully documented (open question,
opendlp.md). - Panelot: leximin computation + simple random pick; client-side anonymisation means the operator never sees identities pre-draw (SRC-0080).
- Algorithm vs operator discretion: auditability is a claim about the whole pipeline (frame → invitations → funnel → algorithm → replacement), not the algorithm alone (SRC-0123).
Cicada relevance
- The lottery is a five-stage method (frame → invitation → response → stratified draw → replacement/audit); Cicada fields should capture each stage, since failure can occur at any single one.