# CATI Quality Control Plan — [Agency name]

> Fill-in template. Replace every [bracketed field], delete guidance notes (in
> blockquotes like this one), and version the result in your document system.
> Companion guide: https://vericati.com/blog/cati-quality-control-guide
> Template version: 1.0 (June 2026) — provided by VeriCATI (vericati.com), no attribution required.

- **Document owner:** [name, role]
- **Applies to:** [all CATI studies / studies over N completes / client X]
- **Last reviewed:** [date] · **Next review due:** [date]

---

## 1. Scope and roles

| QC layer | Owner | Backup |
| --- | --- | --- |
| Pre-field controls | [name] | [name] |
| Live monitoring | [floor supervisor] | [name] |
| Post-hoc audio review | [QC lead] | [name] |
| Back-checks | [name] | [name] |
| Data-file checks | [data manager] | [name] |

## 2. Pre-field controls

- [ ] Questionnaire programming tested end-to-end, including skip logic, by [role]
- [ ] Pilot batch of [20–30] recorded interviews audited with the production rubric (§6) before full launch
- [ ] Interviewer briefing covers: protocol, consent/recording disclosure script, and **the fact that work is recorded and validated** (a documented deterrent — AAPOR 2003)
- [ ] Protocol ambiguities found in pilot resolved and re-briefed before scale-up

## 3. Live monitoring (during fieldwork)

| Parameter | Value | Reference floor |
| --- | --- | --- |
| Share of calls live-monitored | [ %] | IQCS 2023 allows 5% for telephone-centre validation during fieldwork |
| Who monitors | [role] | — |
| Same-shift correction expected within | [minutes/hours] | — |

## 4. Post-hoc audio review

| Parameter | Value | Reference floor |
| --- | --- | --- |
| Share of completes reviewed | [ %] | IQCS 2023 minimum validation: 10% of fieldwork |
| Minimum reviews per active interviewer per week | [≥2] | Avoids proportional-sampling blind spots |
| New interviewer: share of first [5–10] completes reviewed | [100%] | Habits set in week one |
| Discretionary capacity allocated by | [paradata signals: length outliers, completes/hour, refusal shifts] | — |
| **Maximum review lag (interview → finding reaches interviewer)** | [48h / same-day] | The lag is the damage multiplier — measure it (§7) |

## 5. Back-checks (re-contact validation)

| Parameter | Value | Reference |
| --- | --- | --- |
| Share of completes back-checked | [ %] | DIME: 10–20%; J-PAL: ≥10% |
| Coverage rule | [every interviewer ≥1 per study; new interviewers and paradata outliers oversampled] | J-PAL: every interviewer at least once |
| Script | [link to back-check script] | Keep under 3 minutes: interview occurred + length, 2–3 re-asked factual items, incentive delivered, open conduct question |
| Discrepancy handling | [logged where; reconciled by whom; escalation trigger] | Document and reconcile every discrepancy |

## 6. Error-severity rubric

> Re-weight to your protocols; keep the structure: one critical finding **or**
> accumulation to the threshold cancels the interview. Pattern adapted from
> LAPOP's QuAC (cancel at 20+; falsification = 100).

**Cancel/replace threshold: [20] points per interview.**

| Finding | Points |
| --- | --- |
| Interview falsified, in whole or part (triggers §8 protocol) | [100] |
| Ineligible respondent / screener not properly administered | [25] |
| Mandatory question skipped but an answer recorded | [20] |
| Recorded answer contradicts what the respondent said | [20] |
| Consent or recording disclosure omitted | [20] |
| Leading probe / suggesting an answer | [10] |
| Scale or answer options not read as scripted | [8] |
| Vague answer accepted without clarification | [6] |
| Wording deviation that preserves meaning | [4] |
| Rushed delivery, talking over the respondent | [2] |
| [Protocol-specific item] | [ ] |

- Reviewers may reclassify or dismiss findings with a logged reason.
- Inter-reviewer calibration: [monthly]; [2–3] shared calls scored independently; drift discussed at [meeting].

## 7. KPIs and review cadence

> Reviewed [weekly] by [role], per study and per interviewer. Require a minimum
> of [N] reviewed calls before an interviewer-level rate triggers action.

| KPI | Definition | Target / alert |
| --- | --- | --- |
| Time-to-feedback | Median hours, interview end → interviewer receives finding | [< 48h] |
| Critical-error rate | Critical findings per 100 completes | [trend; alert at X] |
| Repeat-error rate | Same finding type recurring after feedback | [alert at X] |
| Review coverage & fairness | % completes reviewed; min reviews per interviewer | [per §4] |
| Back-check discrepancy rate | % back-checks with material discrepancy | [alert at X] |
| Reviewer dismissal rate | % of flags dismissed on review | [investigate above X] |

## 8. Feedback loop and escalation

- Findings delivered to interviewers by [role], via [channel], within the §4 maximum lag
- Coaching threshold: [e.g., any critical finding, or warning pattern across N calls]
- **Falsification protocol:** flag → [second reviewer confirms on audio] → [supervisor + HR involved] → [interviewer's recent work re-reviewed at 100%] → [affected completes replaced; client informed per contract]. A falsification finding never reaches an interviewer's file on the strength of a flag alone.
- Repeated critical findings: [escalation path]

## 9. Client-facing documentation checklist

- [ ] This plan (current version) available on request
- [ ] Per-study QC summary: rates achieved, findings by severity, actions taken
- [ ] Validation evidence retained for [N months]: review logs, verdicts, back-check records
- [ ] Standard responses prepared for: "What share of interviews do you validate?" / "What is your falsification procedure?" / "Which standard does your QC follow?"