Fieldwork QC is a live management process. The goal is to detect deviations early, investigate them fairly and take documented corrective action before poor-quality data spreads through the sample.

Quality control should be designed before launch

Fieldwork quality cannot be improvised after the first suspicious interviews appear. The research plan should define expected interview duration, location rules, back-check rates, supervisor responsibilities, permitted replacements, escalation thresholds and the evidence required to reject an interview.

These rules protect both the dataset and the field team. Enumerators should know the standards on which their work will be reviewed.

Certify interviewers rather than only training them

Training should cover study objectives, respondent selection, consent, questionnaire logic, difficult concepts, neutral interviewing, device use and field protocols. Role plays and mock interviews reveal problems that classroom explanation misses.

Certification can require interviewers to demonstrate correct respondent selection and questionnaire administration before they begin live work. Retraining should be available when errors are correctable.

Use paradata to see how interviews were conducted

Paradata such as timestamps, duration, GPS, device ID, call attempts and edit history can reveal patterns that responses alone cannot. Very short interviews, repeated coordinates, impossible travel, identical timing or unusual skip patterns should trigger review rather than automatic accusation.

Expected ranges can differ by questionnaire, region and respondent type, so flags need context.

Combine automated checks with human verification

Automated dashboards can identify duplicates, outliers, straight-lining, inconsistent responses and enumerator-level anomalies. Supervisors can then use call-backs, back-checks, spot checks or accompaniment to verify what happened.

Audio review may be powerful where it is lawful, ethically approved and covered by informed consent. It should be targeted and securely managed, not collected casually.

Monitor interviewer effects

Compare response distributions across interviewers. One interviewer may report unusually high eligibility, unusually low refusal, repeated response patterns or category distributions that differ sharply from peers working in similar areas. These can be signs of misunderstanding, poor probing or fabrication.

Investigation should consider assignment differences before conclusions are drawn. A fieldworker assigned to a different type of location may legitimately produce different data.

Create a corrective-action loop

Every significant flag should lead to a documented outcome: accepted after review, corrected, recontacted, rejected, replaced, retrained or escalated. Daily feedback prevents the same error from continuing for several days.

The final fieldwork report should summarise quality procedures, issues encountered, interviews removed or replaced and any limitations that remain. High-quality data is not data that never triggered a warning; it is data produced by a process capable of detecting and resolving warnings.

Set measurable quality standards before launch

A quality plan should define what acceptable fieldwork looks like. This can include interview duration ranges, GPS expectations, maximum missingness, back-check rates, allowable quota deviations, audio-review procedures where appropriate and thresholds for supervisor observation.

Standards should be chosen because they are relevant to the method, not because a platform happens to produce the metric. The team should also define what happens when a threshold is breached: investigate, retrain, pause, reject or replace. Predetermined rules make quality decisions more consistent and defensible.

Monitor interviewer behaviour, not only completed cases

Completion counts are a poor measure of field quality on their own. Researchers should examine patterns by interviewer, such as unusually short interviews, repeated answer sequences, low refusal rates, repeated coordinates, excessive use of 'other' responses or improbable productivity.

These indicators are signals rather than automatic proof of misconduct. Supervisors need to review the circumstances, compare with field notes and conduct targeted verification. The combination of paradata and human investigation is more reliable than either one alone.

Control questionnaire and sample changes

Many data-quality problems arise from uncontrolled changes during fieldwork. A supervisor may reinterpret an eligibility criterion, a local team may add a replacement rule or a questionnaire edit may be sent informally to only some interviewers.

Version control should therefore cover scripts, translations, sample files and operational instructions. Any change should identify who approved it, when it took effect and which cases were affected. This allows analysts to assess whether the change introduced bias or needs to be accounted for in cleaning and reporting.

Use layered validation

No single quality check is sufficient. Strong validation combines preventive controls, live monitoring and post-field review. Preventive controls include training, programmed skips and range checks.

Live monitoring uses supervisor observation, dashboards, spot checks and back-checks to detect problems while they can still be corrected. Post-field validation includes duplicate detection, consistency checks, open-end review, review of paradata and reconciliation against the sample plan. Layers are especially valuable because sophisticated errors may pass one control while being detected by another.

Document exclusions and protect the analysis

When interviews are removed, the project should record why, who made the decision and whether replacement was permitted. Analysts need to know how exclusions affect sample composition and whether particular locations or respondent groups are disproportionately lost. A clean dataset should be accompanied by a data-quality or fieldwork note summarising the validation process and remaining limitations. Transparency does not weaken the research; it helps users understand the strength of the evidence and prevents questionable cases from quietly influencing conclusions.

Create an incident and fraud-response protocol

Quality systems should distinguish ordinary interviewer mistakes from patterns that suggest fabrication or deliberate protocol violations. The project can define how suspected cases are investigated, who has authority to suspend an interviewer, what related interviews are reviewed and how evidence is documented.

Decisions should rely on multiple signals where possible rather than one unusual metric. Field teams also need a safe way to report pressure, misconduct or operational problems without concealing them to protect productivity targets. A transparent incident process protects both respondents and credible field staff.

Make quality ownership explicit

Data quality improves when responsibility is clear at every level. Interviewers are responsible for following the protocol, supervisors for observation and immediate correction, central quality teams for reviewing patterns, and project leadership for deciding on exclusions or major remedial action. Clients should also know how and when serious issues will be reported. This division of responsibility prevents quality from becoming an undefined task that everyone assumes someone else is handling. A short quality matrix can list each control, its owner, review frequency, threshold and required response. That structure makes the system easier to manage consistently across interviewers, locations and countries.

How Surveysphere Africa can support

Surveysphere Africa uses centralised field dashboards, supervisor verification, back-checks and documented QC protocols to protect data quality throughout fieldwork.