Field Quality Control: Catching Fabrication in Uzbekistan

The most expensive mistake in field research is discovering a problem after fieldwork has already ended. By then the interviewers have scattered to other projects, the respondents in a distant mahalla — the neighborhood community — can't be brought back, and the budget is spent. When your team is stretched from Tashkent to the Fergana Valley and Karakalpakstan, speaks four languages, and works offline in places where you physically cannot stand behind anyone, quality control stops being a final checkbox and becomes a daily operation. That's why it has to run alongside collection, not after it.

This article is about managing a dispersed team and catching a fabricated interview while the field is still open. Not theory — what actually happens in the Uzbek field.

Why quality control in Uzbekistan is geography management

Field research in Uzbekistan is almost always geographically dispersed. One supervisor runs interviewers in the capital, another in the dense, traditional Fergana Valley, a third coordinates a trip to Karakalpakstan, where distances run into the hundreds of kilometers and connectivity dies in the desert. You cannot supervise a team like that by eye: you don't see who actually knocked on a door and who filled in a questionnaire at a teahouse.

Hence the first principle: quality control isn't distrust — it's a tool for managing distance. Digital collection (CAPI) gives you what paper never could — a stream of metadata on every interview in real time. But metadata is only trustworthy in offline logic: the interviewer's phone records GPS, time, and duration locally, then syncs them at the first connection. Control is built on that trace.

If you're still working with paper questionnaires, it's worth understanding why offline data collection changes the rules of control: on paper, fabrication is nearly impossible to catch in time because there's neither a GPS pin nor a timestamp.

Real-time monitoring: a dashboard instead of an end-of-week report

The first line of defense is live monitoring. While collection is happening, you should see, per interviewer: how many questionnaires gathered today, at what pace, at which points on the map, with what average duration. This isn't an end-of-week report — it's a dashboard you open in the morning and at midday.

What to look for:

• Anomalous pace. One interviewer "closing" three times as many questionnaires as everyone else in the same mahalla. Maybe a genius, but more likely not walking the doors.
• Sharp productivity jumps for one person: four questionnaires yesterday, twenty today.
• Time clustering: ten interviews back to back within an hour — you can't physically cover people that fast.

Live monitoring is also valuable because it sets the standard from day one. When an interviewer knows the supervisor sees their map and pace that same evening, the temptation to cut a corner drops sharply.

GPS verification against Uzbek geography

GPS is the most powerful control tool, but in Uzbek conditions you can't apply it naively. Several realities break the "coordinates matched, so it's honest" logic:

Addressing is weak and ambiguous. Many mahallas have no conventional house numbering; in old quarters houses stand densely, wall to wall, and a 15–30 meter GPS error easily "throws" a pin from one yard to the next. So a rigid "the pin must match the address to the meter" rule produces false alarms.

Connectivity drops. In remote mahallas, the mountains of Surkhandarya, the Karakalpak desert, there may be no network at the doorstep itself. The GPS coordinate must be captured offline at the moment of the interview and reconciled later, at sync — not relied on for an online fix in the moment.

What you actually catch. Not micro-displacement, but gross mismatch: a pin at the café next to the interviewer's home instead of the assigned mahalla, a cluster of fifteen interviews at a single point, a coordinate in the neighboring district. Compare GPS against the assigned area, not the exact address, and flag only clear outliers.

Time and duration work alongside GPS:

• Duration can't be several times shorter than the questionnaire's average. If the median interview is 28 minutes and one comes in at 6, that's almost certainly a click-through.
• Time of day should be plausible: a run of interviews at 2 a.m. or in the peak of 40 °C heat, when no normal person opens the door, is worth a look.
• The gap between one interviewer's questionnaires should leave time to walk between houses.

Compare the GPS pin against the assigned area, not the exact address to the meter. In a dense mahalla an honest interview and a fabrication differ by tens of meters — but a faker usually misses by kilometers.

Suspicious patterns in the answers themselves

Metadata tells you where and when; the answers tell you how the questionnaire was filled in. Fabricated interviews often give themselves away in the structure of the data:

• Straight-lining: the same score across an entire battery of scales. A real respondent hesitates; an inventor puts "4-4-4-4-4."
• Too-clean data: a complete absence of missing values, a zero share of "hard to say." Live fieldwork always yields "don't know" and partially completed open-ended answers.
• Templated open answers: identical or near-identical wording in open-ends across different "respondents" of the same interviewer.
• Impossible combinations: filters that should have closed a block, but the block is filled; contradictions between linked questions.

Some of this is caught automatically (flags on straight-lining and a zero missing-value share), but the final call belongs to a human. Data quality is set back at the questionnaire design stage: well-thought-out check questions and logic filters make fabrication more visible.

Callbacks and back-checks: verifying the interview happened at all

The most reliable way to confirm an interview took place is to contact the respondent again. Uzbekistan is favorably placed here: mobile penetration is very high, there are more SIM cards than people, and most households can be reached by phone. But there are nuances that are easy to miss.

What a callback verifies:

1. The fact of the visit — the interviewer actually came.
2. The respondent's identity — the person who answered is the one declared, not a neighbor or a relative standing in.
3. Key answers — a few anchor questions (age, household composition, two or three substantive items) match what was recorded.
4. Who was actually home — whether the within-household respondent selection procedure was followed.

The Uzbek specifics that break a naive callback. Because of labour migration (millions work in Russia and Kazakhstan, especially from the Fergana Valley), many households are missing their adult men. If your sample required a male head of household and, on the callback, his wife answers and says he has been in Moscow for six months — that's not necessarily a sign of fabrication, but a sign to investigate: who was actually interviewed. Gender norms add a layer: in some households a woman won't speak to a strange man on the phone, and a back-check needs a female interviewer — exactly as the interview itself does.

How much to verify. Practice is calling back 10–15% of questionnaires, weighting toward suspicious metadata (short duration, a GPS outlier) and new interviewers. As trust in a specific person grows you can lower the share, but never zero it out.

Managing a dispersed, multilingual team

Quality control begins before the first call — with placing people correctly. A mistake in team composition produces not fabrication but quiet, low-grade error that's harder to catch.

Match the interviewer's language to the respondent's. Russian dominates in Tashkent and among urban and older respondents; Uzbek is stronger in the Fergana Valley and the rural south. In Samarkand and Bukhara a meaningful share is more comfortable in Tajik; in Karakalpakstan you need Karakalpak — a separate Turkic language, and the questionnaire there must exist in a Karakalpak version. A language mismatch isn't always visible in the data, but it systematically degrades answers.

Match gender to topic and respondent. On sensitive topics (health, family) and to reach women in conservative households, female interviewers are often essential. This is a question not only of ethics but of data quality.

Structure supervision around geography. One supervisor can't run Tashkent, the Fergana Valley, and Karakalpakstan with equal closeness. It's sensible to assign a supervisor per region and distribute the back-check load so that new interviewers in distant locations are checked more often. Remember the cluster effect: in a cluster sample, a single fabricated area in one mahalla can distort estimates for an entire region, because it represents dozens of real households. So an isolated anomaly in Karakalpakstan is more dangerous than the same one in Tashkent, where there are more clusters and each carries less weight. If clusters and their weights are still abstract for you, see how sampling methods work in the Uzbek context.

Mahalla access: when "too many questionnaires" means "cut a corner"

Access to households in Uzbekistan runs through the mahalla, whose gatekeeper is the chairman (raisi). A support letter from the hokimiyat (district administration) or the client, and a courtesy visit to the raisi, open doors; their absence leads to refusals and even calls to the police. But the same mechanics create a specific quality risk.

An interviewer "closing" an implausible number of questionnaires in a mahalla may not have walked the houses but cut a corner through a single contact — taking a list from the raisi and filling in questionnaires from his account, or from a few familiar families. Formally there was a visit, the GPS is in the right mahalla, but this isn't a household sample — it's one informant's survey replicated across forms. This kind of error is caught by a combination of signals: high pace + tight GPS clustering at one point + uniform answers + a back-check where the "respondent" doesn't remember the interview. We cover the subtleties of working with local authorities and the mahalla separately — field logistics and access through the hokimiyat and mahalla.

Season and pace: where slow is honest, not lazy

Before reading a low pace as faking, account for the calendar. The Uzbek field lives by seasons, and a slow day can be entirely legitimate:

• Ramadan (Ramazon): daytime energy is low and evenings are taken by iftar — interview windows shrink.
• 40 °C heat in the south (Surkhandarya, Kashkadarya) and the Karakalpak desert makes a midday round impossible and unsafe.
• The cotton harvest and spring planting pull rural people into the fields — fewer people home in daytime.
• Friday (midday prayers), wedding (toʻy) season, and school exams also shift availability.

The control takeaway: set fair pace expectations by region and season. The norm for Tashkent on an ordinary day and the norm for a mountain mahalla during Ramadan are different numbers. Otherwise you either pressure honest interviewers — pushing them toward the temptation to fabricate — or you miss a genuine shortfall.

A no-blame culture that still stops fabrication

The most delicate part of field control is separating an honest mistake from deceit without turning the team into suspects. If every returned questionnaire sounds like an accusation, honest interviewers start defending themselves, hiding problems, and — paradoxically — drifting toward the very corner-cutting you fear. The goal isn't to "catch the guilty party," it's to protect the whole study.

Separate slow from dishonest. A low pace is not evidence in itself. An honest day during Ramadan, a round in 40 °C heat in Surkhandarya, a mahalla where half the men are in Moscow and almost no one is home by day — these are legitimately slow, not lazy, and certainly not fabricated. Fabrication is given away not by low output but by a combination of signals: high pace plus a GPS cluster at one point plus uniform answers plus a failed back-check. One signal is a reason to look; four together are a reason to act.

Fast feedback instead of a delayed verdict. In the first field days, work through disputed questionnaires quietly and concretely: show the interviewer the specific flag, ask what happened, let them fix it. Often an "anomaly" has an innocent explanation — a mixed-up area, a dead phone, an honest refusal logged the wrong way. That conversation both teaches and retains: the person sees that you're watching, but judging fairly.

Retraining or removal — by evidence, not by mood. A newcomer's single error is grounds for retraining. Proven fabrication is grounds for immediate removal and a review of all their questionnaires. The line between the two runs along the facts, and holding it steady is part of the culture:

• Retrain first when there's no sign of deceit, only procedural slips.
• Remove and audit the entire output when a back-check confirms an invented interview.
• Protect the cluster before the person: one fabricated mahalla distorts estimates for an entire region, so a suspect cluster is pulled from the analysis and re-measured even while the conversation with the interviewer is still under way.

Such a culture, paradoxically, curbs deceit more than fear does: people who are trusted and helped quickly cut fewer corners, while the few who do fabricate are identified by evidence, not by suspicion.

Close the feedback loop — especially in the first days

Control without feedback is useless. The cycle is simple: found a problem → returned the specific questionnaire to the interviewer with a specific reason → recorded the resolution → checked the fix. Speed matters: an error caught and discussed on the first or second day of fieldwork re-trains the interviewer; the same error found two weeks later means you're reworking hundreds of questionnaires.

That's why the first days of any project are the most important for control. Check newcomers almost wall to wall, give feedback the same evening, and the quality standard locks in for the whole project. This is especially valuable for a dispersed team: an interviewer in Nukus you've never met in person learns from how fast and how specifically you react to their first questionnaires.

You can set up this loop — live monitoring, GPS and timing, answer flags, a back-check queue — directly in the AISurvey builder: metadata is collected offline on the interviewer's side and pulled to you at sync, so fieldwork and control live in one tool.

A short field-QC checklist

1. Open the dashboard morning and midday: pace, map, average duration per person.
2. Compare GPS against the assigned area, not the exact address; flag gross outliers.
3. Flag short duration, straight-lining, and a zero missing-value share.
4. Call back 10–15%, weighting toward suspicious questionnaires and newcomers; verify identity and who was home.
5. Account for season and region when judging pace.
6. Return the specific questionnaire with a reason and check the fix — the same day.