Free tool

Survey data quality check

The dataset is back from the field, and before you analyse it you want to know how many speeders, straightliners, gibberish texts and copy-paste answers it holds. Drop the file here: you get the matching GDQ codes per case, with the reason, and a protocol for the method report.

Free
No upload
No signup
GDQ code frame

Drop your dataset here (.csv, .tsv, .txt, .sav) or Choose a file The file stays in your browser. Nothing is uploaded.

The free survey data quality check scans a survey dataset (CSV, TSV or SPSS .sav) right in your browser for speeders, excessively long interviews, straightlining, gibberish and duplicates in open ends, and identical answer patterns. Every hit carries the code of the GDQ Feedback Loop code frame. The tool flags and explains; it removes nothing, and the file is never uploaded.

Seven checks, coded with the GDQ Feedback Loop

In 2026 the Global Data Quality initiative (GDQ) published a shared code frame of 18 reasons for removing a participant from a study. The check applies the seven that can be computed from the dataset alone. Rules and defaults are the same as in DataLion’s survey monitor, so a protocol from the tool and one from the product mean the same thing.

Every threshold can be adjusted before the run. GDQ recommends calibrating the speeder cut-off on soft-launch behaviour and fixing it only then.

GDQ code	Check	Rule (default)
6 Speeding/Racing	Speeder	Interview duration below 33 % of the median, only once 10 cases carry a duration
7 Excessive interview length	Long interview	Duration above 3 times the median; weak evidence per GDQ
8 Straight Lining/Flat Lining	Straightliner	In grids with at least 4 items the same answer in every answered row; flagged from 50 % of a case’s grids
13 Open End - Poor Quality	Open-end gibberish	Repeated character (aaaa), keyboard rows (asdf, qwer, zxcv), consonants only from 5 letters. Short answers alone never count.
14 Open End - Duplicate	Open-end duplicate	Same text (from 15 characters) in two open ends of one case, or identical text (from 25 characters) to another case in the same question
16 Duplicate Responses In Closed Questions	Identical answer pattern	The same answer vector across the closed questions as another case, from 8 closed answers
4 Participant Duplication	Duplicate IP fingerprint	Two cases with the same value in an IP or fingerprint column; only when you map one

What the check cannot see

Ten of the 18 GDQ codes need information that an export does not contain: a honeypot at survey entry, location data, the questionnaire with its trap questions, or a reconciliation with the panel system. Those codes stay with manual coding, in the tool as in DataLion.

Code 18 (Unspecified Issue/Other) is the catch-all for anything that follows no rule.

GDQ code	Why not
1 Bot Detection	needs a honeypot or captcha at survey entry
2 Third Party Fraud Tool Failure	needs an external fraud tool
3 Geo-location Check	needs location data per case
5 Suspicious Survey Entry Time	needs entry timestamps and the field plan
9 Inconsistent/Contradictory Answers	needs the logic between questions that only the questionnaire knows
10 Red Herring/Explicit Trap Question	needs a trap question in the questionnaire
11 Knowledge Question	needs a knowledge question with a known answer
12 Over-qualification/Over-claiming	needs the screener and quotas
15 Open End – AI Completed	there is no reliable detector for AI text
17 Ghost Complete	needs reconciliation with the panel system

What a flag means

For each case the check lists the codes with a reason, such as “95 s, that is 16 % of the median (600 s)”. GDQ treats most codes as a reason to look closer: speed alone is weak evidence, long interviews often have valid causes, and uniform answers in a grid can be plausible. Immediate removal is justified, in GDQ’s words, only for duplicate participants and honeypot failures.

To put your share in context the tool shows the figures from the GDQ Data Quality Benchmarking, Wave 2 (H1 2026; 46 companies, 13 countries, about 1.8 million records). For German research agencies the report gives 40.7 % removal before and during the survey, mostly fraud and behaviour terminations in the field. Post-survey cleanout of completed interviews is 3.9 % for agencies and 27.4 % for German B2C. Your export usually holds completes only, so the second figure is the fitting comparison. A higher rate can mean better detection, a lower one looser rules; the report itself warns against league tables.

Several codes on one case weigh more than a single one
The evidence strength per code is in the table: weak, medium or strong
The benchmark line compares orders of magnitude, it does not grade your file
The CSV with flag columns filters on in SPSS, R or Excel

Cleaning a dataset: the options compared

You can clean by hand in SPSS or Excel, rely on your panel vendor’s flags, use this tool, or work in DataLion’s survey monitor, which computes the flags for every completed case while fieldwork is still running. The table shows what sets the options apart.

Option	When	Codes	Good for
By hand in SPSS or Excel	after fieldwork	your own	small studies with custom rules
Panel vendor flags	before and during fieldwork	vendor-specific	panel sample, fraud before survey entry
Survey data quality check	after fieldwork	GDQ 4, 6, 7, 8, 13, 14, 16	a quick look at an export, no upload
DataLion survey monitor with removal log	during fieldwork	GDQ 4, 6, 7, 8, 13, 14, 16 automatically, all 18 manually	live studies: removal with code, quota freed, log as CSV

From flag to decision

Run the check after the soft launch, then fix the thresholds. GDQ advises setting the speeder cut-off on observed behaviour and checking whether a time is humanly possible at all before a case is dropped. After that you decide case by case, usually on two or more codes, and record the main reason as a GDQ code.

For the method report, download the protocol: thresholds, hits per code, share of flagged cases and the date. The German industry standards of ADM, ASI, BVM and DGOF (2023) require quality measures and their results to be documented; the protocol is the first building block for that.

In DataLion the same check runs during fieldwork in every survey’s monitor: flags for each completed case, removal with GDQ code and note, the quota opens up again, and the log downloads as CSV. The reasoning behind each check is in the whitepaper Data quality in online surveys.

Common questions

Is my data uploaded?

No. The check runs as local JavaScript in your browser. There is no upload, no server and no caching; both downloads are built in the browser too.

Which file formats can I check?

CSV and TSV with a header row (comma, semicolon, tab or pipe; UTF-8 or Windows-1252) and SPSS system files (.sav). Compressed .zsav files are not supported; re-save them once as .sav in SPSS or PSPP. Beyond 200,000 rows the tool checks the first 200,000.

How are speeders defined?

As cases whose interview duration is below 33 % of the median of all cases with a duration, computed only once 10 such cases exist. You can change the value. GDQ calls speed alone weak evidence: also check whether the time is humanly possible at all, and combine it with other codes.

Is 40.7 % flagged cases normal?

The 40.7 % in the GDQ benchmark (Wave 2, German research agencies) covers removals before and during the survey, mostly fraud and behaviour terminations in the field on panel sample. Your export usually holds completes only; for those the report gives 3.9 % post-survey cleanout for agencies and 27.4 % for German B2C. A share above or below says nothing about your study on its own; the rules and the questionnaire shape it too.

Does the tool delete rows?

No. It flags and explains. Removal stays a decision for people, following GDQ usually only when several codes coincide. The CSV with flag columns leaves every filter open to you.

Where do the rules come from?

From the GDQ Feedback Loop code frame v1.1 (2026) by MRS and SampleCon with its 18 codes, and from DataLion’s survey monitor, which runs the same seven checks with the same defaults. The heuristics for gibberish and duplicates are deliberately simple so you can describe them in two sentences in the method report.

Does the check detect AI-written open ends?

No. GDQ code 15 (Open End – AI Completed) stays with manual coding; there is no reliable detector. The check finds duplicates and gibberish. Where a text came from, it cannot tell.

Flags during fieldwork instead of afterwards

In DataLion the survey monitor runs the same checks for every completed case. You remove with a GDQ code and a note, and the log goes into the method report as CSV. You start on the Free plan, no credit card required.

Start for free Data quality whitepaper