Free Survey Response Quality Checker

Free, private, and built for better research

Most data-cleaning tools want your response file on their server, and most of them charge per row. This one does neither. The checks are ordinary statistics: counting repeats, comparing each duration against your own median, measuring vowel ratios. They run on your machine in milliseconds, with no model and no API behind them.

What the checker looks for

Five deterministic checks, run on every row:

1. Straight-lining fires when the same value repeats across at least 80% of a row's answered scale-like columns and there are five or more such columns. The report names the repeated value.
2. Speeding fires on a completion time under 40% of the median for your own dataset. It needs a duration column, and it accepts raw seconds, mm:ss, and hh:mm:ss.
3. Gibberish text fires on an answer of 10 characters or more with a vowel ratio below 0.15 or above 0.85, a character repeated five times in a row, a consonant run of six or more, or no spaces at all beyond 25 characters.
4. Duplicate responses are two or more rows whose answers are identical across every answer column. The report gives the size of each duplicate group.
5. Low-effort text is an answer of two characters or fewer, or a single word repeated, such as "test test test".

Each row is banded by how many checks fired: Clean at zero, Review at one or two, Likely low quality at three or more. No row is ever labelled a bot. The honest reading of three flags is "likely low quality, review manually", and a clean row can still be a bad response the checks did not catch.

Worked examples: two flagged rows

Here is a small export where the median completion time across all respondents is 184 seconds, so the speeding cut-off lands at 74 seconds. Two rows get flagged, for entirely different reasons.

Why row 3 was flagged

Straight-lining. The value 5 appears in 6 of 6 answered scale columns, which is 100% and well past the 80% threshold, and there are at least five scale columns for the check to run on. The duration is normal and the comment reads fine, so this is a single flag: Review, not a removal. Someone who genuinely liked everything produces exactly this row, which is why straight-lining alone is a prompt to look rather than a reason to delete.

Why row 7 was flagged

Speeding and gibberish. 38 seconds is under the 74-second cut-off, which is 40% of the 184-second median. The comment "hjkllkjhgfds" trips two gibberish rules at once: it contains no vowels at all, so the vowel ratio is 0.00 against a floor of 0.15, and all twelve characters form one unbroken consonant run against a limit of six. Note how close the thresholds are: a near miss like "asdkjhasdkjh" has a vowel ratio of 0.17 and a run of 5, so it would pass both rules and never be flagged. The scale answers vary, so straight-lining does not fire. Two flags still means Review, but this one is far more likely to be a real problem than row 3.