Skip to content
Contact SightX support
  • There are no suggestions because the search field is empty.

Response Quality Analysis (powered by ReDem)

Better data, less cleaning: SightX and ReDem work together to flag low-quality responses for you.

SightX is now integrated with ReDem, RepData's data quality platform. ReDem reviews your survey responses across multiple quality signals, including open-ended answers, consistency, timing, grid patterns, and how answers were entered, and provides a keep-or-remove recommendation for each response.

This article explains what ReDem evaluates and how you can use the results in SightX to ensure data quality.

What is ReDem?

ReDem is an AI-based data quality service for survey research. When a response is submitted, SightX sends it to ReDem, which analyzes the respondent's answers along with metadata from their survey session like time spent per page and mouse movement patterns. ReDem returns a quality evaluation for that response, and depending on your project settings, the respondent will either be immediately disqualified, or will be accepted but flagged in the dataset.

What ReDem evaluates

ReDem combines five different checks into a composite quality score for each response. Together, they give a comprehensive view of whether the respondent was a real, attentive person giving thoughtful and consistent answers.

1. Open-ended answers

ReDem reads open-ended (text) answers to judge whether they are valid, relevant answers. It looks for:

  • No answer – things like "n/a", "none", or "I don't know."
  • Off-topic answers – responses unrelated to the question.
  • Gibberish – random characters.
  • Likely AI-generated text – overly polished, generic wording with no personal detail.
  • Bad or offensive language – profanity or hostile wording.
  • Duplicate answers – answers that exactly match other respondents' answers, or that repeat the same respondent's own earlier answers.

ReDem also rates the effort behind valid answers (low, medium, or high), and adjusts its checks to fit the question. For example, short-label questions like brand recall aren't penalized for brief or similar answers.

2. Consistency across the survey

ReDem looks at the whole interview to check that a respondent's answers make sense together. It compares related answers, checks whether the overall pattern of answers is realistic, and looks for direct contradictions, such as saying you never use a product in one question and use it weekly in another. A single misclick or brief lapse isn't meant to sink an otherwise good response.

3. Time spent

ReDem looks at how long each respondent spent on the survey as a whole, and how long they spent on each page. It compares those times to the median for your survey and flags respondents who were far too fast, far too slow, or irregular in their pacing.

To build that median, ReDem uses the first 30 responses to your survey. Those first responses establish the baseline, and every response after that is scored against it. This means the time check starts working once your survey has 30 responses, so it won’t factor into scoring early in fielding.

4. Grid questions

ReDem looks for patterns in grid questions, such as picking the same answer in every row (straight-lining) or zigzagging between the first and last options. This check applies to grids with enough rows to evaluate.

5. Behavior while answering

ReDem analyzes how answers were entered to tell human input from automated or artificial input. For typing, it looks at speed, rhythm, and whether large blocks of text were pasted in. For mouse movement, it looks for unnatural patterns such as instant jumps or perfectly straight lines. Typing with a phone's swipe keyboard or voice dictation is treated as natural.

How ReDem scores a response

Each of ReDem's checks gives a score from 0 to 100. Those individual scores are then combined into one overall quality score, which is a weighted composite of the individual check scores. Think of it as a trust index for each response: the higher the score, the more confident you can be in it.

ReDem looks at all of the scores and recommends whether a response should be kept in the dataset, or excluded. Responses with overall quality scores below 60 are recommended to exclude, along with any response that failed a single check very badly.

Why is a response with a good score recommended for exclusion?

You may notice some responses with a high overall score that ReDem still recommends excluding. This is intentional.

The overall score blends many checks together, so a response can look strong on most of them. But some failures are serious enough that they overrule everything else. For example, if a response fails the behavior check for mouse movement and appears to have been submitted by an automated script, ReDem recommends excluding it even if its open-ended answers, timing, and consistency all scored well.

When this happens, check the exclusion reason for the response to see which check triggered the recommendation.

How SightX uses ReDem's evaluations

For each survey, you decide how SightX handles responses that ReDem recommends excluding. You make this choice in the survey's Distribute settings. There are two options:

  • Disqualify them in the survey. As soon as a respondent submits, SightX checks their response with ReDem. If ReDem recommends excluding it, the respondent is disqualified and won't be counted in your dataset. This saves you the most cleaning time, but it's permanent: disqualified respondents can't be viewed in your dataset, and their survey response can't be recovered.
  • Accept them and flag them in the dataset. Every response is accepted and flagged in the dataset; you review the flagged responses and remove them yourself during data cleaning.

If you aren't sure which to choose, accepting and flagging is the safer starting point, because you can always review a flag but you can't bring back a disqualified response.

Turning ReDem on or off

ReDem response quality can be turned on or off for each project. Your account's default setting determines whether new projects start with it on, and you can change it for any individual project.

How SightX applies ReDem's thresholds

SightX applies a standard set of ReDem cleaning settings to every project, tuned to recommend excluding clearly low-quality responses. These settings are not configurable on a per-project basis.

Good to know

  • Responses are scored once. ReDem evaluates a response when it's submitted. If a response is missing its evaluation, it won't show a recommendation.
  • Some checks need enough data. The time check, for example, needs the first 30 responses to your survey to establish a median before it can score anything, and grid checks only apply to grids with enough rows. As a result, a response may be evaluated on fewer checks than another.
  • A recommendation isn't a verdict. ReDem uses AI, and no automated check is perfect. For example, answering "nothing" to "Is there anything else you'd like to add?" can occasionally be flagged even though it's a perfectly valid answer. If you accept and flag responses, you can use your judgment and keep any response that looks wrongly flagged.

For more detail on how ReDem calculates its scores, see the ReDem Score documentation on ReDem's support site.