Response‑Quality Cleaning
SurveyMars Response Quality Cleaning automatically flags low-quality responses—such as those completed too quickly, submitted as duplicates, or containing meaningless text—without altering your original data, and explains why each response was flagged. After you review and confirm, you can remove recommended exclusions from valid analysis data. The process is traceable and reversible. It is especially useful for report delivery, academic submissions, or surveys with many open-ended questions or large response volumes.
Overview
Response Quality Cleaning is a data quality tool in the statistical analysis stage. It is not simply about deleting abnormal responses—it is a workspace for detect, explain, review, apply, and restore. The system calculates a quality score for each response based on your chosen strictness and rules, and lists match reasons and evidence. After you review responses individually or in bulk, click "Apply Cleaning" to affect analysis data. Original responses are always kept, and mistakes can be undone within a limited time window.
How to Access
1. Open your target survey and go to "Statistics".
2. Switch to the "View & Download Responses" tab.
3. In the toolbar above the response list, click "Clean" to open Response Quality Cleaning.
Note: If the survey already has scan results, you go straight to the review page. On first use or when you need a new scan, you start on the "Start Response Quality Cleaning" settings page. For large response volumes, scanning runs in the background and the page shows progress.
How to Use
Part 1: Select strictness and start scan
1. After you open the cleaning panel, the page shows how many responses will be scanned (for example, "100 responses will be scanned").
2. Choose a strictness level from the three mode cards (first-time users should keep the default "Standard" selection):
- Lenient: Cleans only obvious issues (trap question errors, severe speeders, duplicate responses). About 95% of responses are expected to be retained.
- Standard (recommended): Flags responses with an overall quality score below 70 that also match recommended cleaning rules. About 85% of responses are expected to be retained.
- Strict: Cleans suspicious responses more aggressively to keep high-quality samples. Best for academic papers and competition submissions. About 70% of responses are expected to be retained.
3. (Optional) Click "Expand Advanced Settings" to adjust detection rules, thresholds, and whitelists as needed (see "Configuration Options" below).
4. Confirm the note at the bottom—"Responses that match selected rules will enter cleaning review"—then click the green "Start Scan" button. Small samples finish synchronously; large samples enter a background queue and automatically open the cleaning review results page when the scan completes.
Part 2: Review and process responses on the review page
1. The top of the results page shows flagged count, recommended exclusion count, average quality score, and the "Quality Score Threshold". To change rules, click "Filter Settings" to reconfigure and scan again.
2. Use status tabs to switch list views:
- Flagged: Matched rules but not yet processed.
- Pending Cleaning: Recommended for exclusion; waiting for you to click "Apply Cleaning".
- Cleaned: Cleaning applied; removed from valid analysis.
- Ignored: Responses you chose to keep and not treat as matches.
3. The list shows response number, status, match reason, rule, quality score, and evidence summary. Click "View" to open the detail drawer and verify the full original response, matched rules summary, in-question highlights, and evidence notes (for example, "Matched Question 4: meaningless text '123'").
4. For individual responses, use the actions on the right:
- Clean: Move the response to "Pending Cleaning" and wait for batch application (it is not removed from the valid sample immediately).
- Ignore: Confirm the match can be kept; the response moves to "Ignored". Identical match signatures can be ignored in bulk.
5. Select multiple responses to clean, ignore, or restore in bulk. The "Match Reasons" panel on the right groups matches by reason so you can review by issue type.
Part 3: Apply cleaning and restore
1. After moving responses to "Pending Cleaning", click "Apply Cleaning" in the upper-right corner and confirm the affected count in the prompt to formally remove them from valid analysis data.
2. After application, affected responses move to "Cleaned" status. You can view them by cleaning batch in the "Recycle Bin" under "View & Download Responses".
3. If you made a mistake, undo or restore from the "Cleaned" tab. The system enforces a defined time window—act promptly.
Configuration Options
Strictness and quality score threshold
- The three mode cards sync recommended thresholds and expected retention rates. On the results page, you can also adjust the "Quality Score Threshold" separately to control which responses are more likely to be recommended for exclusion.
Basic rules
- Minimum completion time (seconds): Set in Advanced Settings (default is 15 seconds). Responses completed noticeably faster than required may match "Completed Too Quickly".
Trap Question Validation / Custom Contradiction Logic
- Off by default. After you turn on "Enable Trap Question Validation" or "Enable Custom Contradiction Logic", complete question and condition setup in the survey before clicking "Start Scan". Otherwise, the system prompts you to finish configuration first.
Advanced: Detectors
Enable the following detectors as needed (each rule outputs structured evidence such as match reason, question number, and matched text):
- Completed Too Quickly: Completion time is noticeably shorter than the minimum reasonable time estimated from question types, count, and strictness.
- Suspected Duplicate Responses: Closely matches other responses in the same survey by answer patterns, IP, timing, or answer similarity.
- Incomplete Response: Missing answers, skipped items, or structural gaps (such as unanswered required questions).
- Meaningless Text: Invalid, garbled, or low-quality text in open-ended questions.
- AI Patterns: Matches common AI-generated text patterns.
- Cross-Response Similarity: Open-ended text is highly similar across multiple responses.
Whitelists
- Question-level whitelist: When enabled, add specific questions to skip text-quality detection for those questions.
- Survey text quality whitelist: When enabled, add plain text or regex entries for the current survey. This affects text quality rules for this survey only. Off by default—you must explicitly enable it to avoid unintentionally relaxing rules.
FAQ
Q: What do "Start Scan", "Clean", and "Apply Cleaning" do?
A: "Start Scan" analyzes responses and flags matches without changing the valid sample. "Clean" in the list moves a response to "Pending Cleaning". "Apply Cleaning" in the upper-right corner formally removes pending responses from analysis data.
Q: Can I restore responses after applying cleaning?
A: Yes. Incorrectly excluded responses can be undone and restored by batch from the "Cleaned" tab or the "Recycle Bin".
Q: Why is normal open-ended text flagged as "Meaningless Text"?
A: Check the evidence question number and text in the detail drawer. If it is a false positive, click "Ignore", or configure a question-level or text quality whitelist in Advanced Settings. Setting the correct attributes for textbox questions (such as phone or email) also helps reduce false positives.
Q: Why can't I start a scan after enabling trap questions or contradiction logic?
A: Complete question and trigger condition setup in the survey first, or the system prompts you to finish configuration. If you do not need these rules, turn off the corresponding switches to scan normally.
Q: What if scanning is slow or fails with a large response volume?
A: Large samples scan asynchronously in the background. Reopen the page later to check progress. If the scan fails, use "Filter Settings" to reduce enabled rules or lower strictness and try again. If it still fails, contact support with your survey ID.
