| Flag with dummy variable |
Create a column (e.g., suspect = 1) to mark questionable respondents. Keep them in the dataset but exclude from primary analysis. |
When you’re uncertain about data quality and want to run sensitivity checks. |
| Quarantine and compare |
Run your analysis twice — once with all data, once excluding flagged cases. See if conclusions change. |
When sample size is tight and you can’t afford to lose cases without knowing the cost. |
| Weight down rather than delete |
Assign lower weights to suspect respondents rather than removing them entirely. |
When you have a weighting scheme and want to reduce influence without elimination. |
| Segment and report separately |
Treat suspect respondents as their own group. Report their patterns alongside the clean sample. |
When “bad” responses might actually reflect a real population (e.g., disengaged customers). |
| Set thresholds in advance |
Define exclusion rules before looking at the data (e.g., “anyone completing in under 2 minutes”). Document in your analysis plan. |
Always. Prevents post-hoc fishing for rules that conveniently support your hypothesis. |
