Multi-Response Questions
"Which of these did you experience? Tick all that apply" produces a cell holding several answers at once — Noise;Dust;Heat. The row is still one person, but that person now sits in three categories instead of one.
SenSight only splits such a column when you tick Multi-response in the cross-tab or the Pareto panel and choose the separator. Nothing else changes: the dataset, the other analyses and the export never see the split values. The base stays the number of people, so a column of percentages can add up to well over 100% — here that is correct, not a bug.
What you will notice is that the significance letters and the chi-square line disappear. That is deliberate. Both tests rest on a single assumption — every person falls into exactly one cell — and a multi-response question breaks it. Run them anyway and they see more independent observations than really exist: p-values shrink and letters appear for differences that are not there. The numbers would look entirely reasonable and be wrong. Showing nothing is the honest answer. Compare the shares by eye, or recode the question into a single-answer column ("main issue") when you genuinely need a test.
SenSight only splits such a column when you tick Multi-response in the cross-tab or the Pareto panel and choose the separator. Nothing else changes: the dataset, the other analyses and the export never see the split values. The base stays the number of people, so a column of percentages can add up to well over 100% — here that is correct, not a bug.
What you will notice is that the significance letters and the chi-square line disappear. That is deliberate. Both tests rest on a single assumption — every person falls into exactly one cell — and a multi-response question breaks it. Run them anyway and they see more independent observations than really exist: p-values shrink and letters appear for differences that are not there. The numbers would look entirely reasonable and be wrong. Showing nothing is the honest answer. Compare the shares by eye, or recode the question into a single-answer column ("main issue") when you genuinely need a test.
Try it in the app
Try: load "Employee Survey — Engagement" → Cross-Tab → tick "Rows hold several answers" → rows = Improvements, columns = Department. Read the parsed-answer preview first, then look for the letters in the table — and read why they are not there.