Check the records and define the analysis scope.
Review completion status, missing values, source, language and study version. Decide whether partial responses belong in the analysis and record the decision. Use eligible, usable responses for the question being answered.
In Nuvopoll exports, skipped, unanswered and declined questions have different codes. Do not treat these as ordinary numbers or assume that every empty field means the same thing.
Read the distribution and choose the denominator.
A rating average can hide a split between highly satisfied and dissatisfied respondents. Inspect the distribution alongside any mean. Keep inapplicable answers outside a numeric scale.
For multiple-choice questions, distinguish the proportion of participants selecting an option from the proportion of all selections. These denominators answer different questions. State which one the chart uses, and show its response count.
Compare groups on the same measure.
A cross-tabulation can compare the same question by a relevant grouping variable. Check counts, option codes and recruitment differences. Language, country and source should not be treated as interchangeable variables.
Small groups can make a percentage unstable or expose individuals. Use the display settings and consider additional aggregation before publication.
Read open answers and define models deliberately.
A word cloud can point to recurring terms; it does not replace reading statements in context. Distinguish someone’s account from a researcher’s interpretation.
When several items form an index, specify the formula, reverse coding and weighting assumptions before interpreting the score. Nuvopoll supports models, intervals and group differences, subject to the current model limits. The tool does not validate the construct for you.
Know which report components a filter changes.
Main-study question charts can respond to permitted reader filters. Components from other studies and precomputed model results retain their own data. A filtered question distribution and a model score may therefore describe different participant bases.
An evaluation workflow to use.
- Separate event satisfaction from test performance.
- Report before/after counts and the matching method.
- Inspect distributions and meaningful subgroup differences.
- Read explanations in open answers and verify transcripts where used.
- Export labelled SPSS or CSV data when further statistical work is needed.
- Publish an understandable finding with its method and limits.