Take the files the analysis needs.
| Output | What it preserves |
|---|---|
| SPSS .sav | Variable names, question labels, value labels and defined missing-value codes. |
| CSV | A response table you can open in Excel or another analysis tool. |
| Codebook | The questions, variables and response-option definitions. |
| Flow diagram | The conditional paths that explain which questions participants could see. |
All language versions enter one response table. Submission language, recruitment source and study version remain available as context variables. CSV is an Excel-readable format; it is not a native .xlsx workbook.
A missing answer can mean different things.
| Code | Meaning |
|---|---|
| −99 | The question was not shown because of survey flow. |
| −98 | The question was shown but left unanswered. |
| −97 | The participant declined to answer. |
These codes are defined as missing values in the SPSS file, rather than being treated as numeric observations in an average. Check how your chosen analysis software handles them.
Build a measurement from several questions.
Define concepts, dimensions and the formula that combines them. Reverse-code the relevant items. A satisfaction index, trust score or study-specific scale can use a repeatable calculation instead of a formula assembled again for each dataset.
The model tools support population-based weighting, trimming extreme weights, effective sample size and confidence intervals. Group score differences can be examined with their own confidence interval. Small breakdowns below the configured display threshold are hidden; the default threshold is 15 participants.
When a model is reused, compatibility is checked. Model versions identify the rules behind a calculation. A model is an analytical specification, not a guarantee that the underlying construct or sample is valid.
Repeat the measurement without losing its meaning.
Duplicate a study as a new wave and retain its question structure. Comparability checks help identify changes before collection. Each wave remains a separate study with its own data and notice version.
Use consistent wording, options and recruitment decisions when comparing periods. A trend graph cannot resolve a changed sampling method or a different question meaning.
Show calculated model results in the report.
Add precomputed model results as a score card or dimension chart. A reader selecting a report filter does not recalculate the model. Keep the base of the model result separate from the base of a filtered question chart.
Current limits: measurement models operate on up to 10,000 responses. Their results can be viewed in Nuvopoll but cannot currently be exported as a model-result file. Use labelled raw-response exports when further statistical analysis is needed.
AI naming and data export are separate operations.
When a survey template is saved, AI can suggest variable names for questions whose codes you left blank. You can supply your own codes instead. Participant responses do not exist in that template-naming step.
The SPSS file is created on the server. Exporting it does not send the response table, open answers or identifying fields to an external AI model. Voice-response transcription is a different data flow.