What did multiple imputation add to a vaping study’s estimate?
Follow the completed dataset versions, separate model estimates and final pooled uncertainty.
No Vape · Updated

Find the analysis that used the imputation
In a JAMA study of vaping cessation, the authors describe a primary missing-outcome rule and a separate sensitivity analysis using multiple imputation. The supplementary material has a dedicated methods section for the imputation. When two estimates appear together, first write down which analysis each one belongs to. This guide uses the methods distinction only.
Next identify which fields were missing in the analysis you are reading. A paper can impute an outcome, a background measurement or both. A general note that missing data were handled gives too little information to decide whether the quoted result includes completed versions of that particular field.
Imagine three possible completed versions
Consider a fictional worksheet with one unavailable survey response. A model supplies a plausible value for that field in each of three versions of the worksheet. The recorded entries are still the same entries in all three. The analysis calculates an estimate for each version and then combines the estimates. The number of people studied has not tripled.
Sterne and colleagues’ BMJ methods account describes creating several completed versions, analyzing each and combining the resulting estimates. The values supplied to complete them are conditional on the model. Ask which information the authors used to generate those values, and keep the specific missing-data assumption visible next to the method.
Read the uncertainty at two levels
Each completed version has its own uncertainty around its estimate. The estimates can also differ between versions because missing values were filled differently. Pooling rules can reflect both. If a summary gives only the point estimates, their spread alone will not reproduce the reported standard error.
Set up three labels in a reading note. Record what was imputed, what was estimated in each completed version and how the reported estimate and uncertainty were combined. When an appendix names a specific rule, copy its name rather than guessing from how many versions the authors created.
Keep assumptions and observations separate
If a paper presents results under different assumptions about the people who did not respond, label the results with those assumptions. They help a reader see whether a conclusion changes under the scenarios the authors considered. They do not reveal the missing participants’ actual outcomes.
The result of this reading task is a clear account of what the pooled number represents. For your own record, a day without an observation can remain unresolved without reconstructing it from the surrounding days.
Questions about this step
Is one completed version enough to read the uncertainty from multiple imputation?
One version leaves out the variability across plausible completions. Check the reported pooling procedure before interpreting the combined interval.
Sources and scope
Sources inform the guidance. The worksheets and examples are No Vape editorial suggestions, not tested treatments.
Written with AI assistance and checked against the cited sources. No clinical review is claimed. This guide covers practical planning, not individual medical treatment. How we research and write