Two questions most statistical software never asks: who did not answer, and who is missing from your sample altogether. Both change what your results are about, and neither is repairable downstream.
Every statistics package will tell you how much data is missing. Almost none will tell you who is missing, and that is the question that decides what your results describe.
Two distinct problems live here. Differential nonresponse is when the people who skipped a question differ systematically from those who answered. Composition bias is when the sample you recruited does not resemble the population you are claiming to describe.
Neither is a statistical problem you can fix later with a better model. Both are facts about your data that change the scope of every claim you make from it. Quanta treats them as a first-class analytical step asked before analysis, rather than as a limitation paragraph written afterwards.
The missingness-by-group check compares how much data is absent across the levels of a grouping variable. If one group skipped a question far more often than another, comparisons on that question are not comparing like with like.
Quanta reports whether nonresponse looks even and flags variables whose missingness differs across groups by ten percentage points or more, which turns a vague worry into a specific list.
The point it makes explicitly is the one that gets missed: differential nonresponse can bias comparisons even when overall missingness looks small. Five percent missing overall sounds harmless. Five percent concentrated entirely in one condition is not, and a single overall figure conceals exactly that.
The check needs a grouping variable with at least two groups of ten or more cases, which is itself informative. If your groups are too small to check, they are too small to compare.
The composition check compares your sample's distribution on a characteristic against population percentages you supply, giving the sample percentage, the population percentage, and the difference in percentage points for every category, plus a goodness-of-fit test against those benchmarks.
The verdict is stated plainly as either a composition match or a representation gap.
The benchmarks are yours to supply and yours to justify. Quanta says so directly, reminding you to cite their source when reporting the comparison. That is the right division of labour: the software can compute the comparison, but only you can defend the population you are claiming to represent.
Where a gap exists, post-stratification weighting can adjust estimates for it, and Quanta points to that rather than leaving the gap as a dead end. Weighting is a partial remedy, not a cure: it can correct for characteristics you measured and cannot correct for the ones you did not.
A representation gap found before analysis changes what you run. A gap found after changes only what you write in the limitations, by which point the analysis has already been designed around a sample you now know is skewed.
It also prevents a specific and common error: subgroup analyses on categories too thin to support them. A group of three respondents will produce an adjusted mean, a group difference, and a confidence interval, all printed to two decimal places. Seeing the composition first is what stops that analysis being run.
And it makes the limitations section concrete. "Older respondents were underrepresented relative to the 2020 census, 1.2% versus 16.9%" is a stronger and more credible sentence than "the sample may not be representative."
Open Data Quality. The representation checks sit alongside the missing-data diagnostics rather than in a separate corner of the app.
For missingness by group, choose the grouping variable you intend to compare on. Run this before the comparison, not after.
For sample composition, supply the population percentages for the characteristic you want checked, and keep the citation with them.
Upload categorical fields as text rather than numeric codes, or they will not be available as characteristics.
Use the action bar to add the result to a report, copy the table, or copy the APA text.
Report representation in the sample description, not only in the limitations.
Sample composition differed from national benchmarks, χ²(5, N = 250) = 134.77, p < .001, with respondents aged 25-34 overrepresented (36.4% versus 17.1%) and those aged 65 or over nearly absent (1.2% versus 16.9%). Findings should not be extended to older workers. Missingness did not differ across conditions by more than 3 percentage points.
Cite the source of your benchmarks. An uncited population percentage is an assertion.
Report the nonresponse check even when it is clean. Stating that missingness was even across groups is evidence you looked, and it costs one sentence.
Name who the findings describe. That is the honest version of a generalizability claim, and it is more useful to a reader than a hedge.
Every result carries a ReliCheck Intelligence card with one action: explain this result in plain language. It runs entirely on your Mac with Apple Intelligence. No upload, no API key, nothing to configure. The model only explains the result Quanta has already computed and validated. It never computes a statistic, and it never applies a decision on your behalf.
From the action bar you can add the result to a report, copy the table, or copy the APA text. Exported reports carry the APA table, its note, and a caution specific to that analysis beneath it.
Quanta's engine is pure Swift on Apple's Accelerate framework, with no R or Python at runtime. Every engine is checked against an independent reference rather than internal consistency alone, and the automated suite exceeds 500 checks. Full record on the validation page.
Data Quality is part of every Quanta subscription, $9.99 a month or $79.99 a year after the 30-day trial. One price for everyone. See pricing.
The analysis, the report builder, and the on-device explanations all work with the network off. Your dataset is never uploaded.
Chi-square goodness of fit the test behind the benchmark comparison · Demographics tables where imbalance first becomes visible · Little's MCAR test whether missingness is random · the complete list on the analyses page.