Our Approach to Analysis
Analysis you can review.
Useful findings need figures you can check, methods that fit the question, and your judgment.
How are results calculated?
Calculations before conclusions
For supported quantitative analysis, structured calculations produce counts, percentages, averages, tables, and statistical results before AI explains them.
Checks vary by task: variables, filters, bases, missing data, groups, and methods. AI-assisted interpretation and open-ended summaries are distinct from this calculation process.
The method should fit the question
Match the method to your question, variables, and available observations. Request the analysis you need, or use Brief and Ask to explore.
A descriptive table and a statistical test answer different questions.
What can I check?
The base matters
Review the denominator: valid responses, filters, and subgroup bases determine which records contribute. Missing values aren't silently treated as zero.
A result for people who answered isn't necessarily a result for everyone in the file.
Quality checks, not silent changes
Quality checks flag missing values, exact duplicates, inconsistent types, empty or constant columns, and unexpected values without changing your source.
An exact duplicate can be a legitimate repeated observation. Small groups, sparse categories, and missing data can limit conclusions; you decide what needs attention.
Inspect the available evidence
Evidence varies by task and plan: supporting figures, tables, quality information, statistical output, and supplied Python.
Where supplied, Python should reflect the displayed calculation: filters, grouping, missing-value treatment, and denominator. It's not a promise that every result includes a reproducible script.
What are the limits?
Keep statistical evidence in context
Keep these distinctions in view:
- An observed difference isn't automatically statistically significant.
- Statistical significance isn't practical importance; sample size matters, and a p-value isn't the whole finding.
- An association alone doesn't establish causation.
Human review remains part of the analysis
Review supporting figures, bases, and limitations before using a finding. Context and subject-matter knowledge still matter.
You decide whether the result answers your question and what action it supports.
Faster analysis. Reviewable evidence. Your judgment.
Try your own data · Explore the analysis tasks →