Verify that your findings survive outlier removal and data winsorization — or discover where they are fragile.
Sensitivity analysis tests whether your statistical conclusions change when you modify your data in defensible ways. The most common form re-runs your primary analysis under three conditions: (1) the original data, (2) after removing extreme outliers (values beyond ±3 standard deviations), and (3) after winsorizing at the 5th and 95th percentiles.
If the conclusion (significance of the primary finding) is the same across all three conditions, the result is considered robust — the finding does not depend on a handful of extreme values. If the conclusion changes (e.g., p < .05 on original data but p > .05 after outlier removal), the result is fragile and requires careful scrutiny of those extreme cases before drawing conclusions.
Sensitivity analysis is increasingly required by journals and is recommended by the APA Publication Manual for any analysis where outliers may be influential. StatRyx automates the entire comparison and produces a clear verdict: Robust or Sensitive.
A sensitivity analysis re-runs the main analysis under different reasonable choices — with and without outliers, with winsorized values, with a different missing data method — and checks whether the conclusion holds. If the effect keeps its direction and significance, the finding is robust; if it disappears when three cases are removed, readers need to know. Pre-specifying the sensitivity checks avoids the impression of choosing the most favourable version.
| Condition | n | Effect estimate | p-value | Conclusion |
|---|---|---|---|---|
| Original data | 248 | β = 0.41 | p = .003 | Significant |
| Outliers removed (|z|>3) | 244 | β = 0.38 | p = .007 | Significant |
| Winsorized (5th–95th %ile) | 248 | β = 0.39 | p = .005 | Significant |
A sensitivity analysis was conducted by repeating the primary regression with outliers removed (|z| > 3.0, n = 4 cases excluded) and with 5th–95th percentile winsorization. The main effect of the predictor remained significant across all three conditions (all p < .05), indicating robust results.
Yes. You can run it without an account (3 free analyses), and a free account adds 10 more — a one-time allowance, no payment needed. Every result includes the APA 7 write-up. Paid plans (Student from $9/month) add more analyses and Word, PDF and Excel export.
Winsorizing replaces extreme values with the value at the 5th and 95th percentile boundary. Unlike removing outliers, it retains the same sample size while limiting the influence of extreme observations.
StatRyx currently supports sensitivity analysis for t-tests, ANOVA, correlation, and linear regression. More analysis types can be requested by describing them in the chat.
Describe each alternative analysis and whether the conclusion changed: "Excluding four cases with |z| > 3 did not change the result (B = 0.42, p = .003, versus B = 0.45, p = .001 in the full sample)." A small table comparing the estimates is often clearest.
No. The term is used for two different things: a robustness check of results (this page), and a sensitivity power analysis, which computes the smallest effect a given sample can detect. Make clear which one you mean.
Run a Sensitivity Analysis on your data →