Turn measurements into a comparison you can inspect.

Start with the question, then choose the method.

What does this dataset look like?

Review raw measurements or occurrence-frequency input, save the dataset in the workspace, then inspect descriptive statistics, histograms, boxplots, an empirical cumulative distribution or a Normal Q-Q plot. Counts of occurrences are not arbitrary observation weights. A plot helps you inspect data; it does not establish a statistical model's assumptions.

Do these measurements differ?

Use a one-mean t procedure when estimating variation from a sample. For two independent samples, deliberately choose Welch or pooled t; pooled t assumes equal population variances. For before/after or matched measurements, use genuine paired rows. The paired route uses second column minus first column; independent comparisons use A minus B. Matching row counts alone do not make samples paired.

If a population standard deviation is assumed known separately, use a one-mean or independent-two-means z procedure. Known population SD is not sample SD or standard error, and SumWise does not switch to z automatically because a sample is large.

Do several groups differ?

Compare two to four independent groups with ordinary or Welch one-way ANOVA. Ordinary ANOVA assumes equal population variances; Welch uses its distinct unequal-variance method. An accepted group investigation can produce the complete all-pairs family: Holm-adjusted tests and Bonferroni confidence intervals have different meanings. Choose the method from the study design and assumptions, not whichever gives the smallest p-value.

Do I have raw observations or summaries?

Mean procedures and group comparisons support deliberate Raw data or Summary statistics input. A supplied summary contains n, mean and sample SD; z uses separately identified known population SD. Summary tuples do not create missing observations, medians, raw-data plots or normality evidence. Paired inference still requires actual paired rows.

Worked example: a mean with known population SD

In Statistics, open the Inference tab and choose Known-SD one-mean z, Summary statistics and Test and two-sided interval. Enter one summary row labelled Example, with n and mean below. Leave its sample-SD cell blank; put the known sigma in the separate Known population SD sigma A field. This block is a field checklist, not a calculator command.

n = 25
mean = 52
known sigma = 5
null mean = 50
alternative = two-sided
confidence = 0.95

Choose Review mean procedure, inspect the inputs and assumptions, then Run test. The standard error is 1, z = 2 and the two-sided p-value is approximately 0.0455003. The 95% two-sided interval for the population mean is approximately [50.0400, 53.9600].

This assumes independent sampling and an appropriate normal model or justified approximation, with sigma genuinely known. A small p-value does not measure practical importance, prove the model or establish causation. For an interval without a test, choose Two-sided interval only; no null or p-value is invented.

Worked example: compare three summary groups

From Statistics, open Probability, counts and group means, then Compare group means. Choose Summary statistics and the ordinary equal-variance method deliberately. Enter each row as a separate group:

Group A
n = 3, mean = 2
sample SD = 1

Group B
n = 3, mean = 3
sample SD = 1

Group C
n = 3, mean = 4
sample SD = 1

Set Significance alpha to 0.05 and, for this synthetic independent-group example, check I am comparing independent groups (unit weights). Choose Review selected groups, then Run group comparison. Ordinary one-way ANOVA gives F = 3, degrees of freedom (2, 6) and p = 0.125. At a 5% significance level, this does not reject equal population means; it does not prove equality.

The example assumes independent groups, an appropriate normal model and equal population variances. These summaries suffice for the calculation, but cannot verify those assumptions. Welch is a different method and gives a different result for these same inputs.

Fit a relationship, then inspect what the curve misses.

Fit a straight line, proportional model, supported polynomial, logarithmic, exponential or power model. Inspect the accepted scatter/fitted view and residual diagnostics, then request a prediction explicitly. Keep the fitting scale and extrapolation in view; a higher R-squared alone is not a reason to choose a higher-degree polynomial.

Confidence intervals for a fitted mean and prediction intervals for a new observation answer different questions. Available intervals depend on the model; exponential and power fits provide back-transformed point predictions without uncertainty intervals.

Prefer the direct calculator route? The established five-point vector regression tutorial remains available. It demonstrates slope 0.6, intercept 2.2 and R-squared 0.6.

See the data and keep the numerical result.

Use plots alongside their retained numerical tables. Sampled fitted curves are approximate drawings, not exact mathematical curves. Summary-origin group means and valid difference-interval views do not contain raw observations.

Reopen the work. Share selected results.

Save supported datasets and accepted inference results in the workspace, then use File → Save to write the file to disk. Keeping a dataset or result inside the workspace is a separate step. Reopening shows historical snapshots without rerunning or freshly verifying the numbers.

Export supported figures as vector PDF or raster PNG, or select records for a searchable PDF report. Report figures may include raster content; 3D stays raster. Exports do not replace saving the workspace. Follow the save-and-share guide.

Local mathematics, with clear boundaries.

Calculations run locally. Microsoft Store handles installation, trial, purchase and license verification; offline calculation does not promise indefinitely offline licensing. App interfaces are available in English, Spanish, French and German. Some technical reports and examples remain in English; use the documented decimal-point input syntax.

Choose within the supported scope.

SumWise supports bounded statistical workflows, not a complete replacement for R, SPSS or specialist statistical software. Supported sample sizes, numerical ranges and work limits can produce an explicit refusal. Do not drop observations, change methods or reinterpret assumptions just to force a result. Multiple regression, general time-series modelling and unrestricted data import are not promised.

Python calculation is unavailable in this Store release; SVG export is unavailable. Native algebra and the statistical tools described here remain available. Statistical significance is not practical importance, causation or professional validation.

Before running your own comparison

Are the samples paired? Decide from how observations were collected. Separate means and SDs cannot supply the covariance needed for a paired test.

Which SD field? Sample SD estimates variation from the sample. Known population SD is a separate modelling assumption. Neither field is an SE or a variance.

Will opening an old result correct it? No. Saved numbers stay unchanged; explicitly Review/Run to use the current methods. Keep the original record until a replacement succeeds.

Try one task with your own question.

3.2.1 start-to-report guide · Current release notes · Support