app

T-Test Wizard

T-Test Wizard captures two or more numeric groups, compares every group after the first with the first group, and draws one publication-oriented mean and standard-deviation plot. The first group is always the reference group.

Open T-Test Wizard

From the LabKit launcher, select T-Test Wizard and choose Open. From a source checkout, run:

labkit_TTestWizard_app

The App requires Base MATLAB and the LabKit App Framework. It does not require Statistics and Machine Learning Toolbox.

Workflow

The left controls use three short pages:

1. Data — open a table, choose where selected values go, or clear data.
2. Test & Plot — choose one t-test, run all comparisons, review result rows, and adjust plot presentation.
3. Export — save all group observations and result rows as separate CSVs.

The right workspace has two user-selectable pages:

• Data keeps the opened source table above the editable analysis table, each using half of the workspace.
• Plot gives the completed statistical figure the full workspace.

Changing data or test settings does not silently recalculate the plot. A compact status beside the plot controls reports OUT OF DATE until Run / refresh comparisons is used again. The short What will run and Result family summaries stay compact so the result table and plot controls receive most of the vertical space.

Add Groups From A Table

The Data step accepts .csv, .tsv, .xlsx, and .xls. Workbook worksheets can be selected from the dropdown. Headers and labels remain visible in the source grid.

For each group:

1. select its numeric cells;
2. leave Source selection to at (new group), or choose an existing group;
3. choose Add selected source cells;
4. repeat for every additional group.

Finite numeric cells and numeric text are accepted. Blank and label cells are skipped with a visible count. NaN and infinities block that selection until they are excluded or corrected. Opening another file replaces the source grid but keeps all groups already captured.

For a new group, the App proposes a name from the nearest useful header above the selection. It follows merged-style headings to the left and combines up to two useful levels, such as Treatment A - Metric Y. Dates, row numbers, and numeric summary cells are not used as names. The proposed label remains editable in the analysis table.

The lower analysis table uses one row per observation:

Type or paste directly into either column. Blank rows are available for manual entry without a source file. Group order follows the first appearance of each label in this table; the first group is the reference. To change the reference, reorder the corresponding rows and edit any cell so the new order is committed. The interface deliberately does not expose an unexplained N column.

To relabel many observations at once, select their rows in the analysis table, choose an existing group under Change selected rows to, and use Change selected rows. Selection is committed when the completed table selection changes, so dragging across a range does not repeatedly rebuild the App while the pointer is still moving. Use Delete selected rows to remove only the selected observations; if every observation in a group is deleted, that empty group is removed. Clear all data remains the separate full-reset action.

Choose The Tests

One selected method is applied to every comparison:

The Question dropdown offers two-sided, first greater than comparison, or first less than comparison alternatives. Difference direction is always:

first/reference group minus comparison group

Alpha defaults to 0.05. No multiple-comparison correction is applied. The App reports each requested t-test exactly as configured; whether correction, ANOVA, a mixed model, or another method is scientifically appropriate remains the user's responsibility.

Every group needs at least two finite observations. Paired testing additionally requires each comparison group to have the same length as the first group. Pairing follows displayed row order and is never inferred from a filename, header, or equal length.

Results

Run / refresh comparisons creates N - 1 immutable result snapshots for N groups. Each row reports:

• comparison and reference labels;
• sample counts;
• group means;
• reference-minus-comparison difference;
• t statistic, degrees of freedom, and exact p-value;
• NS, *, **, ***, or ****;
• stable success or failure status.

The thresholds used for the shorthand plot labels are:

The exact p-value remains available in the result table and CSV. Editing groups or test settings afterward does not rewrite the completed family; rerun to create a new family.

Plot Style

The plot uses the copied completed-result observations, not live editor values. Choose Mean bars with SD or Box plot without recalculating the tests. Both styles use numeric y-axis ticks, ordinary horizontal x-axis tick labels, and the same annotation geometry. Figure Studio can adjust typography and label presentation after the plot is popped out.

The bar visual contract is:

• one pastel bar per group;
• bar height equal to the group mean;
• dark, capped sample-standard-deviation error bars;
• a zero-anchored y-axis for all-positive or all-negative bars, including after wheel or toolbar zoom;
• black boxed axes on a white background;
• Helvetica labels and no grid;
• stacked brackets from the first bar to every later successful comparison;
• significance shorthand on each bracket.

The first four colors are soft green, orange, blue, and yellow; the palette cycles for additional groups. The box style uses the same group colors with subtler fill, dark outlines, whiskers, and medians. When individual values are shown, the box chart's own outlier markers are suppressed so observations are not drawn twice. Optional controls overlay individual values, hide SD bars on bar plots, hide brackets, and edit the title or y-axis label.

Initial limits include bars or boxes, visible individual values, error bars, and significance brackets. Reset plot view recomputes that fitted viewport without rerunning the tests or changing project data. Switching between bar and box styles also accepts the new style's fitted limits; ordinary redraws otherwise preserve the user's current zoom.

Use Plot > Pop out all plots for a standalone MATLAB figure. The popout can be passed to Figure Studio for further visual styling without recalculating the tests.

CSV Outputs

Export group data CSV writes one group per column and pads unequal groups with blank cells:

Row,Reference,Treatment 1,Treatment 2
1,1.2,1.8,1.1
2,1.4,1.7,1.2

Export comparison results CSV writes one row per later group. Columns are:

Test, Alternative, Alpha, Reference Group, Comparison Group,
N Reference, N Comparison, N Pairs,
Mean Reference, SD Reference, Mean Comparison, SD Comparison,
Difference Reference-Comparison, SE, T, DF, P,
CI Lower, CI Upper, Significance, Status, Message

The imported source is never modified or overwritten.

Programmatic Use

Run a complete first-versus-each family without opening the GUI:

groups = struct( ...
    "label", {"Reference", "Treatment 1", "Treatment 2"}, ...
    "values", {[1.2 1.4 1.3], [1.8 1.7 2.0], [1.1 1.2 1.4]});
options = struct( ...
    "method", "welch", ...
    "alternative", "two_sided", ...
    "alpha", 0.05);
results = ttest_wizard.testRun.runGroupTTests(groups, options);

runGroupTTests returns one canonical runTTest result per later group.

Project Recovery

Project schema version 2 stores ordered groups, test settings, plot settings, completed comparison results, and source references. Version-1 A/B projects migrate by preserving A as the first group and B as the second group.

Assumptions And Limitations

• Observations must share a meaningful measurement unit.
• Independent tests assume independence within and between groups.
• Paired tests use the displayed row order for every first-versus-group pair.
• The App does not infer experimental units, exclusions, independence, biological replication, or the correct statistical model.
• No multiple-comparison correction, ANOVA, nonparametric test, regression, or mixed model is performed.
• Plot shorthand does not replace the exact p-value or establish scientific importance.

Related Documentation

• Statistics Apps
• T-Test Wizard design
• Simple Scientific CSV Exchange
• Figure Studio
• App Framework