Stimulus Norming Studio

Step 1

Upload a stimulus CSV

Start with the file you want to inspect. The app previews the data, guesses likely column types, and flags import problems before any statistics run.

CSV upload Drop a CSV here or choose a file Comma, semicolon, and tab-delimited files are supported. The file stays in your temporary browser session.

Choose CSV No file selected

Included examples

Use two_group.csv first for a simple matching task, then three_group.csv if you want to inspect the multi-group workflow.

two_group.csv for 2-group matching
three_group.csv for 3-group ANOVA checks
no_group.csv for QC without group tests

Preview

No dataset loaded

Check whether the parser split the columns correctly and whether the first rows look like the original spreadsheet.

QC warnings

Early import checks

This block highlights file-shape problems such as width mismatches, renamed duplicate headers, or other issues seen during upload.

Step 2

Map columns and confirm the dataset

Tell the app which column contains the stimulus text, which column defines the comparison groups, and which columns should become numeric metrics.

Stimulus text column Group or condition column Stimulus ID column

Numeric metrics

Uploaded quantitative columns

Choose the numeric columns that should be compared across groups, such as ratings, lengths, frequencies, or any precomputed item property.

Ignored columns

Keep them in exports but exclude from analysis

Use this for notes, labels, bookkeeping fields, or anything that should remain attached to each row but not drive the statistics.

Type overrides

Manual fixes when auto-detection is wrong

If a numeric column was read as text, or a note column was mistaken for data, correct it here before saving the mapping.

Column	Inferred type	Example	Override

Step 3

Run normalization analysis

Set the matching criteria here. The results table shows whether each metric passes the current rule and which rows look risky.

Waiting for dataset

Alpha

Alpha is the p-value cutoff used for the statistical comparison. Higher values are more lenient; lower values are stricter.

Max standardized mean delta

This limits how far apart the group means can be after standardization. Smaller values force the groups to be more closely matched.

Max SD ratio

This compares the spread of the groups. A value near 1 means the groups vary by about the same amount.

Optional max raw mean delta

Use this when you want a practical limit in the original units, for example no more than 0.5 rating points or 1 character difference.

Outlier method

Z-score flags values far from the mean. IQR flags values far outside the middle range of the data and is often more robust.

Outlier cutoff

Lower cutoffs mark more rows as unusual. Higher cutoffs are stricter and only flag more extreme values.

Include derived text metrics Exclude flagged outliers from inferential tests Restrict analysis to rows without QC issues

Metric overview

Group means for the selected metric

Use the chart for a quick visual comparison only. The decision still comes from the metric table and its pass/fail checks.

Metric results

Pass/fail, p-values, thresholds

Each row summarizes one metric. Start with the status column, then inspect the test, effect size, and notes if a metric fails.

Row annotations

Inspect missing values, duplicates, parse failures, and outliers

This block explains why individual rows may be risky to keep, for example because a value is missing, duplicated, or unusually extreme.

Step 4

Generate counterbalanced lists and export files

After QC and analysis, create balanced list assignments and export the exact rows you want to carry into the experiment build.

Lists

Choose how many output stimulus lists you want to create for counterbalancing.

Seed

Keep the same seed if you want the exact same randomized assignment again later.

Stratify by

Stratification keeps assignments balanced within selected categories, such as valence class or condition label.

Block by

Blocking groups rows into separate assignment buckets before the lists are generated.

Keep groups separate during assignment

List balance

Assignments by group and list

Check that each list receives a balanced number of rows before exporting the randomized file.

Assignments

Per-row seeded output

The seed makes the assignment reproducible. Keep it if you want to rebuild the same lists later.