Z-Scores for Stylometry

Step 1 The Raw Data: Word Counts

Counting Function Words

Stylometrists typically analyse the most frequent function words — grammatical words like the, a, of, at, to. Unlike content words, function words are used unconsciously and are hard for an author to deliberately control, making them strong markers of individual style. The table below shows raw counts of five such words across several documents, along with each document’s total word count.

The Problem with Raw Counts

Documents have different lengths. A 10,000-word novel will naturally have more occurrences of the than a 2,000-word essay, even if both authors use the word at the same rate. We cannot compare raw counts directly — we first need to normalize them.

Step 2 Normalize to Relative Frequencies

Removing the Effect of Document Length

We divide each raw count by the document’s total word count. The result is a relative frequency (a proportion between 0 and 1), which makes documents of different lengths comparable. Below the data rows, the mean and standard deviation of each column are shown — these are the key ingredients for z-score standardization.

Step 3 Z-Score Standardization

Putting Everything on the Same Scale

The z-score tells us how many standard deviations a value is above or below the mean. After standardization, every column has mean = 0 and standard deviation = 1, regardless of the original scale.

Reading Z-Scores

Now all features are comparable: a z-score of +1 for the and +1 for of both mean “1 standard deviation above average” — even though the underlying frequencies were very different.

An author’s profile is the pattern of green and red cells across the row. Documents that share this pattern are likely written in the same style, or by the same author.

Positive z-score: the value is above the mean for that word. The document uses this word more than average.

Negative z-score: the value is below the mean. The document uses this word less than average.

z = 0: exactly at the mean.

Why This Matters for Stylometry

After z-scoring, each word contributes equally to distance measures and downstream analysis (like PCA or clustering).
A word that an author uses at an unusual rate can now “stand out” just as much as any other word’s pattern.
Try editing the raw data in Step 1 and watch how everything updates!

Step 4 Document Distances

Comparing Documents

Now that every feature is on the same scale, we can compare documents using a distance or similarity metric. Choose a metric below to see its formula and how it changes the heatmap.

Metric:

Distance Heatmap

PCA Plot (2D Projection)

Reading the Visualizations

Heatmap: The diagonal is always 0 (zero distance to itself).
Heatmap: Lower values (cooler colors) mean more similar documents.
PCA: Points closer together in the plot are more similar in their z-score profiles. The axes show the percentage of variance explained by each principal component.
Try switching metrics to see how the heatmap ranking changes.

Step 5 Add Your Own Document

Test Against the Corpus

Enter the raw word counts for a new document below. The tool will normalize, compute z-scores (using the corpus statistics from above), and find the nearest match. This does not change the original corpus — your document is only compared against it.