What Is a Z Statistic?
A z statistic is a z score with a job in a courtroom: it is the standardised distance between what a sample produced and what a null hypothesis predicted, used as evidence. Same arithmetic family, same table — different question being answered.
Short answer: The z statistic is z = (sample result − null value) / (standard error). If the null hypothesis is true, it follows the standard normal distribution, so the z table converts it into a p-value. A plain z score describes one value in a dataset; a z statistic judges a claim about a population.
The difference, side by side
- What gets standardised: a z score standardises an individual observation (X against μ, σ). A z statistic standardises a summary — a sample mean or proportion — against the null value and the standard error.
- What it is compared with: a z score is interpreted as a percentile (where does this value stand?). A z statistic is interpreted as a tail probability under a hypothesis (how surprising is this sample if the claim holds?).
- What follows from it: a z score ends in a description. A z statistic ends in a decision framework: a p-value, a significance verdict, a confidence interval standing nearby.
The shared machinery is why the two share a name and a table — and why they are so often conflated. The cleanest mental split: z scores belong to descriptive statistics, z statistics to inferential statistics. A worked construction of one lives in the post on calculating the z test statistic: sample mean 504 against a claimed 500, standard error 2, z statistic 2.00, p = 0.0456. Nothing in that computation describes any individual item; every number is about the claim.
What its size tells you
Under the null, the z statistic behaves like a draw from the standard normal — which makes its interpretation a pure table exercise. |z| near 0: the sample agrees with the claim. |z| around 1: ordinary wobble (two-tailed p ≈ 0.32). |z| = 2.45: p ≈ 0.0143 — evidence worth publishing at conventional levels. |z| beyond 4: p below 0.0001, at which point experienced analysts check the data and the model before the claim, because statistics that extreme usually indict the assumptions (independence, known σ, honest sampling) before the hypothesis. That scepticism is professional, not perverse: the statistic is only as meaningful as the standard error underneath it, and the standard error is only as meaningful as the study design underneath that.
Related questions
Frequently asked questions
Is a z statistic the same as a z score?
Not quite. Both are standardised distances read on the same distribution, but a z score standardises an individual value for description, while a z statistic standardises a sample summary against a null hypothesis for inference — different inputs (σ vs standard error) and different outputs (percentile vs p-value).
What does a negative z statistic mean?
The sample came out below the null value. For a two-tailed test only its size matters; for a left-tailed test it is exactly the evidence sought. Direction is information — a strongly negative statistic in a test predicting an increase is evidence against the predicted direction, not near-miss support for it.
Also asked as: what is a z statistic in statistics · what is the z statistic · z statistic definition · what is z test statistic.