Z Statistic to P Value: Formulas, Tests, and Code

Definition in One Paragraph

Converting a z statistic to p value is the mathematical procedure of translating a standardized test score from a standard normal distribution into a tail probability under the assumption that the null hypothesis is true. In frequentist hypothesis testing, a z-statistic measures how many standard errors an observed sample metric—such as a mean or proportion—diverges from the hypothesized null value. The resulting p-value quantifies the probability of obtaining a test statistic at least as extreme as the observed score purely through random sampling variability. Researchers compare this probability against a chosen significance threshold (typically α = .05, .01, or .001) to decide whether data warrant rejecting the null hypothesis.

Formula with Every Symbol Explained

Converting a standardized z-statistic into an inferential p-value relies on the cumulative distribution function (CDF) of the standard normal distribution, denoted by Phi (Φ). For any real value z, the cumulative probability is defined by the integral:

Φ(z) = P(Z ≤ z) = ∫_-∞^z (1 / √(2π)) × e^{–t² / 2} dt

Every symbol in this relationship represents a distinct parameter:

z: Standardized test statistic: z = (x̄ – μ₀) / SE for means, or z = (p̂ – p₀) / SE for proportions.
Φ(z): Cumulative distribution function of N(0, 1), measuring total probability from -∞ up to z.
Z: Standard normal random variable (mean μ = 0, standard deviation σ = 1).
π: Archimedes’ constant (≈ 3.14159).
e: Base of natural logarithms (≈ 2.71828).
t: Dummy integration variable along the standardized score axis.

The exact formula used to determine the p-value depends on the alternative hypothesis (H_a):

Two-Tailed Hypothesis Test

For non-directional tests (H₀: θ = θ₀ vs. H_a: θ ≠ θ₀), deviations in either tail count against the null. By bilateral normal symmetry, the two-tailed p-value doubles the upper tail area:

p = 2 × [1 – Φ(|z|)] = 2 × Φ(-|z|)

Right-Tailed (Upper-Tail) Hypothesis Test

For directional upper tests (H₀: θ ≤ θ₀ vs. H_a: θ > θ₀), only positive deviations provide evidence. The p-value is the upper survival probability:

p = P(Z ≥ z) = 1 – Φ(z)

Left-Tailed (Lower-Tail) Hypothesis Test

For directional lower tests (H₀: θ ≥ θ₀ vs. H_a: θ < θ₀), negative deviations provide evidence. The p-value equals the cumulative probability directly:

p = P(Z ≤ z) = Φ(z)

Worked Example with Arithmetic

Consider a laboratory audit evaluating an automated pipetting workstation. Target dispense volume is μ₀ = 50.0 μL with a known historical standard deviation of σ = 1.2 μL.

An audit team draws an independent sample of n = 36 cycles, yielding sample mean x̄ = 50.45 μL. The team performs a two-tailed statistical test at α = .05 (H₀: μ = 50.0 μL vs. H_a: μ ≠ 50.0 μL):

Step 1: Calculate Standard Error

SE = σ / √n = 1.2 / √36 = 1.2 / 6 = 0.20 μL

Step 2: Calculate the Z-Statistic

z = (x̄ – μ₀) / SE = (50.45 – 50.00) / 0.20 = 0.45 / 0.20 = 2.25

Step 3: Convert Z-Statistic to P-Value

Using a standard normal z-table lookup, the cumulative probability for z = 2.25 is Φ(2.25) = 0.9878. The upper tail area is 1 – 0.9878 = 0.0122. For a two-tailed test:

p = 2 × [1 – Φ(|2.25|)] = 2 × 0.0122 = 0.0244

Step 4: Formal Statistical Decision

Because p = .0244 ≤ .05, the result falls in the rejection region. The auditors reject H₀ at α = .05, concluding that dispense volume differs significantly from 50.0 μL.

Interpretation and Reading Guidance

Following the frequentist definition of a p-value, p = .024 indicates a 2.44% chance of observing a mean deviating by ≥ 0.45 μL in either direction if H₀ were true. It does not measure the probability that H₀ is true, nor effect magnitude. National Institute of Standards and Technology (NIST) benchmarks provide interpretation guidelines:

P-Value Range (p)	Test Statistic Range (\|z\|)	Strength of Evidence	NIST Decision Action (at α = .05)
p < .001	\|z\| ≥ 3.29	Overwhelming evidence against H₀	Reject H₀; highly statistically significant
.001 ≤ p < .01	2.58 ≤ \|z\| < 3.29	Strong evidence against H₀	Reject H₀; significant at α = .01 and α = .05
.01 ≤ p < .05	1.96 ≤ \|z\| < 2.58	Moderate evidence against H₀	Reject H₀; significant at standard α = .05
.05 ≤ p < .10	1.645 ≤ \|z\| < 1.96	Weak or marginal evidence against H₀	Fail to reject H₀ at α = .05
p ≥ .10	\|z\| < 1.645	Inconclusive or no evidence against H₀	Fail to reject H₀; retain null baseline

Assumptions and Common Mistakes

Valid conversion of a z-statistic to an inferential p-value requires satisfying three core assumptions:

Known Population Variance (σ): The population standard deviation must be known. If σ is estimated by sample standard deviation s in small samples, researchers must apply Student’s t-distribution vs normal distribution models to prevent anti-conservative bias.
Normality or Large Sample Size: The population must be approximately normal, or sample size must be sufficiently large (n ≥ 30) for the Central Limit Theorem to ensure sampling distribution normality.
Independent Sampling: Observations must be sampled independently at random.

Common analytical mistakes to avoid:

Omitting Two-Tailed Doubling: Reporting a single-tail probability for a two-tailed test halves the reported p-value and inflates Type I errors.
Sign Reversal with Negative Z-Scores: For a negative score (z = -2.15) in a right-tailed test, reporting cumulative lower tail Φ(-2.15) = .0158 rather than survival area 1 – Φ(-2.15) = .9842 inverts the conclusion.
Confusing Significance with Effect Magnitude: Large samples can yield tiny p-values (p < .001) for negligible practical deviations; always report effect sizes alongside confidence intervals.
Post-Hoc Directionality Switching: Selecting a one-tailed test after inspecting data direction constitutes p-hacking; directionality must be preregistered.

How to Report It (APA Example)

The Publication Manual of the American Psychological Association (APA 7th Edition, Section 6.43–6.44) outlines strict standards for reporting statistical tests:

Italicization: Italicize Latin statistical symbols (z, p, n). Do not italicize Greek letters (α, β, μ) or numbers.
Spacing: Insert spaces around mathematical operators (=, <, >).
Decimal Precision: Report z-scores to two decimal places (e.g., z = 2.25) and exact p-values to two or three decimal places (e.g., p = .024).
Leading Zeros: Omit leading zeros before p-values because probabilities cannot exceed 1.0 (write p = .024, never p = 0.024). Retain leading zeros for z-scores (write z = 0.85).
Truncation Threshold: Never write p = .000; report values under .001 as p < .001.

Standard reporting models include:

Narrative format: “A one-sample z-test indicated that the mean dispense volume was significantly greater than specification, z = 2.25, p = .024, 95% CI [50.06, 50.84].”
Parenthetical format: “The laboratory audit revealed a statistically significant calibration drift (*z* = 2.25, *p* = .024).”
Non-significant outcome: “Dispense volume did not deviate significantly from baseline (*z* = 1.15, *p* = .250).”
High significance: “Automated calibration adjustments substantially reduced cycle variance (*z* = 3.82, *p* < .001).”

How to Calculate It in Excel, R, or Python

Modern analytical software computes exact cumulative probabilities without printed tables.

1. Microsoft Excel

In Microsoft Excel, NORM.S.DIST evaluates the standard normal distribution. Setting cumulative to TRUE returns Φ(z). For further details, consult our reference guide on the standard normal distribution in Excel.

# Two-tailed p-value from z in cell A1:
=2 * (1 - NORM.S.DIST(ABS(A1), TRUE))

# Right-tailed (upper-tail) p-value:
=1 - NORM.S.DIST(A1, TRUE)

# Left-tailed (lower-tail) p-value:
=NORM.S.DIST(A1, TRUE)

2. R Language

In R, pnorm() computes standard normal cumulative probabilities. Using abs() handles two-tailed symmetry, while lower.tail = FALSE evaluates upper survival tails directly:

z <- 2.25

# Two-tailed p-value
p_two_tailed <- 2 * pnorm(abs(z), lower.tail = FALSE)

# Right-tailed p-value (upper tail)
p_right_tailed <- pnorm(z, lower.tail = FALSE)

# Left-tailed p-value (lower tail)
p_left_tailed <- pnorm(z, lower.tail = TRUE)

cat(sprintf("Two-tailed p-value: %.4fn", p_two_tailed))
# Output: Two-tailed p-value: 0.0244

3. Python (SciPy)

In Python, scipy.stats.norm provides cumulative distribution (cdf) and survival functions (sf):

import scipy.stats as stats

z = 2.25

# Two-tailed p-value
p_two_tailed = 2 * stats.norm.sf(abs(z))

# Right-tailed p-value
p_right_tailed = stats.norm.sf(z)

# Left-tailed p-value
p_left_tailed = stats.norm.cdf(z)

print(f"Two-tailed p-value: {p_two_tailed:.4f}")
# Output: Two-tailed p-value: 0.0244

Frequently Asked Questions (FAQ)

What is the difference between a z-statistic and a p-value?

A z-statistic measures the number of standard errors an observed sample metric deviates from the null mean. A p-value is the probability of observing a score that extreme assuming the null is true. The z-statistic is an intermediate test calculation; the p-value is the decision metric evaluated against alpha.

How do you convert a negative z-statistic to a p-value?

For two-tailed tests, apply absolute value: p = 2 × [1 – Φ(|z|)]. By normal curve symmetry, z = -2.25 yields the same two-tailed p-value (p = .0244) as z = +2.25. For left-tailed tests, evaluate p = Φ(z); for right-tailed tests, compute p = 1 – Φ(z).

Why do you multiply by 2 for a two-tailed p-value?

Two-tailed alternative hypotheses test deviations in either direction (larger or smaller than baseline). Because extreme outcomes in both directions contradict the null hypothesis, the total rejection probability must include both symmetric tails.

Can a p-value ever equal zero?

Under a continuous normal distribution, a p-value never equals zero because normal curve tails extend asymptotically to infinity. Software output displaying p = 0.000 reflects decimal truncation; report such results as p < .001 in manuscripts.

What critical z-score corresponds to a p-value of .05?

For two-tailed tests at α = .05, the critical thresholds are z = ±1.96 (2.5% per tail). For one-tailed tests at α = .05, the critical cutoff is z = 1.645 (upper tail) or z = -1.645 (lower tail), placing the full 5% rejection region in one tail.

Sources

National Institute of Standards and Technology (NIST) / SEMATECH. (2012). e-Handbook of Statistical Methods. Sections 1.3.6.6.1 and 7.1.3.1. U.S. Department of Commerce. https://www.itl.nist.gov/div898/handbook/prc/section1/prc131.htm
Illowsky, B., & Dean, S. (2020). Introductory Statistics. OpenStax, Rice University. Chapter 9: Sections 9.3 and 9.4. https://openstax.org/books/introductory-statistics/pages/9-4-rare-events-the-sample-decision-and-conclusion
American Psychological Association. (2020). Publication Manual of the American Psychological Association (7th ed.). Sections 6.43–6.44. APA, Washington, DC. https://apastyle.apa.org/style-grammar-guidelines/numbers/statistics