Z-Value Table

Look up cumulative probabilities for any z-score from -3.99 to +3.99 in the standard normal distribution. This interactive table also includes a z-score calculator that converts raw values to z-scores and finds probabilities directly.

Estimated reading time: 28 minutes. This page covers the full z-table, how to read it, worked examples, connections to hypothesis testing, confidence intervals, effect sizes, regression, and sampling distributions.

Z-Score Calculator

How to Use This Z-Value Table

There are two ways to use this page. The quick way is to type your z-score into the calculator above and get an instant result with a visual bell curve. The reference way is to scroll down to the full z-tables and look up values manually, just like you would with a printed table in a textbook.

Using the Calculator

Enter a z-score (like 1.96 or -2.33) and select what type of probability you want. "Left tail" gives you the area to the left of your z-score, which is the standard cumulative probability shown in most z-tables. "Right tail" gives you the area to the right. "Between" gives you the area between -z and +z, which is useful for confidence intervals. "Outside" gives you the combined area in both tails, which is the p-value for a two-tailed hypothesis test.

Using the Raw Value Converter

If you have a raw data value and want to find its z-score, enter the value along with the population mean and standard deviation. The calculator applies the formula z = (x - mu) / sigma and then looks up the probability automatically. This is particularly useful for homework problems and quick data analysis.

Using the Full Tables Below

The positive z-table covers z-scores from 0.0 to 3.9. The negative z-table covers z-scores from -3.9 to 0.0. Find the row that matches the integer and first decimal of your z-score, then find the column matching the second decimal place. The intersecting cell is your cumulative probability.

Positive Z-Value Table (z = 0.0 to 3.9)

This table shows the cumulative probability P(Z ≤ z) for positive z-scores. Values represent the total area under the standard normal curve to the left of the z-score.

Negative Z-Value Table (z = -3.9 to 0.0)

This table shows the cumulative probability P(Z ≤ z) for negative z-scores. These probabilities are always less than 0.5 since negative z-scores fall below the mean.

How the Standard Normal Distribution Works

The standard normal distribution is a bell-shaped probability distribution with a mean of 0 and a standard deviation of 1. It is the foundation of modern statistics, and nearly every statistical test connects to it in some way.

The Bell Curve

The probability density function (PDF) of the standard normal distribution produces the familiar bell shape. The curve is perfectly symmetric around z = 0 (the mean). It peaks at z = 0 and tapers off toward zero as z moves away from the center in either direction. The total area under the entire curve equals exactly 1, which represents 100% probability.

The 68-95-99.7 Rule

One of the most useful properties of the normal distribution is the empirical rule: 68.27% of values fall within 1 standard deviation of the mean (z between -1 and 1). 95.45% fall within 2 standard deviations (z between -2 and 2). 99.73% fall within 3 standard deviations (z between -3 and 3). This rule gives you a quick mental model for interpreting z-scores without even looking at a table. A z-score of 2 means the value is in roughly the top or bottom 2.5% of the distribution.

Cumulative Distribution Function (CDF)

The z-table shows the cumulative distribution function, which is the integral of the PDF from negative infinity to z. There is no closed-form solution for this integral, which is why we need tables (or computers) to look up values. The values in the z-table are computed using numerical integration methods to at least four decimal places of precision.

Mathematical Definition

The PDF of the standard normal distribution is phi(z) = (1 / sqrt(2 * pi)) * e^(-z^2 / 2). The CDF is Phi(z) = integral from negative infinity to z of phi(t) dt. Because this integral cannot be solved in closed form, mathematicians have developed various approximation methods. The most common include Taylor series expansions, rational function approximations, and the error function (erf). This calculator uses the Abramowitz and Stegun approximation, which provides accuracy to 7 significant digits with a single polynomial evaluation.

How to Read the Z-Table Step by Step

Reading a z-table is a skill that every statistics student needs. Here is the exact process with a worked example.

Example: Find P(Z ≤ 1.96)

Step 1: Take your z-score (1.96) and split it into the row part (1.9) and the column part (0.06). Step 2: Go to the positive z-table. Find the row labeled 1.9 in the leftmost column. Step 3: Move across that row to the column labeled .06. Step 4: The value at the intersection is 0.9750. This means P(Z ≤ 1.96) = 0.9750, or 97.50% of values in a standard normal distribution fall at or below z = 1.96.

Finding Right-Tail Probabilities

The table only shows left-tail (cumulative) probabilities. To find the right-tail probability P(Z ≥ z), subtract the table value from 1. For example: P(Z ≥ 1.96) = 1 - 0.9750 = 0.0250. So 2.50% of values fall above z = 1.96.

Finding Probabilities Between Two Z-Scores

To find P(a ≤ Z ≤ b), look up both values and subtract. For example: P(-1.96 ≤ Z ≤ 1.96) = 0.9750 - 0.0250 = 0.9500. This is exactly 95%, which is why z = 1.96 is used for 95% confidence intervals.

Interpolation for Three-Decimal Z-Scores

Standard z-tables only go to two decimal places. If you need the probability for z = 1.965, you can interpolate between the values for z = 1.96 and z = 1.97. P(Z ≤ 1.96) = 0.9750 and P(Z ≤ 1.97) = 0.9756. Linear interpolation gives P(Z ≤ 1.965) = 0.9750 + 0.5 * (0.9756 - 0.9750) = 0.9753. For most practical purposes, rounding to two decimals is sufficient, but interpolation can improve accuracy when needed.

Common Z-Scores and Their Probabilities

Certain z-scores appear so frequently in statistics that they are worth memorizing. Here are the ones you will encounter most often in coursework, research papers, and data analysis.

Z-Score	P(Z ≤ z)	P(Z ≥ z)	Common Use
-2.576	0.0050	0.9950	99% CI lower bound
-1.960	0.0250	0.9750	95% CI lower bound
-1.645	0.0500	0.9500	90% CI lower / one-tail 5%
-1.282	0.1000	0.9000	80% CI lower
-1.000	0.1587	0.8413	1 std dev below mean
0.000	0.5000	0.5000	Mean (50th percentile)
1.000	0.8413	0.1587	1 std dev above mean
1.282	0.9000	0.1000	90th percentile
1.645	0.9500	0.0500	90% CI upper / one-tail 5%
1.960	0.9750	0.0250	95% CI upper bound
2.326	0.9900	0.0100	99th percentile
2.576	0.9950	0.0050	99% CI upper bound
3.090	0.9990	0.0010	99.9th percentile
3.291	0.9995	0.0005	99.9% CI upper bound

In pharmaceutical research, the z-score of 1.96 appears in nearly every clinical trial report. Regulatory agencies like the FDA require 95% confidence intervals for efficacy estimates, making this the single most referenced z-value in applied statistics. The z-score of 2.576 for 99% confidence is standard in high-stakes engineering applications where the cost of failure is extreme, such as aerospace and nuclear safety analysis.

Z-Scores and Confidence Intervals

Confidence intervals are one of the most practical applications of z-scores. When you hear "we are 95% confident," the z-score 1.96 is doing the work behind the scenes.

The Confidence Interval Formula

For a population mean with known standard deviation: CI = sample mean +/- z* times (sigma / sqrt(n)). Where z* is the critical z-value for your confidence level, sigma is the population standard deviation, and n is the sample size.

The z* values for common confidence levels are: 80% confidence uses z* = 1.282. 90% confidence uses z* = 1.645. 95% confidence uses z* = 1.960. 99% confidence uses z* = 2.576. 99.9% confidence uses z* = 3.291.

Worked Example: Manufacturing Quality

A factory produces bolts with a known standard deviation of 0.5 mm. A sample of 100 bolts has a mean length of 25.3 mm. Find the 95% confidence interval for the true mean. CI = 25.3 +/- 1.96 times (0.5 / sqrt(100)) = 25.3 +/- 1.96 times 0.05 = 25.3 +/- 0.098. The 95% confidence interval is (25.202 mm, 25.398 mm). We can be 95% confident that the true population mean bolt length falls within this range.

Margin of Error and Sample Size

The margin of error (ME) is half the width of the confidence interval: ME = z* times (sigma / sqrt(n)). You can rearrange this formula to find the required sample size for a desired margin of error: n = (z* times sigma / ME)^2. For example, to estimate a population mean within 0.02 mm with 95% confidence when sigma = 0.5: n = (1.96 * 0.5 / 0.02)^2 = (49)^2 = 2401 bolts. This sample size calculation is one of the most common uses of z-values in experimental design and survey planning.

Confidence Interval for Proportions

For proportions (like survey results), the formula is: CI = p-hat +/- z* times sqrt(p-hat * (1 - p-hat) / n). Where p-hat is the sample proportion. For example, if 540 out of 1000 surveyed voters favor a candidate (p-hat = 0.54), the 95% CI is: 0.54 +/- 1.96 * sqrt(0.54 * 0.46 / 1000) = 0.54 +/- 0.031. So the 95% confidence interval is (0.509, 0.571), and you can report that between 50.9% and 57.1% of voters favor this candidate.

Practical Examples

Here are several real-world scenarios where z-scores and this table come into play.

Example 1: Exam Scores

A statistics exam has a mean score of 72 and a standard deviation of 8. What percentage of students scored above 88? First, calculate the z-score: z = (88 - 72) / 8 = 2.00. Look up z = 2.00 in the table: P(Z ≤ 2.00) = 0.9772. P(Z ≥ 2.00) = 1 - 0.9772 = 0.0228, or about 2.28% of students scored above 88.

Example 2: Quality Control

A machine fills bottles to a mean of 500 mL with a standard deviation of 3 mL. What fraction of bottles have less than 494 mL? z = (494 - 500) / 3 = -2.00. P(Z ≤ -2.00) = 0.0228, so about 2.28% of bottles are underfilled below 494 mL. If the factory produces 10,000 bottles per day, that means approximately 228 underfilled bottles daily.

Example 3: Height Distribution

Adult male height in the United States follows approximately a normal distribution with a mean of 69.1 inches and standard deviation of 2.9 inches. What is the probability that a randomly selected man is between 66 and 72 inches tall? z1 = (66 - 69.1) / 2.9 = -1.07, z2 = (72 - 69.1) / 2.9 = 1.00. P(-1.07 ≤ Z ≤ 1.00) = 0.8413 - 0.1423 = 0.6990. About 69.9% of adult men are between 5 feet 6 inches and 6 feet tall.

Example 4: SAT Scores

SAT scores are designed to follow a normal distribution with a mean of 1060 and standard deviation of 217. What z-score corresponds to a score of 1400? z = (1400 - 1060) / 217 = 1.57. P(Z ≤ 1.57) = 0.9418, so a 1400 SAT score is at approximately the 94th percentile. This means a student scoring 1400 performed better than about 94% of all test takers.

Example 5: A/B Testing

Your website control group has a conversion rate of 3.2% and the test group has 3.8%. With sample sizes of 5,000 each, is this difference statistically significant at the 5% level? The pooled proportion p = (160 + 190) / 10000 = 0.035. Standard error = sqrt(0.035 * 0.965 * (1/5000 + 1/5000)) = 0.00368. Z = (0.038 - 0.032) / 0.00368 = 1.63. For a two-tailed test at 5%, we need z greater than 1.96. Since 1.63 is less than 1.96, the difference is not statistically significant. You would need more data to confirm whether the 0.6 percentage point difference is real.

Example 6: Process Monitoring

A call center monitors average hold time. The target is 120 seconds with a known standard deviation of 30 seconds. Today's sample of 50 calls shows a mean hold time of 132 seconds. Is this significantly above the target? z = (132 - 120) / (30 / sqrt(50)) = 12 / 4.243 = 2.83. P(Z ≥ 2.83) = 1 - 0.9977 = 0.0023. At the 5% level (or even 1% level), the hold time is significantly above target. Management should investigate the cause.

Common Mistakes When Using Z-Tables

After teaching statistics for several years, I have seen the same errors come up repeatedly. Here are the ones that trip up most students.

Mistake 1: Confusing Left-Tail and Right-Tail

The standard z-table gives left-tail probabilities. If a problem asks "what is the probability of scoring above z = 1.5?", you need to subtract from 1. Forgetting this step is the single most common z-table error. A quick check: if your z-score is positive and your answer is greater than 0.5, you probably have the left-tail probability when you wanted the right-tail.

Mistake 2: Using the Wrong Table Sign

Negative z-scores have their own section in the table. Looking up z = -1.5 in the positive table (or vice versa) gives a completely wrong answer. Always check the sign first. Some tables only show positive z-scores and rely on the symmetry property: P(Z ≤ -z) = P(Z ≥ z) = 1 - P(Z ≤ z).

Mistake 3: Rounding the Z-Score Prematurely

Calculate your z-score to at least two decimal places before looking it up. Rounding z = 1.96 to z = 2.0 changes the probability from 0.9750 to 0.9772, which might not seem like much but can significantly affect hypothesis test conclusions near the boundary of significance.

Mistake 4: Assuming All Data Is Normally Distributed

The z-table only applies to data that follows (or approximately follows) a normal distribution. Income data, for example, is heavily right-skewed and should not be analyzed with z-scores directly. Check for normality before applying these methods. Common checks include histograms, Q-Q plots, and the Shapiro-Wilk test.

Mistake 5: Using Z When You Should Use T

If you do not know the population standard deviation and are estimating it from a small sample (n less than 30), you should use the t-distribution, not the z-distribution. The t-distribution has heavier tails, reflecting the extra uncertainty from estimating the standard deviation.

Mistake 6: Forgetting to Account for Two Tails

In a two-tailed hypothesis test at the 5% significance level, each tail gets 2.5%, not 5%. The critical z-value is 1.96 (not 1.645, which is for one-tailed tests). Mixing up one-tailed and two-tailed critical values leads to incorrect conclusions about statistical significance.

History of the Normal Distribution and Z-Tables

The normal distribution has a rich mathematical history that spans over 300 years. Understanding this history gives context to why the z-table looks and works the way it does.

Abraham de Moivre, a French-born mathematician working in England, first described the normal curve in 1733 as an approximation to the binomial distribution. He was trying to find a faster way to calculate probabilities for games of chance. His work was published in "The Doctrine of Chances." De Moivre's contribution was major because it showed that a continuous curve could approximate a discrete distribution, a concept that remains central to statistics today.

Carl Friedrich Gauss, the German mathematician, independently developed the normal distribution in the early 1800s in the context of astronomical observations. He showed that measurement errors follow a bell-shaped curve, and the distribution became known as the "Gaussian distribution" in his honor. Gauss used the normal distribution to predict the orbit of the asteroid Ceres, which had been observed only briefly before being lost behind the sun. His predictions proved remarkably accurate.

Pierre-Simon Laplace proved the Central Limit Theorem, which states that the sum of many independent random variables tends toward a normal distribution regardless of the original distributions. This theorem is the main reason the normal distribution appears so frequently in nature and statistics. It explains why heights, weights, test scores, measurement errors, and many biological measurements tend to be approximately normally distributed.

The first published z-tables appeared in the late 1800s and early 1900s. Before computers, these tables were computed by hand using Taylor series approximations and were essential reference tools for scientists and engineers. Karl Pearson's tables, published in the early 1900s, were among the most widely used. The British Association for the Advancement of Science published tables computed to 15 decimal places. Today, computers can calculate these probabilities to arbitrary precision, but z-tables remain a standard teaching tool because they build intuition about the shape and properties of the normal distribution.

Z-Scores in Hypothesis Testing

Hypothesis testing is where z-scores go from abstract numbers to practical decision-making tools. The z-test is one of the fundamental procedures in statistics.

The Z-Test Procedure

Step 1: State the null hypothesis (H0) and alternative hypothesis (H1). For example, H0: the population mean equals 50, H1: the population mean does not equal 50. Step 2: Choose a significance level (alpha), typically 0.05 for the 5% level. Step 3: Calculate the test statistic: z = (sample mean - hypothesized mean) / (sigma / sqrt(n)). Step 4: Find the p-value using the z-table. For a two-tailed test, p = 2 * P(Z ≥ |z|). Step 5: If p is less than alpha, reject H0. If p is greater than or equal to alpha, fail to reject H0.

One-Tailed vs. Two-Tailed Tests

A one-tailed test checks if the parameter is greater than (or less than) a specific value. A two-tailed test checks if the parameter is different from a specific value in either direction. The choice between one-tailed and two-tailed should be made before collecting data, based on the research question. Changing from two-tailed to one-tailed after seeing the data is a form of p-hacking that inflates the false positive rate.

For a two-tailed test at alpha = 0.05, the critical z-values are +/- 1.96. For a one-tailed test at alpha = 0.05, the critical z-value is 1.645 (upper tail) or -1.645 (lower tail). The two-tailed test is more conservative because it splits the significance level between both tails.

P-Values and Z-Scores

The p-value is the probability of observing a test statistic as extreme as (or more extreme than) the one calculated, assuming the null hypothesis is true. For a z-test, the p-value comes directly from the z-table. If z = 2.5 in a one-tailed test, p = P(Z ≥ 2.5) = 1 - 0.9938 = 0.0062. This means there is only a 0.62% chance of seeing results this extreme if the null hypothesis were true, providing strong evidence against H0.

Multiple Testing Corrections

When conducting many hypothesis tests simultaneously (common in genomics, neuroimaging, and market research), the chance of at least one false positive increases dramatically. If you run 20 independent tests at alpha = 0.05, the probability of at least one false positive is 1 - (0.95)^20 = 0.64, or 64%. The Bonferroni correction addresses this by dividing alpha by the number of tests. For 20 tests, each test uses alpha = 0.05/20 = 0.0025, requiring z greater than 2.81 for significance. The False Discovery Rate (FDR) procedure by Benjamini and Hochberg offers a less conservative alternative that controls the expected proportion of false positives among rejected hypotheses.

Z-Scores and Percentile Lookup

A percentile tells you what percentage of values in a distribution fall at or below a given value. Z-scores and percentiles are directly connected through the cumulative distribution function. Here is a detailed lookup table for converting between the two.

Percentile	Z-Score	Percentile	Z-Score
1st	-2.326	55th	0.126
2nd	-2.054	60th	0.253
5th	-1.645	65th	0.385
10th	-1.282	70th	0.524
15th	-1.036	75th	0.674
20th	-0.842	80th	0.842
25th	-0.674	85th	1.036
30th	-0.524	90th	1.282
35th	-0.385	95th	1.645
40th	-0.253	97.5th	1.960
45th	-0.126	99th	2.326
50th	0.000	99.9th	3.090

This table works in reverse too. If you have a z-score and want the percentile, find the closest z-score in the right column and read the percentile from the left column. For example, a z-score of 1.282 corresponds to the 90th percentile, meaning the value is higher than 90% of all values in the distribution.

Percentiles are widely used in pediatric growth charts, standardized test score reports, and economic data. When a child's height is at the "75th percentile," it means their height corresponds to a z-score of approximately 0.674, and they are taller than 75% of children their age. Similarly, when household income is reported at the "90th percentile," the z-score equivalent is 1.282 standard deviations above the mean income.

Normal Approximation to the Binomial

One of the most important applications of the z-table is approximating binomial probabilities. When the sample size is large enough, the binomial distribution approaches the normal distribution, and you can use the z-table instead of calculating binomial probabilities directly.

When to Use the Approximation

The rule of thumb is that the approximation works well when both np ≥ 5 and n(1-p) ≥ 5, where n is the number of trials and p is the probability of success. For example, flipping a fair coin 100 times (n="100," p="0.5):" np = 50 and n(1-p) = 50, both well above 5, so the approximation is excellent.

How to Apply It

For a binomial random variable X with parameters n and p, the standardized value is: z = (X - np) / sqrt(np(1-p)). For example, what is the probability of getting 60 or more heads in 100 fair coin flips? The mean is np = 50, the standard deviation is sqrt(100 * 0.5 * 0.5) = 5. Z = (60 - 50) / 5 = 2.00. Looking up z = 2.00: P(Z ≥ 2.00) = 1 - 0.9772 = 0.0228. So there is about a 2.3% chance of getting 60 or more heads, which matches closely with the exact binomial probability.

Continuity Correction

Because the binomial is a discrete distribution and the normal is continuous, adding a continuity correction of 0.5 improves accuracy. Instead of P(X ≥ 60), calculate P(X ≥ 59.5): z = (59.5 - 50) / 5 = 1.90. P(Z ≥ 1.90) = 1 - 0.9713 = 0.0287. This is closer to the exact binomial answer of 0.0284. The continuity correction matters most when n is small or p is far from 0.5.

Normal Approximation to the Poisson

The normal distribution also approximates the Poisson distribution when the rate parameter lambda is large (typically lambda ≥ 10). For a Poisson random variable with parameter lambda, the z-score is: z = (X - lambda) / sqrt(lambda). For example, if a website averages 50 visits per hour (lambda = 50), what is the probability of getting more than 65 visits in an hour? z = (65 - 50) / sqrt(50) = 15 / 7.07 = 2.12. P(Z ≥ 2.12) = 1 - 0.9830 = 0.0170, about 1.7% chance.

Effect Sizes and Z-Scores

In research, z-scores are closely related to the concept of effect sizes, which measure the practical significance of a result beyond just statistical significance.

Cohen's d and Z-Scores

Cohen's d is the most common effect size measure for comparing two group means. It is calculated as: d = (Mean1 - Mean2) / pooled standard deviation. The z-score for a one-sample test is related to d by: z = d * sqrt(n). This means that a small effect size (d = 0.2) requires a large sample to produce a statistically significant z-score, while a large effect size (d = 0.8) can be detected with a smaller sample.

Standard Effect Size Benchmarks

Jacob Cohen proposed these benchmarks for interpreting effect sizes: Small effect: d = 0.2 (e.g., the difference in height between 15-year-old and 16-year-old girls). Medium effect: d = 0.5 (e.g., the difference in IQ between clerical and professional workers). Large effect: d = 0.8 (e.g., the difference in height between 13-year-old and 18-year-old girls). These benchmarks help researchers understand whether a statistically significant result is also practically meaningful. A drug that produces a statistically significant result (p less than 0.05) but has an effect size of d = 0.1 may not be clinically meaningful.

Power Analysis Using Z-Scores

Statistical power is the probability of correctly rejecting a false null hypothesis. Power analysis uses z-scores to determine the required sample size for a study. The formula for a two-sample z-test is: n = ((z_alpha + z_beta) / d)^2 per group. Where z_alpha is the z-score for the significance level (1.96 for alpha = 0.05, two-tailed), z_beta is the z-score for the desired power (0.84 for 80% power), and d is the expected effect size. For a medium effect (d = 0.5) with 80% power at the 5% level: n = ((1.96 + 0.84) / 0.5)^2 = (5.6)^2 = 31.36, so you need about 32 participants per group.

Converting Between Effect Size Measures

Researchers sometimes need to convert between different effect size metrics. Cohen's d can be converted to the correlation coefficient r using: r = d / sqrt(d^2 + 4). For d = 0.5: r = 0.5 / sqrt(0.25 + 4) = 0.5 / 2.06 = 0.243. The odds ratio (OR) can be approximated from d using: OR = exp(d * pi / sqrt(3)), which gives OR = exp(0.5 * 1.814) = exp(0.907) = 2.48 for d = 0.5.

Z-Distribution vs. Other Distributions

The z-distribution (standard normal) is just one of several probability distributions used in statistics. Understanding when to use each one prevents common errors.

Z vs. T Distribution

The t-distribution looks similar to the z-distribution but has heavier tails. It is used when the population standard deviation is unknown and must be estimated from the sample. As the sample size increases, the t-distribution approaches the z-distribution. At n = 30, they are nearly identical. At n = 5, the t-distribution has noticeably fatter tails, reflecting greater uncertainty with small samples. A practical rule: use z when sigma is known and n is large (over 30). Use t when sigma is unknown and you are using the sample standard deviation s.

Z vs. Chi-Square Distribution

The chi-square distribution is the distribution of the sum of squared z-scores. If Z1, Z2, ..., Zk are independent standard normal variables, then Z1^2 + Z2^2 + ... + Zk^2 follows a chi-square distribution with k degrees of freedom. Chi-square tests are used for testing independence in contingency tables, goodness-of-fit tests, and testing variance.

Z vs. F Distribution

The F-distribution is the ratio of two chi-square distributions (each divided by their degrees of freedom). It is used in ANOVA (Analysis of Variance) to compare means across multiple groups and in regression analysis to test overall model significance. While the z-test compares one or two means, the F-test can compare three or more simultaneously.

When Each Distribution Applies

One mean, sigma known, large n: z-test. One mean, sigma unknown: t-test. Two means, sigma known: z-test. Two means, sigma unknown: t-test. Comparing three or more means: F-test (ANOVA). Testing proportions: z-test. Testing variance: chi-square test. Testing independence: chi-square test. Each test ultimately connects back to probability theory and the normal distribution through the Central Limit Theorem, which is why the z-table remains the starting point for understanding all of inferential statistics.

Sampling Distributions and the Central Limit Theorem

The Central Limit Theorem (CLT) is the single most important theorem in statistics, and it is the reason the z-table is relevant to so many different problems. Understanding sampling distributions is essential for using z-scores correctly in practice.

What Is a Sampling Distribution?

A sampling distribution is the distribution of a statistic (like the sample mean) computed from many random samples of the same size drawn from the same population. Imagine drawing 1,000 random samples of 50 people each from a large population and computing the mean height of each sample. The 1,000 sample means would form a distribution of their own, and this distribution would be approximately normal with a mean equal to the population mean and a standard deviation of sigma / sqrt(n).

The Standard Error

The standard deviation of the sampling distribution of the mean is called the standard error (SE). SE = sigma / sqrt(n). This formula reveals two important facts. First, larger samples produce smaller standard errors, meaning the sample mean is a more precise estimate of the population mean. Second, the standard error decreases with the square root of n, so quadrupling the sample size only halves the standard error. Going from n = 25 to n = 100 cuts the SE in half, but going from n = 100 to n = 200 only reduces it by a factor of 1 / sqrt(2) = 0.707.

CLT in Practice

The CLT works for virtually any population distribution as long as the population has a finite variance. The required sample size for the normal approximation to be reasonable depends on how skewed the original distribution is. For symmetric distributions, n = 10 may be sufficient. For moderately skewed distributions, n = 30 is the traditional rule of thumb. For highly skewed distributions (like income or insurance claims), n = 100 or more may be needed.

Why the CLT Matters for Z-Scores

Without the CLT, the z-table would only be useful for data that is already normally distributed. With the CLT, the z-table applies to the sampling distribution of the mean from any population, as long as the sample is large enough. This is why z-tests and confidence intervals work for income data (which is skewed), customer spending data (which is skewed), and many other non-normal variables. We are not assuming the data itself is normal. We are using the fact that the sample mean is approximately normal, which the CLT guarantees.

Z-Scores in Regression Analysis

Z-scores play an important role in regression analysis, both in standardizing coefficients and in evaluating model assumptions.

Standardized Regression Coefficients (Beta Weights)

In multiple regression, the raw coefficients (b values) are in the units of the original variables, making them hard to compare. Standardized coefficients (beta weights) are obtained by converting all variables to z-scores before running the regression. A beta weight of 0.45 means that a one standard deviation increase in the predictor is associated with a 0.45 standard deviation increase in the outcome, controlling for other predictors. This allows direct comparison of the relative importance of different predictors.

Residual Analysis

After fitting a regression model, you check the residuals (prediction errors) to assess model fit. Standardized residuals divide each residual by the standard deviation of residuals, converting them to z-scores. A standardized residual greater than 2 or less than -2 indicates a potential outlier. A standardized residual greater than 3 or less than -3 is a strong outlier that warrants investigation. If the model is correct and the assumptions are met, standardized residuals should follow an approximately standard normal distribution.

Testing Individual Coefficients

In large-sample regression, each coefficient's significance is tested using a z-test (or a t-test, which converges to the z-test for large samples). The test statistic is: z = b / SE(b), where b is the estimated coefficient and SE(b) is its standard error. If |z| ≥ 1.96, the coefficient is statistically significant at the 5% level. Software output typically shows the z-value (or t-value) and the corresponding p-value for each predictor in the model.

Logistic Regression and Wald Z-Tests

In logistic regression, which models binary outcomes (yes/no, success/failure), the Wald test uses z-scores to test whether each coefficient is significantly different from zero. The Wald z-statistic is: z = beta-hat / SE(beta-hat). For example, if a logistic regression of disease status on smoking gives beta-hat = 0.85 and SE = 0.32, then z = 0.85 / 0.32 = 2.66. Since 2.66 is greater than 1.96, smoking is a statistically significant predictor at the 5% level. The odds ratio is exp(0.85) = 2.34, meaning smokers have 2.34 times the odds of disease compared to non-smokers.

Bayesian Perspective on Z-Scores

While z-scores are fundamentally a frequentist tool, understanding the Bayesian perspective provides additional insight into what z-scores really tell us and what their limitations are.

The Frequentist Interpretation

In the frequentist framework, a z-score of 1.96 means that if the null hypothesis is true, there is a 5% chance of observing a result this extreme (two-tailed). The p-value tells you P(data | H0), the probability of the data given that the null hypothesis is true. It does not tell you P(H0 | data), the probability that the null hypothesis is true given the data. This distinction is subtle but critically important.

What the P-Value Does Not Tell You

A common misinterpretation is that a p-value of 0.05 means there is a 95% chance the alternative hypothesis is true. This is incorrect. The p-value makes no statement about the probability of the hypotheses. It only tells you how surprising your data would be under the null hypothesis. Two studies can have the same p-value but very different probabilities of the alternative being true, depending on the prior probability and the study design.

Bayes Factors and Z-Scores

A Bayes factor quantifies the relative evidence for one hypothesis versus another. It can be approximately related to z-scores: for a standard two-sided z-test, the Bayes factor in favor of H1 is approximately BF = sqrt(n) * exp(z^2 / 2) / z, under certain assumptions. At z = 1.96 (p = 0.05), the Bayes factor is often only about 3 to 5, meaning the data are only 3 to 5 times more likely under H1 than H0. Many Bayesian statisticians consider this "weak" evidence. A z-score of 2.5 (p = 0.012) typically yields a Bayes factor around 10 to 15, which is considered "moderate" evidence. This explains why some researchers advocate for a significance threshold of p less than 0.005 (z greater than 2.81) rather than p less than 0.05.

Practical Implications

The Bayesian perspective reminds us that a statistically significant z-score is not the same as proof that an effect exists. The strength of evidence depends on the prior probability of the hypothesis, the sample size, the effect size, and the quality of the study design. In fields where most tested hypotheses are false (like early-stage drug screening), even a z-score of 2.0 may correspond to a relatively low probability that the effect is real. In fields where most tested hypotheses are plausible (like replication studies of well-established effects), the same z-score provides much stronger evidence.

Frequently Asked Questions

Related Tools

References and Sources

• Wikipedia: Standard Normal Table - Mathematical background on z-tables and cumulative distribution functions.
• Wikipedia: Normal Distribution - Full treatment of the normal distribution including history, properties, and applications.
• NIST: Guidelines for Evaluating Uncertainty in Measurement - National Institute of Standards and Technology guidelines on statistical uncertainty and normal distribution applications.
• Wikipedia: Central Limit Theorem - The foundational theorem explaining why the normal distribution appears so frequently.

Browser Compatibility

This z-value table and calculator works in all modern web browsers including Chrome, Firefox, Safari, Edge, and Opera on desktop and mobile devices. The interactive bell curve visualization uses the HTML5 Canvas API. JavaScript must be enabled for the calculator and table generation features. No plugins or downloads are required.