Can a chi-squared value be negative?

The standard distribution is supported on nonnegative values, consistent with its sum-of-squares construction.

Does it always have integer degrees of freedom?

No. The gamma-family definition permits positive real values, although many classical constructions give integers.

Is the distribution normal when degrees of freedom are large?

It becomes approximately normal after appropriate centering and scaling, but it is not exactly normal.

AI Glossary / Statistics and probability / Intermediate

Chi-squared distribution

Definition

The chi-squared distribution is a family of continuous probability distributions used to describe certain sums of squared standardized quantities. For a positive integer k, the sum of squares of k independent standard normal variables has a chi-squared distribution with k degrees of freedom. The family provides reference distributions for many statistical tests and confidence procedures.

What it means

Squaring removes signs, so the distribution takes nonnegative values and is not symmetric like a standard normal distribution. Its shape depends on the degrees of freedom. With few degrees of freedom it is strongly right-skewed; as the number increases, the standardized distribution becomes more nearly normal. The distribution is distinct from a chi-square test: a test uses it as a reference when assumptions justify that use. Many different statistics can share this reference without representing the same scientific question.

How it works

If Z1 through Zk are independent normal variables with mean zero and variance one, form Q = Z1 squared + ... + Zk squared. Then Q has mean k and variance 2k. More generally, the family is defined for positive, not necessarily integer degrees of freedom through a gamma distribution with shape k/2 and scale two. To calculate a probability or critical value, specify the degrees of freedom and the relevant tail. In applications, establish whether the reference is exact or only asymptotic. Fitted parameters and constraints can reduce degrees of freedom, while dependence or incorrect standardization can invalidate a simple sum-of-squares argument.

Three practical examples

Categorical-data testing: use an approximate chi-squared reference for Pearson's goodness-of-fit or independence statistic when its sampling assumptions are appropriate.
Variance inference: use an exact scaled-sample-variance result for independent observations from a normal population.
Statistical simulation: generate sums of squared normal values to understand reference distributions and check numerical calculations.

Think of it this way

Think of adding several independent squared standardized deviations. Each contribution is nonnegative, and more contributions generally produce a larger total. The degrees of freedom tell you which reference scale and shape apply; a total cannot be judged as unusually large without that context.

Worked example

Suppose two independent standard normal variables are squared and added, so Q has two degrees of freedom. One hypothetical realization Z1 = 1 and Z2 = 2 gives Q = 1 squared + 2 squared = 5. That is a single realized value, not the distribution's mean, which is two. For two degrees of freedom, the upper-tail probability is P(Q greater than q) = exp(-q/2). Thus P(Q greater than 6) = exp(-3), approximately 0.04979. The 95th percentile solves exp(-q/2) = 0.05, giving q = -2 ln(0.05), approximately 5.99146. Separately, if eleven independent observations come from a normal population with variance sigma squared, then 10 times their sample variance divided by sigma squared has ten degrees of freedom. The sample-mean estimation creates the reduction from eleven to ten. These results require the stated normal and independence assumptions; squaring arbitrary measurements does not automatically produce a chi-squared distribution.

Illustrative case study

Fictional case: Cedar Quality Analytics reviews a dashboard that applies one chi-square cutoff to several unrelated checks. Analysts discover that the checks use different degrees of freedom and that one relies on a small-sample approximation. They replace the universal cutoff with method-specific calculations, document assumptions and verify results using appropriate simulations. The team also separates variance tests from categorical goodness-of-fit tests. This fictional case illustrates correct use of a reference distribution without reporting an actual quality-control outcome.

Common misconceptions and mistakes

A chi-squared random variable is not the square of every normally distributed quantity: the standard central construction requires zero means, unit variances and independence. Squared nonzero-mean standardized normals lead to a noncentral form. A weighted or dependent sum may have another distribution. Degrees of freedom are not always simply the number of observations or table cells. The density at a point is not the probability of that exact value; for a continuous distribution, exact-point probability is zero. A large statistic is unusual only relative to its specified reference distribution and hypothesis.

Business use

Businesses usually encounter the distribution through statistical software rather than direct modeling of operational outcomes. Verify that the software procedure matches the question, that degrees of freedom reflect fitted parameters and constraints, and that the reported tail is correct. Large-sample likelihood-ratio results can fail under boundary conditions or other departures from regular assumptions. AI-generated explanations should state whether the reference law is exact or approximate and avoid translating a p-value into the probability that a business hypothesis is true.

Check your understanding

Q: Why is six an approximately 5% upper-tail cutoff in the worked two-degree example?

Answer: A: Its upper-tail probability is exp(-6/2), about 0.04979. The exact 5% cutoff is slightly smaller, approximately 5.99146.

Related terms

Degrees of freedom
Standard normal distribution
Gamma distribution
Chi-square test
Sample variance
noncentral chi-squared distribution
quantile and upper-tail probability

Learn next

simulate independent standard normals, square and sum them for several k values, and compare empirical means with k and variances with 2k. Then repeat using correlated variables to see why the original reference distribution no longer follows automatically.

Sources

Sources: NIST Engineering Statistics Handbook, chi-square distribution: https://www.itl.nist.gov/div898/handbook/eda/section3/eda3666.htm ; SciPy, chi2 probability distribution and gamma relationship: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.chi2.html ; NIST, normal-population variance testing: https://www.itl.nist.gov/div898/handbook/eda/section3/eda358.htm .

Previous: Chi-square test Next: Child node