A test statistic is a numerical summary calculated from observed data using the formula for a selected statistical test. It turns the sample into a quantity that can be evaluated against what would be expected under the null hypothesis. Its value alone is not a conclusion: it must be interpreted using the test’s decision rule or null distribution. section 4.1.4
Different tests use different statistics because research questions, data types, and assumptions differ. The statistic is fixed by the chosen test and the observed data; its calculation is not altered after analysis begins. Software can perform the arithmetic, but choosing and interpreting the test remain essential. section 4.1.4 section 4.2.1
Beverage fill volumes
A manufacturer samples 20 containers and finds a mean fill of 11.85 ounces and a sample standard deviation of 0.30 ounces. To assess a null target mean of 12 ounces, a one-sample t-test uses t = (11.85 − 12) / (0.30 / √20) ≈ −2.24. The statistic expresses the departure from the target relative to the estimated variability of the sample mean. In a two-tailed test at α = 0.05, it is compared with critical values in both tails; the rule carries a risk of Type I error. section 4.2.1
Clinical trial recoveries
In a trial, 14 of 20 patients recover. An exact binomial test uses 14 recoveries as the observed statistic and evaluates the probability of 14 or more under a null recovery probability of 50%. That probability is approximately 0.057. At α = 0.05, the rejection region begins at 15 recoveries, so the observed count does not meet the rule for rejecting the null hypothesis. section 4.2.3
A likelihood ratio test provides another use: comparing the maximized likelihoods of nested models to assess whether additional parameters improve the fit sufficiently. Under standard regularity conditions, the resulting statistic has an approximate chi-square distribution; that approximation is part of the basis for assessing significance. section 3.7.1
