Z-Test

views updated

Z-Test

In scientific and social investigations, researchers may need to make decisions through statistical-hypothesis testing guided by underlying theory and empirical observations. The Z -test is one of the most popular techniques for statistical inference based on the assumption of normal distribution. Many social and natural phenomena follow the law of normal (Gaussian) distribution, which was discovered by Carl F. Gauss (1777–1855), a German mathematician, in the early nineteenth century. The normal distribution is one of the fundamental statistical distributions used in many fields of research, and it has a bellshaped density function with μ (mean) representing the central location and σ² (variance) measuring the dispersion. The normal density function is

A general normal random variable X with a mean of μ and variance σ² can be rescaled into the standard normal random variable Z with mean 0 and variance 1 using

where σ is the standard deviation of X. The observed value of Z is called the Z score. Almost all introductory statistics books provide the table for the probability of {Z = c } or its variants for many convenient values of c (Agresti and Finlay 1997). These values are available in all statistical software packages.

Statistical hypothesis testing and inference on the population mean are usually performed through a sample of random variables observed from the population. Let X ₁, X ₂, …, X_n be n independent and identically distributed normal random variables with a mean of μ and a variance of σ², such as the gas mileages of a particular type of vehicle, the annual average income of households in a city, or the vital signs of patients under various treatments.

Denote the sample mean Then

is distributed as the standard normal random variable, where is the standard deviation of X̄. This result provides the base for a one-sample Z -test on the mean (μ ) of the study population when σ² is known. To test the null hypothesis H ₀: μ = μ ₀ (for example, to test if the mean gas mileage [μ ] is 25 [μ ₀] miles per gallon for a particular type of vehicle), one can form the Z -test statistic as

One would reject the null hypothesis H ₀ in favor of the alternative H_a : μ ≠ μ ₀ (the two-sided Z -test) when the observed value of Z is significantly different from 0. In many situations, one may be interested in a one-sided alternative (H_a : μ < μ ₀ or H_a : μ > μ ₀).

In rejecting or accepting the null hypothesis, one could commit two types of errors. The type I error (a ) is the probability of rejecting the null hypothesis when it is true, and the type II error (β ) is the probability of accepting the null hypothesis when it is false. The p -value is the probability of the test statistic as contradictory to H ₀ as the observed Z value. A detailed study on statistical hypothesis testing is given by Erich Lehmann and Joseph P. Romano (2005). For the two-sided Z -test, one may reject μ = μ ₀ in favor of μ ≠ μ ₀ if the observed Z value of the test statistic satisfies

The type I error for this test is less than 0.05. The upper bound of the type I error is called the size, and 1-β is called the power of the test. For a test, if the p -value is less than the size, one may reject the null hypothesis H ₀ . A commonly used size is 0.05. For the two-sided Z -test with size 0.05, the critical region is {|Z | > 1.96}. The Z test is closely related to the constructing of confidence intervals. For example, the 95 percent confidence interval for the mean μ is

for the two-sided estimation. If X ₁, X ₂…, X_n are not independent and identically normally distributed, under some conditions, the central limit theorem shows that

is approximately standard normal when n is large (typically n ≥ 30). That is, one can still use the Z -test when n is large.

In many applications, the standard deviation of the population is unknown. In these cases, one can replace σ with the sample standard deviation s and form a test statistic as

where

is the sample variance. The test follows the t -distribution of n -1 degrees of freedom. As n increases, t converges to Z. Hence, even when the standard deviation is unknown, the Z -test can be used if n is large. In fact, the estimates of parameters from many parametric models, such as regression models, are approximately normally distributed. The Z -test is therefore applicable for statistical inference of these parameters. When comparing the means (μ ₁ and μ ₂) of two normally distributed populations, and , one can construct a two-sample Z statistic for testing H ₀: μ ₁ – μ ₂ = μ ₀:

Where and X_i, Y_i are independent random samples from these two populations respectively.

The two-sample Z -test can be carried out in the same way as the one-sample Z -test. Similar to the case of the one-sample Z -test, the two-sample Z -test corresponds to the two-sample t -test when the population variances are replaced by their sample variances. However, when n and m are large, there is not much difference between the two sample Z -test and the two-sample t -test. The two-sample t -test is a special case of Analysis of Variance (ANOVA), which compares the means of multiple populations. For a small sample size, it is preferable to use nonparametric methods instead of the Z -test, particularly when it is difficult to verify the assumption of normality.

SEE ALSO Student’s T-Statistic; Test Statistics