t-Test

A t-test is a statistical process to assess the probability that a particular characteristic (the mean) of two populations is different. It is particularly useful when data are available for only a portion of one or both populations (known as a sample). In such a case, a t-test will enable an estimate of whether any differences in means between the two groups are reflective of differences in the respective populations or are simply due to chance. This statistical process is called a t-test because it uses a t-distribution to generate the relevant probabilities, typically summarized in a t-table.

There are many types of t-tests, each of which is appropriate for a particular context and nature of the data to be analyzed. Each t-test has its own set of assumptions that should be checked prior to its use. For instance, comparing the average value of a characteristic for independent samples (i.e. when individuals in each sample have no systematic relationship to one another) and for dependent samples (such as when related pairs are divided into two groups) requires different kinds of t-tests.

Frequently, a t-test will involve assessing whether a sample comes from a specifie population or whether average values between two samples indicate differences in the respective populations. In the latter case, three factors determine the results of a t-test: the size of the difference in average value between the two groups; the number of data points, or observations, in each group; and the amount of variation in values within each group.

A t-test is conducted by calculating a t-statistic and by using a t-table to interpret its value. The equation for the t-statistic depends on the context and nature of data. For instance, when comparing the likelihood that a sample belongs to a particular population, the equation for the t-test is where x is the mean for the sample, μ is the known value for the population, s is the standard deviation of the sample, and n is the number of data points in the sample.

As an example, in a study that examines the effectiveness of a new math curriculum, researchers might ask whether the curriculum is related to students' state standardized math test scores. A survey might be used to collect state test score data for students who participate in the new curriculum and for those who use a different curriculum. The researchers would want to make generalizations for all students who use and do not use the curriculum. However, because gathering the test score data for every student might be difficult and expensive, the researchers might send the questionnaire to only a sample of students in each group.

After calculating the average test score of each sample, the researchers could use a t-test to estimate the likelihood that the difference between the two samples' average test scores was really reflective of different test scores between the populations and not simply due to chance. If, for instance, the averages of the two samples were very similar, data were only available for a handful of students in each sample, and students' test scores in each sample varied greatly, then a t-test would likely show that the two populations did not necessarily have different average test scores and that the differences in the samples were simply due to chance. This would be shown by a very low value of the t-statistic. If, on the other hand, the difference in average test scores between the samples was great, there were many students in each sample, and students' scores within each sample did not vary greatly, a t-test would support the conclusion that the populations' average test scores were truly different from one another. This would be evidenced by a very high value of the i-statistic.

...