Parametric tests make assumptions about the data—for example, that observations are independent and normally distributed. You can verify that data is normally distributed with one of the Kolmogorov-Smirnov Tests.
Parametric tests output a p-value to compare to the threshold to determine whether to reject the null hypothesis.
Two-Sample T-Test for Equal Means
- Paired
There must be a one-to-one correspondence between the values in the two samples.
The test assumes the mean differences between corresponding (paired) values are identically distributed normal random independent variables.
The test determines whether the mean differences are significantly different from zero.
- Unpaired
There is no correspondence between the values in the two samples, which may be of equal or unequal size. The test selects the columns with the two unpaired datasets, some of which may be NULL.
The test assumes the following:- The samples are independent of each other.
- Within each sample, values are identically distributed normal random variables.
- The mean differences between the samples are identically distributed normal random independent variables.
- The variances of the samples may be equal (homoscedastic) or unequal (heteroscedastic).
The null hypothesis is that the population means are equal.
- Unpaired with Indicator
Like unpaired, except instead of selecting the columns with the two unpaired datasets, the test selects the column of interest (dependent variable) and an indicator column. If the indicator variable is negative or zero, the test assigns the first variable to the first group. If the indicator variable is positive, the test assigns the first variable to the second group.
The following table shows the formulas that define the two-sample T-tests for unpaired data. (SQL calculates them differently.)
H0: | μ1 = μ2 |
Ha: | μ1 ≠ μ2 |
Test Statistic: | where N1 and N2 are the sample sizes, and are the sample means, and s12 and s22 are the sample variances. |
N-Way F-Test
- F-Test/Analysis of Variance (One-way, with samples of equal or unequal size)
- F-Test/Analysis of Variance (Two-way, with samples of equal size)
- F-Test/Analysis of Variance (Three-way, with samples of equal size)
Use this F-Test on groups defined by the distinct values of the groupby columns. The groups must include two or more treatments.
Use this F-test (also called ANOVA) to determine if significant differences exist among treatment means or interactions. The null hypothesis is no. Acceptance of the null hypothesis implies factor levels and response are unrelated, so further analysis is unnecessary.
- Tukey's Method: Tests all possible pairwise differences of means.
- Scheffe's Method: Tests all possible contrasts at the same time.
- Bonferroni's Method: Tests or puts simultaneous confidence intervals around a preselected group of contrasts.
F-Test/Analysis of Variance (Two-way, with Samples of Unequal Size)
Use this F-Test on the entire dataset. You cannot specify groupby columns. The workaround is run this F-Test multiple times on pre-prepared datasets with group-by variables in each dataset as different constants. The datasets must include two or more treatments.
This test creates a temporary work table in the Result Database and drops it at the end of processing, even if you specify outputtablename.