One-Way Analysis of Variance - Comparisons
main topic     example      

Stat > ANOVA > One-Way > Comparisons

Choose the comparison procedure based on the group means that you want to compare and the error rate that you want to specify. Procedures that do not compare all pairs of groups are more powerful.

The choice of comparison method depends on your desired inference. Using Tukey's all-differences approach is inefficient when Dunnett's or Hsu's MCB is suitable, because Tukey's confidence intervals are wider and the hypothesis tests less powerful for a given family error rate. For the same reasons, Hsu's MCB is superior to Dunnett's if you want to eliminate levels that are not the best and to identify those that are best or close to the best. The choice between Tukey's and Fisher's depends on whether you want to specify the family or individual error rate.

If you do not assume equal variances in the Options dialog, you must use the Games-Howell method.

Dialog box items

Error rate for comparisons: Enter an error rate between 0.001 and 50. The default is 5. The error rate is the individual error rate for Fisher's method and the family error rate for all other methods.

Comparison procedures assuming equal variances

Tukey: Check to compare all pairs of groups, while controlling the family error rate. Tukey's provides confidence intervals that are less precise and hypothesis tests that are less powerful than either Dunnett's or Hsu's MCB.

Fisher: Check to compare all pairs of groups, while controlling the individual error rate. Fisher's is less common than Tukey's because it does not control the family error rate, which often increases to unacceptable levels.

Dunnett: Check to compare the treatment groups to a control group, while controlling the family error rate.

Control group level: Choose the value for the control group.

Hsu MCB: Check to identify the groups that are the best. Hsu MCB compares each group to the group with either the largest or the smallest mean, while controlling the family error rate.

Best:

Largest mean is best: Choose to compare each group to the group with the largest mean.

Smallest mean is best: Choose to compare each group to the group with the smallest mean.

Comparison procedures not assuming equal variances

Games-Howell: Check to compare all pairs of groups, while controlling the family error rate.

Results

Interval plot for differences of means: Check to display confidence intervals for the differences between the group means for each selected comparison method. Use the confidence intervals to determine likely ranges for the differences and to assess the practical significance of the differences.

Grouping information: Check to display a table that highlights the significant and non-significant comparisons for each selected comparison method. No grouping information table is presented for Hsu's MCB.

Tests: Check to display the hypothesis test form of the comparison output, which includes the differences of the means and the adjusted p-values.