PHARM/PCS V5 · Procedure 44 ANOVA and Comparisons
Dunnett's Test (Comparison with a Control)
Compare each treatment group with a single control after ANOVA, the more powerful choice when control comparisons are all you need.
What it computes
Compares each of p treatment group means against a control group mean using Dunnett's method. The control group is the first block entered. Dunnett's d is computed automatically from the multivariate t-distribution, so there is no need to enter values from Table A.9.
Inputs and outputs
You provide
- The control group first, then each treatment group, as separate blocks
- A confidence level of 95% or 99%
- Optionally, your own value of Dunnett's d
PHARM/PCS reports
- The pooled variance and Dunnett's d
- The allowance A
- Confidence limits on each treatment's difference from the control
- A chart of each treatment's difference from the control
When to use it
The right post-hoc after a significant ANOVA (Procedure 33) when every comparison you care about is treatment-vs-control (several doses against vehicle, say). It's more powerful than all-pairs tests (Procedures 36-38) precisely because it tests fewer, targeted comparisons.
Reading the results
Enter the control first: it's the reference every other group is measured against, and the wrong order invalidates the comparisons. Because Dunnett's only controls error for the control-vs-treatment family, don't reach into its output to also compare treatments against each other; that's what the range tests are for.
No Table A.9 needed
The 1986 program asked you to look up Dunnett's d. V5 computes it from the multivariate t distribution, the same approach R and SciPy use, and matches Tables A.9a and A.9b to within 0.01 across 130 cells.
This page describes the procedure. Calculations run inside the licensed PHARM/PCS V5 application. PHARM/PCS is a trademark of MCS Scientific LLC.