Multiple Comparisons and Outliers

Dunnett's Test Table: Critical Values for Comparisons with a Control

Critical values of Dunnett's t for comparing several treatment groups with one control group while controlling the familywise error rate.

  • k = 1 to 12 treatments
  • One- and two-sided
  • Exact calculator

Dunnett's Test Table

Jump to a table below. Rows: error df · Columns: number of treatment groups compared with the control (k) · Cells: critical |t|

Two-sided test, α = 0.05

Two-sided test, α = 0.05
Treatment groups compared with control (k) →
df ↓
123456789101112
52.5713.0303.2933.4763.6153.7273.8213.9003.9704.0324.0874.137
62.4472.8633.0993.2633.3883.4893.5733.6443.7073.7633.8123.857
72.3652.7522.9713.1233.2383.3313.4083.4753.5333.5843.6303.671
82.3062.6732.8803.0233.1313.2193.2923.3543.4083.4573.5003.539
92.2622.6142.8122.9483.0523.1353.2053.2643.3163.3623.4033.440
102.2282.5682.7592.8902.9903.0703.1373.1943.2443.2883.3283.364
112.2012.5322.7172.8452.9413.0193.0843.1393.1873.2303.2683.303
122.1792.5022.6832.8072.9012.9773.0403.0943.1403.1823.2193.253
132.1602.4782.6542.7762.8682.9423.0033.0563.1023.1423.1793.212
142.1452.4572.6312.7502.8402.9122.9733.0243.0693.1093.1443.177
152.1312.4392.6102.7272.8162.8872.9462.9973.0413.0803.1153.147
162.1202.4242.5922.7082.7952.8652.9242.9743.0173.0563.0903.121
172.1102.4102.5772.6912.7772.8462.9042.9532.9963.0343.0683.099
182.1012.3992.5632.6762.7612.8302.8872.9352.9773.0153.0483.079
192.0932.3882.5512.6632.7472.8152.8712.9192.9612.9983.0313.061
202.0862.3792.5402.6512.7352.8022.8572.9052.9462.9833.0163.045
242.0642.3492.5072.6142.6952.7602.8142.8602.9002.9362.9682.996
302.0422.3212.4742.5782.6572.7202.7722.8172.8562.8902.9212.949
402.0212.2932.4412.5432.6192.6802.7312.7742.8122.8452.8752.902
602.0002.2652.4102.5082.5822.6422.6912.7332.7692.8012.8302.856
1201.9802.2382.3792.4752.5472.6042.6512.6922.7272.7582.7862.811
∞1.9602.2122.3492.4422.5112.5672.6132.6522.6862.7162.7432.767

Two-sided test, α = 0.01

Two-sided test, α = 0.01
Treatment groups compared with control (k) →
df ↓
123456789101112
54.0324.6274.9755.2195.4065.5575.6835.7925.8875.9716.0476.116
63.7074.2124.5064.7114.8694.9975.1045.1965.2765.3475.4115.469
73.4993.9484.2084.3894.5294.6424.7364.8174.8884.9515.0085.059
83.3553.7664.0024.1684.2954.3974.4834.5574.6214.6794.7304.777
93.2503.6333.8534.0064.1244.2194.2994.3674.4274.4804.5284.571
103.1693.5313.7393.8833.9944.0844.1594.2234.2794.3294.3744.415
113.1063.4523.6493.7873.8923.9784.0494.1104.1644.2114.2544.293
123.0553.3873.5773.7093.8113.8923.9604.0194.0704.1164.1574.194
133.0123.3353.5183.6463.7433.8223.8883.9443.9944.0384.0774.113
142.9773.2903.4683.5923.6873.7633.8273.8823.9303.9724.0114.045
152.9473.2533.4263.5473.6393.7133.7763.8293.8753.9173.9543.988
162.9213.2203.3903.5083.5983.6713.7313.7833.8293.8693.9053.938
172.8983.1923.3593.4743.5633.6343.6933.7443.7883.8283.8633.896
182.8783.1683.3313.4453.5313.6013.6593.7093.7533.7923.8263.858
192.8613.1463.3073.4193.5043.5723.6303.6793.7223.7603.7943.825
202.8453.1273.2853.3953.4793.5473.6033.6513.6943.7313.7653.795
242.7973.0673.2183.3233.4033.4683.5213.5673.6083.6433.6753.704
302.7503.0093.1543.2543.3303.3913.4423.4863.5243.5583.5893.616
402.7042.9523.0913.1863.2593.3173.3663.4083.4443.4763.5053.531
602.6602.8983.0303.1213.1903.2463.2923.3323.3663.3973.4243.449
1202.6172.8452.9723.0593.1243.1773.2213.2593.2913.3203.3463.370
∞2.5762.7942.9152.9983.0603.1103.1523.1883.2193.2463.2713.293

One-sided test, α = 0.05

One-sided test, α = 0.05
Treatment groups compared with control (k) →
df ↓
123456789101112
52.0152.4402.6812.8482.9763.0783.1633.2363.3003.3563.4073.453
61.9432.3372.5582.7112.8272.9202.9983.0643.1223.1743.2203.261
71.8952.2672.4762.6192.7282.8152.8882.9503.0043.0523.0953.134
81.8602.2172.4162.5532.6572.7402.8092.8682.9192.9653.0063.043
91.8332.1802.3722.5042.6042.6842.7502.8072.8562.9002.9392.974
101.8122.1512.3382.4662.5622.6402.7042.7592.8072.8492.8872.921
111.7962.1272.3102.4352.5292.6052.6672.7212.7682.8092.8462.879
121.7822.1082.2872.4102.5022.5762.6382.6902.7352.7762.8122.845
131.7712.0922.2692.3892.4802.5522.6132.6642.7092.7482.7842.816
141.7612.0792.2532.3712.4612.5322.5922.6422.6862.7252.7602.791
151.7532.0672.2392.3562.4442.5152.5732.6232.6672.7052.7402.771
161.7462.0572.2272.3432.4302.5002.5582.6072.6502.6882.7222.753
171.7402.0482.2172.3322.4182.4872.5442.5932.6352.6732.7062.737
181.7342.0402.2082.3212.4072.4752.5322.5802.6222.6602.6932.723
191.7292.0332.2002.3122.3972.4652.5212.5692.6112.6482.6812.711
201.7252.0272.1922.3042.3892.4562.5122.5592.6012.6372.6702.699
241.7112.0082.1702.2792.3622.4272.4822.5282.5692.6042.6362.665
301.6971.9892.1472.2552.3352.3992.4532.4982.5372.5722.6032.631
401.6841.9702.1252.2302.3092.3722.4242.4682.5062.5402.5702.598
601.6711.9522.1042.2072.2842.3452.3952.4392.4762.5092.5382.565
1201.6581.9342.0832.1832.2582.3182.3682.4102.4462.4782.5072.533
∞1.6451.9162.0622.1602.2342.2922.3402.3812.4172.4482.4762.502

One-sided test, α = 0.01

One-sided test, α = 0.01
Treatment groups compared with control (k) →
df ↓
123456789101112
53.3653.9004.2114.4294.5974.7334.8464.9445.0305.1065.1745.236
63.1433.6073.8764.0644.2084.3244.4224.5054.5794.6444.7024.755
72.9983.4183.6603.8283.9574.0624.1494.2244.2904.3484.4004.448
82.8963.2863.5093.6653.7843.8803.9604.0294.0894.1434.1914.234
92.8213.1893.3993.5453.6563.7463.8213.8863.9423.9924.0374.078
102.7643.1153.3143.4533.5593.6443.7153.7773.8303.8783.9203.959
112.7183.0563.2473.3803.4823.5643.6323.6903.7423.7873.8283.865
122.6813.0083.1933.3223.4203.4993.5643.6213.6703.7143.7533.789
132.6502.9693.1493.2743.3683.4453.5093.5633.6113.6543.6923.726
142.6242.9363.1113.2333.3253.4003.4623.5153.5623.6033.6403.674
152.6022.9083.0803.1983.2893.3623.4223.4743.5203.5603.5963.629
162.5832.8843.0523.1693.2573.3293.3883.4393.4843.5233.5593.591
172.5672.8623.0283.1433.2303.3003.3583.4093.4523.4913.5263.557
182.5522.8443.0073.1203.2063.2753.3323.3823.4253.4633.4973.528
192.5392.8272.9893.1003.1853.2533.3093.3583.4003.4383.4723.502
202.5282.8132.9723.0823.1663.2333.2893.3373.3783.4163.4493.479
242.4922.7672.9213.0273.1073.1713.2253.2713.3113.3463.3783.407
302.4572.7232.8712.9733.0503.1113.1633.2073.2453.2793.3103.338
402.4232.6802.8222.9202.9943.0533.1033.1453.1823.2143.2443.270
602.3902.6382.7752.8692.9402.9973.0443.0853.1203.1513.1793.205
1202.3582.5972.7292.8202.8882.9422.9883.0263.0603.0903.1173.141
∞2.3262.5582.6852.7722.8372.8892.9332.9703.0023.0313.0563.079

Tip: click any value to highlight its row and column.

How to read this table

  1. Count k, the number of treatment groups compared with the control. The control itself is not counted.
  2. Find the error degrees of freedom from the ANOVA: N − (k + 1) for a one-way design with N subjects in total.
  3. For each treatment, compute t = (x̄ᵢ − x̄₀) / √(MSE (1/nᵢ + 1/n₀)).
  4. A treatment differs from control if |t| (two-sided) or t (one-sided) exceeds the table value.

Worked example

A dose-ranging study randomizes 7 patients to each of placebo and three doses (N = 28), so k = 3 and error df = 28 − 4 = 24. Row 24, column k = 3 of the two-sided α = 0.05 table gives 2.507. With t-statistics of 1.85, 2.71 and 3.40 for the low, middle and high doses, the middle and high doses differ significantly from placebo, with the familywise error rate held at 5% across all three comparisons.

For comparison, an unadjusted t-test would use t = 2.064 on 24 df, and a Bonferroni correction for three comparisons would use 2.574. Dunnett's value is smaller than Bonferroni's because it accounts for the correlation created by the shared control group.

Click the underlined links to highlight the value in the table.

Calculator

For values between the rows and columns of the table. Results update as you type.

Formula and how the values were computed

tᵢ = (x̄ᵢ − x̄₀) / √(MSE (1/nᵢ + 1/n₀))  ·  P(max |Tᵢ| < d) = 1 − α

Critical values are quantiles of the maximum of k equicorrelated t-statistics (correlation 0.5, which applies when all groups are the same size), computed by numerical integration over the multivariate t distribution and rounded to three decimals. The k = 1 column equals Student's t.

Frequently asked questions

What if the group sizes are unequal?

The correlation between comparisons is then not exactly 0.5 and the table is approximate. The approximation is good when the control group is not much larger than the others; for other designs, use software that computes the exact multivariate t probability.

Should the control group be larger?

Often, yes. Dunnett showed that allocating about √k times as many subjects to the control as to each treatment maximizes power for a fixed total sample size.

Dunnett or Tukey?

Use Dunnett when only comparisons with the control matter. It has more power for those comparisons than Tukey's HSD, which also tests every pair of treatments.

Related calculators and tutorials