Multiple Comparisons and Outliers

Studentized Range q Table for Tukey's HSD (α = 0.01)

Critical values of the studentized range distribution for Tukey's honestly significant difference test at α = 0.01.

  • k = 2 to 20 groups
  • df 1 to ∞
  • HSD calculator

Tukey q Table, α = 0.01

Columns: number of group means compared (k) · Rows: error degrees of freedom · Cells: critical q

Studentized Range (q) Table for Tukey's HSD, α = 0.01
Number of groups (k) →
df \ k ↓
234567891011121314151617181920
190.02135.04164.26185.58202.21215.77227.17236.97245.54253.15259.98266.16271.81277.00281.80286.26290.43294.33298.00
214.0419.0222.2924.7226.6328.2029.5330.6831.6932.5933.4034.1334.8135.4336.0036.5337.0337.5037.94
38.2610.6212.1713.3214.2415.0015.6416.2016.6917.1317.5317.8918.2218.5218.8019.0719.3119.5519.76
46.518.129.179.9610.5811.1011.5411.9312.2612.5712.8413.0913.3213.5313.7313.9114.0814.2414.39
55.706.987.808.428.919.329.679.9710.2410.4810.7010.8911.0811.2411.4011.5511.6811.8111.93
65.246.337.037.567.978.328.618.879.109.309.489.659.819.9510.0810.2110.3210.4310.54
74.955.926.547.007.377.687.948.178.378.558.718.869.009.129.249.359.469.559.65
84.755.646.206.626.967.247.477.687.868.038.188.318.448.558.668.768.858.949.03
94.605.435.966.356.666.917.137.337.497.657.787.918.038.138.238.338.418.498.57
104.485.275.776.146.436.676.877.057.217.367.497.607.717.817.917.998.088.158.23
114.395.155.625.976.256.486.676.846.997.137.257.367.467.567.657.737.817.887.95
124.325.055.505.846.106.326.516.676.816.947.067.177.267.367.447.527.597.667.73
134.264.965.405.735.986.196.376.536.676.796.907.017.107.197.277.357.427.487.55
144.214.895.325.635.886.086.266.416.546.666.776.876.967.057.137.207.277.337.39
154.174.845.255.565.805.996.166.316.446.556.666.766.846.937.007.077.147.207.26
164.134.795.195.495.725.926.086.226.356.466.566.666.746.826.906.977.037.097.15
174.104.745.145.435.665.856.016.156.276.386.486.576.666.736.816.876.947.007.05
184.074.705.095.385.605.795.946.086.206.316.416.506.586.656.736.796.856.916.97
194.054.675.055.335.555.735.896.026.146.256.346.436.516.586.656.726.786.846.89
204.024.645.025.295.515.695.845.976.096.196.286.376.456.526.596.656.716.776.82
243.964.554.915.175.375.545.695.815.926.026.116.196.266.336.396.456.516.566.61
303.894.454.805.055.245.405.545.655.765.855.936.016.086.146.206.266.316.366.41
403.824.374.704.935.115.265.395.505.605.695.765.835.905.966.026.076.126.166.21
603.764.284.594.824.995.135.255.365.455.535.605.675.735.785.845.895.935.976.01
1203.704.204.504.714.875.015.125.215.305.375.445.505.565.615.665.715.755.795.83
∞3.644.124.404.604.764.884.995.085.165.235.295.355.405.455.495.545.575.615.65

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

How to read this table

  1. k is the number of group means being compared (not k − 1).
  2. The error df is the within-groups df from the ANOVA: N − k for a one-way design.
  3. Read q where k and df meet, then compute HSD = q × √(MSE / n), where MSE is the error mean square and n is the number of subjects per group.
  4. Any two group means that differ by more than the HSD are significantly different.

Worked example

An ANOVA of 3 dose groups with 11 patients each has error df = 33 − 3 = 30 and MSE = 12.0. Column k = 3, row 30 gives q = 4.45, so HSD = 4.45 × √(12.0 / 11) = 4.65. Group means that differ by more than 4.65 units are significantly different at α = 0.01, with the family-wise error rate controlled across all three pairwise comparisons.

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

HSD = qα, k, df × √(MSE / n)  ·  Tukey–Kramer (unequal n): q × √((MSE / 2)(1/nᵢ + 1/nⱼ))

Each cell is the upper-α quantile of the studentized range distribution for k means and the given error df, computed by numerical integration and rounded to two decimals.

Frequently asked questions

What if my groups have different sizes?

Use the Tukey–Kramer adjustment: for each pair, replace √(MSE / n) with √((MSE / 2)(1/nᵢ + 1/nⱼ)). The q value from the table is unchanged.

My error df is not listed. What should I do?

Use the next smaller df, which gives a slightly larger, conservative q. Interpolating linearly in 1/df between neighbouring rows is also accurate.

Tukey or Dunnett?

Tukey's HSD compares every pair of groups. When each treatment is compared only with a single control, Dunnett's test is more powerful.

Related calculators and tutorials