Spearman-Brown Test Length Calculator

How the Spearman-Brown length estimate works

The Spearman-Brown prophecy formula estimates how reliability changes when a test is lengthened or shortened with items that behave like the existing items. It uses the current reliability and the length multiplier, not a response matrix or a new item analysis.

Let N0 be the current number of items, Np the planned number of items, and r the current reliability as a decimal. The planned length multiplier is mp = Np / N0. Predicted reliability is:

rp = (mp x r) / (1 + (mp - 1) x r)

For reverse planning, let t be the target reliability as a decimal. Solving the same equation for the required multiplier gives:

k = t(1 - r) / (r(1 - t))

The exact target length is N* = N0 x k. Because a test cannot contain a fraction of an item, this calculator rounds that exact length upward and then recalculates reliability at the rounded whole-item length. That final check matters: rounding 51.43 items to 52 items gives a slightly higher reliability than the exact target, while rounding down would miss it.

What moves the result most

The current reliability has the strongest effect. Raising a test from 0.70 to 0.80 takes far fewer added items than raising the same test from 0.90 to 0.95, because the denominator includes the remaining unreliability. The current item count also matters directly: the same multiplier means more additional items when the existing test is longer.

If the target is below the current reliability, the inverse formula can produce a multiplier below one. In that case the calculator reports the shortest whole-item length that still meets the lower target, so the item change can be negative. It never reports fewer than one item.

What the calculator leaves out

The estimate assumes the added or removed items are parallel to the existing items: similar variance, quality, difficulty, and relationship to the same construct. It does not test dimensionality, item bias, content coverage, sampling error, or whether the entered coefficient is appropriate for the planned use. A higher predicted reliability also does not prove validity or fairness.

Worked example

Suppose a 30-item test has reliability 70%, and the goal is at least 80%. The inverse formula gives k = 0.80(1 - 0.70) / (0.70(1 - 0.80)) = 1.7143. The exact length is 30 x 1.7143 = 51.43 items, so the minimum whole-item length is 52 items.

The required item change is 52 - 30 = 22 additional items. Reapplying the prediction formula at 52 items gives about 80.18% reliability. If the planned length is 45 items instead, the predicted reliability is about 77.78%.