Plan Test Length with Spearman–Brown

Define the proposed change

Before adding questionnaire items, write down which score is being lengthened, its current number of items, the reliability estimate being used, and the population in which that estimate was obtained. An internal consistency coefficient from one sample is not a fixed property that automatically transfers to another language, setting, or response format.

The Spearman–Brown research discussion describes the classical parallel-parts assumption and extensions beyond it. For a simple planning exercise, assume additional items have comparable measurement properties and errors behave as required by the model. Adding unrelated questions merely to increase a coefficient is not a measurement strategy.

Calculate a transparent scenario

Let r be the current reliability and k the ratio of new to old test length. The forecast is k×r divided by 1+(k−1)×r. In a fictional ten-item instrument with reliability 0.60, doubling the length gives 1.20/1.60 = 0.75. This is a model-based forecast, not a reliability estimate from newly collected responses.

For a target of 0.80, rearranging gives k = target×(1−r)/(r×(1−target)). Here k = 2.6667, corresponding to 26.6667 items. Because items are whole units, a minimum length calculation rounds upward to 27 items. Replacing k with 2.7 gives a forecast of approximately 0.80198; simply rounding to 26 would fall below the target.

The test-length tool makes these scenarios reproducible. Use its output alongside the reliability planning worksheet, which should retain the input coefficient, original length, target, and explanation of why the parallel-item assumption is plausible.

Test the assumption before accepting the forecast

Review the content map of the proposed longer version. Check that extra items preserve the intended construct coverage rather than repeat a narrow phrase. Consider response burden, fatigue, reading difficulty, and whether the new administration changes the conditions under which the original estimate was obtained. These practical changes can invalidate the simple length projection.

Repeat the calculation with plausible lower and higher starting coefficients when the original estimate is uncertain. This sensitivity exercise is preferable to reporting a single item count as guaranteed. A forecasted increase in reliability says nothing by itself about whether the score captures the intended construct or improves decisions.

Record what must happen next

Retain the original and proposed scoring keys, item versions, and missing-item rules. Pilot the revised instrument and estimate reliability using the resulting data and an appropriate model. Evaluate dimensionality and validity separately. Report the planning result as an assumption-dependent estimate, together with the actual evidence needed before using the new score operationally.

Sources checked on 5 November 2026; all numerical examples are fictional and independently recomputed.