How to Use the Spearman-Brown Formula to Improve Test Reliability

The Spearman-Brown formula is used to predict the reliability of a test after changing the length of the test.

The Formal Definition and Variables

The core mathematical structure of the Spearman-Brown formula establishes a proportional relationship between the degree of change in test length and the resulting change in reliability. This prediction is contingent upon knowing the reliability of the original test and the quantitative factor by which the test length is modified.

The formula used to calculate the predicted reliability (r_predicted) is:

Predicted reliability = kr / (1 + (k-1)r)

The variables within the equation are defined as follows:

• k: This critical factor represents the ratio of the new test length to the original test length. It quantifies the change in size. For example, if the original test consists of 10 questions and the proposed new test will contain 15 questions, k is calculated as 15/10, yielding a factor of 1.5. If the test is shortened by half, k = 0.5.
• r: This variable denotes the known or observed reliability coefficient of the original test before any modification. This value must be derived from actual test administration data, typically ranging from 0 (no reliability) to 1 (perfect reliability), with higher values indicating greater measurement consistency.

The formula demonstrates a positive, but non-linear, relationship: as the test length increases (k > 1), the predicted reliability also increases, but at a decreasing rate. Conversely, shortening the test (k < 1) invariably leads to a reduction in predicted consistency.

Example: How to Use the Spearman-Brown Formula to Increase Reliability

Consider a practical scenario where a testing organization needs to significantly boost the quality of its assessment data. They must determine the reliability gain achieved by extending the assessment time and content.

Suppose a company uses a 15-item test to assess employee satisfaction, which is known to have an observed reliability (r) of 0.74. While 0.74 is acceptable, the company aims for higher precision. They decide that increasing the length of the test to 30 items is feasible. The objective is to calculate the predicted reliability of this new, longer test using the Spearman-Brown Formula.

First, we calculate the length factor k by dividing the new item count by the original item count: k = 30 / 15 = 2. We then substitute k = 2 and the original reliability r = 0.74 into the predictive equation:

• Predicted reliability = kr / (1 + (k-1)r)
• Predicted reliability = 2 * 0.74 / (1 + (2 – 1) * 0.74)
• Predicted reliability = 1.48 / (1 + 0.74)
• Predicted reliability = 1.48 / 1.74
• Predicted reliability ≈ 0.85

By doubling the test length, the predicted reliability of the new test increases substantially to 0.85. This figure provides strong evidence that the proposed lengthening strategy will successfully improve the precision and consistency of the measurement tool.

Understanding the Inverse Formula for Required Length

One of the most powerful uses of the Spearman-Brown Formula in professional psychometrics involves determining the required length increase necessary to meet a predefined reliability standard. Rather than asking what reliability results from a length change, this inverse application asks what length change (k) is needed to achieve a target reliability (r_target).

The formula can be algebraically rearranged to solve for k, yielding: k = r_target * (1 – r_original) / (r_original * (1 – r_target)). This rearrangement is crucial for test development teams operating under strict quality mandates, as it transforms reliability optimization from a trial-and-error process into a precise calculation.

For example, if the original test (r = 0.74) must achieve a minimum regulatory standard of r_target = 0.90, substituting these values into the inverse formula would yield the required k factor. If this calculation results in k = 4.2, it means the 15-item test must be multiplied by 4.2, necessitating a new test length of 63 items. This allows organizations to precisely scope the item development and administration efforts required for certification or regulatory compliance.

Critical Assumptions: The Parallelism Requirement

The theoretical gains predicted by the Spearman-Brown formula are entirely dependent on a critical assumption: the principle of parallelism. This principle mandates that any items added to the test must be statistically and psychometrically identical to the original items.

For the predicted reliability to hold true, the new items must measure the exact same construct, have the same true score variance, and possess equal error variance as the items already in the test. In reality, creating items that are perfectly parallel is exceptionally challenging. If the new items are easier, harder, or measure slightly different aspects of the underlying trait, the observed reliability of the final, longer test will be lower than the mathematical prediction.

It is therefore imperative that if a test is lengthened, the new items or questions added must be of equal difficulty and quality to the existing items. Failure to maintain item quality and homogeneity means the increase in test length is merely adding noise, violating the core assumption of the formula and rendering the predicted reliability inaccurate.

Cautions on Using the Spearman-Brown Formula

Based on the mathematical structure of the formula, we can observe that increasing the number of items on a test (k > 1) will mathematically increase the predicted reliability of the test, assuming the original reliability is greater than zero. However, relying solely on this mathematical logic without considering practical constraints can lead to flawed test design.

For example, consider a minimal increase in length. Suppose we increase the number of items on the 15-item test from the previous example to 16 items. Then we calculate k as 16/15, which is approximately 1.067.

The predicted reliability would still show an increase:

• Predicted reliability = kr / (1 + (k-1)r)
• Predicted reliability = 1.067 * 0.74 / (1 + (1.067 – 1) * 0.74)
• Predicted reliability ≈ 0.752

The new test has a predicted reliability of 0.752, which is higher than the reliability of 0.74 on the original test. This suggests that even small increases yield small gains. Using this logic, we might think that increasing the length of the test by a massive amount of items is always a good idea because we could theoretically push the reliability closer and closer to 1. However, this ignores the practical limitations related to administration and test-taker behavior.

Practical Limitations: Fatigue and Diminishing Returns

The theoretical prediction of infinitely increasing reliability as test length increases is constrained by severe practical limits. As a test becomes excessively long, external factors introduce measurement error that is not accounted for by the Spearman-Brown model, leading to observed reliability falling far short of the prediction.

1. The Introduction of Fatigue Effects:

If a test is too long—stretching beyond acceptable time limits—individuals taking the assessment may experience mental fatigue, boredom, or impatience. This phenomenon, known as fatigue effects, causes test-takers to rush, guess randomly, or lose focus on the later items. These inconsistent responses introduce substantial measurement error, causing the observed reliability of the lengthy test to decrease significantly, contradicting the positive prediction of the formula.

In essence, the relationship between test length and reliability is curvilinear, not strictly linear. Reliability increases rapidly at first with added items, but eventually levels off and may even decline when the test length becomes prohibitive due to factors like fatigue. Test developers must therefore prioritize finding the optimal length that maximizes consistency while minimizing test-taker burden.

The following tutorials explain other commonly used terms in statistics: