Banding refers to the procedure of grouping test scores into ranges and treating scores within a particular range as equivalent when making personnel decisions. After an organization collects test scores from candidates who applied for a job, a hiring decision must be made using these scores. There are a number of approaches for making these decisions. One common strategy is called top-down selection: Candidate scores are ranked from highest to lowest and organizations start at the top of the list by selecting the candidate with the highest score, then move to the person with the next highest score, and so on down the list. Another common strategy is the practice of setting cutoff scores. A cutoff score involves setting a passing score where candidates at or above this score are labeled as passing the test, whereas those below are labeled as failing. With a cutoff score those passing are treated as if they performed equally on the test. Banding is an alternative to top-down and cutoff score approaches.
Banding involves creating a defined range within which candidate scores are treated as being the same. This is similar to grouping scores into grades as done in many academic settings (e.g., a score between 90% and 100% is considered an A, a score between 80% and 89% is considered a B, etc.). The concept of banding is based on the idea that small differences between test scores may not translate into meaningful differences in expected job performance. For example, a candidate who scores 94% on a test may not perform noticeably better on the job than a candidate who scores 92%. This is because tests are not perfectly predictive of job performance and have varying degrees of measurement error. Banding is the idea of taking into account this imprecision by creating ranges within which test scores are treated as being the same. Thus for candidates who have scores that fall within the same band, the difference between their scores is viewed as meaningless in terms of predicting meaningful differences in job performance, and therefore the candidates are treated as if they scored equivalently on the test.
Purpose of Banding
One key question is, Why would an organization create bands within which candidate scores are considered equivalent? Critics have argued that banding results in a loss of information and has a negative impact on the utility or usefulness of a test. They state that a top-down approach has the highest utility. In response others have noted that although banding may in some circumstances result in a loss of economic utility, this loss may be negligible and must be weighed against other compelling reasons for banding such as the need to increase workforce diversity.
Banding was first proposed as a method for reducing the adverse impact against protected groups (e.g., minorities, women) that is often associated with a top-down approach to selection test decision making. This is because Whites, on average, tend to outperform certain minorities on commonly used written multiple-choice selection tests measuring factors such as cognitive ability and job knowledge. Given this situation, Whites will be chosen at a substantially higher rate in comparison to members of these minority groups if a strict top-down rank order approach is used. Banding was suggested as a viable strategy for addressing this problem. Banding can reduce adverse impact because a band includes lower-scoring as well as higher-scoring individuals; thus when selection decisions are made regarding whom to choose from a band, other factors such as diversity can be taken into account. That is, if candidates that fall within a band are considered equal, an organization may consider the minority group membership of candidates when deciding whom to hire from a given band rather than just selecting the individual with the highest score. Banding allows an organization the flexibility to consider other factors such as diversity when making hiring decisions, whereas a top-down approach does not.
Creating Bands
Many different methods exist for developing bands. For example, expert or managerial judgments could be used to determine what range of scores on a test should be considered equivalent. Another viable approach is to use historical data on how candidates in the past performed on the test and subsequently on the job to determine what bands should be formed. An additional, yet controversial, method for creating bands is the concept of using indicators test reliability as a basis for creating bands. This approach uses statistical significance testing to determine what size the band should be so that test scores that fall within a band are not considered statistically different.
The most common version of this approach leverages a statistic known as the standard error of the difference (SED) to create bands. This SED procedure specifies a range of test scores that will be treated as statistically indistinguishable at some accepted level of confidence. That is, the bandwidth is a function of the standard error of measurement of the test and the desired level of confidence that scores within a band are not statistically different. This approach leverages the psychometric properties of the test in terms of its reliability to determine proper bandwidth. Critics of this approach state that the logic behind it is fatally flawed and that carrying it out to its conclusion would lead to random selection (i.e., selecting individuals completely at random rather than based on their test scores). However, proponents of this approach note that because selection tests are not perfectly reliable, the degree of unreliability should be taken into account when interpreting test scores. They further state that using indicators of unreliability is a more objective and appropriate way to create bands than doing so based on purely arbitrary decisions or solely relying on expert judgments.
Types of Bands
Bands can be either fixed or sliding. Fixed bands use the top score as the starting point, and the first band consists of all scores that fit within the range of the top score minus the bandwidth. For example, if the top score on a test was 96.0 and the bandwidth based on the SED approach was 5.2, the first band would range from 96.0 to 90.8. All scores that fell within this range would be considered part of band one and they would be treated as if they were equivalent. The second band would be the next highest score after band one minus the bandwidth. Therefore, for the example given earlier, the second band would range from 90.7 to 85.5. Additional bands would be created in a similar manner. With a fixed band approach, all individuals within a given band must be selected prior to moving to the next band. That is, band one needs to be completely exhausted before moving to band two.
Sliding bands also use the top score as an initial starting point, and the band is equal to this starting point minus the bandwidth. However, the difference with sliding bands is that when the top score is selected, a new band is formed using the next highest existing score in the band as the starting point. That is, when a top score in a band is chosen, the band slides down and is established using the next highest score as its anchor point. Using the previous example where the top score was 96.0 and the bandwidth was 5.2, individuals would be chosen from within this band until the top score is chosen, at which time the band
would slide down to be anchored on the next highest score. Thus if the individual with a score of 96.0 was chosen and the next highest score was 94.0, the new band would be set at 94.0 minus the bandwidth of 5.2 (i.e., a range of 94.0 to 88.8). Furthermore, when the current high score of 94.0 is chosen, the band would slide again and anchor on the next highest remaining score. The sliding band allows an organization to consider more people more quickly by moving down the rank order list more rapidly. Unlike with fixed bands, sliding bands do not require that a band be exhausted before moving down the list. Instead, when a top score is chosen, the band slides down and allows the organization to consider new individuals for selection.
Effectiveness and Legality of Banding
Research has shown that depending on varying circumstances, such as bandwidth size, banding can be used to reduce adverse impact. An organization can use banding procedures to group scores and then give preference to certain groups when selecting from a band as a means of increasing the diversity of its workforce. Opponents of banding note the loss in utility from not using a top-down approach, but proponents have responded by stating that the possible loss in economic utility is not substantial. One other key issue is whether banding is a legal practice. Most agree that although banding is legal, choosing individuals from a band based on protected group status (e.g., race, gender) could be problematic. The Civil Rights Act prohibits considering factors such as race and gender when making hiring decisions. Although this issue has not been fully resolved, recent court rulings have upheld the use of different types of banding. However, a review of these cases indicates that when protected group status is the only factor used to make choices from a band, it is less likely to be acceptable to the courts than when it is only one of many factors that are used.
References:
- Aguinis, H. (2004). Test-score banding in human resource selection: Technical, legal, and societal issues. Westport, CT: Praeger Publishers.
- Campion, M. A., Outtz, J. L., Zedeck, S., Schmidt, F. L., Kehoe, J. F., Murphy, K. R., et al. (2001). The controversy over score banding in personnel selection: Answers to 10 key questions. Personnel Psychology, 54, 149-185.
- Cascio, W. F., Outtz, J. L., Zedeck, S., & Goldstein I. L. (1991). Statistical implications of six methods of test score use in personnel selection. Human Performance, 4, 233-264.
- Henle, C. A. (2004). Case review of the legal status of banding. Human Performance, 17,415-432.