In this issue
Profiles in military psychology
By Michael G. Rumsey
It is now over 100 years since the birth of Hubert E. Brogden in 1913 and more than 50 years since his most influential publications. Despite this passage of time, his contributions remain highly relevant to the science and practice of psychology, in general, and military psychology, in particular, in some respects even more so than when they were first communicated. Robert Perloff (1992, p. 263) described him as a person whose “insights and creativity in personnel, personality, and leadership are probably, even today, without parallel Although Brogden's contributions were indeed diverse, it is in the area of personnel, particularly personnel selection and classification, in which his contributions have proven most significant and long lasting. As much as anyone, he could be considered the father of modern utility analysis. That is, he revolutionized the study of selection and classification by showing how the value of these activities could be quantified. His demonstration that the correlation coefficient could be used as a direct measure of selection efficiency was a groundbreaking revelation. Similarly, he advanced the study of personnel classification by both linking classification to mean predicted performance and by developing a means of calculating this metric that went beyond predictive validity. He developed the idea of quantifying the benefits of both selection and classification in terms of actual dollars. This allowed not only a comparison between costs and benefits but also a means of combining different criteria on a common scale.
Brogden received his doctorate from the University of Illinois in 1939 before embarking on his professional career with a year as instructor at Ohio State University. From there, he moved on to work as a statistician with the Louisiana State Public Health Service from 1942 to 1943, and served with the United States Public Health Service for a year after that. In 1943, he joined the Adjutant General's Office (Brogden, n.d.) as it worked to support the World War II effort.
Between 1943 and 1964, Brogden worked in the Personnel Research Branch of the Adjutant General's Office, a predecessor to what is now the Army Research Institute for the Behavioral and Social Sciences. He served first as a research psychologist, then as a research adviser, and, finally, from 1954 to 1964, as director of research (Brogden, n.d.). It was during this time with the Army that he was most productive, generating numerous significant articles explicating his insights on statistics; personnel selection and classification; and personality testing. His prominence in statistical theory and research design was such that he was chosen to author the review of research on these topics for the Annual Review of Psychology in 1954. In 1964, Brogden joined the faculty of Purdue University as professor of psychology. He continued at this post until his retirement in 1979, when he was given the title of professor emeritus. An indication of Purdue's prominence in the field of industrial-organizational psychology is a partial list of those who, besides Brogden, served on its faculty in 1967: Joseph Tiffin, Robert Perloff, William Owens and Ernest McCormick (Schmitt, n.d.). Yet even among these luminaries, Brogden stood out as an expert in tests and measurements. During his time at Purdue, he never lost his quest for learning. Although other professors could be observed reading novels while proctoring tests, Brogden would read measurement books (S. Sellman, personal communication, Sept. 9, 2015).
Brogden served as an inspiration and guide to numerous students who ultimately became prominent in the field of psychology. Although not by nature a gregarious person, Brogden interacted well with his students. Despite his intimidating intellect, he was personally patient, supportive and nonthreatening, and in one-on-one sessions used examples from his personal experience to elucidate key concepts (S. Sellman, personal communication, Sept. 9, 2015).
In some respects, Hubert's career paralleled that of his older brother, Wilfred. Wilfred helped found the Psychonomic Society (Grant, 1975), and Hubert later served as president of this organization (Brogden, n.d.).
Brogden's Approach: Solve the Right Problem
Brogden's extraordinary career accomplishments were founded on psychometric brilliance and an uncanny ability to formulate theoretical questions in terms relevant to operational problems. He described his approach, with examples, in an article appropriately titled “New Problems for Old Solutions” (Brogden, 1957). In this article, Brogden noted that many apparently intractable problems could be solved if appropriately stated. He observed that commonly used statistical approaches were not necessarily designed for practical issues confronting researchers, and that if these issues were properly articulated, the appropriate solution could be found. Throughout his career he applied this approach to generate solutions that had eluded earlier researchers.
Correlation Coefficient as a Measure of Utility
Brogden's first major breakthrough in the history of selection utility was his demonstration that the value of a selection tool, or predictor, could be related directly to the correlation between that tool and scores on an outcome measure, or criterion (Brogden, 1946b). This insight superseded earlier approaches that related utility to the amount of variance accounted for by the predictor — that is, the squared correlation coefficient (Hunter & Schmidt, 1982). This development by Brogden not only led to a more accurate method of calculating test utility but also put the value of testing in a more realistic perspective. If the true utility value of a test with a correlation coefficient of 0.40 were represented by the square of that value, or 0.16, it would be easy to dismiss that value as trivial, as many critics of testing were disposed to do. Brogden acknowledged that there were circumstances in which neither the product-moment nor the biserial correlation would accurately reflect the value of a test, or what he termed selective efficiency. He saw the need for a coefficient with more general applicability, not subject to the same restrictions as conventional correlation metrics, and developed just such a coefficient later in the same year (Brogden, 1949).
The Dollar Criterion
Brogden and Taylor (1950a) took the estimation of utility a step further with the publication of the article “The Dollar Criterion: Applying the Cost Accounting Concept to Criterion Construction.” This article addressed the problem of how to combine multiple criteria into a unitary value. If all criteria could be assessed on a single scale, then they could be meaningfully weighted and combined.
The article also provided a more tangible measure of the value of testing than was yet available. The scale chosen, the dollar scale, was one that users of tests could easily embrace, at least conceptually. Brogden and Taylor proposed an accounting approach to deriving the dollar value of criterion scores. This was not an approach users found very friendly, so it was not until Schmidt, Hunter, McKenzie, and Muldrow (1979) provided a less stringent means applied.
Brogden and Taylor (1950b) continued their examination of criteria in a second major article. They appreciated that a full understanding of criteria required an examination of factors that could lead to faulty or biased criterion measures. They accordingly provided the most thorough and disciplined examination of such factors ever attempted to that point. These included 10 factors divided into four categories: (a) criterion deficiency, (b) criterion contamination, (c) criterion scale unit bias and (d) criterion distortion.
They then offered strategies for dealing with these factors. This article has stood as the foundational treatment of criterion bias in all the years since.
Personnel Classification: Fitting People to Jobs
In its simplest form, personnel selection involves determining which individuals are best qualified to perform a particular job. When multiple applicants are competing for multiple jobs, the problem of determining the best fit of individuals across all jobs is considered to be one of personnel classification. A number of considerations make the problem of optimal classification a complex one. First, the number of possible matches of individuals to jobs multiplies quickly as the numbers of individuals and jobs increases. Second, each individual matching of an individual to a job affects the best cumulative matching of all remaining individuals to all remaining jobs. That is, each optimal person–job match is not independent of all other person–job matches. Third, the determination of how well the person performs once classified will vary according to the job in which he is placed. The criterion of success for a clerk will be different from that for a bricklayer. In determining where to place an individual, one must have some way of comparing these divergent measures of success.
Brogden did not tackle the entire classification problem in his first foray into this arena, but worked toward a solution gradually, beginning with workable segments of the challenge.
First, Brogden tackled the problems of criterion comparability and the interdependence of multiple person–job matches in the article “An Approach to the Problem of Differential Prediction,” in which he considered the problem of allocating personnel across two jobs (Brogden, 1946a). Here he foreshadowed his later work with Taylor (Brogden & Taylor, 1950a) by proposing the use of a dollar metric as a means of achieving criterion comparability. He then proposed a means of allocating personnel across two jobs, given certain assumptions. He acknowledged the difficulty, especially given the limited computational tools available at the time, of simultaneously solving all person–job matches optimally. Rather, he proposed an iterative procedure, in which the easiest, or most obvious, matches were made first, and the matches needed to fill the remaining positions followed. Next, in 1951, Brogden focused more directly on the predictors, examining the relative value of a single predictor versus multiple predictors in a multiple job context (Brogden, 1951). Here he demonstrated that multiple uncorrelated predictors could produce a better outcome for the organization even if the combination of multiple predictors added no validity beyond that obtainable with a single predictor. These predictors could be differentially weighted for different jobs, and the result would be a more favorable selection ratio for each job for the organization. The gain from adding an additional predictor diminished as the intercorrelation between the two predictors increased, but at a surprisingly slow rate until the intercorrelations rose above 0.80.
A Complete Solution
In 1959, Brogden presented his most polished and complete solution to the classification problem. This solution was not limited to two jobs, as was his first approximation in 1946. He determined that the efficiency of classification could be computed as a function of the following:
- Number of jobs.
- Percent rejected.
- Validity and intercorrelation of job performance estimates.
Higher efficiency could be achieved with a higher number of jobs, a higher rejection rate, higher validity of the predictors and lower intercorrelation among the predictors.
Notably, Brogden reaffirmed his conclusion from the 1951 article that the “possibility of an efficient differential classification battery is not lost until the intercorrelations of the estimates of job performance are quite high” (e.g., with intercorrelations of 0.8, efficiency of classification could still be 45 percent as high as with zero intercorrelations; Brogden, 1955, p. 189). This observation was critical to later defenses of the use of classification batteries (e.g., Zeidner & Johnson, 1994) against those who argued that the tests in those batteries were so intercorrelated that their classification value was negligible.1
Brogden's Lasting Influence
Brogden's contributions served as the foundation for several later major developments in selection and classification. His selection utility work was extended by Cronbach and Gleser (1965), such that the result became known as the Brogden-Cronbach-Gleser model (Russell, Colella, & Bobko, 1993). Its Achilles heel was its cumbersome approach to the problem of estimating the dollar value of performance. When Schmidt et al. (1979) generated a short-cut method for providing this estimate, they “sparked a frenzy of research activity compared to levels seen prior to 1979” (Vance & Colella, 1990).
Brogden's (1959) method of calculating classification efficiency has recently been revived and advanced by Zeidner, Johnson, and their associates, who used it as a foundation for their work on differential assignment theory (Johnson, Zeidner, & Leaman, 1992; Scholarios, Johnson & Zeidner, 1994; Zeidner, Johnson, & Scholarios, 1997; Zeidner & Johnson, 1994; Zeidner, Scholarios, & Johnson, 2003; Zeidner, Johnson, Vladimirsky, & Weldon, 2000). Capitalizing on modern computer capacity, these researchers used a multistage simulation-based approach, incorporating multiple samples in a triple cross-validation design, to evaluate classification efficiency under various conditions. Zeidner and Johnson (1994, p. 405) observed that their findings indicated, “classification effects can be much greater than selection effects.” They thus refuted those who devalued the use of classification batteries as measuring “nothing much more than g” (Zeidner & Johnson, 1994, p. 389). The differential assignment theory research paradigm contributed much to a recent reconfiguration of Army composites used for differential assignment (Greenston, 2002, 2012). Curiously, Brogden's legacy includes contributions to validity generalization as well as to classification, which focuses on differentiation rather than generalization. In part, his influence on validity generalization stems from his tenure as a Purdue professor. In this capacity, in a discussion with Frank Schmidt, then a student, “he stated that the military estimates were stable across samples. I asked him why this was not true for civilian estimates, and he said ‘sampling error.'” Nine years later, remembering this conversation, “it occurred to me that you could use the sampling-error formula to estimate how much of the observed variance was due to sampling-error variance” (“An Interview With Frank L. Schmidt,” 2011, p. 2). Thus was born the inspiration for both validity generalization and meta-analysis. Brogden's ideas contributed to both validity generalization and meta-analysis in other ways as well. The correlation coefficient, the metric often used for meta-analysis, and hence validity generalization, was defended based on Brogden's (1946b) demonstration of its link to selection utility (Schmidt et al., 1985). Brogden and Taylor's (1950b) examination of criterion bias highlighted the importance of criterion deficiency and contamination in explaining a portion of the variance not accounted for by the validity coefficient (Schmidt, Hunter, Pearlman, & Shane, 1979).
It is difficult to fully grasp the impact Brogden had on his field, his peers and his students. He was truly a giant, both in psychometrics and selection, as well as in the less-populated field of personnel classification. For those who knew him, he was as impressive as a person as he was a psychologist. Despite his many achievements, he remained a modest, considerate man. One of his former students (S. Sellman, personal communication, Sept. 9, 2015) recalled being at a party where someone remarked, “Where is Dr. Brogden? I do not see him here.” Brogden, sitting at a table nearby, said nothing, presumably not wishing to cause the man embarrassment. One of the man's companions then guided him a few short steps to Brogden's table and said, “I'd like to introduce you to the real Dr. Brogden.”
1 Zeidner and Johnson (1994) contested Jensen's (1986, p. 216) statement that “the rather uniform high g loadings of all the subtests [of the Armed Services Vocational Aptitude Battery, a battery of cognitive tests used for selection and classification of military service members] leave too little non- g variance to obtain sufficiently reliable or predictively valid differential patterns of the subtest scores for individuals.” g is a general factor derived from cognitive tests that is typically associated with general intelligence. Later, Murphy (2009, p. 458) stated that “the consistent pattern of positive correlations among ability tests and criteria means that the choice of which tests to use to predict performance in which jobs will not usually have a substantial impact on the validity of a test battery.”