Are Smart Communicators Better? Soldier Aptitude and Team Performance

Abstract

This article describes research conducted to provide empirically based, quantitative estimates of the relation between soldier aptitude and team performance. This study examines teams of Signal personnel, who operate communications systems allowing command and control among Army units on the battlefield. Soldiers are assigned at random into teams varying in aptitude, and performance is tested and measured in a controlled setting. The study finds that the team members' scores on the Armed Forces Qualification Test predict their ability to install and operate communications networks. Moreover, the aptitudes of all team members contribute to the probability of a successful outcome. The study concludes that lowering accession standards could degrade team performance among Signal operators who must interact with each other to provide usable battlefield communications systems.

Notes

1 AFQT scores are represented as percentile scores and are normed periodically on the U.S. youth population. Five AFQT categories are in turn defined by AFQT scores, as follows: Category I, percentiles 93 to 99; Category II, 65 to 92; Category III, 31 to 64; Category IV, 10 to 30; and Category V, 1 to 9. Category III is subdivided into Category IIIA, 50 to 92, and Category IIIB, 31 to 49.

2 Applicants with AFQT scores in Category V (percentiles 1–9) are excluded by law from military service.

3 Chi-squares and log likelihoods (in parentheses) of the maximum and minimum AFQT are 27.15 (418.37) and 27.36 (418.16), respectively. In addition, we compared the simultaneous effects of average AFQT and maximum or minimum AFQT in two additional models. These models indicate that average AFQT is a stronger predictor of group performance than either the maximum or minimum AFQT within the group, when one of the latter measures is included along with average AFQT. The effect of average AFQT remains statistically significant, whereas the effects of maximum or minimum AFQT are not significant.

4 Olher possibilities that we considered but rejected based on empirical evidence was the effect on the group outcome of single operators in specific positions. Here, we examined whether the group outcome was determined by the aptitude of the member in each terminal or by the relay, respectively. Compared to each of these three cases, the model using average AFQT of the three members provides a better fit, with higher chi-square and smaller log-likelihood values.

5 We evaluated these alternative models by comparing the ratio of the log likelihoods of the baseline model using average AFQT and the model containing the three operators' AFQTs (Maddala, 1988). This ratio has a chi-square distribution with degrees of freedom equal to the number of additional coefficients. Here, the small improvement in predictive power afforded by the individual AFQT model (−2 log L = 412.29 vs. 413.02 for the model using average AFQT) did not warrant the two additional degrees of freedom required by this model.

6 For example, there is some reason to believe that the effect of AFQT would not be linear but rather linear in logs. That is, it is reasonable to hypothesize that the impact of AFQT is greatest among the middle range of scores, and changes in AFQT at the extreme upper and lower end of the AFQT scale would make little difference. Entering group AFQT in log form (as the mean of the logs of AFQT of the three soldiers) produced results negligably different from using the actual AFQT score. There was also very little difference using a model that included multiplicative effects of AFQT (measured as the log of each individual AFQT). We therefore concluded that, at least in the range of AFQT scores of soldiers we tested, the impact of AFQT appears to be linear, and by entering AFQT measures linearly, we would avoid the additional complexities of interpreting the results that would be introduced by log formulations.

7 Figure 2 also shows the estimated likelihood of success for groups whose mean AFQTs are at the midpoint of Category I and Category IV. The differences appear very large, but caution should be exercised in interpreting results for such groups. These predictions are extrapolated from the model, with few groups in the sample actually found in these categories.