Abstract
The subset selection problem is considered for normal populations. Minimax considerations suggest three procedures as the main contenders for this problem. Two of these are the “average-type” procedure and the classical “maximum-type” procedure. The third procedure has not before been considered as a serious contender. Numerical comparisons of the three procedures indicate that the “average-type” procedure is clearly inferior to the other two. The “maximum-type” procedure performs on the whole somewhat better than the new procedure, although for some situations they seem quite comparable.