If there is a reasonable escape from the trap in which Donald Devine finds himself, I don't know what it is.

Devine is the head of the federal Office of Personnel Management (formerly the Civil Service Commission) and the trap is a consent decree, inherited from the Carter administration, that orders him to stop using the government's chief entry-level test until he can prove that it doesn't discriminate against blacks and Hispanics.

The suit that led to the consent decree established that the two minority groups don't do nearly as well as whites on the Professional Administrative Career Examination (PACE), a key point of entry for 118 job categories at the GS5 to GS7 level. About 40 percent of white applicants score at least 70 percent on the exam, which is supposed to be a major route to management jobs, while only 13 percent of the Hispanics and 5 percent of the blacks do that well.

When the government failed to establish that the test was job-related, it was left with no defense against the charge that it was racially biased.

Incredibly, the Carter Justice Department then entered into a consent decree that said henceforth the percentage of blacks and Hispanics who pass the test must be roughly equal to the percentage who take it. Thus, if a third of those taking PACE are black and Hispanic, roughly a third of those who pass it must also be black and Hispanic.

The fule is to apply for three years, during which time OPM is to devise bias-free tests for each of the 118 categories -- bias-free meaning that the tests must produce the right statistical results or else be shown to be directly job related.

Devine's problem is that it is essentially impossible to prove job-relatedness on a general test. He doubts that it is really possible to do so with more specific tests, though it will cost his agency something like a half-million dollars for each of the 118 categories to try.

"We already have studies that show that a high PACE score is correlated with on-the-job success," he said during a reluctantly granted interview last week. "We came up with a correlation [coefficient] of something like.3 or.4. [A perfect correlation would be 1.0 -- for instance, the correlation between sunrise and morning. Many statisticians say that a correlation of less than.5 is too low to eliminate the possibility of pure chance.]

"I know that some people will say our correlation is not very high, but the problem is that correlations in the social sciences are never very high. I used to be a political scientist, and the correlations between, say, party affiliation and actual voting patterns are about.4. You just don't get the.9 or.98. We think our PACE results prove validity -- they show that people who do well on the test do well on the job -- but the correlation is not to the point where anybody would say 'Wow!'"

But even if OPM went to the trouble and expense to devise specific tests for each of the 118 categories, Devine is not convinced that he could establish a convincingly high correlation. And even if he could, he's not sure he would want it. The whole point, he said, is to select people who are potential managers. And applicants who scored well only on a single entry test might have skills that are too limited for that purpose.

He said he has been seeking advice as to how he might abide by the consent decree without abandoning the principle of merit selection. Most of the suggestions, he said, have their own problems. Personal interviews, for instance, are a good device for sizing up applicants. "But last year, we selected 4,670 employees from 102,000 applicants. There's no way we could conduct 102,000 individual interviews. We have to have some preliminary screening device."

Another suggestion is the so-called "outstanding scholar" approach. "That means that you pick people who fall in the top 10 percent of their class standings at a recognized college or university," Devine explained. "In the present case, it means that you go to black colleges with substantial Hispanic representation and pick people for noncompetitive entry. But that's not really merit selection; the schools are simply not that consistent. Besides, our studies have found that the correlation between academic grade level and on-the-job success is only about.17. That means there's basically no relationship.

It is Devine's private opinion that it is futile even to try to devise a test that is at the same time statistically valid, job-related and without adverse impact on blacks and Hispanics. The problem, he says, is not that the tests discriminate against these minorities but that they reflect previous discrimination in terms of acculturation and education.

There is one further problem. If OPM follows the letter of the consent decree and hires blacks and Hispanics in the same proportions in which they take the entry test, the likely result would be to create a "ghetto" of unpromotable black and Hispanic GS5s And if that happened, who could doubt that the next push would be for promotions out of that "ghetto," based not on measurable ability but on pure statistics?