Did Gov. Michael Dukakis's widely discussed job-training program for Massachusetts welfare mothers prove to be a substantial success, with a broad range of beneficial effects for participants, as a study by the Urban Institute found?

Or was Dukakis's employment and training program (ET), which started in 1983, a costly and over-ballyhooed bust, as characterized by economist June O'Neill in a study she made?

The dispute is not just a tempest in an academic teapot. The apparent success of the Massachusetts program -- in lifting welfare mothers off the rolls and into decent jobs through remedial education, job training and placement reinforced with child care and medical assistance -- made it a prime model for the landmark 1988 federal welfare bill revamping the $21 billion federal-state Aid to Families With Dependent Children program.

As a result, billions of dollars in added federal payments to the states for training and other program componments will be made under the 1988 federal legislation with expectations that the outlays will help bootstrap the "underclass" to economic success and eventually cut welfare costs substantially.

Claims by Dukakis supporters during his 1988 presidential campaign that he had achieved an "economic miracle" in his state came under heavy fire on a partisan basis. But O'Neill, an economist who has worked for the Brookings Institution and the Congressional Budget Office, weighed in with the heaviest blow last May in a study for the Pioneer Institute, a conservative think tank based in Boston..

She concluded that the program had not cut the welfare rolls. She found it had little broad beneficial impact and had obtained its ostensible good outcomes in individual cases through "creaming," that is, attracting ambitious and motivated welfare clients who would have achieved the same kind of success even if they had not been in the program.

Now, a large study headed by Demetra Smith Nightingale of the Urban Institute, a middle-of-the-road nonprofit think tank here, has come to the opposite conclusion.

The study, released last week, found that, on the average, welfare mothers who participated in education and training activities in 1986 and 1987 stayed on welfare for a shorter time, and, after completing the program, earned more and had higher employment rates than welfare mothers who did not participate. It found little evidence of creaming.

Nightingale found that in the first six months of 1988, 44 percent of those who had participated in the program had jobs, compared to 36 percent of the nonparticipants; and that, on average, participant earnings were 34 percent (or $360) higher than those of nonparticipants.

She also found that during the last six months of 1987 the average monthly welfare grant for participants was 8 percent (or $26) less than for a comparable nonparticipant and that participants spent four months less time on welfare on average than nonparticipants.

The sharp differences between the O'Neill and Nightingale studies may result from a focus on different things.

O'Neill, who once worked for the Urban Institute, concentrated on whether the Massachusetts employment and job-training program reduced overall welfare case loads and produced significant welfare savings for the state. She concluded that it didn't.

Nightingale focused on the impact on individual participants, concluding they ended up better off than welfare mothers who did not participate. She did not measure the impact on case loads.

If Nightingale and her associates are right, the new federal program's extra outlays of $1 billion or more annually for training programs probably will turn out to be money well spent.

If O'Neill turns out to be right, the public probably will view the extra outlays as another failure in welfare reform.