On measures involving interpersonal and intrapersonal ability. I expect statistically significant but substantively modest gains. On measures of actual knowledge, the experimental group will score dramatically higher than the members of the comparison group, perhaps 30-plus percentile points higher (technically more than a standard deviation). On measures of reading and math achievement, the differences will be no more than 15 to 20 percentile points (about half a standard deviation). Three years after the experiment ends, all of the differences will have shrunk. The differences in reading and math will be no more than 8 to 12 percentile points (no more than a third of a standard deviation) and may have disappeared altogether.
More formally, I predict that the magnitude of each academic effect will be a function of the g loading of the measure. Measures of retention of simple factual material have the lowest g loadings and will show the largest gains. For highly g-loaded measures such as reading comprehension and math, what has been accomplished by the last half-century of preschool and elementary school will be shown to be about as good as we can do, no matter how much money is spent.
This is a decent prediction. The I think that Murray overestimates the ease at which facts can be taught to and retained by low-IQ students and underestimates their ability with respect to math and reading comprehension.
Facts are difficult to learn because facts must be mostly learned on a case by case basis which is not readily amenable to acceleration. Math and reading (decoding and comprehension) are easier to teach because these skills, can be accelerated (even though teaching language and vocabulary remain problematic). But I knew that from the Follow through and the Baltimore Curriculum Project data. The data shows that we can get at least about three-quarters to a standard deviation improvement on average by the end of elementary school, better if we discount the schools that are so incompetent that they are unable to implement well-tested programs with fidelity.
Murray's point with respect to fade-out is well taken, but I'll leave that for another post.