"More accurate" and "true IQ" are themselves rather meaningless terms. There is no "perfect" number tatooed in the kid's genes, and the tendency to assume that the highest number is necessarily the "most accurate" is itself a misconception.
The GAI is one measure of general intelligence, which de-emphasizes the role of working memory and processing speed.
The FSIQ is another measure of general intelligence -- and it is not, by the way, a true statement that WM and PS have nothing to do with general intelligence. Both scores can be valid -- they are just measuring different things.
If we used a different test entirely, we'd get yet a different measure of general intelligence. You cannot separate the score from the test used to generate it.
As for the statements about statistical significance, note that in most cases, the number of points different for statistical significance at the p<.05 level is considerably higher than cited above, particularly when one makes the comparison to the segment of the norming sample with similar overall performance. Also, recognize that just because a difference is statistically significant, that does not necessarily mean that it is particularly uncommon in the population, nor does it necessarily mean that it is clinically significant for that particular individual. That kind of analysis should be done on a thoughtful and individual basis.
What research indicates about identification procedures is that you should match the identification procedures to the intervention programs. That is, if the program is going to require a kid to do stuff that relies on WM and PS, don't use GAI for identification, because you're going to end up getting kids in the program who then can't handle the work, and you're going to exclude kids who should be in the program. This is most notable with the trend to use nonverbal fluid reasoning tests to identify kids for programs with high verbal written output requirements... a recipe for frustration on all sides.
