The way the tests are currently used is a corruption. Here is a good analogy. Lets take reaction time using a clicker. (or any other way to measure it). You would find that reaction time is highly correlated with things like vocabulary and…well intelligence. All that means is that your reaction time speed and, say your vocabulary are likely to place you roughly on the same part of the distribution compared to other people. So if you are in the 70th percentile for RT then you are probably close to the 70th percentile for vocabulary (this is just an example to make a point-the correlations are really not that high-but they are surprisingly high). So, if you took 100 people, you’d find that the two scores are highly correlated and if you know one you can predict the other.
But now let’s pretend you took 3 people and trained them specifically in reaction time. The 3 people worked all day long on a task similar to the one that you used to measure reaction time. Given this intense training, they get better and better at the task. Now their reaction time is no longer a valid way to predict their vocabulary. It was valid only when reaction time reflected how well someone does on the task that they have had the same exposure to as everyone else and where their performance is influenced by their previous life experience-not specifically with the task… But working specifically on reaction time will not increase vocabulary or comprehension or short term memory as shown by digit span tasks. No, Reaction time now says nothing about intellect for these three people. Their RT and vocab won’t be correlated any more. That is what happens with the SATs and ACTs. When performance reflected incidental learning, it was a decent measure-predictive of overall g. but not when people are being trained to master the test itself. Then the test is no longer valid-it is corrupted.