Your general concept is very interesting. The question is reliability - how can you be sure that you're measuring what you're trying to measure. For example, the QB's willingness to pull the trigger is a variable that plays into your results, and this could skew your results.
For example, playing behind similar OL's, Trent Edwards and Jay Cutler may show very different ATT's - but this could be due to Cutler's willingness to throw into coverage in cases where Edwards was not. So, the ATT would measure the overall offensive unit (OL + QB play) rather than a specific position. Also, the type of patterns used by a team would affect the stat.