Misinterpretation of Data – An Architectural Problem?
Do chip architectures worry about Type I and Type II errors? While performing trade-off analysis between power vs performance demands, architects often are working to a set of system-level (final product) operational scenarios. These scenarios reflect how the chips should work in the final product and are based on research data and legacy applications. But what if the data has been misinterpreted or the user cases misunderstood?
In the world of program management of high order systems, the misinterpretation of data falls into the categories of Type I and Type II errors:
Type I – Errors made by rejecting a hypothesis that is true, e.g., an alarm that fails to go off when it should (system fails to do what it is supposed to do). In a very general sense, an example of a Type I error for Project Management would be having a plan that didn’t work.
Type II – Errors made by accepting a false hypothesis, e.g. an alarm that goes off when it shouldn’t (false alarm or unintended errors). In a very general sense, an example of a Type I error for Project Management would be having a plan that worked but on the wrong problem.
Here’s another example: In terms of data entry, Type 1 would be a “failure to capture accurate data.” Type II would be an error “capturing incorrect data.” Or consider the realm of user interfaces (UI): Type I results from drop down boxes full of options, where the set of options doesn’t adequately address all of the potentially valid inputs the user may want to enter. Type II results from UIs with options/deflates that aren’t obvious or intuitive, which “trick” the user into a false entry.
Statistically, Type I is generally considered to be a “false positive”, and Type II a “false negative.” Whether an error in a system is a Type I or Type II error depends on how you choose your hypothesis.
At this point, you might say that these are really statistical question, hardly of concern to most chip architects. But Type I and Type II errors certainly play a part in the study of trade-off scenarios. And both errors are related in that they both represent a misinterpretation of the data which leads to an incorrect decision.
I’d be interested in reading your thoughts about these types of errors. (Thanks to David Carswell for input into this blog.)