Weirdly, when they get to the programmer productivity bit they decide to compare _reported_ times for the scripters and _measured_ times for the non-scripters. Then they go on to try to justify doing this.
Agreed-- this puzzled me as well. I don't really doubt that the scripting languages are faster in terms of coding time but why not just measure them or at least have a 3rd party confirm the developer's time?
In any other study of workers, I'm pretty sure even the author would feel self-reported data would invalidate the results (especially when all of the "best" scores were self-reported).
e.g. The ten chosen Honda dealerships reported that they can fix a Honda in 10-23 minutes compared to our measurements that Ford dealerships can fix a Ford in about an hour and a half.
Why didn't they just measure them all?