Thinking about methods:
When I do new experiments, I want to include 5-10 participants in each condition who do think-alouds, then do protocol analysis on the think-alouds. It probably wouldn't be to hard to ask MTurk participants to record their voice as they go through the task. Otherwise, just bring people into the lab.
Verbal Reports as Data
I also want to be sure to do normative analysis prior to descriptive work, along the lines of Anderson's rational analysis.
The Adaptive Character of Thought