Then I counted the errors in each paper. But I figured it was the concentration of errors that was important, not the absolute number. That is, if a student wrote a single paragraph which contained three errors, that translates into a higher concentration of errors than a student who wrote a ten-page paper with the same number of errors. So I calculated each student's average number of errors per 100 words.
The really tricky part is that I decided to divide the errors into "cognitive errors" and "performance errors." Cognitive errors follow a certain mental logic and are fairly consistent. A student may consistently write "a lot" as "alot." Performance errors are slip-ups. A student may type "they" instead of "the." I hypothesize that procrastination should show stronger correlation to performance errors than to cognitive errors. Performance errors may pop up when a student is rushed and doesn't proofread. But cognitive errors are difficult for students to catch in their writing, since they don't usually realize that they are making a mistake at all. So proofreading is unlikely to help, unless the student gets outside help from a friend or tutor.
So how exactly can I look at a grammatical mistake and know what caused it? How do I know if the student merely made a mistake or whether they misunderstood the rules of writing?
Two ways. One, I looked for patterns. If a student always puts a comma before prepositions, that's a cognitive error. They're clearly using a rule, even if it's not the right rule. On the other hand, if a student only once fails to capitalize the word beginning a sentence, there's no pattern. It's not likely to be a cognitive error.
The second way I determined if errors were cognitive was based on my experience as a writing tutor and teacher. I knew that certain errors were more likely than others to be cognitive errors. For instance, students who violated punctuation rules when joining independent clauses were often obeying the "comma when you take a breath" rule. Knowing that certain types of error were quite likely to be cognitive allowed me to classify an instance as cognitive if a pattern wasn't totally clear. But if I couldn't tell, I classified errors as performance errors.
I also calculated cognitive and performance errors per 100 words.