Analyzing Your Practice Tests, Part 1
The purpose of taking a practice test is two-fold. First, you are testing yourself to see whether you have learned what you have been trying to learn. Second, you are diagnosing your strengths and weaknesses so that you can build a study plan going forward.
This article is an update of one originally posted about 18 months ago. Why am I posting a new version? Two reasons. One, some functionality has been added to our score reports since then, so I think that we have some better ways of interpreting the data now.Two, I think that the old method could be stripped down a bit – it was a little too complicated. Though, honestly… it’s still pretty complicated… there’s a lot to learn from practice tests!
As I did last time, I’ll base my discussion on the metrics that are given in ManhattanGMAT tests, but you can extrapolate to other tests that give you similar performance data (note: you need per-question timing and difficulty level in addition to percentage correct/incorrect data). It takes about 45 to 60 minutes to do this analysis, not counting any time spent analyzing individual problems.
So here’s what I do when I review a student’s test (or tests)!
First, naturally, I look at the score. I also check whether the student did the essays (if she didn’t, I assume the score is a little inflated) and I ask the student whether she used the pause button, took extra time, or did anything else that wouldn’t be allowed under official testing guidelines. All of this gives me an idea of whether the student’s score might be a bit inflated.
Next, I look at the problem lists for the quant and verbal sections; the problem lists show each question, in order as it was given to the student, as well as various data about those questions.
First, I scan down the “correct / incorrect” column to see whether the student had any strings of 4 or more answers wrong. If so, I also look at the time spent; perhaps the student was running out of time and had to rush. I also look at the difficulty levels because sometimes I’ll see this: the difficulty level is high for the first problem or two, and the timing is also way too long. On the later questions, the difficulty level is lower, but the timing is also too fast. Essentially, the person had a sense that she spent too long on a couple of hard questions, so she sped up… and then she not only got the hard questions wrong but she also got the easier questions wrong because she was rushing.
Then I scan down the “Cumulative Time” (how much time you’ve spent cumulatively on the test) and “Target Cumulative Time” (how much time you should have spent cumulatively) columns. I specifically look for periods when the student is more than 2 minutes off of the Target time. When I see that the student is too fast or too slow, I try to figure out what happened – where was the student spending too much time or rushing? What happened on those problems?
Then I scan down the “Time” column, which lists the time spent per question. Even if the student managed to stay on time cumulatively, the student still might exhibit what I call “up and down” timing – spending too long on some and then rushing on others to catch up. Even if you finish the section on time, you still might have a timing imbalance.
Here are the timing metrics by problem type:
*The first RC question includes the time it takes to to read the passage.
Again, I’m looking for patterns. How many times did the student hit the “warning track” or spend way too much time, and how many times did the student move too quickly? What was the cumulative outcome of these statistics?
If there are more than a few (regardless of whether they’re right or wrong), then the student has a timing problem. For example, if the student had 4 questions over 3m each, then I can guarantee you that the student missed other questions elsewhere simply due to speed – that extra time had to come from somewhere. You know those times when you realize you made an error on something that you knew how to do? Well, if you were also moving quickly on that problem, your timing was at least partially a cause of that error.
Alternatively, if there is even one that is very far over the “way too slow” mark, there’s a timing problem. If you have one quant question on which you spent 4m30s, you might let yourself do this on more questions on the real test – and there goes your score. (By the way, the only potentially acceptable reason is: I was at the end of the section and knew I had extra time, so I used it. And my next question would be: why did you have so much extra time?)
For each section, I get a general sense of whether there is
- not a timing problem (e.g., only 1 or 2 questions in the too fast or warning track range)
- a small timing problem (e.g., 3 questions in the warning track range, or 1 problem in the way too slow category, plus a few “way too fast” incorrect questions), or
- a large timing problem (e.g., 4+ questions in the warning track range, or 2+ questions that are way too slow, plus multiple “way too fast” incorrect questions).
Note that I don’t specify above whether the warning track and too slow questions were answered correctly or incorrectly – that doesn’t matter. It isn’t (necessarily) okay to spend too much time on a question just because that question was answered correctly.
If a timing problem seems to exist, I try to figure out roughly how bad the problem is. How many problems fit into the different categories? Approximately how much time total was spent on the “way too slow” problems? How many “too fast” questions did that cost the student? You may also want to examine the problems themselves to locate careless errors. How many of your careless errors occurred on problems when you were rushing?
Also, be flexible with the assessment. For instance, if you answered a quant question incorrectly in 45 seconds, but you knew that you had no idea how to do the question, so you chose to guess and move on – that was a good decision. You don’t need to count that “against” you in your analysis.
Finally, I see whether there are any patterns in terms of the content area (for example, perhaps 80% of the “too slow” quant problems were problem solving problems or two of the “too slow” SC problems were modifier problems). We’re going to run the assessment reports next to dive deep into this content data, but do try to get a high level sense of any obvious patterns.
All of the above allows you to quantify just how bad any timing problems are. Seeing the data can help you start to get over that mental hurdle (“I can get this right if I just spend some more time!”) and start balancing your time better. Plus, the stats on question type and content area will help youto be more aware of where you tend to get sucked in – half the battle is being aware of when and where you tend to spend too much time.
If you do have timing problems, this article on time management should be useful.
Now we’re done looking at the problem lists; in the next article, we’ll analyze the data given in the assessment reports.