Sunday, April 23, 2006

Certified Pi^H^HPMP

My apologies for the recent lack of updates. I've had limited connectivity recently as I out of town doing some intense studying for my PMI PMP (Project Management Professional) exam. Preparing for the exam involved what was probably the most intense studying I've done for a long time, but it paid off and I passed the test on Friday!

Sunday, April 02, 2006

AWA: Attacking the computer

I was recently asked by one of the members of the TestMagic GMAT Forum about how I attacked the AWA (i.e., essay) section of the GMAT. I think that they key to getting a good score is to understand how the essay is graded. First, a computer program assigns a grade to your essay. Then, a human reads the essay and assigns a grade. If they agree on the grade, then that's the one you receive. If they don't, then another human reads the essay and the scores are averaged out.

Clearly, since Pearson doesn't want to pay more human graders than they have to, it's in their best interest to ensure that human-assigned and computer-assigned grades are highly correlated. So rather than guess at how the humans grade essays, let's analyze how the computer probably grades them, and assume that a similar rubric is used by the human graders.

As someone with a very significant background in computer science, I can assure you that the computer simply cannot make any decisions about whether or not the points you present in the AWA make sense. If you write the sentence "The argument presented does not make sense because apples are blue," the computer will not know that it is nonsense. This doesn't mean you can get away with writing nonsense -- the human grader will obviously catch it. However, it does mean that the computer is assigning grades based on something else.

So what can a computer assign a grade based on? The computer can assign grades based on words and sentences, but not ideas or meaning. In fact, I'm willing to bet that the computer uses an algorithm very similar to the Automated Readibility Index or the Flesch-Kincaid Readability Test, probably combined with some kind of grammar checker, to assign grades. Both of these tests calculate their score based on the average length of your sentences and the average length of the words you use.

Based on the assumption that the AWA software uses an algorithm similar to the ones mentioned, we can derive a strategy : All other things being equal, using longer words and longer sentences while maintaining grammatical correctness will boost your AWA score. At the same time, it is also critical that your AWA essays actually make sense--if your essay is gibberish or horribly organized, the human graders will knock points off of your score and it won't matter how long your sentences and words are.

To ace the AWA, you need to present a series of points that make sense to the human readers--but you also need to boost your computer-assigned score. When writing your Analysis of an Argument essay, why say that an argument is 'wrong' when you can say that it's 'fallacious'? Why say that an assumption made in an argument is 'very bad' when you can say that it's 'particularly egregious'? Breaking up two very related ideas into separate sentences will make your essay more readable by the average person; separating the two related ideas with a semicolon will boost your average sentence length, which is a key metric used by computer scoring algorithms.

Let me emphasize one more time that using long sentences, big words, and proper grammar is necessary, but not sufficient, for a high AWA score since the human readers also need to understand your essay. However, by understanding the ways in which computers are able to grade your essay (and the ways in which they aren't able to), you will have a decisive advantage in the AWA section.

In a future post: How to use an on-screen outline to organize your essay during the test