Tools for Automated Grading?

Dont tempt me asks: "As all teachers and students are well aware, it is back to school time. As a math/computer teacher, I am constantly looking for ways to automate repetitive tasks. The one that seems to take up most of my time is grading. As is typical for us nerds, I find myself looking at handwritten tests and thinking 'there's gotta be a better way...' Since I can't find any related open-source projects, I have been thinking about creating one. I have been toying with the idea of using OMR (Optical Mark Recognition) to make my own scannable multiple choice tests. Is anyone doing this? If not, where would be a good place to start? In addition to teachers, this could be a useful technology for questionnaires, or other processes that require manual data entry."

Comments

Scantron

[justanyone]

I taught astronomy at KU as a discussion section leader in 1991. We used scantron machines. These were #2 pencil IBM-card (~3 inches wide by ~8 inches tall) sized.

The machines could NOT have been expensive. Using them was dead simple. We (the section leaders) wrote several tests, and rearranged each test to have different orderings for the choices. Thus, on test version A-1, I had answers (a) Sun, (b) Moon, (c) Earth, then on A-2 I had (a) Moon, (b) Sun, (c) Earth, etc. Then, we looked at their version of the test, and put in the right key.

This kept cheating to a minimum; at the least they had to memorize the answers instead of the answer key. And, memorizing the answers was kind of okay in a sense since they at least paid attention to the subject material.

Re:Scantron

[clifyt]

"If a person has correct answer but got it only by mistake then what good is it?"

Because by knowing how a test works, it tells a LOT about the student. There is a lot of science behind item response theory...for instance, a right answer on a specific item may tell you that the student has guessed. That's right...the reliability of the item may be that over 70% of the people that passed the test got it wrong, while 70% of the people the failed got it right. Quite a bit can be figured out about the student by what they got right or wrong.

In fact, the classic 3 parameter model guessing is a BIG part about the calibration:

http://www.rasch.org/rmt/rmt181b.htm

(difficulty and discrimination being the other two parameters)

Of course, this come more into play when you get into computer adaptive testing, but it still important on the standard multiple choice test (at least if its standardized).

As for math and computer science...I've never designed a CS test using multiple choice, but math is very useful. Again, adaptive testing with multiple choice is better, but its not like math is rocket science...as an instructor mentioned recently, math is not about memorizing formulas or otherwise, its about learning how to think creatively while breaking down a problem and coming up with a solution. If you can figure out the answers that don't belong and figure out the rest without having to do the problem, you've demonstrated knowledge about the ideas behind the problem in front of you...its just like the complaints we got about an essay rater we were working on several years back (it graded structure, not content)...people were submitting brilliantly written nonsense and then complaining that we were scoring it high -- the fact that someone was able to figure out ways around the writing rubric to game the system meant that they pretty much deserved the high scores they received because they were good writers -- regardless of the nonsense (unlike my own writing...this is not a good example of such writing).

All in all, this all depends on what you are after -- demonstration of mastery or for a diagnostic (or maybe to prove your students know as much as the students down the hall or the school across the state). Luckily, most educators are taught testing methods before they leave school and probably have a little more clue as to what they are looking for than students that think a test may just be multiple choice bullshit (though that could be considered negative face value :-)

clif (speaking for myself and not my office)

Not just grading..

[thefirelane]

Why not do automatic test creation as well. If you're going to do multiple choice, you can have a 'pool' of questions. Each question has a score, based on the percentage of students who get it wrong usually.

You could then automatically create tests with a certain percentage of 'A level' questions and so on. This would also let you more-or-less predict the curve... 10% will get an A, and so on.

Since the grading is automated as well, it would feed back into each question's score automatically

This may sound disturbing, but this is what the SATs do... those small sections at the end are just next years questions being tested

I also had a professor in college who did this, but it was through mental calibration over years. Yes, this does mean you can not give out the tests after for the students to review... but the test was surprisingly fair.

Re:Not just grading..

[big ben bullet]

This is actually an excellent idea.

But what really matters is the content. My wife teaches the courses in primary education (6 to 12y olds). She has a hard time making up questions for each course in each grade. If there was an openly licensed 'questions base' (maybe even a wiki) this would help her alot. She would be happy to contribute her existing material ;-)

You might want to start such an initiative along with the actual open source application.

Ohw... and *please* mind the localisation so there's a potential for every language speaking community in the world, and not only the english speaking community.

Re:It was a dark and stormy night...

[ragnarok]

I had a teacher in high school who graded OMR like sheets manually with an overlay that only had the holes cut for the correct answers.

Several times if I didn't know the answer I would mark more than one (you don't want to mark all of them, it stands out too much) and I always got credit.

The teacher seemed like a smart guy too, I wonder if he was doing it intentionally to see if people would figure it out.

If you can automate, should you be grading?

[gozar]

My best math teacher assigned homework every night. She would flip a coin the next day on whether it would be for a grade or not. So 50% of the time she wouldn't have to grade anything.

Assessment should be about the students knowing the material. Stuff like showing your work goes a long way. Math is the easiest to automate, but that would only show you that the student got the correct answer, not where the answer came from (like from a friend!).

To lower you work load, flip a coin on whether the students will hand in the work. If they aren't handing it in, trade with another student and grade it in class. Scantron only sends the message to your students that you are too lazy to look at their work, so why should they put any effort into it.

Scantron Exploit

[Noksagt]

So many of the scantron-style tests had heavy-black marks on either or both sides of the exam. The machine use these marks to locate each question. (A big square would correspond to start/stop of the test and a line would correspond to an individual question or similar). Ths beats having to guesstimate where the questions were by feed-through rate & also led to high-speed machine-grading.

Well, a clever student decided to draw his own marks on the side of the exam. This managed to trick the idiotic machine into thinking all questions were one off & subsequently his test & all tests which were fed through after his had a low grade; the machine tested if question 1 had the answer expected of question 2 and so on.

Re:Scantron Exploit

[bluGill]

Back when I was in school, 15 years ago, my teachers were onto such tricks. It isn't hard to look at scores are you write them into your book. The teacher already knows 'Suzie' is smart, often getting a perfect score, so if he[1] misses most of the questions it is time to re-examine things by hand.

The most popular way to cheat was to mark the little box at the top that set this sheet to the master, which would re-program the machine to take your test as the correct answers.

None of these tricks were hard for a teacher to catch (if you knew about them it was easier, and the principal made sure they knew). Once you catch a student doing this you just write zero in for his score and re-run the tests.

[1]we miss you Johnny Cash

Multiple Guess

[99BottlesOfBeerInMyF]

I've always had a strong dislike for multiple choice and true/false testing. Taking those tests is often more of an exercise in test taking than it is in the subject matter. A good test taker can eliminate a good portion of the answers right away and use fairly intuitive psychology to improve the odds of guessing correctly.

What ever happened to demonstrating competence in a field? Forget multiple choice and true/false. Ask your students to actually solve applied math problems or actually write some code (or pseudo code). Maybe you can't do as much testing that way and maybe you can't shorten the time it takes to grade the papers but at least you will be testing something worthwhile.

Sorry for the rant but after having survived more than a decade of "education" that consisted primarily of memorize foo and the regurgitate, I'm fairly traumatized by the horror that is the educational system. I learned orders of magnitude more useful information by simply reading everything and anything and trying to apply what I learned to my pet projects. I took one too many tests where I knew several multiple choice answers where justifiable and "right" depending upon unspecified information not contained in the question and having to guess what the test author thought the correct answer was. Multiple choice, true/false, and automated testing are big indicators of a "fast food" mentality and I'm firmly against that sort of foolishness. Grumble, grumble, etc.