Primary Education in America: pay and standardized tests

The NYT has an editorial today about how poorly U.S. teachers are paid.

I agree with the main idea of the article (I have a few issues with some of its points), although I think there would need to be a few changes for something like this to be very successful.

As it is, teachers' salaries are below-market for a given education level:

At the moment, the average teacher’s pay is on par with that of a toll taker or bartender. Teachers make 14 percent less than professionals in other occupations that require similar levels of education.

This is based on weekly pay (which seems to be a known metric in the field, so I believe their Summer break is factored in), and it does ignore some things like the benefits in having a job that makes you feel good about yourself. That said, this suggests that those going into teaching are also not the best in the market. (Granted, many good people go into teaching because they are passionate about it, and this is wonderful; but it's not enough). What happens when great people go into education? I have a friend from college who was extremely bright, in the honors program at a top research university; within a few years of teaching inner-city youth, he had decided to quit his job for more reasonable pay. He definitely wanted to help these kids, but he gave me an example of the state of affairs: one of his 8th graders asked him how to spell "hit".

This means that, while there are some great teachers out there, there are simply not enough great ones.

In any case, for higher pay to work well, we need to be open to a few things.

1. First, we would need to be open to laying off underperforming teachers. This is necessary if we are to pay teachers market- or above-market- rates for their education level, since it will increase the pool of talent. Naturally, we'll need to re-evaluate existing teachers in the new talent pool.
2. Focus on standardized testing to calibrate and pay teachers. Many decent people still question standardized testing based on unfounded fears of things that are easily addressed. Unfortunately, teachers and teachers' unions often add to these concerns, when they should be the ones most embracing them. Standard concerns are:
   
   1. How can we fairly compare teachers when the student quality or school quality differ so much? The primary variable we should be using to measure teachers is aggregate improvement in students' test scores compared to peer students. This means we don't reward or penalize a teacher for teaching kids who have problems to start with or for teaching kids who learn more slowly; we reward or penalize them based on how much their students have improved, compared to peer students. It is possible -- nay, easy! -- to measure these things fairly. Several approaches include (a) building variables into a statistical model that account for typical school performance and how quickly students learn (see item-response models, for a simple example of what's possible) at a school; or (b) moving teachers around a district as a form of experiment. The latter would help to calibrate different districts (which, in turn, is used to calibrate teachers). Teachers don't want to move around? They will if it's local and if you're paying them $100k.
      
   2. Standardized tests makes teachers teach to the test. This is not a problem, if the tests are well-designed. In fact, with a well-designed test, that's exactly what we should be getting teachers to do! In particular, a well-designed test should be broad enough to make teaching to it meaningful to the student, and it should be narrow enough to measure coherent skills. Both are easy.
      
   3. Teachers and administrators can cheat on these tests. Then (a) make it a federal offense to cheat on these tests, and (b) fire teachers and administrators who interfere with this testing process. There are many ways to identify cheating test administrators; statistical models are just a few of them.
   4. Tests don't measure the important things. The argument here is that teachers can teach many important things that aren't captured by tests. I'll use a counterexample-by-anecdote: my fifth-grade teacher had won a number of Teacher-of-the-Year awards throughout the years. When I had her, we rarely made it to studying math and never made it to studying science. I discovered why this was the case: she did not understand fundamental math. For example, she couldn't solve a basic algebra problem posed in our 5th-grade textbook (A robot takes a number, adds 3, multiplies by 4, adds 2, and divides by 2. The result is 7. What is the input?). On the other hand, she had a way of getting students to like her. We spent our time in her class playing games and having her chat with us about whatever she felt like chatting about. When she told us that we'd be learning the 50 states and the Gettysburg address, I obediently memorized them, to find later that she never expected us to learn these things. She was good at getting kids to like her and sounding ambitious, which was good for getting her teaching awards. The only science we ever learned in that class was from substitute teachers.
   Some people actually favor written essays instead of multiple choice questions; I'd argue that such an approach must be done very carefully or not at all, since many such essays are about politically charged issues: "Take a side in the dispute about affirmative action and argue your side." "Should state schools teach the humanities? Take a side and argue it." To argue that an essay grader (who works in academia) can take an objective approach to grading the composition of such essays (on material with which the student is not familiar) is silly -- they're really just a way to see whether the student fits into the academic mindset. (For full disclosure, I tended to do worse on the written essays than on multiple choice questions.)

On the topic of standardized tests and education, I'll argue that we should be much more open to experimenting on kids. If we were to implement nationwide experiments in education, we could have coherent ways of improving curricula (examples might be, "if we teach geometry before algebra, how does that affect mean increase in scores 3 years down the road?" "if we minimize memorization in math and focus on applying core ideas like the Pythagorean Theorem and basic algebra, how does that affect scores?"). I envision this as being similar to the way Amazon and Google experiment with their Web traffic with A/B experiments.

Returning to teacher pay and summarizing: I'm a huge fan of increasing teacher pay, but I think it should come hand-in-hand with clear, objective metrics of success and critical performance reviews.