Thursday, March 25, 2010

Class-size effects in PISA : further results

Further to my recent post, I show a somewhat detailed analysis above. The dependent variable is the students reading score which is an arbitrary function of class-size (i.e. in their English class, since this is data for Ireland) and of the teacher/pupil ratio in the school. In addition, I control linearly for the students sex, their age in months (they are all about 15), their birth order, their fathers education, school type (private/public) and a few measures of the home environment (like number of books). These controls don't actually make much difference.
So again we see a positive gradient with students in bigger classes doing better though it seems flatter than previously shown. The teacher/pupil ratio doesn't matter: using the absolute number of pupils gives the same result i.e. there are no economies or diseconomies of scale. Repeating the analysis for the maths score gives much the same results. The positive gradient is a puzzle though this is not the first time it has been found.


Alan Fernihough said...

Personally, I am a huge fan of Ed Lazear's stuff when it comes to the class size debate. Are there any measures of
(or proxies for) class-room disruption in the data?

In addition, I think this may have been asked already, but can you split the data up such that you see the conditional distributions for pass vs honours classes?

Kevin Denny said...

I don't think there are measures of class disruption per se, Alan. There is a variable which measures the level of discipline, I forget the details. How would you utilise it:
Class size to instrument disruption in the production function? Might work.
There is no information on pass vs. honours. Many countries wouldn't have that & PISA is consistent across countries. There might be a school level question about whether streaming occurs - need to check.

Mark McG said...

It might be interesting to look at the standard deviation of test scores by class size, particularly if there's no question on streaming. Could give some indication as to whether there is selection within schools.

Kevin Denny said...

The variance is an interesting but tricky question. Streaming will give a low variance by definition. However, it might also be the case that variance is a function of class size: with a small class maybe the teacher is better able to bring up those at the bottom? A possible benefit of smaller classes might not be a higher mean but a lower variance.
This relates to a point Lazear made that class size effects have heterogeneous effects with disadvantaged students benefiting most.