Wednesday, June 25, 2014

Growth Model Browsing

Gary Rubenstein:

The most controversial thing about Johnston’s education politics is his firm belief in the accuracy of the Colorado Growth Model. This model is used to compare different schools based on ‘growth’ rather than just ‘achievement.’ Colorado has quite a good website for exploring data like this. So I thought I’d see how the Odyssey School did on their ‘growth.’

As it turns out, Rhode Island uses the same "growth browser." It is good for looking up a school's scores, but not much else, despite its apparent sophistication.

First off, just go have a poke around. If you're interested in data, it'll hold your attention for a while.

My main takeaway is that year over year growth data at the school level and below is quite volatile. The system makes it easy to, say, select a cluster of 10 schools within a few points for growth in one subject/year, and then see how they redistribute themselves in other subjects/years. More often than not, in any other subject/year, the schools are scattered across at least a 20% range, which seems like a lot since almost all the schools are within 30% of each other growth-wise. Is that good or bad? Seems bad to my eyeballs.

On the other hand, the data doesn't seem completely random. You just need a big enough sample size. I get the feeling that at the school level, a three year average would be fairly stable. On the other hand, that stability would make it relatively useless for the kind of accountability schemes reformers want. Changes happen too slow for them then.

It also would help if the system included error bars (circles) for selected schools. Or perhaps a way to either change the y-axis or add some color coding to discern how schools with different rates of free and reduced lunch or other disadvantages are doing system-wide.

This bit of self-praise by the designers focuses my criticism:

The SchoolVIEW data visualization application is head and shoulders above what any other governmental education organization has created. Administrators can look at “big picture” summary data for their district or school, while principals and teachers can focus on individual students and show parents information about their child’s growth. In this way, SchoolVIEW enables everyone to make better decisions on where to invest in education.

I would argue that SchoolVIEW's data visualization tool isn't even the best way of looking at Colorado's growth data. Their tabular Growth Summary Reports provide in one page information that otherwise would take dozens of clicks on the visualization tool, with limited direct comparison year over year or between different categories of data.

What the view is good at is showing how schools compare in a specific year and subject, but it is clear that is of limited value as soon as you take advantage of the ability to shift that comparison into different years and subjects and you see how fleeting and volatile those comparisons are.

No comments: