Testing Data

A Few Quick Fixes For School Accountability Systems

Our guest authors today are Morgan Polikoff and Andrew McEachin. Morgan is Assistant Professor in the Rossier School of Education at the University of Southern California. Andrew is an Institute of Education Science postdoctoral fellow at the University of Virginia.

In a previous post, we described some of the problems with the Senate's Harkin-Enzi plan for reauthorizing the No Child Left Behind Act, based on our own analyses, which yielded three main findings. First, selecting the bottom 5% of schools for intervention based on changes in California’s composite achievement index resulted in remarkably unstable rankings. Second, identifying the bottom 5% based on schools' lowest performing subgroup overwhelmingly targeted those serving larger numbers of special education students. Third and finally, we found evidence that middle and high schools were more likely to be identified than elementary schools, and smaller schools more likely than larger schools.

None of these findings was especially surprising (see here and here, for instance), and could easily have been anticipated. Thus, we argued that policymakers need to pay more attention to the vast (and rapidly expanding) literature on accountability system design.

The Cartography Of High Expectations

In October of last year, the education advocacy group ConnCAN published a report called “The Roadmap to Closing the Gap” in Connecticut. This report says that the state must close its large achievement gaps by 2020 – that is, within eight years – and they use to data to argue that this goal is “both possible and achievable."

There is value in compiling data and disaggregating them by district and school. And ConnCAN, to its credit, doesn't use this analysis as a blatant vehicle to showcase its entire policy agenda, as advocacy organizations often do. But I am compelled to comment on this report, mostly as a springboard to a larger point about expectations.

However, first things first – a couple of very quick points about the analysis. There are 60-70 pages of district-by-district data in this report, all of it portrayed as a “roadmap” to closing Connecticut’s achievement gap. But it doesn't measure gaps and won't close them.

When Growth Isn't Really Growth

Let’s try a super-simple thought experiment with data. Suppose we have an inner-city middle school serving grades 6-8. Students in all three grades take the state exam annually (in this case, we’ll say that it’s at the very beginning of the year). Now, for the sake of this illustration, let’s avail ourselves of the magic of hypotheticals and assume away many of the sources of error that make year-to-year changes in public testing data unreliable.

First, we’ll say that this school reports test scores instead of proficiency rates, and that the scores are comparable between grades. Second, every year, our school welcomes a new cohort of sixth graders that is the exact same size and has the exact same average score as preceding cohorts – 30 out of 100, well below the state average of 65. Third and finally, there is no mobility at this school. Every student who enters sixth grade stays there for three years, and goes to high school upon completion of eighth grade. No new students are admitted mid-year.

Okay, here’s where it gets interesting: Suppose this school is phenomenally effective in boosting its students’ scores. In fact, each year, every single student gains 20 points. It is the highest growth rate in the state. Believe it or not, using the metrics we commonly use to judge schoolwide “growth” or "gains," this school would still look completely ineffective. Take a look at the figure below.

A Simple Choice Of Words Can Help Avoid Confusion About New Test Results

In 1998, the National Institutes of Health (NIH) lowered the threshold at which people are classified as “overweight." Literally overnight, about 25 million Americans previously considered as having a healthy weight were now overweight. If, the next day, you saw a newspaper headline that said “number of overweight Americans increases," you would probably find that a little misleading. America’s “overweight” population didn’t really increase; the definition changed.

Fast forward to November 2012, during which Kentucky became the first state to release results from new assessments that were aligned with the Common Core Standards (CCS). This led to headlines such as, "Scores Drop on Kentucky’s Common Core-Aligned Tests" and "Challenges Seen as Kentucky’s Test Scores Drop As Expected." Yet, these descriptions unintentionally misrepresent what happened. It's not quite accurate - or at least highly imprecise - to say that test scores “dropped," just as it would have been wrong to say that the number of overweight Americans increased overnight in 1998 (actually, they’re not even scores, they’re proficiency rates). Rather, the state adopted different tests, with different content, a different design, and different standards by which students are deemed “proficient."

Over the next 2-3 years, a large group of states will also release results from their new CCS-aligned tests. It is important for parents, teachers, administrators, and other stakeholders to understand what the results mean. Most of them will rely on newspapers and blogs, and so one exceedingly simple step that might help out is some polite, constructive language-policing.

A Case Against Assigning Single Ratings To Schools

The new breed of school rating systems, some of which are still getting off the ground, will co-exist with federal proficiency targets in many states, and they are (or will be) used for a variety of purposes, including closure, resource allocation and informing parents and the public (see our posts on the systems in IN, FL, OH, CO, NYC).*

The approach that most states are using, in part due to the "ESEA flexibility" guidelines set by the U.S. Department of Education, is to combine different types of measures, often very crudely, into a single grade or categorical rating for each school. Administrators and media coverage usually characterize these ratings as measures of school performance - low-rated schools are called "low performing," while those receiving top ratings are characterized as "high performing." That's not accurate - or, at best, it's only partially true.

Some of the indicators that comprise the ratings, such as proficiency rates, are best interpreted as (imperfectly) describing student performance on tests, whereas other measures, such as growth model estimates, make some attempt to isolate schools’ contribution to that performance. Both might have a role to play in accountability systems, but they're more or less appropriate depending on how you’re trying to use them.

So, here’s my question: Why do we insist on throwing them all together into a single rating for each school? To illustrate why I think this question needs to be addressed, let’s take a quick look at four highly-simplified situations in which one might use ratings.

When You Hear Claims That Policies Are Working, Read The Fine Print

When I point out that raw changes in state proficiency rates or NAEP scores are not valid evidence that a policy or set of policies is “working," I often get the following response: “Oh Matt, we can’t have a randomized trial or peer-reviewed article for everything. We have to make decisions and conclusions based on imperfect information sometimes."

This statement is obviously true. In this case, however, it's also a straw man. There’s a huge middle ground between the highest-quality research and the kind of speculation that often drives our education debate. I’m not saying we always need experiments or highly complex analyses to guide policy decisions (though, in general, these are always preferred and sometimes required). The point, rather, is that we shouldn’t draw conclusions based on evidence that doesn't support those conclusions.

This, unfortunately, happens all the time. In fact, many of the more prominent advocates in education today make their cases based largely on raw changes in outcomes immediately after (or sometimes even before) their preferred policies were implemented (also see here, here, here, here, here, and here). In order to illustrate the monumental assumptions upon which these and similar claims ride, I thought it might be fun to break them down quickly, in a highly simplified fashion. So, here are the four “requirements” that must be met in order to attribute raw test score changes to a specific policy (note that most of this can be applied not only to claims that policies are working, but also to claims that they're not working because scores or rates are flat):

The Structural Curve In Indiana's New School Grading System

The State of Indiana has received a great deal of attention for its education reform efforts, and they recently announced the details, as well as the first round of results, of their new "A-F" school grading system. As in many other states, for elementary and middle schools, the grades are based entirely on math and reading test scores.

It is probably the most rudimentary scoring system I've seen yet - almost painfully so. Such simplicity carries both potential advantages (easier for stakeholders to understand) and disadvantages (school performance is complex and not always amenable to rudimentary calculation).

In addition, unlike the other systems that I have reviewed here, this one does not rely on explicit “weights," (i.e., specific percentages are not assigned to each component). Rather, there’s a rubric that combines absolute performance (passage rates) and proportions drawn from growth models (a few other states use similar schemes, but I haven't reviewed any of them).

On the whole, though, it's a somewhat simplistic variation on the general approach most other states are taking -- but with a few twists.

The Data-Driven Education Movement

** Also reprinted here in the Washington Post

In the education community, many proclaim themselves to be "completely data-driven." Data Driven Decision Making (DDDM) has been a buzz phrase for a while now, and continues to be a badge many wear with pride. And yet, every time I hear it, I cringe.

Let me explain. During my first year in graduate school, I was taught that excessive attention to quantitative data impedes – rather than aids – in-depth understanding of social phenomena. In other words, explanations cannot simply be cranked out of statistical analyses, without the need for a precursor theory of some kind – a.k.a. “variable sociology” – and the attempt to do so constitutes a major obstacle to the advancement of knowledge.

I am no longer in graduate school, so part of me says: Okay, I know what data-driven means in education. But then, at times, I still think: No, really, what does “data-driven” mean even in this context?