When Dummy Variables aren’t Smart Enough: More Comments on the NJ CREDO Study
This is a brief follow up on the NJ CREDO study, which I wrote about last week when it was released. The major issues with that study were addressed in my previous post, but here, I raise an additional non-trivial issue that plagues much of our education policy research. The problems I raise today not only plague the CREDO study (largely through no real fault of their own…but they need to recognize the problem), but also plague many/most state and/or city level models of teacher and school effectiveness.
We’re all likely guilty at some point in time or another – guilty of using dummy variables that just aren’t precise enough to capture what is that we are really trying to measure. We use these variables because, well, they are available, and often, greater precision is not. But the stakes can be high if using these variables leads to misclassification/misidentification of schools for closure, teachers to be dismissed, or misidentification of supposed policy solutions deserving greater investment/expansion.
So… what is a dummy variable? Well, a dummy variable is when we classify students as Poor or Non-poor by using a simple, single income cut-off and assigning, for example, the non-poor a value 0f “0″ and poor a value of
We’re all likely guilty at some point in time or another – guilty of using dummy variables that just aren’t precise enough to capture what is that we are really trying to measure. We use these variables because, well, they are available, and often, greater precision is not. But the stakes can be high if using these variables leads to misclassification/misidentification of schools for closure, teachers to be dismissed, or misidentification of supposed policy solutions deserving greater investment/expansion.
So… what is a dummy variable? Well, a dummy variable is when we classify students as Poor or Non-poor by using a simple, single income cut-off and assigning, for example, the non-poor a value 0f “0″ and poor a value of
