But policy matters–or the advocates of high-stakes testing hopes it changes behavior–and to pretend it can only provide benefits is inconsistent with the whole theory of action of high-stakes accountability. Essentially, if you think you can change behavior, you can change it in all sorts of ways. Whether it’s beneficial on the whole is an empirical question, unless you want faith-based accountability policy.
So the relevant question for me is, under what circumstances does cheating become a salient behavior in a high-stakes regime? I’m going to go all Popperian on you and make some testable claims.
- There is a measurable variation in cheating no matter what the pressure behind a test is — that is, if we can measure cheating to be able to make some distinctions, there will be differences in cheating rates both in low-stakes environments and high-stakes environments.
- There will be more cheating in high-stakes environments than in low-stakes environments — that does not mean that there aren’t going to be high-cheating settings in low-stakes environment that are worse than some settings in high-stakes environments.
- One of the key factors in rates of cheating will be perceived opportunity, or the salience of cheating as a visible option.
- Another key factor will be the perception that cheating is a more reliable way of attaining goals (including avoiding shame/punishment) than effort.
- Of the two, salience will be more important in determining level of cheating.
This general idea has been dancing in my head a long time, and I'm glad Dorn managed to pin it down for me.
This also relates to the "but people don't have to do test prep to prepare for tests, and if they are smart they won't" argument. If there is a system of incentives that leads, say, 75% of schools to do bad test prep instead of high quality instruction, you need to fix the incentives, or simply get rid of them. Incentives don't have inherent virtue, that's kind of the point. They're only valuable to the extent that they actually, empirically result in the desired outcome.