Why Teachers Don't Trust Mathematical Models

15 Jun 2020 -

In my education career, data was always at the forefront of my decision making. I wanted as much information as I could before making decisions and I kept myself motivated by seeing progress in students. Not everyone I worked with was driven by data, and at first, I thought that this was mainly due to a lack of confidence in their math skills and a lack of understanding in how data informed decisions can work. I came to learn that this wasn’t the reason most of the time.

The main concern is typically context. When teachers are presented with research or a data informed decision, they jump to thinking about the context. In what context was the research planned in, in what context was the data gathered in, how different are those contexts than the one at our school? And this response, in my opinion, was always right. And always wrong. Right, because context always matters, and the local context is always different from the global context, the current context is different from the context when data was gathered, and teachers are always placed where they best know the context of what’s really going on in the school. And wrong, because it is extremely rare where the insights gathered in another context are of no use whatsoever. Thus, dismissing any insight due to perceiving that the contexts are just too different often isn’t the answer.

One thing that surprised me in my initial study of machine learning and artificial intelligence is how imporant context is in every model. These systems that are meant to find patterns that humans can’t, depend so much on the human decisions made in gathering data, prioritizing features, and choosing how to explain models to decision makers and people that will be executing on any new policies. In my experience with the few times mathematical models were used in my schools, this was always clear to the teachers. They wanted to know the context before moving forward. How was the data collected? Why was the specific data chosen? What are the limitations of the model? While this, at times, came across as resistance to change, or a lack of trust in leadership, it was always the right response. Without a chance to understand all these questions, decisions can’t be made from the data, and without that understanding, models shouldn’t be used to impact students’ experiences and outcomes.

Thus, I think it is essential for teachers to be part of the data gathering, feature development, model interpretation, and decision making process. All teachers will still have the questions above when presented with new data and new models, but they will trust the models and conclusions more if they know that current teachers that know current contexts were part of the process all along the way. They will trust that while their specific concerns or insights may have been missing from this current project, that the leadership of their school or district is actively seeking out that input.

Without this teacher involvement, school districts and administrators run the risk of taking a tool that has the power to predict inequalities in the system and monitor the effectiveness of policies to minimize them and turn it into a tool that makes teachers feel disenfranchised. A standard response from administrators is “I hear your concern, I see where it comes from; but, the data (or research) says otherwise.” And while this may seem like a good way to project how informed you are about the situation, teachers know that data and research seldom give such a clear answer. Educators know that the context and limitations of the research and data gathering mattered and that the bias of the decision makers mattered in how they moved from the research and data to their decision. So, I think education leaders need to adjust the language from “The data shows…” to “The data suggest…, and there seems to be some very clear similarities between the context where this data was gathered and the current context at our school, and we have additional confidence due to working closely with teachers at each step of the process.” Stronger models informed by more data will likely have a smaller impact than is possible without a paradigm shift like this one.