James Cantonwine

If someone likes Visible Learning's tables, the EEF's work is more defensible, while still being easy enough to digest by folks out in schools. It's one that I refer people to after my disdain for meta-meta-analyses slips out. (I don't try to hide it anymore, so that's a low bar.)

In my district it's a topic of discussions about enrollment and budgets for next year. In the vein of everything being local, we're seeing different trends than our neighbors: our high school enrollment is actually up (mostly transfers from a neighboring district).

Are you aware of any work that takes potential changes in birth rates into account when looking at post-pandemic trends? In my region, that's a question I hear but haven't seen analyzed.

I love how in his demand for longitudinal data he shows his ignorance of how NAEP works and reveals a belief that SAT/ACT data would somehow improve sampling concerns. It's a really interesting intersection of Gell-Mann amnesia and the Dunning-Kruger effect.

SAT Suite of Assessments – Reports | College Board Download and review student participation and performance results for the tests that make up the SAT Suite of Assessments.

For the SAT: reports.collegeboard.org/sat-suite-pr...

This has a whole different sampling problem, as your tested population is, almost by definition, not representative of all students.

I think NAEP has value and am not discounting the work that continues to go into it, but I strongly disagree that it's the best assessment to use to motivate or measure improvement at the student, school, and district levels. The samples are too small and the test just isn't designed for that.

At the state level, NAEP scores are something you could potentially optimize for, but to really do so at secondary would require 1) giving up the autonomy on standards-writing that your voters presumably like and 2) agreeing that the way NAEP measures achievement is "better" than your other options.

"Impossible" is too strong, but the idea is directionally correct for a host of reasons. At the district or school levels, NAEP scores are something you cannot optimize for. In my own district of ~9,000 students, exactly zero 8th graders were selected to be tested in 2026.

And consistent over time!

Chatbots (mostly) won't change education They're just language calculators.

This one guy wrote a great post about it a couple of years ago that I still send to people. You should check it out, I'm sure you two will have a lot in common.

pershmail.substack.com/p/chatbots-m...

As more assessments went digital (including NAEP), how certain are we that we are still measuring the same things?

Just chiming in that one of the most effective (and least discussed) reasons for 1:1 initiatives at the district and school levels was the transition to digital testing for accountability at the state level. The costs for the number of devices needed made 1:1 look much more attractive.

Actively working to improve math achievement as measured on standardized tests seems less socially acceptable than working to improve reading scores (which still isn't a socially acceptable goal in all places).

I wonder if ideological capture is more pervasive in math. Many policy documents at the state, district, and school levels aren't much different. There are a lot of subtly different takes on inquiry math, but working toward a philosophy and working to increase achievement might not be the same.

It's easier to see gains from the (also important) language development parts of Scarborough's reading rope on these kinds of assessments, but once you've picked that fruit, you're left with something much, much harder.

I think that's pretty easily explained by the background knowledge component of reading and the lack of clearly defined knowledge we intend to teach and measure on a secondary reading test. Even if we did have a state with a clearly articulated knowledge progression, that wouldn't match NAEP.

The algorithm will see you now - Works in Progress Magazine Radiology combines digital images, clear benchmarks, and repeatable tasks. But replacing humans with AI is harder than it seems.

This is a good structure for a piece on why AI is unlikely to replace teachers at scale any time soon.

Those claims always seem to be based on naive understandings of what people actually do in their jobs.
worksinprogress.co/issue/the-al...

To be clear, that's still not good news: it's that the specific causes and interventions aren't identifiable from these data. Even if we had state-by-state comparisons we could examine engagement effects more closely, but we really can't with what we have.

It's hard to measure things with high school students than people realize or care to admit. We don't have evidence that levels of effort on NAEP itself are consistent over time, and we do have circumstantial evidence (attendance rates) that engagement with school overall has been declining.

Google’s New “Learn Your Way” Could Be Very Useful In Making Boring Textbooks More Accessible. Google has had a very difficult time finding useful ways to use Artificial Intelligence.  Their NotebookLM is somewhat useful, and their storybook application is genuinely great (see Google …

Google’s New “Learn Your Way” Could Be Very Useful In Making Boring Textbooks More Accessible. larryferlazzo.edublogs.org/2025/09/16/g...

It's high time for cognitive science to reach math the way it has with reading.

I would have loved to see more examples of the "delivery system without dead ends." New ideas in that space over the past few years seem to have introduced more problems than they solve (pun very much intended).

One of the best pieces of advice on teaching I ever received was to frequently ask the class, "Why would an intelligent person choose ___?"

OPINION: If we are going to build AI literacy into every level of learning, we must be able to measure it Students and workers are being told to “learn AI.” But while definitions of AI literacy are starting to emerge, we still lack a consistent, measurable framework to know whether someone is truly ready ...

I'm not optimistic about the odds of a national definition of AI literacy that is clear enough to actually assess. State level definitions have more promise, but I think this is something that districts likely need to be thinking about themselves.

hechingerreport.org/opinion-scho...

Are the outputs the way I would make them?

No, and that's why I'm using the tool. I feel like I've been stuck in a rut with how I communicate certain things and am looking for alternatives. It's like getting ideas from someone who knows our context but hasn't lived it. Pretty handy so far.

I'm finding that using GPT 5-Thinking for office tasks is a bit like collaborating with a colleague who doesn't like to process things out loud. You ask a question or pose an idea, wait through some silence, and then receive a complete thought.

I think that might depend on the major, and that the ability to do certain large tasks without supervision is sometimes the best (only?) way to assess knowledge "authentically." (I'm not at all convinced that "authentic" assessment is a good goal, but it is certainly some people's goal.)

In the short term, more blue book-style assessment seems like the play in K-12. In the medium term, I'm more convinced by the ideas here.

Some of the best discussions I've had with science PLCs started with, "What content are you going to intentionally not teach?" (This was before de-implementation had its moment.) That always prompted a better discussion than asking what was most important, which always led to a list of everything.

Are they expected to use a specific AI tool or just any AI tool?

I sometimes see folks in K-12 locking themselves into specific tools that might not actually have the capability they wanted or expected.

It's a fun place to watch the range of reactions people have!