On Bloom's two sigma problem (2020)(nintil.com)
nintil.com
On Bloom's two sigma problem (2020)
https://nintil.com/bloom-sigma
14 comments
I was trying to find this blog post when that post whining about pair programming was going around. 1:1 interaction has significant positive effects (even if not quite 2 sigma).
If anyone on HN has ever used an "intelligent tutoring system" (software that mimics 1:1 tutoring with an expert) I'd like to hear about it.
This is quality writing for a blog post. Bet it would be accepted for publication as a review article in a suitable scientific journal.
Why would the author go to that much extra effort to reach fewer people, slower?
I am not saying he should, just an observation about high quality of the text.
But also to your question - it's not the quantity of readers that matters. Reaching the right people, for whom this text will not just be an afternoon entertainment read, is more important.
But also to your question - it's not the quantity of readers that matters. Reaching the right people, for whom this text will not just be an afternoon entertainment read, is more important.
tl;dr
- Bloom's finding of 2 sigmas for tutoring seems to be an outlier
- However, both tutoring and "mastery learning" do have a significant positive effect. It is possible that part of the benefits of these methods can be simulated by software
- The effects are strongest for less capable students. Average and gifted students might not benefit significantly from ML/DI
- It is unclear how much these techniques are overfitting to a test, or whether they generalise to other life skills
- General scepticism about the quality of educational research, the expected value of new methods, and overall value of classroom education in general
- Bloom's finding of 2 sigmas for tutoring seems to be an outlier
- However, both tutoring and "mastery learning" do have a significant positive effect. It is possible that part of the benefits of these methods can be simulated by software
- The effects are strongest for less capable students. Average and gifted students might not benefit significantly from ML/DI
- It is unclear how much these techniques are overfitting to a test, or whether they generalise to other life skills
- General scepticism about the quality of educational research, the expected value of new methods, and overall value of classroom education in general
The main benefit of ML to average/gifted students is long-term, and would indeed not show up in a formal experiment involving a simple "class". Namely, expecting the student to achieve mastery averts problems with educational "gaps" further along the curriculum; and letting the student learn at their optimal pace can help them learn the existing material faster or in more depth, which benefit also compounds throughout a full curriculum.
DI as a method includes a "90% of all taught material must be review" rule which while helpful for the less proficient students who are the main focus of the method, is probably overkill even for an average student let alone a gifted one.[0] What's more relevant is that instruction should be as unambiguous as possible and that the mastery-learning principle should be followed.
[0] It's also a plausible factor behind the common complaint that DI involves less "creativity, joy and spontaneity" than more popular and commonplace educational methods. This plausibly reflects the outlook of above-average students, but likely not of less proficient ones, for whom 'review' and a tight feedback loop is very helpful indeed!
DI as a method includes a "90% of all taught material must be review" rule which while helpful for the less proficient students who are the main focus of the method, is probably overkill even for an average student let alone a gifted one.[0] What's more relevant is that instruction should be as unambiguous as possible and that the mastery-learning principle should be followed.
[0] It's also a plausible factor behind the common complaint that DI involves less "creativity, joy and spontaneity" than more popular and commonplace educational methods. This plausibly reflects the outlook of above-average students, but likely not of less proficient ones, for whom 'review' and a tight feedback loop is very helpful indeed!
I might be misremembering, but isn't Oxford's preferred teaching model similar to 1-1 tutoring? If so, doesn't that go at odds with "Average and gifted students might not benefit significantly from ML/DI"? I assume there must be some benefit.
Possibly selection bias - is there evidence that a student good enough for Oxford would get worse outcomes when attending, say, Harvard, and missing out on tutoring?
One-on-one tutoring provides much, much better outcomes than one-to-thirty teaching. If you really need statistical evidence for that, then maybe you need statistical evidence for the fact that jumping from a helicopter in flight leads to death [1].
[1] https://www.bmj.com/content/363/bmj.k5094
[1] https://www.bmj.com/content/363/bmj.k5094