“Can we devise teaching-learning conditions that will enable the majority of students under group instruction to attain levels of achievement that can at present be reached only under good tutoring conditions?”
This is the 2 sigma problem as defined by Benjamin Bloom, an educational psychologist who pioneered mastery learning.
This article is inspired from a Tweetstorm I did six months ago. You can find the original thread here.
One year ago, I found Lambda School on Twitter. While researching them online, I came across a thread from Austen talking about introducing Mastery Based Progression to their cohorts. This led me to look into Bloom’s two sigma problem, why it wasn’t solved yet and how technology can help scale effective learning.
Bloom’s 2 sigma Problem
Benjamin Bloom’s research found that one on one tutoring could help achieve 2 sigma improvement over conventional teaching. His objective was to find practical, yet scalable methods of attaining two sigma improvement since tutoring wasn’t scalable in its present form as defined by physical constraints.
A brief overview of the problem.
Why is Bloom’s 2 sigma problem important?
Effectiveness of one on one tutoring indicates that aptitude achievement correlation can be reduced from 0.6 to 0.25, i.e. effective learning levels the playing field for all those who wish to learn irrespective of skill level. It would represent an order of magnitude higher impact as compared to traditional methods.
This is how the achievement distribution looks like for the three cases mentioned in the overview.
Finding Solutions for Bloom’s 2 sigma problem
Looking for solutions to the 2 sigma problem is a four step process.
- Look at the entire teaching/learning process
- Break down into mutually independent pieces
- Find intervention variables and their effect sizes
- Combine variables to achieve intended two sigma improvement
Bloom used this approach to look for viable solutions.
The most important variables with high effect sizes were: Feedback and corrective measures, Cooperative learning, Classroom participation, Tutorial instruction Reinforcement and Graded homework. A combination of these variables enabled by the introduction of suitable tech seemed like a viable solution.
The Evolution of Ed Tech
Education is one of the toughest markets to crack for startups. Regulation, long sales cycles, fragmented markets and illegibility to technology (until now) make ed tech startups difficult to scale and largely outside the purview of venture funding. Until now.
Traditional education was ripe for disruption by MOOCs but this failed miserably. Only the most motivated students could finish courses and completion rates were less than 5%. Asynchronous, do it yourself modes of learning are not effective in their current form. This quote describes why: “The big MOOCs mostly employed smooth-functioning but basic video recording of lectures, multiple-choice quizzes, and unruly discussion forums. They were big, but they did not break new ground in pedagogy.” A richer discussion on the topic can be found here.
Given the advances in tech enabled solutions for various aspects of the learning process such as lesson delivery, feedback mechanisms, ability to replay videos, instant communication, TA support via Zoom/Slack, many people predicted the end of the search for a solution to the 2 sigma problem. They were proven incorrect. Instead of the predicted need of fewer teachers, we require more to achieve 2 sigma improvement with the following caveat: teachers should utilized in an effective manner using tech as leverage.
It takes a few iterations before people find a way to execute correctly. The same pattern can be seen here. The most common startups in ed tech focused on getting scale and distribution with little consideration for summative achievement. Getting courses online wasn’t enough. This article looks at the bastardization of the 2 sigma problem. The following quote summarizes the mistake succinctly:
“So here’s where the bastardization comes in to play. I think that there has been so much focus on the tutoring part, that we’ve lost sight of the learning part. “Will it scale?” is arguably the most important question that an ed tech investor will ask. That’s great, and that’s an absolutely justifiable question. But if you read through Bloom’s paper, there are two parts to the question. Will it scale, AND, will it improve learning levels over the conventional baseline. Like with many “shiny object” technologies, we tend to focus on the scale part and gloss over the improvement part (or worse yet, just “assume” the learning will happen).”
Lambda School’s Foray into Ed Tech
A brief description of Lambda School for the uninitiated: A live, fully online school that trains people to become software engineers, data scientists and designers which is free until you get a job. Instead, students pay a percentage of their income each year after they’re employed, the maximum of which is capped at $30k.
A combination of multiple factors enabled Lambda school’s success at finding a product-market fit: rising student debt and default rates, reliable communication tools like Slack and Zoom and an innovative business model based on Income Share Agreements to name a few.
Lambda managed to crack the problem by innovating on pedagogy and identifying an arbitrage opportunity in the labor market (supply demand gap for CS grads to start with). The latter is important because it allows lambda to do what they do. Traditional education has no skin in the game and perverse incentives. Lambda is forced to act in the students best interests because it has aligned incentives. It earns only when students get high paying jobs.
A few reasons why lambda school works so well:
1. Small class sizes – each cohort is divided into sections with 8-10 students.
2. World class teachers – practitioners who devised OG curricula at places like Apple
3. Cooperative learning – group learning, capstone projects and build weeks
4. Hungry students – gritty students who will do the work (full time for 9 months)
5. Mastery learning – repeat parts of the course until students demonstrate mastery
6. Extensive TA support – feedback and corrective measures
These variables have high effect sizes as indicated by Bloom’s research. This innovation on the pedagogy front is the moat and is scalable across verticals. More from Caleb, VP of Learning at Lambda here and here. By reverse engineering the job market requirements and using their pedagogical infrastructure Lambda can provide effective education in almost any domain legible to online learning.
Lambda School works. The outcomes and demand are proof. The Series B they raised is going to help them scale. On a recent podcast, Austen mentioned their plans to have 3000 students in 2019 and a few cohorts in the EU. Scaling to millions will be an operational challenge and it will be interesting to see how Lambda tackles this.