Why I Keep Starting Over

Online course return rate falls to 7%

The internet made learning easy to start, but staying long enough to finish remains the scarce part.

▼ Falling80% confidencehorizon Dec 2026data as of May 2018

Confidence-weighted forecast · history → projection

Tracking share of first-time HarvardX and MITx online learners who enrolled in another course the next year (percent of first-time learners)

ObservedProjectionshaded band = confidence cone

80%

Confidence

2026

Horizon

Jun 2026

First called

Tracking

Status

The call

Starting a course gives you the feeling of becoming someone new. Finishing asks for the boring part: repetition, confusion, and slow improvement.

The number to notice

Next-year return among first-time HarvardX and MITx learners fell from 38% for the first cohort to 7% for the 2016-2017 cohort.

What’s driving this

courses are easy to start
motivation fades faster than skill builds
free or low-cost access lowers commitment
learners chase the next better resource instead of finishing reps

Sources · check usOpen

The MOOC Pivot ↗Science · 2019-01-11study The MOOC pivot ↗MIT Teaching Systems Lab · 2019study Study offers data to show MOOCs didn't achieve their goals ↗Inside Higher Ed · 2019-01-16news

For you

Your advantage is not finding the perfect course. It is finishing enough reps that the skill becomes real when motivation wears off.

What would change our mind

This call weakens if large online-learning platforms publish comparable cohort data showing next-year learner return above 20% and course completion rising without restricting access only to highly motivated paid learners.

Behind the numbersOpen

Reich and Ruipérez-Valiente analyzed HarvardX and MITx courses on edX from October 2012 to May 2018, covering 565 course iterations, 12.67 million course registrations, and 5.63 million learners. They reported that second-year retention fell from 38% in the first cohort to 7% in the 2016-2017 cohort, and that low completion rates had not meaningfully improved over six years. This is not all online learning, and Harvard/MIT learners are not representative of everyone, but it is one of the strongest public datasets on course starting versus sticking.