◂ CONTENTS SKILL · METRICS AND LEARNING ANALYTICS FOUND IN 3 OF 4 PROJECTS

SKILL 8 OF 9 · METRICS AND LEARNING ANALYTICS

Learning analytics

Data about what learners do, shaped so that a teacher, a delivery team or a programme board can act on it. I go into the data myself, in SQL on BigQuery?BIGQUERYGoogle’s data warehouse, where the Beeline website analytics are queried with SQL. and in Python.

WHERE IT SHOWS UP · PICK A PROJECT

PROJECT 04The Artificial Leaf
ALSO IN OTHER WORK ·Digital Education Hub ▸
WHAT I MEASURED

One grid, four contexts

WHERE
WHAT WAS MEASURED
WHO READS IT
PROJECT 02 · LEARNING METER
Engagement data behind the gamified loops. False passes, agreement, latency and cost for the AI scoring.
Before launch, to decide what shipped
PROJECT 01 · BEELINE
Fourteen metrics on a ladder: five Learning Moments rolling up into a Meaningful Next Step Rate.
A delivery team, a programme board and a community partner
PROJECT 03 · JUMP MATH
An xAPI data spec, and diagnostic, formative and summative views.
Teachers, next lesson
OTHER WORK · EDEH
Learning and space analytics: what a space can tell you about learning.
Educators and policymakers across Europe
PROJECT 01 · BEELINEFROM P. 3–8

Making fuzzy outcomes measurable

Engagement, Comprehension, Decision Quality, Confidence and Follow Through. Each metric climbs one rung at a time, from named to trended, and only when the evidence allows it. July 2026 is the baseline for job-search clicks: 1,925.

LM-01Engagement
LM-02Comprehension
LM-03Decision Quality
LM-04Confidence
LM-05Follow Through
OPEN THE METRIC LADDER ▸
PROJECT 01 · BEELINE · PYTHON AND SQLFROM P. 3–8

Analysing the gateways myself

I use BigQuery over the Google Analytics data to analyse how young people move through the gateway pages. The randomised gateway order test came out at z = −6.75, p < 0.001. The analytics tool could only compare three segments at once, which I flagged as a tool limit, and I asked for a user-level export.

  1. 01QuestionTied to one metric on the ladder.
  2. 02QuerySQL on BigQuery.
  3. 03AnalysePython.
  4. 04ReadThe Ladder of Misinference?LADDER OF MISINFERENCEAlex Edmans’ four checks before trusting a finding: is it accurate, is the sample representative, could something else explain it, and does it hold beyond this case?, before anything is claimed.
OPEN THE DECISION TRAIL ▸
PROJECT 03 · JUMP MATH LABFROM P. 15–18

An xAPI spec the dashboards read

An xAPI?XAPIA data standard that records what learners do inside a course, so dashboards can use it. data spec, with needs separated from wants for the developers, so the teacher dashboard could show diagnostic, formative?FORMATIVEChecks that happen during learning, to adjust teaching. Summative checks measure what was learned at the end. and summative views.

ROUGH WORK · SKETCHED FIRST · 1 SKETCH ▾
OPEN PROJECT 03 ▸
PROJECT 02 · LEARNING METERFROM P. 9–14

Engagement data into a redesign

I redesigned the gamified loops from engagement data and A/B tested the redesign. Learner participation rose by 40%.

OPEN PROJECT 02 ▸
TOOLS
PythonSQLBigQueryGoogle AnalyticsxAPIJupyter notebooksGoogle Colab
RELATED SKILLS
WORK WITH MEMetrics and learning analyticsMetric frameworks, dashboards and A/B tests
ASK ME FOR A DAY RATE ▸