Collection

Check what a study can actually tell you

Inspect sampling, uncertainty and analysis choices before converting a statistical result into a decision.

15 elements · Explore in any order.

3 checklists · 4 concepts · 4 protocols · 3 questions · 1 template

Narrow your search

15 elements

Question

Check who could enter the sample

A million answers from the wrong doorway still came through that doorway.

Read the idea
Checklist

Treat missing responses as an unanswered question

Silence is not a neutral rating waiting to be filled in.

Read the idea
Concept

Ask what was randomized

A random draw into a group and a random allocation inside it solve different problems.

Read the idea
Protocol

Hide condition labels from the person judging the outcome

The evaluator should not have to fight the label before judging the work.

Read the idea
Concept

Distinguish variation in cases from uncertainty in a mean

Customers can have wildly different experiences while the average is estimated very precisely.

Read the idea
Concept

Read a confidence interval as a procedure with assumptions

The interval is useful because of how it was made, not because 95 is a reassuring number.

Read the idea
Concept

Keep the condition attached to a p-value

The small number does not answer every question that can be phrased with 'chance'.

Read the idea
Template

Name the smallest change worth acting on

Detectable is not the same as worth the migration.

Read the idea
Question

Ask what a nonsignificant study could have missed

An uncertain answer should not be promoted to a confident zero.

Read the idea
Checklist

Count the comparisons behind the winning result

The winning p-value may have had a large audition pool.

Read the idea
Protocol

Timestamp the analysis plan before seeing the answer

Memory is a generous editor of what you supposedly expected.

Read the idea
Protocol

Keep a fresh dataset for the discovery's next test

Data that helped invent the rule cannot also be completely surprised by it.

Read the idea
Protocol

Preserve the pair in a before-and-after comparison

Two columns may contain a relationship that their separate averages conceal.

Read the idea
Question

Do not mistake frequent readings for independent evidence

Recording the same slow-moving process every second does not create a new world every second.

Read the idea
Checklist

Keep the ingredients of a ratio

The same ratio can hide two very different stories.

Read the idea