← All workshopsRead Separating a Useful Prediction From a Causal Claim

Causal Inference and Experiments

Compare the observed group mixtures

Give both groups the same composition

Botie's treatment and control groups contain different proportions of dry and moist beds. Their overall means mix condition effects with group composition.

BOTIE

Within-stratum means stay fixed but both groups become half dry and half moist. Why can the aggregate comparison reverse?
The inputs we are comparing

Treatment stratum means

Same in both cases

  1. 04
  2. 19

Control stratum means

Same in both cases

  1. 03
  2. 18

Treatment proportions

Reversal

  1. 00.9
  2. 10.1

Equal composition

  1. 00.5
  2. 10.5

Control proportions

Reversal

  1. 00.1
  2. 10.9

Equal composition

  1. 00.5
  2. 10.5
Result for reversal
-3
Choose an explanation to test
A better treatment looks worse in the pooled data

This check uses the pieces from “Expose each group's own mixture”.

BOTIE

Botie compares recovery rates for mild and severe cases, in that order. Each treatment group has its own proportions of those cases. Calculate the treated group's pooled rate minus the control group's pooled rate using the supplied group-specific proportions.
Treatment stratum means
  1. 00.9
  2. 10.5
Control stratum means
  1. 00.8
  2. 10.4
Treatment proportions
  1. 00.2
  2. 10.8
Control proportions
  1. 00.8
  2. 10.2

Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.

Build a construction in the workspace above first. This case will test those pieces.

What prevents the original aggregate difference from being an immediate causal conclusion?

An observed contrast depends on who entered each group as well as their outcomes.

This is an observed difference. A causal interpretation requires additional design and assumptions.

Work through the Python challenges →
Continue the courseMaking a Fair Comparison by Assignment →
← Scoring Connections We Have Not Observed