Conducting an experiment (internal)
Internal 3 credits Experimental design Random allocation Causation re-randomisation ConfoundingAS 91265, 3 credits. The one statistics standard where you can legitimately claim causation — but only because of one design feature: random allocation to treatment groups. Everything else follows from that.
1 Reading the notation
The situation is given, but the data collection is the student's own — unlike 91264's sampling from an existing data set.📖 Terms the schedule grades you on
3 terms- Conduct an experiment (Achieved)
- Evidence of using EACH component of the investigation process.
- With justification (Merit)
- Link components to context; explain considerations; support findings with evidence from the experiment.
- With statistical insight (Excellence)
- Integrate statistical and contextual knowledge throughout — reflecting on process, or considering other variables.
2 Why an experiment can claim causation
Random allocation is the entire logical basis of this standard. ✕Letting people choose their group
Self-selection destroys the causal claim, because whatever made someone choose the treatment may also affect the outcome. It converts a genuine experiment into an observational study — and the conclusion must then drop the word caused.3 Planning the design
The plan must be visible as a plan, written before the data — allocation and control are named explicitly.The plan the standard expects to see
- The investigative question, naming the treatment and the measured outcome.
- The experimental units, and how many in each group.
- The allocation METHOD — and that it was random, with how the randomisation was done.
- The variables held constant, listed.
- How the outcome will be measured, identically for both groups.
4 The re-randomisation test
The Level 2 tool for deciding whether an observed difference is more than chance.Reading a re-randomisation result
- The actual difference sits FAR OUT in the shuffled distribution → treatment likely caused it.
- The actual difference sits comfortably INSIDE the shuffled distribution → chance is plausible.
- State roughly how many of the shuffles were as extreme as the real result.
5 Ethics and the conclusion
What the conclusion can — and cannot — legitimately claim.| Grade | What it takes | The half people drop |
|---|---|---|
| Achieved | Evidence of using EACH component of the process | Missing one component entirely — every one needs visible evidence |
| Merit ("with justification") | Link components to the context; support findings with evidence from the work | Asserting a decision instead of explaining WHY, for this context |
| Excellence ("with statistical insight") | Integrate statistical and contextual knowledge THROUGHOUT | Treating it as one closing paragraph rather than woven through every component |
"With statistical insight" often means reflecting on the process itself, or naming another variable/explanation the data cannot rule out.
Practice
Recall it first, then use it — testing yourself builds the memory, where rereading the notes above only feels easier.
Recall and application shuffled together — you have to work out what each item wants before you can answer it, which is the part the exam actually tests. Both kinds feed the same spacing schedule, so anything you miss comes back sooner.
Say the answer out loud before flipping — that attempt is where the learning happens, even when you get it wrong.
Exam-style only. Instant feedback, shuffled each attempt, best score saved on this device.