Counterbalancing
In a within-subjects study each person does every condition, so the order can change the results: people get practiced or tired, and one condition can carry over into the next. Counterbalancing gives different people different orders, so those effects even out across the study.
Make balanced orders for your own conditions below. The generator uses the balanced Latin squares the ConductExperience study builder counterbalances with, and runs in your browser.
Balanced Latin square generator
4 conditions: 4 orders. Each condition is once in each position, and directly follows every other condition equally often.
| Order | 1 | 2 | 3 | 4 |
|---|---|---|---|---|
| 1 | A | B | D | C |
| 2 | B | C | A | D |
| 3 | C | D | B | A |
| 4 | D | A | C | B |
Ways to counterbalance
| Design | What it does |
|---|---|
| Every order | Each possible order once: 6 for 3 conditions, 24 for 4 conditions and 120 for 5 conditions. It balances everything, but the number of orders grows fast. |
| Latin square | As many orders as conditions, with each condition once in each position. Position is balanced, but a condition can follow the same one in every order. |
| Balanced Latin square | Also called a Williams design. Each condition is once in each position and directly follows every other condition equally often. An even number of conditions needs as many orders as conditions, an odd number twice as many. |
| A random order for each person | Each person gets a shuffled order. It balances on average, so it needs more people, and one study can still come out uneven. |
How many people
Use a multiple of the number of orders, so each order is used by the same number of people.
| Balanced Latin square | Orders | People |
|---|---|---|
| 3 conditions | 6 | 6, 12, 18 |
| 4 conditions | 4 | 4, 8, 12 |
| 5 conditions | 10 | 10, 20, 30 |
| 6 conditions | 6 | 6, 12, 18 |
Counterbalancing in ConductExperience
- Set the study's sections, or the parts inside a section, to a counterbalanced order, and the study builder orders them with a balanced Latin square, for 2 to 12 of them.
- Each new person gets one of the orders the fewest people have so far, so the orders stay even as people join.
- The Participants table shows each person's order, and the export records it as order_index.
- A tasting's samples are counterbalanced the same way, so each person gets one serving order of the square.
- A section can instead shuffle its parts for each person: the random order in the table above.
- A section's Each person sees can show each person some of its parts, drawn at random once and kept when the person comes back and at each visit, in the section's order. The draw is not balanced across people: a study that needs balance uses groups.
Limits
- The generator takes up to 12 conditions, the most a counterbalanced section takes.
- A balanced Latin square evens out the condition that comes directly before each one, not conditions two or more steps back.
- Counterbalancing spreads order effects across people. It does not remove them from any one person's data.
Related
Study builder
Groups, sections and parts in fixed, shuffled or counterbalanced orders.
Sensory testing
Tastings on tablets with three-digit codes and balanced serving orders.
Online studies
One link for a panel or a mailing list.
Templates
Studies to start from, with their orders already set.
Planning a study?
Start one now, or tell us what you want to measure.