Counterbalancing and order effects

If you present conditions A, B, C in the same order to every participant, you cannot tell whether a difference between C and A is due to the manipulation or the fact that C always comes after fatigue, practice, and context effects have kicked in. Counterbalancing is how you sort that out.

What breaks without it

• Practice effects — participants get faster on later blocks regardless of condition.
• Fatigue — they get slower and less accurate toward the end.
• Carry-over — the response mapping used in block 1 biases responses in block 2.
• Context effects — the first stimulus anchors judgement for everything that follows.

Latin-square

An N × N Latin square is a grid where each condition appears once per row and once per column. Assign participants to rows and you guarantee that each condition appears in each serial position the same number of times. With three conditions (A, B, C), the simplest balanced Latin square is:

P1: A B C
P2: B C A
P3: C A B

For N > 3, use a balanced Williams design so every condition is preceded and followed by every other condition equally often.

Blocked versus interleaved

• Blocked: run all A trials, then all B trials. Preserves within-condition learning and makes the condition manipulation salient. Good for psychophysics, adaptive staircases, training paradigms.
• Interleaved: trials from all conditions are shuffled within a block. Cancels warm-up and fatigue at the block level. Good for reaction-time paradigms where task-switching is not the variable of interest.

Mix them — interleave trials within blocks while counterbalancing block order between participants — and you get both benefits.

What SciBLIND does by default

The builder applies Williams-balanced Latin squares for block order and Fisher-Yates shuffling for within-block trial order, seeded from the per-session RNG seed so the order is deterministically replayable. If you disable counterbalancing in Trial Settings, the methodology PDF calls it out so a reviewer knows.

When you should turn it off

Rarely. The only legitimate reasons are:

1. The condition order itself is the manipulation (»study the word list, then a distractor, then recall«).
2. You are running a single-session adaptive staircase where the trial order is procedurally determined.
3. You are replicating a classic study that prescribed a specific order and you want to match it.

Otherwise — leave counterbalancing on.