A fixed condition order can lead to a wrong conclusion about why responses differ. Counterbalancing in psychology research varies the order of conditions across participants to address that problem.
In a repeated-measures comparison, the same participant experiences each condition, but different participants can receive different orders. For three conditions, check both where each condition occurs and which condition immediately precedes another. A plan can balance positions while leaving adjacent pairs unbalanced. The worked paper audit below shows the difference and provides a blank allocation record; it does not prescribe a sample size or guarantee that carryover effects disappear.
- Name the conditions before listing their orders.
- Count positions and directed neighboring pairs separately.
- Weight those counts by the number assigned to each route.
- Reconsider the design if an earlier condition causes lasting change.
Use the Research Methods guide for the wider design context. If the conditions or outcomes are still vague, complete the operational-definition worksheet first.
What does counterbalancing change?
Counterbalancing changes who encounters each condition order. It does not mean that each participant completes only one condition: in the repeated-measures plan here, each completes A, B and C once, in their assigned sequence.
The BCcampus experimental-design chapter describes practice, fatigue and context as ways an earlier experience can influence later responses. If everyone receives ABC, the identity of a condition is tied to its position. Varying routes makes that fixed association avoidable, subject to the allocation and design assumptions.
For this original paper exercise, let A, B and C stand for three clearly specified task conditions. Letters identify conditions; a route such as BCA specifies their order. Colors in the photographs are a visual aid, not a proposed manipulation or measurement.
Keep recruitment separate from route assignment. Randomly assigning the available participants to planned route slots does not make their recruitment a random population sample. Also distinguish unrestricted random ordering, which need not produce equal route counts in a small group, from deliberately balancing route allocations.
Which orders balance the three positions?
Using all six permutations with equal numbers on each route balances the first, second and third positions. The smaller cyclic set ABC, BCA and CAB also balances these positions when its three routes receive equal numbers.
These are original finite counts for a hypothetical plan, with each condition presented once per route. Each row below contributes once to the unweighted audit; it is not a participant result or a recommended class size.
On a narrow screen, swipe a table sideways to read its remaining columns.
| Route | First | Second | Third | In the cyclic subset? |
|---|---|---|---|---|
| ABC | A | B | C | Yes |
| ACB | A | C | B | No |
| BAC | B | A | C | No |
| BCA | B | C | A | Yes |
| CAB | C | A | B | Yes |
| CBA | C | B | A | No |
To check position balance, count down each column. For example, A is first in ABC and ACB, second in BAC and CAB, and third in BCA and CBA.
| Condition | Full set: first / second / third | Cyclic set: first / second / third |
|---|---|---|
| A | 2 / 2 / 2 | 1 / 1 / 1 |
| B | 2 / 2 / 2 | 1 / 1 / 1 |
| C | 2 / 2 / 2 | 1 / 1 / 1 |
The cyclic arrangement forms a Latin square: each condition appears once in each row and each position column. That property describes its layout. It does not by itself establish balance for immediately preceding conditions, the distinction made in Penn State’s crossover-design notes, section 12.3.
Why can balanced positions still leave adjacent pairs unbalanced?
The cyclic set uses only A→B, B→C and C→A as immediate transitions. Its position table looks balanced, but the reverse transitions never occur.
Count directed adjacent pairs within each route. ABC contributes A→B and B→C. It does not contribute A→C, because those conditions are not neighbors. It does not contribute C→A, because the route ends at C. Do not connect the last condition of one participant to the first condition of another.
For the same one-count-per-route hypothetical audit:
| Immediate pair | All six routes | ABC, BCA, CAB only |
|---|---|---|
| A→B | 2 | 2 |
| A→C | 2 | 0 |
| B→A | 2 | 0 |
| B→C | 2 | 2 |
| C→A | 2 | 2 |
| C→B | 2 | 0 |
There are two transitions in each three-condition route. The full set therefore contributes 12 transitions, distributed equally across the six directed pair types. The cyclic set contributes six transitions, concentrated in three pair types. These totals are arithmetic checks of the listed routes, not measurements of psychological effects.
For example, A→B occurs in ABC and CAB, whereas B→A occurs in BAC and CBA. Looking only at the cyclic routes would never expose A immediately after B.
Penn State calls equal predecessor frequencies balance with respect to first-order carryover, meaning carryover from the immediately preceding period. In the full six-route example, every distinct directed pair occurs twice. This is a property of the schedule under equal route weights; it does not show that carryover is absent, equal in size, short-lived, or harmless under every analysis.
How do you audit actual route allocations?
Count each route as many times as it is actually represented. A list of six available routes is insufficient if most people receive the same one.
Copy this record onto paper or use the browser’s print function. Fill the planned column before allocation and retain completed counts separately. Leave unknown entries blank or write “not recorded”; a missing count is not a zero. This is an original teaching record, not a validated research instrument.
| Route | Planned number | Completed number | Deviation or missing record |
|---|---|---|---|
| ABC | _____ | _____ | _____ |
| ACB | _____ | _____ | _____ |
| BAC | _____ | _____ | _____ |
| BCA | _____ | _____ | _____ |
| CAB | _____ | _____ | _____ |
| CBA | _____ | _____ | _____ |
If a route is intentionally unused, enter zero and state why. If an attempt stops partway through, record its actual condition history; do not count an unfinished route as a complete sequence. Keep anonymous route records rather than names for this paper exercise.
Hypothetical uneven-allocation example: suppose the plan has seven route slots, with one on each of the six routes and the extra slot on ABC. This is a calculation example, not a sample-size recommendation. The weighted position counts become A: 3/2/2, B: 2/3/2 and C: 2/2/3. A→B and B→C each occur three times; the other four directed pairs occur twice. The route list is complete, but its allocations no longer yield the equal counts in the earlier tables.
For an original planning exercise, write the route-allocation rule before viewing outcomes—for example, random assignment to a prepared set of route slots. When the total cannot be divided equally among the chosen routes, record the extra allocations explicitly and recalculate both tables. Calling counts “approximately equal” should not conceal which positions or pairs are more frequent.
Finish the audit with a bounded statement:
“Using the recorded route counts, our plan has _____ occurrences of each condition in each position and _____ occurrences of each directed adjacent pair. It balances _____. It leaves _____ unequal or unresolved.”
The Stroop demonstration planning sheet is one place to apply this reasoning to named conditions, with timing, errors and interruptions recorded separately.
When should you stop and reconsider repeated measures?
Stop treating an order swap as a solution when an earlier condition changes what a later condition can mean. If one task teaches the answer or reveals a strategy that persists, doing another task afterward cannot recreate a naive starting point merely by waiting or reversing the order.
Penn State’s section 12.2 on crossover disadvantages explicitly discusses permanent learning from educational tests. The BCcampus chapter likewise explains why lasting changes can make a within-subjects comparison unsuitable. A proposed pause needs a task-specific justification; this worksheet supplies no universal waiting period.
For that situation, discuss an alternative design with the instructor or research supervisor, such as a between-subjects comparison. Changing designs introduces other tradeoffs and does not automatically solve recruitment, measurement or confounding problems. This page is a paper design exercise and does not authorize collecting data from classmates.
Method and sources
The definitions and limits were checked against BCcampus’s Research Methods in Psychology, “Experimental Design,” using its full text archived October 11, 2026, and Penn State Department of Statistics’ live STAT 509, Lesson 12, sections 12.2–12.3, accessed October 16, 2026. The BCcampus live URL returned an access challenge during this update; its archived chapter was used rather than the challenge page.
The route tables, counts, blank record, uneven-allocation example and conclusion prompt are original editorial teaching materials. Counts were checked by enumerating every permutation of A, B and C and tallying only within-route neighboring pairs. They are finite arithmetic examples, with no participant data, classroom validation, normative scores or claimed learning gains. The key check is to state exactly which feature is balanced before interpreting a condition difference.
