Reduce avoidable bias by separating condition assignment from nuisance factors and expectations.
Field diagramA fair comparison
Estimated reading
10 min
Estimated practice
25 min
Equipment
Paper, random-number generator, or shuffled cards; supplied assignment table.
Progress stays in this page session. Allow lesson storage in Cookie settings to remember it; browser settings may also prevent saving.
Objective
By the end, you can:
Recognize confounding and selection bias.
Use simple randomization or blocking when appropriate.
Explain when blinding is feasible and what it protects.
See the hidden difference
A confounder changes with the condition and can also affect the outcome. If every vessel A trial is performed in the morning and every vessel B trial in the afternoon, time is entangled with vessel. If the tidiest samples are assigned to the preferred method, selection bias enters before measurement. More trials do not automatically repair a biased assignment; they may make a consistent bias look precise.
Randomization assigns conditions using a chance rule so that known and unknown nuisance factors are less likely to align with one condition. Blocking first groups comparable units, such as position or day, then randomizes within each group. Blinding keeps the person handling, measuring, or judging from knowing an assignment when expectations could influence actions or ratings. Blinding is a design choice, not a virtue to claim when it cannot be implemented.
Randomize order to distribute time, fatigue, and drift across conditions.
Block when a strong nuisance factor is known, then compare within blocks.
Use coded labels when the observer could unconsciously read a preferred result.
Record the assignment rule and code key so the design can be audited.
Use judgment, not ritual
Randomization is not useful when it violates a safety or instrument constraint. A balance may need warm-up before any sample; document that fixed condition and randomize the sample order after readiness. A blinded operator may be impossible when vessels look different, but an independent person can code files or score images. State what was blinded, when, and what remained visible.
Predefine the allocation list and preserve it before outcomes are known.
Do not call convenience order random because it feels mixed.
If a deviation breaks the allocation, record it and assess its effect rather than hiding it.
Worked example
Blocked randomization
The learner compares two water-volume methods over three days. Day is a known source of drift, so each day contains one trial of each method in randomized order.
01Block: day 1, day 2, day 3; treatment: method A or B.
02Within each day, a shuffled order assigns the first and second trial.
03A file name hides method during data entry; the key is opened after raw records are locked.
04Conclusion compares within-day differences and reports the limited number of days.
Materials
Set out what you need.
Assignment worksheet
Cards or a reproducible random-number list
Supplied paired dataset
Safety & stop conditions
Pause if the work no longer fits the plan.
This is a planning exercise. Randomization never overrides safe handling, required equipment warm-up, sample stability, or stop conditions.
Method
Work through the steps.
01
List nuisance factors
Write time, operator, location, batch, instrument position, and any factor that could affect the outcome.
02
Choose the design
Randomize order when factors are transient; block when a factor is strong and measurable; blind the judgment step when expectations may matter.
03
Freeze the assignment
Save the assignment list and coding key before data collection, then log deviations with reasons.
Checkpoints
Observe, record, investigate.
Each condition is represented across relevant time or blocks.
The assignment mechanism is reproducible from the record.
The plan names what the operator can and cannot see.
Common mistakes
Correct the process, preserve the record.
Randomizing after seeing early results.
Generate and save the allocation before outcome review; later changes are deviations.
Blocking on a factor that is actually part of the treatment.
Do not remove or average away the variable you intend to study; distinguish design factor from nuisance factor.
Claiming blinding when sample appearance reveals the group.
Describe the remaining visibility and blind an independent stage if feasible.
Practice
Repair an assignment
A supplied table records 12 measurements: all method A trials were measured by person 1 and all method B trials by person 2. Design a follow-up that separates method from operator.
Which factor is confounded with method?
Would random order alone be enough if operators cannot switch?
Answer: It reduces systematic alignment between a condition and nuisance factors or selection preferences. It cannot remove all variation or guarantee balance in a small sample.
02Why use blocks?
Answer: Blocking keeps comparisons within similar levels of a strong nuisance factor, such as day, then randomizes inside each block. It can improve comparability when the factor is not the target.
03If the operator knows the sample group but the final scorer does not, is the study blinded?
Answer: Yes, but only at the scoring stage. Report exactly who was blinded and recognize that handling bias could remain.
