Project DeliveryResearch ProjectsAdvancedQuantitative Research
Controlling Bias: Randomisation and Blinding
Random selection, random assignment and blinding are three different controls on three different biases. From the week's set reading — a published article in another discipline — comes the clearest account of each in the whole supplied dataset.
Reading time11 minutes
LevelAdvanced
Topic streamQuantitative Research
Source materialWeek 8 Set Reading
Updated2026-08-16
Where this comes from
What bias actually is
That parenthesis is the most useful sentence about bias anywhere in the supplied dataset, and it is worth slowing down on. Bias is not a property of your result. It is a property of your method — of what would happen on average if you ran the same procedure many times.
Control one — random selection, against selection bias
Selection bias arises when your sample is not representative of the population you want to talk about. The article's remedy is a random selection procedure, or a stratified random one where you need proportional representation of subgroups.
Control two — random assignment, against group bias
This is a different bias from the first and a different remedy. Random selection decides who enters the study. Random assignment decides which group they land in once they are in. You can have one without the other, and the two protect different claims.
WHICH CONTROL PROTECTS WHICH CLAIM
Control
What it protects
What you lose without it
Random selection from the population
That your sample resembles the population
The right to generalise beyond the people you actually studied
Random assignment to groups
That your groups differ only by the treatment
The right to attribute a difference to the treatment rather than to the groups
Blinding
That knowledge of the treatment does not create the effect
The right to separate the treatment's effect from the expectation of one
The three-way separation is this library's arrangement of the article's material; the article treats them in sequence without tabulating them.
Control three — blinding
The third control addresses a bias the first two cannot touch: that knowing which treatment you received changes the outcome, for the subject and for the researcher measuring them. Blinding withholds that knowledge — from the subjects, from the researchers, or from both.
What blinding buys
SUBJECT
Expectation does not become effect
A subject who knows they received the intervention may improve because they expect to. Blinding them separates the treatment from the expectation of it.
RESEARCHER
Measurement does not drift
A researcher who knows which group they are assessing may score ambiguous cases in the direction they expect. Blinding the assessor removes the opportunity.
UNBLINDED
The magnitude becomes measurable
In an unblinded experiment, mechanism variables can help define how much of the effect was expectation rather than treatment — which is why the article recommends measuring them.
What survives into project research
What you can still do
IfYou cannot randomly select but can define your population precisely
ThenDo that, and bound every claim to it. A precisely bounded claim about one sector is worth more than a vague one about industry.
IfYou cannot randomly assign but can measure the pre-existing differences
ThenMeasure them and report them. A reader can then judge how much of your difference the groups already carried.
IfYou cannot blind anyone
ThenSay so, and consider whether an outcome measure exists that expectation cannot move — a recorded date rather than a satisfaction rating.
IfYou have none of the three
ThenYou have a descriptive study. Report associations, decline to claim causes, and say why in the limitations. That is a defensible study, honestly described.
Getting it right
The bias section of your methods chapter
You have described the selection procedure precisely enough for a reader to judge its bias
Your response or participation rate is reported, with who is likely missing from it
If you have groups, you have said how people ended up in them
If assignment was not random, the pre-existing differences between groups are measured and reported
The pre-test value of your outcome variable is reported by group
Every control you could not apply is named, with the claim it withholds
No figure from another discipline is carried into your work as a threshold
What to carry forward
Bias is a property of your method, not of your result. You cannot inspect one result and detect it, which is why the defence is entirely at design time.
Random selection, random assignment and blinding protect three different claims. Losing one does not cost you the others.
Where assignment is not random, balance on the pre-test value of the outcome matters more than balance on anything else.
The compliance figure in the set reading is that article's report of editorial practice in its own field. It is not a threshold for your study.
Name the controls you could not apply and the claims they withhold. That is stronger than a methods chapter that claims everything.
Frequently asked questions
What is the difference between random selection and random assignment?
Random selection decides who enters your study, and protects your right to generalise to the population. Random assignment decides which group they land in once inside, and protects your right to attribute a difference to the treatment. They are separate controls on separate biases and you can have either without the other.
Is the compliance rate in the set reading a target I should meet?
No. It is that article's report of what journal editors in sport and exercise science accepted around the time it was written. It is not a project research standard and the library does not carry it forward as a rule. The teaching material's own return-rate figure is unattributable and is not a benchmark either.
Why does the pre-test value of the outcome matter most for balance?
Because if your groups already differ on the very thing you are measuring, the difference at the end may be the difference they started with. Balance on other variables cannot rescue that comparison, which is why the article singles it out.
What can I do if I cannot randomise or blind anything?
Run a descriptive study, report associations rather than causes, and say so. Measure the pre-existing differences between whatever groups you have so a reader can judge them. A methods chapter that names each missing control and the claim it withholds is stronger than one that claims everything.
Do participants need to know assignment is random?
The article says so directly — people may not be happy about being randomised, so it must be stated clearly as a condition of taking part. That is a consent obligation as much as a design one.
References and source attribution
Hopkins 2000, 'Quantitative Research Design', a sport and exercise science web journal, vol. 4, issue 1, approximately 4,318 words — the Week 8 set reading. Retained as a published bibliographic citation; the journal, the reviewer and all institutional affiliations are scrubbed per the library's naming rules.
The supplied teaching source: weekly study notes, slide decks and assessment activities for a master's-level research methods subject in project management. Author, institution and year not stated in the supplied files.
Suggested questions for Ask KEVOS
What is the difference between selection bias and group bias?
How do I control bias when I cannot randomise anything?
Why is a statistic biased rather than a result?
What should I balance my groups on if I can only balance on one thing?
What claims do I lose without randomisation and blinding?
