2-2 Comparison groups should be similar

If people in the treatment comparison groups differ in ways other than the treatments being compared, the apparent effects of the treatments might reflect those differences rather than actual treatment effects. Differences in the characteristics of the people in the comparison groups might result in estimates of treatment effects that appear either larger or smaller than they actually are. A method such as allocating people to different treatments by assigning them random numbers (the equivalent of flipping a coin) is the best way to ensure that the groups being compared are similar in terms of both measured and unmeasured characteristics.

Be cautious about relying on the results of non-randomized treatment comparisons (for example, if the people being compared chose which treatment they received). Be particularly cautious when you cannot be confident that the characteristics of the comparison groups were similar. If people were not randomly allocated to treatment comparison groups, ask if there were important differences between the groups that might have resulted in the estimates of treatment effects appearing either larger or smaller than they actually are.

Browse by Key Concept

Back to Learning Resources home

Filter these resources:

Clear Filters

Know Your Chances

This book has been shown in two randomized trials to improve peoples' understanding of risk in the context of health care choices.

| 0 Comments | Evaluated

Sunn Skepsis

Denne portalen er ment å gi deg som pasient råd om kvalitetskriterier for helseinformasjon og tilgang til forskningsbasert informasjon.


How Science Works

Definitions of terms that students have to know for 'How Science Works' and associated coursework, ISAs, etc


CEBM – Study Designs

A short article explaining the relative strengths and weaknesses of different types of study design for assessing treatment effects.



In this 5 min audio resource, Neeraj Bhala discusses systematic reviews and the impact of mega-trials.


Psychiatric disorders

Regrettably, research is not always well done or relevant. Take the example of a distressing condition known as tardive dyskinesia. […]


In an ideal world

‘In an ideal world, wherever possible, we could be gathering anonymised outcome data and comparing this against medication history, making […]


Comparing like with like

In this sub-section Comparisons are key (this page) Treatments with dramatic effects Treatments with moderate but important effects Comparisons are […]


No Resources Found

Try clearing your filters or selecting different ones.