Consideration of different interpretations of results from fixed effects and random effects analyses brings the issue of generalisation to the fore. Results are often generalised from the situation in which they were sampled to other situations. However, strictly speaking, results should only be generalised when the study sample has been taken at random from the whole population of interest. Since centres are rarely sampled at random, even the global results from a multi-centre trial cannot be formally generalised to the population of possible centres.
There are analogies, though, with a single-centre study. In such studies, patients are not usually selected at random from the potential population of patients available at the centre. However, results are usually seen as some indication of those expected in the future both in the same centre and elsewhere.
Thus, in practice, generalisation needs to be by degree and will always to some extent involve subjective judgements, e.g. of how well the centres (or patients) sampled represent their full potential populations. Multi-centre studies would usually be considered more generalisable than single-centre studies even though the centres are not randomly sampled, and even if a fixed effects model is employed.
Was this article helpful?