Inference 1: chance, bias, and confounding

Robert Stewart

in Practical Psychiatric Epidemiology

Published on behalf of Oxford University Press

Published in print August 2003 | ISBN: 9780198515517
Published online March 2013 | e-ISBN: 9780191754289 | DOI:
Inference 1: chance, bias, and confounding

More Like This

Show all results sharing these subjects:

  • Psychiatry
  • Public Health and Epidemiology
  • Epidemiology


Show Summary Details


A fundamental process in interpreting one's own or another's research is to consider what the observations ‘mean’, that is, what can be inferred from them. This involves a series of questions and considerations (Fig. 12.1). The first step theoretically is to decide whether observations can be believed in the first place. It is the duty of the guarantor for any submitted research paper to ensure that the data and results derived from analyses are valid—and to withdraw a submission promptly if there are any concerns over this. Readers have no choice but to assume the integrity of the raw data. The first formal stage of appraisal is to decide to what extent the observations in the sample are likely to apply to the source population. The principal considerations here are chance and bias. If the reader is happy that a population-level association is likely, the next stage is to consider what can be inferred concerning cause and effect. This begins with considering whether the association between proposed exposure and outcome is a direct one and not confounded by other factors. After this point, there are a series of more complex decisions regarding causal pathways, which may not always be inferred from a single study but may require a more broad knowledge of the background literature. These are discussed in Chapter 13 and include the direction of causality (whether an association between A and B is because A causes B or vice versa), mediating factors (does A cause C because A causes B which in turn causes C, or are there other pathways by which A and C are related?), and effect modification (does A cause B to a uniform extent across the population or does the strength of association depend on other factors being present?). Finally, the implications of these inferences need to be considered with respect to developing new hypotheses and investigations, as well as for clinical practice and Public Health.

Chapter.  6655 words.  Illustrated.

Subjects: Psychiatry ; Public Health and Epidemiology ; Epidemiology

Full text: subscription required

How to subscribe Recommend to my Librarian

Buy this work at Oxford University Press »

Users without a subscription are not able to see the full content. Please, subscribe or login to access all content.