by A practice-based account of causal bias No estimated causal results are thus affected solely by the intervention but by many other background attributes and conditions that can give rise to bias between, within or across trial groups. A number of these influence a treatment’s estimated causal effects both within and outside a trial setting. That these and other such demanding preconditions (concauses) would be entirely satisfied for all participants is a foundational assumption implicit in the epistemic practice of randomised experimentation. In the real world, we are however not able to make sure that such concauses are present and evenly distributed among trial groups, since they are at times either known but we cannot easily collect data on them or
Topics:
Lars Pålsson Syll considers the following as important: Statistics & Econometrics
This could be interesting, too:
Lars Pålsson Syll writes The importance of ‘causal spread’
Lars Pålsson Syll writes Applied econometrics — a messy business
Lars Pålsson Syll writes Feynman’s trick (student stuff)
Lars Pålsson Syll writes Difference in Differences (student stuff)
A practice-based account of causal bias
No estimated causal results are thus affected solely by the intervention but by many other background attributes and conditions that can give rise to bias between, within or across trial groups. A number of these influence a treatment’s estimated causal effects both within and outside a trial setting. That these and other such demanding preconditions (concauses) would be entirely satisfied for all participants is a foundational assumption implicit in the epistemic practice of randomised experimentation. In the real world, we are however not able to make sure that such concauses are present and evenly distributed among trial groups, since they are at times either known but we cannot easily collect data on them or may be unknown … Causation in practice, and the logic of experimental reasoning, need to thus be viewed more broadly than the particular disease or problem and its treatment.
Outliers and heterogeneity of treatment effects always exist because, for instance in medical trials, people are of different age, gender and physical health, experience different conditions, treatment needs and responses, develop different levels of resistance to the treatment etc. They are thus not a resolvable statistical or epistemic problem. They are a common consequence of studying dynamic biological, behavioural and social phenomena that involve complex processes. Such dynamic phenomena like diseases and economic policies are continually changing – their scope, their intensity, their duration etc. – and are better understood as (what I call) evolving causes rather than precisely measurable, static causes amenable to statistical analysis. This can contribute to a further degree of uncertainty in the level of accuracy of a trial’s causal estimates across the entire sample. Issues related to complexity, heterogeneity and evolving causes are however often not directly assessed as trials are designed specifically to estimate the average causal effect among the distribution.