Brief Note: Introducing the Language of Causal Analysis
Journal of Healthcare Engineering
The paper "A simple method for causal analysis of return on IT investment"  is welcome. The authors have taken an important step forward by introducing an explicit assessment of causation into their analysis. Traditionally, researchers have been timid in their approach to causality, tending to avoid causal language. This is misplaced. It is true that in the statistical/empirical sciences such as epidemiology, associations are all that we can observe, and that these are not necessarily
... But it is impossible to make sense of them without invoking causal relations -for example, a long-established rule of thumb in multivariate analysis has been to adjust for covariates that are not on the causal pathway of interest. Besides, in most cases, the importance of an associationscientifically and also practically -is mainly derived from what it tells us about a causal process. And yet, the practice has typically been for explicit causal thinking to occur not in the design of the study or the set-up of the analysis, but only afterwards, in assessing the findings; or at least, papers are written in this way, the causal inference being left until the Discussion section of a paper, where it is "smuggled in" , rather than being part of the Methods section. It is more appropriate to develop and use causal language in a rigorous fashion. The paper employs the strategy of looking at the three main variables of interest, noting that six time orders are logically possible but not all are plausible in the real world, and then discarding those that are implausible. First, the options where use of IT precedes IT investment are discarded, whereas those where IT use follows investment are retained. In the second stage, the authors also exclude the possibility in which IT use occurs after IT investment, but where this is mediated through revenue growth, i.e., IT investment → revenue growth → IT use. The paper uses substance-specific knowledge of causal relationships to make an a priori judgement as to whether a particular set of causal relationships should be a candidate hypothesis for investigation and estimation. Put like this, in the abstract, it sounds suspect to someone who is used to the mind-set that inferring relationships from data is the only legitimate procedure, and that any kind of a priori imposition of causal structure is unwarranted. Clearly this strategy is valid only to the extent that the a priori judgement is sound; it needs to be explicitly justified in terms of previous evidence and substantive knowledge, and/or (as here) by plausibility in the light of how the world works.