Instrumental variables in epidemiological research: an assessment of the adoption rate and future trends.

In epidemiological research, the unavailability of possible confounders may lead to inconsistent estimates when a randomized experimental design is not feasible. A related challenge can be possible reverse causal orders between the dependent and the independent variables. This is a so-called endogeneity problem, which is well known in econometrics. However, if possible confounders are unavailable, or we suspect reverse causal orders, the appropriate use of instrumental variables (IVs) can generate consistent estimates. IVs have the properties that they are correlated with the explanatory variables and uncorrelated with the error term. Here I map and compare the use of IVs in epidemiological and economics research over the last decades. My aim is to uncover what may cause the adoption of IVs in epidemiology. Is it related to previous adoptions in epidemiology or economics? If yes (or no), what may the implications be? Figure 1 illustrates the relative use of the term ‘instrumental variable*’ in the title, the abstract or the keywords on manuscripts indexed in the Social Science Citation Index within the subject areas economics and epidemiology (i.e. public, environmental and occupational health). I deleted Economics and Human Biology from the analyses since the journal is indexed within both subject areas. We observe an increasing trend of published items applying the term in economics, reaching a relative frequency of 1.05% in 2010. In epidemiology, the term first appeared in 1988, and at least since the mid-1990s we may observe a modest increasing trend. To take account of the parameters’ trends, I model yearly change in the use of IVs in epidemiology, yt, as yt yt 1 (t is 1 year). I follow a similar procedure to model yearly change in economics, xt. I apply yt as dependent variable. Unreported analyses show that both yt 1 and yt 2 are partially correlated with yt, and I apply them as independent variables. Further unreported analyses show that among the parameters xt, xt 1, xt 2, xt 3 and xt 4, only xt 2 has an effect on yt. Table 1 reports the results of yt regressed on yt 1, yt 2 and xt 2. The negative results of yt 1 and yt 2 indicate that the adoption rate of IVs in epidemiology varies around its trend within a time window of 2 years. Said differently, previous adoptions in epidemiology have no effect on later adoptions. The effect of xt 2 is positive, which shows that a positive change in economics of a given year tends to have a positive effect on the change in epidemiology 2 years later. The variance inflation factors (VIFs) are low indicating no multi-collinearity problems. In an unreported model, I control for year trend, but this effect is low (Std. b1⁄4 0.206; P1⁄4 0.145) and the other parameters are substantially unchanged. One possible explanation of xt 2’s positive effect can be that methodological developments take place in economics, which are applied in epidemiology