Content area
Full text
Difference-in-difference (DiD) research designs have been widely used in health economics, health services research, and health policy evaluations to estimate the effect of policy changes or interventions. Careful implementation of DiD designs have the potential to strengthen the use of observational data in answering questions relevant to the nursing profession.
Estimating causal effects using observational data is often challenging. In contrast to experimental studies where treatments or interventions are allocated randomly, in observational studies the treatment or policy change of interest is not under the control of the researcher. Often, data were collected for other purposes, such as during the course of medical care in the case of electronic health records and medical claims data. Under the right circumstances, however, these data can answer questions about the impact of policies or interventions on a variety of outcomes in real-world settings. The main challenge of observational studies is to appropriately account for confounding factors or variables related to the non-random allocation of treatment, particularly when the factors that determine selection into treatment are not observed.
In a previous article, we discussed the challenges of causal inference and described research designs developed to estimate causal effects using observational data (Perraillon, Welton, & Jenkins, 2019). The objective of this article is to describe in more detail difference-in-difference (DiD) research designs, which are commonly used to estimate the effects of policy changes over time. DiD designs are an extension of before-after comparisons of outcomes that incorporate a suitable control group. Contrary to randomized experiments, DiD designs do not require a control group that is similar in observed and observed characteristics. However, to be valid, other assumptions must be met, some of which can be tested empirically while others require in-depth subject knowledge to ascertain their plausibility.
Two Periods and Two Groups
One of the oldest and well-known examples of a DiD design is Snow's (1854) study showing how cholera was transmitted by water rather than air. Snow exploited a change in water supply from one source of water to another, thought to be contaminated, in one London district. He first showed that after the change in supplier the death rate due to cholera increased significantly. Although intriguing, other contemporaneous factors could have explained the increase in death rates,...