Menu
Log in


SA Branch October meeting 2020: Instrumental variables: a review of the rationale for their use, assumptions and validation methods with application to estimating the impact of health spending on health outcomes

26 Nov 2020 8:30 AM | Marie-Louise Rankin (Administrator)

Laura Edney, Research Fellow in Health Economics at Flinders University, spoke to our October meeting on her own work about the motivation for the use of Instrumental variables (IVs), the assumptions they make and how these can be appropriately tested either directly or through sensitivity analyses by examining the impact of assumption violation. This is the topic of her current research, following on from many years of work on various fields on health economics.

At the beginning of her talk, Laura described the IVs and their application to make causal conclusions from observational data when randomized controlled trials are not feasible. IV methods have been widely used to estimate the impact that spending on health has on health outcomes due, in part, to historic health outcomes representing an important unmeasured confounder that, if unaccounted for, will result in biased coefficients on spending in standard regression models. She addressed few examples of IVs and illustrated to determine variation that is exogenous in treatment and to estimate causal inferences. In particular, smoking and health outcome relationship, an IV cigarette price is not logically directly related to health. The only logical relation is an indirect one: price affects cigarette use that, in turn, affects health.

To illustrate the key concepts of IVs, Laura reviewed the application of IVs to estimating the impact of health spending on health outcomes focusing on eight publications that have estimated this relationship nationally. Using two-stage least squares estimation (2SLS), health spending has a significant impact on health outcomes in presence of IVs. IV quantile regression, an extension of IV approach, was also discussed about estimating across the full distribution compared to unweighted IV 2SLS estimates of the mean effect.

In conclusion, Laura mentioned two key questions to consider: (i) Is the instrument meaningfully related to the predictor variable? (ii) Does the instrument directly or indirectly

influence the outcome variable? Merits and demerits of IVs were also highlighted at the end of the presentation. The meeting was held in virtual platform Zoom video communication with almost 25 attendees in the meeting.

By Shahid Ullah

Powered by Wild Apricot Membership Software