Abstract:
|
Encouragement designs have been used to analyze causal effects of a treatment, policy, or intervention on an outcome of interest when randomizing the treatment was considered impractical (e.g. the causal effect of maternal smoking on child's birthweight (Sexton and Hebel, 1984). In a similar vein, in network data or data with interference, randomizing a treatment to study its causal effect on an outcome may be difficult.
The talk introduces some examples in both large-scale social network data as well as small-scale network data where randomizing treatment may be infeasible. Inspired by these examples, we propose a new experimental design called peer encouragement design that attempts to analyze the direct and spillover effects of a treatment on an outcome in these settings. We provide identification results for both average direct and spillover effects under the design and we explore the application of the new design in real data.
|