# Research

## ** **

# Work in progress

**“Randomized Controlled Trials for models with dyadic outcomes”** (with Bryan Graham and Michael Jansson) **[Abstract]** * We study experimental designs for contexts where the outcome variable is double indexed and dyadically dependent. Whe show that it is enough to administer the treatment randomly across pairs, with a vanishing propensity score, to estimate the average treatment effect as precisely as if the propensity score were constant, or if the treatment were administered at the individual rather than the pair level. Allowing for vanishing propensity scores is important in at least two regards. First, when treatments are costly, that significantly reduces the costs of experimentations. Second, vanishing propensity scores attenuate equilibirum (or spillover) effects. ***“The k-composite likelihood estimator”**

**[Abstract]** *We extend the idea of the composite likelihood estimator on dyadic random graph models. The estimator proposed is obtained by maximizing the average likelihoods of all $k-$ node subgraphs of the entire $N- $ node graph. The properties of this estimator are studied for different regimes of $k$.*