Software – Dean Eckles

PlanOut – A framework for online experiments (A/B tests etc.)

Multiway bootstrap – R code for data with multiway dependencies

icsw: Inverse Compliance Score Weighting – R package for instrumental variables estimation

There are some additional pedagogical notebooks and code on Github.

Some analysis code and data is archived in the Facebook repository at Harvard Dataverse.

Algorithmic transparency and assessing effects of algorithmic ranking.
Testimony before the US Senate Subcommittee on Communications, Media, and Broadband.
This addresses the questions: How does algorithmic ranking in social media work? How can we assess algorithmic impact & amplification? What can policy-makers do here?

Providing normative information increases intentions to accept a COVID-19 vaccine. Nature Communications.
When does sharing information about others’ behaviors cause more of that same behavior? This is both a central question in the social sciences and also important for effecting preventative health behaviors, including vaccination. We use a randomized experiment embedded in the international survey on COVID-19 we conducted with Meta and JHU to address this, providing evidence that these descriptive norms increase intentions to get vaccinated.

Bias and high-dimensional adjustment in observational studies of peer effects. JASA.
Are observational studies informative about social contagion? We use a massive randomized experiment in Facebook News Feed to identify peer effects in link sharing and compare these results to those using only observational data. The richest models are able to recover the experimental “gold standard” peer effect, at least for the most popular classes of behavior.

Estimating peer effects in networks with peer encouragement designs. PNAS.
We employ a novel experimental design to modulate how many likes and comments are received by 49 million Facebook users, allowing us to estimate network effects in adoption and use of communication technologies.