Does online availability boost citations? Using a panel of citations to economics and business journals, we show that the enormous effects found in previous studies were an artifact of their failure to control for article quality, disappearing once fixed effects are added as controls. The absence of aggregate effects masks heterogeneity across platforms: JSTOR has a uniquely large effect, boosting citations around 10%. We examine other sources of heterogeneity, including whether JSTOR disproportionately increases cites from developing countries or to “long-tail” articles. Our theoretical analysis informs the econometric specification and allows citation increases to be translated into welfare terms.

Preventives are sold ex ante, before disease status is realized, while treatments are sold ex post. Even if the mean of the ex ante distribution of consumer values is the same as that ex post, the shape of the distributions may differ, generating a difference between the surplus each product can extract. If, for example, consumers differ only in ex ante disease risk, then a monopolist would have more difficulty extracting surplus with a preventive than with a treatment because treatment consumers, having contracted the disease, no longer differ in disease risk. We show that the ratio of preventive to treatment producer surplus can be arbitrarily small, in particular when the distribution of consumer values has a Zipf shape and the disease is rare. The firm’s bias toward treatments can be reversed, for example, if the source of private information is disease severity learned ex post. The difference between the producer surplus earned from the products can result in distorted R&D incentives; the deadweight loss from this distortion can be as large as the entire producer-surplus difference. Calibrations for HIV and heart attacks based on risk factors in the U.S. population suggest that the distribution of disease risk is sufficiently Zipf-similar to generate substantial differences between producer surplus from preventives and treatments. Empirically, we find that proxies for the Zipf-similarity of the disease-risk distribution are associated a significantly lower likelihood of vaccine development but not drug development.