How Strong is Gravity? Separating Home Bias from Transport Costs Using Hospital Choice

In health care markets, patients typically do not travel very far to visit their provider. For example, Gowrisankaran et al. (2015) find that a five minute increase in travel time to a hospital reduces demand between 17 and 41 percent.

Economists typically assume that distance reflects transport costs, implying a large disutility for travel, but an alternative explanation for the role of distance is that unobserved consumer preferences for hospitals are correlated with distance. We refer to this correlation as “home bias”, because it implies that consumers prefer goods or services produced closer to home. Because, in most settings, it is difficult to separate the effect of transport costs from home bias, researchers typically interpret a distance elasticity as a transport cost elasticity.

In this paper, we examine a unique environment in which we can separate transport cost from home bias, patients' choice of hospital for childbirth. Childbirth is a good context to study this question because the services that the patient needs are similar across multiple childbirths, but patients do not typically need to return to the same hospital where they previously received care. Our empirical strategy, a panel data fixed effects approach from Chamberlain (1980), uses variation from women changing residence and hospitals between births. Thus, patients' distance to hospital providers changes over time independently from persistent unobserved hospital preferences, allowing us to control for such preferences.

We apply this estimation procedure to data on inpatient childbirths in Florida between 2006 and 2014, and compare it to a discrete choice model using individual patient choices that does not control for individual patient-hospital interactions.

The transport cost elasticity derived from the fixed effects estimator is about half those derived from the standard logit approach. The transport cost elasticity is -1.71 from the standard discrete choice framework for hospitals with a minimal share of the market. However, for women that switch hospitals, distance explains much less of the variation in the conditional probability of the sequence of their choices than implied by estimates from a standard discrete choice model. Thus, the estimate of the elasticity using the fixed effect estimator is substantially lower at -0.83. Without controlling for home bias, one will substantially overestimate the transport cost elasticity.

After demonstrating the importance of home bias, we explore three potential explanations for it: switching costs, catering to local demand, and referral patterns. Our estimates of the transport cost elasticity using a standard logit estimator do not differ between first and second births, suggesting that switching costs do not play an important role. Our estimates do not change after conditioning on multiple ways in which a hospital could cater to local demand. Instead, we find that referral patterns from peers and/or clinicians are likely an important determinant of home bias.

We then show how our structural estimates of the distance cost elasticity lead to different conclusions on antitrust market definition for hospital mergers, and for the demand and welfare effects of a planned hospital move.