Leveraging Peer-to-Peer Networks in Pharmaceutical Marketing - Innovation and Marketing in the Pharmaceutical Industry - page 456

A key issue in identifying peer effects is that of defining peers. Manchanda et al.
( 2008 ) use physical distance between physicians to identify physician networks.
Trusov et al. ( 2010 ) use observed “friends” lists on online social networking web-
sites to define the network and the activity (log-on data) to estimate the peer effect.
Other research uses surveys and/or experiments to identify the opinion leaders;
examples include Valente et al. ( 2003 ), Lomas et al. ( 1991 ), Celentano et al. ( 2000 ),
Dufflo and Saez ( 2003 ), Nair et al. ( 2010 ), and Iyengar et al. ( 2011 ). Bhatia and
Wang ( 2011 ) use patient movements between physicians to identify the physician
network. Wuyts et al. ( 2010 ) provide a summary of the data used in the literature to
study customer networks.

15.3 Models of Physician Social Networks

15.3.1 Identifying Physician Networks and Opinion Leaders

Various ways have been used in marketing literature to identify opinion leaders in
industry as well as in academia using surveys (Nair et al. 2010 ; Iyengar et al. 2011 ),
distance between physician office locations (Manchanda et al. 2008 ), and patient
movements between physicians (Bhatia and Wang 2011 ). The clinical leaders can
be identified by surveying physicians (Nair et al. 2010 ; Iyengar et al. 2011 ). The
market leaders can be identified by the referral patterns of physicians (Bhatia and
Wang 2011 ).
There are three main difficulties in identifying the effect of opinion leaders on
other physicians as pointed out in, which provides a recent and broad summary of the
various models of social interactions. The three difficulties are endogenous group
formation, correlated observables, and simultaneity. The endogenous group formation
problem arises because physicians with similar tastes may tend to form social groups;
hence, subsequent correlation in their behavior may reflect these common tastes, and
not a causal effect of one's behavior on another. General practitioner physicians tend
to meet and form their relationships with specialist physicians at conferences, some
hosted by drug companies, which are organized around specific disease conditions
and therapeutic treatment options. These relationships between physicians may have
correlated prescriptions due to similar tastes rather than due to the opinion leader
effect. One solution to the endogeneity of group formation is facilitated by the avail-
ability of panel data. With panel data, one can control for endogenous group forma-
tion via agent fixed effects (e.g., Nair et al. 2010 ).
A second source of correlation is correlated unobservables that drive the actions of
all physicians in a reference group similarly. There may be common location and time
period specific effects which affect all physicians in the group. For example, if some
geographical region such as Florida or Arizona has a higher percentage of elderly
patients with higher incidence of hypertension or high cholesterol, or if a certain
region has a higher ethnic concentration, the physicians may be prescribing more of