Leveraging Social Media in the Pharmaceutical Industry - Innovation and Marketing in the Pharmaceutical Industry - page 482

htXht X
i

(| ) = ( )exp( b )

i o i

where ht

o () is the baseline hazard function and β is a parameter vector. Researchers
have tested different forms of social contagion measures such as adoption, usage,
and volume of usage of the new drug by peer physicians.
Nair et al. ( 2010 ) use a linear response model to study the impact of nominated
opinion leaders' prescriptions on the nominating physicians' prescriptions of an
established drug. The linear model is in the form below:

y Dx
it

=+ + bd g

it ji t (), it

where y it and X ji t

(), denote the new prescriptions written by physician i and physi-
cian i i's opinion leaders at time t . D it denotes the number of details on physician i
at time t . Prescriptions of opinion leaders serve as a proxy for the opinions held by
the opinion leaders toward the prescribed drug. The parameter d measures the
influence of opinion leaders' prescriptions on physicians' prescribing decisions.
The linear model is a reduced form of the prescription behavior process. It serves as
an approximation of a nonlinear model and is appropriate for a wide range of situa-
tions (Hartmann et al. 2008 ).
There are significant methodological challenges for researchers to infer social
influence from physician prescription behavior. Correlation in physician prescrip-
tion behaviors can arise from three possible sources other than social contagion
effects: endogenous group formation, correlated unobservables, and simultaneity.
These sources pose significant challenges for detecting the existence and measure
the magnitudes of social contagion (Manski 1993 ; Nair et al. 2010 ).
Endogenous group formation often referred to as “homophily,” occurs when
people with similar “tastes” tend to come together. Their inherit similarities may
cause them to take similar actions independently even without social contagion.
Thus, such a phenomenon could impede the identification of true social contagion.
This identification problem can be overcome by including individual-specific fixed
effects when panel data is available (Aral et al. 2009 ). Analysis based on cross-
sectional data alone may lead to spurious contagion effects.
Simultaneity happens when product sales or diffusion changes the amount of
WOM while WOM also affects the level of sales or diffusion process. Moreover,
WOM is dynamic; it is an autocorrelated time series with carryover across time.
Researchers have used vector auto regression (VAR) models to capture the dynam-
ics and simultaneity aspects of WOM analysis (Rui et al. 2011 ; Trusov et al. 2009 ).
Social contagion can occur via verbal communications (WOM) and observational
learning. The concept of observational learning stems from social learning studies in
psychology (Bandura 1977 ). Chen et al. ( 2011 ) attempt to disentangle the impact of
WOM and observational learning on the sales of digital cameras using a field experiment.
An intriguing finding of their study is that while negative WOM is more influential than
is positive WOM, positive observational learning information significantly increases
sales, while negative observational learning information has no effect on sales.