Pharmaceutical Detailing Elasticities: A Meta-Analysis - Innovation and Marketing in the Pharmaceutical Industry - page 535

elasticities can vary due to differences in data measurement intervals. In the detailing
context, sales cycles are fairly short. In such settings, short-term temporal variations
in both selling effort and resulting sales can occur in part due to the prevalent use of
nonlinear incentive mechanisms with periodic (monthly or quarterly) incentive pay-
outs (e.g., Zoltners et al. 2006 , p. 222; Mantrala et al. 1994 ). Similar to Tellis ( 1988 ),
we expected that short-term temporal variation will not be picked up in annual data
as much as by shorter interval data.
We dummy coded two variables to capture temporal aggregation. We coded one
variable to be 1 when the temporal aggregation was quarterly and 0 otherwise, and
a second dummy variable to be 1 when the temporal aggregation was yearly and 0
otherwise. We found a main effect for the quarterly dummy as reported in Table 18.3
( b = 0.373, p < 0.01). Also, we found a weak signifi cant two-way interaction between
the quarterly dummy variable and the geographic setting variable ( b = −0.344,
p < 0.10). Since we found other interaction effects, we calculated the difference in
detailing elasticities between monthly and quarterly data, holding all other modera-
tors at their mean value. We fi nd that, holding all moderators at their average level,
quarterly detailing elasticities are lower than monthly detailing elasticities by 0 . 34 .
Also, we found no signifi cant difference between elasticities estimated with yearly
data and monthly data ( b = 0.071, ns).
The next six fi ndings are with respect to researcher ' s model specifi cation choices .
Inclusion or not of lagged output effects and lagged input effects . Detailing effort
has signifi cant carryover effects. For example, based on sales force studies at 50
pharmaceutical companies, Sinha and Zoltners ( 2001 ) report that the aggregate
sales carryover from selling effort in 1 year is 75, 80 % the next year, 62-78 % the
year after, and 52-70 % in the fourth year. Thus, the effectiveness of current-period
detailing would be overstated if lagged leads (lagged output effects) or past effort
(lagged input effects) are omitted.
Accordingly, we employ two dummy variables for capturing the inclusion of
lagged input and output effects in a past study's model specifi cation: the fi rst of
these has a value 1 if lagged input was included, 0 otherwise; the second has a value
1 if lagged output was included, 0 otherwise. As can be seen in Table 18.3 , we found
that detailing elasticities from response models that include lagged input and/or
lagged output effects were smaller than those from response models that exclude
these effects by 0.155 ( p < 0.05) and 0.118 ( p < 0.05) respectively.
Accounting for endogeneity of detailing effort . Endogeneity refers to a correlation
between the input variable and the error term of the estimated response model
which, for example, arises if management allocates sales effort strategically or uses
rules such as effort allocations proportional to past sales. Some researchers have
accounted for endogeneity in model estimation, e.g., via the use of instrumental
variables (e.g., Chintagunta and Desiraju 2005 ), while many have not. If an input
such as detailing is treated as exogenous when it is indeed endogenous, then both
theoretical analyses and empirical evidence show its elasticity may be overesti-
mated (e.g., Manchanda et al. 2004 , p. 472). Therefore, we expected that detailing
elasticities from models that account for endogeneity will be lower than those from
models that do not account for endogeneity.