Pharmaceutical Detailing Elasticities: A Meta-Analysis - Innovation and Marketing in the Pharmaceutical Industry - page 536

We use a dummy coded variable to capture the inclusion or not of endogeneity in
response models (1 if included, 0 if not) and investigated its effects in our meta-
analytic model. The results shown in Table 18.3 indicate that while the main effect
is not signifi cant ( b = 0.047, ns), the effect of the interaction between stage of PLC
and the inclusion of endogeneity was negative and signifi cant ( b = −0.193, p < 0.10).
Since we found other interaction effects, we calculated the difference in detailing
elasticities between models which account for endogeneity vs. those that do not. We
fi nd that, holding all moderators at their average level, models that account for
endogeneity report detailing elasticities to be lower by 0.045. That is, when endoge-
neity is accounted for in response models involving products in the early stage of
the PLC, detailing elasticities tend to be signifi cantly lower.

18.4.1 Accounting for Heterogeneity in Response to Detailing

As several studies have shown over the last decade, physician response to detailing
is likely to be heterogeneous due to both observable and unobservable factors (e.g.,
Manchanda and Chintagunta 2004 ; Skiera and Albers 1998 ). However, a number of
past studies do not allow for heterogeneity in detailing response parameter esti-
mates. To assess the effect of not accounting for heterogeneity in response among
past studies in our database, we dummy coded a variable for the inclusion of hetero-
geneity in response to detailing effort in the model (1 if included, 0 if not) and
incorporated this in our meta-analysis. The result shown in Table 18.2 indicates no
signifi cant differences in elasticity parameter estimates from models that account
for heterogeneity in sales response vs. those that do not ( b = −0.062, ns).

18.4.2 Variables Whose Omission Could Bias Detailing

Elasticity Estimates

If the response model specifi cation in a particular study setting omits a relevant and
plausible infl uencer of sales (e.g., advertising, promotions) then the resultant personal
selling elasticity estimate can be biased, although any predictions with respect to the
direction of the bias may be unfounded (e.g., Clarke 2005 ). Therefore, we examined
if there are any systematic biases in detailing elasticity observations due to omissions
of promotions or advertising (either journal or DTC) from the original models.
Accordingly, we dummy coded two variables for the inclusion of promotions
(1 if included, 0 if not) and advertising (1 if included, 0 if not) in the response
model. For promotions, the main effect in Table 18.2 shows no signifi cant differ-
ence in detailing elasticities for models that include promotions in the response
model vs. those that do not ( b = 0.097, ns). However, we fi nd a signifi cant interac-
tion between the promotions variable and the stage of the PLC ( b = 0.315, p < 0.01).
Since we found other interaction effects, we calculated the difference in detailing