The hypothesized relationships are investigated using cross-sectional data from the medical instruments and supplies industry (SIC 384). A list of firms that conducted product development in the industry was obtained from the FDC Gray Sheet reports. Using the Company Profiles Database we were able to find the addresses of 308 firms that comprised the sample for this study. The base model tested is a moderated regression model specified as follows:
N = a + b1.C + b2.D + e
where
b1 = b11 + b12.A + b13.S
and
b2 = b21 + b22.A + b23.S
N measures the efficacy of new product development, C measures the concentration of firms in the same industry for the geographical region in which the firm is located, D measures the diversity of the industrial base in the geographical region in which the firm is located, S measures the total economic activity in the geographical region, and A measures firm absorptive capacity.
Before marketing a new product, firms in the medical equipment and supplies industry must file either a 510K or a PMA form with the Federal Drug Administration. This reporting requirement means that the innovative output of firms in this industry in unusually clear. We measured the success of a firm’s new product development efforts by the total number of 510k forms filed with the FDA. We collected this data for the years 1993 and 1994.
In order to construct measures of concentration (C) and diversity (D), we first assigned each firm to a Metropolitan Statistical Area (MSA) based upon its zip code. Thus, we defined geographical regions or locales on the basis of Metropolitan Statistical Area. This was selected as our measure of geographic region for two main reasons. First, there is an economic rational behind the construction of an MSA; they have an economic metropolis (city) at their center. Second, MSA has been used as the primary unit of analysis in a number of other studies looking at research spillovers (e.g., Glaeser, et al. (1992), and Jaffe (1989)). We constructed measures of concentration, diversity, and total economic activity, for each MSA using the County Business Patterns (CBP) data base for 1992. Our measures are defined as follows:
Concentration in MSAi | = | Number of Employees in the Industry in MSAi |
Diversity in MSAi | = | Fraction of
the MSA’s employment accounted for by the five largest industries other than SIC 384. |
Total economic activity in MSAi | = | Total MSAi employment. |
The reader should note that the diversity measure is reverse scored. That is, the higher the score on this measure, the less diverse the industrial base of the MSA. These measures are similar to those used by Glaeser et. al. (1992).
The descriptive statistics for our sample are given in Table 1.
TABLE 1 |
|||||
Variable | N | Mean | Std. Dev. | Minimum | Maximum |
Total MSA Employment | 78 | 646465.9 | 698721.7 | 40104 | 3536963 |
MSA Employment for SIC 384 | 78 | 2250.52 | 2426.29 | 2 | 10924.6 |
Diversity measure per MSA (excluding SIC 384) | 78 | 0.167 | 0.039 | 0.108 | 0.316 |
Number of Products developed per Firm in 1993 | 308 | 1.194 | 1.633 | 0 | 12 |
Number of Products developed per Firm in 1994 | 308 | 0.539 | 1.227 | 0 | 9 |
The hypothesized relationships were tested using generalized linear regression with a Poisson model. A Poisson model was selected for a number of reasons. First, the dependent variable is a count measure over a period of time which is often a Poisson process. Second, recent work has modeled a similar response variable, the number of patents, as a Poisson process (e.g., Henderson & Cockburn, 1995). One of the complications of analyzing count data is the likely presence of overdispersion. Overdispersion arises when the variance of the response (in this case the number of products developed) exceeds the nominal variance -- in this case the nominal poisson variance (McCullagh & Nelder, 1989). Overdispersion results in biased estimates of standard errors for the coefficients. To correct for overdispersion in our data, the standard errors of the coefficients were scaled using an estimate of the overdispersion parameter. The net result of this correction is larger standard error estimates for each coefficient.
To evaluate the first three hypotheses we constructed measures of the average number of products developed per firm for each MSA and fitted a model of the form below. By analyzing these relationships at the "MSA" level and using average products developed per firm in an MSA we avoid the problem in our data of having a different number of firms in each MSA.
Model 1: A MSA level
analysis of knowledge spillovers on average
firm product development
Average number of products developed per firm in MSAi | = | b1 . MSAi industry concentration | + | b2 . Diversity in MSAi | + | b3. Total economic activity in MSAi | + | b4 . MSA_{i} industry concentration .Total economic activity in MSAi |
The second model is a firm level analysis of knowledge spillovers. This model tests hypotheses 4a and 4b. The primary interest here is to test for the presence of a moderating effect of absorptive capacity on the relationship between knowledge spillovers and the number of products developed. Interaction terms between absorptive capacity and diversity and between absorptive capacity and concentration are included. This leads to a model of the following form:
Model 2: A firm level
analysis of knowledge spillovers, absorptive capacity,
and, product development
Number of
products developed by firmj in MSAi |
= | b1 . MSAi industry concentration | + | b2 . Diversity in MSAi | + | b3 . Absorptive capacity for firmj | ||||
+ | b4 . MSAi industry concentration . Absorptive capacity for firmj |
+ | b5 . Diversity in MSAi Absorptive capacity for firmj |
The final model investigates how firm size affects the relationship between knowledge spillovers, absorptive capacity, and product development. To investigate this effect we split our sample into three sub-samples based on firm size. The size classes were organizations with up to 20 employees, organizations with between 20 and 100 employees, and organizations with over 100 employees. These classes were suggested by Smilor (1995). Model 2 was re-estimated for each sub-sample.
Previous Page | Main Menu | Next Page
© 1997 Babson College All Rights Reserved
Last Updated 1/15/97 by Geoff Goldman & Dennis Valencia
To sign-up for the Center for Entrepreneurial
Studies' publication lists,
please register with the Entrepreneurship
WebTeam.