99 years of economic insights for Indiana

The IBR is a publication of the Indiana Business Research Center at IU's Kelley School of Business.

Executive Editor, Carol O. Rogers
Managing Editor, Brittany L. Hotchkiss

Appendix

Return to main article: Knowledge Creation and Innovation in the Hoosier State

For the empirical investigation, an OLS regression of the output measure, number of patents per 1,000 workers, was run on the set of inputs listed in Table 1. Table 1 provides the complete estimation results. The three models represent different university knowledge spillover scores calculated for three distance radiuses: 50, 100 and 250 miles. All other inputs remain the same across models. Four counties (Kosciusko, Lake, Martin and Porter) are excluded from the analysis, as they are considered to be outliers, reducing the effective sample size to 88.

The numbers presented in the table are the marginal effects on patents from increasing one unit of the corresponding input. For a binary input (taking values 0 or 1), the number represents what impact having the input is on patents vs. not having it. Robust standard errors are in parentheses. The adjusted R2 indicates that roughly 25-28 percent of the variation in patents can be explained by the variation of these inputs. An input whose marginal effect is marked with an asterisk suggests that the impact from that input on patents is statistically significant (more asterisks indicate a smaller p-value and higher statistical significance).

Opposing signs in binary (negative) and level (positive) measures indicate that patents are mainly driven by the level, while small/low levels embrace no advantage over not having the input at all.

Take Model KSPL-50, for example. Counties that have STEM programs on average generate 1.4 fewer patents per 10,000 workers, compared to counties that do not. On the other hand, among counties that have STEM programs, an increase of 1 percent in the share of STEM graduates leads to an increase of 2 patents per 10,000 workers. The net increase is thus 0.6 patents per 10,000 workers (or 6 patents per 100,000 workers). The larger the share of STEM graduates, the more magnified its positive impact on patents. If, however, the share of STEM graduates at a county is negligible, it bears no advantage over counties that do not have STEM programs at all. (Figure 8 in the article gives a visual presentation of this concept.)

In order to see which matters more, proximity or occurrence, the actual dollar amount of university R&D expenditure at the county level is also included as an input. The final model, however, only includes the binary measure of having R&D expenditure, because the dollar amount is highly correlated with other inputs, such as the share of college degrees, and thus causes multicollinearity, a statistical problem that makes the estimates unreliable.

The effect of having university R&D expenditure on patents is negative, which again is outweighed by the share of college degrees and STEM graduates at a county, both of which have positive correlation with R&D expenditure and have positive impacts on patents.

Table 1: The Regression Results of Inputs on Patents (Full Model)

INPUTS KSPL-50 KSPL-100 KSPL-250
Knowledge spillover
(50-mile cutoff)
0.00277+
(0.00174)
Knowledge spillover
(100-mile cutoff)
0.000267
(0.000378)
Knowledge spillover
(250-mile cutoff)
0.000197
(0.000346)
Has university R&D spending
(0 or 1)
-0.115*
(0.0670)
-0.0843
(0.0604)
-0.0829
(0.0608)
Has STEM programs
(0 or 1)
-0.136*
(0.0790)
-0.110+
(0.0718)
-0.110+
(0.0712)
Population share of STEM graduates
(percent)
0.208**
(0.0974)
0.181**
(0.0900)
0.178*
(0.0905)
Population share of bachelor’s and above degrees
(percent)
0.0330***
(0.00820)
0.0311***
(0.00829)
0.0310***
(0.00827)
Employment share in high-tech industries
(percent)
-0.0141
(0.0111)
-0.0159
(0.0110)
-0.0163
(0.0114)
Employment share in technology occupations
(percent)
0.0366+
(0.0226)
0.0359+
(0.0228)
0.0348+
(0.0226)
Has large high-tech firms
(0 or 1)
-0.130
(0.0971)
-0.147+
(0.0979)
-0.145+
(0.0984)
Number of large high-tech firms per
1,000 workers
5.312***
(1.812)
5.592***
(1.817)
5.686***
(1.876)
Number of small high-tech firms per
1,000 workers
0.0457
(0.0826)
0.0527
(0.0905)
0.0506
(0.0895)
Small high-tech establishment
quotient
0.338*
(0.202)
0.288
(0.202)
0.289
(0.201)
Share of establishment birth
(percent)
0.00165
(0.0333)
0.0115
(0.0325)
0.0116
(0.0325)
Share of employment from establishment birth
(percent)
-0.00962
(0.0273)
-0.00992
(0.0266)
-0.00947
(0.0265)
Share of proprietorship
(percent)
-0.00754**
(0.00356)
-0.00654*
(0.00353)
-0.00652*
(0.00353)
Has VC investment
(0 or 1)
0.134*
(0.0795)
0.110
(0.0883)
0.106
(0.0889)
VC investment
per $1,000 GDP
-0.212
(0.170)
-0.140
(0.166)
-0.132
(0.172)
Log of population
density
-0.0772+
(0.0475)
-0.0752+
(0.0498)
-0.0765+
(0.0500)
Intercept -0.232
(0.296)
-0.208
(0.320)
-0.220
(0.324)
Observations 88 88 88
Adjusted R-squared 0.282 0.250 0.249

Significance Levels: *** p-value < 0.01, ** p-value < 0.05, * p-value < 0.1, and + p-value < 0.15
Source: Indiana Business Research Center