Workforce education and corporate innovation

Abstract

The average workforce education near firms’ research centers facilitates firms’ matching with innovation talents and acquisition of knowledge. This study documents a positive association between the average education level in the metropolitan statistical areas (MSAs) where firms’ research centers are located and the quantity and quality of innovation outputs. The results are confirmed by controlling for various measures of MSA-level economic, population, and employment conditions, as well as research-center level analyses which control for firm-year fixed effects. We further find that local workforce education is more important for firms that are large, less labor-intensive, in non-high-tech industries and located in low education regions. The evidence highlights the importance of having access to well-educated local workforce for corporate innovation.

JEL Classification: G30, J24, O31

Keywords

Corporate innovation human capital patents and citations rank-and-file employees workforce education

1. Introduction

Innovation is crucial to firms’ competitive advantage (Porter, 1992) and aggregate economic growth (Solow, 1957). The view that workers play an essential role in corporate innovation is widely shared among the most innovative firms. For instance, Google’s first core principle of innovation is that “innovation comes from anywhere,” which argues that “innovation is ... everyone’s responsibility. Ideas come from anyone from the very top of the organization to lower ranks and it’s everyone’s responsibility to innovate.”¹ While prior literature recognizes the role of human capital in firm innovation (Chang et al., 2015; Chen et al., 2016; Mao and Weathers, 2019), we examine whether and how the supply of educated workforce affects the quantity and quality of firm’s innovation outputs.

Education represents formal training for individuals to build knowledge base, skill, self-motivation, and capacity to learn, absorb, and explore new knowledge. Prior studies suggest that a firm’s innovation process and performance are dependent on its knowledge base (Laursen and Salter, 2006; Miller et al., 2007; Nieves et al., 2016). Better educated employees have a richer pre-existing knowledge and are better at assimilating and transforming new knowledge (Knudsen and Schleimer, 2020; McDonough, 1993; Minbaeva et al., 2003; Un, 2017; Vega-Jurado et al., 2008; Vinding, 2006). Thus, educated workforce represents the most important input to corporate innovation, and the supply of such resource affects firms’ ability to conduct innovation.

We examine the association between corporate innovation activities and the average educational attainment in the metropolitan statistical areas (MSAs) where a firm conducts innovation (Beck et al., 2018; Call et al., 2017). We focus on the location-based average workforce education for two reasons. First, we view the supply of educated labor in close proximity to a firm as a scarce input to corporate innovation. Higher average level of educated labor provides firms with better chances of employing the required talents. Second, corporate innovation is multi-dimensional and requires input from employees with diverse skills in technologies, management, marketing, industrial design, and so on. A broad-based measure of workforce education is required to reflect the supply of talents in various occupations.

We predict a positive relationship between local workforce education and firms’ innovation outputs. This hypothesis is based on the agglomeration theory of innovation (Carlino and Kerr, 2015), which links the agglomeration of resources with innovation through three mechanisms, matching, knowledge spillovers, and input sharing. Specifically, matching refers to firms’ ability to match their innovation needs with local talent pool. Knowledge spillover theory suggests that agglomeration promotes the spread and adoption of knowledge. Furthermore, firms can benefit from sharing locally concentrated inputs to production and innovation. We argue that a higher level of local workforce education enables firms to find better matches in local labor markets and benefit from knowledge spillovers in that region.² Thus, workforce education positively contributes to innovation activities of firms located in the region.

Following Hombert and Matray (2018) and Matray (2021), we identify locations of firms’ research centers using inventors’ addresses. These research laboratory locations are then matched to the MSA-level workforce education information. We start by showing a significantly positive relation between the average workforce education level in the MSAs of a firm’s research centers and forward patent counts and average citation counts for a sample period of 1980–2008.³ The results are robust to a battery of sensitivity tests including (1) using originality, generality, the market value of patents, new product introductions, and non-self-citations as alternative measures of innovation quality; (2) using the educational attainment in the MSA where the firm is headquartered; (3) controlling for employee treatment; (4) controlling for managerial and employee incentives; and (5) controlling for measures of urban agglomeration (i.e. the concentration of urban employment and industry).

Because firms do not randomly select the location of research centers, we conduct analyses at the research center level and control for firm-year fixed effects to address this firm-year-level selection issue. This design should account for firm-year-specific characteristics, including the selection of research centers at the firm-year level. We find evidence suggesting that within the same firm, research centers exposed to better supply of educated workforce produce more innovation outputs and higher quality innovation. While our tests reduce endogeneity concern, we note that we cannot fully eliminate the selection bias associated with the selection of individual research centers.

Next, we study how the association between local workforce education and corporate innovation varies among firms. We find that the relation is more pronounced for firms that are larger, have fewer employees, in non-high-tech industries and located in areas with below-college average education level. These results show that workforce education is more beneficial to the corporate innovation activities of these firms.

This research extends our knowledge on the role of human capital in corporate innovation. First, while prior literature has documented the effect of the skill level of management on innovation (Barker and Mueller, 2002; Chemmanur et al., 2019; Islam and Zein, 2020; Israelsen and Yonker, 2017), this study suggests that local supply of educated workers affects many aspects of corporate innovation. In a related research, Chen et al. (2020) show that large migration of low-skilled workers from rural areas to cities decreases corporate innovation. This effect can be due to low-skilled workforce and/or an expansion of local labor supply. We complement Chen et al. (2020) to provide direct evidence that workforce education level is associated with corporate innovation, after controlling other location-based labor characteristics.

Second, prior studies find that corporate innovation is affected by inventor fixed effects (Liu et al., 2017), employee satisfaction and treatment (Chen et al., 2016; Mao and Weathers, 2019; Ramdani et al., 2021), compensation incentives (Chang et al., 2015), and labor laws (Acharya et al., 2013). This study adds to prior studies by focusing on the education level of local workforce. The supply of educated labor has a first-order effect on innovation, which is dependent on skills and knowledge.

Third, prior studies show the importance of education for individual inventors (Aghion et al., 2018; Toivanen and Väänänen, 2016) and regional innovation activities (Andersson et al., 2005; Kantor and Whalley, 2014). We extend these results to the firm level since individual inventors rarely innovate on their own and corporate innovation requires contributions from rank-and-file employees. In addition, we examine the cross-sectional variations in the relation between workforce education and corporate innovation to identify firm and regional characteristics that affect the education–innovation relationship. Our findings have implications for firms’ matching with local talents and selection of research center locations.

Finally, this research contributes to the literature studying the effect of geographic location on firm performance and decision making (Garcia and Norli, 2012). Our results indicate that innovation is a human-capital-intensive activity that relies on the educational attainment in the relevant geographic location to supply talented employees. This evidence extends Beck et al. (2018) and Call et al. (2017) who document the important implications of workforce education for financial reporting quality and audit quality. In contrast, we document a real effect of workforce education on corporate innovation.

The remainder of this article proceeds as follows. Section 2 develops the hypothesis. Section 3 describes the research design and summary statistics. Section 4 reports the baseline empirical results and the robustness tests. Section 5 explores the heterogeneity of our baseline results. Section 6 concludes the article.

2. Hypothesis development

We argue that local workforce education facilitates corporate innovation through the following non-mutually exclusive ways. First, better educated employees have a richer prevailing knowledge since they accumulate knowledge through years of formal education (Knudsen and Schleimer, 2020; McDonough, 1993). Knowledge is one of the main inputs in the innovation process and the relationship between knowledge and innovation has been well documented (Hargadon, 2002; Laursen and Salter, 2006; Leiponen, 2006; Miller et al., 2007; Nieves et al., 2016; Wolfe, 1994). Roper et al. (2008) argue that internal knowledge sourcing activities provide an important input to corporate innovation. Thus, highly educated workers can enrich the knowledge base of a firm, which is essential to the innovation process.

Second, highly educated workers are better at acquiring and transforming new knowledge (Knudsen and Schleimer, 2020; Minbaeva et al., 2003; Un, 2017; Vega-Jurado et al., 2008; Vinding, 2006). Prior studies suggest that a firm’ innovation process and performance are dependent on its absorptive capacity (Cohen and Levinthal, 1990; Knudsen and Schleimer, 2020; Zahra and George, 2002). Absorptive capacity is “the ability of a firm to recognize the value of new external information, assimilate it and apply it to commercial ends” (Cohen and Levinthal, 1990: 128). It consists of two components: acquisition and assimilation of knowledge and transformation and exploitation of knowledge (Zahra and George, 2002). Prior related knowledge facilitates the learning of new knowledge and the development of problem-solving and decision-making skills (Cohen and Levinthal, 1990; Criado-Perez et al., 2020). Therefore, employees’ educational level is a key aspect of firms’ absorptive capacity, which contributes to firms’ innovation outcomes.

Third, based on the agglomeration theory of innovation (Carlino and Kerr, 2015), local workforce education can facilitate corporate innovation through better matching human resources with firms’ innovation needs. Corporate innovation is a human-capital-intensive activity and consists of several subprocesses. It requires not only specialized personnel who have matching skills in the particular technical domain, but also other talents in diverse areas such as industrial design, marketing, and customer relations to facilitate collaborative innovation (Greer and Lei, 2012; Leiponen, 2006; Moenaert et al., 1994). Matching these talents with firms’ innovation needs demands access to a large pool of skilled workers. A more educated local workforce improves the likelihood of firms hiring the desired talents in many occupations and reduces their search, hiring, and information frictions, which enhances innovation efficiency.

Fourth, the agglomeration theory of innovation also predicts that local workforce education can help create knowledge spillovers and externalities in the relevant region (Glaeser, 1999; Moretti, 2004; Rauch, 1993). Employees located in a region with higher educational attainment can enrich local knowledge base and generate new ideas by interacting with other well-educated individuals within the same region. Therefore, to the extent that local workforce education contributes to knowledge spillovers in the area, firms can benefit from a large stock of knowledge available locally and increase the chance of innovation success.

Based on the above arguments, we develop the following hypothesis:

Hypothesis: Workforce education is positively related to corporate innovation.

It is possible that we may not observe a significant association between the average level of local workforce education and corporate innovation. First, for firms that compete in the national and global level for talent, they can hire the best quality employees in the country and around the world. For these firms, local workforce education level may not restrict their ability to attract human capital. Second, corporate innovation may be solely driven by inventor employees, who have extremely high qualifications in specialized sectors. In this case, the average education level of the workforce may not have a significant impact on firms’ access to innovation-related human resource. Third, although education is an important determinant of a worker’s knowledge and skills (Bacolod et al., 2010; Carlino and Kerr, 2015), other factors such as innate abilities, experiences, and training also contribute to the development of skills that are valuable to corporate innovation.

3. Research design and descriptive statistics

3.1. Data sources and sample selection

To measure local workforce education, we use the education data collected by the US Census Bureau in its American Community Survey (ACS). We obtain the data from the University of Minnesota’s Integrated Public Use Microdata Series (IPUMS-USA) (Ruggles et al., 2010). IPUMS data provide information on each respondent’s education (IPUMS data item EDUC), the MSAs where each respondent works (PWMETRO) and lives (METAREA), and the sampling weights (PERWT). We use workers’ place of work as the primary location to calculate MSA-level educational attainment (Call et al., 2017).⁴ The IPUMS education variable (EDUC) measures each respondent’s highest level of education, ranging from 0 (N/A or no schooling) to 11 (5+ years of college).⁵ The sampling weights (PERWT) capture the estimated number of people within the MSA who have similar attributes to the respondent.

We collect data on patents and citations from Kogan et al. (2017). This dataset provides information on the patent identifier, citations, technological class, patent application and grant year, and assignees’ identifiers for patents filed and granted by the US Patent and Trademark Office from 1926 to 2010. To identify the possible locations of a firm’s research centers, we follow Hombert and Matray (2018) and Matray (2021) to use the Harvard Patent Network Dataverse to obtain inventors’ addresses (Li et al., 2014).

We control for firm-level variables and MSA-level variables in our model. Information about board of directors’ education level is retrieved from the BoardEx database. Firms’ financial data and stock return data are obtained from Compustat and CRSP, respectively. MSA-level data are collected from the following sources. The State Coincident Economic Activity Index (SCI) is published by the Federal Reserve Bank of Philadelphia.⁶ The consumer price index (CPI) and unemployment data are collected from the Bureau of Labor Statistics.⁷ MSA wage data are obtained from the ACS and are weighted by sample weights reported by the IPUMS. MSA population data come from the Census Bureau.

We begin our sample from 1980, when place of work data became available in IPUMS-USA. Given that the patent data are available until 2010 and we examine the effect of workforce education on future innovation outputs, we end our sample period in 2008.⁸ We apply several data restrictions. First, we exclude firm-year observations with non-US headquarters. Second, we require all the control variables in our main regressions to be non-missing. Third, we exclude firms in utilities and financial industries, whose four-digit SIC codes are 4900–4999 and 6000–6999, respectively. Finally, because the measurement of our education variable is based on firms’ inventor locations, we drop observations without patent filing information in the preceding 5 years (year t − 4 to year t). The final sample consists of 20,401 firm-year observations between 1980 and 2008.

3.2. Variable measurements

3.2.1. Measuring workforce education

The average level of local workforce education proxies for the ease of firms’ access to skilled and knowledgeable workers. To construct this measure, we match firms’ research laboratory locations to the nearest MSA data.⁹ We first calculate the average educational attainment for all MSAs annually. Following Call et al. (2017), the weighted education level of each MSA is computed as follows

E D U C_{m, t} = \frac{\sum_{j = 1}^{n} E D U C_{j, t} \times P E R W T_{j, t}}{\sum_{j}^{n} P E R W T_{j, t}}

(1)

where $E D U C_{m, t}$ is the average educational attainment at MSA m in year t, $E D U C_{j, t}$ is the highest education level of an individual respondent j who works or lives in MSA m in year t, and $P E R W T_{j, t}$ is the sampling weight of respondent j in year t. To match survey respondents to MSAs, we use “place of work: metropolitan area” (PWMETRO) in the IPUMS dataset. In line with Call et al. (2017), for the respondents whose place of work is not identifiable, the detailed metropolitan area (METARED) in which they live is adopted instead. Given that data used to calculate $E D U C_{m, t}$ are only available in the years 1980, 1990, 2000, and 2005 to 2008, we estimate the MSA-level educational attainments for the missing years using linear interpolation.¹⁰ We use this simple method because education level often gradually increases over time.¹¹

We next identify the locations of firms’ research centers. We obtain inventors’ location information for patents assigned to each firm in a given year from the Harvard Patent Network Dataverse and assume that corporate research facilities are located in the same MSA areas as inventors’ addresses.¹² Given that an individual research laboratory may not file patents every year, we consider inventors’ locations associated with patents filed in the preceding 5 years. That is, for each inventor location associated with firm i’s filed patents, we assume that a research center is located in the MSA for the next 5 years. Prior studies also use inventor locations to identify corporate research facilities (Hombert and Matray, 2018; Lychagin et al., 2016; Matray, 2021; Mukherjee et al., 2017). We provide some basic statistics for the location and number of research centers in Table 6 and Table 7 of Appendix 4. We then match these inventor locations to the MSAs to obtain education data. Finally, we calculate the local workforce education for a firm by averaging $E D U C_{m, t}$ for all research laboratory locations in year t.

Our measure of local workforce education based on research center locations may not reflect the supply of educated workforce to the entire firm. First, this measure ignores the educational attainments in locations where a firm’s non-research operations are located. To the extent that the innovation process requires cooperation across firms’ research and non-research functions, our measure does not fully reflect the human resource requirement for corporate innovation. To alleviate this concern, we also perform a sensitivity analysis by using workforce education near corporate headquarter (see Section 4.2.2). Second, our measure captures only one local education characteristic. Other location-based education characteristics, such as the diversity of skill sets, and knowledge specialization can also affect corporate innovation. It is however beyond the scope of this article to examine all characteristics of local educational attainment. We focus on studying the association between the average education level of local workforce and corporate innovation.

3.2.2. Measuring corporate innovation

Following the existing innovation literature (e.g. Chang et al., 2015; Fang et al., 2014), we use corporate patenting activity to proxy for innovation outputs. We construct two variables: (1) the natural logarithm of one plus the truncation-adjusted number of patents filed (and eventually granted) by a firm in year t + 1 $(L N P A T_{i, t + 1})$ and (2) the natural logarithm of one plus the average truncation-adjusted number of forward citations received on the patents filed (and eventually granted) by a firm in a given year $(L N C I T_{i, t + 1}) .$ The number of patents reflects the quantity of innovation outputs, while the average citation reflects the quality (scientific value) of innovation outputs (Acharya et al., 2013; Pakes and Schankerman, 1984).

Compared with the grant year, the application year of a patent better reflects the actual innovation time, as it is close to the time when the innovation activity occurred (Griliches, 1998). However, patent application data encounter an inherent truncation problem since there is a “lag” between the filing year and the grant year (Fang et al., 2014; Hall et al., 2001, 2005). Hence, following the innovation literature (Dass et al., 2017; Fang et al., 2014; Hall et al., 2001), we adjust for the patent truncation problem. The details are explained in Appendix 2. Citations are also exposed to a truncation problem, as a patent may receive subsequent citations for many years after it is granted. Following Hall et al. (2001, 2005), we address the truncation problem of citations. Appendix 3 presents the details.

3.3. Empirical model

We estimate the following baseline model

\begin{array}{l} I N N O V A T I O N_{i, t + 1} \\ = α_{0} + α_{1} E D U C_{i, t} + α_{2} B O D_E D U C_{i, t} + α_{3} B O D_M I S S_{i, t} + α_{4} L N A S S E T S_{i, t} \\ + α_{5} R & D_{i, t} + α_{6} R O A_{i, t} + α_{7} L E V E R A G E_{i, t} + α_{8} L N A G E_{i, t} + α_{9} L N P P E_{i, t} \\ + α_{10} L N S A L E S_{i, t} + α_{11} M T B_{i, t} + α_{12} S A L E S G R_{i, t} + α_{13} C A S H_{i, t} + α_{14} B H R_{i, t} \\ + α_{15} S T O C K V O L_{i, t} + α_{16} H H I_{i, t} + α_{17} H H I_{i, t}^{2} + α_{18} S C I_{i, t} + α_{19} C P I_{i, t} \\ + α_{20} W A G E_{i, t} + α_{21} L N P O P_{i, t} + α_{22} U N E M P_{i, t} + α_{23} M S A_R O A_{i, t} \\ + α_{24} M S A_R O A V O L_{i, t} + Y E A R_{F E} + I N D U S T R Y_{F E} + ε_{i, t + 1} \end{array}

(2)

The key test variable is $E D U C_{i, t}$ —our research location-based measure of workforce education. The dependent variable $I N N O V A T I O N_{i, t + 1}$ refers to our patent and citation variables: $L N P A T_{i, t + 1}$ and $L N C I T_{i, t + 1} .$ A positive coefficient on $E D U C_{i, t}$ would support the hypothesis that local workforce education is positively associated with corporate innovation.

We control for firm-level determinants of corporate innovation. These standard controls included firm size $(L N A S S E T S_{i, t}),$ R&D expenditures scaled by total assets $(R & D_{i, t}),$ ¹³ profitability $(R O A_{i, t}),$ leverage $(L E V E R A G E_{i, t}),$ firm age $(L N A G E_{i, t}),$ capital intensity $(L N P P E_{i, t}),$ labor productivity $(L N S A L E S_{i, t}),$ growth opportunities ( $M T B_{i, t}$ and $S A L E S G R_{i, t}),$ cash holdings $(C A S H_{i, t}),$ past stock returns and volatility ( $B H R_{i, t}$ and $S T O C K V O L_{i, t}),$ and industry concentration ( $H H I_{i, t}$ and $H H I_{i, t}^{2}) .$ We further control for the average education level of all the executives and directors listed in the BoardEx database $(B O D_E D U C_{i, t}) .$ Following Call et al. (2017), we also include a dummy variable $(B O D_M I S S_{i, t}),$ which equals one if the board of directors’ education information does not appear in the BoardEx database.

Given that our workforce education measure is location-based, it is important to control for MSA-level economic and population characteristics. Consistent with Call et al. (2017), we control for the state coincident index $(S C I_{i, t}),$ consumer price index $(C P I_{i, t}),$ average wages in the MSA $(W A G E_{i, t}),$ natural log of the workforce population $(L N P O P_{i, t}),$ unemployment rate $(U N E M P_{i, t}),$ and average profitability and earnings volatility (in the preceding 5 years) for all listed firms headquartered in the MSA ( $M S A_R O A_{i, t}$ and $M S A_R O A V O L_{i, t},$ respectively). Appendix 1 provides detailed variable definitions. The model also includes year fixed effects and industry fixed effects based on the Fama-French 48 industries.¹⁴ The baseline regressions are estimated using standard errors clustered by firm.

3.4. Sample description

Table 1 presents descriptive statistics for all variables. All continuous variables are winsorized at the 1st and 99th percentiles. The mean values of $P A T N U M_{i, t + 1}$ and $C I T N U M_{i, t + 1}$ indicate that the average innovation-intensive firm files (and eventually receives) 38 patent applications per year, and patents are on average cited 36 times.¹⁵ $L N P A T_{i, t + 1}$ $(L N C I T_{i, t + 1})$ has a mean of 2.024 (1.792) and a median of 1.792 (1.486). Firms on average have 9.36 research centers. Firms’ research centers are in MSAs with an average education scale of 7.36, corresponding to an average highest level of education between 1 and 2 years of college (scale value of 7 and 8, respectively). A standard deviation of 0.350 for $E D U C_{i, t}$ suggests that most firms are exposed to above college-level local educational attainments. The distribution of $E D U C_{i, t}$ is similar to that reported in Call et al. (2017), who use a more recent sample.

Table 1.

Descriptive statistics.

Variables	Mean	SD	P25	Median	P75
Innovation variables
PATNUM	38.310	191.900	2.000	5.000	18.000
CITNUM	36.390	163.100	1.245	4.584	17.770
LNPAT	2.024	1.657	0.693	1.792	2.996
LNCIT	1.792	1.650	0.247	1.486	2.838
ORIGINAL	0.461	0.324	0.000	0.530	0.735
GENERAL	0.359	0.332	0.000	0.376	0.657
MV_PAT	2.551	2.415	0.420	1.936	4.119
NPROD	0.118	0.354	0.000	0.000	0.000
LNNSCIT	1.725	1.601	0.216	1.422	2.741
# of RESEARCH CENTER	9.360	13.337	2.000	5.000	10.000
Test variable
EDUC	7.361	0.350	7.143	7.372	7.613
Firm-level controls
BOD_EDU	1.605	1.454	0.000	2.400	2.914
BOD_MISS	0.437	0.496	0.000	0.000	1.000
LNASSETS	5.955	2.112	4.411	5.825	7.461
R&D	0.087	0.121	0.013	0.044	0.111
ROA	–0.027	0.233	–0.024	0.042	0.082
LEVERAGE	0.194	0.187	0.033	0.168	0.291
LNAGE	2.616	0.961	1.946	2.639	3.332
LNPPE	3.575	0.968	2.943	3.486	4.151
LNSALES	5.010	0.824	4.571	5.041	5.487
MTB	2.039	2.112	0.911	1.353	2.313
SALESGR	0.195	0.583	–0.012	0.088	0.228
CASH	0.210	0.234	0.031	0.108	0.319
BHR	0.045	0.600	–0.308	–0.050	0.232
STOCKVOL	0.132	0.085	0.075	0.109	0.164
HHI	0.248	0.183	0.118	0.194	0.322
HHI²	0.095	0.144	0.014	0.038	0.104
MSA-level controls
SCI	154.200	34.540	131.800	157.700	180.600
CPI	117.900	24.780	99.970	120.100	137.800
WAGE	30.920	10.220	23.510	31.200	38.430
LNPOP	13.890	0.656	13.640	13.970	14.290
UNEMP	5.820	1.696	4.600	5.480	6.750
MSA_ROA	–0.063	0.093	–0.099	–0.044	–0.000
MSA_ROAVOL	0.099	0.059	0.054	0.086	0.129

SD: standard deviation.

This table presents descriptive statistics. The sample consists of 20,401 firm-year observations. PATNUM is a firm’s truncation-adjusted number of patents filed (and eventually granted) in a given year. CITNUM is a firm’s average truncation-adjusted number of citations received on the firm’s patents filed (and eventually granted) in a given year. Appendix 1 provides detailed variable definitions. Continuous variables are winsorized at the 1% and 99% levels.

4. Results

4.1. Baseline empirical results

Table 2 reports our baseline results. Columns (1) and (2) report the results on the effect of the average workforce education on the number of patents filed (and eventually granted) and average citations per patent in year t + 1, respectively. Column (1) shows that our inventor location-based measure of workforce education is positively associated with the number of patents in year t + 1 (coeff. = 0.414, p-value < 0.001). The economic significance of the relation is sizable. For example, a one standard deviation increase in $E D U C_{i, t}$ is associated with an estimated increase of patent counts by 15.6 (calculated as e^{0.350 × 0.414} − 1) percentage points in year t + 1. We also examine the associations between $E D U C_{i, t}$ and patent counts in year t + 2 and t + 3. Unreported results show a steadily increasing trend in the coefficient of $E D U C_{i, t}$ from year t + 1 to year t + 3, suggesting that workforce education has a long-term effect on corporate innovation. Overall, the results are consistent with workforce education having a positive effect on the quantity of innovation outputs, supporting our hypothesis. We further note that board education level $(B O D_E D U C_{i, t})$ is also positively associated with the number of patents.

Table 2.

Baseline results—workforce education and corporate innovation.

Variables	(1)	(2)
	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
EDUC	0.414***	0.534***
	(0.000)	(0.000)
BOD_EDU	0.152***	0.075
	(0.001)	(0.144)
BOD_MISS	0.268*	–0.019
	(0.058)	(0.900)
LNASSETS	0.546***	0.524***
	(0.000)	(0.000)
R&D	2.558***	2.288***
	(0.000)	(0.000)
ROA	0.254***	0.254***
	(0.002)	(0.004)
LEVERAGE	–0.448***	–0.475***
	(0.000)	(0.000)
LNAGE	0.083***	0.034
	(0.000)	(0.168)
LNPPE_EMP	0.057**	0.034
	(0.030)	(0.217)
LNSALES_EMP	–0.070***	–0.060**
	(0.007)	(0.030)
MTB	0.060***	0.079***
	(0.000)	(0.000)
SALESGR	–0.014	0.020
	(0.318)	(0.233)
CASH	0.289***	0.288***
	(0.001)	(0.003)
BHR	0.010	–0.009
	(0.458)	(0.558)
STOCKVOL	0.171	0.325*
	(0.301)	(0.087)
HHI	0.036	0.344
	(0.919)	(0.385)
HHI²	–0.009	–0.405
	(0.983)	(0.402)
SCI	0.007**	0.009***
	(0.033)	(0.005)
CPI	0.002	0.002
	(0.298)	(0.411)
WAGE	–0.032***	–0.035***
	(0.000)	(0.000)
LNPOP	0.115***	0.125***
	(0.000)	(0.000)
UNEMP	0.008	0.041**
	(0.574)	(0.013)
	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
MSA_ROA	–0.904***	–1.053***
	(0.000)	(0.000)
MSA_ROAVOL	0.578**	0.375
	(0.038)	(0.220)
Constant	–5.658***	–6.783***
	(0.000)	(0.000)
Year FE	Included	Included
Industry FE	Included	Included
Observations	20,401	20,401
R ²	0.539	0.454

This table reports the regression results of the relation between workforce education and corporate innovation. Column (1) presents the results for the number of patents, while column (2) presents the results for the average number of citations. Appendix 1 provides detailed variable definitions. Year and industry fixed effects are included in all models. p-values shown in parentheses are estimated using standard errors adjusted for heteroskedasticity and clustered by firm. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

Column (2) indicates that $E D U C_{i, t}$ is significantly and positively associated with average citations per patent (coeff. = 0.534, p-value < 0.001). Regarding the economic significance, a one standard deviation increase in $E D U C_{i, t}$ is associated with 20.6 percentage-point increases in the average citations of patents in year t + 1. The results support the notion that local workforce education positively affects the quality of corporate innovation.¹⁶ We find no evidence that the average board education is associated with average citations. Thus, it appears that workforce education is more important for patent quality than is director education.

4.2. Robustness tests

4.2.1. Alternative measures of innovation quality

In this section, we examine whether our baseline results are robust to using several alternative measures of innovation quality. Hall et al. (2001) introduce patent originality and generality as measures of scientific value of innovation. Specifically, patent originality captures the extensive use of a diverse range of technological knowledge at the patent level. It is calculated as one minus the Herfindahl index of the distribution of patents cited by a firm’s filed patents in a given year, across three-digit technological classes. Patent generality reflects the usefulness of patents for future innovation across many technological classes. It is calculated as one minus the Herfindahl index of the received citations of newly filed patents of a firm in a given year, across three-digit technological classes.

Our third alternative measure of patent quality is the economic value of patents. We use the market value of patents provided by Kogan et al. (2017). This measure is estimated as a firm’s excess return over the market return in the 3-day period surrounding the patent approval date, multiplied by the market value of the firm at the beginning of the 3-day period (Kogan et al., 2017).

Next, we consider new product introductions as important final innovation outcomes for firms. In addition to inventors, new product introductions require contributions from many skilled professionals in marketing, design, distributions, manufacturing, and sales. Thus, workforce education should have implications for new product introductions that are beyond the channel of technological innovation. We measure new product introductions using the data provided by Mukherjee et al. (2017). Specifically, Mukherjee et al. (2017) use textual analysis to identify news of firms’ new product announcements. They then count the number of major product announcements which attract cumulative abnormal returns above the 75th percentile in a given year. Finally, we distinguish between self-citations and non-self-citations. According to Hall et al. (2001), self-citations refer to citations that cite patents invented by the same assignees. As a robustness test, we remove self-citations to only focus on non-self-citations.

We summarize the results of using alternative measures of innovation quality in Panel A of Table 3. Specifically, the associations between education and originality, generality, market value of patents, new product announcements, and non-self-citations are presented in columns (1) to (5), respectively. We find that all five alternative measures of innovation quality are positively associated with workforce education. These results confirm our main finding.

Table 3.

Robustness analyses.

Panel A: Alternative measures of innovation quality
Variables	(1)	(2)	(3)	(4)	(5)
Variables	$O R I G I N A L_{t + 1}$	$G E N E R A L_{t + 1}$	$M V_P A T_{t + 1}$	$N P R O D_{t + 1}$	$L N N S C I T_{t + 1}$
EDUC	0.097***	0.128***	0.431***	0.085***	0.508***
	(0.000)	(0.000)	(0.000)	(0.006)	(0.000)
Controls	Included	Included	Included	Included	Included
Year FE	Included	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included	Included
Observations	20,401	20,401	20,401	12,303	20,401
R ²	0.227	0.358	0.693	0.125	0.453
Panel B: Measuring employee education near firms’ headquarters
Variables				(1)	(2)
Variables				$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
EDUC_HQ				0.079*	0.083**
				(0.057)	(0.037)
Controls				Included	Included
Year FE				Included	Included
Industry FE				Included	Included
Observations				77,281	77,281
R ²				0.430	0.360
Panel C: Controlling for employee treatment
Variables		(1)	(2)	(3)	(4)
Variables		$L N P A T_{t + 1}$	$L N C I T_{t + 1}$	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
EDUC		0.506***	0.528***	0.504***	0.526***
		(0.000)	(0.000)	(0.000)	(0.001)
EMPTREAT		0.566***	0.551***
		(0.004)	(0.008)
EMPTREAT _STR				1.025***	1.064***
				(0.000)	(0.000)
EMPTREAT _CON				–0.048	0.062
				(0.884)	(0.812)
Other Controls		Included	Included	Included	Included
Year FE		Included	Included	Included	Included
Industry FE		Included	Included	Included	Included
Observations		8767	8767	8767	8767
R ²		0.551	0.446	0.553	0.448
Panel D: Controlling for management and employee incentives
Variables		(1)	(2)	(3)	(4)
Variables		$L N P A T_{t + 1}$	$L N C I T_{t + 1}$	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
EDUC		0.889***	1.235***	0.900***	1.253***
		(0.000)	(0.000)	(0.000)	(0.000)
DELTA_TOP5		–0.011	0.005	–0.005	0.013
		(0.741)	(0.887)	(0.877)	(0.719)
VEGA_TOP5		–0.019	0.012	–0.027	–0.004
		(0.617)	(0.781)	(0.488)	(0.931)
OPT_ EMP		0.151***	0.203***
		(0.000)	(0.000)
OPT_TOP5		–0.001	–0.018
		(0.981)	(0.517)
AVEOPT_EMP				0.144***	0.198***
				(0.000)	(0.000)
AVEOPT_TOP5				0.047*	0.051
				(0.097)	(0.107)
Other Controls		Included	Included	Included	Included
Year FE		Included	Included	Included	Included
Industry FE		Included	Included	Included	Included
Observations		6238	6238	6254	6254
R ²		0.526	0.483	0.523	0.478
Panel E: Controlling for urban employment intensity and industry concentration
Variables		(1)	(2)	(3)	(4)
Variables		$L N P A T_{t + 1}$	$L N C I T_{t + 1}$	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$
EDUC		0.380***	0.508***	0.379***	0.507***
		(0.001)	(0.000)	(0.000)	(0.000)
MSA_EMPINT		–0.023	–0.012	–0.022	–0.011
		(0.144)	(0.482)	(0.160)	(0.497)
MSA_INDCON				0.526	0.236
				(0.502)	(0.785)
Other controls		Included	Included	Included	Included
Year FE		Included	Included	Included	Included
Industry FE		Included	Included	Included	Included
Observations		20,145	20,145	20,145	20,145
R ²		0.539	0.454	0.539	0.454

Panel A reports results using alternative measures of innovation quality: patent originality, generality, the market value of patents, new product introductions, and non-self-citations. Panel B reports results using the education level of the MSAs where corporate headquarters are located (EDUC_HQ). Panel C reports results after controlling for employee treatment. Panel D reports results after controlling for management and employee incentives. Panel E reports results after controlling for urban employment intensity and geographic industry concentration. Appendix 1 provides detailed variable definitions. Year and industry fixed effects are included. p-values shown in parentheses are estimated using standard errors adjusted for heteroskedasticity and clustered by firm. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

4.2.2. Workforce education based on corporate headquarter location

Following Call et al. (2017), we use the average education level of the workforce in the MSA of a firm’s headquarter location as an alternative measure of workforce education. We identify firms’ historical headquarter locations by searching firms’ historical 10 K filings. While corporate innovation activities typically occur in a firm’s research centers, employees in the firm’s headquarter can influence corporate innovation strategies and internal information flows that are important for knowledge sharing. We find that for 16,302 out of 20,401 (79.9%) firm-year observations in our baseline sample, headquarters are in the same MSA as a research center.¹⁷ This is consistent with headquarters playing an important role in the innovation process. Furthermore, measuring workforce education near corporate headquarters expands our sample to include non-patent-filing firms. Thus, we also test the sensitivity of our results to a broader sample in this analysis.

The results are reported in Panel B of Table 3. Education level near headquarters is positively and significantly related to patent counts and citation counts at the 5% or 10% level. The effects are smaller than the baseline results. Overall, workforce education near headquarters is important for corporate innovation; however, the effects may be smaller than for workforce education near corporate research centers.

4.2.3. Controlling for employee treatment

Our results are subject to endogeneity concerns. In the following three sections, we report tests aiming to address omitted variable problems. Prior literature (Chen et al., 2016; Mao and Weathers, 2019) finds that better employee treatment promotes corporate innovation. Educated workforce may demand better employee treatment, which affects innovation through this effect. We control for employee treatment $(E M P T R E A T_{i, t})$ and the strength and concern components of employee treatment ( $E M P T R E A T_S T R_{i, t}$ and $E M P T R E A T_C O N_{i, t})$ in our regressions. Employee treatment is measured using the employee treatment scores from the KLD database.

The results are reported in Panel C of Table 3. We find that the coefficients on $E M P T R E A T_{i, t}$ and $E M P T R E A T_S T R_{i, t}$ are significantly positive. Moreover, the coefficients on workforce education $(E D U C_{i, t})$ are positive and significant at 1% level in all regressions. Thus, both workforce education and employee treatment are important for corporate innovation.

4.2.4. Controlling for management and employee incentives

Compensation incentives affect innovation (Chang et al., 2015; Manso, 2011). To the extent that workforce education affects firms’ compensation design, management and employee incentives can be considered as omitted variables. We control for management and employee incentives. We measure management incentives using delta and vega of the top five managers’ compensation.¹⁸ We also calculate the natural logarithm of total and average option values granted to the top five managers. To measure employee incentives, we use the natural logarithm of total and average option values granted to employees. Panel D of Table 3 reports these results. We find that $E D U C_{i, t}$ is still significantly and positively associated with patent applications and average patent citations. The results also reveal a positive association between employee incentives and corporate innovation (Chang et al., 2015).

4.2.5. Controlling for urban employment intensity and geographic industry concentration

Our location-based measure of workforce education can be correlated with other location-based factors that affect the agglomeration of innovation. In the baseline model, we have controlled for several MSA-level economic and population measures. In this analysis, we further control for two aspects of urban agglomeration: urban concentration of employment and industry (Carlino and Kerr, 2015). Prior literature argues that urban employment density and the colocations of firms in the same industry facilitate knowledge sharing, which is essential to innovation (Carlino et al., 2007). We measure urban employment concentration as the number of full-time workers per square mile of urban land in the MSA (Carlino et al., 2007). We measure industry concentration in MSAs as the number of same industry firms per one thousand population. Similar to our education measure, we aggregate the research-center level measures of employment and industry firm density to the firm level. Panel E of Table 3 reports the results of this sensitivity analysis. We find that controlling for these MSA-level measures have little effects on the relationship between workforce education and firm innovation.

4.2.6. Other robustness tests

We conduct a battery of additional tests to assess the sensitivity of our results. These analyses are reported in Appendix 5. First, we control for firms’ corporate social responsibility (CSR) performance and product market strategies. We measure firms’ product market strategies using their profit margin. Innovation-lead product differentiation should positively affect profit margin. Controlling for firms’ CSR performance and profit margin do not materially affect our results (Table 8). Second, we control for employee intensity and the natural log of R&D spending. The results are reported in Table 9. We find that including employee intensity has little effect on the education–innovation relationship. Controlling for the natural log value of R&D expenses causes the effect of education on patent count to become insignificant. However, citation count is still significantly and positively associated with workforce education after controlling for the log value of R&D expenses. The results suggest that high workforce education improves the quality of corporate innovation but not innovation efficiency (i.e. turning R&D expenditures into patents). Finally, controlling for industry-year fixed effects and forward R&D outlays do not affect the main results (Table 10 and Table 11).

4.3. Research-center level analyses

Our second endogeneity concern relates to the selection bias problem. That is, our results may be driven by unobservable firm policies to select research center locations and employ highly educated employees. We note that this would not contradict our argument that highly educated workforce is an important input to corporate innovation. Nevertheless, in this section, we attempt to control for unobservable firm-year level characteristics. Our strategy is to use a sample of research-center-firm-year observations to explore within-firm variations.

Our workforce education variable ( $E D U C_{c, i, t},$ where c indexes research center)¹⁹ is now measured as the education level in the MSA where a research center is located, and the dependent variables ( $L N P A T_{c, i, t + 1},$ $L N C I T_{c, i, t + 1},$ $O R I G I N A L_{c, i, t + 1},$ $G E N E R A L_{c, i, t + 1}$ and $M V_P A T_{c, i, t + 1}$ )²⁰ are measured as the quantity and quality of innovation outputs produced by a research center. Similar to $E D U C_{c, i, t},$ the MSA-level control variables are measured using the MSA where a research center is located. We control for firm-year (firm × year) fixed effects and drop all firm-level control variables in the regressions. This design is attractive because all firm-year level unobservable characteristics are subsumed by the firm-year fixed effects. Thus, we effectively explore the within-firm-year variations in the workforce education measure across different research centers. This helps address the concern that there exist firm-year specific selections that can explain the link between workforce education and corporate innovation.

The results are reported in Table 4. Consistent with our expectations, we find that the coefficients on $E D U C_{c, i, t}$ are positive and significant at the 1% level in all regressions, suggesting that a research center’s workforce education has a significantly positive effect on that research center’s innovation outputs and quality. The results are qualitatively similar if we control for firm fixed effects and year fixed effects in the regressions and if we restrict the sample to research centers with non-zero patents in a given year (unreported).

Table 4.

Research-center level analyses.

Variables	(1)	(2)	(3)	(4)	(5)
Variables	$L N P A T_{t + 1}$	$L N C I T_{t + 1}$	$O R I G I N A L_{t + 1}$	$G E N E R A L_{t + 1}$	$M V_P A T_{t + 1}$
EDUC	0.041***	0.027***	0.016***	0.021***	0.077***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
SCI	0.001***	0.001***	0.001***	0.001***	0.004***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
CPI	0.002***	0.000***	0.000***	0.000***	0.002***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
WAGE	0.022***	0.004***	0.005***	0.002***	0.040***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
LNPOP	0.059***	0.012***	0.011***	0.009***	0.099***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
UNEMP	0.016***	0.006***	0.004***	0.005***	0.050***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
MSA_ROA	–0.001	–0.000	–0.007	–0.001	–0.104**
	(0.960)	(0.998)	(0.317)	(0.914)	(0.038)
MSA_ROAVOL	0.034	0.015	0.022**	–0.009	0.189**
	(0.250)	(0.192)	(0.033)	(0.240)	(0.011)
Constant	–1.784***	–0.518***	–0.448***	–0.366***	–3.213***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Firm × Year FE	Included	Included	Included	Included	Included
Observations	163,060	163,060	163,060	163,060	163,060
R ²	0.163	0.243	0.189	0.245	0.216

This table reports regression results for research-center level analyses. The sample is based on location-firm-year observations. Firm-year fixed effects are included in all models. Appendix 1 provides detailed variable definitions. p-values are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

Appendix 6 reports several additional sensitivity tests. Specifically, Table 12 shows that the results are robust to including both firm fixed effects and industry-year fixed effects. We also find that results are similar when we use industry-adjusted measures of innovation outcomes (Table 13). Since a significant proportion of observations are located in California or Massachusetts, we drop these observations and find consistent results (Table 14).

Although the research center analyses alleviate the concern that firm-level and firm-year-level selection biases influence our results, we acknowledge that we are not able to fully address firms’ choices associated with the selection of individual research centers. Specifically, there may exist research-center level variations of the same firm in areas such as strategic focus, culture, employee hiring, and training. Firms are also likely to allocate different levels of resources to different research centers. Furthermore, research-center level characteristics may change over time as firms adjust to local economic and social changes. This suggests that any shocks to local educational attainment may attract unobservable corporate policy reactions including hiring and investment changes which correlate with both workforce education and innovation activities at the research center. Thus, we caution readers not to interpret our results as evidence of causality.

5. When is workforce education more important for corporate innovation?

In this section, we examine the cross-sectional variations in the relation between workforce education and corporate innovation. Specifically, we investigate whether the relationship between workforce education and innovation is affected by firm size, labor intensity, competition environment, high-tech industries, and educational attainment.

5.1. Firm size and the workforce education–innovation relationship

It is unclear whether the supply of educated workforce is more important for large or small firms’ innovation. Large firms may be more capable of exploring employees’ talents, given their vast corporate resources and extensive experience in innovation and human resource management. Large firms may also be able to attract the best local talents. However, small firms may rely more on local workforce for knowledge and innovation inputs. To test the effect of firm size, we partition our sample into two subsamples based on the median value of firm size and estimate our baseline regressions for each subsample. Panel A of Table 5 indicates that the effects of workforce education on future patent counts and average citations are only significant for large firms. The differences between large and small firms are significant. Overall, it appears that workforce education plays a greater role in corporate innovation for large firms.

Table 5.

Cross-sectional variations in the workforce education–innovation relation.

Panel A: Firm size and the workforce education–innovation relation
Subsamples	$L N P A T_{t + 1}$		$L N C I T_{t + 1}$
	(1)	(2)	(3)	(4)
	Large firms	Small firms	Large firms	Small firms
EDUC	0.648***	0.062	0.895***	0.130
	(0.001)	(0.418)	(0.000)	(0.158)
Controls	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Observations	10,370	10,031	10,370	10,031
R ²	0.527	0.263	0.493	0.192
Test of coeff. diff. (p-value)	9.28*** (0.002)		13.53*** (0.000)
Panel B: Employee intensity and the workforce education–innovation relation
Subsamples	$L N P A T_{t + 1}$		$L N C I T_{t + 1}$
	(1)	(2)	(3)	(4)
	High emp.	Low emp.	High emp.	Low emp.
EDUC	0.162	0.686***	0.291**	0.784***
	(0.111)	(0.000)	(0.011)	(0.000)
Controls	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Observations	9876	10,525	9876	10,525
R ²	0.531	0.541	0.468	0.456
Test of coeff. diff. (p-value)	11.17*** (0.000)		7.70*** (0.006)
Panel C: Market competition and the workforce education–innovation relation
Subsamples	$L N P A T_{t + 1}$		$L N C I T_{t + 1}$
	(1)	(2)	(3)	(4)
	Low comp.	High comp.	Low comp.	High comp.
EDUC	0.365***	0.466***	0.416***	0.648***
	(0.002)	(0.000)	(0.001)	(0.000)
Controls	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Observations	10,195	10,206	10,195	10,206
R ²	0.579	0.517	0.491	0.440
Test of coeff. diff. (p-value)	0.40(0.526)		1.64(0.200)
Panel D: High-tech industries and the workforce education–innovation relation
Variables	$L N P A T_{t + 1}$		$L N C I T_{t + 1}$
	(1)	(2)	(2)	(4)
	Low tech.	High tech.	Low tech.	High tech.
EDUC	0.500***	0.087	0.581***	0.187
	(0.000)	(0.476)	(0.000)	(0.169)
Controls	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Observations	10,456	9945	10,456	9945
R ²	0.596	0.569	0.495	0.477
Test of coeff. diff. (p-value)	5.66** (0.017)		4.33** (0.038)
Panel E: Average educational attainment and the workforce education–innovation relation
Variables	$L N P A T_{t + 1}$		$L N C I T_{t + 1}$
	(1)	(2)	(3)	(4)
	Low edu.	High edu.	Low edu.	High edu.
EDUC	1.059***	0.186*	1.052***	0.371***
	(0.000)	(0.100)	(0.000)	(0.003)
Controls	Included	Included	Included	Included
Year FE	Included	Included	Included	Included
Industry FE	Included	Included	Included	Included
Observations	3290	17,111	3290	17,111
R ²	0.596	0.548	0.549	0.449
Test of coeff. diff. (p-value)	10.91*** (0.001)		5.52** (0.019)

This table reports cross-sectional firm-level analyses of the relationship between workforce education and corporate innovation. Panel A partitions the sample based on the median of total assets. Panel B partitions the sample based on the median of employee number to total assets ratio. Panel C partitions the sample based on the median of the Herfindahl index at the four-digit SIC industry level. Panel D partitions the sample into high-tech and non-high-tech industries. High-tech industries belong to the following list of three-digit SIC codes: 283, 357, 366, 367, 382, 384, 481, 482, 489, 737, and 873. Panel E reports separate regression results for firms located in areas with below-college and above-college level average workforce education. Year and industry fixed effects are included in all models. p-values shown in parentheses are estimated using standard errors adjusted for heteroskedasticity and clustered by firm. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

5.2. Employee intensity and the workforce education–innovation relationship

Local workforce education may be more important for firms that rely on fewer employees to manage their assets. Thus, we expect the workforce education–innovation relationship to be stronger for firms that have lower employee intensity. To examine this conjecture, we split the sample at the median value of the number of employees scaled by total assets. As shown in Panel B of Table 5, the coefficients on $E D U C_{i, t}$ are significantly larger for less employee-intensive firms than more employee-intensive firms in both the patent and citation regressions. This suggests that the education level of local workforce is more important for firms that rely on fewer employees to manage their operations.

5.3. Product market competition and the workforce education–innovation relationship

Prior studies argue that firms innovate to escape product market competition (Aghion et al., 2005; Bloom et al., 2016; Hombert and Matray, 2018). We predict that the supply of highly educated workers would be more valuable for firms facing high product market competition. We use the Herfindahl index at the four-digit SIC industry level to partition the sample into a high industry concentration (low competition) sample and a low industry concentration (high competition) sample. As reported in Panel C of Table 5, we find that the effects of workforce education on future patents and average citations are slightly higher for firms facing high competition than for firms facing low competition. However, the differences are not statistically significant.

5.4. High-tech industries and the workforce education–innovation relationship

A fourth cross-sectional test examines whether local workforce education is associated with innovation of both high-tech and non-high-tech firms. We split the sample into high-tech industries and other industries based on the classification scheme of Kile and Phillips (2009). The results are reported in Panel D of Table 5. We find that the effect of education on innovation is only significant for non-high-tech firms and the differences in coefficients are statistically significant. This result appears to be surprising. However, high-tech firms compete on innovation, which may require them to fiercely attract for top talents in the national or even global level. Thus, the education level of employees may not differ significantly across surviving high-tech firms. In contrast, local average educational attainment is more important for the innovation of non-high-tech firms.

5.5. Average educational attainment and the workforce education–innovation relationship

Finally, we split the sample based on MSA’s average educational attainment. Specifically, we separate MSAs into below-college level average education (EDUC < 7) and above-college level average education (EDUC >= 7). The results reported in Panel E of Table 5 show that workforce education is positively related to innovation outputs in both subsamples. In addition, the effect of average workforce education on corporate innovation is more pronounced for low education regions. This suggests that our results are not simply driven by workers with extremely high qualifications.

6. Conclusion

We find that the average education level in the MSAs where a firm’s research centers are located is significantly and positively related to future corporate innovation performance. We provide evidence that firms exposed to higher educated workforce generate more patents and higher quality patents. Workforce education also contributes positively to new product introductions and the economic value of innovation. Furthermore, the positive association between workforce education and corporate innovation is more pronounced for firms that are large, use fewer employees to manage operations, in non-high-tech industries, and located in areas with below-college average education.

The findings of this study have two important implications. First, the results highlight the significant effect of proximity to highly educated workers in corporate innovation. Second, this research suggests that local social and economic infrastructure—particularly education infrastructure—provides important support for firms’ innovation activities.

Footnotes

Appendix 1

Variable definitions.

Variables
Dependent variables
$L N P A T_{i, t + 1}$	Natural logarithm of one plus the truncation-adjusted number of patents filed (and eventually granted) by a firm in year t + 1.
$L N C I T_{i, t + 1}$	Natural logarithm of one plus the average truncation-adjusted number of citations received on the patents filed (and eventually granted) by a firm in year t + 1.
$O R I G I N A L_{i, t + 1}$	One minus the Herfindahl index of the number of patents cited by a firm’s filed patents in year t + 1, across three-digit technological classes defined by the patent database.
$G E N E R A L_{i, t + 1}$	One minus the Herfindahl index of all future patents citing the patents filed by a firm in year t + 1, across three-digit technological classes.
$M V_P A T_{i, t + 1}$	Natural logarithm of total market value of a firm’s patents filed (and eventually granted) in year t + 1.
$N E W P R O D_{i, t + 1}$	A firm’s major product introductions, which is the number of new product announcements with cumulative abnormal returns of firm i above the 75th percentile in year t + 1. The data are available from Professor Alminas Žaldokas’ website (http://www.alminas.com/).
$L N N S C I T_{i, t + 1}$	Natural logarithm of one plus the average truncation-adjusted number of non-self-citations received on the patents filed (and eventually granted) by a firm in year t + 1.
Test variable
$E D U C_{i, t}$	The weighted-average education level of respondents to ACS in the MSA where a firm’s research centers are located in year t. Research centers are identified based on inventors’ locations. MSA-level education level is calculated as follows: $E D U C_{m, t} = \frac{\sum_{i = 1}^{n} E D U C_{i} * P E R W T_{i}}{\sum_{i = 1}^{n} P E R W T_{i}}$ where m, t, i, and n refer to a specific MSA, year, individual respondent, and the total number of respondents in the MSA, respectively. PERWT is the estimated number of workforce members in the MSA with similar attributes (provided by the US Census Bureau). $E D U C_{i, t}$ is the mean of the weighted-average education level of respondents to ACS in the MSAs where firm i’s inventors are located in the previous 5 years. Inventors’ locations are obtained from patents filed by firm i in the preceding 5 years in the Harvard Patent Network Dataverse. We estimate the value of $E D U C_{i, t}$ by interpolating the weighted-average education level during the periods 1980 to 1990, 1990 to 2000, and 2000 to 2005, respectively, where education data and respondents’ work locations are only available for the years 1980, 1990, 2000, and 2005 from ACS. From 2005 onwards, this variable is calculated directly from the Integrated Public Use Microdata Series (IPUMS-USA) database.
Firm-level control variables
$B O D_E D U_{i, t}$	Educational attainment of a firm’s executives and directors listed in BoardEx, measured by the average degree index. It equals 0 for a degree lower than an associate degree, 1 for an associate degree, 2 for a bachelor’s degree, 3 for a master’s degree, and 4 for a degree higher than a master’s degree (e.g. PhD or JD). If the education information of the board of directors does not appear in BoardEx, we assign a value of 0.
$B O D_M I S S_{i, t}$	Dummy variable that equals 1 if the education information of the board of directors does not appear in BoardEx, and 0 otherwise.
$L N A S S E T S_{i, t}$	Firm size, defined as the logarithm of total assets (AT).
$R & D_{i, t}$	A firm’s total R&D expenditures (XRD) scaled by total assets (AT) in a given year. $R & D_{i, t}$ equals 0 if XRD is missing.
$R O A_{i, t}$	Return on total asset, which equals earnings before interest (EBITDA) divided by total assets (AT).
$L E V E R A G E_{i, t}$	Leverage ratio, defined as book value of debt (DLC + DLTT) divided by total assets (AT).
$L N A G E_{i, t}$	Firm age, measured by the logarithm of one plus the number of years since the firm first reported in CRSP.
$L N P P E_{i, t}$	Capital intensity, measured by the natural logarithm of the book value of firm i’s property, plant, and equipment (PPENT) scaled by number of employees (EMP).
$L N S A L E S_{i, t}$	Labor productivity, measured by the natural logarithm of the book value of firm i’s total sales (SALE) scaled by number of employees (EMP).
$M T B_{i, t}$	Market to book value, which equals the sum of market value of equity (CSHO*PRCC_F), book value of debt (DLC + DLTT), preferred stock (PSTKL), and Deferred Taxes and Investment Tax Credit (TXDITC), divided by total assets (AT).
$S A L E S G R_{i, t}$	Growth rate of sales, defined as the changes in sales (SALE) divided by the sales in year t − 1.
$C A S H_{i, t}$	Cash to assets ratio, which is the sum of cash and short-term investments (CHE) divided by total assets (AT).
$B H R_{i, t}$	A firm’s buy-and-hold return over the fiscal year.
$S T O C K V O L_{i, t}$	Stock volatility, which is the standard deviation of monthly stock return (RET) over the fiscal year.
$H H I_{i, t}$	Herfindahl index of the four-digit SIC industry where a firm belongs, measured at the year end.
$H H I_{i, t}^{2}$	Square of the Herfindahl index.
$P M A R G I N_{i, t}$	Profit margin, which is the ratio of net income (IB) to sales (SALE).
$L N R D_{i, t}$	The natural log of one plus R&D expenditures (XRD).
$L N E M P_{i, t}$	The natural log of one plus the number of employees (EMP).
$D E L T A_T O P 5_{i, t}$	The average dollar change in the value of the top-five executive’s stock and option portfolio for a 1% change in stock price.
$V E G A_T O P 5_{i, t}$	The average dollar change in the value of the top-five executive’s stock and option portfolio for a 0.01 change in standard deviation of stock returns.
$O P T_E M P_{i, t}$	The natural log of the total value of stock options granted to all employees who were not among the top-five executives. The total value of stock options granted to all employees is estimated by using the executive’s share of total option grants (PCTTOTOPT) and the Black-Scholes option values reported for the executive grants (BLKSHVAL).
$O P T_T O P 5_{i, t}$	The natural log of the total value of stock options granted to the top-five executives.
$A V E O P T_E M P_{i, t}$	The natural log of the average value of stock options granted to all employees who were not among the top-five executives.
$A V E O P T_T O P 5_{i, t}$	The natural log of the average value of stock options granted to the top-five executives
$E M P T R E A T_{i, t}$	The sum of strength indicators under the employee treatment category of the KLD database scaled by the total number of available categories minus the sum of concern indicators under the employee treatment category of the KLD database scaled by the total number of available categories.
$E M P T R E A T_S T R_{i, t}$	The sum of strength indicators under the employee treatment category of the KLD database divided by the total number of available categories.
$E M P T R E A T_C O N_{i, t}$	The sum of concern indicators under the employee treatment category of the KLD database divided by the total number of available categories.
$C S R_{i, t}$	Firm i’s CSR performance in year t. It is the sum of yearly adjusted community activities, corporate governance, diversity, employee relations, environmental record, human rights, and product quality and safety KLD STATS corporate social responsibility scores. Adjusted CSR is estimated by scaling the raw strength and concern scores of each category by the number of items of the strength and concern of that category in the year and then taking the net difference between adjusted strength and concern scores for that category.
MSA-level control variables
$S C I_{i, t}$	The mean of the Coincident Index for the state where firm i’s inventors are located in the previous 5 years. Data are available from: https://www.philadelphiafed.org/research-and-data/regional-economy/indexes/coincident/.
$C P I_{i, t}$	The mean of the CPI—a measure of cost of living—in the MSAs where firm i’s inventors are located in the previous 5 years. Data are available from the Bureau of Labor Statistics.
$W A G E_{i, t}$	The mean of the weighted-average wages for the workforce in the MSAs where firm i’s inventors are located in the previous 5 years. Wages are obtained from the ACS and weighted by sample weights reported by the IPUMS. The estimated coefficient on this measure is divided by 1000, as presented in the tables.
$L N P O P_{i, t}$	The natural log of the mean of workforce population in the MSAs where firm i’s inventors are located in the previous 5 years. Data are available from the Census Bureau.
$U N E M P_{i, t}$	The mean unemployment rate of MSAs where firm i’s inventors are located in the previous 5 years. Data are available from the Bureau of Labor Statistics.
$M S A_R O A_{i, t}$	Average profitability in MSA, which is equal to the average return on assets for firms in the MSA. The mean of ROA in the MSAs where firm i’s inventors are located in the previous 5 years is then calculated.
$M S A_R O A V O L_{i, t}$	Earnings volatility in MSA, which is computed as the average standard deviation of return on assets for firms in the MSA over the previous 5 years. The mean of the earnings volatility in the MSAs where firm i’s inventors are located in the previous 5 years is then calculated.
$M S A_E M P I N T_{i, t}$	The mean employment intensity in MSAs where firm i’s inventors are located in the previous 5 years. It is the ratio of MSA full-time equivalent employment (in thousands) to urban land areas (in square miles).
$M S A_I N D C O N_{i, t}$	The mean of industry concentration in MSAs where firm i’s inventors are located in the previous 5 years. It is the number of innovative firms in the same industry as firm i scaled by MSA population (in thousands).

MSA: metropolitan statistical area.

Appendix 2

Appendix 3

Appendix 4

Appendix 5

Appendix 6 Acknowledgements

We would like to thank Yaowen Shan (Associate Editor) and two anonymous referees for their helpful comments and suggestions. We also thank S. Ghon Rhee (discussant) and conference participants at the 7th Paris Financial Management Conference.

Final transcript accepted 15 January 2022 by Yaowen Shan (AE Accounting and Auditing).

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Lingwei Li

Notes

References

Acharya

Baghai

Subramanian

(2013) Labor laws and innovation. Journal of Law and Economics 56: 997–1037.

Aghion

Akcigit

Hyytinen

, et al. (2018) The social origins and IQ of inventors. Working paper, Harvard University, Cambridge, MA.

Aghion

Bloom

Blundell

, et al. (2005) Competition and innovation: An inverted-U relationship. The Quarterly Journal of Economics 120: 701–728.

Andersson

Quigley

Wilhelmsson

(2005) Higher education, localization and innovation: Evidence from a natural experiment. Working paper, Royal Institute of Technology, Stockholm.

Bacolod

Blum

Strange

(2010) Elements of skill: Traits, intelligences, education, and agglomeration. Journal of Regional Science 50: 245–280.

Barker

III Mueller

(2002) CEO Characteristics and firm R&D spending. Management Science 48: 782–801.

Beck

Francis

Gunn

(2018) Public company audits and city-specific labor characteristics. Contemporary Accounting Research 35: 394–433.

Bloom

Draca

Van Reenen

(2016) Trade induced technical change? The impact of Chinese imports on innovation, IT and productivity. Review of Economic Studies 83: 87–117.

Call

Campbell

Dhaliwal

, et al. (2017) Employee quality and financial reporting outcomes. Journal of Accounting and Economics 64: 123–149.

10.

Carlino

Kerr

(2015) Agglomeration and innovation. In: Henderson

Strange

(eds) Handbook of Regional and Urban Economics, vol. 5. Amsterdam: Elsevier, pp. 349–404.

11.

Carlino

Chatterjee

Hunt

(2007) Urban density and the rate of invention. Journal of Urban Economics 61: 389–419.

12.

Chang

Low

, et al. (2015) Non-executive employee stock options and corporate innovation. Journal of Financial Economics 115: 168–188.

13.

Chemmanur

Kong

Krishnan

, et al. (2019) Top management human capital, Inventor mobility, and corporate innovation. Journal of Financial and Quantitative Analysis 54: 2383–2422.

14.

Chen

Hsu

, et al. (2016) Be nice to your innovators: Employee treatment and corporate innovation performance. Journal of Corporate Finance 39: 78–98.

15.

Chen

Gao

Luo

, et al. (2020) The effects of rural–urban migration on corporate innovation: Evidence from a natural experiment in China. Financial Management 49: 521–545.

16.

Cohen

Levinthal

(1990) Absorptive capacity: A new perspective on learning and innovation. Administrative Science Quarterly 35: 128–152.

17.

Criado-Perez

Collins

Jackson

(2020) Enablers of evidence-based management: Clues from the absorptive capacity literature. Australian Journal of Management 45: 468–487.

18.

Dass

Nanda

Xiao

(2017) Truncation bias corrections in patent data: Implications for recent research on innovation. Journal of Corporate Finance 44: 353–374.

19.

Fang

Tian

Tice

(2014) Does stock liquidity enhance or impede firm innovation? The Journal of Finance 69: 2085–2125.

20.

Garcia

Norli

(2012) Geographic dispersion and stock returns. Journal of Financial Economics 106: 547–565.

21.

Glaeser

(1999) Learning in cities. Journal of Urban Economics 46: 254–277.

22.

Greer

Lei

(2012) Collaborative innovation with customers: A review of the literature and suggestions for future research. International Journal of Management Reviews 14: 63–84.

23.

Griliches

(1998) R&D and Productivity: The Econometric Evidence. Chicago, IL: University of Chicago Press.

24.

Hall

Jaffe

Trajtenberg

(2001) The NBER Patent Citation Data File: Lessons, Insights and Methodological Tools. Cambridge, MA: National Bureau of Economic Research.

25.

Hall

Jaffe

Trajtenberg

(2005) Market value and patent citations. The RAND Journal of Economics 36: 16–38.

26.

Hargadon

(2002) Brokering knowledge: Linking learning and innovation. Research in Organizational Behavior 24: 41–85.

27.

Hombert

Matray

(2018) Can innovation help U.S. manufacturing firms escape import competition from China? Journal of Finance 73: 2003–2039.

28.

Islam

Zein

(2020) Inventor CEOs. Journal of Financial Economics 135: 505–527.

29.

Israelsen

Yonker

(2017) Key human capital. Journal of Financial and Quantitative Analysis 52: 175–214.

30.

Kantor

Whalley

(2014) Knowledge spillovers from research universities: Evidence from endowment value shocks. Review of Economics and Statistics 96: 171–188.

31.

Kile

Phillips

(2009) Using industry classification codes to sample high-technology firms: Analysis and recommendations. Journal of Accounting, Auditing & Finance 24: 35–58.

32.

Knudsen

Schleimer

(2020) The role of prevailing individual absorptive capacity versus absorptive capacity development for different innovation outcomes. Knowledge Management Research & Practice. Epub ahead of print 7 July. DOI: 10.1080/14778238.2020.1787801.

33.

Kogan

Papanikolaou

Seru

, et al. (2017) Technological innovation, resource allocation, and growth. The Quarterly Journal of Economics 132: 665–712.

34.

Laursen

Salter

(2006) Open for innovation: The role of openness in explaining innovation performance among U.K. manufacturing firms. Strategic Management Journal 27: 131–150.

35.

Leiponen

(2006) Managing knowledge for innovation: The case of business-to-business services. Journal of Product Innovation Management 23: 238–258.

36.

Lai

D’Amour

, et al. (2014) Disambiguation and co-authorship networks of the U.S. patent inventor database (1975–2010). Research Policy 43: 941–955.

37.

Liu

Mao

Tian

(2017) The role of human capital: Evidence from corporate innovation. Working paper, University of Pennsylvania, Philadelphia, PA.

38.

Lychagin

Pinkse

Slade

, et al. (2016) Spillovers in space: Dose geography matter? Journal of Industrial Economics 64: 295–335.

39.

McDonough

III (1993) Faster new product development: Investigating the effects of technology and characteristics of the project leader and team. Journal of Product Innovation Management 10: 241–250.

40.

Manso

(2011) Motivating innovation. Journal of Finance 66: 1823–1860.

41.

Mao

Weathers

(2019) Employee treatment and firm innovation. Journal of Business Finance & Accounting 46: 977–1002.

42.

Matray

(2021) The local innovation spillovers of listed firms. Journal of Financial Economics 141: 395–412.

43.

Miller

Fern

Cardinal

(2007) The use of knowledge for technological innovation within diversified firms. Academy of Management Journal 50: 307–325.

44.

Minbaeva

Pedersen

Björkman

, et al. (2003) MNC knowledge transfer, subsidiary absorptive capacity and HRM. Journal of International Business Studies 34: 586–599.

45.

Moenaert

Souder

Meyer

, et al. (1994) R&D–marketing integration mechanisms, communication flows, and innovation success. Journal of Product Innovation Management 11: 31–45.

46.

Moretti

(2004) Human capital externalities in cities. Handbook of Regional and Urban Economics 4: 2243–2291.

47.

Mukherjee

Singh

Žaldokas

(2017) Do Corporate taxes hinder innovation? Journal of Financial Economics 124: 195–221.

48.

Nieves

Agustín

Javier

(2016) Organizational knowledge and collaborative human resource practices as determinants of innovation. Knowledge Management Research & Practice 14: 237–245.

49.

Pakes

Schankerman

(1984) The rate of obsolescence of patents, research gestation lags, and the private rate of return to research resource. In: Griliches

(ed.) R & D, Patents, and Productivity. Chicago, IL: University of Chicago Press, pp. 73–88.

50.

Porter

(1992) Capital choices: Changing the way America invests in industry. Journal of Applied Corporate Finance 5: 4–16.

51.

Ramdani

Guermat

Mellahi

(2021) The effect of downsizing on innovation outputs: The role of resource slack and constraints. Australian Journal of Management 46: 346–365.

52.

Rauch

(1993) Productivity gains from geographic concentration of human capital: Evidence from the cities. Journal of Urban Economics 34: 380–400.

53.

Roper

Love

(2008) Modelling the innovation value chain. Research Policy 37: 961–977.

54.

Ruggles

Alexander

Genadek

, et al. (2010) Integrated public use microdata series: Version 5.0. Machine-readable Database, University of Minnesota, Minneapolis, MN.

55.

Solow

(1957) Technical change and the aggregate production function. The Review of Economics and Statistics 39: 312–320.

56.

Toivanen

Väänänen

(2016) Education and invention. Review of Economics and Statistics 98: 382–396.

57.

(2017) Absorptive capacity and R&D outsourcing. Journal of Engineering and Technology Management 43: 34–47.

58.

Vega-Jurado

Gutiérrez-Gracia

Fernández-de-Lucio

(2008) Analyzing the determinants of firm’s absorptive capacity: Beyond R&D. R&D Management 38: 392–405.

59.

Vinding

(2006) Absorptive capacity and innovative performance: A human capital approach. Economics of Innovation and New Technology 15: 507–517.

60.

Wolfe

(1994) Organizational innovation: Review, critique and suggested research directions. Journal of Management Studies 31: 405–431.

61.

Zahra

George

(2002) Absorptive capacity: A review, reconceptualization, and extension. Academy of Management Review 27: 185–203.