TMT Family Members’ Education and Firm Innovation: Evidence from Chinese Family Firms

Abstract

This study investigates the effect of the educational level of top management team (TMT) family members on firm innovation among publicly listed family firms in China. Using a panel of 14,338 firm-year observations from 2015 to 2023, this study employs fixed effects regressions to show that the educational background of family members positively influences firm innovation, measured by the proportion of R&D personnel and capitalized R&D expenditures. Moreover, this positive effect is more pronounced under greater industry competition, higher transparency, and smaller firms. The mediation analysis identifies potential channels of asset tangibility, ownership concentration, and management fees through which family education influences firm innovation. Sectoral heterogeneity reveals a more pronounced effect within the manufacturing and service sectors, while no statistically significant relationship emerges in the agriculture sector. Concerns over endogeneity are mitigated using lagged family education, two-stage least squares regressions, and panel vector autoregressions. The baseline result remains robust when firm innovation is alternatively measured by the number of patents. These findings contribute to the literature on innovation in family firms and offer implications for investors, corporate decision-makers, and policymakers in emerging markets.

1. Introduction

Family-owned businesses generate nearly three-quarters of global gross domestic product (GDP), accounting for $60−70 trillion in annual revenue, and employ approximately three-fifths of the global workforce (Sligh, 2024). The 500 largest family-owned firms collectively represent the world’s third-largest economic force when ranked by total revenue (Ernst & Young, 2021). A defining feature of these firms is the persistence of family control across generations, which reflects a long-term orientation in strategic and organizational decision-making (Walsh & Seward, 1990; Zellweger, 2007; Ward, 2011). This characteristic has particular relevance as economies such as China pivot from rapid GDP growth to more sustainable, quality-driven development grounded in technological advancement and industrial upgrading (Betti, 2024). In such dynamic and competitive environments, innovation becomes critical to sustaining firm performance over time (Duran et al., 2016; Qi & Chau, 2018). Family firms—despite their global prominence—face unique governance-related challenges and advantages in pursuing innovation, shaped by their ownership structures and intergenerational priorities.

Despite substantial research on governance and innovation in family firms, limited attention has been paid to the influence of family member education as a determinant of innovation. The extant literature presents mixed findings regarding the impact of family control on R&D investment and innovation outcomes. Some studies find that family firms underinvest in R&D (Block, 2012), potentially due to heightened risk aversion (Chen et al., 2020). And family-owned firms may cultivate self-perceptions in which family-oriented goals serve as negative signals to prospective talented candidates (Basco et al., 2023). In contrast, interview data indicate that higher technological content in products is associated with higher educational attainment among company founders (Dou et al., 2019; Madsen et al., 2003). Moreover, family CEOs with substantive control rights appear more willing to invest in R&D (Zulfiqar et al., 2021). This study contributes to the ongoing debate by explicitly analyzing the impact of family member education on innovation and evaluating the moderating roles of industry competition and information transparency.

Using a dataset comprising 14,338 firm-year observations from 3097 unique Chinese publicly listed family firms between 2015 and 2023, this study presents four key findings. First, top management team (TMT) family members’ education positively influences firm innovation inputs, measured by the proportion of R&D personnel and capitalized R&D investment. Second, the positive association between family member education and firm innovation input is stronger when industry competition intensifies, information transparency is high, or firm size is relatively small. Third, family member education influences firm innovation partly through the channels of reduced asset tangibility, reduced ownership concentration, and increased management fees. Fourth, the positive effect of family member education on firm innovation is primarily observed in the manufacturing and service sectors, with no significant relationship found in the agriculture sector. Finally, endogeneity concerns are alleviated through the use of lagged independent variables, two-stage least squares (2SLS) regressions, and panel vector autoregressions. The results remain robust when firm innovation is alternatively measured by the number of authorized patents.

This study contributes to the literature in several distinct ways. First, it deepens the understanding of the determinants of firm innovation (He & Tian, 2020; Wan et al., 2005). Prior research highlights the role of external conditions, including financial market development (Hsu et al., 2014), government corruption (Ayyagari et al., 2014), government subsidies (Ding & Zhang, 2023), and regional gambling preferences (Chen et al., 2014). Internal mechanisms also matter, such as CEO characteristics (Barker & Mueller, 2002; Li et al., 2023) and corporate taxation (Mukherjee et al., 2017; Atanassov & Liu, 2020). This study fills a gap by identifying the educational background of TMT family members as a novel determinant of innovation in Chinese family firms.

Second, this study advances the literature on the implications of TMT educational background. Prior research links TMT educational background to firm performance (Jalbert et al., 2002; Diaz-Fernandez et al., 2014), corporate risk-taking (Zhang et al., 2023), corporate social responsibility (Wen et al., 2024), default risk (Zheng et al., 2024), entrepreneurial strategic orientation (Yang & Wang, 2014), strategic consensus (Knight et al., 1999), and innovation performance (Xu & Hu, 2023; Daellenbach et al., 1999). This study addresses a gap by documenting firm innovation as a novel outcome of TMT educational background in Chinese family firms.

Third, this study investigates the moderating roles of industry competition and information transparency, thereby offering a more granular understanding of the external contingencies that facilitate or hinder innovation. It further analyzes how the educational backgrounds of family members shape both the composition of R&D personnel and the scale of innovation investment across diverse industries, thereby generating practical implications for corporate managers, institutional investors, and policymakers aiming to develop targeted strategies that enhance innovation capacity in family-controlled enterprises. In addition, this study also sheds light on internal corporate governance mechanisms. The educational background of TMT family members represents an informal governance channel that shapes strategic decision-making and innovation investment, particularly in emerging market family firms. These insights align with the broader discourse on evolving corporate governance structures.

The remainder of the paper is structured as follows. Section 2 reviews the relevant literature and formulates the testable hypotheses. Section 3 outlines the data sources and empirical methodology. Section 4 reports the main empirical findings. Section 5 discusses the results and their implications. Section 6 concludes, noting limitations and directions for future research.

2. Literature Review and Hypotheses Development

2.1. Determinants of Firm Innovation

Firm innovation is influenced by a variety of external factors that shape a firm’s strategic behavior and resource allocation. Prior research has highlighted the significant role of contextual environments in driving innovation outcomes (He & Tian, 2020; Wan et al., 2005). External factors such as financial market development (Hsu et al., 2014), government corruption (Ayyagari et al., 2014), and government support—including subsidies and tax incentives—have been shown to affect firms’ R&D investment and innovation performance (Zuo & Lin, 2022; Xia, 2024; Kang & Park, 2012). Corruption, in particular, may reduce innovation incentives by distorting resource allocation (Huang & Yuan, 2021). In contrast, well-designed government subsidies can reduce innovation costs and stimulate technological advancement (Ding & Zhang, 2023). Political ties also significantly influence firm innovation in China, with central government connections enhancing innovation performance more effectively than local government ties (Kwak et al., 2023). Moreover, social and cultural dimensions such as regional gambling preferences have also been found to correlate with corporate risk-taking and innovation behavior (Chen et al., 2014). Lastly, access to external finance and the alleviation of financing constraints remain critical for innovation, especially for younger and smaller firms (Brown et al., 2009).

Firm innovation is also strongly influenced by internal characteristics, including financial strength, governance structures, and human capital. Greve (2003) adopts a behavioral theory and shows that superior performance reduces both R&D intensity and innovation output. In contrast, profitability stimulates subsequent innovation among firms in high-technological-opportunity industries (Audretsch, 1995). Innovation outcomes are also shaped by firm size and age (Hansen, 1992; Stock et al., 2002). Iqbal et al. (2022) find that financial leverage impedes input-side innovation but enhances output-side innovation, as proxied by patent counts. Using panel data on Chinese listed firms from 2002 to 2011, Rong et al. (2017) document that institutional ownership promotes innovation, particularly under heightened market competition. Meta-analytic evidence suggests that board characteristics—specifically meeting frequency and the proportions of independent and external directors—exhibit the strongest associations with firm innovation (Sierra-Morán et al., 2024). Empirical findings confirm a positive and statistically significant relationship between board diversity and innovation outcomes (Makkonen, 2022). Firms with gender-diverse boards generate more patents, more novel patents, and demonstrate higher innovation efficiency (Griffin et al., 2021). Organizational culture also plays a critical role, with developmental cultures—characterized by flexibility and external orientation—being most conducive to innovation, while hierarchical cultures—marked by control and internal focus—are least aligned with innovative goals (Büschgens et al., 2013). Finally, CEOs’ multicultural backgrounds are also shown to influence firm-level innovation (Li et al., 2023).

2.2. Impacts of TMT Educational Background

Jalbert et al. (2002) document a significant relationship between CEO educational background and firm performance, as measured by both return on assets (ROA) and Tobin’s Q, suggesting that educational attainment may shape managerial ability and strategic decision-making. Drawing on primary data from two Spanish databases, Diaz-Fernandez et al. (2014) find that diversity in the education levels of TMT members exerts a negative and statistically significant effect on corporate performance. Using a sample of firms in the financial sector listed on the Nigerian Stock Exchange, Saidu (2019) reports that CEO educational background contributes positively to firm profitability, implying that higher education enhances managerial competence in complex, regulated environments. In contrast, Wang et al. (2015) show that TMT functional heterogeneity fails to improve firm performance and is significantly negatively associated with short-term outcomes. Focusing on Taiwanese publicly listed firms disclosing corporate social responsibility reports, Lee et al. (2021) find that heterogeneity in TMT educational background has a detrimental impact on firm performance, with the adoption of greenhouse gas emission strategies mediating this relationship.

In addition to affecting firm performance, TMT educational backgrounds also influence corporate risk metrics. Utilizing a large panel of 4681 firm-year observations from 2012 to 2020, Zhang et al. (2023) find that CEO educational background is negatively associated with corporate risk-taking, suggesting that better-educated executives adopt more prudent financial policies. Similarly, Zheng et al. (2024) demonstrate that higher levels of executive education reduce corporate default risk, supporting the notion that educational attainment enhances risk management capabilities. Herrmann and Datta (2005) show that firms with greater international diversification tend to be led by TMTs with more advanced educational qualifications, indicating that global strategic initiatives may require cognitively complex leadership.

Beyond performance and risk, educational attributes also influence firms’ social and strategic orientations. Wen et al. (2024) reveal that CEO gender, age, education, and career experience are positively associated with corporate social responsibility engagement, suggesting that diverse life experiences enrich social awareness and stakeholder responsiveness. Yang and Wang (2014) examine whether educational heterogeneity within TMTs affects firms’ entrepreneurial strategic orientation, finding mixed evidence of its influence on strategic boldness and proactivity. Complementing this view, Knight et al. (1999) report that TMT diversity undermines strategic consensus, implying that although diversity may increase knowledge breadth, it can simultaneously reduce alignment in strategic vision.

2.3. Impacts of TMT Education on Firm Innovation

Based on a sample of 437 publicly listed Taiwanese firms from 2006 to 2012, Kuo et al. (2018) conclude that director education significantly influences firm R&D investment, with higher educational attainment among directors associated with greater R&D intensity. Likewise, Çerez et al. (2024) find that a graduate-level educational average among board members positively affects firm-level R&D expenditure. Analyzing a sample of 1,374 listed Chinese companies, Wang et al. (2019) report that a higher proportion of doctorate-holding directors significantly increases the likelihood of R&D engagement. Firms led by CEOs with advanced science-related degrees also exhibit elevated R&D spending (Barker & Mueller, 2002). Liu et al. (2010) provide evidence that returnee entrepreneurs—those educated abroad—are more innovation-oriented than their domestically educated counterparts. Using firm-level data on Chinese listed firms and TMT scholarly publication records from 2000 to 2021, Xu and Hu (2023) find that the academic competence of TMTs contributes positively to firms’ innovation performance. Daellenbach et al. (1999) report a positive association between the technical orientation of the TMT and elevated levels of R&D intensity relative to industry averages.

TMT educational diversity is also economically relevant. Midavaine et al. (2016) demonstrate that educational diversity within corporate boards contributes positively to R&D investment. Drawing on a sample of 3888 Swedish firms, Mohammadi et al. (2017) argue that higher education diversity within the workforce is linked to enhanced radical innovation outcomes. Bolli et al. (2018) find that vertical educational diversity significantly increases the extensive margin of R&D activity and the incidence of product innovation, particularly with respect to novel product development. Bello-Pintado and Bianchi (2020) document a robust, positive, linear, and statistically significant relationship between horizontal educational diversity and the intensity of technological innovation activities.

Duran et al. (2016) argue that family firms—due to concentrated ownership, heightened control, and the salience of nonfinancial objectives—exhibit greater innovation output relative to nonfamily firms. Block (2012) similarly finds that founder ownership has a positive impact not only on R&D intensity but also on R&D productivity. Consistent with this line of reasoning, the following hypothesis is proposed:

Hypothesis 1:

TMT family members’ education positively impacts Chinese family firms’ innovation.

2.4. Moderating Role of Industry Competition

While TMT educational background consistently promotes firm innovation, recent research emphasizes that industry competition substantially moderates this relationship. Intense rivalry amplifies environmental uncertainty and compels firms to innovate swiftly to maintain competitiveness and ensure survival (Tang et al., 2015). In such high-stakes contexts, TMTs with stronger educational credentials are better positioned to identify nascent technological trajectories, make informed strategic decisions amidst complexity, and manage R&D activities with greater efficiency and foresight (Barker & Mueller, 2002). As a result, the value of human capital embedded in TMT education becomes more salient under conditions of heightened competitive pressure.

Empirical findings support this contingency framework. Basit et al. (2022) demonstrate that intensified foreign competition significantly strengthens the effect of knowledge-based competencies on innovation, particularly within service industries relative to manufacturing. Le (2024) finds that competitive intensity positively moderates the impact of knowledge sharing on product innovation. Cross-sectional evidence further reveals that the positive influence of CEOs’ multicultural backgrounds on innovation output is more pronounced when firms operate in highly competitive industries (Li et al., 2023). Similarly, Singh and Chakraborty (2024) show that family firms increase R&D investment in response to growing domestic product market rivalry. Motivated by the empirical patterns and contextual reasoning outlined above, the following hypothesis is proposed:

Hypothesis 2:

The positive impact of TMT family members’ education on Chinese family firms’ innovation is stronger under greater industry competition.

2.5. Moderating Role of Information Transparency

In Chinese family firms, the educational background of TMT family members equips them with advanced knowledge and strategic capability, both of which are important drivers of innovation. However, whether this potential is realized often depends on the level of information transparency (Zhong, 2018). Information transparency is vital in corporate governance, as it mitigates information asymmetry between insiders and outsiders (Bushman & Smith, 2003; Bhattacharya & Ritter, 1983). When transparency is high—characterized by greater analyst coverage and timely disclosures—external stakeholders such as investors are better able to evaluate the competence and strategic intent of family executives (He & Yin, 2019). In such contexts, the educational attainment of TMT family members is more likely to be perceived as a credible signal of leadership quality, enhancing legitimacy and facilitating access to capital and support for innovation. Conversely, under low transparency, even highly educated TMT members may be perceived as beneficiaries of nepotism, and their innovation-related decisions may be met with skepticism or underinvestment. Therefore, higher information transparency strengthens the positive relationship between TMT family members’ education and firm innovation by improving external perception, resource availability, and decision-making credibility.

The empirical evidence supports the preceding logic. The association between CEOs’ multicultural backgrounds and firm innovation is more pronounced in settings with high, rather than low, information transparency (Li et al., 2023). Transparency positively moderates the link between knowledge sourcing and innovative behavior (Che et al., 2019), as greater openness enhances the internalization of external knowledge into innovation outputs. Shen et al. (2020) show that the academic backgrounds of TMT members mitigate information asymmetry, thereby fostering innovation. Brown and Martinsson (2019) report significantly higher levels of R&D and patenting in more transparent information environments. These effects are most salient in industries dependent on external equity, suggesting that transparency reduces the information costs associated with arm’s-length financing. Wang and Zhang (2023) further argue that transparency-oriented firms exhibit a stronger propensity for innovation. Driven by the empirical patterns and contextual rationale discussed above, the following hypothesis is advanced:

Hypothesis 3:

The positive impact of TMT family members’ education on Chinese family firms’ innovation is stronger under higher levels of information transparency.

3. Data and Methodology

3.1. Methodology

Following Liu et al. (2023), this study measures the dependent variable, firm innovation, along two dimensions: (1) the proportion of R&D personnel to all employees (RDPerson); and (2) the ratio of R&D investment to operating income (RDInvest).1

The key explanatory variable, TMT family members’ educational background (FamilyEdu), is measured as the average education level of all TMT family members within each firm-year. The original data classify education into seven categories: (1) secondary vocational school and below, (2) junior college, (3) bachelor’s degree, (4) master’s degree, (5) doctoral degree, (6) others, and (7) MBA/EMBA. Following Zhang et al. (2023), and recognizing that doctoral degrees typically carry greater academic prestige than MBAs, this study recodes MBA as five, doctoral as six, and excludes a small number of observations classified as “others.”

To test Hypothesis 1, the following fixed-effects regression model is estimated to examine the impact of TMT family members’ education on firm innovation:

$$\begin{array}{l}{\mathit{RD}}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}+{\beta }_4{Leverage}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_5{Tangibility}_{i,t}+{\beta }_6{SOE}_{i,t}+{\beta }_7{SaleGrow}_{i,t}+{\beta }_8{ROA}_{i,t}+{\beta }_9{Loss}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{10}{InstiOwn}_{i,t}+{\beta }_{11}{BoardSize}_{i,t}+{\beta }_{12}{BoardIndep}_{i,t}+{\beta }_{13}{Duality}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{14}{TMTPay}_{i,t}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(1)

where i denotes firm i and t denotes year t. RD refers to either the proportion of R&D personnel among total employees (RDPerson) or the ratio of R&D expenditure to operating revenue (RDInvest). FamilyEdu denotes the average educational level of TMT family members. FirmSize is defined as the natural logarithm of total assets. FirmAge is measured as the natural logarithm of one plus the number of years since the firm’s inception. Leverage is the firm’s total liabilities divided by total assets. Tangibility is the ratio of net fixed assets to total assets. SOE is a dummy variable equal to one if the firm is a state-owned enterprise and zero otherwise. SaleGrow is the percentage growth rate in sales. ROA is the return on assets, defined as net income divided by total assets. Loss is a dummy variable equal to one if the firm reports negative net income and zero otherwise. InstiOwn measures institutional ownership, defined as the shares held by institutional investors divided by total shares outstanding. BoardSize is the natural logarithm of the number of directors on the board. BoardIndep is the proportion of independent directors, calculated as the number of independent directors divided by the total number of directors. Duality is a dummy variable equal to one if the CEO simultaneously serves as board chairman, and zero otherwise. TMTPay is the natural logarithm of total annual compensation paid to the TMT, including directors, supervisors, and senior executives. IndustryFE denotes industry fixed effects, YearFE denotes year fixed effects, and ε is the error term.2 The definitions of all variables are provided in

Table 1. Variable definitions.

Variable	Definition
RDPerson	The proportion of R&D personnel, calculated as the number of R&D employees divided by the total number of employees.
RDInvest	The ratio of R&D expenditure to operating revenue.
FamilyEdu	The average educational level of TMT family members, with individual education level classified as follows: (1) secondary vocational school and below; (2) junior college; (3) bachelor’s degree; (4) master’s degree; (5) MBA/EMBA; (6) doctoral degree.
FirmSize	The natural logarithm of the firm’s total assets.
FirmAge	The natural logarithm of one plus the number of years since the firm’s establishment.
Leverage	The firm’s total liabilities divided by its total assets.
Tangibility	The ratio of net fixed assets to total assets, used to capture the firm’s asset tangibility.
SOE	A dummy variable equal to 1 if the firm is a state-owned enterprise, and 0 otherwise.
SaleGrow	The percentage change in sales revenue from the previous year.
ROA	Return on assets, calculated as net income divided by average total assets.
Loss	A dummy variable equal to 1 if net income is negative, and 0 otherwise.
InstiOwn	The ratio of total shares held by institutional investors to the total number of shares outstanding.
BoardSize	The natural logarithm of the number of board directors.
BoardIndep	The number of independent directors divided by the total number of directors.
Duality	A dummy variable equal to 1 if the chairman and CEO are the same person, and 0 otherwise.
TMTPay	The natural logarithm of total annual salary paid to the top management team (TMT), including directors, supervisors, and senior executives.
HHIRev	Herfindahl-Hirschman Index based on main business revenue, measuring industry concentration by summing the squared market shares of all firms, where each firm’s share is its main business revenue relative to the industry total.
Transparency	Information disclosure rating of a listed firm, based on evaluations by the Shanghai and Shenzhen Stock Exchanges. Ratings are classified as A (Excellent), B (Good), C (Pass), and D (Fail).
Top5	The proportion of shares held by the five largest shareholders to total shares outstanding.
ManageFee	The ratio of management expenses to operating revenue.
NumPatents	The natural logarithm of one plus the number of patents authorized.

.

3.2. Data

The data source is the China Stock Market & Accounting Research (CSMAR) Database. The empirical analysis is limited to publicly listed family firms in China. Data on TMT family members’ educational backgrounds are drawn from the sub-dataset of “Family Firms” in CSMAR. R&D personnel ratios and investment intensity are extracted from the sub-dataset of “Listed Firms’ R&D and Innovation” in CSMAR. After merging the independent and dependent variables, the final sample period spans 2015 to 2023, as observations outside this range account for less than 1% of the data.

Control variables are also retrieved from CSMAR to capture firm characteristics, including firm size, firm age, leverage, asset tangibility, a state-owned enterprise dummy, sales growth, ROA, and a negative net income dummy. Additional control variables include board-level governance features, such as board size, board independence, and CEO duality. Institutional ownership and total TMT compensation are also included as controls.

Special-treatment stocks are excluded from the sample due to their elevated delisting risk. Following the removal of observations with missing data, all variables are winsorized at the 1st and 99th percentiles. The resulting sample includes 14,338 firm-year observations, representing 3097 unique A-share listed Chinese family firms.

For the key explanatory variable, Figure 1 presents the time trend of each educational degree category among Chinese family firms from 2015 to 2023. The number of bachelor’s and master’s degrees increases significantly over time, whereas other degrees remain relatively stable.

4. Results

4.1. Descriptive Statistics

Table 2. Descriptive statistics.

Variable	Obs	Mean	SD	Min	P25	Median	P75	Max
RDPerson	14,338	18.855	13.939	0.710	10.530	14.845	22.980	72.460
RDInvest	14,338	6.487	6.114	0.090	3.230	4.630	7.470	37.020
FamilyEdu	14,338	3.403	1.011	1.000	3.000	3.333	4.000	6.000
FirmSize	14,338	21.842	1.027	19.777	21.115	21.712	22.439	26.440
FirmAge	14,338	2.920	0.296	2.079	2.708	2.944	3.135	3.638
Leverage	14,338	0.354	0.182	0.049	0.205	0.338	0.482	0.924
Tangibility	14,338	0.177	0.123	0.002	0.081	0.157	0.252	0.708
SOE	14,338	0.039	0.193	0.000	0.000	0.000	0.000	1.000
SaleGrow	14,338	0.164	0.361	−0.653	−0.025	0.112	0.278	3.808
ROA	14,338	0.047	0.073	−0.375	0.019	0.049	0.084	0.254
Loss	14,338	0.122	0.327	0.000	0.000	0.000	0.000	1.000
InstiOwn	14,338	0.326	0.242	0.001	0.114	0.277	0.513	0.920
BoardSize	14,338	2.058	0.189	1.609	1.946	2.079	2.197	2.708
BoardIndep	14,338	38.116	5.239	28.570	33.330	37.500	42.860	60.000
Duality	14,338	0.447	0.497	0.000	0.000	0.000	1.000	1.000
TMTPay	14,338	15.464	0.650	13.357	15.027	15.433	15.860	17.573

Note: This table reports descriptive statistics for all variables. The variable definitions are provided in Table 1. The sample includes 14,338 firm-year observations from 3097 unique firms over the period spanning 2015 to 2023.

presents descriptive statistics for the variables. The mean value of RDPerson is 18.855%, indicating that family firms place considerable emphasis on R&D personnel. The standard deviation of RDPerson is 13.939%, suggesting substantial variation in the deployment of R&D staff. Similarly, RDInvest has a mean of 6.487% and a standard deviation of 6.114%, also reflecting considerable heterogeneity. The average FamilyEdu score is 3.403, consistent with Figure 1, which shows that most TMT family members hold either a bachelor’s degree (FamilyEdu = 3) or a master’s degree (FamilyEdu = 4). Regarding firm characteristics, the average log of firm size is 21.842, and the mean log of firm age is 2.920. The mean leverage ratio is 35.4%, while the average asset tangibility is 17.7%. The mean value of the SOE dummy is 3.9%, implying that 96.1% of family firms are non-state-owned, as expected. The average sales growth rate and ROA are 16.4% and 4.7%, respectively, suggesting moderate profitability and strong revenue growth. Negative earnings are observed in 12.2% of firm-years. Institutional ownership averages 32.6%. The mean log of board size is 2.058, with independent directors comprising 38.116%. Finally, 44.7% of CEOs concurrently serve as board chairmen, and average logged TMT pay is 15.464.

Table 3. Pairwise correlations.

	RDPerson	RDInvest	FamilyEdu	FirmSize	FirmAge	Leverage	Tangibility	SOE	SaleGrow	ROA	Loss	InstiOwn	BoardSize	BoardIndep	Duality	TMTPay
RDPerson	1
RDInvest	0.639 ***	1
FamilyEdu	0.160 ***	0.183 ***	1
FirmSize	−0.135 ***	−0.167 ***	0.101 ***	1
FirmAge	−0.102 ***	−0.083 ***	−0.048 ***	0.139 ***	1
Leverage	−0.203 ***	−0.273 ***	0.027 ***	0.463 ***	0.116 ***	1
Tangibility	−0.357 ***	−0.223 ***	−0.097 ***	0.080 ***	0.091 ***	0.129 ***	1
SOE	0.015 *	−0.038 ***	0.004	−0.003	−0.022 ***	0.061 ***	−0.036 ***	1
SaleGrow	0.008	−0.088 ***	0.019 **	0.102 ***	−0.103 ***	0.072 ***	−0.031 ***	0.028 ***	1
ROA	−0.030 ***	−0.137 ***	−0.035 ***	−0.005	−0.086 ***	−0.360 ***	−0.061 ***	−0.062 ***	0.302 ***	1
Loss	0.054 ***	0.172 ***	0.046 ***	−0.021 **	0.078 ***	0.212 ***	0.035 ***	0.009	−0.218 ***	−0.691 ***	1
InstiOwn	−0.091 ***	−0.047 ***	0.063 ***	0.265 ***	−0.047 ***	0.045 ***	0.087 ***	−0.021 **	0.069 ***	0.140 ***	−0.087 ***	1
BoardSize	−0.073 ***	−0.046 ***	0.047 ***	0.160 ***	0.020 **	0.053 ***	0.042 ***	0.044 ***	0.034 ***	0.023 ***	−0.033 ***	0.096 ***	1
BoardIndep	0.047 ***	0.037 ***	−0.023 ***	−0.054 ***	0.014 *	−0.003	−0.012	−0.034 ***	−0.013	−0.009	0.025 ***	−0.060 ***	−0.645 ***	1
Duality	0.067 ***	0.087 ***	−0.001	−0.106 ***	−0.065 ***	−0.054 ***	−0.046 ***	−0.031 ***	0.008	0.020 **	0.004	−0.032 ***	−0.091 ***	0.098 ***	1
TMTPay	0.041 ***	0.098 ***	0.128 ***	0.513 ***	0.115 ***	0.127 ***	−0.039 ***	−0.060 ***	0.046 ***	0.132 ***	−0.043 ***	0.206 ***	0.186 ***	−0.099 ***	−0.072 ***	1

Note: This table reports pairwise correlations among the variables. The variable definitions are provided in Table 1. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

presents pairwise correlations among the variables. The results indicate that RDPerson and RDInvest are significantly positively correlated with FamilyEdu at the 1% level, suggesting that higher levels of family education enhance firm innovation. Among the control variables, firm size, firm age, leverage, asset tangibility, ROA, institutional ownership, and board size exhibit significant negative correlations with firm innovation at the 1% level. In contrast, the negative earnings dummy, board independence, CEO duality dummy, and TMT pay are significantly positively associated with firm innovation at the 1% level. Most explanatory variables display correlation coefficients below 0.5, implying that multicollinearity is unlikely to pose a concern.

To further examine the issue of multicollinearity, the variance inflation factor (VIF) test is performed, with results presented in

Table 4. Multicollinearity tests.

	VIF Values
FamilyEdu	1.04
FirmSize	1.87
FirmAge	1.07
Leverage	1.60
Tangibility	1.06
SOE	1.02
SaleGrow	1.17
ROA	2.39
Loss	1.95
InstiOwn	1.13
BoardSize	1.78
BoardIndep	1.73
Duality	1.03
TMTPay	1.48

Note: This table reports the VIF values to assess multicollinearity among the independent variables used in the regression models.

. The VIF test quantifies the extent to which the variance of a coefficient is inflated due to linear dependence among regressors. According to Hair et al. (2013), a VIF value exceeding 10 indicates severe multicollinearity that may bias estimation outcomes. The VIF values reported in Table 4 are well below this threshold, suggesting that multicollinearity is not a significant concern.

4.2. Baseline Regressions

Table 5. Baseline regressions.

	(1)	(2)	(3)	(4)
	RDPerson	RDPerson	RDInvest	RDInvest
FamilyEdu	1.357 ***	1.242 ***	0.737 ***	0.678 ***
	(0.097)	(0.093)	(0.049)	(0.045)
FirmSize		−0.212 *		−0.377 ***
		(0.127)		(0.054)
FirmAge		−1.992 ***		−1.469 ***
		(0.352)		(0.157)
Leverage		−10.711 ***		−8.905 ***
		(0.720)		(0.356)
Tangibility		−19.108 ***		−2.747 ***
		(0.808)		(0.394)
SOE		0.479		−0.699 ***
		(0.483)		(0.179)
SaleGrow		1.255 ***		0.057
		(0.296)		(0.152)
ROA		−11.951 ***		−17.171 ***
		(2.222)		(1.192)
Loss		1.479 ***		1.431 ***
		(0.404)		(0.204)
InstiOwn		−0.530		0.551 ***
		(0.401)		(0.175)
BoardSize		−1.854 **		0.241
		(0.741)		(0.296)
BoardIndep		0.039		0.048 ***
		(0.025)		(0.010)
Duality		0.752 ***		0.591 ***
		(0.187)		(0.082)
TMTPay		1.530 ***		1.651 ***
		(0.181)		(0.090)
Constant	14.236 ***	10.887 ***	3.978 ***	−7.317 ***
	(0.331)	(3.325)	(0.161)	(1.403)
Industry FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	14,338	14,338	14,338	14,338
Adjusted R2	0.363	0.409	0.312	0.417

Note: This table reports the results of univariate and multivariate fixed-effects regressions of firm innovation on TMT family members’ education. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

reports the results of univariate and multivariate fixed-effects regressions of firm innovation on the education level of TMT family members, as specified in Equation (1).3 The estimates reveal that TMT family member education exerts a significantly positive effect on firm innovation at the 1% level, indicating that firms with more highly educated TMT family members allocate greater resources to R&D personnel and expenditures. This evidence supports Hypothesis 1 and aligns with recent findings that more educated directors are perceived as more committed to R&D (Kuo et al., 2018; Çerez et al., 2024; Wang et al., 2019). The coefficient on family education is also economically significant. For example, based on the multivariate result in Column 2, a one-standard-deviation increase in family education corresponds to a 1.256% (=1.242 × 1.011) increase in RDPerson, which accounts for 9% (=1.256/13.939) of the standard deviation of RDPerson. Similarly, based on Column 4, a one-standard-deviation rise in family education corresponds to a 0.685% (=0.678 × 1.011) increase in RDInvest, representing 11% (=0.685/6.114) of the standard deviation of RDInvest.

Regarding the control variables, firm size and firm age display a significant negative relationship with firm innovation, implying that smaller and younger firms engage more heavily in R&D activities. As expected, leverage and asset tangibility are negatively related to innovation, suggesting that lower leverage and greater liquidity foster R&D activities. ROA shows a negative coefficient, while the Loss dummy yields a positive coefficient, indicating that more profitable firms allocate less to R&D activities. CEO-chairman duality and TMT compensation are both positively associated with firm innovation.

4.3. Moderating Effect of Industry Competition

This section investigates whether industry competition moderates the relationship between TMT family members’ education and firm innovation. The Herfindahl-Hirschman Index (HHI), a widely adopted metric of industry concentration, is computed as the sum of squared market shares of all firms in a given industry. A higher HHI reflects greater market concentration and lower competitive intensity (Rhoades, 1993). HHI data are obtained from CSMAR, where HHIRev measures industry concentration by summing the squared revenue shares of firms. Following Hu et al. (2023) and Li et al. (2023), this study incorporates HHIRev and its interaction with family education into the regression model:

$$\begin{array}{l}{RD}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t}\times {HHIRev}_{i,t}+{\beta }_2{FamilyEdu}_{i,t}+{\beta }_3{HHIRev}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_4{FirmSize}_{i,t}+{\beta }_5{FirmAge}_{i,t}+{\beta }_6{Leverage}_{i,t}+{\beta }_7{Tangibility}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_8{SOE}_{i,t}+{\beta }_9{SaleGrow}_{i,t}+{\beta }_{10}{ROA}_{i,t}+{\beta }_{11}{Loss}_{i,t}+{\beta }_{12}{InstiOwn}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{13}{BoardSize}_{i,t}+{\beta }_{14}{BoardIndep}_{i,t}+{\beta }_{15}{Duality}_{i,t}+{\beta }_{16}{TMTPay}_{i,t}\hspace{0.33em}\\ \hspace{1em}\hspace{1em}\hspace{1em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(2)

Table 6. Moderating effect of industry competition.

	(1)	(2)
	RDPerson	RDInvest
FamilyEdu × HHIRev	−4.310 ***	−1.075 ***
	(0.935)	(0.348)
FamilyEdu	1.678 ***	0.791 ***
	(0.134)	(0.064)
HHIRev	17.171 ***	5.301 ***
	(3.511)	(1.258)
FirmSize	−0.222 *	−0.389 ***
	(0.128)	(0.055)
FirmAge	−2.009 ***	−1.476 ***
	(0.353)	(0.158)
Leverage	−10.891 ***	−8.927 ***
	(0.725)	(0.359)
Tangibility	−19.224 ***	−2.803 ***
	(0.811)	(0.396)
SOE	0.590	−0.676 ***
	(0.480)	(0.180)
SaleGrow	1.297 ***	0.066
	(0.298)	(0.152)
ROA	−12.607 ***	−17.406 ***
	(2.248)	(1.208)
Loss	1.401 ***	1.392 ***
	(0.406)	(0.205)
InstiOwn	−0.524	0.581 ***
	(0.402)	(0.176)
BoardSize	−1.778 **	0.294
	(0.743)	(0.297)
BoardIndep	0.046 *	0.049 ***
	(0.025)	(0.010)
Duality	0.750 ***	0.583 ***
	(0.187)	(0.082)
TMTPay	1.579 ***	1.645 ***
	(0.182)	(0.091)
Constant	8.401 **	−7.605 ***
	(3.342)	(1.418)
Industry FE	Yes	Yes
Year FE	Yes	Yes
Observations	14,223	14,223
Adjusted R2	0.408	0.417

Note: This table presents fixed-effects regression results on the effect of industry competition on the relationship between firm innovation and TMT family members’ education. Note that HHIRev denotes the Herfindahl-Hirschman Index of industry concentration, which is inversely related to the industry competition measure. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

reports the results on the moderating effect of industry competition. The interaction term coefficients are negative and statistically significant at the 1% level. As industry concentration (HHIRev) decreases—indicating heightened industry competition—the positive effect of TMT family member education on firm innovation becomes more pronounced. These findings lend support to Hypothesis 2. The coefficient on FamilyEdu remains positive and statistically significant. The coefficients on control variables and the adjusted R-squared are comparable to those reported in the baseline regression in Table 5.

4.4. Moderating Effect of Information Transparency

This section examines how information transparency moderates the relationship between TMT family members’ education and firm innovation (Li et al., 2023). Firm-level transparency data are obtained from CSMAR, where companies are categorized into four transparency levels: A (Excellent), B (Good), C (Pass), and D (Fail). The sample is partitioned based on contemporaneous transparency ratings, and the fixed effects regressions specified in Equation (1) are re-estimated for each subsample.

Table 7. Moderating effect of information transparency.

	(1)	(2)	(3)	(4)
	Transparency = D (Fail)	Transparency = C (Pass)	Transparency = B (Good)	Transparency = A (Excellent)
	RDPerson	RDPerson	RDPerson	RDPerson
FamilyEdu	0.386	0.593 *	1.183 ***	1.544 ***
	(0.942)	(0.327)	(0.115)	(0.243)
FirmSize	−1.457	−1.321 ***	−0.259	−1.022 ***
	(1.230)	(0.396)	(0.161)	(0.346)
FirmAge	−8.737 **	−2.062	−1.740 ***	−0.571
	(4.325)	(1.269)	(0.420)	(1.015)
Leverage	1.924	−8.733 ***	−11.020 ***	−6.981 ***
	(5.671)	(2.020)	(0.886)	(2.277)
Tangibility	−20.516 **	−20.005 ***	−18.342 ***	−24.525 ***
	(9.303)	(2.538)	(0.975)	(2.311)
SOE	−3.259	1.575	0.139	4.883 **
	(3.315)	(1.235)	(0.543)	(2.035)
SaleGrow	−3.277	0.926	1.023 ***	0.798
	(2.433)	(0.863)	(0.361)	(0.773)
ROA	2.388	−4.058	−16.177 ***	−15.529 **
	(11.177)	(5.204)	(3.025)	(6.047)
Loss	−2.500	1.329	1.594 ***	3.667
	(2.345)	(0.967)	(0.494)	(3.117)
InstiOwn	−2.998	−1.069	−0.699	0.545
	(5.770)	(1.435)	(0.479)	(1.010)
BoardSize	5.127	−0.555	−1.262	−4.205 **
	(5.122)	(2.376)	(0.907)	(2.057)
BoardIndep	0.087	0.122	0.080 **	−0.176 ***
	(0.261)	(0.082)	(0.031)	(0.057)
Duality	5.197 **	1.236 *	0.423 *	1.474 ***
	(2.138)	(0.656)	(0.224)	(0.525)
TMTPay	0.213	0.848	1.360 ***	2.327 ***
	(1.866)	(0.641)	(0.223)	(0.472)
Constant	60.355 *	41.460 ***	11.424 ***	24.657 ***
	(32.738)	(12.020)	(4.347)	(9.188)
Industry FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	190	1384	9717	2211
Adjusted R2	0.298	0.408	0.404	0.488

Note: This table presents fixed-effects regression results on the effect of information transparency on the relationship between firm innovation and TMT family members’ education. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

presents the results. As transparency improves, the coefficient on FamilyEdu increases progressively from 0.386 to 0.593, 1.183, and 1.544. Statistical significance also strengthens, rising from insignificance to the 10% level and then to the 1% level. Note that the coefficient on FamilyEdu is only marginally significant at the 10% level when transparency equals C (Pass). TMT family members’ education exerts a statistically significant positive effect on RDPerson under high information transparency (excellent or good), but not under lower transparency (pass or fail). These findings support Hypothesis 3, which posits that the positive effect of TMT family member education on firm innovation is amplified under greater information transparency.

4.5. Moderating Effect of Firm Size

This section examines whether firm size moderates the relationship between TMT family members’ education and firm innovation (K. Wang et al., 2024). Firms are classified into small and large-firm subsamples each year based on the median firm size. The fixed effects regressions specified in Equation (1) are then re-estimated for each subsample.

Table 8. Moderating effect of firm size.

	(1)	(2)	(3)	(4)
	Small Firms	Large Firms	Small Firms	Large Firms
	RDPerson	RDPerson	RDInvest	RDInvest
FamilyEdu	1.380 ***	0.999 ***	0.809 ***	0.519 ***
	(0.135)	(0.129)	(0.063)	(0.064)
FirmSize	0.041	−0.469 **	−0.596 ***	−0.374 ***
	(0.330)	(0.187)	(0.148)	(0.076)
FirmAge	−2.230 ***	−1.864 ***	−1.465 ***	−1.609 ***
	(0.505)	(0.497)	(0.225)	(0.224)
Leverage	−10.445 ***	−11.072 ***	−9.984 ***	−7.883 ***
	(1.040)	(1.011)	(0.502)	(0.515)
Tangibility	−20.058 ***	−18.990 ***	−2.606 ***	−3.645 ***
	(1.166)	(1.143)	(0.554)	(0.585)
SOE	0.408	1.086	−0.733 **	−0.607 ***
	(0.692)	(0.686)	(0.297)	(0.228)
SaleGrow	1.960 ***	0.649 *	0.267	−0.155
	(0.499)	(0.362)	(0.251)	(0.189)
ROA	−19.709 ***	−4.186	−19.108 ***	−15.166 ***
	(3.222)	(3.082)	(1.729)	(1.668)
Loss	1.525 **	1.277 **	2.281 ***	0.655 ***
	(0.611)	(0.524)	(0.320)	(0.254)
InstiOwn	0.695	−1.540 ***	−0.064	1.294 ***
	(0.585)	(0.545)	(0.245)	(0.260)
BoardSize	−1.411	−3.303 ***	0.628	−0.139
	(1.065)	(1.010)	(0.421)	(0.422)
BoardIndep	0.106 ***	−0.066 **	0.044 ***	0.052 ***
	(0.036)	(0.033)	(0.014)	(0.014)
Duality	0.624 **	0.879 ***	0.635 ***	0.607 ***
	(0.267)	(0.263)	(0.119)	(0.114)
TMTPay	1.792 ***	1.411 ***	2.304 ***	1.187 ***
	(0.291)	(0.234)	(0.141)	(0.118)
Constant	−0.838	25.529 ***	−13.004 ***	0.695
	(6.908)	(4.910)	(3.223)	(1.995)
Industry FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	7168	7167	7168	7167
Adjusted R2	0.375	0.456	0.416	0.414

Note: This table presents fixed-effects regression results on the effect of firm size on the relationship between firm innovation and TMT family members’ education. Firms are classified into small and large subsamples each year based on the annual median firm size. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

reports the results. As firm size increases, the coefficient on FamilyEdu declines from 1.380 to 0.999 for RDPerson, and from 0.809 to 0.519 for RDInvest. Statistical significance remains at the 1% level. These results indicate that the positive effect of TMT family education is stronger among smaller firms.

4.6. Mediation Analysis

To better understand the mechanism through which the education of TMT family members influences firm innovation, a three-step regression approach is employed to examine the mediating effect (Baron & Kenny, 1986):

$$\begin{array}{l}{RDPerson}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}+{\beta }_4{Leverage}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+{\beta }_5{SOE}_{i,t}+{\beta }_6{SaleGrow}_{i,t}+{\beta }_7{ROA}_{i,t}+{\beta }_8{Loss}_{i,t}+{\beta }_9{InstiOwn}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+{\beta }_{10}{BoardSize}_{i,t}+{\beta }_{11}{BoardIndep}_{i,t}+{\beta }_{12}{Duality}_{i,t}+{\beta }_{13}{TMTPay}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(3)

$$\begin{array}{l}{Mediator}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}+{\beta }_4{Leverage}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}+{\beta }_5{SOE}_{i,t}+{\beta }_6{SaleGrow}_{i,t}+{\beta }_7{ROA}_{i,t}+{\beta }_8{Loss}_{i,t}+{\beta }_9{InstiOwn}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}+{\beta }_{10}{BoardSize}_{i,t}+{\beta }_{11}{BoardIndep}_{i,t}+{\beta }_{12}{Duality}_{i,t}+{\beta }_{13}{TMTPay}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(4)

$$\begin{array}{l}{RDPerson}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t}+{\beta }_2{Tangibility}_{i,t}+{\beta }_{3}Top{5}_{i,t}+{\beta }_4{ManageFee}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+{\beta }_5{FirmSize}_{i,t}+{\beta }_6{FirmAge}_{i,t}+{\beta }_7{Leverage}_{i,t}+{\beta }_8{SOE}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+{\beta }_9{SaleGrow}_{i,t}+{\beta }_{10}{ROA}_{i,t}+{\beta }_{11}{Loss}_{i,t}+{\beta }_{12}{InstiOwn}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+{\beta }_{13}{BoardSize}_{i,t}+{\beta }_{14}{BoardIndep}_{i,t}+{\beta }_{15}{Duality}_{i,t}+{\beta }_{16}{TMTPay}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.33em}\hspace{0.33em}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(5)

where Mediator denotes asset tangibility (Tangibility), total ownership by the five largest shareholders (Top5), or the management fee ratio (ManageFee), respectively.

Table 9. Mediation analysis.

	(1)	(2)	(3)	(4)	(5)
	RDPerson	Tangibility	Top5	ManageFee	RDPerson
FamilyEdu	1.356 ***	−0.006 ***	−0.017 ***	0.004 ***	0.992 ***
	(0.095)	(0.001)	(0.001)	(0.000)	(0.093)
Tangibility					−19.607 ***
					(0.799)
Top5					−7.725 ***
					(0.780)
ManageFee					30.910 ***
					(2.359)
FirmSize	−0.145	−0.003 ***	−0.030 ***	−0.011 ***	−0.104
	(0.129)	(0.001)	(0.001)	(0.001)	(0.130)
FirmAge	−2.478 ***	0.025 ***	−0.033 ***	−0.002	−2.180 ***
	(0.362)	(0.003)	(0.004)	(0.002)	(0.346)
Leverage	−12.323 ***	0.084 ***	−0.033 ***	−0.065 ***	−8.928 ***
	(0.736)	(0.006)	(0.008)	(0.004)	(0.723)
SOE	0.725	−0.013 ***	−0.010 **	−0.004	0.518
	(0.492)	(0.004)	(0.005)	(0.003)	(0.469)
SaleGrow	1.395 ***	−0.007 ***	−0.012 ***	−0.006 ***	1.343 ***
	(0.301)	(0.003)	(0.003)	(0.002)	(0.293)
ROA	−9.754 ***	−0.115 ***	0.402 ***	−0.257 ***	−0.956
	(2.260)	(0.019)	(0.022)	(0.012)	(2.270)
Loss	1.479 ***	0.000	−0.007 *	0.024 ***	0.674 *
	(0.411)	(0.004)	(0.004)	(0.002)	(0.401)
InstiOwn	−1.013 **	0.025 ***	0.238 ***	−0.006 ***	1.516 ***
	(0.408)	(0.004)	(0.004)	(0.002)	(0.443)
BoardSize	−2.330 ***	0.025 ***	−0.053 ***	0.003	−2.332 ***
	(0.758)	(0.006)	(0.007)	(0.003)	(0.737)
BoardIndep	0.019	0.001 ***	0.001 ***	0.001 ***	0.024
	(0.025)	(0.000)	(0.000)	(0.000)	(0.024)
Duality	0.885 ***	−0.007 ***	0.010 ***	0.000	0.818 ***
	(0.189)	(0.002)	(0.002)	(0.001)	(0.184)
TMTPay	1.608 ***	−0.004 **	0.003	0.014 ***	1.119 ***
	(0.184)	(0.002)	(0.002)	(0.001)	(0.181)
Constant	8.165 **	0.142 ***	1.304 ***	0.111 ***	17.605 ***
	(3.364)	(0.031)	(0.037)	(0.015)	(3.488)
Industry FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Observations	14,338	14,338	14,338	14,338	14,338
Adjusted R2	0.390	0.331	0.319	0.408	0.426

Note: This table reports fixed effects regression results examining the mediating roles of asset tangibility (Tangibility), ownership concentration (Top5), and management expense ratio (ManageFee) in the relationship between TMT family members’ education and firm innovation. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

presents the mediation analysis based on the three-step approach. In Column 1, the coefficient on FamilyEdu is 1.356 and significant at the 1% level, capturing the total effect on firm innovation (RDPerson). Columns 2−4 show that FamilyEdu significantly decreases asset tangibility (Tangibility, −0.006) and ownership concentration (Top5, −0.017), while increasing the management fee ratio (ManageFee, 0.004), with all effects significant at the 1% level. Column 5 indicates that these mediators are each significantly associated with innovation at the 1% level. The estimated indirect effects suggest that FamilyEdu fosters innovation by reducing asset tangibility (−0.006 × −19.607/1.356 = 8.7% of the total effect), lowering ownership concentration (−0.017 × −7.725/1.356 = 9.7%), and increasing the management fee ratio (0.004 × 30.910/1.356 = 9.1%). Collectively, these three indirect channels suggestively explain approximately 27.5% of the total effect.

The mediation analysis identifies three significant channels through which TMT family members’ education influences firm innovation. First, Tangibility accounts for 8.7% of the total effect and is negatively associated with education, indicating that more highly educated family executives reduce the proportion of tangible assets. This shift reflects a greater allocation of resources toward intangible assets such as R&D, patents, and brands, which are critical for innovation. By prioritizing knowledge- and technology-based investments, highly educated executives foster conditions conducive to innovation. Second, Top5 (ownership concentration of the five largest shareholders) accounts for 9.7% of the total effect and is also negatively related to education, suggesting that higher-educated family executives are linked to more dispersed ownership structures. Lower ownership concentration can lessen the influence of large shareholders’ potentially conservative or risk-averse preferences, thereby allowing greater managerial discretion, fostering more independent decision-making, and enabling the firm to pursue innovative projects with greater strategic flexibility. Third, ManageFee (management fee ratio) accounts for 9.1% of the total effect and is positively associated with education, implying that highly educated family executives allocate more resources to administrative, R&D management, and marketing activities. Such expenditures often reflect the operational costs of innovation-related initiatives, thereby directly supporting higher innovation output.

Finally, it is important to note that while the mediation analysis provides insight into potential pathways through which family education influences innovation, the results should be viewed as suggestive rather than conclusive. Given the observational design, mediation estimates cannot fully establish causality, but they highlight plausible mechanisms that complement the baseline findings.

4.7. Different Industries

Chinese firms are broadly classified into three main sectors: primary industry (agriculture), secondary industry (manufacturing), and tertiary industry (services). Following Liu et al. (2024), the sample is classified into three industry subsamples and Equation (1) is re-estimated to examine industry effects. The result, presented in

Table 10. Different industries.

	(1)	(2)	(3)
	Agriculture Sector	Manufacturing Sector	Service Sector
	RDPerson	RDPerson	RDPerson
FamilyEdu	−0.384	1.342 ***	0.862 **
	(0.589)	(0.089)	(0.336)
FirmSize	−0.922	−0.760 ***	1.443 ***
	(0.716)	(0.126)	(0.355)
FirmAge	−4.420	−1.646 ***	−2.556 **
	(3.623)	(0.351)	(1.080)
Leverage	−0.183	−8.159 ***	−19.713 ***
	(3.840)	(0.743)	(1.895)
Tangibility	−14.267 ***	−22.125 ***	−3.748
	(3.606)	(0.819)	(2.819)
SOE	8.852 ***	−0.777 *	3.855 ***
	(1.936)	(0.449)	(1.260)
SaleGrow	−1.057	1.324 ***	0.509
	(1.617)	(0.309)	(0.757)
ROA	7.237	−16.965 ***	2.712
	(8.569)	(2.248)	(5.938)
Loss	5.014 ***	1.000 **	3.459 ***
	(1.637)	(0.411)	(1.129)
InstiOwn	5.868 **	−0.667 *	0.843
	(2.484)	(0.381)	(1.519)
BoardSize	12.756 ***	−2.334 ***	−0.438
	(4.788)	(0.730)	(2.263)
BoardIndep	0.281	0.009	0.130 *
	(0.188)	(0.024)	(0.076)
Duality	−0.744	0.970 ***	−0.122
	(1.362)	(0.178)	(0.606)
TMTPay	−0.349	1.958 ***	0.311
	(0.786)	(0.177)	(0.584)
Constant	10.854	15.617 ***	−2.619
	(15.586)	(3.189)	(10.598)
Industry FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	126	11,315	2897
Adjusted R2	0.522	0.296	0.429

Note: This table reports fixed effects regression results of firm innovation on TMT family members’ education across three industries. The 2012 CSRC industry classification codes are employed to classify firms into three broad sectors: Agriculture (first letter = A), Manufacturing (first letter = B/C/D), and Service sectors (all other cases). Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

, indicate that the coefficient on FamilyEdu loses significance within the agriculture sector. In contrast, the coefficients on FamilyEdu remain positive and statistically significant at the 1% level in the manufacturing sector and at the 5% level in the service sector. The magnitude of the FamilyEdu coefficient is greater in the manufacturing sector than in the service sector. However, the adjusted R-squared in the service sector exceeds that of the manufacturing sector. Collectively, these findings suggest that the positive effect of TMT family members’ education on firm innovation is present in the manufacturing and service sectors but is not evident in the agriculture sector.

4.8. Lagged Independent Variable

Endogeneity concerns arise in the baseline regression specified in Equation (1). Reverse causality is plausible, whereby firms with higher innovation may attract TMT family members with higher educational attainment. Alternatively, unobserved third-party factors may simultaneously influence both TMT family member education and firm innovation, resulting in a spurious positive association. Following Liu et al. (2024) and L. Wang et al. (2024), this concern is addressed by re-estimating Equation (1) using a one-year lag of the independent variable. If the coefficient on the lagged independent variable remains significant, it alleviates concerns about reverse causality. The following regression specification is employed:

$$\begin{array}{l}{RD}_{i,t}={\beta }_0+{\beta }_1{FamilyEdu}_{i,t-1}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}+{\beta }_4{Leverage}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_5{Tangibility}_{i,t}+{\beta }_6{SOE}_{i,t}+{\beta }_7{SaleGrow}_{i,t}+{\beta }_8{ROA}_{i,t}+{\beta }_9{Loss}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{10}{InstiOwn}_{i,t}+{\beta }_{11}{BoardSize}_{i,t}+{\beta }_{12}{BoardIndep}_{i,t}+{\beta }_{13}{Duality}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{14}{TMTPay}_{i,t}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(6)

Consistent with expectations, the results presented in

Table 11. Lagged independent variable.

	(1)	(2)
	RDPerson	RDInvest
Lagged FamilyEdu	1.192 ***	0.694 ***
	(0.107)	(0.051)
FirmSize	−0.404 ***	−0.431 ***
	(0.145)	(0.062)
FirmAge	−1.628 ***	−1.467 ***
	(0.417)	(0.190)
Leverage	−10.024 ***	−8.709 ***
	(0.825)	(0.387)
Tangibility	−19.224 ***	−2.659 ***
	(0.940)	(0.450)
SOE	0.731	−0.768 ***
	(0.576)	(0.206)
SaleGrow	0.897 ***	−0.362 **
	(0.332)	(0.169)
ROA	−8.465 ***	−14.139 ***
	(2.531)	(1.277)
Loss	1.418 ***	1.366 ***
	(0.442)	(0.218)
InstiOwn	−0.488	0.465 **
	(0.466)	(0.207)
BoardSize	−1.832 **	0.425
	(0.871)	(0.346)
BoardIndep	0.044	0.061 ***
	(0.029)	(0.012)
Duality	0.848 ***	0.574 ***
	(0.214)	(0.095)
TMTPay	1.431 ***	1.561 ***
	(0.209)	(0.104)
Constant	15.490 ***	−5.636 ***
	(3.926)	(1.638)
Industry FE	Yes	Yes
Year FE	Yes	Yes
Observations	10,979	10,979
Adjusted R2	0.413	0.413

Note: This table reports the results of fixed-effects regressions of firm innovation on one-year-lagged TMT family members’ education. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

indicate that the one-year-lagged FamilyEdu exerts a significantly positive effect on firm innovation. The coefficients on lagged FamilyEdu are 1.192 and 0.694, closely aligned with the 1.242 and 0.678 reported in Table 5 for RDPerson and RDInvest, respectively. And both coefficients are statistically significant at the 1% level. The adjusted R-squared values are 0.413 and 0.413, also comparable to the 0.409 and 0.417 in Table 5 for RDPerson and RDInvest, respectively. Overall, the findings in Table 11 lend support to Hypothesis 1, suggesting that the education of TMT family members positively influences firm innovation rather than reflecting reverse causality.

4.9. Two-Stage Least Squares Regressions

To further address the endogeneity issue, a 2SLS regression is implemented. Consistent with Liu et al. (2024) and Hu et al. (2023), the annual industry-average value of the independent variable (FamilyEdu) is employed as the instrumental variable.

$$\begin{array}{l}{FamilyEdu}_{i,t}={\beta }_0+{\beta }_1{MeanFamilyEdu}_{i,t}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.17em}+{\beta }_4{Leverage}_{i,t}+{\beta }_5{Tangibility}_{i,t}+{\beta }_6{SOE}_{i,t}+{\beta }_7{SaleGrow}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.17em}+{\beta }_8{ROA}_{i,t}+{\beta }_9{Loss}_{i,t}+{\beta }_{10}{InstiOwn}_{i,t}+{\beta }_{11}{BoardSize}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.17em}+{\beta }_{12}{BoardIndep}_{i,t}+{\beta }_{13}{Duality}_{i,t}+{\beta }_{14}{TMTPay}_{i,t}\hspace{0.33em}\\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{0.17em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(7)

$$\begin{array}{l}{RD}_{i,t}={\beta }_0+{\beta }_1{PredFamilyEdu}_{i,t}+{\beta }_2{FirmSize}_{i,t}+{\beta }_3{FirmAge}_{i,t}+{\beta }_4{Leverage}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_5{Tangibility}_{i,t}+{\beta }_6{SOE}_{i,t}+{\beta }_7{SaleGrow}_{i,t}+{\beta }_8{ROA}_{i,t}+{\beta }_9{Loss}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{10}{InstiOwn}_{i,t}+{\beta }_{11}{BoardSize}_{i,t}+{\beta }_{12}{BoardIndep}_{i,t}+{\beta }_{13}{Duality}_{i,t}\\ \hspace{1em}\hspace{1em}\hspace{1em}+{\beta }_{14}{TMTPay}_{i,t}\hspace{0.33em}+IndustryFE+YearFE+{\epsilon }_{i,t}\end{array}$$

(8)

In the first stage, Equation (7) regresses FamilyEdu on the industry-average FamilyEdu (MeanFamilyEdu), which serves as the instrumental variable. The second stage then uses the predicted value (PredFamilyEdu) from the first stage to estimate firm innovation in Equation (8). Prior studies suggest that industry-average independent variables are valid instruments (Liu et al., 2024; Hu et al., 2023). The rationale is that firms within the same industry tend to exhibit similar FamilyEdu, making a firm’s FamilyEdu more influenced by industry-wide factors than by firm-specific characteristics. This instrumental variable is presumed to be exogenous. Specifically, firm innovation is unlikely to be associated with industry-average FamilyEdu, as innovation is a firm-level rather than industry-level behavior. Similarly, prior studies also suggest that the lagged independent variable may serve as a valid instrument (Lu et al., 2025; Ye & Zhang, 2025). In Equation (7), the instrument variable, MeanFamilyEdu, may be replaced with a one-year-lagged family education variable, which satisfies relevance because it is strongly correlated with current education owing to serial correlation. It also satisfies exogeneity because, given the temporal gap, it is unlikely to be directly related to the current dependent variable, firm innovation, except through its correlation with current education.

Columns 1−3 of

Table 12. Two-stage least squares regressions.

	(1)	(2)	(3)	(4)	(5)	(6)
	FamilyEdu	RDPerson	RDInvest	FamilyEdu	RDPerson	RDInvest
PredFamilyEdu		7.914 ***	3.534 ***		1.710 ***	0.936 ***
		(0.487)	(0.186)		(0.123)	(0.058)
MeanFamilyEdu	0.985 ***
	(0.057)
LagFamilyEdu				0.949 ***
				(0.004)
FirmSize	0.076 ***	−1.293 ***	−0.843 ***	0.007 *	−1.115 ***	−0.747 ***
	(0.011)	(0.162)	(0.067)	(0.004)	(0.163)	(0.067)
FirmAge	−0.235 ***	−1.451 ***	−0.862 ***	−0.017	−2.269 ***	−1.332 ***
	(0.030)	(0.439)	(0.189)	(0.012)	(0.467)	(0.204)
Leverage	0.069	−13.829 ***	−10.422 ***	−0.042 *	−14.032 ***	−10.552 ***
	(0.061)	(0.868)	(0.401)	(0.023)	(0.901)	(0.416)
Tangibility	−0.549 ***	−29.344 ***	−5.670 ***	0.033	−35.063 ***	−7.963 ***
	(0.080)	(0.989)	(0.421)	(0.031)	(0.985)	(0.409)
SOE	0.071	0.980	−0.580 **	0.003	1.252 *	−0.621 ***
	(0.047)	(0.661)	(0.253)	(0.020)	(0.663)	(0.232)
SaleGrow	0.065 **	1.783 ***	0.219	0.015	1.746 ***	−0.063
	(0.026)	(0.387)	(0.178)	(0.012)	(0.377)	(0.180)
ROA	−0.704 ***	−18.435 ***	−17.821 ***	−0.024	−18.974 ***	−16.149 ***
	(0.175)	(2.730)	(1.374)	(0.066)	(2.800)	(1.388)
Loss	0.053	0.532	1.250 ***	0.012	1.169 **	1.514 ***
	(0.035)	(0.516)	(0.243)	(0.014)	(0.504)	(0.242)
InstiOwn	0.122 ***	−3.907 ***	−0.498 **	0.003	−2.993 ***	−0.148
	(0.036)	(0.506)	(0.218)	(0.013)	(0.524)	(0.227)
BoardSize	0.155 ***	−4.240 ***	−0.638 *	0.010	−2.904 ***	−0.016
	(0.059)	(0.912)	(0.358)	(0.025)	(0.995)	(0.376)
BoardIndep	0.001	0.038	0.039 ***	−0.000	0.070 **	0.059 ***
	(0.002)	(0.032)	(0.012)	(0.001)	(0.033)	(0.013)
Duality	−0.005	0.913 ***	0.674 ***	0.003	1.095 ***	0.696 ***
	(0.017)	(0.239)	(0.101)	(0.006)	(0.247)	(0.105)
TMTPay	0.089 ***	1.554 ***	1.578 ***	0.002	2.348 ***	1.861 ***
	(0.016)	(0.241)	(0.107)	(0.006)	(0.240)	(0.113)
Constant	−2.593 ***	18.891 ***	−4.253 **	0.057	23.085 ***	−2.184
	(0.354)	(4.270)	(1.727)	(0.114)	(4.619)	(1.774)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	14,338	14,338	14,338	10,979	10,979	10,979
F-statistic	1050.09			103,991.62
LM-statistic	766.28			3107.21
Adjusted R2	0.101	0.008	0.068	0.908	0.200	0.241

Note: This table reports two-stage least squares regression results of firm innovation on TMT family members’ education, employing the industry-average family education and the one-year-lagged family education as instrumental variables. Variable definitions are provided in Table 1. All regressions include year fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

present the results of the 2SLS regression using industry-average family education as the instrumental variable. The first-stage estimates, consistent with Equation (7), indicate that the coefficient on the industry-average FamilyEdu is 0.985 and statistically significant at the 1% level, mitigating concerns regarding weak instruments. It also shows that the instrument, MeanFamilyEdu, is strongly correlated with the firm’s own FamilyEdu (Cragg-Donald Wald F-statistic = 1050.09), far exceeding the Stock and Yogo’s (2002) 10% maximal IV size critical value of 16.38, indicating no weak instrument concern. The Kleibergen-Paap rank LM statistic (χ² = 766.28, p < 0.001) rejects the null of underidentification, confirming the instrument’s relevance. In the second stage, following Equation (8), the results in Columns 2−3 reveal a statistically significant positive association between the predicted FamilyEdu and firm innovation, with coefficients of 7.914 and 3.534 at the 1% level for RDPerson and RDInvest, respectively.

Columns 4−6 of Table 12 present the results of the 2SLS regression using one-year-lagged family education as the instrumental variable. The results are similar to the preceding case using industry-average family education as the instrument. The large F-statistic indicates that the lagged family education is not a weak instrument. The second-stage coefficients are also positive and statistically significant at the 1% level. Overall, the instrumental variable regression results are consistent with the baseline regression estimates, offering additional support for Hypothesis 1. The findings reinforce a causal interpretation running from TMT family members’ education to firm innovation, rather than the reverse.

4.10. Panel Vector Autoregression and Granger Causality Tests

To further address concerns regarding dynamic endogeneity and potential reverse causality, this study estimates panel vector autoregression models using the system-GMM approach developed by Abrigo and Love (2016). Specifically, the joint dynamics are examined between TMT family members’ educational background (FamilyEdu) and two firm innovation input measures, R&D personnel ratio (RDPerson) and R&D investment intensity (RDInvest). The models use one lag and employ lagged instruments from periods 1 to 3. The results are given in

Table 13. Panel vector autoregression and Granger causality Tests.

		Coefficient	Std.Err	z-Statistic	p-Value	Granger χ2	p-Value
RDPerson	RDPerson	0.837	0.085	9.83	0.000
	FamilyEdu	−3.179	1.605	−1.98	0.048	3.921	0.048
FamilyEdu	RDPerson	0.004	0.006	0.55	0.585	0.298	0.585
	FamilyEdu	0.963	0.133	7.22	0.000
RDInvest	RDInvest	0.596	0.145	4.11	0.000
	FamilyEdu	4.199	2.131	1.97	0.049	3.881	0.049
FamilyEdu	RDInvest	0.011	0.009	1.18	0.236	1.402	0.236
	FamilyEdu	0.855	0.153	5.58	0.000

Note: This table reports results from panel vector autoregression models examining the dynamic relationship between TMT family members’ educational background and firm innovation. The models are estimated using system-GMM with one lag and instruments drawn from lags 1 to 3. Granger causality is tested using Wald χ² statistics.

.

Granger causality Wald tests indicate that FamilyEdu Granger-causes both RDPerson (χ² = 3.921, p = 0.048) and RDInvest (χ² = 3.881, p = 0.049), while the reverse directions are statistically insignificant (p = 0.585 and p = 0.236, respectively). These findings reinforce the directional interpretation from TMT education to firm innovations and bolster the identification strategy employed.

4.11. Alternative Measure of Firm Innovation

The preceding analysis employed R&D personnel (RDPerson) and capitalized R&D expenditures (RDInvest), which capture firm innovation inputs rather than actual innovation outputs or outcomes. To ensure robustness, we replace RDPerson or RDInvest in Equation (1) with the number of authorized patents (NumPatents), which provides a more complete picture of firm innovation. The results, reported in

Table 14. Alternative measure of firm innovation.

	(1)	(2)
	NumPatents	NumPatents
FamilyEdu	0.092 ***	0.042 **
	(0.024)	(0.019)
FirmSize		0.524 ***
		(0.032)
FirmAge		−0.016
		(0.073)
Leverage		0.345 **
		(0.147)
Tangibility		−0.495 **
		(0.197)
SOE		−0.579 ***
		(0.110)
SaleGrow		−0.114 *
		(0.069)
ROA		0.031
		(0.423)
Loss		−0.025
		(0.087)
InstiOwn		−0.011
		(0.089)
BoardSize		−0.215
		(0.158)
BoardIndep		0.000
		(0.005)
Duality		0.108 ***
		(0.041)
TMTPay		0.249 ***
		(0.044)
Constant	3.047 ***	−11.694 ***
	(0.084)	(0.749)
Industry FE	Yes	Yes
Year FE	Yes	Yes
Observations	2215	2215
Adjusted R2	0.144	0.402

Note: This table presents the results of univariate and multivariate fixed-effects regressions of firm innovation on TMT family members’ education, using the number of patents as an alternative measure of firm innovation. Variable definitions are provided in Table 1. All regressions include year and industry fixed effects. Robust standard errors, adjusted for heteroskedasticity, are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

, show that the coefficient on FamilyEdu is 0.092 and 0.042, statistically significant at the 1% and 5% levels in univariate and multivariate fixed-effects regressions, respectively. These findings further support Hypothesis 1, namely, that TMT family members’ education positively influences innovation in Chinese family firms. Note, however, that the sample size for NumPatents (2215 observations) is considerably smaller than for RDPerson and RDInvest (14,338 observations) due to missing patent data in the CSMAR database, which explains why we use R&D inputs as our baseline dependent variable.

5. Discussion

This study reinforces and extends prior evidence linking TMT characteristics to firm innovation by uniquely focusing on the educational backgrounds of family-member executives in Chinese family firms. The findings are consistent with earlier studies, though the target groups differ. Several studies report a positive association between board-level education and innovation (Kuo et al., 2018; Çerez et al., 2024; Wang et al., 2019). Barker and Mueller (2002) and Liu et al. (2010) similarly document a positive link between CEO education and firm innovation. However, those studies focus narrowly on board members or CEOs, whereas the present analysis encompasses the broader TMT, including directors, supervisors, and senior executives. Xu and Hu (2023) and Daellenbach et al. (1999) also find a positive association between TMT education and innovation, yet the present sample differs by restricting attention to family firms, where TMT roles are occupied specifically by family members. These results offer theoretical support for the view that human capital embedded in TMTs, particularly within family-controlled contexts, plays a central role in shaping innovation outcomes.

This study also adds nuance to the existing literature by identifying critical boundary conditions that moderate the relationship between family education and innovation. Specifically, the results indicate that this positive relationship is intensified in environments characterized by greater industry competition, consistent with prior evidence that external pressures enhance the role of human capital in driving innovation (Li et al., 2023; Le, 2024; Basit et al., 2022). Moreover, it is shown that higher levels of information transparency also strengthen this relationship, aligning with earlier findings that transparency reinforces the effectiveness of human capital in shaping innovation outcomes (Li et al., 2023; Che et al., 2019). This result also echoes broader evidence that transparency improves firm-level innovation capacity (Brown & Martinsson, 2019; Wang & Zhang, 2023). These findings offer theoretical insight into how contextual factors interact with managerial attributes to influence innovation, and suggest that institutional reforms promoting transparency and competition can amplify the benefits of executive human capital.

Moreover, the results reveal sectoral heterogeneity in the strength of this relationship: while manufacturing and service firms exhibit a strong and statistically significant association between family education and R&D intensity, no such relationship is observed in the agricultural sector. This finding mirrors sector-specific results reported by Liu et al. (2024), who show that the positive effect of employee stock ownership plans on environmental, social, and governance ratings is significant only in manufacturing and service firms, but not in agriculture. The present results suggest that the strategic value of family human capital may be context-dependent, shaped by sectoral variation in technical demands and capital intensity.

In our 2SLS analysis, we recognize that the industry-average FamilyEdu instrument may risk violating the exclusion restriction if industry-level human capital spillovers directly influence firm innovation. To address this concern, we also employ the one-year-lagged FamilyEdu as an alternative instrument, which is less likely to be affected by contemporaneous industry spillovers and thus provides a cleaner source of identification. Importantly, the results remain highly consistent across both instruments, lending confidence to the robustness of our causal interpretation. The choice of the two IVs reflects a trade-off between data breadth and instrument purity.

Finally, the generalizability of the present findings may be shaped by China’s distinctive institutional and cultural context. In the Chinese capital market, family-controlled firms represent a substantial subset of publicly listed companies, where founding families often maintain significant ownership stakes and executive influence (Walsh & Seward, 1990; Zellweger, 2007; Ward, 2011). The Confucian emphasis on education, hierarchy, and seniority may enhance the perceived legitimacy and strategic authority of highly educated family executives (Liao & Liu, 2024; Chen et al., 2022). Moreover, China’s evolving market institutions, which combine government intervention with private sector expansion, create an environment where family executives with advanced education can more effectively manage regulatory complexity and mobilize external resources (Peng et al., 2008). These cultural and institutional features may intensify the observed link between family education and innovation. In institutional contexts characterized by more dispersed ownership and stronger formal governance, this relationship may be attenuated. As such, the Chinese setting may condition the external validity of the results and highlight the role of family-based human capital in emerging markets.

6. Conclusions

Using a comprehensive dataset comprising 14,338 firm-year observations from 3097 publicly listed Chinese family firms over the period 2015−2023, this study yields six key findings. First, the education levels of TMT family members positively influence innovation inputs, measured by the proportion of R&D personnel and capitalized R&D investments. This suggests that educated family members are more likely to prioritize and allocate resources toward innovation as part of their managerial strategy. Second, the positive effect of TMT family member education on innovation intensifies in highly competitive industries. In such environments, firms face greater pressure to innovate for survival and growth, and educated family members are better equipped to identify opportunities, adjust strategic direction, and deploy resources effectively, thereby enhancing innovation outcomes. Third, high levels of information transparency strengthen the positive effect of TMT family member education on innovation inputs. In more transparent firms, greater information availability reduces asymmetry and enhances oversight, enabling educated family members to make more informed, accountable, and strategically aligned decisions. Fourth, the moderating role of firm size is evident: the positive relationship between TMT family member education and innovation inputs is more pronounced among smaller firms, where resource constraints and limited managerial depth make the influence of educated family members more consequential. Fifth, several potential mechanisms are identified: TMT family member education promotes firm innovation partly through reduced asset tangibility, reduced ownership concentration, and increased management fees, with these suggested pathways explaining approximately 27.5% of the total effect. Sixth, the influence of TMT family members’ education varies across sectors: it has a significant positive impact on R&D personnel in the manufacturing and service sectors but is insignificant in the agricultural sector. Finally, endogeneity concerns are mitigated using lagged independent variable regressions, 2SLS estimation, and panel vector autoregressions. The results remain consistent when firm innovation is measured alternatively by the number of authorized patents.

This research contributes to the existing literature in several ways. First, it adds to the literature on the determinants of firm innovation by identifying the educational background of TMT family members as a novel driver of innovation in Chinese family firms. Second, it contributes to the literature on the effects of TMT educational background by documenting firm innovation as a previously unexplored outcome in this context. Third, it examines the moderating roles of external factors such as industry competition and information transparency, offering a more nuanced understanding of the conditions under which innovation in family firms is enhanced. Fourth, it investigates the heterogeneous effects of family members’ education on firm innovation across industries, yielding practical insights for managers, investors, and policymakers seeking to promote innovation in family-controlled firms. Finally, from a governance perspective, the findings suggest that family member education functions as a form of internal governance leverage. It reflects an evolving mode of human capital-based governance, particularly salient in emerging markets where formal governance mechanisms may be complemented or substituted by family influence. In this regard, the study contributes to the broader discourse on the transformation of corporate governance in the context of socio-cultural and institutional change.

Despite these contributions, the study is subject to certain limitations. First, the analysis focuses solely on publicly listed family firms in China, which may constrain the generalizability of the findings to private firms or family businesses in other national settings. Future research could address this limitation by extending the analysis to include private family firms or by exploring similar dynamics across different countries and cultural environments. Second, the study relies on quantitative proxies for education and innovation, which, although informative, may not fully capture the qualitative dimensions of these constructs. For example, informal learning, managerial experience, or cultural perceptions of education and innovation may offer valuable supplementary context. Future studies could incorporate qualitative methods, such as interviews or case studies, to provide deeper insights into the intricate relationships among family member education, innovation, and external conditions. Third, the innovation measures employed in this study are focused on input-based proxies, namely R&D personnel and investment intensity, due to their broad availability and consistency across firms. While these metrics capture managerial commitment to innovation, they may not fully reflect realized outcomes or the effectiveness of innovation activities. Future research could incorporate output-based indicators, such as the number of patents granted, patent citations, or innovation efficiency, to offer a more comprehensive understanding of how educational backgrounds influence firm-level innovation performance. Fourth, while this study uses robust empirical methods, including fixed-effects, instrumental variable regressions, and dynamic panel modeling, the possibility of omitted variable bias or unobserved heterogeneity remains. By addressing these limitations, future research could build upon the present study to advance our understanding of how family firms pursue innovation amid an increasingly competitive global landscape.