*3.3. Sample Selection*

In view of the fact that manufacturing enterprises and information transmission, software, and information technology service enterprises (referred to as the information industry) belong to technology-intensive industries that tend to place high importance on R&D activities, this paper takes the manufacturing and information listed by companies in Shanghai and Shenzhen A-share markets in 2014 as the primary sample. To mitigate

the lag of the effect of independent variables on R&D investment, this paper defines the measurement of the independent variable as 2013.

The enterprises in these two industries that voluntarily release their social responsibility reports are defined as the voluntary group (processing group), while the enterprises that do not release their social responsibility reports are classified as the non-release group (control group). ST stocks are the "specially treated" stocks of companies with abnormal financial or other conditions. On 22 April 1998, the Shanghai and Shenzhen Stock Exchanges announced that they would carry out special treatment on the stock trading of these companies. After deleting the ST stock class and missing data, a total of 1912 sample observations were obtained, including 186 observations of the voluntary group and 1726 observations of the control group.

### *3.4. Description of Variables*

Following Chen and Tang (2012) [4], the natural logarithm of enterprise innovation expenditure is used to represent the dependent variable innovation input.

The voluntary release of social responsibility reports is a dummy variable, where CSR = 1 for enterprises that voluntarily release social responsibility reports and CSR = 0 for enterprises that do not release social responsibility reports. The CSR score of the Runling Global Rating Agency was used as a proxy variable for the CSR performance of enterprises that voluntarily issued CSR reports.

Using the method of Caliendo and Kopeinig (2008) [44], this paper selects those variables that simultaneously affect the voluntary release of CSR reports, along with the innovation input of enterprises for matching. Referring to similar studies [49,50], variables such as financial leverage, enterprise size, enterprise age, ownership attributes, operating performance, enterprise growth, organizational redundancy, ownership concentration, freecash-flow level, and industry were selected as the screening basis. Following the example of Lian et al. (2011) [51], "voluntary release of social responsibility report" is taken as the dependent variable of the logit model, and the combination of variables with the highest quasi-R2 and the area under the receiver operating characteristic (ROC) curve, or the area under the ROC, is selected as the covariate of the propensity score-matching model. See Table 1 for details of the covariates, outcome variables, and explanatory variables.


**Table 1.** Variable descriptions.


**Table 1.** *Cont.*

### **4. Empirical Results**

*4.1. Propensity Score-Matching Hypothesis Test and Empirical Results*

4.1.1. Matching Effect Test of the Voluntary Group and Control Group

Two preconditions must be satisfied for the empirical test when using the propensity value-matching method: the common support hypothesis and the balance hypothesis. In this paper, the nonparametric K-density distribution method was used to describe the propensity distribution of the voluntary group and the control group. Figure 1 shows the kernel density distributions of the propensity scores of the matching voluntary group and control group before and after the match. Before matching, the control group (dashed line) has the highest frequency, around a propensity score of 0.06, and the mode of the voluntary groups' propensity score (solid line) is about 0.12. The gap between the two density curves suggests a significant difference between the groups. After matching, in Figure 1, the two density distributions moved significantly closer to each other, indicating that the matching process to alleviate the differences in the two groups has relatively ideal match results.

In addition, after calculating the propensity scores of enterprises to release social responsibility reports voluntarily, it is necessary to further examine the post-matching distribution of each covariate of the two groups of samples. Only when the propensity distribution of the voluntary group and the control group is balanced and there is no systematic difference can the externality of "voluntary release of social responsibility" be addressed. The statistical method of a T-distribution test (bilateral) was used to compare the inter-group differences of sample covariates between the two groups, before and after matching, so as to evaluate the balance effect of matching. Table 2 shows the results of the balance test; the absolute deviation of financial leverage, enterprise age, business performance, enterprise growth, organizational redundancy, and other variables after matching is less than 5%, and there is no statistical significance between the voluntary group and the control group. There were significant differences between the voluntary group and the control group before pairing, and the deviations were reduced to 8.8% and 10.7%, respectively, after treatment; there were no significant differences between the two groups. According to Rosenbaum and Rubin (1985) [43], matching can be considered effective if the absolute deviation is less than 20%.

**Table 2.** The covariate balance test.


4.1.2. Matching Results Analysis of the Voluntary Group and the Control Group

In this paper, the kernel matching method is used to explore the average processing effect of voluntary CSR reporting; the corresponding T value is reported in Table 3. The pre-matching effect is about 0.624, which decreases to 0.237 after matching, indicating that the OLS model may lead to a high estimation coefficient due to endogeneity, while the PSM method makes the results more accurate because it addresses the problem of sample self-selection. In addition, the ATT values of a pair of two-nearest-neighbor matching and radius (caliper) matching robustness test change slightly.

**Table 3.** The treatment effect of a voluntary social responsibility announcement.


Note: \*\*\* represents *p* < 0.01; \*\* represents *p* < 0.05; \* represents *p* < 0.1.

## *4.2. The Empirical Test of Social Responsibility Performance and Innovation Investment*

Table 4 reports the mean value, standard deviation, and correlation coefficient of each variable. In the correlation between the two variables, CSR performance is significantly positively correlated with innovation input (*ρ* = 0.312, *p* < 0.01), which preliminarily conforms to the presented hypothesis. In addition, financial leverage, firm size, firm attributes, organizational redundancy, and other control variables are significantly correlated with the dependent variable of innovation input, and there is not a high correlation between the two variables. Innovation input is reported in logarithmic terms; investment in innovation averaged about CNY 43.1 million per enterprise. Note that the average performance of

CSR is about 37.580 (full marks is 100), indicating that the social responsibility of listed companies in China is at a low average development stage.


**Table 4.** Description statistics and correlation coefficient matrix (*N* = 186).

Note: \*\*\* represents *p* < 0.01; \*\* represents *p* < 0.05; \* represents *p* < 0.1.

#### *4.3. Analysis of the Regression Results*

We took the enterprises that voluntarily issued CSR reports as samples and considered innovation input and CSR performance, respectively, as the dependent variable along with one of the independent variables in the OLS multiple linear regression and quantile regression models. The OLS regression results in column 1 of Table 5 show that the estimated coefficient of CSR performance is about 0.033 (*p* < 0.01), indicating that CSR has a significant positive impact on innovation input, and the innovation input is roughly 3.3 percentage points higher for every one-point increase in the social responsibility performance index. Thus, hypothesis H2 can be verified. Column 2 of Table 5 shows that the regression coefficients of independent variables at 20% of the innovation input are about 0.060 (*p* < 0.01); similarly, the estimated coefficients on social responsibility performance at the 40th, 50th, and 60th percentiles of the dependent variable are 0.032 (*p* < 0.01), 0.030 (*p* < 0.01) and 0.020 (*p* < 0.05), respectively. The results show that social responsibility performance at these percentiles had a positive effect on innovation investment and that the positive effect gradually decreased with the increase in innovation input percentiles. At the 80th percentile, the coefficient estimate of social responsibility performance fell to about 0.001, and the result is not significant. This is plausibly due to the product or service superiority of high-innovation enterprises, and these firms tend to focus more on innovation. Thus, the social responsibility performance of such enterprises does not result in a significant effect on innovation. Based on the discussion of quantile regression, the original hypothesis H3a can be verified, indicating that with the increase in innovation input, the positive effect of CSR performance on innovation input gradually weakens.

As a robustness check, this paper also employs Tobit regression, using the overall sample of enterprises that do not disclose social responsibility and those that voluntarily disclose social responsibility. We use the left-censored processing method; that is, the social responsibility performance of an enterprise that does not disclose responsibility information is regarded as 0. The results show that CSR performance has a significant positive impact on innovation investment (*p* < 0.01). Financial leverage and enterprise age have a significant weakening effect on innovation investment (*p* < 0.01). Variables such as growth and organizational redundancy still have a positive leading effect on the dependent variables (the *p*-value of the organizational redundancy variable is 0.03, and the regression coefficient *p*-value of other variables is less than 0.01). The model is also significant as a whole (*p* < 0.01) (see Table 6 for details). This result shows no substantial change compared to the OLS regression results above, implying that the conclusion for hypothesis H2 is relatively robust; that is, the performance of CSR can have a positive impact on innovation investment.


**Table 5.** OLS and quantile regression results (*N* = 186).

Note: Standard errors are reported in brackets below the coefficient estimates. \*\*\* represents *p* < 0.01; \*\* represents

*p* < 0.05; \* represents *p* < 0.1.

#### **Table 6.** Tobit regression result.


Note: \*\*\* represents *p* < 0.01; \*\* represents *p* < 0.05.

#### **5. Conclusions**

In this paper, manufacturing and IT-related publicly listed companies in the Shanghai and Shenzhen A-share markets were selected as the overall research samples, and the samples were divided into the voluntary group and control group, according to whether they voluntarily issued social responsibility reports. The propensity score-matching method is used to empirically test the impact of a voluntary social responsibility report on innovation input. The results show that in the voluntary group, it has a positive effect, and the innovation input of the voluntary group is significantly higher than that of the control group. This conclusion shows that the active release of social responsibility information by enterprises has a positive effect on innovation input, possibly due to the stakeholders' increased attention and recognition. Enterprises may use R&D and innovation as a response strategy to meet the demands of stakeholders. At the same time, trust and feedback from stakeholders will also offer effective incentives for enterprises to strengthen innovation investment. This conclusion is contrary to the research findings of Pan et al. (2021) [6], and this may be related to the choice of variables. Pan et al. (2021) [6] took one dimension of corporate social responsibility (carbon emissions) as a proxy variable and found that it has a weakening impact on R&D investment. Additionally, our findings also differ from those of Mithani (2016) [49], which suggest that enterprises' efforts in the ecological environment will weaken the positive effects on R&D. Mithani's sample is based on the Indian market, and the study focuses on the environmental dimension of CSR. Thus, the differing conclusions likely arise from the different institutional backgrounds and variable measurements. On the other hand, our finding is consistent with some of the existing literature [15,52], and our study further provides empirical evidence for the positive relationship between CSR and R&D intensity.

In addition, we discussed the relationship between CSR performance and innovation investment in the voluntary group; the OLS regression results showed that CSR performance contributed to the increase in average R&D investment. This conclusion is in contradiction to the research results of Pan et al. (2021) [6], although their study was also based on a sample of Chinese firms in the context of economic transition, which showed a significant weakening effect of corporate carbon dioxide emission reduction policies on the intensity of R&D investment as the policy may lead to higher cost effects, thus affecting the intensity of innovation investment. In contrast, the apparently opposite findings are not contradictory, as carbon dioxide reduction is only one aspect of corporate social responsibility, and there are many other dimensions of corporate social responsibility. After gaining positive responses from stakeholders through the implementation of comprehensive social responsibility, companies will have more motivation to sustain development in R&D and innovation. In addition, the study by Gallego-Álvarez et al. (2011) [13] also presents the opposite conclusion to this paper, selecting 500 European companies and 500 non-European companies for their study; the conclusion shows that CSR has a significant negative impact on R&D investment. On the one hand, this may be due to the global scope of the study sample and the large differences in the degree of marketization of firms across countries (regions). On the other hand, the study defines CSR as a dummy variable, compared to the CSR variables measured by the score rating method, which can provide a more accurate picture of CSR performance.

The findings of the study are more similar to those of Ho et al. (2016) [26], who chose the Kinder Lydenburg Domini (KLD) rating index as a proxy variable for CSR, which covers a more comprehensive and extensive content and has high credibility and reference value in Western capital markets [53], and found that the social responsibility performance of companies in European and American capital markets has an R&D investment intensity that has a significantly positive predictive effect. In addition, the findings of this paper are consistent with the view of Husted and Allen (2007) [42] that "CSR provides opportunities for innovation". The above discussion indicates that after more than a decade of development, the development of CSR in China is becoming more and more mature; favorable CSR performance is becoming a medium of interaction between enterprises

and their stakeholders, and it is gradually becoming an important driving force for R&D innovation and competitiveness.

At the same time, the quantile model was also used to explore the effect of CSR performance on the different quantiles of innovation input. We found that with an increase in innovation input, the effect of CSR performance on innovation input gradually diminishes. This result is in line with the expected assumption of the resource-based theory. Under the premise of limited resources, enterprises with high innovation investment tend to attract customers and other stakeholders through high-quality differentiated products, and such enterprises lack the pressure and motivation to "please" stakeholders through social responsibility. In contrast, enterprises with insufficient investment in innovation and low product differentiation tend to practice social responsibility, which is a prudent way to convey the message of "benevolence" to society.

Meanwhile, in the Tobit model of Table 6, we find that corporate financial leverage has a significant negative effect on innovation investment, which finding is similar to previous research [4], in which the asset-liability ratio indicates a firm's external financing capacity. The lower the asset-liability ratio, the more funds a firm can borrow, and the more investment it will make in its innovation activities [4]; firms with a low asset–liability ratio usually have a large amount of potentially redundant resources, which can help enterprises in the process of selecting R&D projects, alleviate the urgency of pursuing immediate short-term results, and motivate enterprises to try high-risk strategies and innovation projects; in addition, the R&D innovation of enterprises is usually coherent and the projects are interrelated. The existence of redundant resources enables enterprises to invest in new projects when faced with environmental changes, thus ensuring the continuity of R&D. Moreover, the age of the firm has a significant negative effect on innovation investment, which is consistent with the random effects model of Ju et al. (2013) [54]. With the growth of enterprise survival time, enterprise knowledge and experience and organizational systems may become more and more solidified; all kinds of organizations face the problem of organizational inertia and this inertia will continue to increase over time, which is manifested in the organization's operation of conformity and the old-fashioned over-reliance on the original resources, thus affecting positive enthusiasm for R&D and innovation investment [55].

Compared with the existing literature, the value of this paper may be reflected in the following aspects. First, the application of the propensity score-matching model alleviates the endogenous bias caused by the self-selection of samples in traditional regression methods and adds more convincing empirical evidence when discussing the relationship between CSR and innovation investment. Second, the use of a quantile regression model on innovation input helps shed light on the varying or unequal effects of CSR performance, given the level of innovation input. In particular, our results reject the knowledge-based view and conclude that higher social responsibility performance is not statistically associated with higher innovation investment.

At the same time, the practical implications of this study lie in the following areas. First, both CSR and innovation investment are welfare-enhancing strategies for an enterprise. Specifically, innovation investment may entail the development and application of energy-saving and environmentally friendly technologies that increase the consumer's utility and improve the workers' working conditions, as well as increase the efficiency of resource use. This is the indirect embodiment of CSR. Therefore, an enterprise's innovation investment decision may be based on the specific needs and goals of the enterprise and its stakeholders, in order to optimize the overall effect of the two strategies. Second, because CSR and innovation investment have positive externalities, establishing common platforms to facilitate information-sharing in technology and management and helping enterprises to reduce the cost of social responsibility and the risk of failure in R&D and innovation. Such platforms could also help guide enterprises to develop complementary social responsibility and innovative investment strategies.

In spite of the aforementioned theoretical significance and practical enlightenment, however, there are still some imperfections in this paper. In future research, we will select multiple years to verify the above assumptions in this paper, using panel data samples. Meanwhile, we could also further explore the impact and mechanism of corporate R&D investment on corporate social responsibility, along with the boundary of contingency factors influencing the above two relationships.

**Author Contributions:** L.C., S.H.L., S.X. and Y.L. conceived the idea of the paper and designed the research; S.X. and Y.L. reviewed the related previous literature; L.C. and S.H.L. analyzed the data. All authors wrote and reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This paper is supported by China Philosophy and Social Sciences Office, Grant No. 20AJY008.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Sample data could be accessed from the Wind Database, the website of Wind Database: "https://www.wind.com.cn/". The database needs to be used for a fee and the authors did not have any special access privileges that other users would not have.

**Conflicts of Interest:** The authors declare no conflict of interest.
