Innovation through invisible diversity
‘Are LGBT-friendly firms more innovative?’
By
Eva Bosma (s2503506)
MSc BA Strategic Innovation Management
University of Groningen Faculty of Economics and Business
Supervisor: K. McCarthy Co-assessor: E. Huizingh
June 2018 Word count: 14061
Abstract
This study emphasizes on the invisible form of diversity. Since invisible diversity compasses many characteristics, this study will focus on the invisible sexual orientation measured by focusing on the inclusiveness of the LGBT-community (lesbians, gay, bisexual, transgender). The expectation is that invisible diversity will positively influence the innovation performance of the firm, as have been proven to be correct for the visible form of diversity. Besides the hypothesized relationship, LGBT inclusive signaling is expected to positively moderate this relationship. By using data collected from the Human Rights Campaign Foundation, CSR Hub, the European Patent office and the ORBIS database, the results show no support for the hypotheses. The results show a partially negative influence of the invisible diversity on the innovation performance while the effect of the moderator is unproven. However, when specifying the research per separate industry we do notice some different outcomes. This result indicates that the variables in certain industries do matter.
1. Introduction
In the last decade a lot of focus has been on differentiation and integration. Not only society, but businesses as well, made some remarkable changes regarding this differentiation debate. This resulted in an increase in employee resource groups (ERGs) in which minorities within an organization are grouped together to create social support, network opportunities and to provide them with a platform (Githens, 2009). Not only visible minorities, but many invisible minority groups raised their voices to start making the invisible more visible. Invisible diversity is about features that make us unique, but which are not directly visible or which are easy to hide. Someone might have bad vision or be deaf, someone can be gay or transgender, or someone can have a mental health issue and hide all of this.
Society and governments have acknowledged the importance of making the invisible more visible. This resulted in some remarkable legal and political gains for the LGBT-community (lesbians, gay, bisexual, transgenders) in the United States, including the freedom to marry. Despite this change in law, discrimination regarding the LGBT-community has only increased in the past years (U.S. Equal Employment Opportunity Commission (EEOC), year-end-summary 2016) and this discrimination might be a reason why, according to the Australian Workplace Equality Index (AWEI), almost 30% of the sexually diverse people is not out at work.
nondiscrimination act) state apply for more patents compared with firms located in non-ENDA states (Geo and Zhang, 2015). Another study about (invisible) diversity and inclusion shows, using a qualitative case study, that a firm who embraces a culture where employees can fully be themselves leads to a more productive, effective and creative work environment (Fullerton, 2013).
When researching the current diversity literature we find a lot of information about the influence of diversity on firm performance. However, most findings are often conflicting. Some authors find a positive relation between diversity and performance, others a negative, and some do not even find a significant relationship at all (Jackson et al., 2003). When specifying the influence of diversity on the innovation performance we find empirical evidence showing that a diverse workforce has a direct positive influence on the innovation performance of a firm. Since the relation between diversity and innovation performance does provide clear results, this relationship is a well discussed topic in the current literature (e.g. Richard, 2000; Richard, McMillan, Chadwick and Dwyer (2003); Hoffman (1985)).
Since the invisible form of diversity is getting more attention from society and governments it is important that our literature starts paying more attention to the invisible forms of diversity as well. Therefore, the focus of this study will be the influence of the invisible diversity. Since there is empirical evidence showing a positive influence of diversity on the innovation performance, this research will raise the question whether the same applies for the invisible form of diversity. Therefore, the main research question of this paper is:
‘What is the influence of invisible diversity on the innovation performance of a firm?’
this research also have a focus on the signaling theory. Being inclusive towards invisible forms of diversity and acting upon this statement are two different things. To check whether signaling invisible diversity inclusion influences the relationship between invisible diversity and the innovation performance, will the invisible inclusive signal be added as a moderator in this research. Using an invisible inclusive signal as a moderator in this research will contribute to the signaling literature, since the current literature does not provide any information regarding the relationship between invisible diversity signals and the innovation performance.
Furthermore, the outcomes of this study will provide several societal/managerial implications. When this study provides a positive outcome, it might result in creating a zero-tolerance approach regarding discrimination since being open to all forms of diversity could enhance the innovation performance. Therefore, a positive outcome might provide the invisible diverse minorities with new job opportunities since it could stimulate firms to incorporate several invisible diverse minority groups in their employee search. This results in the potential of this research to not only contribute to the literature, but to society and businesses as well.
2. Literature review 2.1 Diversity
Diversity can be seen as a really broad and narrow concept at the same time. A narrow view on diversity defines diversity only as a term related to Equal Employment Opportunity and Affirmative Action (EEO/AA) (Carrell, Mann, and Sigler, 2006), while a broader concept includes all the different options about how people can be different. According to Gomez-Mejia, Balkin and Cardy (2007) can diversity be described as ‘human characteristics that make people different from one another’ (p.119). Roberge, Lewicki, Hietapelto and Abdyldaeva (2011) state that ‘diversity refers to differences between individuals on any attributes that may lead to the perception that another person is different from the self’ (p.1). Carrell et al. (2006) believe that ‘modern definitions of workforce diversity focus on the ways that people differ that can affect a task or relationship within an organization’ (p. 6). The study of Carrell et al. (2006) shows that in 2004 firms were mostly divided with their definition about diversity, almost half of the firms researched viewed diversity narrowly and the other half viewed diversity in a broader sense. The results from their research about the components of diversity are shown in figure 1.
Figure 1: Components of Diversity (Carrel et al., 2006, p.9).
diversity discloses attributes which are likely to relate to knowledge, skills and abilities needed in the workplace. According to Jackson et al. (2003) are most studies in the field of diversity theories focused on the relations-oriented diversity. The attributes researchers include most often are gender diversity, race-ethnic diversity, and age diversity.
The relationship between relations-oriented diversity on the workforce and performance gives many different outcomes. Pazy and Oron (2001) show in their study about military officers that gender diversity has a positive influence on the performance of woman, but not on the performance of men. Another study from Fenwick and Neal (2001) shows, using a simulation, that gender composition influences some measurements of performance, but not all measurements. In general, we see that some studies about gender diversity show a positive relationship between gender diversity and performance (Jackson and Joshi, 2003; Rentsch and Klimoski, 2001), some show a negative relationship (Jehn and Bezrukova, 2003), and some studies do not find a significant relationship (Richard, 2000; Watson, Johnson and Merritt, 1998). The same applies for age diversity. Many scholars show mixed results about the relationship between age diversity and firm performance (Kilduff, Angelmar and Mehr, 2000; Bunderson and Sutcliffe, 2002). The relationship between race-ethnic diversity and firm performance is only researched limited in the current literature. The current literature shows some negative relationships (Jackson et al., 2003; Kirkman, Tesluk and Rosen, 2001) and only one study (Richard, 2000) found a positive relationship.
2.2 Diversity and Innovation
firm (Gomez-Meji et al., 2006). Okoro and Washington (2012) also show in their research that the best way to maintain a steady flow of innovations within the firm is by hiring and retaining diverse employees (e.g. backgrounds, race and nationalities). Besides, the current diversity literature shows that having a diverse group of people working together within an organization might enhance creativity (Woodman, Sawyer and Griffin, 1993). Since innovation is the successful implementation of creative behavior (Amabile, Conti, Coon, Lazenby and Herron, 1996), we can argue from this point of view that having diversity on the workforce might indirectly influence the innovation performance of the organization in a positive way. Based on the above mentioned studies, we can say that diversity in the workforce not only directly correlates with innovation (Richard, 2000; Richard, 2003; Hoffman, 1985), but also indirectly influences the innovation performance of a firm (Gomez-Meji et al., 2006; Woodman et al., 1993).
2.3 Invisible diversity
As shown in the previous sections did we discuss diversity and its relation with the innovation performance of an organization. However, we only discussed the traditional view on diversity where the focus is on gender, age and race-ethnicity (e.g. Jackson et al., 1995), which can be called visible diversity according to Tsui and Gutek (1999). A less common form of diversity is the ‘invisible diversity’. Invisible diversity is the type of diversity we usually do not see when we meet and interact with others. Examples of this form of diversity can compass disabilities, religion, class, regionalism, sexual orientation and other characteristics (Clair et al., 2005). A lot of research has been done about the visible form of diversity and only limited research has been conducted about the invisible forms of diversity.
students, that in general LGBT (lesbian, gay, bisexual and/or transgender) students feel disenfranchised and invisible. This same statement can be made from the interviews conducted by the working paper of Hendriks et al. (2018). Furthermore, the Centers for Disease Control show that as of 2012, about half of the adults in America (117 million people) had one or more chronic health conditions. Chronic health conditions are often invisible, yet many people within this category actively participate in the workforce (e.g. Pinder, 1988; Vickers, 2001).
When comparing visible workforce diversity with invisible workforce diversity, we see that these different types have different interaction experiences at work (Clair et al., 2005). When people can fully be themselves at work, they experience a feeling of authenticity. However, when people feel they have to conceal personal information, and therefore keep certain things invisible, will this negatively influence one’s authentic self-presentation (Creed and Scully, 2000; Moorhead, 1999; Reimann, 2001). One way to incorporate people who are invisibly diverse is to call for better policies and practices from organizations and organizational leaders (Sabat, Lindsey, Membere, Anderson, Ahmad, Kind and Bolunmez, 2014).
2.4 Invisible diversity and Innovation
mentioned in the previous section, do Sabat et al. (2014) argue that to incorporate the invisible diverse people is to call for better policies and practices. Therefore, we could argue whether it is the ENDA state that influences the innovation performance in a positive way, or whether it is the (equality) policies of the specific firms in those ENDA states.
2.5 Effect of equality-supportive policies
That equality supportive policies have become more important over the years, has to do with several additions to the law. Not only the Equal Employment Opportunity and Affirmative Action (EEO/AA) (Carrell et al., 2006) had an influence, but also different equality expressing acts made this change possible. In the UK, for example, is equality expressed via the Sex Discrimination Act (1975), the Race Relations Act (1976) and the Disability Discrimination Act (1995). In 2006 the (invisible) disability diversity was even acknowledged again and therefore the Convention on the Rights of Persons with Disabilities and its Optional Protocol was adopted at the United Nations. However, in this part we will mostly focus on how the inclusion of invisible diversity in a firm's policy can have a different impact on the performance of the firm.
The question rises what the influence is of an equality-supportive policy on for example firm performance and firm employees. Badgett, Durso, Kastanis and Mallory (2013) show in their research that equality-supportive policies result in less discrimination and a better workplace environment. Furthermore, they also find a relation between equality-supportive policies and improved employee’s health. Blau (1964) shows another finding of equality-supportive policies, namely an increase in job satisfaction and commitment, which he explains by the social exchange theory. This theory is based on reciprocity were the focus is on exchanging things in return for mutual benefits. In the case of equality-supportive policies it could mean that employees might feel more appreciated and supported by the firm and in return will be more committed to the firm and work harder, which will result in an increase in productivity (Wang and Schwarz, 2010).
argue that firms who exploit an equality-supportive policy create a better corporate image. It signals that the firm is operating in a socially responsible manner and is providing a safe environment for all kinds of employees. They argue that an environment like that might attract more qualified and higher educated people, which in turn will improve firm performance. Furthermore, they argue that such a policy might positively influence the buying behavior of the invisible diverse community, since they prefer to make purchases from an equality-supportive firm.
However, there are still states in the US who do not have passed the ENDAs, meaning that there is still a significant portion of the US population who argue that some members of the invisible-diverse community act immoral and should be treated differently (Johnston and Malina, 2008). This means that when firms have an equality-supportive policy they might also exile potential customers, investors and partners. The same applies for potential employees, as some of them might not want to work for a firm who hires all kinds of (invisible) diverse employees. Furthermore, Johnston and Malina (2008) mention that becoming a fully equality-supportive firm does come with some substantial costs. According to the Australian Human Rights Commission are the barriers (potential) employees with disabilities face, regarding to the workplace environment, in need to be changed. These changes are referred to as ‘reasonable adjustments’. These changes could include bigger computer screens for employees with bad vision or different work schedules. However, firms are not obliged to make these changes if they can prove that these are too expensive, difficult or time consuming. This indicates that, if a firm wants to be fully equality-supportive, this might come at a high price. Furthermore, one criteria from the Human Rights Campaign (HRC) foundation is that there are equal benefits for LGBT workers and their families (HRC report, 2018), meaning that a firm should provide domestic partner benefits. Since, in the US, employees can cover their partners under the health insurance of its employer, a LGBT-inclusive policy might result in an increase in health insurance costs for the firm.
H1: Incorporating invisible diversity in the firm’s policy has a positive influence on the innovation performance of the firm.
2.6 Signaling theory
As mentioned in the previous section do Wang and Schwarz (2010) believe that incorporating an equality-supportive policy creates a better corporate image, which signals that the firm is operating in a socially responsible manner. However, according to Colgan, Creegan, McKearney and Wright (2007) is there too often a difference between policies and practices. This means that sometimes the policy does not match with how the firm is operating in practice. To incorporate this element in this research, we will refer to the signaling theory.
The signaling theory is about the communication between two different parties when they both have access to different information. The sender decides what kind of information is being signaled and how and when this communication is happening. The receiver decides on how to interpret the received information (Connelly, Certo, Ireland and Reutzel, 2011). Examples of such signaling are for instance a start-up with a board consisting of highly respected board members in their field of work. Having such a board can signal that they are a reliable start-up and can therefore be seen as attractive by potential investors. However, signaling can also have a ‘down-side’. According to Connelly et al. (2011) might some organizations share their private information to show that they are a high quality organization, while low quality organizations might not share all their information, so they eventually might end up higher on the quality scale.
might get a dishonest reputation among their stakeholders. We talk about a honesty signal when the signal represents the actual association of that specific signal (Durcikova and Gray, 2009).
Westphal and Zajac (1998) argue that too little research has been conducted about the incentives of signalers. They argue that signals might become weaker, when the signaler has the incentive to delude the potential receivers. Therefore, they suggests that researchers should investigate the relation between the extent to which signals of the firm are actually representative to the firm. This is the issue Colgan et al. (2007) highlighted as well. The signal is in this case the policy and the actual representation are the practices of the firm. When these two aspects do not match, we could say that there is a dishonest signal.
Miller and Triana (2009) show in their research that the signaling theory can be used to signal diversity of an organization. Their article shows that signaling gender diversity in a board has a positive influence on the innovation performance of the firm. Signaling the diversity of an organization, by for example having multiple races in the firm’s board, shows that the organization understands multiple different stakeholders and that they value all their interests. Furthermore, Fombrun and Shanley (1990) argue that firms can signal their ethical nature by implementing Corporate Social Responsibility (CSR) initiatives. They say that firms who invest in certain charities and different public foundations can signal that the firm is operating in a socially responsible way which, according to them, has a positive impact on the firm’s reputation. This ‘corporate signal’ can be seen as the commitment of the company to certain social aspects (Aqueveque, 2005). With this CSR way of working they can signal their corporate values, sense of justice and congruence with its words (which are for example written down in their policies). This signal is related to the corporate image of the firm and is, in general, more observable then the firm's policy.
Based on the above literature we phrase the following hypothesis:
2.7 Conceptual model
3. Methodology 3.1 Measure invisible diversity
As discussed in the literature section of this paper does invisible diversity compass several characteristics. Since focusing on all aspects will be too time consuming, will the focus in this research be on the invisible diversity regarding the sexual orientation. The reason for this focus is mostly due to the fact that this aspect is currently really present within society. Furthermore, not only people, but businesses as well are becoming more aware that focusing on the equality regarding sexual orientation is demanded by society. Therefore, in 2002, the HRC Foundation started ranking firms on how LGBT-inclusiveness they are operating. Not only did the HRC acknowledge this, in 2013 Credit Suisse added the LGBT-equality index as another ethical investment product to the existing mainstream and therefore acknowledged the importance of this, invisible, stakeholder as well.
Therefore, to measure the invisible diversity policies and signals of the firms, we will measure the equality regarding this specific stakeholder. Specifically, this research will use the LGBT inclusiveness of the firm as a measure of the invisible diversity of the firm.
3.2 Data collection
500 largest publicly traded businesses, American Lawyer magazine’s top 200 revenue-grossing law firms (AmLaw 200) and hundreds of publicly and privately held mid- to large sized businesses (CEI-2018, full report). The CEI score is measured on a scale from zero to 100 and is composed of four different criteria for which the firms are given points. A CEI score of 100 is the perfect score. The higher the CEI score the more the firm's policies are focused on the inclusiveness of the LGBT employees. Every year the HRC Foundation sends out a CEI survey to previous and prospective respondents. As mentioned before, these firms will be scored on four different criteria including non-discrimination policies and equal employee benefits. The CEI score will be based on the overall score of the firm. The HRC foundation is not connected to any of the responding firms in any way, this to guarantee that there are no incentives to manipulate the results. Even though firms are allowed to fill in the survey by themselves and are therefore able to provide fake answers, the HRC foundation still makes sure that the information gathered from the different firms is objective and will provide a reliable CEI score. This is organized in two different ways. First, the HRC foundation collects publicly available data and uses that information to double check all the received surveys to make sure that there are no false assumptions. Afterwards the firms will receive their official CEI score. Second, to make sure that firms cannot just ignore any of the requests to fill in the survey to avoid a bad score, the HRC foundation still scores these firms with the use of secondary data sources. These secondary data sources consists of publicly available information, as well as information submitted to HRC from unofficial LGBT employee groups or individual employees (HRC report 2018, p.92). This information will be used to provide the firms with an unofficial CEI score. With these actions taken, the report from the HRC foundation includes reliable, objective information and CEI scores of both LGBT- and non-LGBT-inclusiveness firms. This results in a lot of variance in scores and guarantees that firms, who on forehand believe they would get a bad score, cannot just avoid taking part in this research. Therefore, the emergence of ‘greenwashing’, where corporations creatively manage to signal certain elements and avoid others to gain the best reputation (Laufer, 2003), is decreased to a minimum.
LGBT-inclusiveness which are observable by the general public. This results in a focus on the CSR rating for these specific firms, since the CSR rating can be seen as the corporate signal of the firm. This data will be collected from the CSRHub, which is the largest sustainability business intelligence database. The CSRHub collects data from 556 industry-leading CSR/ESG data sources and combines that data to provide the different firms with a score ranging from zero to 100. The higher the score, the more the firm is signaling a CSR way of operating. Since a general focus on the CSR will focus too much on the non-LGBT related aspects, will the focus be on the subcategory ‘Employees’ and within that subcategory on the sub-subcategory ‘Diversity & Labor Rights’. The focus in this sub-subcategory is, among others, on proactive management initiatives, the structure of the boards (e.g. background, skills, minorities) and the representation of minorities in employees.
The innovation performance data is collected via the European Patent Office (EPO), which is the patent office for Europe that issues patents to inventors and business for their inventions, and trademark registration for product and intellectual property identification. Within this database of the EPO we specifically used the Global Patent Index since it enables detailed patent searches.
The data from the HRC Foundation, the CSRHub and the EPO is used over the years 2008 - 2017. The research will start with data from 2008 so that we will be able to see the differences over a time period of 10 years. This will provide us with insights about the change based on the different scores and therefore we will be able to see how things changed when the LGBT-community became an even more important stakeholder.
3.3 Measurement
TheInvisible diversity policy
of the firm is measured by focusing on the LGBT-inclusiveness of the firm.
Therefore, the CEI score will be used as a measure. Several researchers use the CEI score as a measure of LGBT-inclusiveness of a firm (e.g. Johnston and Malina, 2008; Everly and Schwarz, 2015). The different firms get a CEI score between zero and 100. The scores will be compared per year.
The Invisible diversity signal
of the firm is measured by focusing on whether the firms have a
LGBT-inclusiveness signal. This will be measured by focusing on the CSR rating, specifically in the sub-subcategory ‘Diversity & Labor Rights’, since this is the most objective way of measuring a corporate signal. This score can range from zero to 100.
3.4 Method of analysis
3.4.1 Negative binomial regression
The dependent variable in this research is the innovation performance of the firm, measured in the amount of patent applications. The dependent variable needs to be treated as a count variable, since the innovation performance of the firm can only take the non-negative integer values {0, 1, 2, 3, …}. Count data shows the number of occurrence of a certain action or behavior. Since the innovation performance is a type of data which cannot get a value below zero, the use of a count data analysis is necessary (Coxe, West and Aiken, 2009). When the standard deviation is bigger than the mean of the count variable, the data is over-dispersed. In this case the literature suggests using a negative binomial regression instead of a Poisson regression. Both types of analysis have the same structure, but the confidence intervals for the negative binomial regression are in general narrower compared to those of the Poisson regression (Hilbe, 2007; Gardner, Mulvey, & Shaw, 1995). In this study we expect the that the dependent variable (innovation performance) is over-dispersed and therefore a negative binomial regression will be used. Furthermore, the data used for this research is quite large (204 firms), which also implies that it is sensible to use a negative binomial regression. We should notice that the negative binomial regression does not contain an equal R squared measure (Long and Freese, 2006). Therefore, to measure the fit of the data to the model being used, the log-likelihood will be included. The value of the log-likelihood on itself will not have a meaning, but the comparison of the value with the different models will. The more maximized the log value is (the closer the value gets to zero), the better fitted the model.
collection part of this research is the data collected over a time period of ten years (2008 - 2017), therefore time series data will be needed to analyse the data.
3.4.2 Data sample
A total of 204 different firms, from 10 different industries are used for this research. The different industries, including the amount of firms in those industries, are shown in table 1.
Industry Number of firms
1. Advertising and Marketing 20
2. Aerospace and Defense 14
3. Apparel, Fashion, Textiles, Dept. stores 16
4. Energy and Utilities 27
5. Chemicals and Biotechnology 16
6. Manufacturing 39
7. Automotive 25
8. Computer Software 17
9. Internet Services and Retailing 17 10. Computer Hardware and Office equipment 13 Table 1: Overview industries
4. Results
Variable Observations Mean Std. Dev. Min Max (1) Inn. performance 2,040 88.31569 332.2991 0 3622 (2) CEI score 1,403 71.91162 31.64148 0 100 (3) CSR Div./Labor 1,900 62.15263 8.852716 42 82 (4) Firm age 2,008 62.41633 47.03071 0 203 (5) RD expenses 984 1016391 1988660 2047 2.26E+07
Table 2: summary of statistics
Table 2 shows that the amount of observations fluctuates over the different variables, ranging from 2040 observations for the innovation performance to only 984 observations for the R&D expenses. This means that, when running the analysis over all the variables, not all the data will be available for use. Finally, the standard deviation and min / max values of the R&D expenses variable are measured in a thousand USD. Next a pearson correlation is conducted to measure the strength of the linear relationship between the different variables. The results of this correlation are shown in table 3.
Variable (1) (2) (3) (4) (5) (1) Inn. performance 1.0000 (2) CEI score 0.2294 1.0000 (3) CSR Div./Labor 0.1210 0.4979 1.0000 (4) Firm age -0.1138 -0.1556 0.0569 1.0000 (5) RD expenses 0.4368 0.3075 0.1043 -0.0757 1.0000
Table 3: Pearson correlation results
Based on the results of table 3, we see that the variables are not highly correlated with other variables (Camm and Evans, 1996). This means that we can use the different variables together in the following negative binomial regression analysis.
4.1 Negative binomial regression
firm, the variables are classified in a certain hierarchy. The first model contains the control variables. The second model includes the independent variable, to test for hypothesis 1. The third model measures the effect of the independent variable (CEI score) and the independent effect of the CSR score on the innovation performance. The fourth model will include the interaction effect between the CEI score and the CSR score, this output will be used to test for hypothesis 2.
Before we can run the negative binomial regression, we need to make sure that the data is measured per firm over the ten years (2008 - 2017). Therefore, a time series data analysis needs to be used. The data, for all the industries, is analyzed per firm over ten years. The results of negative binomial regression analysis are shown in table 4.
Variables Model 1 Model 2 Model 3 Model 4
R&D expenses -3.39E-08* -8.74E-08*** -9.23E-08*** -9.12E-08*** (1.81E-08) (2.21E-08) (2.28E-08) (2.30E-08)
Firm age 0.0021116 -0.0037386* -0.0038083* -0.0037145* (0.0014638) (0.0021861) (0.0021914) (0.0022066) CEI score -0.0027862 -0.0032927* 0.0029755 (0.0018227) (0.0018988) (0.0181837) CSR 0.0127265 0.020596 (0.0129278) (0.0260421) CEI x CSR -0.0001053 (0.0003036) Observations 974 725 725 725 Wald's Chi-square 5.34 21.25 21.91 22.17 Wald's p-value (0.0693) (0.0001) (0.0002) (0.0005) Log-likelihood -3284.1894 -2440.9431 -2440.4609 -2440.4011
Standard Errors in the parentheses: ***p<0.01, ** p<0.05, *p<0.1 Table 4: results negative binomial regression
performance (measured in the amount of patents) by 0.000000039. The control variable firm age is not significant.
Model 2, used to test for hypothesis 1, shows negative significant results for the control variables. These results are not in line with what was expected. The model shows no significant results for the independent variable, this implies that there is no influence of the CEI score on the amount of patents. However, in model 3 we do notice a significant result for the CEI score on the amount of patents. This relation shows significance on a 10% level and the relation is found to be negative. This means that an increase in the CEI score by one unit will result in a decrease in the innovation performance by 0.0032927. This result is not in line with what was expected. Therefore, no evidence has been found to support for hypothesis 1. Instead, we do find partially support for the contrary of hypothesis 1, namely that the relation between the dependent and the independent variable is partially proven to be negative.
Model 4 is used to test for hypothesis 2. This model shows significant negative results for both control variables. Furthermore, the model shows no significant results for the other variables. This implies that the interaction effect of the CEI score and the CSR score does not influence the relation between the dependent and independent variable. Therefore, no evidence has been found to support for hypothesis 2.
The log-likelihood is included in the model. This number on its own does not tell anything about the model but comparing it with the other log-likelihood values shows the fit of the model. The closer the log-likelihood gets to zero, the better fitted the model. Finally, the Wald’s chi-square and p-values are included in the table. The Wald’s p-value is significant for all models, therefore the null hypothesis in all models can be rejected which indicates that the coefficients for the variables is not simultaneously equal to zero. This result shows that the variables in the model create a statistically significant improvement in the fit of the overall model.
4.2 Analysis per industry
Since the total amount of observations, as shown in table 2, ranges from 2040 to 984 we could argue that the previous analysis of the overall model does not contain the data for all the firms and industries.
Since some firms do not have R&D expenses, they fall out of the analysis. The database shows that for 3 industries no R&D data is available. This implies that the previous analysis has only been run over 7/10 industries. Therefore, for purely technical reasons, we will double check the results by checking model 2, 3 and 4 per industry. In this case the R&D expenses will be left out of the analysis for industries 1, 3 and 4, the other industries will be individually analyzed keeping the R&D expenses as control variable. The results of these analysis are shown in appendix A (for industries 1, 3, 4) and in appendix B (for industries 2, 5, 6, 7, 8, 9, 10).
As shown in appendix A, the results for industry 1 are insignificant (for model 2) and not available for the other models. Therefore, both hypotheses are not supported for industry 1. For industry 3 we see significant results for model 3. These results show that the control variable (firm age) is negatively related (on a 10% level) with innovation performance, this result is not in line with what was expected. Furthermore, the CEI score has a positive significant effect on the innovation performance, this is in line with hypothesis 1. However, this positive relation only occurs in model 3 and not in model 2. In other words, the model only shows a positive significant relation when the CSR is also added to the model. Therefore, hypothesis 1 is only partially supported. The CSR shows a significant negative result. For industry 4 we do not find significant results, meaning that both hypotheses are not supported for this specific industry.
6 hypothesis 1 is unsupported and that model 4 even shows signs for the contrary effect. For industry 7 the CEI score shows a positive significant relation with innovation performance, but this relation only occurs in model 4. Therefore, hypothesis 1 is only partially supported for industry 7. We focus on model 4 to test for hypothesis 2. For this hypothesis we only find a negative significant relationship in industry 7, which shows that hypothesis 2 is not supported. All the other industries do not show significant results, which implies that the moderator does not influence the relation between the dependent and the independent variable. Remarkable is that, for industry 7 we find positive significant relations between the CEI score and the innovation performance and between the CSR score and the innovation performance (model 4), but when used as an interaction is shows a negative result. Furthermore, industry 6 shows a positive significant relation between the CSR score and the innovation performance (model 3).
Dishonest signaling
Industry number Average CEI score Average CSR score 1 90,38 74,55 2 81,65 60,78 3 81,69 59,00 4 71,71 65,00 5 67,06 61,38 6 65,78 62,15 7 56,22 61,48 8 89,02 62,59 9 81,51 57,06 10 64,66 60,62
Table 5: average CEI score compared with the average CSR score
5. Discussion
As shown in table 4 and discussed in the results section, the negative binomial regression results are not in line with the expected hypotheses. Therefore, no evidence has been found to support for hypothesis 1 and 2. Model 3 in table 4 even shows significant results to support for a negative relationship between the dependent and independent variable. However, when specifying the negative binomial regression to the individual industries we find mixed results. Industries 3 and 7 show partially support for hypothesis 1 and industries 2, 5, 6, 10 show negative significant relations (the contrary of hypothesis 1). Hypothesis 2 is not supported in an individual industry. Only industry 7 shows a significant result regarding the moderator, although this relationship is found to be negative which is the contrary of hypothesis 2.
5.1 Invisible diversity and innovation
As mentioned in the literature section do Geo and Zhang (2015) find evidence showing that firms located in an ENDA state are more innovative. Their reasoning taught us that those firms were more focused on non-discriminating activities, more focused on attracting the less visible forms of diversity and that those actions resulted in more patents and patent citations. The results from their study provided us with reasoning to predict hypothesis 1. However, they focused their research on firms being located in certain states. In this research the focus is on firms trying to incorporate invisible diversity through firm policy (the CEI score). Therefore, we can argue that being located in a state which includes invisible diversity has a positive influence on the innovation performance and that having an invisible diversity inclusive policy does not. Based on this result we can argue that there is a difference between an invisible inclusive location and an invisible inclusive policy and that they both have different effects on the innovation performance.
Furthermore, in the US there are still states who did not pass the ENDAs, this indicates that a lot of people still argue that being part of the LGBT community means acting in an immoral way (Johnston and Malina, 2008). Firms who include all forms of diversity might therefore scare away potential employees, investors and customers. This could be a reason why incorporating the LGBT community in the inclusive policy of the firm does not result in a higher innovation performance and could even result in a lower innovation performance. When we specify the results to the individual industries, we notice different outcomes. Some industries show a clear negative relation while others show a partially positive relation. This difference could for example occur because some firms in those industries are located in ENDA states while others are not. Although this is only speculation, future research should take this part into account.
result in less available resources for the R&D department, which eventually could result in a decrease in innovation performance.
5.2 Moderation effect
The results show that the overall model does not provide support for hypothesis 2. The results show an insignificant result what means that hypothesis 2 is unproven. However, when specifying the analysis per industry we find a significant negative effect of the moderator in industry 7. Although this finding is not in line with what was expected, does the literature review of this research provide us with some potential arguments for this occurrence. Since the literature review focused on the negative effect of a dishonest signal, an additional analysis has been conducted. The results of this analysis are shown in table 5. Table 5 shows that all industries have an average CEI score which is higher than the average CSR score, except for industry 7.
In the literature section we discussed the negative influence of a dishonest signal. When a firm’s policy does not match the way they work in practice, we talk about a dishonest signal (Colgan et al., 2007). A dishonest signal could result in a dishonest reputation among a firm’s stakeholder, which eventually will have a negative effect on the firm’s performance (Durcikova and Gray, 2009). Tabel 5 shows that industry 7 in general signals a high amount of LGBT inclusiveness, while the policy does not match this high score of LGBT inclusiveness. Although this does provide evidence for the negative finding for hypothesis 2 in industry 7, it is hard to say whether this has been the incentive of the signalers and whether all relevant stakeholders have interpreted the signals as being dishonest. Westphal and Zajac (1998) argue that the incentives of signalers have only limited been researched in the literature and although this research provides proof that dishonest signaling negatively influences the innovation performance, further research in the field of dishonest signaling is recommended.
diversity to find out what the effect of signaling diversity is on the innovation performance of the firm. They found that signaling diversity has indeed a positive influence on the innovation performance, since it shows that a firm understands and values multiple different stakeholders. The results of their research have been providing us with the arguments to predict hypothesis 2. However, as mentioned in the previous part, this research has a focus on the invisible form of diversity. As we already concluded in the discussion section of hypothesis 1, we can argue that diversity and invisible diversity do not have the same influences on the performances of firms. Multiple empirical studies have found that organizations are becoming more attractive to minorities when they include diversity in their advertising campaigns (e.g. Avery, Hernandez and Hebl, 2006; Walker, Feild, Giles, Armenakis, & Bernerth, 2009). However, this does not automatically mean that signaling diversity, or invisible diversity, makes a firm more attractive to both minorities and majorities. In some cases, for example in non-ENDA states, majorities might even feel less attractive to a firm when they actively signal (invisible) diversity.
5.3 Control variables
The expectations regarding the control variables are in general not met for this research. In the overall analysis, firm age shows no significance in model 1 and negative significant relations in models 2, 3 and 4. Furthermore, R&D expenses is found to be negative significant for all the different models in the overall analysis. When specifying the results to the individual industries we find mixed results. Some industries show significant positive results for the control variables, other significant negative results and some industries do not find significant results.
positively with the innovation performance. Furthermore, Hannan and Freeman (1984) provide reasons why firm age can negatively relate to the innovation performance of firms. They argue that mature firms have a higher probability of incumbent inertia compared to younger firms. This implies that older firms tend to do nothing or remain unchanged, which could mean that they are not willing to invest in innovation projects. Finally, firm age in this research is measured by the date of incorporation. Measuring firm age this way could have resulted in unreliable regression results, this due to the fact that the date of incorporation can in some cases be reset. When firms are derived from a merger their incorporation date is reset, this means that sometimes old firms might have a recent incorporation date and therefore a young firm age. This implies that the results from the firm age variable should be interpreted with caution.
financial resources as well. Therefore, the fourth industrial revolution is blurring the line between R&D expenses and innovation performance (Daemmrich, 2017).
5.4 Dependent variable
Based on the unexpected outcomes, and unproven hypotheses, the question arises whether the dependent variable in this research has been measured in the most accurate way. The use of patents as a quantitative measure for the innovation performance has been widely used in the literature since Scherer (1965) and Griliches (1981). However, according to Danguy, de Rassenfosse and van Pottelsberghe de la Potterie (2009), do patents vary greatly in their value. Most of the patents are being labeled as having low value and only a few are labeled as high value patents. This implies that even if firms have a lot of patents, it does not automatically mean that they are an innovative firm.
Where in the past patenting was solely a way to capture value from an innovation, patenting nowadays is also driven by another factor, namely the strategic behavior of the firm (Danguy et al., 2009). Using patents as strategic behavior is what is called strategic patenting. Strategic patenting means that firms use patents to block competitors (Blind, Edler, Frietsch and Schmoch, 2006). In this case, the use of patents is not only a measure for the innovation performance of a firm or its technical success, but patents are also a measure for the strategic behaviour of firms (e.g. Blind et al. 2006; Cohen, Nelson and Walsh, 2000; de Rassenfosse and Guellec 2009).
Furthermore, as mentioned before, there are many alternative ways to protect the value from an innovation. Even though patenting is the only mechanism which can be measured in a quantitative way, other appropriation mechanisms are widely used (e.g. secrecy, lead time). Harabi (1995) shows in his research that some firms try to avoid filling patents. The reason for this occurrence is that filling for a patent requires a firm to disclose a lot of information regarding the invention. Sometimes this disclosed information enables competitors to invent around the patent, which makes patenting unattractive for the inventor.
provide information about whether a firm has created an invention. Since being innovative means that a firm actually commercialized an invention, it remains unclear whether patents are a reliable representation of the innovation performance. We will elaborate on this part more in the limitation section of this paper.
6. Conclusion
The main objective of this paper is to provide insights in the effect of invisible diversity on the innovation performance. This study focuses on answering the question: ‘ What is the influence of invisible diversity on the innovation performance of a firm?
’. Invisible diversity compasses several forms of diversity. The
focus in this study has been on the invisible sexual orientation, in particular the LGBT community. Empirical evidence is obtained by using a sample of 204 firms from 10 different industries and so a longitudinal study is conducted between the years 2008 to 2017. By using a negative binomial regression analysis, both predicted hypothesis are unsupported. Incorporating invisible diversity does not show a positive relation with innovation performance. By analyzing all industries together the relationship is even found to be partially negative. Furthermore, the moderation effect of the invisible signaling is found to be unsupported. This implies that signaling invisible diversity has no indirect influence on the innovation performance. However, when analyzing the individual industries separately we find mixed results. This implies that some industries do benefit from an invisible diversity inclusive policy, while others do not. Furthermore, this separate analysis shows that invisible diversity signaling has no influence on the innovation performance unless the signal is found to be dishonest.
6.1 Theoretical implications
that claim. On the contrary, we do find partially support for a negative relation between invisible diversity and innovation performance. This implies that focusing too much on invisible diversity will have a negative effect on the innovation performance. Reasons for this outcome contain the fact that some firms might be located in non-ENDA states, which could mean that even though the firm embraces invisible diversity, the people in those states might react negatively towards the inclusive policy. Furthermore, becoming a fully LGBT-inclusive firm comes at a substantial price (Johnston and Malina, 2008), which could result in less available resources for other departments.
When researching the moderation effect we do not find significant results. This means that a LGBT-inclusive signal does not indirect influence the innovation performance. However, we do find one negative significant effect of the invisible diversity signal. This negative effect occurs when the industry uses a dishonest signal. These results contribute to the study of Westphal and Zajac (2001) who expect that firms might get a negative reputation when they have dishonest signals. Since this research shows that a dishonest signal has a negative indirect effect on the innovation performance, it contributes to their research. It also contributes to the research of Colgan et al. (2007), who wonder what the effect is of a mismatch between signal and practice. Our study shows that signaling less inclusiveness, compared with the policy, has no influence on the innovation performance while signaling more inclusiveness, compared with the policy, has a negative influence on the innovation performance. With this result we contribute to the study of Colgan et al. (2007) by showing that signaling a better practice compared to the policy has a negative influence on the innovation performance.
6.2 Practical implications
influence on the innovation performance. This shows that firms should be careful with what invisible diversity signals they send to their stakeholders.
Furthermore, since in general the relation between diversity and innovation has been proven to be positive, the results from this study might remind businesses that this does not contain all forms of diversity. Since, according to this research, invisible diversity does not always result in higher innovation performance is this something managers should pay attention to.
6.3 Limitations
This study contains several limitations. First of all, the innovation performance in this research is measured by the amount of patents a firm has applied for in a specific year. Even though we already discussed the use of patents as a measure of innovation performance in the discussion part of this research, there is another limitation regarding the use of patents. Namely, the EPO does not make a distinguishing between the different types of inventions. Therefore, the applied patent could be either a radical or incremental invention. Furthermore, it is unsure whether the patents applications are actually granted and eventually being commercialized. Together with the statements made in the discussion section of this paper, we could question whether the use of patents as a measure for innovation performance is the most accurate method.
Furthermore, the measure of the moderator is a limitation as well. The CSRHub measures the CSR score of the firms based on a score between zero and 100, although they use the same scale as the CEI score, it does not automatically mean that the scores could be compared with each other. It is unsure whether a perfect CEI score has the same value as a perfect CSR score. This results in the question whether the dishonest signal (as shown in table 5) is actually representative of a dishonest signal. Besides that does the CSRHub only provide the CSR scores at the current moment and not per year, this resulted in the use of the CSR score as a constant score over the years 2008 till 2017, while the other variables changed over time.
interpreted with caution. Finally, this research uses multiple different databases. This results in being really dependent on the available information in those databases.
6.4 Further research
A future research in the field of (invisible) diversity should put more emphasis on the differences between the invisible inclusive policy and the invisible inclusive states. Firms located in an ENDA states show a positive influence on the innovation performance, while an invisible diversity inclusive policy does not. Since the possibility rises that a non-ENDA state could influence the relation between an inclusive policy and innovation performance, a future research should exclude this.
Furthermore, a future research in the field of the signaling theory should put more focus on the incentives of the sender and the interpretation of the receiver, since these elements are not taken into account in this research.
7. References
Avery, D.R., Hernandez, M. and Hebl, M.R. (2006) Who’s Watching the Race? Racial Salience in Recruitment Advertising. Journal of Applied Social Psychology
34(1)
Badgett, M.V.L., Durso, L.E., Kastanis, A., Mallory, C. (2013) The Business Impact of LGBT Supportive Workplace Policies. Retrieved from The Williams Institute website
Blau, P. M. (1964). Exchange and power in social life. New York: Academic Press.
Blind, K., Edler, J., Frietsch, R. and Schmoch, U. (2006). “Motives to patent: Empirical evidence from Germany”. Research Policy, (35:5), pp. 655-672.
Camm, J.D. and Evans, J.R. (1996). Management Science: modeling analysis and interpretation.
Carrell, M. R., Mann, E. E., and Sigler, T. H., (Spring 2006). Defining workforce diversity programs and practices in organizations: a longitudinal study. Labor Law Journal. Spring 57 (1): 5-12.
Clair, J.A., Beatty, J. E. and Maclear, T.L. (Jan., 2005). Out of Sight but Not out of Mind: Managing Invisible Social Identities in the Workplace. The Academy of Management Review
. 30(1): 78-95
Claudio Aqueveque, (2005) "Signaling corporate values: consumers' suspicious minds", Corporate Governance: The international journal of business in society, Vol. 5 Issue: 3, pp.70-81
Colgan, F., Creegan, C., McKearney, A., Wright, T. (2007) Equality and diversity policies and practices at work: lesbian, gay and bisexual workers, Equal Opportunities International
, 26(6) 590-609
Connelly, B.L., Certo, S.T., Ireland, R.D. and Reutzel, C.R. 2011. Signaling Theory: A Review and Assessment. Journal of Management,
37(1), 39-67
Coxe, S., West, S. G., & Aiken, L. S. (2009). The analysis of count data: A gentle introduction to Poisson regression and its alternatives. Journal of Personality Assessment, 91: 121–136.
Creed, W. E. D., & Scully, M. A. 2000. Songs of ourselves: Employees' deployment of social identity in workplace encounters. Journal of Management Inquiry. 9: 391-413.
Daemmrich, A. (2017). Invention, Innovation Systems and the Fourth Industrial Revolution. National Academy of Inventors.
18(1): 257-265
Danguy, J., de Rassenfosse, G. and van Pottelsberghe de la Potterie, B. (2009). The R&D-patent relationship: An industry perspective,EIB Papers, European Investment Bank (EIB), Luxembourg, 14(1), 170-195
de Rassenfosse, G. and Guellec, D. (2009).”Quality versus quantity: Strategic interactions and the patent inflation”. Paper presented at the 4th annual conference of the EPIP association.
Durcikova, A., & Gray, P. 2009. How knowledge validation processes affect knowledge contribution.
Journal of Management Information Systems, 25(4): 81-107.
Ehie, I., and Olibe, K. (2010). The effect of R&D investment on firm value: An examination of US manufacturing and service industries. International Journal of Product Economies,
Fombrun, C., and Shanley, M. (1990). What's in a name? reputation building and corporate strategy. Academy of Management Journal,
33(2), 233-258.
Freeman, C., Soete, L., 1997. The Economics of Industrial Innovation 3rd ed. MIT Press, Cambridge, MA.
Frietsch, R., Schubert, T. and Neuhäusler, P. (2017). Secular trends in innovation and technological change, Studien zum deutschen Innovationssystem, No. 7-2017, Expertenkommission Forschung und Innovation (EFI)
, Berlin
Long, J. S., Freese, J. (2006). Regression Models for Categorical Dependent Variables Using Stata, Second Edition. College Station, TX: Stata Press.
Fullerton, M., (2013) Diversity and inclusion – LGBT inclusion means business. Strategic HR Review
,
12(3), 121-125
Gardner, W., Mulvey, E. P., & Shaw, E. C. (1995). Regression analyses of counts and rates: Poisson, over dispersed Poisson, and negative binomial models. Psychological Bulletin, 118: 392–404.
George Pohle, Marc Chapman, (2006) "IBM's global CEO report 2006: business model innovation matters", Strategy & Leadership, Vol. 34 Issue: 5, pp.34-40
Githens, R.P., (2009). Capitalism, Identity Politics and Queerness Converge: LGBT Employee Resource Groups. New Horizons in Adult Education and Human Resource Development
23(3), 18-31
Hannan, M.T. and Freeman, J. (1984). Structural inertia and organizational change. American sociological review
: 149-164
Hansen, J. (1992). Innovation, Firm Size, and Firm Age. Small Business Economics,
4(1), 37-44
Harabi, N. (1995). “Appropriability of technical innovations: An empirical analysis”. Research Policy, (24:6), pp. 981-992.
Hilbe, J. M. (2007). Negative binomial regression. New York: Cambridge
Hirschey, M., and Weygandt, J. (1985). Amortization Policy for Advertising and Research and Development Expenditures. Journal of Accounting Research
, Vol 23, 326-335
Hoffman, E. 1985. The effect of race ration composition on the frequency of organization communication. Social Psychology Quarterly 48(1): 17–26.
Hudson, S.W. jr. (2014). Diversity in the Workforce. Journal of Education and Human Development. 3(4): 73-82.
Huergo, E. and Jaumandreu, J. (2004), How Does Probability of Innovation Change with Firm Age?, Small Business Economics
, 22, (3_4), 193-207
Jayakumar, U.M. (2009). The Invisible Rainbow in Diversity: Factors Influencing Sexual Prejudice Among College Students. Journal of Homosexuality
Johnstone, R. A., & Grafen, A. 1993. Dishonesty and the handicap principle. Animal Behaviour, 46: 759-764
Johnston, D., & Malina, M. A. (2008). Managing sexual orientation diversity: The impact on firm value. Group and Organization Management, 33(5), 602–625.
Laufer, W.S. (2003). Social Accountability and Corporate Greenwashing. Journal of Business Ethics
, 43:
253-261
Miller, T., & Triana, M. D. C. 2009. Demographic diversity in the boardroom: Mediators of the board diversity— firm performance relationship. Journal of Management Studies, 46: 755-786.
Moorhead, C. 1999. Queering identities: The roles of integrity and belonging in becoming ourselves. Journal of Gay. Lesbian, and Bisexual Identity. 4: 327-343
Okoro, E. A. and Washington, M. C., (2012). Workforce diversity and organizational communication:
Analysis of human capital performance and productivity. Journal of Diversity Management.
Pinder, R. 1988. Striking balances: Living with Parkinson's disease. In R. Anderson & M. Bury (Eds.), Living with chronic illness: 67-88. London: Unwin Hyman
Pollitt, C. and Summa, H. (1997) Reflexive Watchdogs? How Supreme Audit Institutions Account for Themselves. Public Administration, 75:2 pp313–36.
Porter, M. E., 1981. The contributions of industrial organization to strategic management. Academy of Management Review
Quevedo, J., Pellegrino, G., Vivarelli, M. (2014): R&D drivers and age: Are young firms different? Research Policy 43 (9): 1544–1556.
Reimann, R. 2001. Lesbian mothers at work. In M. Bernstein & R. Reimann (Eds.), Queer families, queer politics: Challenging culture and the state: 254-271. New York: Columbia University Press
Roberge, M., Lewicki, R. J., Hietapelto, A., Abdyldaeva, A. (2011). From theory to practice:
Recommending supportive diversity practices. Journal of Diversity Management. 6 (3): 1-20.
Richard O. 2000. Racial diversity, business strategy, and firm performance: a resource-based view.
Academy Management Journal. 43(2): 164–177.
Richard, O., McMillan, A., Chadwick, K., and Dwyer, S. 2003. Employing an innovation strategy in racial diverse workforces: effects on firm performance. Group and Organization Management. 28: 107–126.
Sabat, I., Lindsey, A., Membere, A., Anderson, A., Ahmad, A., King, E., & Bolunmez, B. (2014). Invisible Disabilities: Unique Strategies for Workplace Allies.Industrial and Organizational Psychology, 7
(2), 259-265
Shefer, D. and Frenkel, A. (2005). R&D, Firm Size and Innovation: an Empirical Analysis. Technovation, 25(1): 25-32
Thiederman, S. (2008). Making diversity work. NY, NY: Kaplan Publishing
Uma M. Jayakumar PhD (2009) The Invisible Rainbow in Diversity: Factors Influencing Sexual Prejudice Among College Students, Journal of Homosexuality, 56:6, 675-700,
Vickers, M.H. 2001. Work and unseen chronic illness. London: Routledge.
Walker, H. J., Feild, H. S., Giles, W. F., Armenakis, A. A., and Bernerth, J. B. (2009). Displaying employee testimonials on recruitment web sites: Effects of communication media, employee race, and job seeker race on organizational attraction and information credibility.Journal of Applied Psychology, 94
(5),
1354-1364.
Wang, P., & Schwarz, J.L. (2010). Stock price reactions to LGBT nondiscrimination policies. Human Resource Management, 49, 195– 216.
Westphal, J. D., & Zajac, E. J. 1998. The symbolic management of stockholders: Corporate governance reforms and shareholder reactions. Administrative Science Quarterly, 43: 127-153.
8. Appendices
Appendix A: negative binomial regression analysis per industry (1, 3, 4)
Variables M2: industry 1 M3: industry 1 M4: industry 1 M2: industry 3 M3: industry 3 M4: industry 3 Firm age -0.0204157 x x 0.0702977 -0.0490049* x (0.0230784) x x (0.0450244) (0.0281111) x CEI score -0.0119554 x x 0.0201805 0.0268363* x (0.0461216) x x (0.0144646) (0.0160372) x CSR x x -0.3513562* x x x (0.2114137) x CEI x CSR x x x x x x Observations 78 x x 90 61 x Wald's Chi-square 1.06 x x 5.86 6.81 x Wald's p-value (0.5874) x x (0.0534) (0.0782) x Log-likelihood -21.788835 x x -107.51074 -28.73462 x
Standard Errors in the parentheses: ***p<0.01, ** p<0.05, *p<0.1
Variables M2: industry 4 M3: industry 4 M4: industry 4 Firm age -0.0089036 -0.008653 -0.0116826 (0.009199) (0.0095335) (0.0097239) CEI score -0.0051834 -0.0052713 0.1372016 (0.0078756) (0.008272) (0.0912094) CSR -0.0483346 0.0923674 (0.0644322) (0.1004991) CEI x CSR -0.0020746 (0.0013204) Observations 244 244 244 Wald's Chi-square 1.27 1.58 4.26 Wald's p-value (0.5303) (0.6632) (0.3721) Log-likelihood -156.68331 -156.3575 -155.10862
Appendix B: negative binomial regression analysis per industry (2, 5, 6, 7, 8, 9, 10) Variables M2: industry 2 M3: industry 2 M4: industry 2 M2: industry 5 M3: industry 5 M4: industry 5 R&D expenses 9.64E-08 7.24E-08 7.46E-08 9.47E-07*** 1.03E-06** 8.14E-07*
(9.09E-08) (0.0038778) (9.47E-08) (3.65E-07) (4.23E-07) (4.42E-07)
Firm age 0.0188734** 0.0221059*** 0.0219921*** -0.0417622*** -0.0418202*** -0.0410847*** (0.0058517) (0.0062321) (0.0062521) (0.0069244) (0.0068341) (0.0069806) CEI score -0.0171559*** -0.0197035*** -0.003895 -0.0135237** -0.0130162** -0.1286772 (0.003355) (0.0038778) (0.0558014) (0.0062675) (0.0053534) (0.089614) CSR 0.073899 0.0834848 -0.0368618 -0.1181493 (0.0492535) (0.0761349) (0.0889552) (0.108223) CEI x CSR -0.0001545 0.0018783 (0.0009232) (0.0014528) Observations 89 89 89 93 93 93 Wald's Chi-square 32.33 33.38 33.91 57.45 58.17 55.89 Wald's p-value (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Log-likelihood -367.34401 -366.32557 -366.3152 -172.42771 -172.34134 171/4853
Standard Errors in the parentheses: ***p<0.01, ** p<0.05, *p<0.1
Variables M2: industry 6 M3: industry 6 M4: industry 6 M2: industry 7 M3: industry 7 M4: industry 7
R&D expenses -5.64E-08 1.26E-07 2.09E-07 -4.52E-08 x -1.23E-07
(1.79E-07) (1.76E-07) (1.83E-07) (8.45E-08) x (1.67E-07)
Standard Errors in the parentheses: ***p<0.01, ** p<0.05, *p<0.1 Variables M2: industry 8 M3: industry 8 M4: industry 8 M2: industry 9 M3: industry 9 M4: industry 9 R&D expenses -1.36E-07*** -1.00E-07** -1.03E-07** -1.67E-07*** -1.68E-07*** -1.71E-07***
(4.08E-08) (4.89E-08) (4.94E-08) (5.55E-08) (5.66E-08) (5.76E-08)
Firm age -0.0080614 -0.0065654 -0.0064512 0.0316158 0.0317414 0.0286028 (0.0059453) (0.006001) (0.0061143) (0.0364643) (0.0366863) (0.0374561) CEI score 0.0085073 0.0109579 0.1011509 -0.0003435 -0.0001944 0.0734824 (0.0082453) (0.0083808) (0.2685316) (0.0127675) (0.0129275) (0.0734824) CSR -0.0613173 0.0897957 -0.0061447 0.1078687 (0.0418094) (0.4501466) (0.072779) (0.1886268) CEI x CSR -0.0014803 -0.0013025 (0.004397) (0.0019721) Observations 99 99 99 43 43 43 Wald's Chi-square 13.54 15.80 16.09 9.04 9.12 9.32 Wald's p-value (0.0036) (0.0033) (0.0066) (0.0288) (0.0582) (0.0969) Log-likelihood -498.65643 -497.53899 -497.48448 -160.91203 -160.90842 -160.6905
Standard Errors in the parentheses: ***p<0.01, ** p<0.05, *p<0.1
Variables M2: industry 10 M3: industry 10 M4: industry 10 R&D expenses 4.56E-08 3.67E-08 3.42E-08
(2.93E-08) (2.98E-08) (3.01E-08)
Firm age -0.0057032 0.001133 0.0038625 (0.0058658) (0.00917) (0.011572) CEI score -0.0097847* -0.0093435 0.0166978 (0.0057376) (0.0056954) (0.0627206) CSR -0.042993 -0.0171345 (0.043016) (0.0753109) CEI x CSR -0.0004203 (0.0010042) Observations 63 63 63 Wald's Chi-square 5.92 7.26 7.31 Wald's p-value (0.1158) (0.226) (0.1986) Log-likelihood -332.67933 -332.16807 -332.08055
