Optimizing effects of firms’ technological and non-technological processes on export-led innovation

ABSTRACT

This study contributes to research by evaluating the optimizing effects of innovation approaches on export-led innovation to ascertain a more efficient outcome. It uses the probit model with binary endogenous regressors to test the effects on 4,049 firms’ observations in Norway (1,784 manufacturing and 2,265 service firms). Results reveal that as export-led innovation induces a superior firm's innovation, it optimizes with innovation approaches – technological (STI) and non-technological (DUI) processes. While export-led innovation optimizes with STI in manufacturing firms’ product and process innovations, the optimizing effect is traceable to service firms’ process and service innovations. Nevertheless, export-led innovation optimizes with DUI in manufacturing firms’ service and New-to-Market innovations, and the effect only associates with service firms’ product innovations. Results show that while firms’ and export-led innovations optimize with the individual influence of STI and DUI, their interaction is inadequate. The study indicates that innovation methods are crucial for optimizing export-led innovation and present constructive policy implications that entrench invaluable decision-making that can enhance business performance.

KEYWORDS:

• Export-led innovation
• firms’ innovations
• STI/DUI processes
• optimizing effects
• probit model

1. Introduction

Abundant research reports have shown that exporting firms are more innovative than non-exporting firms (e.g. Antonelli 2019; Cieślik et al. 2021; Damijan, Kostevc, and Rojec 2017), and such exporting effect, which improves firms’ innovations, has been termed export-led innovation (Fassio 2018). The export-led innovation, referred to as the learning by-exporting hypothesis, entails that firms can access unique knowledge in foreign markets, facilitating experiences that lead to innovation (Bustos 2011; Fassio 2018; Lileeva and Trefer 2010). Studies investigating export-led innovation are evolving in this context, considering the influence of firms’ technological and non-technological processes. Specifically, the knowledge contribution on whether a firm's technological process – Science, Technology, and Innovation (STI onward) primarily facilitate firms’ and export-led innovations or improves with the inclusion of non-technological processes – Doing, Using, and Interacting (DUI afterward) to approach innovation is limited. Although there is an argument that combining the two approaches makes firms more innovative (Jensen et al. 2007), the literature suggests that the DUI approach to innovation tends to be dominantly associated with service firms (Chaminade, Lundvall, and Haneef 2018; Jensen et al. 2007; Sundbo and Gallouj 2000). With this assertion, it is tempting to argue that the DUI may not optimize firms’ and export-led innovations, for example, in exporting manufacturing firms. To this end, given that STI and DUI are knowledge facilitators and affect firms’ innovativeness (Crespi, Criscuolo, and Haskell 2008), this paper scrutinizes how STI and DUI can optimize export-led innovation. It aims to verify whether there is a more efficient way to achieve export-led innovation.

Innovation entails every scientific, commercial, and organizational activity deployed to facilitate technologically novel or superior products or services (OECD/Eurostat 1997). The STI has often used investment in research and development (R&D) activities to capture innovation, but critics believe the approach is not comprehensive. Presuming some firms may be more innovative than others based on their innovation methods, we presume that STI and DUI will optimize export-led innovation. While existing studies (Haus-Reve, Fitjar, and Rodríguez-Pose 2019; Jensen et al. 2007) have attempted this view, they are not directed to export-led innovation. Jensen et al. (2007) established that the complementary effect of STI and DUI produces higher innovative efforts than only one of the learning methods, but their finding does not confirm whether it applies to export-led innovation. Haus-Reve, Fitjar, and Rodríguez-Pose (2019) explicitly treated STI and DUI as collaboration methods, and their effects are primarily evaluated on product innovation. While their findings contrasted with Jensen et al. (2007), it does not verify export-led innovation. A study by Fassio (2018) investigating export-led innovation reveals that export destinations incentivize product and process innovations from technological learning and the foreign demand effect perspectives. While this is evident, the effects of STI and DUI are conspicuously not studied. Hence, our study is unique because it fills a vacuum by capturing the firm's STI – DUI interaction and their individual effects on export-led innovations and explicitly establishes its knowledge contribution from industry perspectives. In order to realize the aim of the study, we stimulate the investigation with the following research questions: How does export-led innovation affect firms’ innovative behavior? How do STI and DUI affect firms’ and export-led innovations? Are the STI and DUI optimizing effects industry specific?

Using the data obtained from the Community Innovation Survey (CIS afterward) and the accounting data from Norway for 2018, we apply a probit model with the binary endogenous regressors to test a model relating the impact of STI and DUI on export-led innovations in 4,049 Norwegian firms. Our analysis primarily contributes to the literature on the relationship between the firms’ STI and DUI and export-led innovations. As a pioneering study developing its analysis with a recent CIS measurement approach, our findings imply that innovation methods robustly optimize firms’ and export-led innovations as applied to a specific industry, illuminating a more efficient export-led innovation. Also, the study validates extant research that exporting firms are more innovative than non-exporting firms and enhances the generalizability of the evidence in a country like Norway. Moreso, as one of the emerging empirical studies on the self-selection hypothesis in a developed country, its evidence entrenched a foundation for valuable insights. Its contribution suggests a constructive economic policy implication enabling innovation approaches for business performance enhancement.

Next, we structure the remaining part of the paper, starting with the theoretical perspective in Section 2. The paper proceeds in Section 3 with the method, and Section 4 presents the results. Finally, Section 5 concludes with research and practical economic implications.

2. Theoretical perspective

2.1. Innovation modes

Practical results from various businesses have proven that firms presently performing well will likely surpass their current performance (Kemp et al. 2003). An important reason behind such differences in firm behavior is that innovation enables them to generate unique knowledge. Since knowledge is tantamount to learning, innovation can be regarded as a functional form of knowledge acquisition through learning. Meanwhile, the various knowledge-bases along the lines of innovation success are traceable to the two unrelated approaches to learning – STI and DUI (Jensen et al. 2007). While STI indicates an innovation mode ‘based on the production and use of codified scientific and technical knowledge’ (Jensen et al. 2007, 680), the DUI mode signifies an innovation emanating from what is learned from experiences that a product introduces while making or using it interactively (Jensen et al. 2007). Specifically, the STI approach to innovation primarily refers to how firms use and continue to develop codified scientific knowledge to conceptualize their innovative activities. One primary source of developing this understanding is linked with the R&D activities of large firms (Mowery and Oxley 1995). The learning technique associated with the STI mode can be used extensively as a global knowledge for innovation if unrestricted with intellectual property rights (Jensen et al. 2007).

While the STI learning approach supports the technological and quantifiable practice, the DUI learning mode is non-technologically oriented. The DUI learning approach is mainly regarded as innate knowledge that is highly restricted because it is based on experiences that are not easily codified (Jensen et al. 2007). This type of learning develops as an unexpected outcome of the firm's production and marketing undertakings in a relational manner of learning by doing, using, and interacting. Thus, it is a type of knowledge acquired on the job as work personnel develops new skill sets in solving the problems they face (ibid). However, as STI and DUI innovation modes differ, they use internal and external knowledge sources for innovation collaborations. Focusing on external knowledge partnerships, while research firms and universities are crucial to the STI innovation mode, the DUI innovation mode hinges on cooperation with customers and suppliers (Parrilli and Heras 2016).

2.2. Export-led innovation

There is evidence of reverse causality between innovation and export, and accordingly, the influence of innovation on export is referred to in the literature as the self-selection hypothesis (Bravo-Ortega, Benavente, and González 2014). In other words, some innovative domestic firms self-select themselves into the export markets since they become more productive than others with the ability to produce and introduce products that can compete in the foreign market. Their innovative capacity makes the international markets’ entry cost affordable and triggers foreign direct investments in the destination markets. This view suggests that innovation is necessary to be more efficient and, in this way, enables firms to access export markets (Wakelin 1998). On the contrary, the influence of export on innovation, regarded as the learning by-exporting hypothesis in the literature, implies that exporting allows firms to access new knowledge and acquire experience in enabling further innovation (Bustos 2011; Fassio 2018; Lileeva and Trefer 2010). The innovation effect of exporting embeds in international trade and is triggered by countries’ comparative advantages in world production opportunities (Ricardo 1817). Aghion et al. (2018) substantiate the positive effects of exporting on innovation, revealing that while high-productivity firms experience strong export shocks, the effect on low-productivity firms is negative. Antonelli and Feder (2021) posit that export drives the creative response and innovation activities leading to the firm adoption of technological innovation that amplifies productivity and export performance.

Evidence shows that export enables innovation from different perspectives and makes exporting firms more innovative. From the demand-pull perspective, the greater the demand for the exporting firm's products, the greater the motivation to introduce innovations (Fassio 2018; Rivera Batiz and Romer 1991 Woerter and Roper 2010;). Regarding competition, the more rivalry in the international markets, the more exporting firms are induced to continuously innovate to maintain a competitive advantage (Aghion et al. 2005; 2018; Schumpeter 1942). In this regard, firms’ exporting intention or behavior will increase R&D activities (Castellacci 2011; Siedschlag and Zhang 2015). In the context of knowledge spillover, as exporting firms interact with rivals in the international markets, the knowledge from more sophisticated rival firms becomes knowledge gained to become innovative (Andersson and Lööf 2009). Lastly, empirical evidence on the learning-by-exporting hypothesis confirms that exporters benefit from innovation-induced opportunities through learning (Crespi, Criscuolo, and Haskell 2008). The reason is that they have relations with experienced customers, especially in high-tech foreign markets (Fassio 2018).

2.3. Hypotheses

2.3.1. Innovative performance in exporting and non-exporting firms

In the attempt to test the effect of firms’ internationalization on innovation and productivity performance in Ireland, an argument supporting the innovative superiority of exporting firms shows that exporting firms are more innovative than non-exporting firms (Siedschlag and Zhang 2015). Similarly, Damijan, Kostevc, and Rojec (2017) confirm the innovative superiority of exporting firms over non-exporting firms in a study investigating how exports affect different innovation activities. De Fuentes, Niosi, and Peerally (2021) corroborated that exporting firms are more innovative based on investigating Canadian firms’ innovation and export relationships. Similarly, in the case of the United States, Bernard et al. (2007) prove that exporting firms are more innovative and vital than non-exporting firms. From this background, evidence suggests that the more exporting activities are performed, the quicker innovations are introduced and adopted, resulting in a higher productivity growth rate (Antonelli 2019). Following this view, we hypothesize that:

H1: The innovative performance of exporting firms is higher than that of non-exporting firms, confirming a positive export-led innovation.

2.3.2. Firms’ STI, export-led innovation, and innovative performance

In a study of the EU countries, Bravo-Ortega, Benavente, and González (2014) posit that firms employing R&D activities have higher incentives for exporting, especially in medium-sized firms, than in small firms. Regarding the firms’ strength in innovation, manufacturing firms tend to incline to technology and R&D competencies (STI) by highlighting efficient production methods (processes), but dominant service firms emphasize their workforce to consider organizational orientation to innovation (Tether 2005). This view suggests that while the R&D measures associate more with goods manufacturing firms, other innovation measures, such as competencies and expertise relating to non-technological innovation, are associated with dominant service firms (Sundbo and Gallouj 2000). In this context, it is plausible that goods exporting firms innovate primarily through the STI approach. Supposing exporting has an innovative effect on firms, and firms’ exporting behavior will increase R&D activities because of competition in the foreign market (Castellacci 2011; Siedschlag and Zhang 2015), we hypothesize that:

H2: Export-led innovation is positively associated with products and process-oriented innovations and is higher with the firms’ STI approach than with the DUI approach.

2.3.3. Firms’ DUI, export-led innovation, and innovative performance

When determining a more innovative firm between exporting and non-exporting firms, most studies have used investment in R&D (STI approach) to capture innovation. However, the Oslo Manual (OECD/Eurostat 1997) stresses that capturing the firms’ innovative efforts with the STI approach alone is not holistic. Similarly, Roberts and Vuong (2013) reveal that previous investment in R&D (STI approach adoption) may not always make a firm innovative, and their study shows no significant difference between firms that adopted the STI approach and those that did not (ibid). Further, Sundbo and Gallouj (2000), Jensen et al. (2007), and Chaminade, Lundvall, and Haneef (2018) argue that the STI approach to innovation is missing some firm innovative efforts associated with DUI and investments in human capital, noting that most of the service firms do not adopt the STI approach and are still innovative.

However, evidence suggests that firms primarily engaging in the DUI approach to innovation are dominant service firms (Chaminade, Lundvall, and Haneef 2018; Jensen et al. 2007; Sundbo and Gallouj 2000). This evidence suggests that the DUI approach may not be sufficient for some firms, for example, manufacturing firms, since it is dominantly associated with service firms. Nevertheless, manufactured products can be seen as goods and services, suggesting that the innovative performance visible to manufacturing firms innovating through the DUI approach can be traceable to services. The reason is that when a manufacturing firm produces a product, it needs considerable expertise in marketing orientation and organizational methods to market them. In this context, they need a knowledge base found with service firms by adopting a robust DUI approach. For example, goods export firms will witness stiff competitive services in foreign markets and will likely intensify innovative strategies to sustain a competitive advantage. In such a competition, they can adopt a robust DUI approach by investing in human capital and organizational method capable of optimizing service provision to secure a higher sales volume. From this background, we contend that:

H3: Export-led innovation is positively associated with service-oriented innovation and is higher with firms’ DUI approach than with the STI approach.

2.3.4. The STI and DUI interaction, export-led innovation, and innovative performance

Restricting innovation to the investment in R&D (STI) is insufficient; hence, it is not holistic to take the scientific approach as the primary source of innovation, productivity, and growth (Chaminade, Lundvall, and Haneef 2018, 34). In this regard, there is a claim that the interaction of STI and DUI approaches as an innovation strategy makes firms more innovative than the STI approach alone (Jensen et al. 2007). Again, Haus-Reve, Fitjar, and Rodríguez-Pose (2019) show that the interaction between the STI and DUI approaches produces a negative effect. Accordingly, the STI and DUI approaches are individually significant for firm innovation, and export does not influence their significance. Accordingly, instead of deploying the two innovation approaches simultaneously to fashion innovation, they best serve as substitutes (ibid). Parrilli and Elola (2012) stressed that combining the two approaches does not add any benefits in the absence of a competent workforce, and the STI approach is more relevant than the DUI approach. In the preceding, we note that there are conflicting pieces of claim on the interaction effect of STI and DUI on firms’ innovative performance and assume there is a need for more empirical evidence. However, supposing firms’ innovative performance is higher when they apply the STI and DUI individually, we posit that:

H4: The impact of export-led innovation on firm performance is higher when STI and DUI do not interact than when there is interaction.

3. Methods

3.1. Data

Based on the available data, variables like research intensity and firm size cannot be computed from one source alone. For this reason, we merged Eurostat's CIS 11 (period 2016-2018) provided by Statistics Norway and accounting data obtained from The Brønnøysund Register Center in Norway based on the 3-year reference period 2016–2018 to achieve the measures. The CIS is an innovation survey activity in enterprises that offer statistics relative to countries, types of innovators, economic activities, and size classes. The survey is designed to offer information on the innovativeness of sectors by type of firms, the different types of innovation, and various aspects of the development of an innovation, such as the objectives, sources of information, public funding, and innovation expenditures. We limit our analysis to the CIS data for 2018 because it complies with the current standard recommended by the Olso manual in the 4th Edition with an improved measurement approach. The accounting data presents non-financial limited companies’ overview of income statements, balance sheets, and analysis of figures by industries and regions.

3.2. Variables

3.2.1. Dependent variables

The dependent variable, which we denoted as Y in this study, may represent any innovation type. First, product innovation (prodINN) is defined as a new or enhanced good or service that varies significantly from the firm's prior goods or services and has been introduced on the market. Next, process innovation (procINN) is defined as implementing a new or considerably enhanced production or delivery method. The third dependent variable, service innovation (servINN), is defined based on the survey's response question about whether the company in 2016–2018 introduced new or significantly improved services. This definition is an improved version of the CIS that observes product innovation differently from service innovation. The dependent variable Y is assigned the value of ‘1’ if firms successfully innovate in the period surveyed and ‘0’ otherwise. Also, Y includes New-to-Market and New-to-Firm product innovation as the fourth and fifth innovation types, where the value of ‘1’ is for firms with such innovation in the period surveyed and ‘0’ otherwise (Ibid).

3.2.2. Independent variables (exogenous)

The independent variable is a vector of exogenous covariates in our study. The firm's absorptive capacity is one characteristic that enables it to generate and build on existing external knowledge and adapt it for innovative activities. That is, the firm's absorptive capacity is tantamount to learning-enabled capacity. However, it is a multi-item measurement variable, and it appears there is no consensus on the specificities of its measure. To this end, we recognize that firms can adopt different learning methods or capacities to gain external knowledge through collaboration factors that affect innovation, and Jensen et al. (2007) classify these learning methods into the STI and DUI approaches. Thus, we identify the first set of variables, STI and DUI, from the context of absorptive capacity. Following this view, Haus-Reve, Fitjar, and Rodríguez-Pose (2019) observe that the individual application of the two approaches is better for firms’ innovation than their interaction. More so, there is evidence that a firm's learning approach (the STI and DUI) can affect its innovativeness (Crespi, Criscuolo, and Haskell 2008). Hence, we include the STI and DUI denoted with (tech_STI) and (nontech_DUI), respectively, and their interaction among the independent variables. As Haus-Reve, Fitjar, and Rodríguez-Pose (2019) measured the variables, we determined the tech_STI with firms that reported using universities, research institutes, and consultancy firms (scientific collaboration) as the most critical partners for innovation. The value ‘1’ represents firms that engage in the STI approach, and ‘0’ otherwise. We defined the nontech_DUI with firms that reported using linkages with suppliers and customers (supply chain collaboration) as the most critical partners for innovation and denoted the firms that engage in the DUI approach with ‘1’ and ‘0’ otherwise.

Following Damijan, Kostevc, and Rojec (2017), another variable from the context of absorptive capacity is the share of total R&D expenditure in firm turnover (research intensity). We include research intensity measured by the share of R&D expenditure (external and in-house) divided by the firm's turnover for the surveyed period. While the information on the share of R&D expenditure is obtained from the CIS data, we obtained the firm's turnover from the accounting data. Further, given that innovative success increases with firm size (Knott and Vieregger 2015), one of our independent variables emanating from the accounting data include firm size, measured as the firm's number of employees during the survey period. We classified firm size into the small firm (Sfirm), medium firm (Mfirm), and large firm (Lfirm), with 10 - 49, 50 - 249, and 250 and above, respectively, where the reference category is classified as firms less than 10. Another covariate in our independent variables includes firms’ market concentration (maktCON). We classified the market concentration into local and export markets to determine whether they affect innovation differently. We apply the Hirschmann – Herfindahl index to determine local and export market concentrations with the industry index provided in the CIS data classifications. This approach toes Damijan, Kostevc, and Rojec (2017) computation of a measure of market strength for the firms using the percentage of firm turnover locally and abroad for each market concentration type.

Also, one of the firms’ characteristics that can enhance innovative capacity is embedded in labor competencies. We identify labor competencies as one of the absorptive capacity measures, and Haus-Reve, Fitjar, and Rodríguez-Pose (2019) and other extant literature have measured this variable as the share of the firm's workers with university degrees. In this study, we try to be more direct by looking into the employees’ education and skills associated with the firms’ activities instead of merely looking into their education degrees as a yardstick. Therefore, as one of the independent variables, we measured labor competencies with firms that improve their workers by setting time aside for further education, skills development, and training within the company to organize work. Firms that improve their workers are represented with ‘1’ and ‘0’ otherwise.

Investing in advanced technologies and more efficient equipment will likely increase production capacity, improve product quality, and cut costs. It presents manufacturing firms with the capacity to manufacture commodities and provide services to customers promptly. Thus, innovating firms will leverage technology investments to improve productivity and efficiency to achieve competitive advantage (Mechling, Pearce, and Busbin 1995). We used two survey questions to capture new technology investment (newtechINV): during the surveyed period, had the company procured machinery, equipment, or software based on technology already used in the enterprise? Also, during the surveyed period, had the company procured machinery, equipment, or software based on new technology unique to the enterprise? We denoted the fulfilling firms with ‘1’ and ‘0’ otherwise. Following Golovko and Valentini (2014) and Fassio (2018), export-led innovation cannot be measured directly in our study but as a measure of various innovation outcomes.

Following some scholars (e.g. De Fuentes, Niosi, and Peerally 2021), innovation strategies, which this study identified as business expansion strategies (BUSexpansion) and included in the exogenous variables, are usually essential determinants of innovation. Since innovation strategies have economic implications, we relied on the following question from the CIS survey to measure BUSexpansion: how important were the following strategies for the company's financial results, (i) improvement of existing goods and services, (ii) development and launch of entirely new goods and services, (iii) low price of goods or services (price-leading strategy), (iv) high-quality goods or services (quality management strategy) (v) offer of a wide range of products or services, (vi) focus on a single / a few essential products/services, (vii) prioritizing and serving established customer groups, (viii) attempts to reach new customer groups, (ix) standardized/fixed products or services, and (x) development of customer-specific solutions. We indicate firms that engage in business expansion strategies with a value of ‘1’ and ‘0’ otherwise (see Appendix A1 for details of measurement).

3.2.3. Endogenous covariate and instrument variables

Evidence suggests that as exporting activities improve firms’ innovation, it is not an exogenous factor (De Fuentes, Niosi, and Peerally 2021). The reason is that firms self-select themselves into export markets, and such selection bias creates a self-selection-based endogeneity problem that violates the strict exogeneity assumption. In dealing with such a problem, the empirical context associated with the instrumental variable approach suggests that the selection equation ensures that the endogenous covariate constructs are significant with unbiased coefficient estimates in the outcome equation. In this context, we identify two instruments that can directly affect export (i.e. the endogenous covariate) and indirectly affect innovation activities (the outcome equation). According to García-Vega and Huergo (2019), foreign R&D outsourcing exclusively incentivizes exporters’ process innovation. Although the available data for this study did not distinguish between foreign outsourcing and domestic outsourcing R&D, we only included outsourcing R&D in the model as an instrument for export status. We assume that because of competition, exporting firms will conduct R&D activities in the foreign market to capitalize on various capabilities, reduce the risk and failure of costly innovation, or realize cost advantages. Specifically, concerning marketing strategies, exporters will source innovative knowledge (market orientation) about foreign markets to beat the competition. Such orientation facilitates an efficient production process that supports export. We relied on the question: did firms buy R&D services from others during the surveyed period to proxy firms that outsource R&D? We designate the performance with ‘1’ and ‘0’ otherwise.

An indicator in the data used for this study gave information that firms receive external funds not channeled toward innovation activities. Since this information indicates that the financial support is not for innovation purposes, we suspect it might capture the government's financial support for firms on export activities. Following De Fuentes, Niosi, and Peerally (2021), we include this as information on the government's financial support (Finnsupp) for firms on export activities. Thus, we relied on the following questions to capture ‘Finnsupp’ for firms on export activities: Did the company receive financial support from local/regional authorities or government-run institutions during the period? Did the company receive financial support from national authorities or government-run institutions? Did the company receive financial support from EU authorities or institutions during the period? The firms that receive external funds are denoted with ‘1’ and ‘0’ otherwise (see AppendixA1 for variable measures).

3.3. Model specification and estimation procedure

Several studies have shown a correlation between innovation and exports (Cieślik et al. 2021; Damijan, Kostevc, and Rojec 2017), and these studies (Aghion et al. 2018; De Fuentes, Niosi, and Peerally 2021) have tested the interplay between innovation and exports to investigate causality. While they evidenced that export increases the incentives for innovation, it creates endogenous covariates problems in the association. While we acknowledge the two-way relationship between innovation and export, we only focus on the direction from export to innovation. In this paper, innovation and exports are binary (discrete outcome and endogenous variables), and when a probit model contains discrete outcomes and endogenous covariates, a standard Two-Stage Least Square (2SLS) estimator poses complications (Maddala 1983; Mroz and Guilkey 1992). Also, as 2SLS and Maximum-Likelihood (ML) Estimators are deemed consistent, only the ML Estimator is asymptotically efficient (Bollen, Guilkey, and Mroz 1995, 118). Hence, we adopt the extended probit model (eprobit) that uses ML Estimator for binary endogenous covariates and discrete outcomes.

Adapting to Arendt and Holm (2006), we posit that a probit model with the identified parameters for outcome Y with binary endogenous covariates S has the structural form of equations 1(1) $Y=1\left(\mathrm{\alpha S}+\mathit{X}{\text{β}}_i+v_Y\right),>0$(1) and 2(2) $S=1\left(\mathit{Z}{\gamma }_i+v_S\right),>0$(2) : (1) $Y=1\left(\mathrm{\alpha S}+\mathit{X}{\text{β}}_i+v_Y\right),>0$(1) (2) $S=1\left(\mathit{Z}{\gamma }_i+v_S\right),>0$(2) $(v_Y,v_S|X,\mathit{Z})\sim N\left(0,0,1,1,\rho \right).$

In equation 1(1) $Y=1\left(\mathrm{\alpha S}+\mathit{X}{\text{β}}_i+v_Y\right),>0$(1) , Y is a binary indicator capturing firm innovation types in our study. X is a vector of exogenous covariates affecting innovation type Y. S is a binary endogenous covariate, as is then estimated with instrument variables Z. While α is the regression coefficient for export status, β_i, and γ_i are vectors of regression coefficients for the exogenous and instruments variables. As subscript i represent firms in the model, 1(.) is the indicator function with the value ‘1’ if the expressions in the brackets are true and ‘0’ otherwise. Also, while ν_Y and ν_s are unobserved errors, N(0,0,1,1, ρ) is the basic bivariate normal distribution with correlation coefficients ρ = Corr (ν_Y, ν_s).

An eprobit model fits models for cross-sectional data and a binary-outcome probit regression model that accommodates any combination of endogenous covariates with exogenous variables (i.e. binary, continuous, and ordinal endogenous covariates). The estimators implemented are ML Estimators of Wooldridge’s (2010, 470–471).¹ It accounts for endogenous covariates problems whether they arise independently or in combination. We follow the eprobit procedure to fit equations 1(1) $Y=1\left(\mathrm{\alpha S}+\mathit{X}{\text{β}}_i+v_Y\right),>0$(1) and 2(2) $S=1\left(\mathit{Z}{\gamma }_i+v_S\right),>0$(2) simultaneously by regressing Y on X with an endogenous specification that regresses S on instrument variables Z with probit and robust standard error options, where ν_Y and ν_s are multivariate normal with mean 0 and covariance. $\sum =\left[\begin{array}{cc}1& {\rho }_{1s}\\ {\rho }_{1s}& \sum _s\end{array}\right]$We tested the four hypotheses developed in this study from three modeling perspectives. First, we generally specify a baseline model accommodating the endogenous and exogenous variables, emphasizing the log of firm size, STI, and DUI. In order to identify how firm size explicitly affects firms’ innovation in the estimations, we specify a second model highlighting the dummy for firm sizes. Thus, the study's analysis baseline and firm-size dummy models foster hypotheses 1–3. In the third modeling perspective, we tested the interaction between the STI and DUI to verify whether their combined effect is higher than their separate effect on export-led innovation. This perspective tested hypothesis 4, and we obtained the results using margins post-estimation command for eprobit regression in Stata 17. The eprobit procedure assumes that if the Wald test p-value of the null hypothesis that export status (S) is endogenous cannot be rejected in the model, there must be a significant correlation (p < 0.05) between the outcome equation and the endogenous equation errors (i.e. equations 1(1) $Y=1\left(\mathrm{\alpha S}+\mathit{X}{\text{β}}_i+v_Y\right),>0$(1) and 2(2) $S=1\left(\mathit{Z}{\gamma }_i+v_S\right),>0$(2) ) assuming joint normality. If this is the case, it is statistically assumed that the model is identified.

3.4. Descriptive statistics

The CIS survey 2018 observes small firms as firms between 10–49 employees and medium and large firms between 50–249 and 250 and above 250, respectively. The study focuses on firms in the service and manufacturing industries. Tables 2 and 3 illustrate the informative statistics of the variables included in the models for 1,784 manufacturing and 2,265 service firms, totaling 4,049 sampled firms. All variables except research intensity, firm size, export market, and local market concentration are binary. The variance inflation influence is less than 5 (VIF < 5), both on the average and across all variables indicating that multicollinearity is not an issue for the estimates. The Breusch–Pagan test (p < 0.000) shows the data is heteroscedastic; hence, we use the vce robust standard error option to evaluate the statistically significant estimates.

Table 1. Summary of hypothesized relationships.

Hypotheses	Relationships
H1	The innovative performance of exporting firms is higher than that of non-exporting firms, confirming a positive export-led innovation.
H2	Export-led innovation is positively associated with products and process-oriented innovations and is higher with the firms’ STI approach than with the DUI approach.
H3	Export-led innovation is positively associated with service-oriented innovation and is higher with firms’ DUI approach than with the STI approach.
H4	The impact of export-led innovation on firm performance is higher when STI and DUI do not interact than when there is interaction.

Table 2. The characteristics of the CIS sample (2016 - 2018) by firm size.

Firm type	Obs.	Mean	Std. Dev.	Obs.	Mean	Std. Dev.
Manufacturing firms Service firms
Small firms	790	0.44	0.49	1,103	0.48	0.49
Medium firms	608	0.34	0.47	641	0.28	0.45
Large firms	104	0.06	0.23	89	0.04	0.19

The firm sizes are dummy variables; the baseline category is the firms with less than ten employees and is not included in the model.

4. Results

In order to determine the instrument's validity and whether the estimator used is appropriate, we verified the exports’ exogeneity and the instruments’ relevance. The LM Wald test proposed by Lochner and Moretti (2015) rejects the exogeneity (p < 0.05) in all but one model. The LM test is similar to the Durbin – Wu – Hausman (DWH) test but has proven more relevant for the exogeneity test in nonlinear models (Babington and Cano-Urbina 2016). Nevertheless, when we consider the test results from the Durbin – Wu – Hausman (DWH) perspective, the exogeneity of our estimator was rejected in all the models (p < 0.05). Further, we establish that the instruments are relevant by testing whether they are weak using the First-Stage F Statistic. Following Stock and Yogo (2001), the first-stage F statistic must be large, usually above 10, for the reliable 2SLS assumption. The test result (F ≥ 12.64) indicates that the instruments used are relevant; thus, we proceeded with the estimation.

Table 3. Basic characteristics of the variables from CIS sample (2016 - 2018).

Variables	Sample for service firms			Sample for Manufacturing firms
	Obs Mean Std. Dev			Obs Mean Std. Dev
Dependent variables
Service innovation	2,265	0.44	0.49	1,784	0.21	0.41
NEW-to – MKT product innovation	2,265	0.33	0.46	1,784	0.30	0.45
NEW-to – FIRM product innovation	2,265	0.34	0.47	1,784	0.36	0.48
Product innovation	2,265	0.36	0.47	1,784	0.50	0.50
Process innovation	2,265	0.56	0.49	1,784	0.53	0.49
Independent variables
Log of research intensity	2,265	−0.31	1.10	1,784	–0.87	1.70
Log of firm size	2,265	3.35	1.15	1,784	3.56	1.19
Labour Competence	2,265	0.93	0.25	1,784	0.92	0.26
Technological approach (STI)	2,265	0.18	0.38	1,784	0.20	0.40
Nontechnological approach (DUI)	2,265	0.15	0.36	1,784	0.15	0.36
EXPORT market concentration	2,265	0.04	0.09	1,784	0.05	0.10
LOCAL market concentration	2,265	0.00	0.00	1,784	0.00	0.00
Business expansion strategies	2,265	0.98	0.11	1,784	0.98	0.12
Outsourcing	2,265	0.25	0.43	1,784	0.29	0.45
Export status	2,265	0.41	0.49	1,784	0.46	0.49
Financial support	2,265	0.20	0.43	1,784	0.29	0.45
New technology investment	2,265	0.69	0.46	1,784	0.70	0.45

4.1. Manufacturing firms

The estimation results show significant correlations (p < 0.05) between the model's primary outcome and endogenous equation errors, indicating that the model is identified. For our binary independent variables, marginal effects suggest the degree to which predicted probabilities change from 0 to 1 at the sample mean. The marginal effect on continuous variables expresses the extent to which the change in the outcome variable will occur based on a unit change in the predictor variable. The marginal effect at sample means presented in Table 4 shows that research intensity is insignificant and that firm size only significantly affects the New-to-Firm product innovation. In Table 5, where we use a dummy firm size, our estimate further reveals that research intensity is still insignificant, and firm size remains significant for New-to-Firm product innovation, considering it is large. The finding suggests that when larger firms are developing a new product that needs to succeed in the market, a unit decrease in investment in innovation still attracts a probability that the product thrives. Supporting our finding, Crespi and Zuniga (2012) noted that after large firms decide to invest in innovation, it does not affect the intensity after the decision is taken. The reason is that they are at an advantage due to economies of scale associated with production and R&D entrenched in their enormous human resources, resulting in innovation. Also, the finding that research intensity is insignificant suggests that it may not always be vital for innovation. This view may correlate with Roberts and Vuong’s (2013) assertion that R&D investments do not necessarily make a firm more innovative. However, since the impact of R&D activities is not usually immediate, its insignificance for innovation in this study can be due to time constraints.

Table 4. Average marginal effects and baseline estimation for manufacturing firms.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	0.0036	0.0141	0.0118	0.0126	0.0035
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Firm size(log)	–0.0008	0.0000	0.0025	0.0274**	–0.0106
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Labor competence	0.0354	0.1520***	0.1410**	0.0618	0.0770
	(0.04)	(0.04)	(0.05)	(0.05)	(0.04)
Tech. approach (STI)	0.1840***	0.2030***	0.0953***	0.1550***	0.1170***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Non-tech. approach (DUI)	0.2910***	0.2160***	0.1520***	0.2330***	0.1940***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Export market conc.	0.3050*	0.0379	0.0989	0.0480	0.2380*
	(0.13)	(0.13)	(0.11)	(0.13)	(0.11)
Local market conc.	7.9340*	9.4690**	4.0670	9.4110**	5.3390
	(3.13)	(3.31)	(2.77)	(3.27)	(3.02)
Business expansion	0.5430***	0.3730*	1.3830***	1.8370***	0.3080*
	(0.16)	(0.15)	(0.07)	(0.07)	(0.14)
New tech. investment	0.0502	0.1610***	0.0807**	0.0730**	0.0309
	(0.03)	(0.03)	(0.02)	(0.03)	(0.03)
export	0.4970***	0.2820***	0.1590*	0.2010*	0.3940***
	(0.04)	(0.08)	(0.07)	(0.09)	(0.05)
Log likelihood	−2103.64	−2176.41	–1960.74	–2173.26	–2062.47
chi-square test	805.95***	354.66***	1084.42***	1528.83***	444.46***
Wald test of exogeneity	34.19***	10.70***	7.00***	4.46**	33.40***
No. of observations	1784	1784	1784	1784	1784

Marginal effects; Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

Table 5. Average marginal effects. Estimation with firm class dummy.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	0.0041	0.0145	0.0110	0.0115	0.0026
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Large firm	0.0591	0.0664	0.0373	0.1540**	0.0124
	(0.06)	(0.06)	(0.05)	(0.06)	(0.05)
Medium firm	–0.0467	0.0084	–0.0400	0.0329	–0.0560
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Small firm	–0.0185	0.0378	–0.0193	–0.0064	0.0134
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Labor competence	0.0369	0.1470***	0.1450**	0.0689	0.0721
	(0.04)	(0.04)	(0.05)	(0.05)	(0.04)
Tech. approach (STI)	0.1780***	0.2000***	0.0930***	0.1520***	0.1130***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Non-tech. approach (DUI)	0.2910***	0.2150***	0.1540***	0.2360***	0.1940***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Export market conc.	0.3040*	0.0318	0.0974	0.0498	0.2300*
	(0.13)	(0.13)	(0.11)	(0.13)	(0.12)
Local market conc.	8.2660**	9.4310**	4.4590	9.7410**	5.5740
	(3.12)	(3.31)	(2.77)	(3.26)	(3.007)
Business expansion	0.5440***	0.3710*	1.3950***	1.8470***	0.2970*
	(0.16)	(0.15)	(0.10)	(0.08)	(0.14)
New tech. investment	0.0509	0.1620***	0.0836***	0.0747**	0.0344
	(0.03)	(0.03)	(0.02)	(0.03)	(0.03)
Export	0.5030***	0.2820***	0.1530*	0.1960*	0.3900***
	(0.04)	(0.08)	(0.07)	(0.09)	(0.05)
Log likelihood	−2100.81	–2174.89	–1958.37	–2171.57	–2058.15
chi-square test	814.15***	361.00***	887.04***	1403.33***	445.16***
Wald test of exogeneity	34.19***	10.70***	7.00***	4.46**	33.40***
No. of observations	1784	1784	1784	1784	1784

Marginal effects; Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

In Tables 4 and 5, while the marginal effect of labor competence is positively significant for process and service innovation, new technology investment is significant for process, service, and New-to-firm product innovations at their sample mean. The statistical significance of the marginal effect of the labor force aligns with Adewumi (2022), who found that a quality labor force positively mediates disruptive innovations across countries. The result found for technology investment is unsurprising, and the ground is that firms will invest in advanced technologies to remain in business since it facilitates production capacity and minimizes production costs. Hence, this finding confirms that innovative firms will leverage technology investments to enhance productivity and efficiency for competitive advantages (Mechling, Pearce, and Busbin 1995).

Similarly, as the marginal effect of concentration in the export market is positively significant for product and New-to-Market product innovation, concentration in the local market is statistically significant for the product, process, and New-to-Firm product innovation, as presented in Tables 4 and 5. The positive effects contrasted with Damijan, Kostevc, and Rojec (2017), who showed a negative effect in their study. Given that evidence on market concentration is inconsistent, our results support the positive findings for studies in developed countries (e.g. Schmutzler 2013; Tishler and Milstein 2009). Our study reveals that the marginal effect of the local market is significantly higher than any other covariate in the model, and the probability that the local market affects the Norwegian firms’ innovation successes is higher than the export market. While this result is a surprise, a possible explanation for this outcome is that the dimension of the feedback effect caused by the buyer-seller interaction in the Norwegian market puts the firms under intense competitive pressure to invest in innovative ways of satisfying their customers locally more than the exporting firms. For example, Norway has a market size of about 5.4 million people, suggesting it is relatively small. Thus, as local firms compete with importers to satisfy the market, non-exporting firms are more pressurized to intensify innovative ways to gain market shares and remain in business. On the other hand, since exporting Norwegian firms are concerned majorly with the foreign market, their innovative effort may be less pronounced, mainly if they compete in a less competitive foreign market.

In Tables 4 and 5, the marginal effect at sample means indicates that the likelihood of adopting the DUI approach to firms’ innovative performance is higher than the STI approach in Norway. Also, the result shows that the positive and statistical significance of the marginal effects of STI and DUI approach to innovation and business expansion strategies has the probability of improving all innovation types. The evidence aligns with Clausen et al.(2012) and De Fuentes, Niosi, and Peerally (2021) that STI and DUI's marginal effects are crucial for firms’ innovation, as displayed in Tables 4 and 5. Accordingly, the variations in these innovation strategies are critical factors for maintaining innovation activities across firms (Clausen et al. 2012). Further, the STI and DUI approach to innovation as a measure of information sources indicates that absorptive capacity is also critical for innovating firms.

Tables 4 and 5 indicate that export has significant and positive marginal effects on all innovation types. The positive association with innovation validates hypothesis 1, which states that exporting firms are more innovative than non-exporting firms. The result suggests that being an exporter has the probability of increasing the firm's innovations, confirming a positive export-led innovation. It supports the view that as exporting firms compete in the international markets, the knowledge from more experienced competitors becomes knowledge gained that enables superior innovation efforts (Andersson and Lööf 2009). Thus, the finding confirms existing studies that the innovative performance of exporting firms is higher than non-exporting firms (Antonelli 2019; Antonelli and Feder 2021; Cieślik et al. 2021; Damijan, Kostevc, and Rojec 2017; De Fuentes, Niosi, and Peerally 2021; Siedschlag and Zhang 2015). It further corroborates previous studies that export positively influences innovation (Aghion et al. 2018; De Fuentes, Niosi, and Peerally 2021).

However, controlling for R&D intensity and other variables like business expansion strategies, firm size, and other absorptive capacity measures intervene in export's effect on innovation in the model specification. These covariates enable us to assume that export status is not capturing a scale effect associated with competition advantages relating to exporting. Also, including variables such as labor competence and the DUI associated with suppliers and customers make us exclude that technological-oriented factors mainly drive our results.

Hypothesis 2 posits that export-led innovation is positively associated with products and process-oriented innovations and is higher with the STI approach than the DUI approach. The result shows that the hypothesis is confirmed as predicted but with an exception. The upper left of the panel in Figure 1 shows the point where firms are engaging in STI (blue line, STI = 1) and DUI (red line, DUI = 1) approaches. It suggests that the association between export-led and product innovations is higher with firms’ STI marginal effect of 57.6 percent than with the DUI marginal effect of 48 percent. Also, the marginal effect provided on process innovation at the panel's upper right is similar to product innovation. The association between export-led and process innovations is higher with an STI marginal effect of 31 percent than with the DUI marginal effect of 24.4 percent. Further, in the panel's middle right, the positive effect of export-led innovation on New-to-Firm product innovation optimizes with an STI marginal effect of 23 percent than the DUI marginal effect of 21 percent. The exception to hypothesis 2 is that at the lower left of the panel in Figure 1, the DUI has a higher marginal effect of 47 percent than the STI marginal effect of 42 percent on the association between export-led innovation and New-to-Market product innovation.

Figure 1. Marginal and interaction effects of the STI and DUI approaches for manufacturing firms.

In the preceding, export influences the significance of our findings. Our result improves on Haus-Reve, Fitjar, and Rodríguez-Pose (2019), who conducted a similar study and demonstrated that export has no effect. Our finding reveals that a discrete increase in the previous use of the firm's DUI approach will not increase the probability of using the STI approach in the cases of product, process, and New-to-Firm product innovations. Thus, as a discrete increase in export-led innovation has the probability of improving firms’ innovation in manufacturing firms, a discrete increase in the application of the STI approach has the probability of optimizing the effect, given that the innovation is product and process-oriented. However, while this finding supports Clausen et al. (2012) that the STI approach to innovation has a higher probability of innovation success, this does not hold in the case of New-to-Market product innovation.

The result in the panel's middle left in Figure 1 confirms hypothesis 3, which states that export-led innovation is positively associated with service-oriented innovation and is higher with firms’ DUI than with the STI approach. As presented in the panel's middle left of Figure 1, the impact of export-led innovation on service innovation has a probability of optimizing with DUI's marginal effect of 21 percent than the STI's marginal effect of 15 percent. The result indicates that the previous use of the STI approach is more robust at improving the DUI approach regarding service innovation. Here, DUI optimizes the marginal effect of export-led innovation on service innovation in manufacturing firms. In this context, we identify that hypothesis 2 strengthens the view that the influence of STI (technological processes) on export-led innovation optimizes product innovations in manufacturing firms (Bravo-Ortega, Benavente, and González 2014; Castellacci 2011). Also, by comparing STI's relative impact on export-led innovation, findings complement Clausen et al. (2012) that exporting firms engaging in STI are innovating more than those engaging in DUI.

To test hypothesis 4, we estimate a specification for the interaction of STI and DUI and present the results in Table 6.² First, we ignored the export-led innovation, and the interaction has significant negative marginal effects on all innovation types. However, the result is similar but statistically more significant when considering the earlier ignored export-led innovation. Based on that, we allow the effect of export in our result. The finding in Table 6 suggests that at the sample mean, the marginal effect of STI – DUI interaction has a probability of decreasing firms’ innovative performance and the possible optimizing effect of STI and DUI on export-led innovation. Since results are similar for all innovation types, we use product innovation for our analysis. At the lower right of the panel in Figure 1, we present the result for product innovation. It demonstrates that when STI and DUI interact (STI*DUI = 1), the probability that export-led innovation influences product innovation has a marginal effect of 0.00 percent (see Table 7 for manufacturing firms). Thus, the left-to-right downward slope indicates the predicted probability effect of STI – DUI interaction resulting from the discrete change from 1 to 0 on firms’ innovation and the optimizing effect of export-led innovation.

Table 6. Average marginal effects of STI and DUI interaction for manufacturing firms.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	0.0049	0.0136	0.0120	0.0127	0.0039
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Firm size(log)	−0.0016	0.0007	0.0018	0.0268**	−0.0115
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Labor competence	0.0346	0.1530***	0.1400**	0.0610	0.0762
	(0.04)	(0.04)	(0.05)	(0.05)	(0.04)
STI-DUI interaction	–1.5670***	−1.3340***	–0.8350***	–1.2910***	–1.0490***
	(0.15)	(0.15)	(0.15)	(0.17)	(0.15)
Export market conc.	0.2970*	0.0416	0.0960	0.0436	0.2330*
	(0.13)	(0.13)	(0.11)	(0.13)	(0.12)
Local market conc.	8.0270*	9.3960**	4.1160	9.4550**	5.4090
	(3.13)	(3.30)	(2.78)	(3.26)	(3.02)
Business expansion	0.5310***	0.3800*	1.3780***	1.8270***	0.3000*
	(0.15)	(0.15)	(0.09)	(0.07)	(0.14)
New tech. investment	0.0505	0.1600***	0.0814***	0.0742**	0.0319
	(0.03)	(0.03)	(0.02)	(0.03)	(0.03)
Export	0.4970***	0.2930***	0.1490*	0.1890*	0.3850***
	(0.04)	(0.08)	(0.07)	(0.09)	(0.05)
Log likelihood	–2104.52	–2176.90	–1961.18	–2173.65	–2063.22
chi-square test	794.27***	357.81***	1092.24***	1533.74***	429.75***
Wald test of exogeneity	34.19***	10.70***	7.00***	4.46**	33.40***
No. of observations	1784	1784	1784	1784	1784

Marginal effects; Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

Table 7. Average marginal effects of export on STI and DUI approach interaction for manufacturing firms.

DUI*STI = 0|export status	Outcome = 0.4100
DUI*STI = 1|export status	Outcome = 0.0000

In Figure 1, when STI and DUI approaches do not interact (STI*DUI = 0), the marginal effect of the probability that export-led innovation influences product innovation is 51.6 percent. When the export-led innovation is ignored, the marginal effect of the association between non-interaction of STI and DUI and product innovation is 46.0 percent. Here, the result confirms hypothesis 4, which states that the impact of export-led innovation on firm performance is higher when STI and DUI do not interact than when there is interaction. A point of departure from Haus-Reve, Fitjar, and Rodríguez-Pose (2019) is that while export-led innovation is not a significant factor in their findings, it affects results in the current work. However, whether the export-led innovation is ignored or not, the finding supports Haus-Reve, Fitjar, and Rodríguez-Pose (2019), who posit that the STI and DUI approach to innovation is more relevant to innovations independently. Also, it corroborates that the STI and DUI interaction significantly affects firms’ innovation negatively (Haus-Reve, Fitjar, and Rodríguez-Pose 2019).

4.2. Service firms

When we turn our view to service firms, results for labor competence, STI, DUI, new technology investment, and business expansion strategies are similar to manufacturing firms, as presented in Table 8. However, research intensity and firm size affect innovations negatively in service firms. In Table 9, where we include a dummy firm size in the estimation, the negative effect of research intensity and firm size on innovations persists. Meanwhile, while the firm size affects product innovations negatively, it is insufficient for other innovations. The result for the marginal effects presented in Tables 8 and 9 indicates that a unit increase in research activities of service firms will lead to a lower probability of innovativeness, given that they are large firms. Regarding market concentration, while the marginal effect of the local market seems insufficient for innovation in service firms, a unit increase in the export market concentration has a probability of decreasing process innovation. For service firms, this result supports some studies, e.g. Schumpeter (1934), that competition negatively affects innovation.

Table 8. Average marginal effects and baseline estimation for service firms.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	−0.0032	−0.0263*	−0.0356***	−0.0229*	−0.0092
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Firm size(log)	−0.0310***	0.0042	0.0075	0.0088	−0.0171*
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Labor competence	0.0670	0.1970***	0.1660***	0.2110***	0.0654
	(0.04)	(0.04)	(0.04)	(0.05)	(0.04)
Tech. approach (STI)	0.1760***	0.1490***	0.2130***	0.1190***	0.1810***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.02)
Non-tech. approach (DUI)	0.1370***	0.2450***	0.2180***	0.0821**	0.2240***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.02)
Export market conc.	0.0559	−0.3180*	−0.1630	−0.1940	−0.0559
	(0.11)	(0.13)	(0.13)	(0.12)	(0.10)
Local market conc.	−2.3970	4.1220	5.4740	3.1270	1.7220
	(2.84)	(3.06)	(3.02)	(3.02)	(2.73)
Business expansion	1.8320***	0.0849	0.4460**	1.7170***	0.3210*
	(0.11)	(0.09)	(0.16)	(0.04)	(0.15)
New tech. investment	0.0664**	0.1500***	0.0753***	0.0535*	0.0795***
	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)
export	0.5860***	0.3730***	0.4050***	0.2200*	0.5350***
	(0.02)	(0.09)	(0.06)	(0.09)	(0.03)
Log likelihood	–2739.81	−2847.57	−2841.52	−2841.89	−2675.65
chi-square test	3009.71***	419.12***	496.36***	3786.04***	1116.41***
Wald test of exogeneity	74.77***	8.69***	18.22***	2.79	91.57***
No. of observations	2265	2265	2265	2265	2265

Marginal effects; Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

Table 9. Average marginal effects with a firm class dummy for service firms.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	−0.0033	−0.0270*	−0.0366***	−0.0240*	−0.0087
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Large firm	−0.1250***	−0.0011	0.0008	−0.0049	−0.0404
	(0.03)	(0.03)	(0.03)	(0.03)	(0.03)
Medium firm	−0.0424	0.0191	0.0501	0.0866	−0.0815
	(0.05)	(0.05)	(0.05)	(0.06)	(0.05)
Small firm	−0.0547*	0.0090	0.0014	−0.0027	−0.0075
	(0.02)	(0.03)	(0.03)	(0.03)	(0.02)
Labor competence	0.0692	0.1970***	0.1680***	0.2130***	0.0638
	(0.04)	(0.04)	(0.04)	(0.05)	(0.04)
Tech. approach (STI)	0.1750***	0.1500***	0.2130***	0.1200***	0.1820***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.02)
Non-tech. approach (DUI)	0.1340***	0.2450***	0.2170***	0.0807**	0.2250***
	(0.03)	(0.03)	(0.03)	(0.03)	(0.02)
Export market conc.	0.0561	−0.3140*	−0.1600	−0.1910	−0.0557
	(0.11)	(0.13)	(0.13)	(0.12)	(0.10)
Local market conc.	−2.5230	4.0820	5.3610	2.9400	1.7930
	(2.83)	(3.06)	(3.03)	(3.02)	(2.73)
Business expansion	1.8390***	0.0864	0.4460**	1.7190***	0.3190*
	(0.11)	(0.09)	(0.16)	(0.04)	(0.14)
New tech. investment	0.0663**	0.1510***	0.0759***	0.0539*	0.0795***
	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)
Export	0.5850***	0.3670***	0.4010***	0.2100*	0.5340***
	(0.03)	(0.09)	(0.06)	(0.09)	(0.03)
Log likelihood	–2735.25	−2847.53	−2841.44	−2840.81	−2675.360
chi-square test	3223.25***	413.09***	486.96***	3230.40***	1125.41***
Wald test of exogeneity	74.77***	8.69***	18.22***	2.79	91.57***
No. of observations	2265	2265	2265	2265	2265

Marginal effects; Standard errors in parentheses. *p < 0.05, ** p < 0.01, *** p < 0.001

Tables 8 and 9 indicate that hypothesis 1 is supported similarly in service and manufacturing firms. The result shows that exporting service firms are more innovative than non-exporters, confirming a positive export-led innovation in the previous studies (Antonelli 2019; Antonelli and Feder 2021; Cieślik et al. 2021; De Fuentes, Niosi, and Peerally 2021). Contrary to the evidence in manufacturing firms, the result for hypothesis 2 is not confirmed for service firms, as predicted. In the upper left of Figure 2, the effect of export-led innovation on products-oriented innovations is higher with DUI marginal effect than the STI, except for process innovation (blue line, STI = 1, red line, DUI = 1). However, the result for the marginal effect reveals that previous engagement in STI has a probability of increasing the effective use of DUI. The evidence in the upper right of Figure 2 shows that the effect of export-led innovation on process innovation is higher with an STI marginal effect of 41 percent than a DUI marginal effect of 34 percent in service firms, and previous engagement in DUI does not have the probability of increasing the effective use of STI approach.

Figure 2. Marginal and interaction effects of the STI and DUI approaches for service firms.

The panel's middle left in Figure 2 presents the result for hypothesis 3 regarding service firms. As indicated, the result did not confirm the hypothesis that the positive association between export-led and service-oriented innovation is higher with firms’ DUI approach than with the STI approach. While this view conforms with manufacturing firms, the evidence provided for service firms shows that the relationship will best optimize with an STI marginal effect of 45.4 percent. The evidence of the marginal effect reveals that the previous use of the DUI approach does not have a probability of increasing the effective use of the STI approach regarding service innovation. Thus, the positive association between export-led innovation and service innovation has a likelihood of optimizing using the STI approach in service firms.

Regarding hypothesis 4, the estimation procedure in manufacturing firms was repeated for service firms. In Table 10, findings suggest that the interaction of STI and DUI decreases firms’ innovative performance and the likelihood of the optimizing effect of STI and DUI on export-led innovation. Similar to the earlier illustration for manufacturing firms, at the lower right of the panel in Figure 2, we present the result for product innovation. The result indicates that when STI and DUI interact (STI*DUI = 1), the probability that export-led innovation influences product innovation has a marginal effect of 0.00 percent (see Table 11). When STI and DUI approaches do not interact (STI*DUI = 0), the probability of export-led innovation influencing product innovation optimizes with a marginal effect of 58.8 percent. When the export-led innovation is ignored, the marginal effect of the association between the non-interaction effect of STI and DUI and product innovation is 32.4 percent. Thus, in service firms, the result also confirms hypothesis 4 that the impact of export-led innovation on firm performance is higher when STI and DUI do not interact than when they do. It further validates Haus-Reve, Fitjar, and Rodríguez-Pose (2019) that the STI – DUI interaction negatively affects firms’ innovations.

Table 10. Average marginal effects of STI and DUI interaction for service firms.

Variables	(1) Product innovation	(2) Process innovation	(3) Service innovation	(4) New-to-firm prod. Innov.	(5) New-to-market prod. innov.
Research intensity(log)	−0.0045	−0.0258*	−0.0363***	−0.0238*	−0.0093
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Firm size(log)	−0.0305***	0.0036	0.0080	0.0097	−0.0171*
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Labor competence	0.0675	0.1960***	0.1670***	0.2110***	0.0655
	(0.04)	(0.04)	(0.04)	(0.05)	(0.04)
STI-DUI interaction	−0.9660***	−1.3030***	−1.3770***	−0.6180***	−1.3150***
	(0.14)	(0.16)	(0.15)	(0.15)	(0.13)
Export market conc.	0.0675	−0.3230*	−0.1560	−0.1860	−0.0550
	(0.10)	(0.13)	(0.12)	(0.12)	(0.10)
Local market conc.	−2.5820	4.2070	5.3840	2.9490	1.7120
	(2.84)	(3.06)	(3.02)	(3.02)	(2.73)
Business expansion	2.0750***	0.0871	0.4400**	1.7050***	0.3210*
	(0.136)	(0.0898)	(0.162)	(0.0438)	(0.145)
New tech. investment	0.0654**	0.1510***	0.0748***	0.0528*	0.0795***
	(0.02)	(0.02)	(0.02)	(0.02)	(0.02)
Export	0.5920***	0.3590***	0.4140***	0.2390**	0.5360***
	(0.02)	(0.09)	(0.06)	(0.09)	(0.03)
Log likelihood	–2743.058	–2848.47	–2842.56	–2843.75	–2675.67
chi-square test	2318.53***	405.69***	510.89***	3030.15***	1118.47***
Wald test of exogeneity	74.77***	8.69***	18.22***	2.79	91.57***
No. of observations	2265	2265	2265	2265	2265

Marginal effects; Standard errors in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001

Table 11. Average marginal effects of export on STI and DUI approach interaction for service firms.

DUI*STI = 0|export status	Outcome = 0.58
DUI*STI = 1|export status	Outcome = 0.00

5. Conclusions

This study investigates how firms’ STI and DUI optimize export-led innovation and firms’ innovative efforts. It posits that as export-led innovation improves firms’ innovations, its marginal effect improves with STI and DUI differently in manufacturing to service firms. The evidence provided in the study shows that as the marginal effect of export-led innovation on product and process innovations optimizes with STI in manufacturing firms, the STI-optimizing effect is found with service and process innovations in service firms. Also, as the marginal effect of DUI optimizes the effect of export-led innovations on service and New-to-Market innovations in manufacturing firms, the enhancing effect of DUI in service firms is found with the product, New-to Market, and New-to Firm product innovations. Finally, the study confirms that while the STI – DUI interaction is counterproductive to firms’ innovative success, the individual influence of the STI and DUI tends to maximize the effect of export-led innovation on the specific type of firms’ innovations.

The findings present significant perceptions. First, we find empirical support from previous studies noting that exporting firms are more innovative than non-exporting firms (Antonelli and Feder 2021; Cieślik et al. 2021; Clausen and Pohjola 2013; Damijan, Kostevc, and Rojec 2017; Siedschlag and Zhang 2015). Second, the study aligns with Clausen et al. (2012) that the STI has a higher marginal effect on a firm's product innovation but with a contradiction in service firms. Third, the finding that the STI – DUI interaction negatively affects firms’ innovations confirms a study by Haus-Reve, Fitjar, and Rodríguez-Pose (2019). In other words, the STI – DUI complementarity can be counterproductive, and firms may need to focus on one of the two approaches to optimize firms and export-led innovations based on a specific innovation type. This finding contrasted with Jensen et al. (2007), who observed that the STI – DUI interaction optimizes firms’ innovations.

Generally, insights from this study further show that as STI and DUI can optimize firms’ and export-led innovations, the optimizing effects vary with firm and innovation types. Evidence from this study suggests that while STI may be more significant for most innovations in manufacturing firms, DUI seems to be critically more essential for the success of most service firms’ innovations. Also, as firms’ innovative success is synonymous with large firms in manufacturing firms, the study demonstrates that it is peculiar to small service firms in Norway. It additionally reveals that while R&D intensity may not always make manufacturing firms innovative, its lower degree can be essential for innovation in service firms. Finally, the positive effect of market concentration confirms that firms’ innovations improve through learning (Crespi, Criscuolo, and Haskell 2008) by competing continuously to take advantage (Aghion et al. 2005; 2018; Schumpeter 1942). However, the study findings reveal that as competition continuously improves manufacturing firms’ innovations, it reduces innovation efforts in service firms.

5.1. Implications for research

While this research demonstrates interesting perceptions, we acknowledge some limitations. First, despite the distinct perception of the data with a better measurement approach, it is missing some lag effects of variables like innovation and R&D, which might affect the results. The exclusion of the lag variables may limit the generalizability of the findings; hence future studies that include the variables are encouraged to cover the dynamic aspect we have not studied and may also cover countries other than the one we have explored.

Second, potential studies may include longitudinal analysis of factors affecting the export-led innovation, STI, and DUI relationship as the new realities evolve. Hence, replication analyzes might reveal critical temporal results with an improved understanding of the association. In this regard, the inconsistent effects of R&D intensity and firm size might yield evidence other than what is currently posited in our study. A significant reason is that it takes time for investment in R&D activities to become significant for firms’ innovation, and our study is cross-sectional.

Third, the evidence in this study is limited to service and manufacturing firms, implying that the results may not be significantly similar to firms in other industries. Specifically, the findings may not be generalizable until similar future research considering the export-led innovation, STI, and DUI relationship confirms our view. Moreso, a longitudinal investigation applying a propensity score matching method estimating the heterogeneous treatment effects of self-selection may reveal an improved analysis that allows for a more flexible policymaking perspective.

5.2. Practical and economic implications

The findings have practical and economic implications. First, our results show that while STI is more significant for manufacturing firms’ product innovation, DUI is critically more essential for the success of product innovation in service firms. This finding suggests that as Norwegian firms choose to innovate to remain competitive, firms must carefully employ the use of STI and DUI processes to optimize performance that sustains profitability. For example, an incumbent service firm may leverage the combination of first mover advantage, particularly from a technical change, allocative efficiency of the innovation modes, and market experience perspective to beat the competition. If not, it may be difficult for a product from a new entry service firm engaging in DUI to compete, considering the knowledge of the optimizing tendencies of the firm's innovation modes is misjudged.

Second, the finding that the effect of export-led innovation on service innovation optimizes with DUI in manufacturing firms and STI in service firms has some implications. While exporting manufacturing firms may need to leverage the DUI associated with human resource investments and proper organizational methods to optimize the provisions of products’ services, service firms must invest reasonably in R&D activities to improve their services. One reason is that some services are intangible, and humans have the explicit knowledge to deal with exceptional cases that machines (technological innovations) cannot handle. While this case may apply more to manufacturing export firms, service export firms will need advanced technologies to market commoditized services. Hence, the firms’ inability to appropriately determine when the STI and DUI maximize export-led innovation will increase their service costs and may affect profitability negatively.

Finally, the finding suggests that the STI and DUI complementarity tends to be counterproductive to firms’ innovative efforts. Since the STI and DUI individually optimize export-led and specific firms’ innovations, the narrative suggests that Norwegian firms should use the STI or DUI based on the targetted innovation. Economically, their proper applications can facilitate efficient production methods or services that affect higher and sustainable profit margins. On the contrary, their misuse can attract high production and service costs, hamper competition, and reduce firms’ profitability and growth.

Acknowledgments

This research is a part of the large regional dynamics and innovation capabilities in non-metropolitan contexts’ (REDINN) project.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The use of the probit 2SLS estimator is not efficient since the probit model contains discrete outcomes and endogenous covariates (see Bollen, Guilkey, and Mroz 1995, 118). Hence, we adopt the eprobit model that implements Wooldridge's IV maximum likelihood estimators, with the advantage that it needs not to be correctly specified to obtain consistent and efficient parameter estimates (Bollen, Guilkey, and Mroz 1995, 118).

2 The interaction is inestimable with the main effects due to the nature of the data. Also, the use of a 2SLS estimator is not efficient for this estimation because the probit model contains discrete outcomes and endogenous covariates (Bollen, Guilkey, and Mroz 1995, 118).

Appendix A1.

Variable measures.

Table

Variable	Measurement description
Product innovation	The dummy variable takes the value ‘1’ for firms that introduced new or significantly improved goods and ‘0’ otherwise, taken from CIS data.
Process innovation	The dummy variable takes the value ‘1’ for firms that introduced new or significantly improved processes or methods for innovation in business processes and ‘0’ otherwise, taken from CIS data.
Service innovation	The dummy variable takes the value ‘1’ for firms that introduced new or significantly improved services and ‘0’ otherwise, taken from CIS data.
New-to-market product innovation	The dummy variable takes the value ‘1’ for firms whose product or service innovations are new to the company's market ‘0’ otherwise, taken from CIS data.
New-to-firm product innovation	The dummy variable takes the value ‘1’ for firms whose product or service innovations are new to the firm ‘0’ otherwise, taken from CIS data.
Export	The dummy variable takes the value ‘1’ for firms that sold to the Norwegian market and other markets beyond Norway and ‘0’ otherwise, taken from CIS data.
Research intensity	Share of R&D expenditure (external and in-house) in a total turnover in the surveyed year. It is computed with the CIS and accounting data.
Local market concentraion	Percentage of firm turnover locally based on Hirschman – Herfindahl concentration index, taken from CIS data.
Export market concentration	Percentage of firm turnover abroad based on Hirschman – Herfindahl concentration index, taken from CIS data.
Financial support	The dummy variable takes the value ‘1’ for firms that receive government funds for export activities, specifically, not indicated for innovation, and ‘0’ otherwise, taken from CIS data.
Labor competence	The dummy variable takes the value ‘1’ for firms that deem it important or fairly important for education and skill development to organize work ‘0’ otherwise, taken from CIS data.
Outsourcing	The dummy variable takes the value ‘1’ for firms that buy R&D services from others during the period and ‘0’ otherwise, taken from CIS data.
New technology investment	The dummy variable takes the value ‘1’ for firms that procure new machinery, equipment, or software based on technology during the period ‘0’ otherwise, taken from CIS data.
Technological approach (STI)	The dummy variable takes the value ‘1’ for firms that used universities, research institutes, and consultancy firms as partners for innovation and ‘0’ otherwise, taken from CIS data.
Non-technological approach (DUI)	The dummy variable takes the value ‘1’ for firms that used suppliers and customers as partners for innovation ‘0’ otherwise, taking from CIS data.
Business expansion strategies	The dummy variable takes the value ‘1’ for firms that responded with 3, 2, and 1 as strategies for financial results in the period and ‘0’ otherwise, taken from CIS data.
Firm size	The number of employees in the company at the time of the survey. Taking from accounting data