R&D and innovation collaboration between universities and business: A PLS-SEM model for the Spanish Province of Huelva

García-Machado, Juan J.; Sroka, Włodzimierz; Nowak, Martyna Article R&D and innovation collaboration between universities and business: A PLS-SEM model for the Spanish Province of Huelva Administrative Sciences Provided in Cooperation with: MDPI – Multidisciplinary Digital Publishing Institute, Basel Suggested Citation: García-Machado, Juan J.; Sroka, Włodzimierz; Nowak, Martyna (2021) : R&D and innovation collaboration between universities and business: A PLS-SEM model for the Spanish Province of Huelva, Administrative Sciences, ISSN 2076-3387, MDPI, Basel, Vol. 11, Iss. 3, pp. 1-27, https://doi.org/10.3390/admsci11030083 This Version is available at: https://hdl.handle.net/10419/240174 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ administrative sciences Article R&D and Innovation Collaboration between Universities and Business—A PLS-SEM Model for the Spanish Province of Huelva Juan J. García-Machado 1, Włodzimierz Sroka 2,3,* and Martyna Nowak 4   Citation: García-Machado, Juan J., Włodzimierz Sroka, and Martyna Nowak. 2021. R&D and Innovation Collaboration between Universities and Business—A PLS-SEM Model for the Spanish Province of Huelva. Administrative Sciences 11: 83. https://doi.org/10.3390/ admsci11030083 Received: 17 July 2021 Accepted: 9 August 2021 Published: 17 August 2021 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). 1Department of Financial Economics, Accounting and Operations Management, University of Huelva, 21071 Huelva, Spain; [email protected] 2Department of Management, WSB University, 41-300 D ˛abrowa Górnicza, Poland 3College of Business and Economics, University of Johannesburg, Johannesburg 1809, South Africa 4Department of Applied Sciences, The Hong Kong Polytechnic University, Hong Kong, China; [email protected] *Correspondence: wsr[email protected] Abstract: In the last decade we have witnessed a growing amount of interest for developing better ‘exchange’ between universities, research centres and technology parks and companies, governments and other institutions. The biggest aim of those projects is, on the one hand, to make sure that valuable research does not stay hidden in the ivory tower of academia, and, on the other, that there are clear indications for what kinds of solutions are needed in the market. Due to the lack of empirical research in the topic, the focus of this paper is to establish and explain which factors determine the demand for technological services and how they can contribute to the promotion of greater university–business collaboration in R&D and innovation. To achieve that goal, we applied the PLSSEM (Partial Least Squares Structural Equation Modelling) method in order to create a theoretical model, which was then verified through the application of the CTA (Confirmatory Tetrad Analysis) with the purpose of evaluating whether the specification of the chosen measurement model based on the theoretical rationale was supported by data. The test run was performed on 96 companies from the Spanish region of Huelva. It showed that only four of the considered factors, namely influence of the environment, market conditions, technology adoption decision and economic characteristics of the company, constituted 65.76% of the variance of the endogenous latent Demand for Technological Services. We believe that thanks to the proposed model and its adaptivity, it is possible to design relevant policies and undertakings aimed at promoting the research-business collaboration at the regional, national and international levels. Keywords: technology demand; technology adoption; R&D collaboration; university; business 1. Introduction The last decade has witnessed the increase of activities undertook by the scientific and research institutions in the developed countries (such as R&D centres, universities and technology parks) with the objective of presenting their latest research results and practical solutions to governments, companies, and other institutions around the world, in order to find more sources of investment for the continuation of the projects and research. The issue of broadly understood collaboration between business and science is one of the key issues constituting the foundations for the modern knowledge-based economy (Doma´nska 2018). One of the examples is the MINATEC Campus in France, where an approach to microand nanotechnology research based on the triple helix of higher education–research– industry has been adopted. The MINATEC innovation campus is home to 3000 researchers, 1200 students, and 600 business and technology transfer experts on a 20-hectare state-ofthe-art campus with 13,000 m 2 of clean room space where it generates up to 350 patents and 1600 scientific articles every year (Allan et al. 2019). The main reason for this kind of Adm. Sci. 2021,11, 83. https://doi.org/10.3390/admsci11030083 https://www.mdpi.com/journal/admsci Adm. Sci. 2021,11, 83 2 of 27 actions is not only the willingness of the scholars and researchers to path the way for their work beyond the so-called Ivory Tower of Academia (Coulter 1999), but also the fact that public spending and investment in R&D is not sufficient and is not allowing the desired progress. The case of Spain seems to reflect this latest change quite clearly. According to the latest Cotec report (Cotec Foundation for Innovation 2020), investment in R&D has reported its fifth consecutive year of growth, surpassing, for the very first time, the amount of EUR 15 million. It is important to highlight that the main contributor to this growth was not public administrations, but rather companies that drove the advance in research spending by increasing their investment by 8.2%. Even though Spain has recovered to the levels of investment in R&D achieved prior to the crisis of 2008, the percentage of GDP dedicated to R&D is still below the average for European Union countries, which are currently at the level of 2.12%, compared with 1.8% in 2006 (Europe Press Agency 2018). For Spain, the highest level was reported in 2010, at 1.36%, and in 2018 it reached only 1.24% (and 1.21% in 2017). From 2010 to 2018, investment in R&D in Spain decreased by 8.82%, placing the country in the third quartile of the EU-28, very far from the target of 2% set for 2020 agreed by the Government of Spain with the EU (Maqueda 2019), and the 3% target set by the EU in its Europe 2020 strategy (European Commission 2010). With the ongoing consequences caused by the global pandemic and the efforts that must be undertaken in order to facilitate the economic revival of the economy, it seems quite possible that spending on R&D will not increase significantly. Another observation made for the case of Spain with respect to research and knowledge transfer from Spanish universities points out a slight increase in expenditure R&D hires, reaching an average price of EUR 77,000 per contract (compared to EUR 71,000 per contract in 2016). However, at the same time, a downward trend in average prices in R&D contracts can be observed, particularly for technical support and service provision—from EUR 15,000 per contract in 2010 to EUR 3800 per contract in 2017. Similarly, the average prices of R&D on request (hiring for R&D projects, characterized by the generation of new knowledge), has fallen from EUR 44,000 per contract in 2010 to EUR 32,000 per contract. This is a reflection of the reduction of the scientific–technical scope of R&D contracts and services, and its replacement, in many cases, by agreements with consulting and advisory purposes. Therefore, the expenditure on R&D with a certain level and scope seems to take place only within the framework of “subsidized grants”. Quite striking is also the fact that the exchange is very ‘local’—67% of contracts are carried out with entities whose headquarters are in the same Autonomous Community, 27% with entities whose headquarters are in other parts of Spain, 6% with companies located in Europe (4%) and only 2% with companies on other continents (Conde-Pumpido Touron and Cerezo García 2019). Given the lack of empirical studies on the subject, the objective of this work is to fill in the knowledge gap by discovering and explaining which factors determine companies’ demand for technological services, and in what way, and how this could contribute to the promotion of greater university–business collaborations in R&D. Based on the authors’ previous work, in which the complex initial theoretical model was developed using the PLS-SEM methodology, this paper examines and tests its practical implementation and predictive relevance using a sample of companies from the Huelva region of Andalusia, Spain. For the design, evaluation and predictive relevance of this model, the Structural Equation Models based on Variance (PLS-SEM) methodology (Hair et al. 2019a) and the statistical software SmartPLS, version 3.2.9 was used (Ringle et al. 2015). Out of 13 initially considered factors (constructs), just four of them, namely the influence of the environment, market conditions, technology adoption decisions and, to a lesser extent, the economic characteristics of the company, explain 65.76% of the variance of the central construct of this research, that is, demand for technological services. The authors believe that the implementation of this type of study can be an important step and act as a basis for designing relevant policies and actions aimed at promoting research–business cooperation at regional, national and international levels. Adm. Sci. 2021,11, 83 3 of 27 2. Literature Review and Theoretical Framework Technological development is one of the factors that has a significant effect on the economic development of enterprises and countries. One of the most important reasons for organisations commencing R&D cooperation is to obtain an innovative product that allows them to obtain a competitive advantage (Cygler and Wyka 2019), and therefore it can be claimed that innovation capability plays an important role in international competitiveness (Klein et al. 2021). Technological collaboration makes it possible for companies to overcome the complexity of international markets, thus boosting the internationalisation of the firm (Serrano et al. 2021). Gónzalez Hermoso de Mendoza (2011), focusing on the importance of innovation, highlighted how companies in Spain can benefit from contracting part of their R&D from universities and public research centres, but despite the boost that the Spanish Public Administration has given to promote relations between the scientific and business sectors, only 2% of Spanish companies collaborate regularly with universities and public research centres. The reasons for such a low exchange ‘rate’ include the great differences in mentality between researchers from public centres and businesspeople, which often makes these relationships difficult. Since the prestige of a university or a research centre, as well as the professional careers of the researchers themselves, is dependent on the number and quality of publications, the objective is to publish and popularize the research as fast as possible. However, it is often crucial for a company’s competitiveness to keep its R&D activities confidential, so it is imperative that the researchers do not publish their results without having first protected the research through a patent or other form of industrial protection. On top of that, the civil servant status of most Spanish public researchers is, in many cases, an added difficulty for collaborations with the private sector. The internal organization of research centres, rather than facilitating engagement, makes it difficult for researchers to engage in collaborative projects with companies. However, overcoming these obstacles and developing the technological cooperation agreements with universities (or other type of research centres) can directly translate into important strategic advantages being obtained, especially for small companies (Chastenet et al. 1990). Although the pace of change is slower than desired, important advances have been made, and collaboration between universities and public R&D centres and companies has been increasing. Undoubtedly, the restrictions on public funding for universities and public centres contribute to this, as they are forced to seek funding in other ways— one of those being collaboration with companies through research contracts. This gives the companies a competitive edge and helps to improve innovative processes, while, at the same time, providing the universities with the resources necessary for the constant modernization of laboratories and for the gratification and incentivisation of its agents (Nieto Antolín and Rodríguez Duarte 1998). However, according to a report from the Spanish Chamber of Commerce (2020), the financing of university R&D by companies has been falling since 2008 (with a turning point in 2017). The uptake of resources as a result of university–company collaborations via licenses decreased between 2016 and 2017 (the most recent data available when the study was prepared), and the number of spin-offs (companies born in the university) created was at its lowest between 2007 and 2017. This occurred despite the abundant “production” of 453,489 articles in the last four years, placing Spain as one of the main research states, with 3.3% of the world’s total academic publications. The report highlights that the number of applications for patents owned by universities was at the level of 327 in 2018, 25% less than a year earlier, although these data may be influenced by legal changes. It is, according to the Spanish Chamber of Commerce, “the reality of considerable excellence in publications and, on the contrary, a scarce transfer to the productive sector” (Spanish Chamber of Commerce 2020). López Hurtado (2014) examined an interpretive model of relationships in which he highlighted the need for the three main axes of the economy, State–Company–University, to be interrelated. He reviewed the theoretical approaches regarding said interactions and the impact they have on society, highlighting the concepts of the scientific–technological triangle and the triple helix model. The former, known also as the Sábato Triangle Approach, Adm. Sci. 2021,11, 83 4 of 27 reflects the relationship between the government, the scientific–technological infrastructure, and the productive structure (Sábato 1997;Vega-Jurado et al. 2007;Marone and González del Solar 2007). The later distinguishes between the academy, industry, and the government (Etzkowitz and Leydesdorff 2000;Etzkowitz 2003;Gonzáles 2009;Leydesdorff 2011). The intent of both theories is to describe innovation systems, and although they differ in their approach with respect to knowledge generation and the interpretation of innovation (in linear and non-linear terms), they agree that innovation does not depend solely on the capabilities that the public sector, industry and universities possess, but rather that it results from mutual relationships between agents and interactions that are established within the framework of National Innovation Systems (López Hurtado 2014). Due to the aforementioned scarcity of empirical research on the topic of interaction among companies and scientific and research organizations, we embarked on the task of investigating the factors on which the demand for technological and scientific solutions by companies depends, using a sample of companies form the Spanish province of Huelva. To design an explanatory model of the demand for scientific and technological services by companies in the province of Huelva, we used the works of García-Machado et al. (2012), who empirically examined an extension of the Technology Acceptance Model in the context of online financial commerce, Roldán and Sánchez-Franco (2012), who applied it to the context of social networks, and García-Machado (2017), who proposed a PLS-SEM model for the study of secure online trading services. The results presented show that there is a positive, direct, and statistically significant relationship between the expectations of personal results, the perceived relative advantages, a shared vision and mutual trust based on the economy gains, and the quality of knowledge provided. Another work we used for the design of the initial theoretical model was the study of Magotra et al. (2018), which analysed the relationship between the perception of customer value and technology adoption behaviour with reference to online banking customers. Furthermore, this relationship was examined through the development of an Integrated Technology Adoption Model through the application of the Structural Equation Models (SEM) approach. The extrapolation of the factors gathered in these studies, regarding the demand for technological services, led to the creation of a first study model, made up of 13 latent variables (or constructs): • Facilitating Conditions (FC): Venkatesh and Zhang (2010) placed the Facilitating Conditions as one of the factors that directly affect the final construct of the demand for technological services. Yu (2012), on the other hand, defined the Facilitating Conditions as the degree to which an individual believes that there is an organizational and technical infrastructure supporting the use of technology. •Behaviour Intention towards Technology Adoption Decision (BITA): Several authors have considered this “Intent” from various perspectives. For example, Lee (2009) investigated a model that measures the factors affecting the adoption of online banking from a risk/benefit perspective, integrating two techniques: TAM (Technology Acceptance Model) and TPB (Theory of Planned Behaviour). In this model, the “Intention” is considered to be “Intention of Use”, placing the variable as a final endogenous construct. Venkatesh and Davis (2000) extended the TAM model to the TAM2, also applying the model in other fields of study. Regarding the variable addressed in this work, they placed the “Behavioural Intent” as an intermediate variable that collects relationships of various constructs with the final construct. These authors defined it together with the “Facilitating Conditions” as the direct determinant of adoption behaviour. Legris et al. (2003) defined BITA as an intermediate variable that gathers information from the constructs: beliefs and evaluations, attitude towards behaviour, normative belief and motivation to comply, and subjective norms. Alsajjan and Dennis (2010) explained that attitude and behaviour are so closely related that they could be considered in certain studies to be the same variable. Attitude should predict actual behaviour, as should intentions, but attitude avoids the bias that often marks measurements of intentions. Sharma and Govindaluri (2014), in their structural equations Adm. Sci. 2021,11, 83 5 of 27 model, placed this variable as one more construct that may or may not influence the relationship with another construct, the so-called “Technology Adoption Decision”, making it a step variable towards final adoption. These studies, enriched by the work of Rawashdeh (2015), were used as a basis for the preparation of the questionnaire. • Attitude towards Technology Performance (ATP): Lai and Li (2005) focused on the attitude of the different agents towards the adoption of Internet banking. Rogger (2003) specified this as the “disposition of the individual to experience an innovation” and indicated that it could be considered the disposition of an individual towards experiencing the acquisition of new technology. •Perceived Utility (PU); and • Perceived Ease of Use (PEU): Legris et al. (2003) used a model in which both “Perceived Ease of Use” and “Perceived Utility” appear—two variables that are quite interesting and important in any model of adoption of technology. Alhassany and Faisal (2018) used both variables in their model, framing it within what they referred to as the “technology dimension”. They defined “Perceived Utility” as the beliefs of users that the adoption of technology will improve their productivity and performance. The “Perceived Ease of Use” is based on the entrepreneur’s perspectives and the evaluations of facilities/difficulties in the execution of the product. • Technological Attributes (TAT): According to Magotra et al. (2018), the construct “Technological Attributes” can be defined by the two previous constructs, PU and PEU. “Technological Attributes”, in turn, influences the “Technology Adoption Decision” and the “Demand for Technology Services”, because if a technology has perfect attributes for reinforcing or improving a certain area, and it is also easy to use, a company will consider adopting it. This attribute influences the decision-making process of the responsible person. In their study, Sharma and Govindaluri (2014) also confirmed that PU and PEU define TAT. • Business Predisposition Towards the Adoption of Technology (BPTAT): Yu (2012) exposed the idea of the adoption of online banking through the Unified Theory of Acceptance and Use of Technology (UTAUT), showing that, among other factors, it is influenced by the “Perceived Financial Cost” and the “Performance Expectation”. These, although not directly, would contribute to what would come to be “Business Predisposition Towards the Adoption of Technology”. In principle, it is assumed that the higher the financial cost, the less business predisposition, or the higher the expectation of performance, the greater the predisposition. • Economic Characteristics of the Company (ECC): This is mentioned, inter alia, in the study by Magotra et al. (2018). It suggests that the economic attributes of the company could be an essential factor to consider in our research. Labra Lillo (2015) confirmed this by stating that one of the most important factors for investment in R&D is the size and the economic nature of the company, which are always related. • Technology Adoption Decision (TAD): Magotra et al. (2018) designed a model where FC, TAT and BPTAT were related to this construct. However, it also depends on two more relationships: those of ATP and ICAT. Therefore, in some way, this endogenous latent variable could be understood as being “intermediate” or “regulatory” when it comes to relating all the model variables with the final construct. Following the explanation of Porras Bueno (2016), the adoption decision is the core of several variables, and it is within a cause–effect system that ranges from the antecedents of the adoption decision to the impact of the business owners. Other authors consulted were Verhoef et al. (2009) and their construct “Self-Service Technology”. • Demand for Technological Services (DTS): This is the final dependent variable at which all of the relationships of the model will arrive. At first, no references were found that included this final construct, neither as such nor from another perspective that could be subject to adaptation, as in the case of the previous constructs. However, every technology acceptance model has a final dependent variable. For example, Sharma and Govindaluri (2014), with, among others, questions about the customers’ Adm. Sci. 2021,11, 83 6 of 27 intentions as a measure to know whether they would be willing to demand a variety of services, better defined the DTS construct. Other items used were collected from Verhoef et al. (2009). • Marketing Actions (MKTA): Figueroa-García et al. (2018) pointed out that government organizations, through their marketing actions, are main actors in the education and dissemination of the information to promote a sustainable consumer behaviour. Kollmuss and Agyeman (2002) noted that institutional factors, that is, how the actions of institutions affect caring for the environment, are located among the external factors. Transferred to our study, marketing actions that can be carried out by different scientific and research organizations should have an impact on the Demand for Technological Services by companies. Taken to the field of DTS, this would mean: “How do the actions of institutions outside the organization affect this demand?” • Influence of the Environment (IE): Figueroa-García et al. (2018) stated that there are external aspects to the person (such as education and sociodemographic variables, among others) that have an influence on environmental sustainability. Contextualizing it for this research, this could be defined by aspects such as education, socio-economic, demographic, geographical and even political variables among many others, which also might influence the demand for technological services. • Market Conditions (MKC): as stated by Francis (2010), the market is volatile and changes quickly. As such, it will affect the final decision regarding new products, new marketing actions and new technological resources that lead the company to decide to adopt technical services or to lag behind its competitors. 3. Methodology 3.1. Sample Characteristics This study is based on a sample of 96 companies from the Spanish province of Huelva. Initially, a database was prepared from a list of companies provided by the Office for the Transfer of Research Results (OTRI) of the University of Huelva and the SABI database, completed with a direct search by municipalities through Google Maps. Out of a total of 467 companies, those that were not operational were eliminated (145). The remaining ones that provided contact information (email, telephone, fax or postal address) were invited to participate in the investigation. A total of 96 valid questionnaires were received, which represents a response rate close to 30%. The most relevant characteristics of the companies that make up the sample, according to their location, type of company, number of employees, seniority, turnover, and activity sector are shown in Figures 1–5. In general, most companies are based in the Huelva area (47%) and are private limited companies (59%). Most of the companies have been run for 20 years or more (46%) and were rather small in size, with a turnover of less than EUR 500,000 per year (41%). As for the sector of activity, agriculture and food industry companies constituted 18% of the sample, followed by wholesale trade (6%), specialized construction (6%), and building construction (5%). 3.2. Data Collection The data were collected from May to November 2019. For a company to become part of the study, two requirements had to be met: first, they had to be based in the province of Huelva, and second, the responsible person (entrepreneur, manager, technical or administrative director) had to fill in an online questionnaire regarding the demand for technological services on behalf of the company. This stage of the investigation was the most challenging due to the reluctance of companies to provide identifying information, opinions, or lack of time, so we consider a response rate of 30% to be a great achievement. Furthermore, as we will show later, a sample of 96 companies represents a sufficient size to validate the results. Adm. Sci. 2021,11, 83 7 of 27 Adm. Sci. 2021, 11, x FOR PEER REVIEW 7 of 27 Figure 1. Geographic distribution of the companies. Figure 2. Distribution of companies by number of employees. Figure 3. Distribution of companies by age of the company. 46% 30% 15% 9% 0 - 10 11 - 50 51 - 250 250+ 5% 4% 10% 10% 8% 16% 47% 0-1 yo 2-3 yo 4-7 yo 8-11 yo 12-15 yo 16-19 yo 20+ yo Figure 1. Geographic distribution of the companies. Adm. Sci. 2021, 11, x FOR PEER REVIEW 7 of 27 Figure 1. Geographic distribution of the companies. Figure 2. Distribution of companies by number of employees. Figure 3. Distribution of companies by age of the company. 46% 30% 15% 9% 0 - 10 11 - 50 51 - 250 250+ 5% 4% 10% 10% 8% 16% 47% 0-1 yo 2-3 yo 4-7 yo 8-11 yo 12-15 yo 16-19 yo 20+ yo Figure 2. Distribution of companies by number of employees. Adm. Sci. 2021, 11, x FOR PEER REVIEW 7 of 27 Figure 1. Geographic distribution of the companies. Figure 2. Distribution of companies by number of employees. Figure 3. Distribution of companies by age of the company. 46% 30% 15% 9% 0 - 10 11 - 50 51 - 250 250+ 5% 4% 10% 10% 8% 16% 47% 0-1 yo 2-3 yo 4-7 yo 8-11 yo 12-15 yo 16-19 yo 20+ yo Figure 3. Distribution of companies by age of the company. Adm. Sci. 2021,11, 83 8 of 27 Adm. Sci. 2021, 11, x FOR PEER REVIEW 8 of 27 Figure 4. Distribution of companies by revenue. Figure 5. Distribution of companies by economic sector. 3.2. Data Collection The data were collected from May to November 2019. For a company to become part of the study, two requirements had to be met: first, they had to be based in the province of Huelva, and second, the responsible person (entrepreneur, manager, technical or administrative director) had to fill in an online questionnaire regarding the demand for technological services on behalf of the company. This stage of the investigation was the most challenging due to the reluctance of companies to provide identifying information, opinions, or lack of time, so we consider a response rate of 30% to be a great achievement. Furthermore, as we will show later, a sample of 96 companies represents a sufficient size to validate the results. We constructed a very complete data set that initially included 77 indicators or manifest variables and a size of 96 observations from companies in the province of Huelva. In total, they added to 7392 datapoints. The questions included manifest variables at the microeconomic level of the company (location, type, seniority, number of employees, turnover, etc.) and items that have been taken and/or adapted from previous studies. All the items were measured with a Likert scale of 1 to 7 points, from 1 = strongly disagree to 7 = strongly agree, where 4 is interpreted as a point of indifference. The detailed questionnaire can be seen in Table 1. 0 5 10 15 20 25 30 35 40 45 < 500,000 € 500,000 € - 1,000,000 € 1,000,000 € - 2,000,000 € 2,000,000 € - 5,000,000 € 5,000,000 € - 10,000,000 € 10,000,000 € - 20,000,000 € > 20,000,000 € No. of Companies Revenue 13% 6% 25% 6% 13% 13% 6% 8% 10% Food & Beverage Graphic Arts Agriculture Fishing Construction Commerce Chemical Metalurgy Real Estate Figure 4. Distribution of companies by revenue. Adm. Sci. 2021, 11, x FOR PEER REVIEW 8 of 27 Figure 4. Distribution of companies by revenue. Figure 5. Distribution of companies by economic sector. 3.2. Data Collection The data were collected from May to November 2019. For a company to become part of the study, two requirements had to be met: first, they had to be based in the province of Huelva, and second, the responsible person (entrepreneur, manager, technical or administrative director) had to fill in an online questionnaire regarding the demand for technological services on behalf of the company. This stage of the investigation was the most challenging due to the reluctance of companies to provide identifying information, opinions, or lack of time, so we consider a response rate of 30% to be a great achievement. Furthermore, as we will show later, a sample of 96 companies represents a sufficient size to validate the results. We constructed a very complete data set that initially included 77 indicators or manifest variables and a size of 96 observations from companies in the province of Huelva. In total, they added to 7392 datapoints. The questions included manifest variables at the microeconomic level of the company (location, type, seniority, number of employees, turnover, etc.) and items that have been taken and/or adapted from previous studies. All the items were measured with a Likert scale of 1 to 7 points, from 1 = strongly disagree to 7 = strongly agree, where 4 is interpreted as a point of indifference. The detailed questionnaire can be seen in Table 1. 0 5 10 15 20 25 30 35 40 45 < 500,000 € 500,000 € - 1,000,000 € 1,000,000 € - 2,000,000 € 2,000,000 € - 5,000,000 € 5,000,000 € - 10,000,000 € 10,000,000 € - 20,000,000 € > 20,000,000 € No. of Companies Revenue 13% 6% 25% 6% 13% 13% 6% 8% 10% Food & Beverage Graphic Arts Agriculture Fishing Construction Commerce Chemical Metalurgy Real Estate Figure 5. Distribution of companies by economic sector. We constructed a very complete data set that initially included 77 indicators or manifest variables and a size of 96 observations from companies in the province of Huelva. In total, they added to 7392 datapoints. The questions included manifest variables at the microeconomic level of the company (location, type, seniority, number of employees, turnover, etc.) and items that have been taken and/or adapted from previous studies. All the items were measured with a Likert scale of 1 to 7 points, from 1 = strongly disagree to 7 = strongly agree, where 4 is interpreted as a point of indifference. The detailed questionnaire can be seen in Table 1. All the indicators and data were computed in an Excel spreadsheet and then converted into CSV format to be able to run it using SmartPLS v.3.2.8 software (Ringle et al. 2015) to apply PLS-SEM modelling. 3.3. Estimation of the Theoretical Model An initial theoretical model was developed concerning the possible determinants of the demand for technological services by companies. The latent exogenous and endogenous variables, as well as their relationships, are represented in the initial proposed model (Figure 6) and are summarized in Table 1. Adm. Sci. 2021,11, 83 15 of 27 loadings, where we can see how the already-eliminated MKTA4 and ECC4 indicators do not meet this criterion either. Of those remaining, only ECC1 does not meet it, since its load is less than 0.5 and it also does not present statistical significance. Therefore, we eliminated it from the ECC measurement model. Table 7. Significance and relevance of outer loadings. Formative Constructs Indicators Outer Loadings t-Value p-Value 95% BCa Confidence Interval Is It Significant? (p< 0.05) Marketing Actions (MKTA) MKTA1 0.764 4.153 0.000 [0.425; 0.972] Yes MKTA2 0.868 4.126 0.000 [0.679; 0.997] Yes MKTA3 0.569 2.387 0.017 [0.070; 0.899] Yes MKTA4 0.435 1.799 0.072 [−0.071; 0.822] No Economic Characteristics of the Company (ECC) ECC1 0.181 0.967 0.334 [−0.198; 0.540] No ECC2 0.417 2.080 0.038 [0.023; 0.767] Yes ECC3 0.682 4.000 0.000 [0.389; 0.908] Yes ECC4 0.304 1.079 0.281 [−0.313; 0.805] No ECC5 0.786 4.527 0.000 [0.526; 0.970] Yes Note: Bias-Corrected and Accelerated (BCa) bootstrap confidence intervals for 5000 subsamples, no sign changes, and two-tailed test. For the measurement models of the MKTA and ECC formative constructs formed by MKTA1, MKTA2, MKTA3, ECC2, ECC3 and ECC5, it was found that both present convergent validity and their retained indicators do not present collinearity problems. As such, they are relevant and statistically significant. After debugging the items, the PLS algorithm was run again, converging after nine iterations, thus finding a faster and more stable solution. 4.3. Evaluation of the Structural Model Once it has been verified that both the validity and the reliability of the measurement models meet the requirements indicated above, and once the non-significant indicators of MKTA and ECC have been refined, the next step consisted of evaluating the structural model, which represents the relationships hypothesized between the constructs (GarcíaMachado 2017). This involved examining the predictive capacity of the model, for which PLS-SEM was originally designed, as well as the relationships between the constructs. The key criteria for evaluating the structural model are the algebraic sign, the significance and relevance of the path coefficients, the level of the values of R 2 , the effect size f 2 , the predictive relevance Q 2 , and the effect size q 2 (Hair et al. 2011,2017,2019b). However, before doing so, it is advisable to examine the possible multicollinearity between the constructs of the structural model. Table 8shows the results of the VIF values of all the sets of predictors. For this analysis, most authors recommend that the FIV values should be below 5, or what is the same, a tolerance greater than 0.20, although, recently, Hair et al. (2019a) recommended a threshold of 3 for assessing the VIF. In any case, for the proposed model, there are no collinearity problems, since all the VIFs are below these thresholds. Continuing with the evaluation of the structural model, we analysed the coefficients of determination or the R 2 values of the endogenous latent variables. This value measures the amount of variance in the endogenous constructs explained by all the exogenous constructs linked to them, and it is the most frequently used measure for checking the predictive power of the model (Hair et al. 2019a). While several authors argue that a valid R 2 should be greater than 0.1 (Hair et al. 2017;Falk and Miller 1992), the interpretation of R 2 will depend on the model and field of study. In general, R 2 values can be described as substantial, moderate and weak, depending on whether their value is 0.75, 0.5 or 0.25 (García-Machado 2017). To avoid the bias produced by the increase in the number of exogenous constructs, it is usually also used in adjusted coefficient of determination (R 2adj ). Adm. Sci. 2021,11, 83 16 of 27 Table 8. Collinearity assessment: VIF values in the structural model. MKTA TAT + ATTA ECC MKC TAD DTS FC PEU IE BITA PU Marketing Actions (MKTA) 1.000 Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) 1.349 Economic Characteristics of the Company (ECC) 1.044 Market Conditions (MKC) 2.082 Technology Adoption Decision (TAD) 2.323 Demand for Technological Services (DTS) Facilitating Conditions (FC) 1.273 Perceived Ease of Use (PEU) 1.000 Influence of the Environment (IE) 1.333 1.742 Behaviour Intention Toward Technology Adoption (BITA) 1.324 Perceived Utility (PU) 1.333 As can be seen in Table 9, TAT + ATTA has a value close to 0.75, the highest value, which can be considered substantial. Then, DTS, TAD and PU follow with 0.657, 0.642 and 0.449, respectively, values that are around 0.5, which can be considered moderate, leaving ECC with 0.176, which is rather weak. The R 2adj values do not present much difference with respect to the previous ones. Table 9. Explained variance (R2). Endogenous Latent Variables R2R2adj Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) 0.744 0.738 Economic Characteristics of the Company (ECC) 0.176 0.167 Technology Adoption Decision (TAD) 0.642 0.630 Demand for Technological Services (DTS) 0.657 0.642 Perceived Utility (PU) 0.449 0.443 To assess whether the omission of an endogenous construct has a substantial impact on the model, the effect size f 2 is used (Albort-Morant et al. 2018;Hair et al. 2017;Ali et al. 2018;Müller et al. 2018). Values of f 2 above 0.02, 0.15 or 0.35 are considered to be a small, medium, or large effects, respectively (Cohen 1988). Table 10 shows the results of the effect size f 2 . For example, the largest effect size is PU on TAT + ATP (1.708), followed by PEU on PU (0.814) and BITA on TAD (0.677), which have large effects, and by MKTA on ECC (0.213), ECC on DTS (0.179) and IE over DTS (0.173), which have moderate effects. Table 10. f2Effect Sizes. MKTA TAT + ATTA ECC MKC TAD DTS FC PEU IE ICAT PU Marketing Actions (MKTA) 0.213 Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) 0.135 Economic Characteristics of the Company (ECC) 0.179 Market Conditions (MKC) 0.116 Technology Adoption Decision (TAD) 0.077 Demand for Technological Services (DTS) Facilitating Conditions (FC) 0.056 Perceived Ease of Use (PEU) 0.814 Influence of the Environment (IE) 0.086 0.173 Behaviour Intention Toward Technology Adoption (BITA) 0.677 Perceived Utility (PU) 1.708 To analyse the significance and relevance of the relationships between the constructs in the structural model, we looked at the algebraic sign, which provided us with the path coefficients. In Figure 7, it can be observed that all of the signs are positive, which indicates a direct relationship between them. The greatest relative importance among the constructs with a direct effect on the demand for technological services (DTS) is, in order of importance, the influence of the environment (IE), followed by market conditions (MKC) and the economic characteristics of the company (ECC). At first glance, it appears Adm. Sci. 2021,11, 83 17 of 27 that technology adoption decisions (TAD) would rank fourth and last. Regarding the exogenous constructs that act on DTS, through the mediating variables ECC, TAD and TAT + ATTA, the greatest importance is perceived ease of use (PEU), followed by the intention to adopt technology (BITA) and marketing actions (MKTA). Facilitating conditions would also occupy the fourth and last place in this classification. Adm. Sci. 2021, 11, x FOR PEER REVIEW 17 of 27 Table 10. f2 Effect Sizes. MKTATAT + ATTA ECC MKC TAD DTS FCPEUIE ICAT PU Marketing Actions (MKTA) 0.213 Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) 0.135 Economic Characteristics of the Company (ECC) 0.179 Market Conditions (MKC) 0.116 Technology Adoption Decision (TAD) 0.077 Demand for Technological Services (DTS) Facilitating Conditions (FC) 0.056 Perceived Ease of Use (PEU) 0.814 Influence of the Environment (IE) 0.086 0.173 Behaviour Intention Toward Technology Adoption (BITA) 0.677 Perceived Utility (PU) 1.708 To analyse the significance and relevance of the relationships between the constructs in the structural model, we looked at the algebraic sign, which provided us with the path coefficients. In Figure 7, it can be observed that all of the signs are positive, which indicates a direct relationship between them. The greatest relative importance among the constructs with a direct effect on the demand for technological services (DTS) is, in order of importance, the influence of the environment (IE), followed by market conditions (MKC) and the economic characteristics of the company (ECC). At first glance, it appears that technology adoption decisions (TAD) would rank fourth and last. Regarding the exogenous constructs that act on DTS, through the mediating variables ECC, TAD and TAT + ATTA, the greatest importance is perceived ease of use (PEU), followed by the intention to adopt technology (BITA) and marketing actions (MKTA). Facilitating conditions would also occupy the fourth and last place in this classification. To assess whether these relationships are truly significant, as well as to analyse the total, direct and indirect effects that a latent variable exerts on the key objective variable DTS, we ran a bootstrapping process with corrected and accelerated bias (BCa) for 5000 sub-samples, without sign changes, and a two-tailed test with a significance level of 0.05. Tables 11 and 12 show the results of the significance tests for direct effects and total effects (combination of direct effect plus indirect effects). Figure 7. Debugged theoretical model. Figure 7. Debugged theoretical model. To assess whether these relationships are truly significant, as well as to analyse the total, direct and indirect effects that a latent variable exerts on the key objective variable DTS, we ran a bootstrapping process with corrected and accelerated bias (BCa) for 5000 sub-samples, without sign changes, and a two-tailed test with a significance level of 0.05. Tables 11 and 12 show the results of the significance tests for direct effects and total effects (combination of direct effect plus indirect effects). Table 11. Results of the significance test for the path coefficients (direct effects). Path Coefficients t-Value p-Value 95% BCa Confidence Intervals Is It Significant? (p< 0.05) MKTA →ECC 0.419 4.959 0.000 [0.178; 0.543] Yes TAT + ATTA →TAD 0.255 2.667 0.008 [0.091; 0.461] Yes ECC →DTS 0.253 3.334 0.001 [0.113; 0.408] Yes MKC →DTS 0.288 3.22 0.001 [0.097; 0.455] Yes TAD →DTS 0.247 2.206 0.027 [0.013; 0.448] Yes FC →TAD 0.160 2.244 0.025 [0.002; 0.284] Yes PEU →PU 0.670 10.146 0.000 [0.468; 0.765] Yes IE →TAT + ATTA 0.171 2.634 0.008 [0.043; 0.296] Yes IE →DTS 0.321 3.681 0.000 [0.149; 0.487] Yes ICAT →TAD 0.566 5.992 0.000 [0.354; 0.725] Yes PU →TAT + ATTA 0.764 10.527 0.000 [0.579; 0.872] Yes Adm. Sci. 2021,11, 83 18 of 27 Table 12. Results of the significance tests for the total effects. Path Coefficients t-Value p-Value 95% BCa Confidence Intervals Is It Significant? (p< 0.05) MKTA →DTS 0.106 2.567 0.010 [0.032; 0.197] Yes TAT + ATTA →DTS 0.063 1.660 0.097 [0.010; 0.162] No FC →DTS 0.040 1.410 0.159 [0.000; 0.107] No PEU →TAT + ATTA 0.512 6.111 0.000 [0.265; 0.634] Yes PEU →TAD 0.131 2.211 0.027 [0.040; 0.272] Yes PEU →DTS 0.032 1.573 0.116 [0.006; 0.091] No IE →TAD 0.044 1.686 0.092 [0.009; 0.115] No IE →DTS 0.332 3.989 0.000 [0.167; 0.492] Yes ICAT →DTS 0.140 2.099 0.036 [0.020; 0.284] Yes PU →TAD 0.195 2.565 0.010 [0.068; 0.365] Yes PU →DTS 0.048 1.710 0.087 [0.009; 0.122] No As shown in Table 11, assuming a significance level of 5%, all the relationships of the structural model are significant (with many even at a 1% level), which gives an idea of the robustness of our model. The most significant relationships, which show the great significance of the coefficients path, are found between the market actions and the economic characteristics of the company, the perceived ease of use with the perceived utility, the influence of the environment on the demand for technological services, the intention towards the adoption of technology with the decision technology adoption and the perceived utility with attitude toward technology adoption. Regarding the total effects shown in Table 12, of the exogenous variables MKTA, FC, IE, BITA and PEU on the endogenous constructs TAT + ATTA, TAD y DTS, it can be seen that the relationships MKTA → ECC → DTS, PEU → PU → TAT + ATTA, PEU → PU → TAT + ATTA → TAD, IE → TAT + ATTA → TAD → DTS, ICAT → TAD → DTS, and PU → TAT + ATTA →TAD are all significant at the 5% level, and in some cases even at a 1%. After debugging the non-significant indicators and relationships, as well as performing a reorganization of the constructs, Figure 8shows our final proposal for a more parsimonious explanatory model on business demand for technology services in the province of Huelva. Table 13 shows the contribution of each latent variable to the final construct (DTS) through the decomposition of the explained variance, where we verified that the variables that most influence the demand for technological services by the companies are the influence of the environment and market conditions, followed by the decision to adopt technology. It is important to highlight that the economic characteristics of the company variable has the least influence with 8.7% of R2. Table 13. Decomposition of the explained variance of the endogenous latent variable DTS. Latent Variable Path Coef. Correlation R2 Economic Characteristics of the Company (ECC) 0.253 0.344 8.70% Market Conditions (MKC) 0.288 0.672 19.35% Technology Adoption Decision (TAD) 0.247 0.653 16.13% Influence of the Environment (IE) 0.321 0.672 21.57% Total R265.76% 4.4. Assessment of the Relevance and Predictive Power of the Model Until relatively recently, the Stone-Geisser (Q 2 ) test was used to evaluate predictive relevance, by applying a blindfolding procedure to predict deliberately omitted data within a sample and then comparing the resulting estimates with the real values (Hair et al. 2019a). In this case, it is possible to calculate the relative impact or the effect size q 2 when omitting an exogenous construct and see its influence on the endogenous construct (Q 2 inclusive and Q2excluded). Tables 14 and 15 show the values of Q2and q2. Adm. Sci. 2021,11, 83 19 of 27 Adm. Sci. 2021, 11, x FOR PEER REVIEW 19 of 27 After debugging the non-significant indicators and relationships, as well as performing a reorganization of the constructs, Figure 8 shows our final proposal for a more parsimonious explanatory model on business demand for technology services in the province of Huelva. Table 13 shows the contribution of each latent variable to the final construct (DTS) through the decomposition of the explained variance, where we verified that the variables that most influence the demand for technological services by the companies are the influence of the environment and market conditions, followed by the decision to adopt technology. It is important to highlight that the economic characteristics of the company variable has the least influence with 8.7% of R2. Figure 8. Proposed final model. Table 13. Decomposition of the explained variance of the endogenous latent variable DTS. Latent Variable Path Coef. Correlation R2 Economic Characteristics of the Company (ECC) 0.253 0.344 8.70% Market Conditions (MKC) 0.288 0.672 19.35% Technology Adoption Decision (TAD) 0.247 0.653 16.13% Influence of the Environment (IE) 0.321 0.672 21.57% Total R2 65.76% 4.4. Assessment of the Relevance and Predictive Power of the Model Until relatively recently, the Stone-Geisser (Q2) test was used to evaluate predictive relevance, by applying a blindfolding procedure to predict deliberately omitted data within a sample and then comparing the resulting estimates with the real values (Hair et al. 2019a). In this case, it is possible to calculate the relative impact or the effect size q2 when omitting an exogenous construct and see its influence on the endogenous construct (Q2 inclusive and Q2 excluded). Tables 14 and 15 show the values of Q2 and q2. Figure 8. Proposed final model. Table 14. Predictive relevance (Q2values). Endogenous Construct Q2 Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) 0.607 Economic Characteristics of the Company (ECC) 0.046 Technology Adoption Decision (TAD) 0.433 Demand for Technological Services (DTS) 0.490 Perceived Utility (PU) 0.391 Table 15. q2Effect Sizes. MKTA TAT + ATTA ECC MKC TAD DTS FC PEU IE ICAT PU Marketing Actions (MKTA) 0.048 Technological Attributes + Attitude towards Technology Adoption (TAT + ATTA) −0.055 Economic Characteristics of the Company (ECC) 0.080 Market Conditions (MKC) 0.057 Technology Adoption Decision (TAD) 0.025 Demand for Technological Services (DTS) Facilitating Conditions (FC) 0.023 Perceived Ease of Use (PEU) 0.642 Influence of the Environment (IE) 0.051 0.086 Behaviour Intention Toward Technology Adoption (BITA) 0.300 Perceived Utility (PU) 0.504 As can be seen, the Q 2 values of all endogenous constructs are above zero, with TAT + ATP being the one with the highest value, followed by DTS, TAD and PU. Hair et al. (2019b) proposed a new rule of thumb for measuring predictive relevance according to the value of Q 2 : low (Q 2 > 0), Medium (Q 2 > 0.25) and high (Q 2 > 0.5). Accordingly, TAT + ATP would have high relevance, DTS, TAD and PU medium, and ECC low. Regarding the effect size q 2 , values of 0.02, 0.15 and 0.35 would indicate a small, medium or large predictive relevance of an exogenous construct over an endogenous one (Hair et al. 2019a). In our case, the largest effect size is PEU over PU (0.642) and PU over TAT + ATP (0.504). BITA over TAD (0.300) has a medium size effect, and the rest have a small effect. Adm. Sci. 2021,11, 83 20 of 27 Shmueli et al. (2016) pointed out that neither the value of Q 2 nor that of q 2 provides highly interpretable results in terms of the magnitude of the error, and they do not provide us with anything related to the precision of the model for predicting the values of new cases outside the sample. They recommend a procedure called PLSpredict (Shmueli et al. 2019), which has been implemented in the SmartPLS software since version 3.2.6 (Ringle et al. 2015). For a better understanding of this and other procedures, the works of Shmueli et al. (2016,2019), Evermann and Tate (2016), Sharma et al. (2018,2019), and Danks and Ray (2018) can be consulted. Following the procedure developed by Shmueli et al. (2016), we applied the PLSpredict algorithm implemented in SmartPLS (Ringle et al. 2015). The method uses training and hold-out samples to generate and evaluate predictions from the PLS path model estimates. In the settings, we select k = 3 folders or sections (96/3 = 32), since each folder must contain a minimum of 30 data points. The algorithm then predicts each section or folder (holdout sample) with the remaining k − 1 = 2 subsets, which, in combination, become the training sample. This process is repeated 10 times by default. The number of repetitions indicates how often the PLS prediction algorithm performs k-fold cross-validation on random splits of the complete data set in k sections (folds). Traditionally, cross-validation only uses a random division in k-folds. However, a single random division can make predictions highly dependent on this random assignment of data (observations) in k-folds. Due to the random split of the data, runs of the algorithm at different points in time may vary in their predictive performance measures (for example, mean square error, mean absolute percentage error, etc.). Repeating the k-fold cross validation with different random data partitions and calculating the mean between the repeats ensures a more stable estimate of the predictive performance of the PLS path model. Based on the procedures suggested by Shmueli et al. (2016), the current implementation of the PLS prediction algorithm (PLSpredict) in the SmartPLS software enables researchers to obtain cross-validated prediction error statistics and summaries of prediction errors such as the root mean squared error (RMSE), the mean absolute error (MAE) and mean absolute percentage error (MAPE) to assess the predictive performance of a PLS path model for manifest variables (MV or indicators) and latent variables (LV or constructs). These three criteria are available for the results of the indicators, whereas it is only possible to calculate RMSE and MAE for construct results. These criteria allow the predictive performance of alternative PLS path models to be compared. Furthermore, to evaluate the results of a specific PLS path model, its predictive performance can be compared using two new indices: (1) The Q 2 value in PLSpredict compares the prediction errors of the PLS path model with the simple mean predictions. To do this, we used the mean value of the training sample to predict the results of the holdout sample. The interpretation of the results of the Q 2 value is similar to the evaluation of the Q 2 values obtained by the blindfolding procedure in PLS-SEM. If the Q 2 value is positive, the prediction error of the PLS-SEM results is less than the prediction error of simply using the mean values. In that case, the PLS-SEM models offer better predictive performance. (2) The linear regression model (LM) provides summary statistics and prediction errors that ignore the specified PLS path model. Instead, the LM approach returns all exogenous indicator variables with each endogenous indicator variable to generate predictions. Thus, a comparison with the PLS-SEM results provides information on whether using an established theoretical model improves (or at least does not worsen) the predictive performance of the available indicator data. Compared to LM results, PLS-SEM results should have a smaller prediction error (for example, in terms of RMSE or MAE) than LM. Consider, as mentioned previously, that the LM prediction error is only available for the manifest variables, and not for the latent variables. In our solution, following the suggestions of Roldán and Cepeda (2020), we first run the algorithm and check that the values of Q 2predict of the indicators of the dependent Adm. Sci. 2021,11, 83 21 of 27 variables of interest are all positive (Q 2predict > 0). The results of our analysis can be seen in Table 16. Table 16. Q2predict values. Indicator PLS RMSE MAE Q2predict ATTA1 0.935 0.602 0.319 ATTA3 0.946 0.695 0.357 ATT1 0.968 0.656 0.395 ATTA4 0.913 0.582 0.315 ECC2 1.397 1.248 0.015 ECC3 0.971 0.765 0.039 ECC5 2.187 1.889 0.061 TAD6 0.944 0.722 0.470 TAD2 1.065 0.784 0.395 TAD5 0.929 0.744 0.548 TAD7 1.177 0.877 0.338 TAD3 1.214 0.917 0.414 DTS4 0.943 0.732 0.433 DTS1 1.22 0.998 0.324 DTS2 1.123 0.895 0.405 DTS3 1.019 0.8 0.480 PU4 1.039 0.744 0.322 PU1 0.991 0.68 0.371 PU3 1.083 0.763 0.357 The next step was to check whether the prediction errors were symmetrically distributed—to do so, we analysed the skewness. If the asymmetry in the absolute value is less than 1, the RMSE should be used as a criterion for the prediction error; otherwise, the MAE should be applied. Table 17 shows the descriptive statistics of the indicators of the dependent variables of interest. Table 17. Descriptive statistics and choice of the error prediction criterion. Mean Median Min Max Standard Deviation Kurtosis Asymmetry Decision ATTA1 −0.014 0.121 −5.452 1.888 0.935 10.03 −2.375 MAE ATTA3 −0.01 0.151 −4.17 1.977 0.946 2.897 −1.284 MAE ATT1 −0.013 0.16 −5.097 2.277 0.968 6.413 −1.902 MAE ATTA4 −0.014 0.087 −5.37 2.441 0.913 10.957 −2.262 MAE ECC2 −0.005 0.45 −3.301 2.48 1.397 −1.069 −0.439 RMSE ECC3 −0.004 −0.177 −1.469 2.788 0.971 0.065 0.917 RMSE ECC5 −0.003 −0.658 −4.178 5.905 2.187 −0.873 0.492 RMSE TAD6 0.007 0.112 −2.597 3.458 0.944 1.189 −0.003 RMSE TAD2 0.013 0.177 −4.77 4.667 1.065 4.017 −0.426 RMSE TAD5 0.009 0.008 −2.675 2.583 0.929 -0.21 −0.144 RMSE TAD7 0.008 0.183 -4.677 4.335 1.177 2.172 −0.596 RMSE TAD3 0.013 0.257 -4.829 3.982 1.214 2.521 −0.976 RMSE DTS4 0.004 0.06 −3.591 2.932 0.943 1.243 −0.486 RMSE DTS1 0.004 0.234 −3.775 2.515 1.22 0.136 −0.631 RMSE DTS2 0.003 0.033 −3.777 2.762 1.123 0.44 −0.548 RMSE DTS3 0.002 0.034 −3.198 2.714 1.019 0.438 −0.464 RMSE PU4 −0.009 0.15 −5.068 2.905 1.039 4.433 −1.148 MAE PU1 −0.01 0.08 −5.084 2.128 0.991 5.888 −1.566 MAE PU3 −0.009 0.104 −4.859 2.322 1.083 3.274 −1.182 MAE Finally, we calculated the differences in the errors (as used as the RMSE or MAE criteria) between the predictions made using PLS and those of the linear regression model (LM) ignoring the specified path PLS model. For the predictions of PLS to be more accurate than those of LM, the errors of the latter must be greater, and, therefore, the differences Adm. Sci. 2021,11, 83 22 of 27 when subtracting them from the former must be negative. Table 18 shows the differences in prediction errors between both models. Table 18. The difference in RMSE or MAE errors between PLS and LM (predictive power). Indicator PLS LM PLS-LM RMSE MAE Q2predict RMSE MAE Q2predict RMSE MAE Decision ATTA1 0.935 0.602 0.319 1.05 0.763 0.141 −0.115 −0.161 MAE ATTA3 0.946 0.695 0.357 1.053 0.821 0.204 −0.107 −0.126 MAE ATT1 0.968 0.656 0.395 1.021 0.786 0.327 −0.053 −0.130 MAE ATTA4 0.913 0.582 0.315 0.915 0.691 0.312 −0.002 −0.109 MAE ECC2 1.397 1.248 0.015 1.918 1.558 −0.856 −0.521 −0.310 RMSE ECC3 0.971 0.765 0.039 1.18 0.94 −0.42 −0.209 −0.175 RMSE ECC5 2.187 1.889 0.061 2.645 2.258 −0.373 −0.458 −0.369 RMSE TAD6 0.944 0.722 0.470 1.074 0.795 0.315 −0.130 −0.073 RMSE TAD2 1.065 0.784 0.395 1.232 0.907 0.191 −0.167 −0.123 RMSE TAD5 0.929 0.744 0.548 1.092 0.834 0.374 −0.163 −0.090 RMSE TAD7 1.177 0.877 0.338 1.3 0.977 0.192 −0.123 −0.100 RMSE TAD3 1.214 0.917 0.414 1.331 1.032 0.296 −0.117 −0.115 RMSE DTS4 0.943 0.732 0.433 1.181 0.855 0.11 −0.238 −0.123 RMSE DTS1 1.22 0.998 0.324 1.429 1.124 0.072 −0.209 −0.126 RMSE DTS2 1.123 0.895 0.405 1.458 1.101 −0.003 −0.335 −0.206 RMSE DTS3 1.019 0.8 0.480 1.326 0.983 0.12 −0.307 −0.183 RMSE PU4 1.039 0.744 0.322 1.031 0.78 0.333 0.008 −0.036 MAE PU1 0.991 0.68 0.371 1.02 0.764 0.333 −0.029 −0.084 MAE PU3 1.083 0.763 0.357 1.126 0.874 0.306 −0.043 −0.111 MAE Thanks to this, we can see that the model shows great predictive power for DTS, TAD and ECC. It has positive Q 2 predict and negative differences for RMSE (recommended). It also has it for TAT + ATP and PU, and positive Q 2 predict and negative differences for MAE. Therefore, the model meets all the criteria and has high predictive power, that is, the ability to predict new results. 5. Discussion and Conclusions Due to the 2008–2014 crisis and its aftermath, Spain reduced its investment in R&D by 8.82% relative to its highest value of 1.36% of GDP in 2010. Despite that since 2016 it has been increasing (currently it is at a level of 1.24%), it is still far from the 3% target set by the EU in its Europe 2020 strategy and the countries of central and northern Europe that led this ranking (Spain occupies the 16th position). However, despite this decline in investment in R&D (Instituto Vasco de Estadística 2020), we agree with Yoldi (2016) and Ametic (2017) regarding the hopeful trend of cooperation with the aim of taking advantage of synergies between the public and private sectors. However, although efforts have been made, the health crisis caused by COVID-19 unleashed a new scenario of total exceptionality that will most probably lead to future cuts to combat the subsequent economic crisis produced by the pandemic. It therefore seems even more important for companies to strengthen their cooperation with universities as the best mean to promote, share and complete basic and applied research developed by both, to attract talent, to hire researchers, to use specialized equipment and scientific instruments at a reduced cost, to gain experience in the field of project management and direction, and to keep up to date with international scientific developments. To ease and contribute to the growth and promotion of closer collaborations between universities and companies in R&D, this research focused on exploring and discovering what factors of businesses (based in the province of Huelva) explained and determined the demand for technological services, and in what way. Through a cross-sectional study carried out on a sample of 96 companies, the most relevant characteristics of the same were analysed, according to their area, location, type of company, number of employees, Adm. Sci. 2021,11, 83 23 of 27 seniority, turnover, and activity sector. First, based on the theoretical and literature review, we designed a complex initial theoretical model, using the PLS-SEM methodology, which posed eight exogenous constructs as possible determinants of the demand for technological services by companies (the economic characteristics of the company (ECC), the attitude towards the performance of the technology (ATP), the perceived ease of use (PEU), the market conditions (MKC), the marketing actions (MKTA), the facilitating conditions (FC), the intention of the behaviour towards the adoption of technology (BITA), and the influence of the environment (IE), as well as four intermediate endogenous constructs (the business predisposition towards the adoption of technology (BPTAT), the technological attributes (TAT), the perceived utility (PU), and the decision to adopt technology (TAD), which were modelled in mode A (previously reflective). This model was tested using a questionnaire with 77 indicators adapted to the Huelva context. After successive phases of analysis and evaluation, both measurement models and the structural model, as well as the global adjustment of the model, different modifications were made, ranging from the purification of indicators and non-significant relationships, and reorganization of constructs, to changes in the measurement models of some latent variables that began to be modelled in mode B (previously formative). The later was carried out by means of a Confirmatory Tetrad Analysis (CTA-PLS) with an empirical foundation additional to the theoretical one on the measurement models, especially those modelled as B or formative mode (MKTA and ECC). The final model proposed for the business demand for technological services in the province of Huelva is a mixed model of factors and compounds, which is more parsimonious, and which is explained by four endogenous and six exogenous constructs. The variables that most influence the demand for technological services by companies were the influence of the environment (21.57%), market conditions (19.35%), and the technology adoption decision (16.13%). The economic characteristics of the company represented only 8.70% of the explained variance. These four variables alone explained 65.76% of the variance of the endogenous latent variable “Demand for Technological Services (DTS)”. Other important relationships were also identified, showing that 74.4% of the variance of the construct “Technological Attributes + Attitude Towards Technology Adoption (TAT + ATTA)” was also explained by the influence predictors of environment and perceived utility. Additionally, 64.2% of the variance of the construct “Decision to Adopt Technology (TAD)” was explained by the Intention Predictors of Behaviour Towards the Adoption of Technology, Facilitating Conditions, and TAT + ATTA. As for the measure of goodness of global fit of the model, it shows a proper fit of the proposed model, both in the three statistics suggested in the context of PLS-SEM, and in the exact fit tests based on bootstrap. The present study raises some interesting questions. Given that it is based on a sample of 96 companies in the province of Huelva, it would be interesting to analyse how the model would work in terms of predicting results for other companies and for those located in other geographical areas, especially those with a stronger presence of R&D centres. Additionally, further studies should be conducted to verify whether some of the economic characteristics of the companies, such as size, location, type of company, age, turnover, or activity sector, could act as mediating or moderating variables in the demand for technological services. It also seems justified to conduct a more qualitative analysis regarding the variable of marketing actions to confirm that the flow of the “information” is as good as the questionnaire suggests. An added benefit of conducting this kind of study is that it might help to analyse the supply of an institution’s scientific and technological infrastructure by carrying out an inventory of available resources, an idea that was inspired by the Office for the Transfer of Research Results of the Geological and Mining Institute of Spain, which issues annual catalogues of Technological and Scientific Offers (Instituto Geológico y Minero de España 2013), and by the Fundación Campus Tecnológico de Algeciras, which has a website where they gather everything related to transfers, and which can be easily accessed. Adm. Sci. 2021,11, 83 24 of 27 We hope that the results presented in this study will help lead to a better understanding of the motivating forces that drive the decision making of companies regarding the demand for technology services. Consequently, public bodies, universities, agencies, and research centres, as well as companies interested in innovation, development, and adoption of new technologies, will be able to work together with the aim of designing strategies to obtain a more desirable and positive response in relation to basic and applied research for a better use of resources, University–Business cooperation, the economy, and society in general. Author Contributions: Conceptualization, J.J.G.-M. and M.N.; methodology, J.J.G.-M.; software, J.J.G.-M.; validation, J.J.G.-M. and W.S.; formal analysis, M.N.; investigation, J.J.G.-M.; resources, J.J.G.-M.; data curation, J.J.G.-M.; writing—original draft preparation, M.N.; writing—review and editing, W.S.; visualization, J.J.G.-M.; supervision, J.J.G.-M.; project administration, M.N.; funding acquisition, J.J.G.-M. All authors have read and agreed to the published version of the manuscript. Funding: This research received no external funding. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Not applicable. Conflicts of Interest: The authors declare no conflict of interest. References Albort-Morant, Gema, Jorg Henseler, Gabriel Cepeda-Carrión, and Antonio L. Leal-Rodríguez. 2018. Potential and Realized Absorptive Capacity as Complementary Drivers of Green Product and Process Innovation Performance. Sustainability 10: 381. [CrossRef] Alhassany, Hiba, and Faisal Faisal. 2018. Factors Influencing the Internet Banking Adoption Decision in North Cyprus: An Evidence from the Partial Least Square Approach of the Structural Equation Modeling. Financial Innovation 4: 29. [CrossRef] Ali, Faizan, S. Mostafa Rasoolimanesh, Marko Sarstedt, Christian M. Ringle, and Kisang Ryu. 2018. An assessment of the use of partial least squares structural equation modelling (PLS-SEM) in hospitality research. International Journal of Contemporary Hospitality Management 30: 514–38. [CrossRef] Allan, Jacqueline E. M., Annarita Ferreri, Sergio Grande, Roberto Giannantonio, and Francesco Matteucci. 2019. Technology Transfer in Nanotechnology. Luxembourg: Publications Office of the European Union. Alsajjan, Bander, and Charles Dennis. 2010. Internet Banking Acceptance Model: Cross-Market Examination. Journal of Business Research 63: 957–63. [CrossRef] Ametic. 2017. Objetivo Aprovechar Sinergias del Sector Público y Privado. Available online: http://ametic.es/es/noticias/participaen-el-mapa-de-demanda-temprana-en-el-sector-tic-salud (accessed on 27 April 2020). Andreev, Pavel, Tsipi Heart, Hanan Maoz, and Nava Pliskin. 2009. Validating Formative Partial Least Squares (PLS) Models: Methodological Review and Empirical Illustration. In Thirtieth International Conference on Information Systems. Phoenix: ICIS Procedings, pp. 1–17. Avkiran, Necmi K. 2018. Rise of the Partial Least Squares Structural Equation Modeling: An Application in Banking. In Partial Least Squares Structural Equation Modeling. Edited by Necmi K. Avkiran and Christian M. Ringle. Cham: Springer, pp. 1–29. Chastenet, Denis, Bernard Reverdy, and Eric Brunat. 1990. Les Interfaces Universitès Enterprises. París: ANCE/Les Editions D’Organisation. Chin, Wynne W. 2010. How to Write Up and Report PLS Analyses. In Handbook of Partial Least Squares: Concepts, Methods and Applications. Edited by Vincenzo Esposito Vinzi, Wynne W. Chin, Jörg Henseler and Huiwen Wang. London and New York: Springer, pp. 655–90. Cohen, Jacob. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Hillsdale: Lawrence Erlbaum Associates. Conde-Pumpido Touron, Rogelio, and Lidia Cerezo García. 2019. Investigación y Transferencia de Conocimiento en las Universidades Españolas 2017. Madrid: CRUE Universidades Españolas. Available online: http://www.crue.org/Documentos%20compartidos/ Publicaciones/Encuesta%20Investigación%20y%20Transferencia%20de%20Conocimiento/Resumen%20I+TC%202019_V5.pdf (accessed on 7 March 2021). Cotec Foundation for Innovation. 2020. Evolución de la I+D. Available online: https://cotec.es/en/observacion/evolucion-de-la-i-d/ 169d9768-f54d-821f-7c17-2bae4afff56e (accessed on 24 May 2021). Coulter, David. 1999. The epic and the novel: Dialogism and teacher research. Educational Researcher 28: 4–13. [CrossRef] Cygler, Joanna, and Sylwester Wyka. 2019. Internal barriers to international R&D cooperation: The case of Polish high tech firms. Forum Scientiae Oeconomia 7: 25–45. [CrossRef] Danks, Nicholas P., and Soumya Ray. 2018. Predictions from Partial Least Squares Models. In Applying Partial Least Squares in Tourism and Hospitality Research. Edited by Faizan S. Ali, Mostafa Rasoolimanesh and Cihan Cobanoglu. Bingley: Emerald Publishing Limited, pp. 35–52.