Exploring science park location choice

.


Introduction
Science parks (hereafter: SPs) are the result of supply-driven policy that aims to facilitate networking and economic activity among technology-based firms, universities, and research institutes (Edler and Georghiou, 2007).An SP can be defined as: "a real estate or area development, managed by an on-site management company.It is home to knowledge organisations, such as research institutes, higher educational institutions and technology-based firms in all business development phases.Resident organisations can make use of a wide range of shared or private facilities, such as R&D facilities, business support, leisure and other amenities" (Ng et al., 2019a, p. 726).In addition, SPs might offer a bridging function between academic knowledge and industry (Lamperti et al., 2017).Existing SP literature focuses largely on proving the impact of presence on an SP on tenant firm performance and therefore policy effectiveness (Albahari et al., 2010).Studies include explanatory variables, such as characteristics of technology-based firms, but attention to characteristics of the SPs themselves is often missing (Ramírez-Alesón and Fernández-Olmos, 2018).Ng et al. (2019a) distinguished SP types in Europe on the presence of knowledge-intensive organisations, facilities offered, size and ownership characteristics, and showed that different configurations exist, which is generally ignored.
From an economic perspective, the supply of real estate is inelastic as a result of the long lifespan of buildings, while the demand for adequate space and quality can be more dynamic and related to the needs of users (Geltner and Miller, 2001).Within the SP context the supply-side consists of design-related attributes like geographical location, buildings, facilities, and services.The demand-side consists of the needs of technology-based firms and other resident organisations towards what an SP can supply to them.The mismatch between supply and demand has been acknowledged by SP managers as troublesome (Albahari et al., 2019).It might negatively affect tenant performance as they do not have access to adequate facilities and services that optimally support their R&D and business activities.
The demand-side of SPs is often studied through the perceived benefits by its users or their most important motivations to move to a SP (e. g.Link and Scott, 2003;Van Dierdonck et al., 1991;Westhead and Batstone, 1999).In addition, studies have compared growth rates and/ or perceived location benefits of new technology-based firms on and off SPs (e.g.Ferguson and Olofsson, 2004;Dettwiler et al., 2006).A third, more rare type of studies explored specifically how certain SP facilities and services create value/benefits for their tenant firms (Albahari et al., 2019;Ng et al., 2019bNg et al., , 2021)).These studies assess the conditions that make an SP more or less valuable to its tenant firms (Gwebu et al., 2019).However, these studies focus mostly on value creation once a firm is located on the SP and starts interacting with other residents, using facilities, and so forth.We argue that it is also important to understand the motives of firms to locate at a specific SP in the first place.Research about location choice is very limited and little is known about how tenant firms choose between locations they could locate to.As far as we know only a few studies offer insights in SP location decisions (e.g.Westhead and Batstone, 1998;Lindelöf and Löfsten, 2003;Wright et al., 2008;Van Der Borgh et al., 2012).Although several of these studies rank location factors for on and off park firms on perceived importance, their approaches do not allow to identify trade-offs that tenants make or their willingness to pay.
Gaining a better understanding about how firms choose between SP locations is first of all important for SP managers.When operating in a dynamic environment, SPs continuously need to develop and differentiate themselves from other competing new SPs (Koh et al., 2005).To be successful, they must attract the most suitable and promising firms to reside on the park (Lecluyse et al., 2019).Especially when attracting multinationals and their research centres there might be a severe competition between sites to become the most attractive location (Van Dierdonck et al., 1991).The latter is not an easy task.SP management needs to balance aspirations for offering the most attractive location benefits with limited resources to achieve that objective.Park managers face the challenge of designing the SP in such a way that they can better fulfil the needs of their potential residents while knowing that it is impossible to design it in such a way that all desired attributes are simultaneously offered and utilised at the desired levels.Moreover, heterogeneity in needs of potential future residents prevents a one-sizefits-all solution.Hence, there are limits to SP attraction strategies, highlighting the need to make better use of the resources available (Etzkowitz and Zhou, 2017).This is only possible if SP management enhances its understanding of potential residents' location choice preferences.
Second, policy-makers trying to increase economic development through SP strategies will profit from the outcome of this study.Potential residents may use different trade-offs to assess the level of attractiveness of each location option under consideration.Based on these trade-offs, new firms might consider other locations that more clearly match their needs, which could jeopardise policy-makers' objectives (Good et al., 2019).As SPs are often (partially) financed by governments in order to reach policy goals, it is vital that these parks are configured effectively (Monck and Peters, 2009).
Our study will offer the needed insights in effective SP configuration by clarifying and comparing the trade-offs that firms make in choosing a potential new SP location.In the context of SPs, trade-offs can be defined as management choices that increase the delivery of one (or more) SP attributes at the expense of the delivery of other SP attributes.We show which physical and non-physical attributes of the SP location, facilities, and services are preferred in a potential resident's location choice.Moreover, we also include a diverse set of organisational characteristics (i.e.covariates) that might impact firms' preferences, especially whether on and off park firms make different trade-offs.
So far, perceived benefits of firms in relation to the SP location have been studied through ranking or Likert measures (e.g.Westhead and Batstone, 1999;Ferguson and Olofsson, 2004).These studies show the importance of various SP attributes separately, but the relative importance or trade-off that technology-based firms make among typical attributes remains unknown.Providing insights into this relative importance is a vital step forward in effectively configuring new SP locations or changing existing SP configurations over time.By including these trade-offs into our analysis, we are able to feature a complexity that is greater than what is often described in SP literature.
We even take it a step further by including the economic amount a (potential) tenant is prepared to pay for locating at the SP.This so-called 'willingness to pay' highlights those aspects that positively influence (potential) tenants to contribute willingly to their location choice.Firms that currently do not have access to this specific SP attribute are willing to pay a certain percentage more for alternative locations that do offer this attribute, while holding all else equal.This information is very useful to estimate the economic costs and benefits of alternative SP configurations, thereby contributing again to a more effective use of the limited available resources.SP managers can use the 'willingness to pay' for example for a better segmentation towards preferred residents or for tailoring their prices to the willingness to pay.As far as we know, SP studies have not included the willingness to pay among firm decision makers for specific park design characteristics yet.
Besides our theoretical contribution highlighting the effective use of resources when developing an attractive SP location, our paper also offers a more methodological contribution.The stated preference approach that is used (commonly used in travel-behaviour research, see Hensher et al., 2015) provides the option to include non-existing SP configurations as well, which yields additional insights for SP managers, policy-makers and SP theory.To our knowledge, the stated choice method has not been used within the context of SP location choice and could pave the way for a new branch of quantitative research in SP demand.
And last, existing SP tenant firms are somewhat biassed due to their choice to select the SP location in the first place, which implies they already have specific preferences (Sanni et al., 2009).This issue should be reduced with a control group of off-park technology firms that are not subjected to influences from the SP location and have not chosen an SP location in the first place (Siegel et al., 2003;Vásquez-Urriago et al., 2016), when studying trade-offs and willingness to pay.
In sum, this exploratory study addresses preferences and trade-offs of both on and off SP technology-based firms when considering important SP design attributes in times of relocation, including their willingness to pay for these attributes.A heterogeneous group of decision-makers from technology-based firms located on different locations, within various business development phases, technology domains and firm ages in the Netherlands is approached to participate.
The paper is structured as follows; first literature on location choice preferences of technology-based firms is reviewed as input for the design of the stated choice experiment.The method (sampling procedure, operationalisation, design of the choice experiment, and discrete choice modelling) are explained in Section 3. Next, the results including the sample description and choice modelling results are discussed.Lastly, in Section 5 the main conclusions of this study, limitations and future research are provided.

Technology-based firm's location choice
In understanding location choices of firms, it is vital to consider the rationality of maximising profits or utility through the choices they make (Pusterla and Resmini, 2007).The resulting uneven distribution of industrial activity among geographical locations is explained through agglomeration economies, where firms experience proximity benefits among stakeholders (Fujita et al., 1999).Firms are largely concentrated when transportation of raw materials or products would otherwise lead to high costs.This basic economic principle would imply that firms that indulge in technology-based activities do not necessarily have to colocate, as the assets of these firms are mainly intellectual property; i.e. ideas and patents and other mechanisms would dictate the transfer of these assets (Stuart and Sorenson, 2003).However, early examples in the US with high concentration of technology-based firms within Silicon Valley and Boston's Route 128 and in Europe the cases of Cambridge in UK and Heidelberg in Germany, showed that knowledge-intensive activity does have a spatial dimension too (Saxenian, 1996;Cooke, 2001).Nonetheless, Classical Marshallian district theory might not fit the SP context for several reasons.Firstly, SPs are not always focused on production processes, which is the case for the Marshallian concept, but more often aim to enhance innovative and creative capacity (Salvador et al., 2013).Secondly, the classical buyer-supplier relationships are less common within the SP area, while partnerships are more readily available between businesses and academia (Ravix, 2014).Lastly, despite firms and institutions being clustered together organisational distance could be a barrier as well (La Rovere and de Jesus Melo, 2012).
Although many innovation networks are geographically concentrated due to transportation costs, the social proximity among actors cannot be ignored (Boschma and Frenken, 2009).The uneven spatial distribution of entrepreneurial activity is likely related to the capabilities of firms, availability of opportunities and human capital (Baptista and Mendonça, 2010).It is argued that the geographical proximity towards multiple knowledge sources (i.e.universities, research institutions and firms) enhances the ease of creating and transferring tacit knowledge (Ponds et al., 2007).Feldman (1999) posited that localised knowledge spillovers contribute to innovation and that these spillovers are represented in patent citations, people, and R&D.Furthermore, new technology-based firms located on SPs seem to have significantly more R&D output than off-park counterparts through more efficient investing in innovations as a result of public policies, location benefits and agglomeration effects (Yang et al., 2009).The following subsections discuss specific aspects that have been shown to affect the location choice of technology-based firms.

Proximity to university
The geographical proximity between a university and firms is likely to reduce search costs for valuable knowledge and to improve chance encounters between individuals that enable innovative opportunities (Feldman, 1999).Moreover, universities and public research organisations could serve as anchor organisations that create value for other closely located organisations (Clarysse et al., 2014).An anchor organisation is highly active in R&D and creates knowledge externalities through proximity (Agrawal and Cockburn, 2003).Geographical proximity has been suggested by Ponds et al. (2007) to be a means to overcome institutional distance between firms and universities for scientific collaboration.Both Ferguson and Olofsson (2004) and Dettwiler et al. (2006) found that new technology-based firms most valued the proximity to university among different perceived benefits.Firms that have previous collaborative agreements with universities are more able to strengthen their absorptive capacity and enhance their innovation performance (Díez-Vial and Fernández-Olmos, 2015).However, Woerter (2012) argued that small firms will likely have lower absorptive capacity than larger firms.Therefore, smaller firms will be more eager to collaborate with universities and a high technology proximity between these firms and university could contribute to knowledge transfer.Similarly, Albahari et al. (2017) posited that firms benefit from university knowledge, but more effort is required for commercialising this knowledge.Moreover, they did not find significantly more collaborations between firms and universities on Spanish SPs.Furthermore, Link and Scott (2006) revealed that employee growth decreases on SPs when geographical distance of universities increases, which is largely related to the limited means for knowledge transfer.

Firm diversity
The effects of (dis)similarities among activities and technology domains of firms have been widely researched.Recently, SPs are seen as science and technology zones that facilitate technology transfer among R&D firms on-park (Spithoven, 2015).In addition, industrial boundaries are blurring with seemingly disconnected firms, but technological compatible, yielding surprising results through collaboration (Munir and Phillips, 2002;Prahalad and Venkatram, 2003).Firms active within a strict technology domain are likely to achieve short-term success within their own area of expertise.In contrast, firms active in different technology domains might explore newer domains and gain success in the long-term (Rosenkopf and Nerkar, 2001).In the SP context, firms are more likely to collaborate with each other through a wider range of activities (Vásquez-Urriago et al., 2016).New technology-based firms who can complement each other with joint collaborating and exporting activities are more likely to have higher innovation performance, whereby the disadvantages of knowledge spillovers are diminished (Ramírez-Alesón and Fernández-Olmos, 2018).Shearmur and Doloreux (2000) argued that the sectoral mix on SPs can be explained by two reasons: as a strategy to contribute to synergy among firms or as a real estate decision to avoid vacant premises.Research in Italy on 'specialised' SPs (i.e.hosting many firms active in a few sectors) and 'general' SPs (i.e.many firms active in different sectors) shows that the former SPs have a positive effect on firm investments, while the latter are more able to improve sales performance of tenant firms (Liberati et al., 2016).
Firms active in technology sectors, such as biotechnology, are more likely to co-locate with other similar firms as a means to get access to the specialised local educated workforce, experience, and specific materials (Stuart and Sorenson, 2003).Regions that co-locate a high number of firms with the same specialisation are exposed to less risk to fail through the proximity of specific suppliers, markets, infrastructure, networks, human talent and potential knowledge spillovers (Renski, 2011).Moreover, Link and Scott (2006) showed that SPs focused only on information technologies grow faster than parks focused only on biotechnology or multiple different sectors.Koçak and Can (2014) found some evidence supporting that more knowledge sharing and client ties are found at SPs with more tenant firms in the same industrial activity group.The close proximity of similar firms active in innovation could lead to both competition and collaboration (Lamperti et al., 2017).
In contrast, a wide range of sectors could lead to technological breakthroughs, as innovation is often a product of the technological recombination of inventions from various technologies, components, or processes from multiple firms (Adner and Kapoor, 2010).A diversity of firms complemented with non-profit knowledge institutions allows for a diverse flow of information and different measures for success (Powell et al., 2010).It is this diversity that allows each individual actor to be more able to recognise new opportunities within their own specialisation, while the collection of organisations will be able to retrieve knowledge from different domains (Van Der Borgh et al., 2012).

SP facilities and services
SPs provide firms, higher educational institutions, and research institutes both configurative and process-related amenities, which are respectively the physical infrastructure (facilities) and the services that contribute to their organisational goals (Albahari et al., 2019).Various types of facilities are provided on the SP area, such as specialised R&D facilities (e.g.laboratories, cleanrooms, piloting rooms), business facilities (e.g.offices, meeting rooms, conference rooms, canteens, restaurants, shops) and leisure facilities (e.g.cinema, sport facilities) (Ng et al., 2019a).Business specific services are for example administrative, marketing and managerial support, venture capital access, training programs, while other services include social networking events, park management, cleaning and maintenance (Ratinho and Henriques, 2010).These facilities and services are provided for private use or shared usage among tenants in which the latter allows for economies of scale and collaborative purposes (Chan and Lau, 2005;Brinkø et al., 2014).Dettwiler et al. (2006) argued that new technology-based firms on SPs are more able to develop their networks through their contractual agreements with the SP and availability of facilities.Similarly, Schiavone et al. (2014) found that SP tenant firms have access to more opportunities and resources within the SP environment which benefits their innovation performance.For new technology-based firms the access of facilities and services within their initial start is beneficial, but when these firms mature, they are less likely to share knowledge as competition rises (McAdam and McAdam, 2008).
In addition, communication infrastructure, such as ICT facilities provided by business incubators is essential for knowledge management and contributes to regional innovation performance (Wang et al., 2020).Specifically, for these smaller and younger firms the networking events and training are of interest as they lack the financial and organisational capacity to do so on their own (Chan and Lau, 2005;Löfsten and Lindelöf, 2005).Networking events enable the interaction between different co-located actors and repeated interactions between these actors is expected to contribute to knowledge sharing (Inomata et al., 2016).SP tenant firms associate both training programs and business networking events with knowledge sharing and collaboration and to be closer to the university and new clients, while networking events also serve the purpose of meeting other firms (Ng et al., 2019b).Soetanto and Jack (2013) posited that within their networking activities new technology-based firms are more interested in seeking intangible resources (i.e.combine technical or market-related knowledge) than tangible resources (i.e.combine assets or use of R&D facilities and equipment).This suggests that these younger firms are less likely to collaborate in a physical sense through shared use of facilities and colocation might be less important for them.However, the transfer of difficult to codify knowledge is eased through repeated face-to-face interactions (Storper and Venables, 2004).
To summarise, the first main driver of the technology-based firm's location choice are location-specific factors.These factors are related to the means of transportation and the urban context of the firm's location, which is related to agglomeration benefits and access to markets.The second driver is proximity of a university, which serves as a potential source for knowledge, collaborative opportunities and human talent.Furthermore, the firm diversity of a location can be focused on firms active in similar or in a wide range of technology sectors with different impacts on collaboration opportunities, knowledge production and valorisation.Lastly, the provision of facilities and services provided by SPs involve specific R&D facilities and equipment, but more general business support facilities/services and leisure facilities, which allow tenant firms to focus on their core business activities, to network and share knowledge.These four main drivers will be further operationalised in the next section in order to incorporate them in the experimental design.

Methods
To gain insight in the design-related preferences of technology-based firms in relation to SPs, a stated choice experiment is designed.The stated choice framework is an experimental approach that allows for the estimation of the utility of attributes within the trade-offs that firms make (Hensher et al., 2015).In this section the sampling of decisionmakers is briefly discussed, followed by an operationalisation of the main location drivers derived from literature for the experimental design.Then the covariates (i.e.organisational characteristics) are operationalised in order to study possible preference differences among firms.Lastly, the analysis procedure of discrete choice modelling is discussed.

Sampling of technology-based firms
The stated choice experiment is administered by an online survey among a sample of firms, in order to gain insight in design-related preferences of technology-based firms.The desired respondents are Clevel representative decision-makers of technology-based firms on and off SPs; i.e.CEO, CTO, CFO, COO, etc.A disadvantage of targeting these respondents is that these representatives often have tight time schedules as a result of daily business operations, which could limit the number of responses (Mintzberg, 1973).However, executive-level employees are considered subject matter experts and a reliable source of information of organisational processes (Norburn, 1989) and, therefore, can also participate as respondents.
In total, 828 SP firms are contacted through email that are located on the ten largest SPs in the Netherlands based on desk research.The majority of SPs are focused on high-tech 1 and/or biotechnology sectors and located within the same area or within the same city of a technical university or university medical centre (see Appendix A).All parks provide facilities and services for shared or private use for technology firms, research institutes and service providers.For the off-park firms, zip codes not similar to those of SPs are considered.Through online desk research a list is established with 482 off-park technology firms with publicly available email addresses to be contacted to partake in this study.Both on-park and off-park technology-based firms are defined by their technological sectoral focus and business activities related to producing new products, services, systems, or processes.The survey was hosted online in the period between December 2018 and April 2019 with reminders sent to firms in January and March 2019.

Design of attributes and experiment
The four main drivers for the technology-based firm's location derived from literature will serve as the input for the operationalisation of the attributes used in the stated choice experiment.In this study the hypothetical alternatives are area developments with typical SP attributes to be evaluated by respondents.For the facilities and events attributes, these are communicated towards respondents as present within the immediate location and therefore within walking distance.The configuration of the hypothetical locations consists of seven attributes with each varying among three levels.The seven attributes and their levels can be found in Table 1.

Design of attributes
The location-specific factors, which involve modes of transportation and accessibility are adopted from De Bok and Van Oort (2011).The distance values (0.8 and 2 km) related to 'station location', 'suburban location' and 'highway location' reflect the densely populated polycentric character of the Netherlands (Burger and Meijers, 2012).The Dutch firm location policy is largely focused on accessibility on three dimensions: the proximity to the central business district, to railway stations and to highways (Schwanen et al., 2004).
Proximity to a university that is relevant for the decision-maker is defined by the geographical distance between the area and the universities common in the Netherlands (see Appendix B).The majority of high-tech focused SPs are located in the same area as a technical university or not in the vicinity at all.Some biotechnology-focused SPs are located in the same area as a university medical centre or in a distance of approximately 50 km (i.e. in a different city).
Firm diversity refers to the (dis)similarity of technology domains; i.e. type of technology sectors, activities and R&D output.This attribute is derived from Liberati et al. (2016) and defined with the levels: 1) all technology domains are present in the area, 2) firms are focused on a limited number of technology domains including that of the decisionmaker and 3) firms are focused on the same technology domain as the decision-maker.
The provision of facilities and services is captured through the attributes R&D facilities, shared facilities (i.e. business support and leisure facilities) and events that the SP location can provide.Business support facilities are for example conference rooms, meeting rooms, dining facilities, while leisure facilities are sport facilities, cinema etc. Events held in the area are networking events and training events held on location.These two types of services are distinguished as, on one hand, networking events aid firms in seeking new opportunities (Löfsten and Lindelöf, 2005) and, on the other, training programs aid firms in their business activities (Albahari et al., 2019).
Finally, the cost attribute that enables the estimation of the willingness to pay for the other attributes (levels) is the location use expenses for firms (i.e.rent).This is an important factor that impacts the overall utility of a location.Specifically, for new technology-based firms who need more development time, the possible rental subsidies from SPs are important, as normal rent prices on SPs are generally higher than market prices (Chan and Lau, 2005).Dettwiler et al. (2006) found that off-park new technology-based firms are more concerned with rental cost than on-park, which the authors attributed to different contractual agreements and facility solutions on SPs.The costs of use include the costs of acquiring or leasing the facilities, operational costs (i.e.maintenance and energy expenses) and the costs of additional services.Saving costs from infrastructure allows firms to allocate resources to their innovation efforts (Durão et al., 2005).Within the SP context renting facilities is arguably the most common arrangement for technology-based firms (Chan and Lau, 2005;Ratinho and Henriques, 2010;Squicciarini, 2008).In general, more central areas are more expensive, while Audretsch et al. (2005) found that for firms the cost of being closely located to universities surpasses the benefits of knowledge spillovers.The cost parameter is defined with three levels which refers to decision-makers' existing total cost of use.This parameter is defined at the levels; 1) plus 10 % on total current cost of use, 2) same as the total current cost of use, and 3) minus 10 % on total current cost of use.Inquiring respondents for their actual total current cost of use was not considered as agreements are individually made and are likely to be influenced by many unobservable factors.Moreover, rental prices of physically similar spaces may differ significantly due to location factors (Geltner and Miller, 2001).

Design of the experiment
The seven attributes are the characteristics that define the hypothetical SP alternatives within the choice tasks presented to decisionmakers of technology-based firms.The explicit question in each choice task is: 'suppose your organisation or branch office should relocate and you could choose between the following two hypothetical location alternatives or choose to not relocate.The three alternatives are identical for all characteristics that are not explicitly mentioned'.See Appendix C for an example choice task as presented to respondents.Each respondent is presented with nine choice tasks each with two hypothetical alternatives and an opt-out option.The use of an opt-out option is suggested to limit the hypothetical bias to some degree and result in more realistic preferences (Ladenburg and Olsen, 2014) and could link the alternatives closer to the respondent's actual situation (Hensher et al., 2015).In the current study, the opt-out option is for firms equal to preferring their current situation.The inclusion of an opt-out option does not increase the required number of choice sets as it is static among all tasks.
In the SP context, an unlabelled choice experiment would be more appropriate as respondents may not fully understand the concept or know its interchangeable names; i.e. science, technology or research park, campus (Hansson et al., 2005).Furthermore, the labels of these park types offer no meaningful associations for the decision-makers compared to, for example, choice experiments regarding transport choice (e.g.bus vs car).In this way, the choice data collected is solely based on the attributes of the unnamed alternatives and leads to design principles also suitable for broader knowledge-based and businessoriented real estate developments.In addition, the use of unlabelled alternatives decreases the required number of choice sets significantly.The number of possible design profiles for unlabelled choice alternatives is L H with L as the number of attribute levels and H as the number of attributes (Hensher et al., 2015).It is crucial to limit the number of design profiles in order to reduce the cognitive burden of respondents.In an unlabelled experiment with seven attributes with each having three levels, the number of all possible design profiles equals 3 7 = 2187.As this number is far too large to present to respondents this should be reduced significantly.Therefore, an orthogonal fractional design is used that consists of a smaller fraction that still allows for the estimation of the utilities for all attributes.The smallest design based on seven attributes each with three levels consists of 27 profiles (Hahn and Shapiro, 1966).This design allows for estimation of all main effects as well as the estimation of interaction effects (i.e.multiplication) between the first and second, and second and third attribute ('location', 'university' and 'firm diversity' respectively).Within each choice task it is crucial to pair two different profiles as two identical profiles do not represent a choice task.Furthermore, choice sets with different unique pairs of profiles are useful to reduce order effects (Hensher et al., 2015).Therefore, each profile in the fractional design is paired with another profile.Profiles 1 to 9 are shown to respondent 1 within the alternative 1 slot and each of those profiles are paired with one of the other 26 profiles in the alternative 2 slot, profiles 10 to 18 for respondent 2, etc.To ensure a somewhat random order, the fractional design profiles are paired eleven times whilst making sure that all profile pairs are unique across all versions.

Operationalisation of covariates
In order to investigate possible differences in preferences between technology-based firms, a number of covariates are included.These covariates are mainly organisational characteristics and characteristics of the respondent's current location (see Table 2).The reason to include these covariates in the online survey is to find possible differences in the base preference for the opt-out option (i.e.current situation) and differences in attribute evaluation among decision-makers.In order to avoid multicollinearity issues the Pearson's correlation test is used to seek potential redundant covariates.Covariates with correlations that exceed a certain threshold can be omitted from the further analysis of W.K.B. Ng et al. the choice data (variables with r ≤ 0.5 are suitable).In the survey the covariates are inquired before the choice experiment as it allows the respondent to familiarise with a majority of the attributes within the stated choice experiment.
The first covariate is to check if the responding firm's headquarter is currently located on an SP in the Netherlands.Larger and more mature firms are likely to have multiple business locations and therefore firm age and size are also included.In terms of technological focus and activity, the covariates 'sectors' and 'new product development activities' are used.In addition to technology-related sectors, SPs can provide a base for operations for other service firms (i.e.consultancy and servicing companies).The last group of covariates consists of the access to different facilities and services.Within their current headquarter location, firms are asked if they have access to the various facilities in their current area.For the access to training events and networking events a Likert scale is used to measure the frequency of relevant events held for these firms in their current location.

Discrete choice modelling
Through the analysis of choice data of the technology-based firms the part-worth utility for each attribute level can be identified.Respondents convey their stated preference through their choice among two hypothetical location alternatives and to not relocate.The main aim of this method is to fit the choice data within a linear function.In this section the utility function, mixed multinomial logit model and the willingness to pay estimation are discussed.

Utility function
The standard utility function for decision-maker n for an alternative i (U in ) is the sum of the part-worth utilities (V ijn ) of all attributes j and an error component ε (Eq.( 1)).This latter component embodies what is unobserved within the choice experiment, as not all relevant attributes can realistically be included (Hensher et al., 2015).In the choice experiment, the decision-makers of technology-based firms receive hypothetical relocation options and a choice of not relocating.The reference utility component α n is included as the utility of not relocating and, hence, represents a base utility U 0n (Eq.( 2)).The two utility equations are defined as: The observed part-worth utility V ijn of an attribute is the product of an attribute preference parameter β j coefficient and attribute level x ij .The β j parameters are estimated and yield numeric values, while x ij corresponds to the discrete values of attribute levels.Effects coding is used for the attributes.Through effects coding, the reference third base level ('different city' for the university example) is fixed in order to estimate the two other level variables (x ij1 and x ij2 ).It follows that 'same area' is coded as x ij1 = 1, x ij2 = 0, 'same city' as x ij1 = 0, x ij2 = 1 and 'different city' is codes as x ij1 = − 1, x ij2 = − 1.The part-worth utility V i, universisty3,n of the third reference level is derived from (− 1* β i,universisty1 ) + (− 1* β i,universisty2 ), in which the sum of the three levels equals zero.Therefore, the estimated utilities for the two parameters are relative to the referenced base level.This method reflects the fact that the choice of the reference level is arbitrary and only the difference between utilities is meaningful.In the current study, the previously mentioned main effect of the attributes are estimated.In addition, the interaction effects of the product of two attributes and the effects of the covariates on the attribute preferences are also estimated.
For the interaction effects of the attributes, only the multiplication of the first three attributes can be estimated as a result of the chosen design (Hahn and Shapiro, 1966).These interactions are: the attribute levels of 'location' * 'university' and 'university' * 'firm diversity'.A preference parameter (δ jj' ) is estimated, which is the added utility of the combination of both attributes in an alternative.Furthermore, preferences among decision-makers are likely to be different as a result of heterogeneity within the group of respondents.De Bok and Van Oort (2011) argued that relocation behaviour is dependent on firm attributes such as firm size, firm age and sector.These covariates (z kn with k as the level of the covariates) enter the standard utility function as interaction effects (δ jk ) of attributes (e.g.'university in the same area' * 'SP firm').In order to limit redundant covariates, only variables are considered that are not strongly significantly correlated through a Pearson's correlation test to avoid multicollinearity issues (Booth et al., 1994).The inclusion of covariates in the utility function allows for estimating differences in part-worth utilities among decision-makers that have specific firm characteristics.
The part-worth utility (Eq.( 3)) for this study is defined as: As this equation indicates, the part-worth utility (V ijn ) of an attribute j and alternative i is the sum of three components; the main effects of the attributes (product of attribute preference parameters (β j ) and attribute levels (x ij ), the interaction effects with other attributes (δ jj' * x ij * x ij' ), and the interaction effects with the covariates (δ jk * x ij * z kn ).

Mixed multinomial logit model
As preferences across firms may vary, the mixed multinomial logit model (MMNL) is a suitable extension of the MNL model to estimate the part-worth utilities.The MMNL assumes that all or some of the parameters are random among all decision-makers.Therefore, this method considers the heterogeneity among respondents and the panel data structure by estimating a distribution rather than a point-estimate for each parameter (i.e.multiple observations per respondent) (Train, 2009;Hensher et al., 2015).Usually, random parameters are expected to follow a normal distribution and represent the difference in preferences for those attribute levels.To estimate the distribution, a simulated log likelihood method is used where the random preference parameters are drawn from a (normal) distribution.In this method, the use of Halton draws results in lower simulation errors (Train, 1999).For this study, models are estimated using 1000 Halton draws (Bhat, 2003).For each attribute, only one level can enter the model as a random parameter; for arbitrary reasons this will be the first level.The goodness-of-fit of the estimated MMNL model is measured through the ρ 2 and the ρ 2 adjusted with the latter considering the number of parameters.The ρ 2 compares the log likelihood of the null model (alternatives have equal chances of 0.33 within a choice task) and the log likelihood of the estimated model.These measures are similar to the R 2 , which expresses the accuracy of the model to predict the data in linear functions.An R 2 of 0.8 for linear functions is considered to be similar to a ρ 2 of 0.4 (Domencich and McFadden, 1975).Therefore, results with ρ 2 between 0.2 and 0.4 suggest a well-fitting model (Louviere et al., 2000).

Willingness to pay estimation
The trade-off between attributes can be expressed in monetary terms as the willingness to pay (wtp), which is the ratio of the utility parameters of an attribute level of interest (x ij ) and the cost attribute (x c ).This ratio expresses the linear relation between 1 unit change of an attribute and how much the cost attribute has to change to keep the total utility constant.Wtp for attribute j is defined as (Hensher et al., 2015); where β j and β c are the marginal utilities for attribute j and the cost parameter, respectively.In order to meaningfully express the willingness to pay for attribute levels, in this case, the estimated values need to be multiplied by 20 % (i.e. the range of the cost attribute from minus 10 % to plus 10 %).

Sample characteristics and covariates
In total the survey was completed by 69 respondents with a majority of them active in a decision-making role, which shows that it reached the relevant persons in the organisation (Table 3).Just above half (54 %) of the responding technology-based firms are currently located on a Dutch SP.The majority of the firms are located either in a suburban or highway location.Both the distribution of firm size and firm age suggest that the sample contains more firms in the smaller and younger categories compared to the overall sample.The sub sample of 37 SP firms is in terms of location type and firm age comparable as there were no significant differences found when comparing it with the population of SPs at the time. 2 The sampled firms are active in a wide range of technology sectors including services with 'computer and software engineering' and 'bio/medtech' being the two most common sectors.For each firm the total number of selected technology sectors is computed to divide the sample into two groups.41 firms are active in one or two sector(s) (i.e.'high technology sectoral focus').While the remaining 28 firms are active in more than two or zero technology sectors (i.e.'low technology sectoral focus').Specifically, firms operating in 'industrial manufacturing', 'sensors', 'agriculture', 'energy' and 'optics' are significantly more active in more than two sectors. 3The vast majority of firms (88 %) is active in one of the phases of new technological product development with more firms active in the latter three phases.For access to facilities and services; meeting rooms, dining facilities, sport centres and laboratories are the most accessible for respondents.The more specialised R&D facilities and larger sport facilities are relatively less often available for the sampled firms.For both training events and networking events, the majority of firms have access to at least one relevant event per month.

Correlation covariates
In order to consider the heterogeneity among the preferences of the decision-makers, correlation tests are conducted among the binary covariates.In general, multicollinearity issues might arise when variables have significant and strong correlations (r) (Booth et al., 1994).Considering the small sample size (n = 69) a strict threshold of r ≤ 0.5 is used to select covariates for further analysis.Strongly correlated covariates are similar and to some degree redundant.The Pearson correlation between whether the firm is already located on an SP ('SP firm') and the access to facilities is above the threshold.Moreover, the correlations between 'SP firm' and a majority of the variables regarding access to facilities exceed the multicollinearity threshold.The following facility variables are excluded for further analysis: 'laboratories', 'clean rooms', 'piloting, 'conference rooms', 'dining facilities', 'sport centre' and 'sporting grounds'.
In a similar way the organisational covariates are also tested for multicollinearity (Appendix D). 'SP firm' is significantly correlated with 'firm size: less than 10 employees' r(69) = 0.354, p = 0.003 and negatively with 'more than 250 employees' r(69) = − 0.267, p = 0.027.This suggests that SP firms in the sample are relatively small in terms of number of employees compared to the respondents currently not located on an SP.Furthermore, for apparent reasons, some 'Firm size' and 'Firm age' variables are significantly correlated and exceed the multicollinearity threshold (i.e.'less than 10 employees', 'between 10 -50 employees' and 'age; 4 or less years' are excluded).The four phases within the new product development process are largely positively correlated.The significant correlations of 'concept development' with 'design and engineer' and 'prototype development and testing' are also too strong in terms of this criterion.In addition, the correlation between 'design and engineer' and 'prototype development & testing' exceeds the correlation threshold.
To summarise, the remaining covariates are not strongly mutually correlated and are used for further analysis.The considered binary covariates that will be used to test interaction effects on attribute parameters are; 'SP firm', 'meeting rooms', 'training events', 'networking events', 'location', 'firm size: less than 10 employees', 'between 50 and 250 employees', 'more than 250 employees' 'firm age: between 5 to 9 years', 'between 10 to 19 years', '20 years and older', 'high technological sectoral focus', 'prototype development and testing' and 'launch'.

Model estimation results
In total, the respondents were presented 1863 options and they made 621 choices.The selection frequencies of the 27 design profiles and the opt-out option are found in Appendix E. The distribution between design profiles and the opt-out option is respectively 70 % and 30 % suggesting that for the majority of the choice tasks the design profiles are relevant enough to prefer those over their current situation.With the Mixed Multinomial Logit model (MMNL) the final model is estimated and discussed here.In this section, the performance of the model and the main effects are discussed.The first level of each attribute that is significant in an initial MNL estimation enters the model as a random parameter in the final MMNL estimation.Looking at the goodness of fit statistics (Table 4) the model with a ρ 2 adjusted of 0.286 fits the choice data well as the ρ 2 adjusted value is higher than 0.2 (Louviere et al., 2000).
In the MMNL estimation results (Table 5), the 'constant' and main effects for the attributes 'networking and training' and 'cost plus 10%' are significantly random, which indicates that the utility for these attribute levels vary among decision-makers.The 'constant' parameter, which represents the base utility of selecting the opt-out option is positive and its utility varies among firms.As expected, among the decisionmakers there exist utility differences in terms of their current location.Moreover, the heterogeneity of the cost parameter indicates significant differences in the evaluation of costs.This underscores the many unobservable factors that influence the respondent's real-life cost (Geltner and Miller, 2001).The differences in utility found for the events attribute, 'networking and training', could be explained that for some firms these are less useful than for others.Training events are in general catered to and specifically useful for younger firms (Albahari et al., 2019).However, for the current study, no significant interaction effect of firm size or firm age was found (see Section 4.3.2).For networking events this could be related to the social personality of individuals which may vary between firms (Koçak and Can, 2014).
Furthermore, the parameters in bold indicate the attribute levels with the highest β-values which means these represent the highest utilities for respondents for each attribute.The parameters in italics indicate the utilities of the reference levels that were calculated for each attribute based on the estimated utilities of the other levels, as explained.The negative sign of the highly significant 'cost plus 10%' means that respondents do not prefer to pay more, which is not surprising.
In the next sections, the main effects, the willingness to pay and the impact of interactions effects on utilities are further discussed.

Main effects and willingness to pay
The MMNL results allow for the estimation of the relative importance value and the willingness to pay for all attributes.However, previous non-significant part-worth utilities should be interpreted as having no effect, i.e. its utility does not deviate significantly from zero.Therefore, the part-worth utilities are set to zero for the not-significant parameters of 'station location', 'university in the same city', 'networking and training' and 'cost same as current'.These part-worth utilities were already close to zero initially (see Table 5).
The relative importance of an attribute (for the choice of location) is indicated by the size of the utility range across the levels of the attribute; the larger the range the larger the impact of the attribute on the overall preference value for the choice of location.The total range of the utility of the cost attribute is the highest with 2.44, which is derived from the highest negative value and the highest positive value of the part-worth utilities.In the same manner, the size of the utility range of all attributes are derived from their respective part-worth utilities (Table 6).As expected, cost of the location is the most important aspect (Chan and  , 2005;Audretsch et al., 2005).Besides the cost attribute, the highest impact design-related attribute is 'university' followed by 'R&D facilities', 'location', 'shared facilities', 'firm diversity' and lastly 'events'.

Lau
The utility impacts for 'location', 'shared facilities' and 'firm diversity' are similar and therefore are somewhat equally important for tenant firms.
The wtp values should be interpreted as what percentage of the total current cost, firms are willing to pay to change from one level to another level.For instance, if firms are currently located in an area with a university in a different city then they are willing to pay 10.11 % more for an alternative with a university in the same area, while holding all else equal.
Looking at the design-related attributes, the relative highest partworth utility is represented by 'university in the same area', followed by 'in the same city' and lastly 'in a different city'.The degree of proximity of the university is in line with previous work of Ferguson and Olofsson (2004) and Dettwiler et al. (2006).This attribute might be attractive for financial reasons.It provides firms access to potential highly educated recent graduates for relative low cost (Audretsch and Lehmann, 2006).Moreover, possible unintended knowledge spillovers are generally less expensive than formal agreements (Chan et al., 2011).
Our results show that when firms are given the choice of none, shared or private use of R&D facilities the most preferred areas are those that allow for private use, which does not outweigh the possible advantages of the shared use of these facilities.Firms that currently do not have access to R&D facilities are willing to pay 7.74 % more for alternative locations that offer R&D facilities for private use.The second most important R&D attribute level is the shared use of these facilities among different firms.The face-to-face interaction between different organisations has been researched as means to build trust and knowledge (Storper and Venables, 2004;Ramírez-Alesón and Fernández-Olmos, 2018).Nonetheless, firms are also likely to be more focused on conducting their core activities and prefer secrecy within the R&D settings (Dettwiler et al., 2006).
For the relatively urbanised station location no significant partworth utility is found, while firms do prefer the 'suburban location' more, where both the station and the highway entrances are somewhat close by.American metropolitan areas are in general attractive for firms for their access to human talent and facility mix (Florida and Mellander, 2016).However, the polycentric nature of the Dutch context is reflected within our findings that firms prefer suburban locations more where the distances are relatively small (Burger and Meijers, 2012).Firms currently located at highway locations are willing to pay 6.08 % more to relocate to suburban locations.
For the shared facilities, all part-worth utilities are significantly different from zero.Technology-based firms prefer areas with only shared business support facilities the most, followed by shared business support and leisure facilities (e.g.cinema, sports facilities), and they least prefer areas where no shared facilities are provided.Firms with no access to shared facilities are willing to pay 5.99 % of their total current cost to upgrade to an alternative location with shared business facilities.The shared use of facilities and services has been suggested to provide for opportunities for cost saving and collaboration with others (e.g.Chan and Lau, 2005;Brinkø et al., 2014).Evidence from the Netherlands revealed that the possible benefits of meeting new people on SPs through shared resources and facilities are known to tenant firms, but at a high cost (Van Der Borgh et al., 2012).Our results show that while SPs can choose to provide more expensive shared facilities to tenants, they seem to prefer the shared use of R&D facilities and business support and leisure facilities to some extent, but the private use of R&D facilities is the most preferred option.Among the levels of the 'firm diversity' attribute, firms are more willing to pay for areas that at least host firms that are in the same technology domains as them.Followed by areas with a wide focus on all technology domains and lastly with a narrow focus.This suggests firms first and foremost prefer to exploit their current core activities, which could reduce risks (Renski, 2011).While collaborative opportunities with firms from different backgrounds comes second (Lamperti et al., 2017).As a collective of firms with different backgrounds such areas could therefore tap into a wide pool of knowledge and allow for exploration of new technological fields (Adner and Kapoor, 2010;Van Der Borgh et al., 2012).However, risks might arise from cognitive distance and the lack of absorptive capacity between firms from different fields (e.g.Boschma, 2005;Ubeda et al., 2019).The higher preference for so-called 'specialised' SPs are suggested to be beneficial for firm investments, while the relatively less desired 'general' SPs could be related to attaining sales goals (Liberati et al., 2016).Lastly, areas providing relevant networking events are most preferred, followed by networking and training events and least preferred are areas which provide no events.Firms that do not have access to any events are willing to pay 4.42 % more for alternative locations where networking events are held.The general purpose of these networking events is likely to share knowledge, seek out collaboration opportunities and to get closer to people from academia and industry (Ng et al., 2019b).Specifically, for smaller firms the dependence of networks is essential to gain access to market and technological knowledge for improving the firm's product offering (Van De Vrande et al., 2009).

Interaction effects
Besides the main effects, the interaction effects include the effects between two attributes and the effects between an attribute and a covariate and allow for a further investigation of the attributes and possible differences in preferences among decision-makers related to firm characteristics.The estimation results on this level are shown in Table 7.The utility of an alternative that is located in the suburbs with the university within the same area increases with 0.372 and 0.618 respectively through the main effects and the two-way interaction increases the overall utility with 0.345, which results in a total utility of 1.335.An alternative with a university close by (0.618) that focuses on a limited number of technology domains (− 0.461) receives an additional utility through its two-way interaction (0.216), but has a relatively lower total utility (0.373).
For the interaction effects between attributes and covariates, descriptive characteristics of the decision-maker such as 'SP firm', 'firm size', firm age', 'location', 'technological sectoral focus', 'prototype development & testing', 'launch' and 'access to meeting rooms' significantly increase or decrease the utility in conjunction with specific attribute parameters.Tenant firms already located on SPs prefer station locations and a university in the same area more than off-park counterparts.Furthermore, the SP location also leads to more preference to areas with shared business support and leisure facilities.Moreover, the Pearson correlation test (Section 4.2) showed that the 'SP firm' covariate is correlated with other variables (i.e.access to a wide range of facilities and that SPs especially attract smaller firms, while larger firms are significantly less present).These findings are in line with Dettwiler et al. (2006) that park tenant firms have access to more facilities and services than off-park counterparts and that they value these shared facilities and services.Their facility management framework is extended through outlining which other aspects are meaningful while considering other covariates.
Furthermore, smaller firms prefer suburban locations and areas with similar cost as their current situation relatively less.Both effects are likely to be related to financial motives.These new technology-based firms might want to target local markets and therefore locate further away from urban areas (Baptista and Mendonça, 2010).Moreover, smaller firms are more likely to find suitable space for their operations in less expensive locations.Furthermore, saving housing costs allows new technology-based firms to redistribute their funds to their core activities (Durão et al., 2005).The high costs of SP services experienced by new technology-based firms have been reported by Westhead and Batstone (1999) and Chan and Lau (2005).
For younger firms (between 5 and 9 years) a relatively lower preference is found for alternatives with a focus on a limited number of technology domains, which for the main effects is the least preferred level of this attribute.For relatively older firms (between 10 and 19 years) the total costs should stay the same as they prefer affordable options.Furthermore, firms already located in a station or suburban location do prefer suburban locations more in their decision-making process.
Firms who are focused on one or two technological sectors prefer areas which are closely located to a university less.This is especially relevant as the main effect of this university attribute parameter represents the highest part-worth utility.For these firms, the close geographical proximity seems less important.Potential obstacles of knowledge transfer between universities and firms could be unawareness, secrecy or a lack of commercial or academic interest.Moreover, the firm's research activities might fit commercial goals, but may not be of interest for academia (Woerter, 2012).Repeated interactions between a firm and a university are beneficial in accumulating absorptive capacity (Díez-Vial and Fernández-Olmos, 2015), but recent work of Ubeda et al. (2019) showed that too much absorptive capacity might reduce mutual learning.
Firms active in developing and testing prototypes are less sensitive to rent price (costs) in the choice of location.The iterative process of prototyping and testing of new innovations tends to be expensive and risky as not all efforts can be valorised.However, these uncertainties could be limited if simultaneous and consecutive prototyping is possible (Teece, 1986).These firms are likely to acknowledge the uncertainties of these types of activities and therefore expect that more expensive locations represent more quality that could aid their business.It should be noted that the current explanation is found within the mental perception of the decision-maker as the alternatives within the experiment are only different through the attributes and their levels.
Furthermore, firms active in launching innovations into the market prefer suburban locations less than those firms not active in launching innovations.This could be explained as that for launching commercially viable products, more space is required, in which the rent for areas closer to city centres are generally higher.Lastly, firms that currently have access to meeting rooms prefer alternatives that are relatively more expensive.It is noted that the meeting room variable is positively related to being located on an SP.The rents on SPs are generally higher than market prices, which might suggest that firms that have access to these facilities are used to the more expensive choices.

Discussion and implications
This paper's contribution to SP literature and innovation policy is based on an exploratory study of stated preferences of Dutch technologybased firms in the context of hypothetical location choice situations and in particular the different trade-offs that firms make based on whether they are located on or off a SP.The hypothetical location alternatives presented in the choice experiment possess typical SP attributes related to the location, proximity to university, firm diversity, facilities and services offered.Moreover, the discrete choice modelling adds to previous research through estimating the trade-off among all seven attributes used.
Although characteristics of real-life alternatives are largely interwoven with their location (i.e. transportation options, distance to the university and space for facilities and services), the current stated choice approach disentangles the separate utility effect of each attribute.Through the mixed multinomial logit model the heterogeneity is considered among preferences of technology-based firms.Preference differences are captured through the inclusion of organisational characteristics and the random parameters in the model.The use of an offpark group reduces the selection bias of the SP tenant firms as they already have made the choice to locate on these parks (Siegel et al., 2003;Vásquez-Urriago et al., 2016).

Implications for theory
Our results fill knowledge gaps in relatively novel research directions within the SP literature; the study of perceptions or stated preferences of firms (Albahari et al., 2019;Lecluyse et al., 2019) and the development of SPs (Mora-Valentín et al., 2018).By focusing on perceptual measures, we address the research call of Lecluyse et al., 2019 (page 575) who argue that: "examining perceptual measures, such as perceived value and satisfaction, will complement our current knowledge on SP contribution and will largely enrich our insights into the actual benefits provided by SPs".In addition, the current literature is relatively silent on how tenant firms choose between alternative locations.Notwithstanding the important insights of existing studies on SP location (e.g.Westhead and Batstone, 1998;Lindelöf and Löfsten, 2003;Wright et al., 2008;Van Der Borgh et al., 2012) location choice preference has not received the necessary attention in the extant SP literature.
Our study of 69 technology-based firms in the Netherlands of which half are located on SPs and half are not, reveals that costs remain the prime consideration in location choice, while considering the trade-off among attributes.The university presence is the second-most important consideration, followed by R&D facilities, location type, shared facilities, firm diversity and, lastly, events.Several of these results confirm the current state of knowledge identified in the literature.First of all, we confirm the finding of Westhead and Batstone (1999) that the costs of premises influence location decisions made by technology-based firms.Second, we find a preference for being located in close proximity to the university, which is in line with previous work of Westhead andBatstone (1998, 1999), Lindelöf andLöfsten (2003), andVan Der Borgh et al. (2012).However, our study reveals some important preference differences based on a diverse set of organisational characteristics.We find that tenant firms already located on SPs prefer a university in the same area more than off-park counterparts.This preference of existing park tenants towards a university is in line with the work of Díez-Vial and Fernández-Olmos (2015).We also see that firms who are focused on one or two technological sectors prefer areas which are closely located to a university less.When considering other proximity dimensions, this lower utility could be explained as that these specialised firms have different processes (institutional distance) or are focused on select domains, while universities often conduct research in a wide array of domains (cognitive distance) (Boschma, 2005).Third, we see that areas providing relevant networking events are most preferred, followed by networking and training events and least preferred are areas which provide no events.This is in line with the finding of Wright et al. (2008) who state that firms base their location choice on the fact that they need access to (university) networks and knowledge.Moreover, from the perspective of the entrepreneur, these types of social events could develop their personal and business relationships with other entrepreneurs, clients, suppliers and the SP management (Xia et al., 2020).
Besides confirming the current state of knowledge our study also extends previous findings.In particular the work of Van Der Borgh et al. (2012), who show that the availability of shared facilities is one of the most important motives for locating at an SP.We specify their findings further by focussing on a diverse set of shared facilities.In addition, we extend the work of Wright et al. (2008) who find a difference in university and non-university SP location choice by small and large firms.Where small firms tend to locate in a university SP location, large firms tend to locate more outside university parks.We find that SPs in general especially attract smaller firms, while larger firms are significantly less present.
We not only confirm and extend previous work, but our findings also highlight some important new insights by including the fact that firms have a specific preference in terms of several additional location aspects.Among the levels of the 'firm diversity' attribute, firms are more willing to pay for areas that at least host firms that are in the same technology domains as them.Followed by areas with a wide focus on all technology domains and lastly with a narrow focus.This answers to some degree why technology-based firms would choose to co-locate (Stuart and Sorenson, 2003).Although for younger firms (between 5 and 9 years) a relatively lower preference is found for alternatives with a focus on a limited number of technology domains.These firms prefer alternatives focused on a large range of technology domains (including their own) in order to explore new technological fields where more opportunities are available (Almeida and Kogut, 1997).

Implications for practice
So, with the help of this study we are able to analyse how on-and offpark technology firms value specific SP design characteristics, how they value different types of value park configurations, and whether these values are held uniformly or not.Although the findings refer to aspects that are sometimes difficult to change in real-life (i.e.location type or proximity of university), important strategic choices remain concerning how to operate these SPs.The results are especially relevant for policymakers as this group plays an important role in allocating public resources towards new SP development.We know that high-tech resources are finite, which makes regional planning especially appropriate in order to avoid market failures (Taylor, 1983).This line of research into SP characteristics contributes to evaluating the policy effectiveness of SPs as the most important trade-offs among different park configurations are made clear.Existing SP literature mainly focuses on the revealed preference in order to prove policy effectiveness with so-far mixed evidence (Albahari et al., 2010).In these studies, limited attention is given to the SPs setups, while these parks can be very different in terms of characteristics (Etzkowitz and Zhou, 2017;Ng et al., 2019a).In particular, our study contributes to the set-up and development of SPs by offering a better insight into the location choice preferences of potential residents.Whereas multiple studies analyse the role of resident selection by SP management (e.g.Link and Link, 2003;Chen et al., 2006;Salvador, 2011), this only explains one side of the park location choice.The potential resident itself also has a very important say in the final location choice, which can be viewed as an important strategic decision (Wright et al., 2008;Gwebu et al., 2019).To some degree our research results are generalisable for other medium-sized SPs in the Netherlands when they reach a similar size as the ten largest SPs used in the current study.In addition, policy-makers in countries with similar national innovation systems as the Netherlands can use the trade-off results to prioritise certain attributes for their SP configuration of greenfield developments.
Next, our results also provide managers of existing SPs with strategic insights in technology focus and the provision of facilities and services.Considering that only a small fraction of the sampled firms is currently located at SPs near station locations, more attention should be given to this unfulfilled need for centrally located knowledge-based areas with adequate means of public transportation.When emphasising R&D facilities, our results demonstrate that the most preferred areas are those that allow for private use.Moreover, technology-based firms prefer areas with only shared business support facilities the most, followed by shared business support and leisure facilities (e.g.cinema, sports facilities), and they least prefer areas where no shared facilities are provided.Our study also demonstrates that tenant firms already located on SPs prefer areas with shared business support and leisure facilities more than off-park firms.For SP management also a crucial role is required for facilitating the interaction between firms from different technology fields in order to avoid cognitive distance and enable mutual learning.Moreover, as the utility differences suggest, the organisation of networking and training events should require attention that caters to specific people and firms (Koçak and Can, 2014;Albahari et al., 2019).For universities our results show that technology-based firms value their presence.In contrast, technology-based firms currently not located on SPs prefer the proximity of a university relatively less.Therefore, for universities that want to attract new firms to their campus, more effort is required to convey their added value towards new tenants.For practitioners, this study allows for benchmarking new and current SPs and knowledge-based area developments through the utility estimation of the considered attribute levels.Finally, our study provides firms with an intention to move a general overview of location preferences of peers as relocating operations are often major business decisions, which affect their workforce and physical assets.
Although the precise conditions that enable SPs to be successful remain unknown (Yang et al., 2009), this study reveals the discrete conditions of SPs that fit the needs of its users.It enhances SP managers' understanding of how firms choose between SP locations and what needs to be provided to attract firms to their SP.This might lead to a more effective and efficient resident selection strategy by SP management.Closing the gap between the tenants' demand and SP configuration is beneficial for both SP managers and tenant firms for achieving both parties' goals, increase performance and possibly attract new tenant firms (Albahari et al., 2019).

Study limitations
In addition to implications to theory and practice, this study is not without limitations.First and foremost, the small sample size did not allow for segmenting groups based on preferences (i.e.latent class analysis).However, the SP sample group did show some similarities with the total SP population in the Netherlands at the time of the survey.Moreover, a considerable share of the responding firms within the SP sample group was relatively younger and smaller than the off-park sample group, which could suggest selection bias.This could be related to the fact that more mature and financially viable firms are more often located outside of SPs.In order to reduce selection bias, we did control for these organisational variables through the significant interaction variables (e.g.firm age and firm size).By adopting a novel approach for the SP context, the aim of this study was mainly exploratory.It should be noted that the estimated preference values refer to averages across firms.The method of discrete choice analysis does not offer insight in how preference values are distributed, e.g., a few firms displaying a very strong preference could result in the same average value as a large number of firms assigning only moderate preferences to a certain attribute.Furthermore, due to the small Dutch sample of technology-based firms the results are not representative for the population of technology-based firms in the Netherlands.A limitation of our study is that we have not included an item of 'offering both a private and a shared R&D facility'.Consequently, the value of a shared R&D facility is now only assessed as a substitute to private use of R&D facilities and not as a complement.Moreover, the use of hypothetical SP alternatives with a restricted number of attributes comes with some drawbacks.The unnamed alternatives are not tied to existing places and the image or brand of SPs are not considered.Especially for younger firms, image benefits are important in order to enhance their legitimacy (Ferguson and Olofsson, 2004;Ng et al., 2019b).

Implications for future research
Our study opens additional venues for future research on the demand-side of SPs.Essential issues of supply and demand of real estate are the inelasticity of the former and the dynamic nature of the latter.In this sense, Díez-Vial and Fernández-Olmos (2017) argued that the value of SPs changes over time, which asks for longitudinal research into the dynamic needs of its users.Other researchers are encouraged to continue investigating the demand of technology-based firms while considering contextual factors such as duration of stay in their current location.Specifically, additional research into the added value of R&D facilities for shared usage is greatly encouraged in order to reveal the tenant's demand within an innovative environment.Future choice experiment studies should at least include attribute levels which cover both private R&D facilities and private and shared R&D facilities in order to assess the explicit value of the shared usage of R&D facilities.Moreover, recently, SP management size has been found to be positively related to the tenant firm's innovation performance (Albahari et al., 2018).Future research could delve into the impact of the SP management and include labelled alternatives related to real-life places.However, for labelled choice experiments, more design profiles are required and real-life alternatives might lead to many unobservable factors that impact the firm's decision-making.Finally, future research might analyse the interplay between initial location choice preferences and preferences at work during the residential period.Initial motives for entering the SP may be reinforced or may break down (i.e.no longer creating value), or even new motives may develop along the way.

Declaration of competing interest
None.

Table 1
Attributes and levels SP choice experiment.
1 High-tech focused SPs include firms active in a wide variety of sectors such as; aerospace, computer science, electronics, new materials, photonics and robotics.W.K.B.Ng et al.

Table 2
Covariates with levels choice experiment.

Table 3
Characteristics of 69 technology-based firms in the Netherlands.
a Executive, Technical, Operations, Commercial.bRespondents were allowed multiple options.2Based on data from the Netherlands Chamber of Commerce, the total population consisted of 1277 SP firms in 2019 among the ten SPs used in this study.The sub sample of 37 SP firms does not differ significantly with the total population on location type (χ 2 (3) = 2.34, p = 0.50) and firm age (χ 2 (3) = 2.52, p = 0.47).

Table 4
Goodness of fit statistics MMNL model.

Table 6
Utility and willingness to pay SP attributes.