Investigating value added from heritage assets: An analysis of landmark historical sites in Wales

We reveal how tourist visitation to similar historical sites supports different levels of local gross value added (GVA). The paper shows how information on tourism activity at few historical sites can be used to analyse causal recipes defining whether sites support relatively high/low levels of GVA. Fuzzy ‐ set qualitative comparative analysis is employed to offer perspectives not possible with other analytical methods. The study reveals that for a set of similar heritage sites, that factors supporting local economic impacts are complex and with this having ramifications for management interventions around sites that seek to boost the economic impacts of visitation.


| INTRODUCTION
This paper provides an analysis of how tourist visitation to a group of heritage assets (historical sites and castles) works to support local gross value added (GVA). The analysis reveals that visitation to similar historical sites actually supports very different levels of economic activity. While the paper works to contribute to the evidence base on the economic impacts of visitation to heritage assets, it also seeks to investigate the reasons why tourism to the different historical sites supports different levels of GVA. The paper also illustrates how even information on tourism spending and activity at relatively few historical sites can be used to systematically analyse the causal recipes that define whether a site will support relatively high or low levels of GVA. In the case the small number of sites examined made traditional econometric and cluster type analysis impractical in furthering the analysis. However, the paper reveals how fuzzy-set qualitative comparative analysis (fsQCA) can be employed to offer analytical perspectives that would not have been possible with other analytical methods.
The fsQCA method is used in the paper to show that the causal conditions defining whether sites might be associated with relatively high-GVA or low-GVA supported per visitor are actually complex.
The method provides a valuable means of understanding the different configurations of activity that can be associated with both high and low levels of GVA per visitor at the case sites. Fundamentally, there are different routes to how heritage sites support tourism value added. The practical implications of the findings for management of heritage sites are investigated in the conclusions to the paper. The paper then seeks to contribute to the research literature on the economic value of heritage tourism in two ways: first, by showing that the factors supporting the local economic effects of tourism at similar heritage sites are complex and vary by site; second, by demonstrating how an economic impact assessment based on input-output methods can be improved upon to gain a structured understanding of the causal recipes leading to impact, and with this offering valuable perspectives for management through the use of the fsQCA method.
The remainder of the paper is structured as follows. Section 2 provides background to the paper in terms of the connections between heritage assets and regional economic development. Section 3 introduces the case material on Wales and examines how visitation to heritage sites in rural parts of Wales supports regional economic activity. The section shows that similar heritage assets support very different levels of economic activity. In Section 4, we introduce the fsQCA method through which the historical heritage assets are grouped in terms of causal recipes that describe their ability to induce regional economic activity from visitor spending (here in terms of high-GVA and low-GVA supported per visitor) and the results from employing the method. Section 5 contains discussion of the findings and concludes.

| BACKGROUND
There has been a great deal of research to explore the economic effects associated with different types of tourist, and how tourism spending supports (or sometimes does not support) local economic activity. In terms of analytical means of understanding the wider economic effects associated with tourism spending, the framework of tourism satellite accounts (TSA) has been particularly useful, and with input-output frameworks also being commonly used to reveal the indirect and induced effects connected with different types of tourism spending (Frechtling, 2010;UNWTO, 2008).
The analytical input-output framework has been widely used in the study of visitation specifically linked to heritage sites and to cultural events. For example, in a recent paper, Parga-Dans and Alonso-Gonsalez (2018) examined the direct and indirect economic effects of reopening a cave at the Altamira World Heritage site and employed an input-output method to estimate the value of activity supported in Cantabria. Bryan, Munday, and Bevins (2012) also used input-output methods to investigate the economic activity supported by visitation to national museums sites in Wales. Çela, Lankford, and Knowles-Lankford (2009) also employed an approach based on input-output frameworks to examine visitation to a national heritage area in the United States. Studies grounded in input-output frameworks can have a role in revealing, to different types of stakeholders, the value of heritage tourism, to gain buy-in for new developments, and as one means of providing evidence of benefits to set against costs, particularly where heritage sites are under different types of pressures. Çela et al. (2009) make the point that the more discretionary characteristics of spending at heritage sites makes it important for tourism planners to understand visitor expenditure patterns.
The analysis of the role of heritage tourism in more peripheral areas might be particularly relevant. Researchers have argued that the analyses of the consumption of heritage as a means of economic regeneration have perhaps been too focused on urban areas. In these cases, heritage tourism might be seen as a means of bringing spending power to cities and looked on positively by inward investors (see Aas & Ladkin, 2005;Chang, Milne, Fallon, & Pohlmann, 1996;MacDonald & Jolliffe, 2003;Richards, 1996). However, these types of effects are also important for more peripheral areas. Heritage tourism has been viewed as a means of reversing the problems of older industrial areas caused by declines in heavy resource intensive industry and manufacturing and as a means of providing new economic opportunity in areas of high unemployment. The ways through which tourism expenditure supports local economic activity may be particularly important in more rural and peripheral areas, or areas where structural change has left a legacy of old industrial assets. Pérez-Alvarez, Torres-Ortega, Díaz-Simal, Husillos-Rodríguez, and De Luis-Ruiz (2016) show, for example, the important role of heritage tourism in boosting economic prospects in old mining areas. Moreover Fonseca and Ramos (2012) draw attention to how heritage tourism can assist in addressing economic issues in more peripheral areas where there are few alternative avenues in improving economic prospects. While heritage tourism may provide a developmental route for periphery areas, positive economic impacts levered from increased, or more targeted, visitation to heritage sites cannot be guaranteed but have to be planned and managed. For example, Urry (1990) showed, in the United Kingdom case, that the growth of the heritage industry has increased tourism (particularly international visits). However, it is only selected sites which fully benefit due to the importance of bespoke factors, including the quality of the attractions, transportation infrastructure, accessibility of media information, and competition from adjacent similar attractions. Urry (1990) however suggests that even in the most disadvantaged, peripheral, or rural locations, heritage can potentially be mobilized to gain competitive advantage in a "race" between places.
Then, the visitation linked to heritage tourism could be important in economic development processes, and with techniques such as input-output analysis one means of establishing the levels of local GVA that might be dependent on tourism spending. However, while visitor expenditure surveys accompanied by input-output economic modelling can show up differences between heritage sites in their ability to support local value added, it may not provide enough information to planners about the causal recipes at different sites. In particular are the characteristics leading to relatively high value added per heritage site visitor the same across sites, or do these factors differ across sites? Although econometric techniques and cluster analysis might provide some insights into the common determinants of the higher value added supported per visitor at a range of sites, such techniques might not be best placed to show planners the different routes that lead to similar outcomes. Moreover, econometric and cluster analysis techniques may be less than useful where the number of site observations is limited.
In the case that follows these problems were very evident. The case that follows reveals a series of historical sites where it was possible to use visitor spending surveys and input-output methods to explore how the sites differed in their ability to support high levels of value added per visitor. However, a further layer of analysis was required to explore whether there were different routes to high value added per visitor. In this process, the fsQCA analysis was found to provide insights not offered by simple cluster or econometric analysis. itor sites, however, presented distinct challenges: Many of the economic impacts of visitation do not occur at the site, but more widely throughout the regional economy, as visitors spend money on accommodation and other services away from the heritage site in question.
Meanwhile, heritage sites themselves will have impacts away from their immediate location through their purchases of goods and labour.
The assessment of the economic impacts of spending by visitors to identified heritage sites was assisted by visitor spending surveys collected at sites (see Table 1 for the breakdown of surveys undertaken to support the research). The sites comprised castles, one bishop's palace, two Iron Age fortress sites, and a UNESCO recognized industrial heritage site.
The funding secured under the Environment for Growth umbrella provided an opportunity to help individual site managers address some of the difficult economic measurement issues they face. At the same time, it provided an opportunity to ensure that impacts were reported consistently, and hence comparably, across the supported projects through the use of a shared suite of questionnaires. The visitor spending survey was designed to record expenditure, by item, for respondents-with these data later being aggregated into categories by the research team (e.g., food and drink, transport, accommodation, souvenirs, high street shopping, recreation, and entertainment). The visitor spending survey was also designed to capture data on non-spend trip related items including the purpose of visit to the heritage site (leisure trip from home, leisure trip as part of a longer break, business purposes, etc.); make-up of party (by age), whether the trip involved an overnight stay; mode of transport to the site; regularity of visit to the site (first-timer, repeat visitor); and site satisfaction (enjoyment of visit, rating of facilities and staff). Visitors who had spent the previous night in Wales away from home were asked supplementary questions on their accommodation and spend.
The direct economic impact of this visitor expenditure in the region occurred, for example, as visitors purchased food and drink, paid for parking, and met accommodation costs. However, an estimate of direct effects only covers part of total impact. There was also a need to consider how the visitor spending supported regional economic activity (in Wales) indirectly. Expenditure by visitors requires outputs from other Welsh industries. For example, visitors stay in local B&Bs/guesthouses, and purchases are made by these accommodation providers from local farms or wholesalers to provide their services.
This regional sourcing then in turn leads to further regional spending by the local farms and so on. The extent of these supplier effects, then, depends on the level of regional sourcing for the particular sector and on the levels of regional sourcing by its suppliers. Additionally, visitor spending adds to local incomes, the large part of which will likely be spent in the region. These induced-income effects can be added to supplier effects to estimate the total indirect consequences of the direct local economic activities (for an explanation of these indirect and induced effects associated with Welsh tourism spend, see, e.g., Jones, Munday, & Roberts, 2003).    Wales. The table also reveals major differences in the proportion of visitors that reported that they were staying away from home during the visit, with this being one factor expected to increase levels of regional spending associated with the trip.   Table 2 may be important for regional authorities investing in heritage assets revealing how site specific visitation links through to regional GVA and employment opportunities.

| Economic impact of visitation
As well as providing indicators in terms of the site visits required to create a Welsh full-time equivalent job, the  Historic character and the renown of sites (rated simply Low 1 to High 6 and based on a combination of identified site renown factors including uniqueness of attraction, key historical events at site, and whether having UN Heritage site status) have been found to be an important factor driving site visits (Kerstetter, Confer, & Graefe, 2001). Moreover, research has revealed that people with an interest in visiting heritage or cultural sites (i.e., "heritage tourists") tend to stay longer, spend more, are more highly educated, and have a higher average annual income than the general traveller (see Travel Industry Association, 1997). It is accepted that this is not a universal conclusion in relation to the value of heritage tourists to the local economy (see Staiff, Bushell, & Watson, 2013) and that there may be important differences between "heritage tourists" and "tourists at heritage sites." Notwithstanding this, indicators of high renown, such as attaining World Heritage Status, have been shown to alter the visitor profile of a site and increase the prospects of drawing higher spending individuals interested in culture.
Within   In what follows, we show how selected of the identified conditions can be used to examine causal recipes leading to high-GVA or low-GVA per visitor outcomes at the identified heritage sites. The notion of high-GVA and low-GVA suggested here is a feature of the employment of the specific method (fsQCA) and its set-theoretic approach to analysis, in that the method allows openness to possible asymmetrical relationships between causal conditions and outcomes (Fiss, Sharapov, & Cronqvist, 2013). Then, the presence and the absence of the outcome, respectively, may require different explanations (see Berg-Schlosser, De Meur, Rihoux, & Ragin, 2009). Therefore, in the next section, analysis is undertaken with regard to each of high-GVA and low-GVA outcomes separately.

| Fuzzy-set qualitative comparative analysis
FsQCA is used to examine selected data from Table 3 to provide a more systematic analysis of the causal recipes linked to the regional GVA supported by the visitation to the historic heritage sites. This method is evolving as an important tool in tourism economic analysis in the analysis of causal configurations (see recently, e.g., Pappas, 2017;Olya & Gavilyan, 2016).
FsQCA offers a set-theoretic approach to causality analysis, in respect of conditions and an outcome (Ragin, Epstein, Duerr, & Kenworthy, 2008). 2 The method is of potential value here because of its flexibility to deal with relatively small data sets (here 14 heritage sites), and the asymmetric nature of analysis, whereby the limits of an outcome variable are considered in separate analyses (as described in FsQCA has particular value here as distinct from regression analysis (see Vis, 2012, for comparative discussions). For example, while a simple regression explores the effects of changes in independent variables on a dependent variable, the fsQCA method allows the investigator to examine the combined conditions that lead to an outcome (such as higher regional GVA levels supported per visitor). Moreover, the fsQCA method is better suited to problems where there are relatively small numbers of observations and allows the analyst to explore whether there are different combinations of factors that lead to a given outcome.
To enable the fsQCA analysis of the heritage site data (seeTable 3), the condition and outcome variable values need to be transformed (calibrated) into representing memberships of relevant sets, using their continuous scale values, in the form of fuzzy membership scores (see Ragin, 2008). This is allowing for the fact that a heritage site's membership of a set might be by degree, rather than strictly "in" or "out" of the set. Each variable has a respective fuzzy membership score within a 0.0-1.0 domain, that is, with the limits representing 0.0 (full exclusion "non-membership" from a set) and 1.0 (full inclusion "membership").
Clearly, calibration is a key issue here for the continuous variables described in Table 3, and we adopt a popular transformation process, the direct method (see Ragin, 2008). This requires three threshold qualitative anchors for full membership (upper threshold), full nonmembership (lower threshold), and the crossover point. These are then used within log-odds formulae to create the necessary membership scores. Evaluation of the three qualitative anchors here follows the approach presented in Andrews, Beynon, and McDermott (2016) and Beynon, Jones, and Pickernell (2016). This approach involves the identification of the 5th percentile (lower threshold), 95th percentile (upper threshold), and 50th percentile (cross-over point) values, based on a constructed probability density function graph for each variable.
Based on the qualitative anchors found for each continuous con-

| RESULTS
In what follows, the fsQCA based findings over the considered four condition variables (adult visitors, first time visitors, other attractions, and renown) are outlined. 3 The first stage of results using fsQCA is elucidation of the associated fuzzy set data, undertaken through a truth table (see Table 4). The truth table reveals all the different combinations of condition attributes that are connected to the outcome, that is, separately considering high-GVA or low-GVA per visitor-as described previously. This is used to synthesize the results of fuzzyset analyses of the logically possible configurations of a given set of causal conditions (see Ragin, 2008). Table 4 represent configurations based on the considered four condition variables, through considering case strong membership to a configuration (membership score values below and above 0.5 are assigned the values 0 and 1, respectively). Recall that there are four condition variables in Table 3 which means there are 16 (2 4 = 16) possible configurations.   Table 4), it can be viewed as the proportion of memberships, in fuzzy terms, in the outcome explained by each logical configuration. The heritage site column gives the names of the built heritage sites associated with a configuration. The last two consistency columns in Table 4 show the respective raw consistency values associating a configuration with GVA (high-GVA and low-GVA, respectively).

Rows in
With respect to the separately considered outcomes, high-GVA and low-GVA, the identification of configurations considered to be associated with either of them (or not) are defined by consistency threshold values. Choice of consistency threshold for the raw consistency measure influences the strength of evidence used in subsequent analysis (see Ragin, 2006), and with configurations above or below the threshold cut-off value designated fuzzy subsets of the outcome or not fuzzy subsets, respectively. The causal combinations that are fuzzy subsets of the outcome delineate the kinds of cases in which the outcome is found.
With reference to the consistency value results in Table 4, a consistency threshold of 0.92 was employed for both when considering high-GVA and low-GVA (using the same threshold value for both high-GVA and low-GVA outcomes, following Andrews et al., 2016). In Table 4, the bold values in the consistency columns show those configurations with such consistency values above the identified consistency threshold.
Moving beyond the truth table elucidation of considered configurations of condition attributes, the fsQCA method moves to associated necessity and sufficiency findings for the high-GVA and low-GVA outcomes. In fsQCA terms, these are described as necessity, if a condition must be present for the outcome to occur (analysis of necessity); and sufficiency, if a given condition or combination of conditions can produce this result. In undertaking fsQCA, Ragin (2008) suggests both necessity and sufficiency should be investigated.
In terms of necessity, in regard to each of high-GVA and low-GVA, no individual condition attributes (including their presence or absence) were identified that had an associated necessity based consistency value (in singular variable terms) above the considered threshold of 0.9 (see Young & Park, 2013). Hence, no condition attributes were considered necessary for the respective outcome to be materialized. Table 5 reports the findings in regard to sufficiency. Table 5 uses a notation prescribed in Ragin and Fiss (2008  Remainders are those configurations neither considered as a fuzzy subset nor not fuzzy subset of an outcome. Moreover, following Rihoux and Ragin (2009), the complex solution is defined as a "minimal formula derived without the aid of any logical remainders" (p. 181) (so not considering Configurations 1, 2, 3, 8, 10, and 11 and with these struck through in Table 4), and the parsimonious solution as a "minimal formula derived with the aid of logical remainders" (p. 183) (so considering Configurations 1, 2, 3, 8, 10, and 11 struck through in Table 4 if enable a more parsimonious solution).
Also shown in Table 5 are further measures, namely, Unique consistency-the degree to which cases sharing a given causal combination (configuration) agree in displaying the relative outcome; Raw coverage-the overall coverage of a causal combination that may overlap with other combinations; and Unique coverage-coverage uniquely due to a causal combination (see Ragin, 2008, for their technical details).
To accompany subsequent discussion of these fsQCA results visualizations of groupings of the considered 14 heritage site are presented in Figure 1.
The established causal recipes in Table 5 Configurations (see Table 4

| DISCUSSION AND CONCLUSIONS
The paper adds to an evidence base exploring the local economic impacts of tourism supported by visitation to heritage assets. The paper also reveals how an impact analysis grounded in an input-output framework can be supplemented by analysis within an fsQCA frame to derive insights into the different routes through which similar heritage sites work to support relatively high levels of GVA. The method employed in the latter analysis is particularly useful where a smaller number of site-level observations may restrict the use of other analytical techniques.
The paper shows selected heritage sites in Wales vary in their ability to support regional GVA. The underlying core conditions defining whether sites might be associated with relatively low or high-GVA supported per visitor are represented by complex configurations. The analysis suggests that over generalizing on the conditions supporting higher levels of regional GVA linked to heritage visitation is inappropriate.
A series of practical implications result from the research. In general, heritage tourism has been identified as a major growth area in Europe and elsewhere (Poria, Butler, & Airey, 2003;Richards, 1996), and the promotion of heritage tourism has formed part of many EU regional development programmes with museums, monuments, and other heritage attractions becoming a focus of regional economic development strategies (see Janiskee, 1996). In this context, making better connections between heritage assets and local economic returns in an era of competing demands from heritage organizations for public funds is important (Bryan et al., 2012). For example, the United Kingdom Culture, Media, and Sport Committee (2011) showed the consequences of public funding cuts for heritage at risk and that "Reductions in public funding alongside restrictions on credit, falling investment returns and the failure of development companies will make it much harder to find viable solutions for our heritage at risk" (p. 141). There is then a need to better understand the factors which contribute to a site's ability to attract relatively high or low levels of GVA levered by visitors.
The research in this paper would suggest that if increasing the impact of visitation is an aim of marketing and planning around heritage, then there is a need to understand that the demand and supply side routes to relatively high-GVA per visitor are complex.
Then, for some sites, causal routes to relatively high-GVA supported per visitor may require different marketing plans and need to jointly focus on elements of the demand and the supply side around heritage assets. Identified here as important conditions in supporting high value added tourism is the higher proportion of adults in total visitors, but the analysis reveals that factors such as the number of first time visitors, the local supply side, and variables describing renown do not always feature in the same way in the high value added configurations.
We accept in the research here that the condition variables require far more exploration. For example, further analysis of visitor demographics at the heritage sites would be useful. However, the analysis reveals that one discrete configuration is inadequate to explain high or low value added per visitor supported at sites, and moreover, in only a few cases are the low/high value added solutions mirror images of one another.
where X i is the degree of membership of an individual i in set X, and Y i is its degree of membership in set Y. Table A1 shows an example data set, for which the consistency measure will be elucidated on (adapted from Ragin, 2008).