The Recreational Value of the Baltic Sea Coast: A Spatially Explicit Site Choice Model Accounting for Environmental Conditions

Abstract

The coast plays a significant recreational role in the nine countries around the Baltic Sea. More than 70% of the population of these countries visit the coast, representing some 80 million recreational visits annually. Understanding the values associated with coastal recreation, and the potential welfare changes resulting from improvements in the state of environmental and infrastructure conditions of the Baltic Sea coast is important for marine environment management in the region. We estimate a spatially explicit travel cost model for Baltic coast recreation to assess the welfare of accessing individual sites, identify recreational hot spots and simulate the welfare changes resulting from improvements in environmental and infrastructure conditions. The total benefits associated with Baltic Sea coast-based recreation amount to 27.5 billion EUR per year with significant variation across sites. Improving water quality and infrastructure boost the recreational value by nearly 6.2 billion EUR, an increase of about a fifth of the existing recreational benefits.

1 Introduction

Recreation in coastal areas has been growing over the past decades, and it has become a key factor in the economic development and social welfare of countries with coastal tourism (Ghermandi and Nunes 2013). At the same time, many coastal recreation opportunities are sensitive to environmental quality (Ahtiainen et al. 2013a, 2013b; Hynes et al. 2018). As a result, studies on the market and non-market benefits of coastal areas play a crucial role in the effective coastal management. This motivated multiple studies that evaluate the recreational benefits of various coastal areas (Ghermandi and Nunes 2013; Voke et al. 2013; Zhang et al. 2015), which highlight the importance of preserving the natural environment. In this study, we estimate the recreational value of the coastal area of the Baltic Sea—one of the largest semi-enclosed bodies of saltwater in the world. Specifically, we focus on identifying the effect of the improvement of environmental quality of the recreational sites.

The Baltic Sea is encompassed by nine European countries, all of which benefit from the recreational opportunities it provides. The value of Baltic coast recreation has been the focus of only a few economic studies to date. Czajkowski et al. (2015) investigated recreation patterns in the countries around the Baltic Sea and applied the single site Travel Cost Method (TCM) to estimate the total economic benefits provided by Baltic coast recreation. This gave a figure of 14.8 billion EUR per year, with the potential of an additional 2 billion EUR, if the environmental status of the sea was improved. Other TCM estimates of coastal sites were provided by Vesterinen et al. (2010) for Finland, and by Sandström (1996) and Soutukorva (2005) for Sweden. In addition, the benefits of improved water quality for recreation have been estimated using stated preference methods. Ahtiainen et al. (2014) estimated the value of alleviating eutrophication in the Baltic Sea as 3.6 billion EUR annually. The economic value of reductions in eutrophication had earlier been measured for Sweden's Stockholm archipelago (Söderqvist and Scharin 2000), and for Lithuania, Poland, and Sweden (Markowska and Żylicz 1999). Tuhkanen et al. (2016) estimated the value of benefits from water quality improvements in Estonia, while Pakalniete et al. (2017) investigated willingness to pay for improvements in the quality of coastal water at recreational sites in Latvia.

Our study contributes to the literature by conducting a spatially-explicit analysis of the economic value of Baltic Sea-based recreation across eight countries around the Baltic Sea. Unlike earlier studies (e.g., Czajkowski et al. 2015), we do not treat the Baltic coast as a single site, but instead model the demand for a set of coastal locations used for recreation in each country. Single-site TCM has been criticized for disregarding substitution possibilities (Fletcher et al. 1990), so instead we implement the random utility model-based TCM (Parsons 2017) which performs substantially better when estimating environmental quality changes (Phaneuf and Smith 2005). Specifically, we employ Hausman et al. (1995) two-stage budgeting model. This approach allows substitution among recreational sites to be explicitly accounted for, and thus makes it possible to accommodate both access and quality changes into the valuation, including the environmental qualities (Zandersen et al. 2007; Termansen et al. 2013) and spatial variability (Bateman et al. 2013; Czajkowski et al. 2017). More generally, our study addresses the policy demand for valuation of recreation and tourism created by the Marine Strategy Framework Directive (2008/56/EC) and the Maritime Spatial Planning Directive (2014/89/EU) ongoing in the European countries. Therefore, we investigate the importance of environmental quality (measured by the Blue Flag status of a given site) for the choice of recreational sites, and simulate the welfare changes resulting from its enhancement, which is a methodologically stronger and more precise approach than in earlier, perception-based estimates (see Czajkowski et al. 2015).

The remainder of the paper is structured as follows. In Sect. 2, we describe our empirical approach—the survey data and econometric framework of the analysis. In Sect. 3, the results of the estimation of the two-stage budgeting model are presented. Section 4 contains our interpretation of these results, in terms of implied welfare measures of Baltic Sea-based recreation, a description of the distribution of recreational value along the Baltic coast, and the simulated change in recreational value resulting from improvements of environmental conditions. The last section presents conclusions.

2 Empirical Approach

2.1 Survey Data

The data used for this study comes from a survey conducted in all of the nine countries¹ around the Baltic Sea between April and June in 2010² (Ahtiainen et al. 2013a, 2013b). In each country approximately 1000 respondents participated in the survey, resulting in a total sample of 9127 observations. The survey consisted of five sections: (i) an introduction, including a short description of the Baltic Sea; (ii) questions about respondents’ relation to the Baltic Sea (e.g., paid work), whether they spent time on leisure or recreational activities near the Baltic Sea, and their place of residence; (iii) details of their most recent visit to the sea, including name of the site that they visited, questions about mean of transport, time spent on getting to the site and the distance; (iv) attitudinal questions; and (v) socio-demographic questions.³ The main survey instrument was created in English and later translated by study partners to the national languages. Specific attention was paid to ensure that the questionnaire, despite the use of different languages, was identical. The data was collected via CATI (computer assisted telephone interviews) and CAPI (computer assisted personal interviews) method in cooperation with a professional public opinion research company and trained interviewers. Table 1 summarizes the information about recreational trips, respondents’ socio-demographics, and coastal sites in each country.

Table 1 Summary of the information about recreational trips and explanatory variables (standard deviations in brackets)

In TCM the distance to each of the analysed sites is the most crucial variable. We assumed that each respondent could have chosen from all the coastal recreational destinations in their country.⁴ Travel distances to each of the sites were measured using ARCGIS (ESRI® ArcMap™ 10.0) and MapQuest. For all survey respondents, we calculated the distances between origin, selected and potential destinations as a corrected Euclidian distance. The correction was calculated using a country-specific scaling coefficient to reflect the road network. The scaling was calculated first by taking a set of randomly selected grids with population > 0 and calculating the Euclidian distances in ARCGIS and road distances using MapQuest; next by averaging the Euclidian and the road network distances per country, and finally by calculating the ratio between the average route distance and average Euclidian distance.⁵

The travel cost⁶ was determined as a vehicle operating cost, (which conservatively included petrol, oil, and tire use only; Hang et al. 2016) and the opportunity cost of time⁷ (Czajkowski et al. 2019) for a two-way journey.⁸

The environmental qualities of coastal recreational destinations were characterized by their eligibility for the Blue Flag eco-label and by their level of compliance with the EU Bathing Water Directive (2006/7/EC). The Bathing Water Directive operates on four levels: excellent, good, sufficient, and poor compliance, or noncompliance. For simplicity of analysis, we grouped the good and sufficient compliance levels into one category: mandatory compliance. Limit values are defined by the presence of microbial parameters (concentration of Intestinal Enterococci (IE)/100 ml and concentration of Escherichia Coli (EC)/100 ml); physical parameters (pH and colour); and biophysical properties (water transparency measured by the Secchi depth, residues, and floating material). The Blue Flag programme requires sites to be at the excellent compliance level over a 4-year period on average, and possess a minimum level of infrastructure (toilet facilities with at least a septic tank, lifesaving facilities, and information on water quality at the site⁹), as well as at least one site per municipality with handicap access. We used data from the European Environmental Agency (EEA) official report for each country. Table 2 shows the quality parameters of bathing water and the values for the respective compliance levels.

Table 2 Water quality parameters according to the EEA bathing water directives and Blue Flag criteria

Respondents’ choices of recreational sites, along with the number of their visits to them, reveal information about their preferences. People choose sites that provide the best recreational opportunities, and how many trips they make is determined by preferences constrained by the budget. This information is sufficient to model demand for recreational trips and investigate how that demand is influenced by various site characteristics.

2.2 Theoretical and Econometric Framework

In our estimation of recreational benefits we follow the framework proposed by Hausman et al. (1995): a two-stage budgeting model, in which an individual first decides how many trips to make, and then decides how to allocate these trips across available recreational sites. This first step is modelled using a count data model, and the second step using a discrete choice model. Linking these two components is a best-practice approach for the estimation of recreational values, which has been in use since publication of the seminal paper by Bockstael et al. (1987)—see Parsons et al. (1999) for a discussion and a comparison with other approaches.

We start the formal description of the model with the second stage described in the previous paragraph. Here, for each trip that an individual $i \in \{ 1, \ldots ,N\}$ has decided to make in the first stage, we assume that a choice of the one of J^c available recreational sites is made to maximize the utility function¹⁰

$$U_{ij}^{c} = \alpha_{j}^{c} - \beta_{i}^{c} TC_{ij}^{c} + \varepsilon_{ij}^{c} .$$

(1)

In (1), $\alpha_{j}^{c}$ are site-specific constants $\left( {j \in \{ 1, \ldots ,J^{c} \} } \right)$,¹¹ with one of them being constrained to 0, and therefore used as a reference level. Inclusion of all possible alternative specific constants makes it impossible to estimate the effects of some site-specific characteristics, such as their environmental qualities, but this approach allows all possible site differences to be controlled for, including the unobserved site-specific attributes (Murdock 2006).¹²$TC_{ij}^{c}$ represents a travel cost equal to the total vehicle operating cost and the opportunity cost of travel time,¹³ whereas $\beta_{i}^{c}$ denotes the marginal utility of money. In order to account for preference heterogeneity and to relax the IIA assumption of a simple multinomial logit (MNL) model, we allow $\beta_{i}^{c}$ to vary across respondents. Because it is not feasible to estimate separate $\beta_{i}^{c}$ for each individual, we instead assume that it follows log-normal distribution in the population, and estimate the parameters of the distribution. This allows us to account for preference heterogeneity across respondents. Lastly, $\varepsilon_{ij}^{c}$ is a stochastic term following extreme value distribution, which leads to the well-known mixed logit formula of the likelihood function:

$$L_{i}^{c} = \int {\prod\limits_{j = 1}^{{J^{c} }} {\left( {\frac{{\exp \left( {\alpha_{j}^{c} - \beta_{i}^{c} TC_{ij}^{c} } \right)}}{{\sum {\exp \left( {\alpha_{l}^{c} - \beta_{i}^{c} TC_{il}^{c} } \right)} }}} \right)^{{y_{ij}^{c} }} } f\left( {\beta_{i}^{c} } \right)} d\beta_{i}^{c} ,$$

(2)

where $y_{ij}^{c}$ is equal to 1 if an individual i has chosen alternative j and 0 otherwise. The integral in (2) is approximated by using the Quasi Monte Carlo method with 1000 scrambled Sobol draws (Czajkowski and Budziński 2019).^14,15

Following Hausman et al. (1995), we define the inclusive value of individual $i$, expressed in monetary terms, as

$$S_{i}^{c} = \int {\frac{1}{{\beta_{i}^{c} }}\log \left( {\sum\limits_{j = 1}^{{J^{c} }} {\exp \left( {\alpha_{j}^{c} - \beta_{i}^{c} TC_{ij}^{c} } \right)} } \right)f\left( {\beta_{i}^{c} } \right)} d\beta_{i}^{c} ,$$

(3)

which corresponds to the expected utility from a trip choice situation, divided by the marginal utility of money.

Next, in order to obtain site-specific welfare estimates, we follow the approach set out in Termansen et al. (2013). The inclusive value when access to site k is lost can be calculated as:

$$\hat{S}_{ik}^{c} = \int {\frac{1}{{\beta_{i}^{c} }}\log \left( {\sum\limits_{\begin{subarray}{l} j = 1 \\ j \ne k \end{subarray} }^{{J^{c} }} {\exp \left( {\alpha_{j}^{c} - \beta_{i}^{c} TC_{ij}^{c} } \right)} } \right)f\left( {\beta_{i}^{c} } \right)} d\beta_{i}^{c} .$$

(4)

This is equivalent to assuming that the travel cost to site k becomes infinitely large. The loss of welfare due to the loss of access to site k is then given as $S_{i}^{c} - \hat{S}_{ik}^{c}$. Analogous calculations can be made for any subset of sites, including all the sites in a country, which means calculating the total recreational value of the Baltic coast for the citizens of the country.

In the first stage of the budgeting model, an individual decides how many trips to the Baltic coast to make. This decision depends on the vector of individual characteristics ${\mathbf{X}}_{i}^{c}$ and the price index. Following Hausman et al. (1995), we employ the inclusive value expressed in monetary terms, $S_{i}^{c}$, as a price index, and assume that the mean number of trips is given by:

$$\lambda_{i}^{c} = \exp \left( {{\mathbf{X}}_{i}^{c} {{\varvec{\uptau}}}^{c} + \phi^{c} S_{i}^{c} } \right).$$

(5)

The number of trips $\left( {T_{i}^{c} } \right)$ is modelled using the negative binomial P model (NBP; Greene 2008), in which the probability of t trips occurring is given by:

$$P\left( {T_{i}^{c} = t} \right) = \frac{{\Gamma \left( {\theta^{c} \left( {\lambda_{i}^{c} } \right)^{{Q^{c} }} + t} \right)}}{{\Gamma \left( {\theta^{c} \left( {\lambda_{i}^{c} } \right)^{{Q^{c} }} } \right)t!}}\left( {u_{i}^{c} } \right)^{{\theta^{c} \left( {\lambda_{i}^{c} } \right)^{{Q^{c} }} }} \left( {1 - u_{i}^{c} } \right)^{t} ,$$

(6)

where $u_{i}^{c} = \frac{{\theta^{c} \left( {\lambda_{i}^{c} } \right)^{{Q^{c} }} }}{{\theta^{c} \left( {\lambda_{i}^{c} } \right)^{{Q^{c} }} + \lambda_{i}^{c} }}$. $\theta^{c}$ and $P^{c} = 2 - Q^{c}$ are the parameters to be estimated, where for $P^{c} = 2$, the model collapses to the standard negative binomial regression (NB). The model is estimated using a weighted maximum likelihood method.

Estimating the total consumer surplus requires integrating the demand function over the price index (Bujosa Bestard and Riera Font 2010):

$$CS_{i}^{c} = \int\limits_{ - \infty }^{{S_{i}^{c} }} {\exp \left( {{\mathbf{X}}_{i}^{c} {{\varvec{\uptau}}}^{c} + \phi^{c} s_{i}^{c} } \right)ds_{i}^{c} } = \frac{1}{{\phi^{c} }}\exp \left( {{\mathbf{X}}_{i}^{c} {{\varvec{\uptau}}}^{c} + \phi^{c} S_{i}^{c} } \right).$$

(7)

The per-trip consumer surplus can then be easily calculated by dividing (7) by the expected number of trips, namely $CS_{c}^{per - trip} = \frac{{CS_{i}^{c} }}{{\lambda_{i}^{c} }} = \frac{1}{{\phi^{c} }}$. As can be seen, the per-trip consumer surplus does not vary across respondents, but does vary across countries. Furthermore, the resulting total change in the consumer surplus related to the loss of access to site k can be calculated as¹⁶:

$$\Delta_{k} CS_{i}^{c} = \frac{1}{{\phi^{c} }}\exp \left( {{\mathbf{X}}_{i}^{c} {{\varvec{\uptau}}}^{c} } \right)\left[ {\exp \left( {\phi^{c} S_{i}^{c} } \right) - \exp \left( {\phi^{c} \hat{S}_{ik}^{c} } \right)} \right].$$

(8)

As noted earlier, when the choice model includes site-specific constants as in (1) it is not possible to estimate the effect of some environmental characteristics of a site, as they do not vary across respondents. Therefore, to calculate the effect of improving the environmental conditions at the Baltic coast we follow Murdock (2006). Specifically, we estimate an auxiliary regression, treating the site-specific constants as a dependent variable

$$\alpha_{j}^{c} = \omega^{c} + \gamma {\mathbf{E}}_{j}^{c} + \eta_{j}^{c} ,$$

(9)

and using site characteristics, ${\mathbf{E}}_{j}^{c}$, and country-specific dummies, $\omega^{c}$, as independent variables. One can then calculate consumer surplus upon change in the environmental conditions, $\Delta {\mathbf{E}}_{j}^{c}$, by calculating updated site-specific constants, $\hat{\alpha }_{j}^{c} = \alpha_{j}^{c} + \gamma \Delta {\mathbf{E}}_{j}^{c}$, and then plugging them into Eqs. (3) and (7).

3 Estimation Results

We start with the results of the site choice models, which are presented in Table 3. For Demark, Finland, Latvia and Sweden a mixed logit (MXL) model was used, assuming log-normal distribution for the travel cost variable as described in the previous section. For the rest of the countries a simple MNL model was chosen with a fixed travel cost coefficient.¹⁷ The site-specific constants are not reported for brevity—they are available in the supplementary materials to this paper. The estimates of the travel cost coefficient, for the countries for which the MNL was used, are significant and of the expected sign. For the rest of the countries, we report the parameters of the log-normal distribution, namely the mean and standard deviation of the underlying normal distribution. The coefficients for the travel cost cannot be compared between the countries, because (as is the case in all discrete choice models) they are confounded with country-specific scale parameters. The lower part of Table 3 presents model diagnostics. The number of estimated parameters corresponds to the number of observed recreational sites minus one and the coefficients for the travel cost. The Ben-Akiva–Lerman pseudo-R² represents the average predicted probability of the site actually selected. It indicates that the explanatory power of the models varies between countries, with the probability of correct predictions ranging from 11 to 75%.¹⁸ Additional explanatory variables, such as the month when the trip was made or the self-reported purpose of the visit, did not significantly improve the models’ fit to the data. We also tried to use different approaches to explain respondents’ choices by including visited site characteristics (such as compliance with EEA bathing water directives and Blue Flag criteria, water quality, or the size of the population in the area) in these models directly, instead of the alternative specific constants. However, these models were worse fitted to the data, relative to the models presented in Table 3.

Table 3 The results of country-specific site choice models

In Table 4 we present the results of linear regression¹⁹ with site-specific constants from choice models as dependent variables, as described by Eq. (9) in the previous Section. We use this approach to investigate the welfare changes associated with improvements in the water quality of a recreational site, rather than the recreational value of individual sites. The model was estimated jointly for all countries, with coefficients for site quality attributes assumed to be the same across countries. This restriction was necessary to ensure identification of the model—for some countries, water quality characteristics either do not vary or, if they do vary, the number of sites is too small, which may result in spurious parameters.²⁰ Such specifications are in line with the approach used by Czajkowski et al. (2015), in which the effect of (perceived) environmental quality was assumed to be homogeneous across countries.

Table 4 Linear regression with site-specific constants from choice models as a dependent variable

The results show that the Blue Flag quality indicator is significant and has an expected sign, indicating that people prefer sites awarded Blue Flag status. On the other hand, water quality indicators (compliance with the EU Bathing Water Directive) as well as indicators for concentration of EC and IE bacteria were not significant. Finally, the model also includes population density at different sites to control for the urban vs. rural character of a site. These coefficients indicate that individuals on average prefer sites with a higher population in a 3 km radius. This would describe small towns, but with relatively high population (and therefore possibly better infrastructure and accessibility).

Table 5 presents the results of the NBP models that were used to model the number of trips to the Baltic coast. The model specification follows that described by Czajkowski et al. (2015), except for the travel cost, which is replaced with the measure of inclusive value obtained from the site choice model, in line with the approach outlined in Sect. 2.2. The results show that the NBP model performs well and offers a significant improvement over the standard negative binomial regression. The estimated coefficients for the inclusive value are all significant and positive, ranging from 1.06 to 4.99. The coefficients associated with the other covariates differ substantially between countries. These disparities reflect a variety of factors, such as cultural differences, different recreational habits, or different substitutes available for recreation. Finally, the differences between the estimates of the parameters $\theta$ and $P$ for different countries indicate that there are variations not only in the mean number of trips, but also in the shape of their overall distribution.

Table 5 The results of the country-specific count data (number of trips) models

4 Interpretation of the Results

The models presented in Sect. 3 allow us to calculate the mean consumer surplus (CS) per trip, predict the number of trips, and calculate the mean consumer surplus associated with the ability to make the predicted number of trips.²¹ These results are presented in Table 6.

Table 6 Estimated economic benefits associated with recreational use of the Baltic coast

We find that Poland and Germany have the highest CS per trip (92–95 EUR), while Finland and Sweden have the highest total CS for all trips (625–672 EUR). On the other hand, Estonia has the lowest total CS of only 95 EUR for all trips. This result is partly driven by the low number of trips in Estonia, and partly by the low value per trip.

The total consumer surplus is presented in the fourth row and can be compared with the results of the simpler (non-spatially-explicit) approach of Czajkowski et al. (2015), presented in the last row. For most countries we find that the current approach produces higher estimates, with Estonia being the only exception, and Sweden having very similar estimates in both approaches. Overall, we find that the total CS associated with the recreational use of the Baltic coast accounts to 27.5 billion EUR annually. This result is almost twice as high as the 14.8 billion EUR estimated by Czajkowski et al. (2015). As our new estimates are based on a spatially-explicit econometric approach these results can be considered more reliable and they also indicate a somewhat different distribution of recreational benefits among countries compared to using a traditional single-site TCM, which disregards substitution possibilities.

4.1 Identifying Hot Spots of Recreational Value

The results allow us to calculate the total yearly welfare per site, which is an indication and illustration of recreational hot spots and associated site values per country around the Baltic Sea. The welfare loss per site is calculated as the mean change in consumer surplus according to formula (8) multiplied by the adult population of each country.

The total annual value of sites ranges from 37.7 thousand EUR per site in Denmark to more than 600 million EUR per site in Sweden and Finland. For Finland, a very high consumer surplus loss per site per average respondent combined with relatively few sites in the survey produce high maximum site values despite a relatively small population. Note that these values only apply to recreational trips undertaken within each country, i.e., no cross-border recreation/tourism is accounted for in this analysis. Table 7 gives a statistical overview and Fig. 1 is a map of recreational site values. Recreational hot spots, defined as Baltic sites that represent the highest recreational values, are situated primarily along the German and Polish coastline, Stockholm in Sweden, Turku and Helsinki in Finland. In Denmark, with many coastal sites and relatively short distances to the coast for a population spread all around the country, site values are necessarily relatively low compared to hot spot areas along the German and Polish coastlines. Nevertheless, site values in Denmark are very heterogeneous, ranging from around 38,000 EUR to more than 54 million EUR. As expected, we find for each country that total site benefits are generally higher in highly populated areas (e.g., Helsinki, Stockholm, Kiel, Gdańsk) as well as in areas that are popular tourist resorts for the national populations (e.g., Rügen in Germany, Lysekil in Sweden, Kołobrzeg in Poland). This is in line with economic theory.

Table 7 Estimated total annual recreational site values

Fig. 1

Hot spots of recreational value of coastal sites

4.2 The Effects of Improvements in Environmental Conditions

Finally, we use the estimated effect of Blue Flag status on the welfare that people derive from recreational trips to the Baltic in order to simulate the welfare change that would result from improvements in water quality and infrastructure.²² Specifically, we estimate the welfare change that would occur if all coastal recreational sites were in compliance with Blue Flag standards. The results are summarized in Table 8.²³

Table 8 Estimated change in economic benefits associated with increased Blue Flag attributes

The first two rows of Table 8 provide information regarding the total number of sites for each country, as well as the share of sites that had Blue Flag status at the time of the survey. As can be seen, Denmark and Poland have the highest share of sites with the status, whereas Finland and Sweden have the lowest. The third row of Table 8 presents the value of total CS, which was previously presented in Table 6. Next, we present changes in CS that would occur if all sites complied with the Blue Flag requirements. The results are shown both in absolute and relative terms. We find that the highest percentage change per trip is observed for Sweden, followed by Lithuania, Finland and Poland.²⁴

Finally, for the sake of comparison, the last row of Table 8 presents the estimated welfare change associated with water quality improvements as described in Czajkowski et al. (2015). These numbers were based on changes in perceived water quality—the 5-level Likert scale question: “In your opinion, what is, on average, the status of the environment in the [respondent’s country] part of the Baltic Sea coast?” Two effects of a unit increase in water quality perception were simulated: the impact on the probability of engaging in the Baltic Sea-based recreation, and on the expected number of trips, which allowed a calculation of the resulting increase in total welfare. In this study, we improve this estimate by linking it to a change in a characteristic that can be objectively observed—Blue Flag status—and we then use respondents’ actual behaviour to predict its welfare effects.

The estimated welfare change, that is the economic value that people attach to upgrading all sites in their country to comply with Blue Flag standards, is 6.2 billion EUR.²⁵ This is considerably more than a previous estimate of 1.09 billion EUR, although as noted earlier in this paper, the two estimates are not fully comparable, as they are based on different measures of environmental change (actual vs. perceived), different methodology (site-specific vs. single-site approach) and somewhat different samples (as described in Sect. 2). Overall, an approximately 6.2 billion EUR per year gain (an increase of almost 22.4%) in recreational value is an inspiring number that can be used for policy purposes when compared with the cost of achieving the corresponding water quality conditions in all recreational sites at the Baltic Sea coast.

4.3 The Effects of Improvements in Environmental Conditions—Practical Example

In the previous section, we noted that improvement of all non-Blue Flag status sites to blue flag status could results in welfare gain of 6.2 billion EUR. While this number can be highly evocative and can emphasize the importance of improving the quality of the environment, it may be hard to improve the water quality conditions in all of the sites to comply with the EEA bathing water directives (and Blue Flag criteria). Therefore, we also conducted a simulation to show how sites with the worst environmental status²⁶ can gain in value by achieving Blue Flag eco-label. The results are presented in Table 9.

Table 9 Change in recreational value of coastal sites associated with increased Blue Flag attributes

The sum of the increase in the value of the eight sites examined amounted to 409.49 million EUR with the biggest improvement for ‘Kaerradal Soedra (2)’ near Varberg in Sweden (with approximately 51.53% recreational site value increase, from 2.28 million EUR to 4.71 million EUR) and ‘Plaża wschodnia’ near Łeba in Poland (with approximately 51.51% recreational site value increase, from 60.96 million EUR to 125.72 million EUR). However, it is worth pointing out, that improving only a subset of recreational sites, as classical economics predicts, will result in decline in value of the other sites (via substitution effect).²⁷ This decrease is mainly related to the improvement in the value of the sites with the worst characteristics, and the lack of changes in the sites with higher environmental quality (e.g., with Blue Flag status). For instance, sites with better characteristics will lose some visitors, who will move to newly-improved sites. Therefore, in our exemplary simulation (of improving 8 sites with the worst environmental characteristics) the adjusted increase in the value of all sites is equal to 82.17 million EUR (0.72% increase of total recreational value of all sites from examined countries).

5 Conclusions

This paper estimates a spatially explicit discrete choice and count data model for coastal recreation in eight of the nine countries around the Baltic Sea. Furthermore, it tests inclusion of site attributes as fixed effects, general water quality attributes based on monitored site-specific data, and indicators of population density. The results present the total recreational value per site, spatially identified recreational hot spots, and the potential welfare effects of improving environmental and infrastructure conditions to the level required by the Blue Flag standard at all sites. The total recreational value of Baltic Sea coastal locations cumulates to 27.5 billion EUR with significant variations within and across countries and reveals a significant heterogeneity across sites and countries. The annual total per site value ranges from slightly more than 38 thousand EUR for a coastal site in Denmark to more than 700 million for a site in Finland. The improvement of site quality to Blue Flag standard has a substantial positive welfare effect of nearly 6.2 billion EUR, indicating a significant welfare improvement (22.4%) derived from enhancing water quality and recreational infrastructure. Through the geographical scale, spatial explicitness and projections, our estimations offer an extension and improvement with reference to previous studies on recreational values at the Baltic coast, which also used revealed preference valuation methods (Sandström 1996; Soutukorva 2005; Vesterinen et al. 2010; Czajkowski et al. 2015).

The limitations of our study stem mostly from the data that we are working with. First, we only have detailed information (including the visited site) about the last recreational trip that the respondent made to the Baltic Sea. As such, it is not possible to base our approach solely on RUM-based choice models (e.g., Fezzi et al. 2023) or use a discrete–continuous approach (e.g., Von Haefen and Phaneuf 2005). Combining the choice model with a count data model allows for spatially-explicit inference, but we acknowledge that it would be beneficial to compare our outcomes with different available methods. Unfortunately, this is not possible with the data that we currently have. Second, for several countries, we identified a limited number of sites, which makes it difficult to directly control for sites’ characteristics on the country level. To identify these effects, one would have to identify a higher number of sites in each country to exploit geographical variation or obtain recreational data for several years to utilize some temporal variation. Alternatively, combining our data with stated preference methods could shed some light on the importance of various characteristics of recreational sites (Englin and Cameron 1996; Adamowicz et al. 1997; Fix and Loomis 1998). As we do not have such additional data, we were only able to estimate the joint effects of sites’ characteristics for all the countries together (consider Table 4). Third, we employ Blue Flag status as a measure of environmental quality. Admittedly, the Blue Flag has several other eligibility criteria related to safety, infrastructure and education that could be also driving our results. In an additional analysis (available in online materials) we found that environmental quality indicators such as concentrations of EC and IE, were not significant, even if the Blue Flag variable was excluded from the model. The possible explanation could be that tourists are unaware of the environmental quality in general, but they treat the Blue Flag status as its proxy. Nonetheless, further research is needed to determine which components of the Blue Flag programme drive tourists’ behaviour. Lastly, we limited the data to only domestic travel, excluding the possibility of international trips. As such, the estimated benefits are likely underestimated. We note, however, that the number of international trips that were reported was rather small in our data. As such, this effect would be difficult to capture. Further research is needed to provide guidance on how to properly incorporate international trips into travel cost valuation studies.

A number of challenges must be noted for the conducting of a future similar study across the eight countries with the data available: too little variation to effectively capture the effects of general water quality on site preferences (or to separate the effect of the Blue Flag status into individual components, including environmental quality, safety, accessibility and instructional criteria), either because of too few sites per country or because of the lack of variance in site characteristics between countries. New types of data may prove relevant and significant when introduced into this type of analysis, such as meteorological data, information on water temperatures, the presence of algal blooms and other information that people may look for when deciding whether to visit a site for recreation. Similarly, information on beach characteristics—whether a beach is a sandy, rocky or grassy, littered, what types of surroundings it has, the degree of naturalness—are known from other stated and revealed preference studies to have a significant impact on welfare. Incorporating such data sources in further analyses could provide a valuable source of additional information, and facilitate more efficient tourism management with increased social welfare benefits.

In this study, we take into account the impact of Blue Flag status, which is information that is both easily available to visitors and can be objectively, and without much environmental knowledge, be assessed as indicating real (not perceived) environmental conditions. We show that Blue Flag status has a large impact on the value that people attach to different recreational sites, which is a strong indicator that visitors have environmental preferences in relation to sites. This should be taken into account when constructing policies for Baltic coast recreation.

In summary, our study estimates the economic value of Baltic Sea-based recreation at 27.5 billion EUR annually, and describes its distribution among the coastal locations. Using spatially explicit travel cost methods significantly improves on value estimates of coastal recreational compared to traditional single-site travel cost approaches. Identifying recreational hot spots and areas with possibly untapped potential for quality improvements, has clear management applications. In addition, we show how recreational behaviour is affected by environmental quality and we simulate welfare changes resulting from its enhancement. This can hopefully provide planners and policy makers with needed argumentation for further improving water quality and coastal recreation infrastructure. Improved water quality and infrastructure boost the recreational value by nearly 6.2 billion EUR in the Baltic Sea region, demonstrating their importance for the management of recreational sites. Overall, it is clear that although the Baltic Sea coast is a major source of recreational value for the eight European countries studied, recreational benefits could be significantly increased through environmental and infrastructure improvements.

Appendix

See Tables 10, 11, 12, 13, and 14.

Table 10 The results of country-specific site choice models when opportunity cost of time is calculated as three-fourths of the average hourly earnings in each country

Table 11 The results of the country-specific count data (number of trips) models when opportunity cost of time is calculated as three-fourths of the average hourly earnings in each country

Table 12 Estimated economic benefits associated with recreational use of the Baltic coast when opportunity cost of time is calculated as three-fourths of the average hourly earnings in each country

Table 13 Estimated total annual recreational site values when opportunity cost of time is calculated as three-fourths of the average hourly earnings in each country

Table 14 Estimated change in economic benefits associated with increased Blue Flag attributes when opportunity cost of time is calculated as three-fourths of the average hourly earnings in each country