A Meta-Regression Analysis of Hunters’ Valuations of Recreational Hunting

Abstract

Hunters’ valuations of recreational hunting have been estimated by a large number of location-specific studies since the early 1970s, but to date there has been no systematic assessment of this research at the global scale. The present study performed a meta-analysis of 80 studies with 588 value-per-day estimates. The assessment showed a high concentration of studies pertaining to the valuation of deer and the valuation of hunting in the USA. The average value was USD 69 /hunting day in 2020 prices, but the variation was large, ranging from USD 4 to 325 /hunting day. The statistical performance of alternative mixed-effect models explaining the estimated value differences was tested with different hunting attributes (targeted game animal); context variables (income/capita, population density, year of study, region of application); and study characteristics (valuation method, publication outlet). The results showed that the type of game animal, income per capita, and valuation method had significant effects on estimated values. The predictive power was high for all models, supporting the application of the meta-analysis results to guide the management of hunting where primary valuation studies have not been undertaken, in particular outside the USA.

1. Introduction

Game animals have been valued as a human food resource throughout history, and hunting is as old as humankind. Nowadays, game animals are a source of recreational benefits, with animal watching and hunting generating significant incomes for individuals and communities [1,2]. Hunting can account for 7% of the gross domestic product in countries where hunting is an important part of the tourism income (review in Gren et al. [3]). Game animals may also perform important functions in the food web and ecosystems and can have value purely by existing. The recreational and other values of hunting can be of considerable importance to local economies, the determination of land use, and the management of game animal populations. Recreational hunting can also threaten species conservation and have a negative impact on local livelihoods [4]. Overabundant game animal populations can generate costs to landowners due to the damage animals cause to crops and young trees, and several game (e.g., moose and deer) present significant physical risks through vehicle accidents (see Gren et al. [3] for a review of the costs of wildlife).

The management of these conflicting interests has been an issue in many countries for a long time (e.g., Pack et al. [5] for a review). The efficient management of game animals requires some weighting of the total value of recreational hunting with the associated costs in terms of, e.g., the threat to biodiversity or the damages incurred by landowners. The total value includes the economic effect on local and regional economies of hunters’ spending on, e.g., equipment and lodging, and the hunters’ perceived valuation in excess of their costs. While hunters’ spending is materialized as incomes for local and regional economies, the hunters’ valuation of the hunting is more difficult to quantify. Economists started to estimate hunters’ valuation of recreational hunting in the early 1960s (see partial reviews in, e.g., [6,7,8,9]). Estimated values (willingness to pay—WTP) for game hunting vary greatly between studies. For example, the Federal-Provincial Task Force [10] reported a value-per-day of USD 4, and Pang [11] estimated a value-per-day of USD 325 in 2020 USD. These differences can be explained, in part, by several factors, such as the differences in the type of game animal hunted, the valuation method, and contextual factors, which are likely to influence the estimated values. Despite the existence of several reviews, there has been no systematic, global-scale review of hunting valuation studies.

The purpose of this study was to undertake a systematic review of studies that estimate hunters’ valuations of recreational hunting across national domains, species hunted, values measured, and valuation methods. Recreational hunting is defined here as the killing of game animals for the main purpose of pleasure and enjoyment and not for commercial or subsistence purposes [3]. This definition excludes studies focusing on the role of hunting tourism in generating incomes for local economies. Three main types of questions are raised: (i) Descriptive—which recreational hunting values have been estimated? Where? When? Which valuation methods have been used? (ii) Analytic—which factors explain the differences in estimated hunters’ valuations? (iii) Policy-relevant—what is the predictive power of these models?

The questions are addressed using meta-regression analysis (MRA), first suggested by Glass [12], which has the appeal of combining empirical evidence from multiple independent studies that use different methods and datasets (see Tipton et al. [13] for a historical review of MRA). MRA has been used in a number of different studies as a means of synthesizing information and data from previous research (e.g., [14,15]). It can offer insight into the determinants of effects when controlling for study characteristics and contextual factors. MRA can therefore provide more information than simply tabulating results from a literature review of previous studies. Another advantage of MRA is the superiority of the method for policy analysis when environmental values from source study sites are transferred to policy sites for which values are not available (e.g., [8,16]). Site-specific valuation studies can be costly and time-consuming, and benefit transfer is then a powerful tool for assessing hunting values at policy sites.

Starting in the early 1990s, a large body of literature has been produced on the MRA of recreational values (see, e.g., Schägner et al. [17] for a brief review). These MRA studies have been applied to different types of ecosystems (e.g., wetlands, coral reefs, and forests) or recreational activities, such as fishing and hunting [18,19,20]. To the best of our knowledge, three previous MRA studies have focused on evaluating differences in estimated hunting values among primary studies, all of which were applied to specific regions or types of game animals [7,21,22]. All three studies measured recreational hunting in per-day values—noting that Kerr and Woods [7] also estimated the value per hunt. Loomis and Richardson [21] and Huber et al. [22] conducted MRAs of hunting values in the USA alone. Kerr and Woods [7] limited their analysis to the value of hunting big game animals but did so at the global scale.

The main contributions of this study are: (i) the updated and enlarged scope of the systematic assessment of studies that estimate the non-market value of game hunting; (ii) an expansion of the geographical areas covered by previous overview studies; and (iii) the comparison of models with alternative specifications for the dependent variable. In addition, the income in the region of application was added as an exploratory variable, since it can be an important determinant of value (e.g., [23]) but was not included in the previous three MRA studies of hunting values.

This study is organized as follows: Section 2 describes the materials and methods, including the collection of studies and data on the dependent and independent variables, the specification of the regression models, and the calculations of predictive power. The results regarding the three questions raised in the study are presented in Section 3, and the study ends with a discussion and concluding remarks.

2. Materials and Methods

2.1. Collection of Studies and Data

Source studies were identified by three different methods: (i) an internet search of various databases using various combinations of the keywords hunting OR game AND willingness to pay OR value; (ii) the collection of studies from existing databases of hunting values, agency web pages, and authors known to have undertaken non-market valuations of hunting; and (iii) the application of the snowball method. This literature search ended in June 2022.

Databases searched with the key words included Scopus, Semantic Scholar, and Web of Science. The Web of Science and Scopus websites provided studies published in journals, and the other search engines identified studies from non-academic institutions and reports and working papers from academic institutions that were not published in journals. Regarding the second method, two main databases of hunting values were used: the Recreational Use Values Database (RUVD) [24] and the United States Geological Service (USGS) [9]. The RUVD [24] database includes studies on hunting- and wildlife-associated recreation values in the USA and Canada from 1958 to 2015. There is considerable overlap between the RUVD [24] and the USGS [9], the latter including USA studies classified according to the type of recreation, with hunting as a separate activity class. The snowball method was particularly useful for identifying many studies from references within or to specific studies. Prior research identified many hunting valuation studies [6,7,8,18,19,21,25,26,27,28,29].

The requirements for the inclusion of a study in the analysis were that it reported (i) the value estimate, (ii) the region of application, (iii) the value estimation method, and (iv) the unit of measurement. Only value estimates of consumer surplus were included, i.e., hunters’ values in excess of the costs of hunting, which is the theoretical foundation for the economic valuation of non-market ecosystem services such as recreational values [30]. In total, 179 studies were identified with this information. After correction for duplicates, 126 usable studies remained, and 80 studies provided information on value-per-day. See Figure A1 in Appendix A for a flow chart of the selection steps.

The first study was conducted in 1961 by Davis [31], and the most recent was a study on hunting values in Italy carried out in 2020 [32]. A list of all the studies can be found in Table S1 in the Supplementary Materials. The 80 included studies provided 588 usable observations, resulting in an average of 7.5 observations per study. This was above the average of 4.9 for environmental economics meta-regression studies reported by Nelson and Kennedy [15], which was mainly based on studies of WTP at the state level in the USA. The largest such study included 90 observations [33]. The reported value estimates were converted to 2020 values based on country-specific consumer price indices [34], followed by conversion into USD using the average purchasing power parity exchange rate for 2020 [35].

A common approach to explain the variation in estimated recreational values between studies using MRA is to include independent variables reflecting the quality of the recreational activity, context, and study characteristics [20]. Quality aspects (such as harvest rates or game abundance) can be important determinants of satisfaction from hunting. The contextual factors reflect general aspects such as income and regional culture, which can be important aspects of hunting participation. Study-specific factors usually include the choice of valuation method and the study design, which are known to affect value estimates.

Hunters’ satisfaction depends on a variety of hunt attributes, which include crowding among hunters, social aspects, nature experience, the challenge, the harvest, and the type of game animal hunted (e.g., [36]). Except for the type of game animal, and sometimes the harvest, data on such attributes were not available in the published studies. With respect to the game animals, most studies valued a specific game animal (such as moose or deer), while other studies did not specify a particular game animal, but instead estimated the value of big game or small game. Seven animal categories and associated dummy variables were identified: ‘moose’, ‘elk’, ‘deer’, ‘bird’, ‘small game’, ‘big game’, and ‘other game’. Some small-game studies included birds, and some big-game studies included moose, deer, or elk. ‘Other game’ was applied to species for which few valuation studies have been conducted (such as wild boar or goats) and when studies did not specify a species. In general, we expected higher values for large game than for small game, because of, e.g., the greater volume of meat harvested and the more prominent role of large game species as trophies.

The contextual variables included in this study were income, population density, the region of application, and the year of study. Most studies did not report hunters’ incomes. Income per capita measured by gross domestic product (GDP) in the country or region was therefore used as a proxy, which is a common approach in MRAs [20,23]. A drawback of this measurement method was that it did not include non-resident hunters, for whom data were not available in most studies. On the other hand, the share of non-resident hunters can be relatively low: for deer, it averages 13% in the USA and 2% in Canada [37], with many of those being out-of-state/province hunters, rather than foreigners. Further, when benefit transfer is of interest, data on hunters’ incomes (rather than population-wide measures of income) are difficult or impossible to obtain, both for the source sites and for the policy sites to which values are transferred. A few studies reported hunting values for different states and provinces in the USA and Canada, respectively, and in those cases the state or province GDP/capita was used instead of the national-level data. For the USA, GDP/capita at the state level was obtained from [38,39], and for Canada, it was obtained from [40]. Data on GDP/capita for the other countries were determined on a national basis [41].

Population density is often introduced in MRAs of recreational values to account for environmental pressure, which can affect the habitats of different animals (e.g., [23]). Data on population density (people per km²) for different states in the USA and provinces in Canada were obtained from [40,42], respectively. Population density data for the other countries, measured at the national scale, were obtained from [43]. Dummy variables identified three regions of application: ‘USA’, ‘Canada’, and ‘other countries’. The variable ‘year’ represented the year of data collection, which covered a period of almost 60 years from 1961 to 2020, with studies published a few years later (from 1964 to 2021).

Study-specific factors were the valuation method and the type of publication outlet. Previous MRAs of environmental values have shown that results vary systematically by valuation method [7,17,18,19,44], so dummy variables were included to enable the identification of valuation method effects. The methods for obtaining estimates of non-market hunting values were divided into revealed and stated preference methods. The travel cost method (‘TCM’) is one of the most frequently applied revealed preference methods. It links unpriced public goods to a priced market good, such as costs for travel to a site for recreation. Another common revealed preference method, the hedonic technique (‘Hedonic’), utilizes the observed market price effects of a non-market good. One example is the impact of environmental conditions on estate prices. However, a limitation of the revealed preference methods is that they cannot be used to elicit non-use or existence values of environmental changes. In order to deal with this limitation, stated preference methods were developed, which are based on surveying individuals to elicit their willingness to pay or accept compensation for hypothetical environmental changes. The most common are the contingent valuation method (‘CVM’) and the choice experiment (‘CE’) method. The CVM provides an estimate of the willingness to pay for, in our case, hunting. The CE method was developed to assess the value of different attributes, such as social aspects and harvest [3].

Following the literature on MRA, we also introduced a dummy for studies published in scientific journals with an independent referee system (e.g., [20,45,46,47,48,49]). This variable reflected the potential existence of publication bias, which may arise as insignificant results or low value estimates may be less likely to be submitted and/or published. The dataset included 33 studies published in the so-called grey literature by non-academic public authorities, 29 studies in journals with a referee system, and 18 studies published at academic institutions as theses or research papers.

In total, 18 explanatory variables were considered for the regression analyses, 7 describing hunt attributes, 6 contextual factors, and 5 study characteristics (

Table 1. Variables used in the MRA.

Variable	Description
Dependent variable	Value per day in the primary studies converted to 2020 prices followed by exchange rate conversion in purchasing power parity USD
Explanatory variables
Hunt attributes;
Elk	Indicator variable = 1 when elk was the target animal, 0 otherwise
Moose	Indicator variable = 1 when moose was the target animal, 0 otherwise
Deer	Indicator variable = 1 when deer was the target animal, 0 otherwise
Waterfowl	Indicator variable = 1 when waterfowl was the target animal, 0 otherwise
Small game	Indicator variable = 1 when small game was the target animal, 0 otherwise
Big game	Indicator variable = 1 when big game was the target animal, 0 otherwise
Other game	Indicator variable = 1 when other game was the target animal, 0 otherwise
Contextual variables
GDP/capita	Continuous; GDP per capita, thousand purchasing power parity USD in 2020
Population density	Continuous; people/km2
USA	Indicator variable = 1 when the study was conducted in the USA, 0 otherwise
Canada	Indicator variable = 1 when the study was conducted in Canada, 0 otherwise
Other country	Indicator variable = 1 when the study was conducted in another country, 0 otherwise
Year	Continuous; the year of the study
Study characteristics
CVM	Indicator variable = 1 when the contingent valuation method (CVM) was used, 0 otherwise
TCM	Indicator variable = 1 when the travel cost method (TCM) was used, 0 otherwise
CE	Indicator variable = 1 when the choice experiment (CE) method was used, 0 otherwise
Hedonic	Indicator variable = 1 when the hedonic valuation method was used, 0 otherwise
Journal	Indicator variable = 1 when the study was published in a journal, 0 otherwise

).

2.2. Meta Regression Modeling and Prediction

MRAs were modeled for the different sub-samples for which there were sufficient data. In general, 10 observations per explanatory variable are regarded as sufficient, although more observations are advised [50]. Applying this simple rule and recognizing that the number of explanatory variables was at least 10 for the sub-samples of value-per-day measurements of game animal and valuation method (six contextual variables plus a minimum of four study characteristics), we required 100 observations. This allowed for regressions on sub-samples of targeted game animal (deer, waterfowl) and valuation method (CVM).

The dataset was hierarchical, comprising studies at the top level and observations at the bottom level, with the associated risk of within-study correlation. Therefore, a mixed-effect model with random effects was used, which is common in MRAs of environmental values [15,51,52]. This method accounts for correlations in observations within and between studies [53].

Tests were conducted using maximum likelihood estimation with the random effect of a specification where the dependent variable, income, and population density were logged and a specification without any logged variables. The results showed that the logged specification provided the best statistical outcomes, as measured by the information criteria and the McFadden pseudo R² for each of the sub-sample models. The regression equation was then written as:

$$ln{V}_{ij}={\beta }_0+{\beta }_{1}lnGDP{C}_{ij}+{\beta }_{2}lnPopden{s}_{ij}+{\beta }_{3}Yea{r}_{ij}+{{\displaystyle \sum }}_h{\beta }_{4h}{X}_{ih}+{\lambda }_{0j}+{\epsilon }_{ij}$$

(1)

where V_i,j is the estimated value of hunting for observation i in study j; GDPC_ij is the GDP/capita; Popdens_ij is the population density; Year_ij is the year of study; and X_ih is the vector of dummy variables h = 1, .., m for region, game animals, and study characteristics. The random effect at the study level in the intercept is presented by the term ${\lambda }_{0.j}$, and ε_ij is the error term.

Regression Equation (1) was estimated for all models with the mixed command for linear regressions in Stata software [51]. For all regressions, ‘USA’, ‘Journal’, ‘CVM’, and ‘Deer’ (when appropriate) were the reference dummy variables. For example, when a deer subsample model was estimated, there was no need for the ‘Deer’ dummy. Smeared estimated values account for bias from the non-zero mean error distribution for logged-dependent variable models, and the predicted value was thus calculated as [54]:

$$Valu{e}_{ij}^{predict}=\exp \left(ln{V}_{ij}\right) \ast \exp \left(\epsilon /2\right)$$

(2)

where ε is the error term [54]. The predictive power of the regression models was estimated for all models, measured as the percentage difference between the predicted and source value for each observation, PD, according to:

$$P{D}_{ij}=\frac{Valu{e}_{ij}^{predict}-Valu{e}_{ij}^{source}}{Valu{e}_{ij}^{source}}\ast 100$$

(3)

where $Valu{e}_{ij}^{source}$ is the source value. Individual-case predicted value distributions accounted for case-specific random errors and delta-method standard errors. For each model, the mean percentage error (MPE) was calculated by averaging all PD scores for all observations. We also report the mean predicted values; their standard error and 95% confidence intervals; and the root mean square error (RMSE), which is a measure of the mean deviation from the data in relative terms. Since RMSE is scale-dependent, it was normalized by dividing it by the mean value, RMSE/µ, where µ is the mean source value for the full and sub-sample models.

3. Results

3.1. Descriptive Data Analysis

A simple data description was used to approach the first question raised in this study, i.e., what has been valued, how, and where? Most of the studies targeted deer, using the CVM valuation method, and were conducted in the USA (

Table 2. Number of source studies overall and by region, game animal, and valuation method.

	USA	Canada	Other Countries	Total
Total	63	9	8	80
Game animal:
Moose	9	1	1	11
Elk	15	0	0	15
Deer	32	2	2	36
Waterfowl	19	3	1	23
Small game	12	2	1	15
Big game	19	4	1	24
Other animals	9	1	1	14
Valuation method:
CVM	35	6	5	46
TCM	34	4	2	40
CE	0	0	1	1
Hedonic	2	0	0	2

).

Note that the sum of the number of studies in each category may exceed the total number of studies, since some studies estimated values for several game animals or used and compared valuation methods. The relatively large number of studies on WTP for hunting in the USA (79%) and deer hunting (45%) was consistent with Minin et al. [3], who conducted an MRA of hunting in general considering studies at the global scale. The CVM and TCM were used in similar proportions, but only a few studies used other methods.

The 80 studies generated 588 observations that were used in the MRA. The dominance of the USA therefore increased, since several of the USA studies contained a relatively large number of observations (

Table 3. Descriptive statistics, N = 588 individual value estimates from 80 studies.

	Mean	St Dev	Minimum	Maximum
Dependent variable:
Hunt value per day, 2020 USD	69.37	41.42	4.2	325.86
Explanatory variables:
Hunt attributes
Game animal:
Moose	0.02		0	1
Elk	0.06		0	1
Deer	0.52		0	1
Waterfowl	0.21		0	1
Small game	0.05		0	1
Big game	0.11		0	1
Other animals	0.02		0	1
Contextual variables
GDP/capita, constant thousand 2020 USD	37.12	10.11	16.02	75.68
Population density, people/km2	45.19	67.21	0.1	467
Region
USA	0.84		0	1
Canada	0.14		0	1
Other countries	0.02		0	1
Year of study	1991	11	1961	2020
Study characteristics
Valuation method:
CVM	0.89		0	1
TCM	0.09		0	1
CE	0.01		0	1
Hedonic	0.01		0	1
Journal publication	0.08		0	1

).

The average value was approximately USD 69, but the variation was large. Note that, despite the relatively large number of studies published in journals (29 of 80 studies), the share of the number of observations was far lower than that for non-journals, since a few studies published in non-journal outlets included large numbers of observations.

3.2. Meta-Regression Results

Breusch–Pagan tests showed concern with heteroscedasticity, so robust standard errors were estimated. The Ramsey specification tests did not reveal the existence of missing moderator variables at the 1% confidence level for any model [55]. Cook’s distance did not reveal observations that were outliers or had high leverage in any of the models. Multicollinearity was not a concern, as none of the variance inflation factors were above five [56]. The results from the overall value-per-day model and the three separate sub-sample models are presented in

Table 4. Regression results of mixed-effect full and sub-sample models with ln value-per-day as dependent variable and robust standard errors (standard errors in parentheses).

	Full	Deer	Waterfowl	CVM
Constant	−9.438 (18.724)	−17.954 (23.296)	29.934 (43.368)	29.401 (22.845)
Elk	0.146 * (0.078)			0.152 * (0.079)
Moose	0.157 (0.226)			0.479 ** (0.197)
Waterfowl	−0.374 *** (0.044)			−0.377 *** (0.042)
Small game	−0.928 *** (0.108)			−0.994 *** (0.099)
Big game	−0.055 (0.11)			−0.069 (0.114)
Other game	−0.928 *** (0.211)			−0.568 *** (0.198)
Ln (income/capita)	0.503 *** (0.092)	0.512 *** (0.115)	0.494 *** (0.089)	0.504 *** (0.095)
Ln (person/km2)	−0.017 (0.015)	−0.027 (0.020)	0.001 (0.027)	−0.012 (0.015)
Canada	0.047 (0.312)	0.423 (0.537)	−0.896 *** (0.340)	0.018 (0.374)
Other country	−0.115 (0.359)	−1.214 ** (0.512)	0.273 (0.663)	0.277 (0.331)
Year	0.006 (0.009)	0.010 (0.112)	−0.014 (0.022)	−0.014 (0.011)
TCM	0.353 ** (0.188)	0.546 ** (0.271)	0.078 (0.142)
CE	1.106 *** (0.351)	2.559 *** (0.328)	NA
Hedonic	1.100 *** (0.188)	NA	NA
Journal	−0.342 (0.214)	−0.406 (0.516)	0.274 (0.244)	−0.235 (0.235)
Random effects
${\lambda }_{0.j}$	0.392	0.467	0.146	0.412
${\epsilon }_{i.j}$	0.125	0.113	0.104	0.113
Model statistics
N	588	307	125	549
Studies	80	36	23	46
McFadden’s R2	0.248	0.123	0.153	0.310
AIC/N	1.107	1.001	1.002	0.898
BIC/N	1.239	1.133	1.232	1.015

Significance: *** p < 0.01, ** p < 0.05, * p < 0.10.

.

All models were statistically superior to models without covariates, as measured by McFadden’s pseudo R² and information criteria, and they each contained statistically significant independent variables. McFadden’s pseudo R² was equal to or exceeded 0.2 for the Full and CVM models, which was very good [57]. The two normalized information criteria, AIC/N and BIC/N, were slightly better (i.e., lower values) for the sub-sample models.

The two models that included game animals as independent variables showed qualitatively similar results: the coefficients were positive for ‘Moose’ and ‘Elk’ and negative for ‘Waterfowl’, ‘Small game’, and ‘Other game’. This was consistent with the higher volume of meat from moose and elk and the absence of trophy value for waterfowl, small game, and other game. It also corroborated the findings of Engelmann et al. [58], who showed that WTP is about 100 times higher for a moose than for a small game animal, and that the WTP for a wild boar is about one sixth of the WTP for a moose among Swedish hunters.

Regarding contextual factors, results common to all models included the significant and positive coefficients on ‘Lngdpc’, consistent with the micro-economic theory. The coefficient of ‘Lngdpc’ is the elasticity of WTP with respect to income (the % increase in WTP from a 1% increase in income), which ranged between 0.49 and 0.51. WTP was lower for ‘Other countries’ compared with the USA in the Full and Deer models, and for ‘Canada’ in the Waterfowl sub-sample model. ‘Population density’ and ‘Year’ were not significant in any model.

With respect to study characteristics, valuation method had significant effects in the Full model and the Deer sub-sample model, where ‘TCM’ raised the value compared with studies using ‘CVM’. ‘Journal’ was not significant in any model.

3.3. Predictions

When comparing the predictions of the mean values and associated 95% confidence intervals from the models with the source data, the predictions were relatively similar and close to the mean source data values for all models (

Table 5. Data and predicted mean value per day and ranges within a 95% confidence interval (2020 USD), MPE, and normalized RMSE for different separate and combined sub-sample models.

	Data Mean	Pred. Mean (s.e.)	95% Confidence Interval:		MPE (%)	RMSE/Mean
			Low	High
Full	69.3	68.7 (1.36)	66.1	71.5	29.7	0.42
Deer	79.5	78.9 (1.33)	76.3	81.5	27.6	0.41
Waterfowl	50.5	49.7 (2.29)	45.1	54.2	25.5	0.32
CVM	66.6	66.2 (1.33)	63.6	68.8	28.3	0.38

).

The predicted means were in the range of 98–99% of the data means, and the data means were within the predicted 95% confidence intervals for all models. The MPE estimates were similar and ranged between 25.5 and 29.7, being lowest for the Waterfowl model. The normalized RMSE (RMSE/Mean) was also the lowest for the Waterfowl model. The normalized RMSE was below unity for all models, confirming the relatively high predictive ability of these models.

However, common to all models was the relatively large deviation of the predicted value from the source at relatively high values, above approximately USD 120 /day (Figure A2). There were a small number of relatively large source values, exceeding USD 150, in all models. These high values were uniformly underestimated by all models. As with the full model, the underestimation of large source values for the Deer and CVM models was evident. The Waterfowl model was less prone to this effect.

4. Discussion and Conclusions

The purpose of this paper was threefold: to summarize the results of the global hunters’ valuation literature, to assess the determinants of hunting values, and to estimate the predictive power of MRA models. To this end, 588 usable observations were obtained from 80 studies. We found that there was a high concentration of studies in the USA, deer was the most studied animal, and contingent valuation was the most common valuation method.

The mean values of the different subsets of hunters’ valuations could be compared with other MRA studies, sometimes by further division of the subsets. Huber et al. [22] estimated a mean value of USD 82 per day (in 2020 prices) for all types of hunting in the USA. Our USA estimate was USD 77 /day, so the Huber et al. [22] estimate was within the 90% confidence interval of our average estimate. Kerr and Woods’ [7] mean estimate for global per-day values for deer was USD 150, which was high compared to the estimate in the present study of USD 80.

With respect to the determinants of the value of hunting, the explanatory power of different hunting attributes (targeted game animal); context variables (income measured as GDP/capita, population density, year of study, region of application); and study characteristics (valuation method, publication outlet) was estimated. Regarding hunting attributes, the targeted game animal had significant effects in all models that included these variables. Elk and moose had higher, and waterfowl and small game lower, values than the deer reference case. The relatively high values for hunting large animals were consistent with their higher volume of meat and the absence of trophy value for the other game animals. This finding supported the results of other valuation studies that estimated hunters’ valuation of different game animals [59].

Regarding the effects of contextual variables, common results for all models were the positive impact of regional/national income on the value of hunting and the non-significant effects of the timing of the study and the region of application. An income variable was not included in the MRAs by Kerr and Woods [7] or Huber et al. [22], precluding comparison. However, our result was consistent with results from studies on the recreational value of nature in general that showed a positive income effect [20]. The estimated differences in income elasticities and the variation in average values between the models implied that the effects of a given change in income would differ depending on the targeted game and the valuation method. An increase in income by, say, 20% would raise the average value per day by USD 6.97, but this would range between USD 4.99 and 8.14 depending on which game animal was targeted.

The results showed that the valuation method had significant effects—the values estimated by the CVM method were lower than those of other methods. This supported the results obtained in several MRA studies on recreational ecosystem values [17,59,60]. Contrary to the expectations expressed in other studies (e.g., [61]), the CVM may not provide higher values than methods based on revealed preferences.

Although the mixed-effect model accounted for clustered variance in the studies, other clustering related to regions could exist [62]. Therefore, clustered variance tests were carried out for the different regions (USA, Canada, and other regions) instead of their representation as dummy variables. The results showed minor differences between the two approaches (Table S2 in Supplementary Materials).

A potentially important determinant of hunt value that this study could not address is unknown hunt-specific attributes, including the number of days hunting (whether for the specific hunt or the season); the abundance of animals; regulations; the ease of access, congestion; scenic amenity; and the availability of substitutes (e.g., other locations, species, or activities). Evidence suggests that, at least up to a point, longer trips are valued higher than shorter ones, and hunters value locations with higher scenic amenity, more abundant game, fewer hunters, and better prospects for obtaining a trophy [63,64]. The inability to include these items in our MRA should temper expectations about its predictive precision.

Nevertheless, all sub-sample models had good predictive ability, with a mean percentage error (MPE) below 30%. The relatively high predictive power of all the models supported benefit transfer to regions other than the source study regions, a topic which has been discussed extensively in the literature on environmental valuation [8]. This could be of particular use for hunting values because of the many countries for which valuation source studies have not been undertaken. Significant levels of hunting participation in those places [65], combined with the magnitude of values estimated here, indicates the importance of a wider consideration of hunting-related values in environmental decision making. However, monetizing hunting ecosystem service values through primary studies can be expensive, time-consuming, and reliant on the availability of valuation expertise. Benefit transfer offers rapid and inexpensive indicators of value.

While the transfer of values to new contexts can be cause for skepticism, our results suggest that cross-context transfers can provide valuable guidance. The determinants of the value of recreational hunting are similar, irrespective of location, although there remain significant locational differences in value magnitudes. Our MRA suggested that the inter-regional transfer of values is justified when the income per capita, type of game animal, and valuation method are included in the transfer model. This could be of high relevance for regions with many hunters but few valuation studies. For example, there are approximately 6 million hunters in Europe [2], but only a few valuation studies have been conducted in this area.

In practice, managing wildlife presents significant challenges because of the diversity of ecosystem services it affects and because the human dimensions of wildlife management are extremely diverse and complex [66]. The value of recreational hunting constitutes one of several factors involved in wildlife management, such as its environmental, cultural, safety, social, and agricultural impacts. The results of the present study facilitate the benefit transfer of hunters’ valuations to regions without source studies. This contributes to decision making by facilitating the understanding and comparison of the total value of recreational hunting, which also includes the effects on the prosperity of local and regional economies due to, among other things, sales of licenses and equipment. The systematic quantification of the total value of hunting at the international scale is an important field for future research that needs to meet the challenges of a lack of consistent data on hunting activities and the associated dispersal effects.