An Assessment of the Impacts of Forest Management on Aboriginal Hunters: Evidence from Stated and Revealed Preference Data Wiktor Adamowicz, Peter Boxall, Michel Haener, Yaoqi Zhang, Donna Dosman, and Juanita Marois ABSTRACT. Assessing the impacts of forest harvesting activities on Aboriginal people and incorporating these considerations into forest management plans is one of the chal- lenges facing Canadian forest managers. In this study, we model hunting behavior using stated and revealed preference data on subsistence use of wildlife resources. We use this framework to assess the impacts of forest management changes on Aboriginal people in northwestern Saskatchewan. Innovative approaches to data collection are employed to address challenges in obtaining data in these contexts. The econometric analysis combines the stated and revealed preference information to account for limitations in the revealed preference data. Monetary measures of welfare are examined, but we also assess resource compensation and zoning as mechanisms for addressing the impact of forest harvesting on subsistence wildlife use. The results also demonstrate the use of geographic informa- tion system information in linking forest management and Aboriginal resource use. FOR. SCI. 50(2):139–152. Key Words: Aboriginal forest use, nonmarket valuation, resource compensation. M OST FOREST MANAGEMENT ACTIVITY in Canada takes place on public land. Forest managers op- erating on public lands are required to take into account nontimber values in addition to timber values when making management plans. One of the major beneficiaries of nontimber values are Aboriginal people (Haener and Adamowicz 2000). The majority of Canada's Aboriginal people live in the forest, and forest resources are culturally Wiktor Adamowicz, Professor, Department of Rural Economy, University of Alberta, Edmonton, AB, T6G 2H1 Canada—Phone: (780) 492-4603; vic.adamowicz@ualberta.ca. Peter Boxall, Associate Professor, Department of Rural Economy, University of Alberta, Edmonton, AB, T6G 2H1 Canada—Phone: (780) 492-5694; peter.boxall@ualberta.ca. Michel Haener, Senior Policy Advisor, Ministry of Aboriginal Affairs, Government of the Northwest Territories, PO Box 1320, Yellowknife NT X1A 2L9 Canada—Phone: (867) 873-7391; michel_haener@gov.nt.ca. Yaoqi Zhang, Post-doctoral Fellow, Department of Rural Economy, University of Alberta, Edmonton, AB, T6G 2H1 Canada; yaoqi.zhang@ualberta.ca. Donna Dosman, Research Associate, Department of Human Ecology, University of Alberta, Edmonton, AB, T6G 2N1 Canada—Phone: (780) 492-3012; ddosman@ualberta.ca. Juanita Marois, Lecturer, Faculty of Physical Education and Recreation, University of Alberta, Edmonton, AB, T6G 2H9 Canada—Phone: (780) 492-5702; juanita.marois@ualberta.ca. Acknowledgments: This research was supported by Mistik Management, the Social Sciences Humanities Research Council, Natural Sciences and Engineering Research Council, and the Canadian Forest Service (SSHRC-NSERC-CFS Research Partnerships Program). The research team also thanks the communities of Green Lake, Waterhen, Canoe Narrows, Jans Bay, Cole Bay, Beauval, and Dillon for their support and partnership in this research program. W. Adamowicz was a Gilbert White Visiting Fellow at Resources for the Future, Washington, DC, when this article was prepared. Manuscript received September 27, 2002, accepted July 26, 2003. Copyright ? 2004 by the Society of American Foresters Forest Science 50(2) 2004 139 as well as materially important for them (Tobias and Kay 1993, Usher 1976, Beckley and Hirsch 1997). Aboriginal people also account for a substantial percentage of northern Canada's population, and their population growth rates are significantly higher than the rest of the Canadian population (DIAND 2001). Policy makers have begun to recognize that the values of Aboriginal people should be incorporated into management planning and forestry activity and are increas- ingly requiring forest managers to consult with Aboriginal people before embarking on management plans. It is not clear, however, how the values of Aboriginal people can be incorporated into forest management plan- ning. Conceptually, these values could be incorporated by adjusting the optimal timber rotation and harvesting plan for nontimber values (Bowes and Krutilla 1985; Mendelsohn 1996). This would involve measuring the values associated with nontimber goods at different forest ages and modifying rotation ages and harvesting plans to account for these values (e.g., Englin 1990). In practice, difficulties arise when collecting general harvest information given there are many nontimber goods to consider, and the spatial nature of the values must be accounted for to be useful from a forest management perspective. A further difficulty is that infor- mation about the use of nontimber goods is privately held by Aboriginal people, and this information on wildlife har- vesting, for example, will not be freely released because of its value to hunters and its sensitive nature. Information may also be withheld because of the cultural importance of certain sites or activities. Aboriginal protests against for- estry activity may also result in difficulties in gathering information on nontimber harvest activities. Given the cul- tural significance of these resources, and that different reg- ulations apply to Aboriginal harvesters, information from recreational hunting licenses and related wildlife harvest statistics that are routinely gathered from non-Aboriginal hunters are not relevant for most Aboriginal hunters. These issues suggest that data gathering techniques used in typical studies of recreational hunting demand are not useful in an Aboriginal hunting context. In addition to obtaining information on activity levels or use of nontimber products, incorporation into economic models typically requires assessment in monetary terms, which introduces further difficulties. First, monetary valu- ation may imply property rights that are rejected by the Aboriginal people. Property rights issues surrounding Ab- original peoples' access and rights to forest resources re- main controversial. A contingent valuation approach that requests a willingness to pay for improved quality of non- timber goods would likely be rejected as it implies that the Aboriginal people do not have the rights to the resource. An approach that involved an offer of compensation would also likely be rejected, in part because of the usual difficulties with compensation-based contingent valuation but also be- cause of the lesser importance of monetary transactions for some segments of the Aboriginal population. In cases where monetary values have been required for compensation, re- placement cost methods have been used (e.g., Usher 1976). Even though these methods have been criticized (Beckley and Hirsch 1997, Brown and Burch 1992, see also Haener et al. 2001b), they continue to be popular in applied work. In this article, we view the impact of forest management ac- tivity as broader than simply a reduced harvest of wild game. In fact, forestry activities may result in increased game populations in some cases, yet they may reduce the welfare of the hunters. Therefore, we choose to model actual behavior and focus on the behavioral changes and implied value changes arising from changes in the forest environment. A behavioral approach also allows us to ex- amine nonmonetary as well as monetary measures of compensation. The challenges described above can be summarized as data collection challenges and valuation challenges. In this study, we describe an integrated approach that attempts to address both issues. The study is carried out in a Forest Management Area (FMA) in Northern Saskatchewan. Mis- tik Management, a not-for-profit corporation, manages this area. Mistik's management's objectives are to manage the 3.3 million-ha area to provide fiber for a pulp mill and sawmill located near Meadow Lake, Saskatchewan. The company is to do this without compromising the nontimber resources flowing from the forest region. A unique aspect of this situation is the fact that a single entity manages a large landscape with the objective to supply mills with different fiber requirements. Furthermore, the Meadow Lake Tribal Council[1] owns the NorSask sawmill to which Mistik Management supplies fiber. Thus, concerns regarding the flow of nontimber resources to Aboriginal people are inte- gral to Mistik's land management approach. The firm has also put considerable effort into developing co-management boards with Aboriginal communities in the region and pro- viding mechanisms for Aboriginal people to benefit from the employment opportunities created in the region. This article begins with a description of a data collection effort that focused on building trust and developing the research program in cooperation with the communities. This unique data collection approach, we believe, is likely the only way to collect data of the type required for assessment of Aboriginal nontimber values.[2] Three types of data were collected. (1) Information on "special sites" identified by members of the Aboriginal community was collected. The community members expected the researchers to pass in- formation on these sites on to the management agency so that action could be taken to avoid harvesting in or near these sites, though no cessation of forestry activity was promised by the researchers or Mistik. This information included identification of areas with high moose popula- tions, cabins, salt licks, and other attributes of the land. This "traditional ecological knowledge" of the hunters is an important component of our modeling strategy. (2) Infor- mation on actual hunting activities was collected from Ab- original hunters. The collection of this revealed preference (RP) information is described below. (3) Stated preference (SP) information was collected to better identify preferences for attributes of wildlife harvesting sites. In addition to the data gathered from the Aboriginal people, data on forest 140 Forest Science 50(2) 2004 characteristics were obtained from geographical informa- tion system (GIS) data provided by Mistik Management. After describing the data collection process, the article continues with a description of the modeling process (com- bining RP and SP data) and the simulation of hunting behavior following forest harvesting plans. The impacts of forest harvesting on hunting are examined using monetary welfare measures as well as a form of resource compensa- tion. In addition, an alternative forest management strategy that involves concentrated forest harvesting (or zoning) is explored. The results suggest that modeling actual behavior can be used as a method of capturing some of the value of non- timber resources accruing to Aboriginal people. These methods also provide significant insight into the impact of forest harvesting on nontimber activities. However, substan- tial investments in data collection are required to obtain data for such analysis. Furthermore, challenges in identifying the opportunity cost of time make the monetary valuation of nontimber resources difficult. Use of resource compensation methods, however, appears to be promising. The methods pre- sented in this article provide insights into the potential for resource compensation as a practical way to incorporate the values of Aboriginal people into forest management. The analysis of zon- ing, or concentrated forest harvesting versus dispersed harvesting, also suggests potential for this as an approach that incorporates Aboriginal values into forest management. The article concludes with a discussion of some of the limitations of the study and challenges associated with incorporating Aboriginal values into forest management planning. Data Collection Study Area and Sample The study area for this project is the Millar Western- NorSask FMA area in northwestern Saskatchewan, which extends along the Alberta-Saskatchewan border, comprising 3.3 million ha of land (see Figure 1). The current population of the FMA area is about 25,000 spread over about 22 communities in and around the FMA area. This population includes people of Cree, Dene, Me ?tis,[3] and European descent. Although NorSask utilizes the softwood and Millar Western utilizes the hardwood in the region, the landscape planning for the region as a whole is undertaken by Mistik Management Ltd. Mistik's mandate is "to provide the mills with a long-term sustainable wood supply while taking into account the many resources and uses of the forest" (Mistik Management Information Booklet). Data Collection Process Initial contacts with the communities and the forest man- agement agency identified that even though Aboriginal peo- ple are engaged in collection and use of many different nontimber forest products, large game harvesting (in partic- ular moose; deer and caribou to a lesser extent) was con- sidered the most important. The community members them- selves were interested in participating in a study of hunting as this activity is important culturally as food for hunters and other members of the community, and as an activity that could be significantly affected by forest management ac- tions. Thus, the initial contacts with the communities to obtain permission to conduct research also helped to focus Figure 1. Maps of Canada and of Alberta and Saskatchewan showing the locations of the NorSask Forest Management Agreement. Forest Science 50(2) 2004 141 the research question on a topic of interest to the commu- nities. An approach that engages the community in defining the research program is emerging as a requirement in con- ducting research with Aboriginal communities. The process by which the data were collected differed from previous hunting research in the economic literature. Data were collected in informal in-person interviews that were arranged by a community contact person who was a resident in each community. While the design of the survey was to be a straightforward question and answer session, the interview that evolved was more of a conversation allowing "story telling." The interviews took approximately 40–120 min and averaged just over 1 hour. Further information on the sample and the interview process can be found in Dosman et al. (2002). Trust was an important factor in the entire process. Community members were apprehensive about discussing hunting and trapping activities with a stranger, especially a non-Aboriginal person. In response, we employed one pri- mary interviewer who became known in the communities and developed relationships with the local people. We also employed a community resident who socially and culturally had access to the hunters. This person helped to arrange interviews, attended them to ease the participants, and trans- lated some unfamiliar concepts. The interviewer lived in the communities for approximately 1 year, which facilitated the development of trust between the interviewer and commu- nity members. The fact that the interviewer lived and par- ticipated in the community was fundamental to the research process. Reciprocity also played a vital role in securing relation- ships with the respondents. Reciprocity for participation in the study was offered at several levels. First, the research group made a commitment to report back the findings to each of the communities. Each hunter was asked during the mapping exercise to identify special sites such as nesting areas, calving areas, burial sites, cabins, or historical sites. A map of special sites was created for each community and presented to community leaders. Second, an incentive of a draw in each community for a gift certificate at a local hunting store was offered to all respondents. Individual respondents responded favorably to the incentive. Third, in interviews with First Nations elders an offering of tobacco was made at the beginning of the interview. This offering is a sign of respect and helps to formalize the relationship between the interviewer and the elder. During the period Oct. 1999 to Sept. 2000, 124 inter- views with Me ?tis and First Nations hunters were conducted in seven communities (Green Lake, Waterhen, Canoe Nar- rows, Jans Bay, Cole Bay, Beauval, and Dillon) represent- ing five co-management areas. Slightly more Me ?tis (59%) hunters were interviewed than First Nations (41%) hunters. In addition, an attempt was made to capture hunters from both northern (37%) and southern (63%) communities, to facilitate investigating the influence of better access to larger commercial centers on harvesting behavior. Revealed Preference Information Hunting trip information for the past hunting season was collected in two complementary formats. The first was a traditional trip log, which recorded the location and fre- quency of each trip. It also included information on the approximate distance traveled, the modes of transportation, with whom the trip was taken, the duration of the trip, the season of the trip, and the number of moose and other game harvested by the individual being interviewed and by the group. The second format was a map of the NorSask FMA area on which the hunters drew their general hunting area and the location of the trips recorded in the trip log.[4] We defined the general hunting area as the entire area in which an individual hunter would consider going to hunt moose. Both the map and the trip log were used simultaneously in collecting information. Some of the more traditional hunters would require significant interpretation of the map because they were not accustomed to seeing two-dimen- sional representations of the land base. Once this was ac- complished, the use of a map as a visual tool worked to make the respondents feel more at ease with the process. Many informants preferred to talk in stories and they would point out the sites and then begin to remember the rest of the trip details that were needed to complete the trip log. Information provided on each respondent's map was transferred into digital form using ArcView. Mistik Man- agement provided digital files for the region including lakes, rivers, roads, trails, FMA, and other planning unit boundaries. This information was used as geographic refer- ences for developing general hunting area and hunting trip locations. Although individual trips were digitized, to en- sure confidentiality, the data were aggregated for each com- munity. Aggregating the individual general hunting areas by community also helped determine the geographical extent of hunting trips for each community. General hunting areas for several communities overlapped, but for the most part they followed the boundaries of Fur Conservation Areas, which are based on traditional trapping areas. Special Sites Map/Information Respondents were also asked to mark "special sites" such as salt licks, cabins, areas of exceptional moose hab- itat, burial sites, and avian nesting areas on the maps.[5] These maps are similar to those generated in traditional land-use mapping exercises and reflect the traditional eco- logical knowledge of the hunters (MacKinnon et al. 1999, Pyc 1999). This information provided a spatial record of characteristics that may be important in explaining hunter site choice. We employ some of this information in devel- oping measures of the attributes of hunting sites. Stated Preference Information We extend existing studies on Aboriginal hunting (e.g., Winterhalder 1983 and 2001, Feit 1987) by incorporating stated preference methods into our survey. We used a choice experiment approach to investigate how Aboriginal people in the region select where they hunt and how their behavior 142 Forest Science 50(2) 2004 might change in response to changes in moose, forestry, costs, and other factors. The design of the choice experiment began with a list of hunting site characteristics used in earlier choice experi- ments designed to capture preferences of southern hunters for hunting sites in central and northern Saskatchewan. This list of potential hunting site characteristics was presented to and discussed with a group of six Aboriginal hunters from four communities in the area. The appropriateness of each attribute was discussed, and culturally appropriate levels were determined. This set of attributes was further vetted through an elder hunter who after some discussion approved the list. From these approved attributes a choice experiment was designed. The attributes themselves did not differ radically from earlier designs used in studying non-Aboriginal hunters (e.g., Adamowicz et al. 1997, Boxall and Macnab 2000). However, the levels of the attributes did (see Table 1). In particular, the levels for the distance traveled and the mode of transportation attributes while hunting at the site differed from earlier surveys reflecting the fact that Aboriginal hunt- ers live in their hunting regions and that past cultural prac- tices influence the mode of transportation for some of them. The prototype choice experiment was initially text based, similar to choice experiments used in mail surveys of li- censed hunters in the south (e.g., see Boxall and Macnab 2000). One elder informant, who was conversant in English but had difficulty reading, found that reading the survey and discerning the specifics of the choice experiments was dif- ficult. It was decided that an illustrative approach would be more appropriate in this setting. Photographs were used for attributes for which a picture would easily illustrate its meaning, such as time since harvest and access to the hunting site. More detail on the choice experiment can be found in Dosman et al. (2002) and Haener et al. (2001b). Socio-Demographic and Cultural Data In addition to the information related to actual hunting behavior and the responses to the choice experiment, we also collected socio-demographic data including age, com- munity born, gender, education, marital status, number of children, employment status, partner's employment status, and Aboriginal status. During the development of the sur- vey tool, we were informed that it would be culturally inappropriate for us to ask respondents to report their annual income. We thought annual income could be imputed from the respondent's employment status and industry in which they worked. However, it became evident during the inter- views that over the course of a year many respondents were employed in a series of jobs that last for a few days to several weeks or months in many different industries rang- ing from forest firefighting to road construction to forestry work. A much more detailed employment record for the year would be needed to be able to impute individual income levels. These features have significant implications for the ac- curacy of value of time calculations derived from travel distances using standard economic approaches (i.e., travel cost models). Because many individuals in our sample changed their employment-related activities over the year, we use the average male income for the region in our computation of a wage rate. We attained average male income levels from the national Aboriginal census (Statis- tics Canada 1998).[6] Methods Descriptions of the behavior of boreal Aboriginal hunters in the anthropological literature suggest that they have con- siderable knowledge of moose biology and that their hunt- ing behavior represents decisions that optimally provide opportunities for harvest. For example, Winterhalder (1983) describes frequent use by hunters of areas in proximity to water and forest areas recently disturbed by fire, and that they adjusted their hunting behavior seasonally to match changes in habitat use by moose. These features correlate well with biologists' analysis of preferred moose habitats (e.g., Saskatchewan Forest Habitat Project 1991). Feit Table 1. Definition of hunting site attributes for the choice experiment administered to the Aboriginal hunters. Attributes Levels How far hunting site is from home 10 km 50 km 100 km 200 km How many people you see at the hunting site Nobody else, except others in my hunting party Other people How many signs of moose you will see each day Signs of less than 1 moose per day Signs of 1 to 2 moose per day Signs of 3 moose per day Signs of more than 4 moose per day How hunters travel while at the site On foot without trails or cutlines By ATVs on old logging roads By 4-wheel drive on new logging roads By boat through interconnected lakes How long it has been since the site was harvested Site just harvested Site logged 3 to 5 years Site logged 10 to 15 years No evidence of logging Forest Science 50(2) 2004 143 (1987) suggested that Cree hunters use indicators of moose populations to guide hunting decisions. This information suggests that models of hunting site choice by Aboriginal hunters should incorporate such indicators and their poten- tial change as a result of landscape alteration through timber harvests. Our modeling framework is illustrated in Figure 2. We employ hunter knowledge and forest characteristics to de- velop a model of moose population or abundance. This model provides measures of one of the most important attributes of the sites. In addition to information on moose populations, forest landscape and road network characteris- tics are directly used as explanatory variables in a RP model. We then employ a combination of RP and SP data to generate a joint model of hunter preferences. Each of these components will be outlined below. Spatial Resolution Because our objective was to develop a hunting site choice model that could be used to simulate the effect of landscape changes on behavior, a spatial framework must be chosen that incorporates landscape and hunting attributes as well as trip behavior. For this spatial scale, the operating area (OA) was selected as the unit of analysis. Mistik Management considers the OA as the smallest spatial unit used to plan forest harvest operations in the FMA. Hunters in our sample took trips to most of the 450 OAs in the FMA, as well as some outside the FMA. From the digital files provided by Mistik Management, the following variables for each OA were developed: lake area (ha), length of rivers (km), length of road (km) by class of road (1–8), and size of the OA (ha). For the OAs in the FMA, the following forest landscape characteristics were available: crown closure class (four classes ranging from open to closed based on percentage cover), age class, the area of recent (5 years or less) and older (?5 years) timber harvests, the area burned in forest fires, nonforest area (e.g., muskeg), productive forest area, and the area not subjected to previous timber harvest operations. Moose Population/Abundance Model One important landscape attribute that was not available was the abundance and availability of moose. Usually aerial survey transects are used by biologists to estimate moose abundance, but these had not been completed in the study region. To overcome this gap in the data, and following the research by Feit (1987), information provided by the hunt- ers was used to develop a moose abundance indicator for each OA. To create this indicator, OAs were identified in which respondents indicated there were exceptional areas for moose. The information on moose populations and for- est landscape attributes were used to construct a model of moose populations. Given the discrete nature of this vari- able (exceptional habitat or not), a logit model was used to estimate the probability that any OA in the FMA had exceptional moose habitat. Landscape features related to moose habitat preferences discussed in the biological literature were included in the model (e.g., variables used to develop moose habitat suit- ability indices in the Saskatchewan Forest Habitat Project (1991)). These included the density of rivers, areas of dis- turbance from fire or previous timber harvests, the area of standing water, the area of muskeg, and the area of forest classified as relatively open (crown closure 0–25%). Other variables were chosen that were related to human use and perception such as the number of cabins within 10 km of an OA and the presence of a salt lick within 10 km of the OA. The independent variables in the model are reported in Table 2.[7] The influential explanatory variables were re- cent anthropogenic disturbance (new cuts), the presence of salt licks, and those related to aquatic habitat (muskeg and water). The crown closure and river density variables were significant at the 10% level. This model was used to esti- mate the probability that each OA in the FMA would contain exceptional moose habitat. Choice Model Development The interest in this article is in developing a model to explain why hunters visit certain OAs over others and how their choices might be affected by timber harvesting. This information represents discrete choice data, which can be analyzed using econometric methods based on random util- ity theory (Louviere et al. 2000). This theory maintains that the utility an individual derives from visiting an alternative site, i, is considered to be associated with the attributes of that alternative. This utility function (Ui) can be represented as Ui ? Vi ? ?i, where Vi signifies a deterministic compo- nent and ?i an unobservable or stochastic component. Vi can be characterized by its attributes. Thus, Vi ? ?kXi where Xi is a vector of k attributes associated with alternative i and ?k is a vector of parameters or taste weights. If the distribution of the stochastic components is characterized as IID Gum- bel, the conditional probability of selecting alternative i from a set, C, of alternative sites is: Figure 2. A summary of the strategy used in modeling Aborig- inal hunting trips in the NorSask FMA. 144 Forest Science 50(2) 2004 Prob(i) ? exp(??kXi) ? jeC exp(??kXj) (1) where ? is a scale parameter and C is the choice set. When a single set of data is used to estimate a model, ? is confounded with the parameter vector and cannot be identified. However, in models in which multiple data sources are merged to estimate the parameter vector, the scale of one data set can be estimated relative to the other (Louviere et al. 2000). We anticipated that the RP attribute data would not be sufficient to capture the preferences of the hunters. In part this arises because data for some important attributes were not available (e.g., moose populations). In addition, the RP attributes are likely correlated and confound effects. For example, the effect of forest harvesting on the esthetics and appearance of a site would be confounded with the impact on moose populations. Finally, it appears that joint SP-RP models can outperform RP models in predicting actual behavior when the SP data are carefully collected (Haener et al. 2001a). Therefore, both data types are employed in modeling the trip locations of Aboriginal hunters. Because the SP data were generated from a controlled design, the number of alternatives in the choice set, C, was 3 and the attribute levels (Table 1) were predetermined. However, this is not the case for the RP data. For the RP data, the choice set for each community determined by the survey was different, ranging from 30 to 207 OAs. To reduce computational burden, each community's choice set size was reduced to 30 OAs by randomly selecting a subset of the relevant OAs for each trip from the full choice set. This procedure has been shown to produce parameters that are not significantly different from those derived using the full choice set (e.g., Parsons and Kealy 1992). The landscape attributes used in the models and their coding are described in Table 3. Because we combined the RP and SP data in a joint model, several variables from the RP and SP data were transformed so that their coding was commensurate. Note that there is not complete overlap between the RP and SP data series. The encounters variable, for example, exists only in the SP data. The variables common to both models are travel cost, moose abundance, two variables for time since forest harvest, and water access. The joint model combines data for the attributes common to Table 2. Parameter signs and significance levels for a logit model used to estimate the probability that an OA contains high moose populations. Variable Description Parameter sign P-value Constant – 0.003 Crown A Percent forest area in the OA with crown closure 0–25% density – 0.058 Salt licks A dummy variable indicating presence of salt lick(s) within the OA ? 0.000 Water Percent OA area covered by water ? 0.021 Muskeg Percent OA area classified as muskeg ? 0.021 New cut Percent OA area classified as new cuts ? 0.000 River density Density of rivers in the OA ? 0.064 Table 3. A description of variables used to estimate choice models using RP and SP information from Aboriginal hunters in the NorSask FMA. Common variables Coding RP data SP data Travel cost $00 Road distance was transformed to a travel cost as follows: [2*(RD ? 3*NRD)*$0.589] ? [1/90*1/3*(income/2040)*2*(RD ? 3*NRD)]a Moose abundance 0–1 Based on predicted probability from the OA moose habitat model in Table 2 0.05, signs of less than 1 moose/day 0.375, signs of 1–2 moose/day 0.750, signs of 3 moose/day 1.000, signs of more than 4 moose/day New cut dummy 0, 1 1 ? new cuts area ? old cut area and ?90% of area uncut; 0 otherwise Site harvested 5 or fewer years ago Old cut dummy 0, 1 1 ? old cuts area ? new cut area and ?90% of area uncut; 0 otherwise Site harvested 10–15 years ago No cut Base ?90% of the area is uncut No evidence of timber harvesting Water access dummy 0, 1 1 ? a river intersects the OA or more than 1% of the OA is covered by a lake; 0 otherwise OA is accessible by water Cabins 0–5 Number of cabins in or within 5 km of the OA NA No hunt constant NA 1 ? stay at home instead of hunting Encounters 1, ?1 NA 1 ? other people are encountered New access 1, 0 1 ? access via new logging roads NA Old access 1, 0 1 ? access via old logging roads NA a RD refers to road distance, NRD refers to nonroad distance, and income represents the average male income for each community reported in the census. Forest Science 50(2) 2004 145 both sets of data, while still allowing the coefficients of the attributes unique to the RP and SP data to be estimated. The model we use for simulation is based only on the attributes common to both models and one variable unique to the RP data (i.e., cabins). Because the SP data are based on an orthogonal design, we can employ a subset of the attributes in estimation and simulation without significant concern about specification error. The combined variables and their resulting codes are listed in rows 2–7 of Table 3. The remaining rows show the variables that are unique to the RP and SP data sets. In modeling hunting site choice, travel cost is commonly used as proxy for the cost of visiting areas (e.g., Adamowicz et al. 1997; Boxall and Macnab 2000). Following the stan- dard procedures in the travel cost literature, it was assumed that travel cost is a function of out-of-pocket expenses related to travel distance and the time costs of traveling (e.g., Boxall et al. 1996). Travel distance was included in the design of the SP choice experiment. However, it was necessary to calculate travel distances for the RP data. To estimate these distances, the shortest road distance between each community and the centroid of each OA in the com- munity general hunting area was determined using the GIS. In several parts of the FMA, the road network is sparse, therefore travel distances include the distance by road and the remaining "nonroad" distance required to reach the OA. The nonroad distance also serves as an indicator of the remoteness of the OA. The operating cost of vehicle use associated with the road distance was estimated at $0.589/km (reported by the Canadian Automobile Associ- ation). The operating costs associated with the nonroad portion of the distance was assumed to be three times this figure, as supported by information suggesting that the fuel mileage of off-road vehicles and snowmobiles is about one-third that of a car or truck.[8] The standard means of incorporating the value of travel time was also used. We assumed an average speed of 90 kph and use one-third the estimated wage rate for each commu- nity. The wage rate was determined by dividing the average male income in each community by the total number of work hours in a year (assuming a 40-hour workweek). We recognize that this method of valuing time may be inappro- priate for this context and suggest that tradeoffs associated with pursuing subsistence activities need to be further investigated. Results Choice Model Parameters The parameter estimates for the joint RP-SP model and the corresponding RP and SP models are reported in Table 4. The RP model only has parameters for those attributes that could be related to the spatial information provided by the hunters and Mistik Management. For the SP model, parameters displayed are only for those attributes used in the choice experiment. The joint model contains parameters for all of the characteristics in the RP and SP models, but recall that these parameters are constrained to be equal across the two sets of data. The measure of goodness of fit (?2 ) is highest for the SP model and lowest for the RP model but indicates reasonable fit for models with this number of alternatives. Because the joint model provides the most complete information on attributes, it is discussed in detail below. However, it should be noted that the signs of those parameters identified as statistically significant are consis- tent across the three models but for one exception.[9] The dummy variable on new cuts (recent harvests) in the RP model is positive and significant, likely a result of con- founding between harvest, access, and moose populations. In the joint model, the travel cost and the encounter parameters are negative and significant. This suggests that the hunters prefer to hunt closer to their communities and that they would prefer fewer encounters with other hunters while hunting. The cabins, water access, and the moose abundance indicator variables are positive and significant. These findings suggest that the hunters prefer to hunt in OAs with or near cabins, that have good water access, Table 4. Parameter estimates for the RP, SP, and joint RP-SP models. Variables RP model SP model Joint model Parameter t statistic Parameter t statistic Parameter t statistic Travel cost –0.1753* –3.357 –0.6721* 11.32 –0.6868* 10.563 Moose probability 1.5336* 7.872 0.8295* 5.788 0.9821* 6.536 New cut dummy 0.1703* 2.031 –1.2783* 11.05 –1.1131* 8.113 Old cut dummy –0.0756 0.232 –0.5055* 3.577 –0.4032* 2.669 Water access dummy 0.2177* 2.482 0.4703* 3.449 0.5395* 3.866 Cabins 0.3154* 11.505 1.7535* 3.798 No hunting dummy –2.0024* 11.048 –1.7958* 9.232 Encounters –0.2464* 4.724 –0.2442* 4.389 New access 0.1159 0.84 0.1568 1.068 Old access 0.0868 0.628 0.1089 0.741 Ln(?) –1.5619* 6.500 Log likelihood –3,124.5 –826.3 –3,989.2 ?2 0.04 0.23 0.08 LL sum (RP?SP) –3,950.8 * Parameter is significant at the 5% level. 146 Forest Science 50(2) 2004 and that contain high moose numbers. Note, however, that the access variables, while positive, are statistically insignificant. The parameters on the timber harvest variables in the joint model suggest a preference pattern where newly har- vested areas are avoided, while those with older are slightly preferred to new harvest, but not as desirable as areas with no harvests (the base). Because moose abundance is also related to forest disturbance patterns (see the new cut vari- able in Table 2), the choice model parameters suggest a complex pattern of preferences for forest age and distur- bance. It appears that there are amenity effects for the forest condition independent of the forest effects on moose abun- dance. Thus, recent timber harvests have a negative effect on utility through the new cut dummy in the joint model, but have a positive effect on utility through their beneficial impacts on moose abundance. Using the Model to Simulate the Effects of Forest Landscape Changes An advantage of using choice models to examine pref- erences over attributes is that the model can be used to examine changes in choice behavior when attributes of one or more alternatives in the choice set change. This can be done in a probabilistic framework using Equation 1 above. In addition, given that a cost variable is included in the model, monetary measures of economic welfare can also be associated with these changes in attributes (see Hanemann 1982). These features were used to simulate the effects of two different timber harvesting plans on the distribution of hunting trips in the general hunting areas of two communi- ties in the study and the associated economic impacts.[10] Once these effects are understood, we then consider whether actions to improve moose populations in certain areas would "compensate" for the effects of harvesting. Simulation of Timber Harvesting To determine the influence of timber harvesting on hunt- ing behavior, two harvesting scenarios for two communities were imposed on the current distribution of trips in the relevant general hunting areas. The first scenario uses Mis- tik Management's harvesting plan (see the dispersed plan in Figure 3). According to this plan, about 3,000 ha of forest, distributed in 19 OAs, will be harvested in the general hunting area for community 1, and about 6,000 ha of forest, distributed in 20 OAs, will be harvested in community 2. A strategy that has been receiving significant attention in forest management recently is the zoning of land areas and the concentration of activities within zones. Rather than attempting to practice sustainable forest management "ev- erywhere" on a landscape, foresters, for example, could emphasize timber production in one zone, leave another zone for wildlife and landscape protection, and manage using multiple-use principles in a third. This zoning strategy is often referred to as the TRIAD (a three zone strategy involving an-intensive forest management zone, a protected area, and a multiple-use management zone). It corresponds to the notion of increasing management effort and capital investment in those regions best suited for forest manage- ment (Vincent and Binkely 1994). We consider an analo- gous strategy here; however, the zoning considered is a concentration of forest harvesting and implicitly identifying zones where hunting could be the main land-use strategy. Based on the zoning concept and considering the distri- bution of uncut forest in the OAs, an alternative forest harvesting plan involving a more spatially concentrated harvest is used for comparison with the dispersed harvesting plan. The 3,000 ha of dispersed harvesting planned for the general hunting area of community 1 is reallocated into 3 OAs, and the 6,000 ha of harvesting planned for the general hunting area of community 2 is reallocated into 4 OAs. If the total number of trips taken in each community in a year is held fixed, then following the timber harvests the distribution of trips across the OAs in each community's general hunting area will change. These changes result from the effects of harvesting on both moose habitat and hunter preferences through changes in the new cut variable in the moose abundance model (Table 2) and changes in the new cut dummy variable in the choice model (Table 4). Figure 3 shows those OAs that experience relatively significant changes in the predicted distribution of trips following the dispersed and concentrated timber harvests. The change in trips is measured by the percentage change in trips taken to the OA (i.e., (postharvest trip – preharvest trip)/preharvesting trips ? 100). The response to the dis- persed forest harvesting plan shows that the most significant impact arises in OAs that have not been previously har- vested. Hunting trips to these sites would decrease by 30–40%. Because most of the forest harvesting in the area of community 2 occurs on such lands, the impacts are more severe on hunters in this community. Hunters are expected to move to other areas, but they reallocate over a large number of OAs instead of simply switching to a small number of other sites. Employing a concentrated forest harvesting plan signif- icantly reduces the impact on hunters. In part this is because of the decreased number of OAs affected, but it is also because most of the sites selected for concentrated harvests had already experienced some degree of timber harvesting. As one would expect, concentration appears to result in an improved overall outcome. The welfare impacts[11] associated with the timber har- vesting plans in each community are presented in Table 5. The harvesting effects are more pronounced in community 2 and are very small in community 1. As mentioned above, the reason for this difference is that timber harvesting has occurred near community 1 for a number of years. How- ever, timber harvesting is just beginning near community 2. Therefore, hunters in community 1 are already hunting in areas containing recent cutblocks, and further harvesting in this area has a limited effect on Aboriginal hunters using the area. In community 2, however, most areas visited by hunt- ers in the community have never been subjected to forestry operations. As a result, timber harvests in this area causes the hunters to substitute away from the newly cut areas. The hypothetical concentrated harvesting plan mediated the neg- ative effects of harvesting in community 2. An unexpected Forest Science 50(2) 2004 147 result occurs in community 1. Dispersed harvesting results in a positive (although insignificant) welfare effect as the positive effect on moose populations dominates the negative effect of harvest on esthetics. However, the concentrated harvests do not positively affect as many sites and result in a very small net loss in welfare. Both these effects are quite small in per trip and aggregate terms. Resource Compensation as an Alternative to Monetary Welfare Measures Forest managers have some ability to counteract the effects of forest harvesting by compensating "in-kind." This could involve enhancing the hunting-related attributes of sites known to be preferred by hunters. This strategy is somewhat analogous to the concept underlying zoning in that land-use specialization is being employed. Strategies that forest managers could employ include removing access (to reduce encounters and congestion), investing in wildlife habitat improvements, or, with cooperation from fish and wildlife management agencies, limiting access by non-Ab- original hunters to certain OAs. We examine resource com- pensation in terms of investments in wildlife habitat that generate increased moose populations in select OAs. Improvements in moose populations could result from restricting human access to lower the moose mortality rates Figure 3. Predicted percent changes in trips of Aboriginal hunters for two forest harvesting plans in two commu- nities in the NorSask FMA. 148 Forest Science 50(2) 2004 from hunting, or landscape alterations through forest man- agement to provide more moose habitat (e.g., Saskatchewan Forest Habitat Project 1993). In this study, it is assumed that actions can be taken to affect moose abundance and that these measures will affect the moose abundance measure described above and in Table 2. We also consider the possibility that different levels of intensity of investment in increasing moose populations could occur. The first is a low intensity plan that increases the probability of an OA having exceptional moose abundance to 0.20. The second is a high intensity plan that will increase the probability to 0.50. The model of hunting preferences along with the infor- mation on site attributes provides a way to determine the best area to invest in habitat improvements to increase moose populations. To determine which OAs were the best candidates for moose improvements, the moose abundance probability was increased to the target level (i.e., 0.2 or 0.5) at each individual OA. For each change, the change in total welfare per trip was calculated. The OAs were then ranked according to the change in total welfare per trip resulting from improving moose probability at that site. To determine how many of the OAs require improvement, the effect of improving moose habitat is simulated for the top 2 ranked sites, then the top 3 ranked sites, then the top 4 ranked sites, etc., until the welfare impact is just enough to offset the impact of the timber harvest plan. This strategy is similar to that employed in resource compensation efforts in the Natural Resource Damage Assessment literature (e.g., Desvousges et al. 2000). The results of these simulations for each community are reported in Table 6. For community 1, in which consider- able harvesting has occurred in the general hunting area in the past, no OAs required improvement to offset the nega- tive impacts of the timber harvest plan because the net impact of timber harvest was positive. Of course the impact of the plan on hunters in this area was also judged to be relatively minor. However, in community 2, 13 OAs re- quired management intervention to achieve improvement in the low-intensity scenario, and 1 OA was required in the high-intensity scenario. Thus, in the area that has not expe- rienced forestry operations in the past, considerable inter- vention is required to compensate hunters. This leads one to question: what are the features of those OAs in which intervention is required for compensation? For community 2, where many OAs have never been har- vested, the best candidates for moose enhancement are OAs where no cutting occurred in the past and no new harvesting occurs. Other candidate OAs would be those with more cabins nearby, are most accessible by water, and, as mea- sured by the moose abundance indicator model, moose habitat is relatively poor. Discussion and Conclusions Increasing importance is being placed on recognizing the values of Aboriginal people in resource management deci- sions. In the context of forest management in Canada, Aboriginal people are often significantly affected by for- estry decisions. Attempts to incorporate Aboriginal values into management have included various co-management strategies and consultation strategies, but even these have not necessarily addressed the challenge adequately. Mone- tary compensation for the impact of industrial activity on traditional land-use activities has been used in some cases, but the methods of determining such monetary compensa- tion are questionable. In this article, we make use of a unique data set that allows us to assess the impact of forest management on Aboriginal hunting activities. We employ a behavioral ap- proach to better understand the tradeoffs that hunters make and the implicit value of changes in the environment. We use this behavioral model to assess impacts of forest har- vesting and to develop monetary measures of this impact. We also use this model to examine alternative strategies for managers including resource compensation and zoning. The first conclusion arising from our study is that invest- ment in the data collection component is critical for the collection of data on use of nontimber resources by Aborig- inal people. Without the investments made to collect the data in an atmosphere of trust and reciprocity, our study would not have been possible. The unique character of our study area that includes a forest management agency that has previously invested in co-management relationships with the Aboriginal people should also not be undervalued. It is unlikely that very many forest management contexts would involve such characteristics. Nevertheless, if the val- ues of Aboriginal people are to be effectively incorporated into management, such investments are necessary. In addition to investing significantly in data collection, we also made the choice to obtain and employ SP data, RP Table 5. The changes in economic values associated with two harvesting plans on hunting trips taken by Aboriginal hunters in two communities in the NorSask FMA. Harvest plan Per trip values Total values Community 1 Community 2 Community 1 Community 2 Dispersed harvesting $0.14 (0.202)a $–1.21 (0.593)a $150.96 $–2791.88 Concentrated harvesting $–0.03 (0.052)a $–0.113 (0.170)a $25.16 $–189.28 a Standard errors computed using the Krinsky-Robb procedure with 1,000 replications. Table 6. The number of OAs in which actions to improve moose abundance are required to compensate Aboriginal hunt- ers for the planned dispersed timber harvest. Intensity of effort Number of OAs required Community 1 Community 2 Low 0 13 High 0 1 Forest Science 50(2) 2004 149 data, and hunter perceptions/knowledge. We believe that the use of traditional ecological knowledge of the hunters im- proved our ability to model choice. We also believe that the use of SP data is important in contexts such as these where the RP data may be highly correlated and where new man- agement strategies may extend landscape conditions beyond those that are currently being experienced. However, issues remain regarding econometric implications of our sequen- tial estimation strategy, the appropriate weighting of RP and SP data in estimation, and the degree to which understand- ing and behavioral prediction is improved by combining data sources. A second conclusion is that it is clear that Aboriginal hunters make tradeoff decisions that are consistent with an underlying optimization framework. This conclusion is con- sistent with the research of Winterhalder (2001) and others who examine Aboriginal hunter behavior in an optimal foraging framework. Thus, Aboriginal hunting behavior can be used to develop measures of value associated with envi- ronmental attributes. This supports the use of behavioral studies in the Aboriginal context. However, it is not clear that we have accurately captured the behavioral relation- ships. In particular, the tradeoffs regarding time use require more investigation. Aboriginal hunters respond to opportu- nities, both market and nonmarket, that arise over time. A more careful assessment of the choice of these opportunities could provide valuable insights into the implicit value of time in hunting activities. Hunting activities are not recre- ation for many Aboriginal people. The choice to invest time in hunting rather than market wage opportunities reflects a type of reservation wage that could be identified from detailed activity data. This we feel is an avenue for future research that could help identify the importance, both culturally and materially, of hunting within Aboriginal societies. While estimated monetary measures of welfare are re- ported in this article, we believe that our investigation of resource compensation and zoning (concentrated forest har- vesting) could be more useful to resource managers. In- creasingly it is recognized that wide-scale multiple-use for- estry may not be the best way to manage forests and that specialization may be an improved management model. Our results suggest that investments in habitat at a few sites could offset the impacts of forest harvesting. It would be interesting to examine the costs of such investments versus the monetary compensation required to offset these impacts. However, while resource compensation addresses the im- pacts in total, it does not address the distribution of impacts. Such distributional impacts have been identified as a diffi- culty in resource compensation exercises or indeed in any form of benefit cost analysis that does not use money as the numeraire (Brekke 1997). The models we develop here and the resource compen- sation approach is also consistent with the notion of main- taining the welfare from traditional nontimber harvesting activities within forest management planning. One compo- nent of recent interpretations of Aboriginal rights include rights to nondeclining streams of benefits from traditional activities. Such a constraint on forest management planning can be implemented using the data and methods we present here. This approach provides an opportunity for forest man- agers to actively incorporate values of Aboriginal people into forest management and recognize emerging policy out- comes on the rights of Aboriginal people. The distribution of impacts across the population of Aboriginal people has been ignored in this study. That is clearly a limitation of our approach to this point. In previous research, the heterogeneity across Aboriginal peoples has been highlighted as a significant feature in the data (Haener et al. 2001b). Incorporating heterogeneity into our modeling approach through the use of mixed logit models or finite mixture models (e.g., Boxall and Adamowicz 2002) will undoubtedly improve our understanding of behavior and provide us with improved measures of welfare and resource compensation. However, incorporating heterogeneity will also increase the complexity of the measurement of welfare and resource compensation. Endnotes: [1] This is an organization consisting of nine First Nations that is charged with providing services and programs to facilitate economic devel- opment for the communities in the region. [2] The data collected at the individual level are confidential. Even representations of predicted activities (maps, etc.) that are based on these individual data are considered confidential. In this article, results are presented in aggregate form or, in cases where more disaggregate results are presented, all identifiers have been removed to prevent any possibility of revealing individual level information. The use of such information in forest management planning clearly requires a similar degree of protection of confidential information by the management agency. Without such agreements, it is unlikely that private information of this type will be made available. [3] Me ?tis are people whose ancestors come from mixed indigenous and European (primarily French) descent. The Me ?tis, First Nations, and Inuit are the three groups of Aboriginal people in Canada. [4] The map provided was approximately 6 ? 3 ft in size so that detailed information could be recorded. [5] Most of these data are not used here for modeling purposes, but these individual maps were compiled into one and this was provided to Mistik Management to be used when drafting their forest harvest plans. We also derived "community-level" special sites maps and presented these to each community during presentations in June 2001. This procedure assisted in the process of building trust during the data collection exercise. [6] Because many individuals in our sample changed their employment- related activities over the year, one could develop a model explaining the changes as a function of characteristics of the opportunities (wages, etc.) and the season of the year (an indicator of the value of nontimber-based activities). This would provide a measure of the implied reservation wage associated (and value of time) with nontimber-based activities. However, our data were not detailed enough to construct such an analysis. [7] The parameter estimates are not presented to maintain the confiden- tiality of this traditional ecological knowledge about the location of abundant wildlife populations. Further details on these estimates can be obtained from the authors and will be provided in a fashion that maintains the confidentiality of the information but allows for repli- cation of the results. [8] This information is reported by Kreag and Moe (2002) and available at www.labaronssports.com/pages/mancominiatv.htm [9] A likelihood ratio test was performed to examine whether the RP and SP parameter vectors are significantly different that those of the joint model. The results (?2 ? 76.8, 4 df. based on 5 equality restrictions less 1 restriction released for the scale parameter) suggest that the joint model parameters are significantly different. This test, however, 150 Forest Science 50(2) 2004 is recognized in the literature as being rather strict and it appears that only one or two parameters drive this result. Nevertheless, we plan on further research to investigate this issue. [10] We do not identify these communities in this article for reasons of confidentiality. [11] Hanemann's (1982) welfare measure was used to assess the economic effects of these harvest plans in each community. Literature Cited ADAMOWICZ, W.L., J. SWAIT, P. C. BOXALL, J. LOUVIERE, AND M. WILLIAMS. 1997. Perceptions versus objective measures of environmental quality in combined revealed and stated prefer- ence models of environmental valuation. J. Environ. Econ. Manage. 32(1):65–84. BECKLEY, T.M., AND B.H. HIRSCH. 1997. Subsistence and non-in- dustrial forest use in the lower Liard valley. Inf. Rep. NOR- X-352. Natural Resources Canada, Canadian Forest Service, Edmonton, Alberta, Canada. 42 p. BOWES, M.D., AND J.V. KRUTILLA. 1985. Multiple use manage- ment of public forestlands. P. 531–569 in Handbook of natural resource and energy economics, Kneese, A.V., and J.L. Sweeney (eds.). Elsevier Science Publishers, New York. 630 p. BOXALL, P.C., B.L. MCFARLANE, AND M. GARTRELL. 1996. An aggregate travel cost approach to valuing forest recreation at managed sites. For. Chron. 72(6):615–621. BOXALL, P.C., AND W.L. ADAMOWICZ. 2002. Understanding het- erogeneous preferences in random utility models: A latent class approach. Environ. Resour. Econ. 23(4):421–446. BOXALL, P.C., AND B. MACNAB. 2000. Exploring the preferences of wildlife recreationists for features of boreal forest manage- ment: A choice experiment approach. Can. J. For Res. 30(12):1931–1941. BREKKE, K.A. 1997. The numeraire matters in cost-benefit analy- sis. J. Publ. Econ. 64(1):17–123. BROWN, T.C., AND E.S. BURCH. 1992. Estimating the economic value of subsistence harvest of wildlife in Alaska. P. 203–254 in Valuing wildlife resources in Alaska. Peterson, G.L., C.S. Swanson, and D.W. McCollum (eds.). Westview Press, Boul- der, Colorado. 357 p. DESVOUSGES, W.H., D.J. MACNAIR, AND G.A. SMITH. 2000. Lower Fox River and Bay of Green Bay: Assessment of po- tential recreational fishing losses and offsets. Triangle Eco- nomic Research, Durham, NC. 74 p. DEPARTMENT OF INDIAN AFFAIRS AND NORTHERN DEVELOPMENT (DIAND). 2001. Basic departmental data. First Nations and northern statistics section. Ottawa, Canada. 82 p. DOSMAN, D., M. HAENER, W. ADAMOWICZ, J. MAROIS AND P. BOXALL. 2002. Assessing impacts of environmental change on Aboriginal people: An economic examination of subsistence resource use and value. Project Report 02-1, Department of Rural Economy, University of Alberta, Edmonton, Alberta, Canada. 33 p. ENGLIN, J. 1990. Backcountry hiking and optimal timber rotation. J. Environ. Manage. 31(2):97–105. FEIT, H.A. 1987. North American native hunting and management of moose populations. Swedish Wildlife Res. Suppl. 1:25–42. HAENER, M.K., AND W.L. ADAMOWICZ. 2000. Regional forest resource accounting: A Northern Alberta case study. Can. J. For. Res. 30(2):264–273. HAENER, M., P.C. BOXALL, AND W.L. ADAMOWICZ. 2001a. Mod- eling recreation site choice: Do hypothetical choices reflect actual behavior? Am. J. Agric. Econ. 83(3):629–642. HAENER, M., D. DOSMAN, W. ADAMOWICZ, AND P.C. BOXALL. 2001b. Can stated preference methods be used to value at- tributes of subsistence hunting by Aboriginal peoples? A case study in northern Saskatchewan. Am. J. Agric. Econ. 83(5):1334–1340. HANEMANN, W.M. 1982. Applied welfare analysis with qualitative response models, Working Pap. No. 241. Univ. California, Berkeley, 26 p. KREAG, G., AND J. MOE. 2002. Minnesota's north shore snowmo- bile trail: What is its value? Minnesota Sea Grant Tourism Outreach, University of Minnesota. (www.seagrant.umn.edu/ tourism/snow.html. May 7, 2002.) LOUVIERE, J., D. HENSHER AND J. SWAIT. 2000. Stated choice methods: Analysis and applications in marketing, transporta- tion and environmental valuation. Cambridge University Press, Cambridge, United Kingdom. 402 p. MACKINNON, L., C. APENTIK, AND M.P. ROBINSON. 1999. Revis- iting traditional land use and occupancy studies: Relevance and implications for resource management in Alberta. Sustain- able Forest Management Network Working Paper 1999-16. Edmonton, Alberta, Canada. 32 p. MENDELSOHN, R. 1996. An economic-ecological model for eco- system management. P. 213–221 in Forestry, economics and the environment, Adamowicz, W.L., et al. (eds.). CAB Inter- national, Wallingford, United Kingdom. 275 p. PARSONS, G.R., AND M.J. KEALY. 1992. Randomly drawn oppor- tunity sets in a random utility model of lake recreation. Land Econ. 68(1):93–106. PYC, C. 1999. The use of traditional knowledge in Cree hunting strategies. Sustainable Forest Management Network Working Paper 1999-14. Edmonton, Alberta, Canada. 11 p. SASKATCHEWAN FOREST HABITAT PROJECT. 1991. Annual Report. Wildlife Branch, Saskatchewan Parks and Renewable Re- source, Prince Albert, Saskatchewan, Canada. 101 p. SASKATCHEWAN FOREST HABITAT PROJECT. 1993. Integrated forest/wildlife management parameters. Wildlife Branch, Saskatchewan Environment and Resource Management, Prince Albert, Saskatchewan, Canada. 27 p. STATISTICS CANADA. 1998. Welcome to Statistical Profile of Canadian Communities. ww2.statcan.ca/english/profil/ (accessed May 1998) TOBIAS, T.N., AND J.J. KAY. 1993. The bush harvest in Pinehouse, Saskatchewan, Canada. Arctic 47(3):207–21. USHER, P.J., AND G. WENZEL. 1987. Native harvest surveys and statistics: A critique of their construction and use. Arctic 40(2): 145–160. USHER, P.J. 1976. Evaluating country food in the northern native economy. Arctic 29:105–120. Forest Science 50(2) 2004 151 VINCENT, J.R., AND C.S. BINKLEY. 1994. Efficient multiple-use forestry may require land-use specialization. Land Econ. 69(4):370–376. WINTERHALDER, B. 1983. Boreal foraging strategies. P. 201–241 in Boreal forest adaptations: The Northern Algonkians, Steeg- mann, A.T. (ed.). Plenum Press, NY. 360 p. WINTERHALDER, B. 2001. Recent work on the behavioral ecology of hunter-gatherers. P. 12–38 in Hunter-gatherers: An interdis- ciplinary perspective, Panter-Brick, C. R. Layton, and P. Rowley-Conwy (eds.). Cambridge University Press, Cam- bridge, United Kingdom. 341 p. 152 Forest Science 50(2) 2004