Abstract. The forest composition we witness today is a product of temporal anthropogenic andnonanthropogenic disturbances. Scholars from geography, anthropology, and other disciplines havelong been aware of the informing nature of spatial relationships: human actions in a previous timeoften leave imprints in today’s landscape. Traditional empirical studies of forest condition typicallyignore this type of information and rely on aggregated forest-level indicators developed from aspatialplot-level analyses. This paper conducts a spatial analysis of one important forest product species,Shorea robusta, in a foraging setting in southern Nepal. Forest plot locations were located usingDifferential Global Positioning Systems (DGPS) and were processed using a Geographic InformationSystem. Three rival hypotheses of the geographic distribution ofShorea robustaare presented: (1) apattern of no human disturbance, (2) a pattern of open access and optimal foraging, and (3) a patternof optimal foraging altered by the geographic configuration of enforced institutions. Multivariateregression models are estimated and optimal foraging patterns are identified. Statistical tests lendsupport to the third hypothesis. Methods such as the ones presented here are important if we areto better understand the geographic implications of institutional design on human behavior and theenvironmental outcomes that result.
Over the past decade, considerable attention has been given to the subject of human-induced forest change and the depletion of specific species in forests (Myers, 1988;Aldhous, 1993; Repetto, 1988; Lovejoy, 1980; Task Force on Global Biodiversity,1989; Norton, 1986; Reid and Miller, 1989). Often these studies take a macroview of the problem, focusing on general political or economic influences (Re-petto, 1988; Richards and Tucker, 1988). Other research shifts attention to theindividual and searches for deeper understanding of influential variables that driveforaging behavior. Some of these ‘micro-scale analyses’ focus on the influence ofinstitutions or rules-in-use that create or modify human incentives and behaviorrelated to forest product consumption (Ascher, 1995; Angelsen, 1995; McKean,1992; Thomsonet al., 1992; Ostrom and Wertime, 1995; Morrow and Hull, 1996).
Micro-level investigations concerned with understanding the human impacts onforest change require some capacity to quantify forest condition and some methodto analyze the change. The traditional method to quantify forest condition is to(1) take a sample of vegetation using a sampling strategy and forest plot meas-urements, (2) calculate aggregate species abundance indicators such as density,dominance and frequency from these data, and (3) use these indicators to describethe current status of the forested area as a whole. Plot-level analyses are alsosometimes conducted (see, e.g., Umans, 1993), but usually without attention to thespatial distribution of the plots. If the researcher isextremelyfortunate, prior datamay have been collected on forest condition and these data can be compared withnewly collected measurements. General conclusions can then be made regardingthe change in forest resources and the impact of current institutional arrangementsand forest policies on human foraging incentives in the region.
Unfortunately, it is rare indeed to find a study location that actually has hadforest condition measures taken at an earlier point in time. In most cases, especiallyin developing world settings, we possess information gaps: no prior data existson the condition of a forest we set out to study. Even in the rare circumstanceswhere a forest inventory has been taken, the data is either not georeferenced, orgeoreferenced in an aggregated form. Understanding change in the resource in thiscontext is quite difficult, for no baseline data exists for comparison.
Scholars from geography, anthropology, and other disciplines have long beenaware of the informing nature of spatial relationships: yesterday’s human actionsoften leave imprints that remain apparent in the landscape of the natural resourceof today (Pickett and Candenasso, 1995; Kelleret al., 1996). In instances wherewe lack longitudinal data we can still extract new information related to changethrough the study of these patterns. Unfortunately, up until very recently, our abil-ity to capture spatial relationships has been hampered by our inability to collectaccurate spatial data. The advent of differential global positioning system (DGPS)technology provides new opportunities for the accurate georeferencing of data.Armed with this new information and digital-processing capabilities supplied bygeographic information systems (GIS) and spatial statistics, we can more easilycollect accurate spatial data and analyze it for expected spatial patterns. A spa-tial analysis provides an opportunity to extract additional information about forestchange in instances where no baseline forest condition data exist.
Further, in addition to overcoming the ‘no baseline data’ problem, a spatialanalysis at a forest plot level may help shed light into community dynamics – some-thing that might be missed using data aggregated at the forest level. For example, inagrarian societies who depend on forest products for subsistence, the existing spa-tial distribution of an important forest product species may reflect human foragingdecision making in response to the physical geography and established harvestinginstitutions, rules, or social norms. Over geographic space, particular forest loca-tions may be subject to heavier harvesting levels as foragers respond to existingcommunity relationships and forest institutional structure. To the researcher trying
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 233
to understand how community inequities and governance arrangements influencethe harvesting behavior of foragers, a spatial analysis could be quite revealing.
The goal of this article is twofold: It is to understand the influence of forestgovernance and human foraging on a particularly important tree species,Shorearobusta, in one empirical Nepalese context. It is also to work toward developingnew institutional analysis methods – by combining recent advances in DGPS, GIS,institutional analysis and regression – to tease out the human dimensions of forestchange using cross-sectional forest plot data.
The article proceeds in the following manner: First, the study site and datacollection methods are described. Second, an overview is provided on the forestgovernance structure at the site and an assessment of human foraging patternsis made based on villager reports and what we witnessed in the field. Given thisknowledge, three rival hypotheses are presented related to the geographic distribu-tion of one particularly important forest species: a ‘no human influence’ pattern,an ‘open access and optimal foraging’ pattern, and an ‘optimal foraging combinedwith institutional influences’ pattern. Third, a traditional aggregate forest plot ana-lysis is presented and it is determined that little information can be garnered toidentify which hypothesized pattern is supported. A focus at the forest plot level isrequired. Fourth, three plot-level multivariate regression count models are presen-ted, one representing each hypothesis. Fifth, statistical methods are described andresults are presented. Statistical tests are conducted to determine which hypothesisis supported. Sixth, substantive and methodological implications are discussed.
2. The study site and data collection
In October 1994, forested areas within the Kair Khola Watershed in the ChitwanDistrict of southern Nepal were chosen for a study of forest governance (Figure 1).The project, a part of the International Forestry Resources and Institutions (IFRI)research program at Indiana University, entailed gathering of information relatedto forest governance, use, and condition along with socioeconomic attributes ofvillagers who utilize these resources (Ostrom and Wertime, 1995). A research teamcomprised of Nepali researchers and the author spent six weeks in the field learningabout villager foraging practices and the institutions governing forest harvestingand management. The research site falls at the juncture of the Kayar and the ShakiRiver systems. Figure 2 presents a scanned and geometrically rectified 1995 topo-graphic map of the region. This map was created by His Majesty’s Government ofNepal through interpretation of 1992 aerial photographs of the region. Grey areasdesignate forests; white areas reflect either degraded forest areas or areas undersome agriculture regime. Four general communities exist in the study area: Milanon the west bank of where the Shakti and Kayar Rivers converge, Shaktikor to the
234 CH. M. SCHWEIK
Figure 1.Study site location within Nepal.
east along the banks of the Kayar, Latauli to north of Shaktikhor, up into the hillsand Chherwan, still higher in the hills and farthest east.1
In general, the villagers in the west village of Milan are relatively more well offthan the other communities. Many households own good land along the river withample access to water resources to irrigate rice fields. Milan also exhibits moreheterogeneous population when compared with the eastern communities with mostmembers from the Chepang, Chettri, and Newar ethnic groups, but others such asBrahmin, Tamang, Gurung, and Magar are also represented. The eastern villagesof Latauli, Shaktikor, and Chherwan also are comprised of subsistence farmersliving in areas where it is more difficult to irrigate given their topographic locations.Consequently, they grow other crops, such as maize, that require less water and areless commercially valuable.
The scanned topographic map in Figure 2 is also helpful, for it identifies house-hold point locations. These point locations were also interpreted from the 1992aerial photos and correspond reasonably well with what we witnessed in the field.In a few instances, we digitized household locations not identified in the map wherewe knew them to exist. These estimates are presented in Table I.
1 Villages’ names used here differ slightly to the names utilized in another study of this generalregion (see Schweiket al., 1997). For consistency purposes in this analysis, village names correspondto the names printed on His Majesty’s Government scanned topographic map displayed in Figure 2.The Schweik, Adhikari, and Pandit study, on the other hand, provides village names as reported bythe villagers themselves.
OP
TIM
AL
FO
RA
GIN
G,IN
ST
ITU
TIO
NS
AN
DF
OR
ES
TC
HA
NG
E235
Figure 2.Map of study site.
236 CH. M. SCHWEIK
TABLE I
Characteristics of the communities in the region (Based on villager estimates, IFRI, 1994)
Community name Associated village names Estimated # of Forests
(from map) from IFRI study households harvested
Milan Sulitar, Kuwapani, 110 Sugabhanjyang,
Sinjali gaun, Bhandari gaun, Latauli,
Sewnjaja towe, Milan Chok Kaswang (rarely)
Shaktikhor Dogara 58 Latauli
Latauli Latauli, Deurali 35 Latauli
Chherwang Chherwang 40 Latauli
2.1. FOREST PLOT DATA COLLECTION
We utilized traditional plot sampling to measure forest condition. The team in-cluded one forester, one botanist, and several assistants. We utilized ten-meterradius circular forest plots for sampling. Due to the steep terrain within theseforests, the team followed trails to reach 50 m altitudinal intervals. At each verticallocation, a random number was used to determine the direction and the distancefrom the trail that the corresponding plot should be taken. Overall, 97 forest plotswere sampled (Figure 2). Data recorded include:
• soil characteristics, such as the depth of the humus layer, and the depth andcolor of the ‘a’ and ‘b’ horizons;• tree identification, including diameter at breast height, height and species type
for each tree within the plot;• plot physiographic information, such as slope (in degrees, measured by a clino-
meter), elevation (using an altimeter), and aspect (the direction the slope faces);• ancillary observations, such as the existence of insect damage, signs of animal
grazing, and evidence of human harvesting.
We also did something rather unusual – but soon to become more prevalent. Wewere fortunate enough to have two eight-channel GPS receivers and a laptop com-puter in the field, which allowed us to collect accurate positional data regardingthese forest plots. Using differential GPS (DGPS), a technique which employs twoGPS machines, one acting as a ‘basestation’ at a known location, and the othercollecting data in the field, we were able to collect forest plot positions in longitude,latitude, and Universal Transverse Mercator (UTM) coordinate systems with anaccuracy of 1–5 m (see Paceet al., 1995). This type of accuracy is required fora plot-level geographic analysis. These positions were converted into a GIS pointcoverage and are overlaid on the georeferenced map presented in Figure 2.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 237
3. Forest Governance, Use and Hypothesized Outcomes
Villagers in each of these three communities are subsistence farmers and dependheavily on forest products for their livelihood. The term ‘forest’, as defined by ourIFRI research program, is land area larger than 0.5 ha, possessing some woodybiomass, subject to the same governance structure and utilized by at least threehouseholds. Using this definition, three forests were identified: to the west, ‘Sug-abhanjyang Forest’; to the east, ‘Latauli Forest’; and to the south, ‘Kaswang Forest’(Figure 2). Each are semi-deciduousShorea robustaclimax forests.
These forests are each designated official ‘government forests’ and fall underthe management of the District Forest Office (DFO). The DFO manages foreststhrough Village Development Committee (VDC) boundaries. The VDC is the smal-lest political unit in the Nepalese administration system. A VDC boundary runsdirectly up the Kayar River in the south-west and then follows the Shakti Rivernorthward, effectively placing the western Sugabhanjyang Forest under a differentVDC jurisdiction from that governing Kaswang and Latauli.
There are three formally established DFO rules related to forest product use.First, anyone who is a member of a VDC is permitted to harvest grass, tree fodder,and deadwood from forests within that VDC to support their daily subsistencerequirements. Second, live tree harvesting can only be conducted if formal permis-sion is received from the DFO prior to harvesting. Third, a ‘no encroachment’ ruleexists that prohibits the conversion of DFO forest to some other land use.
DFO guards stationed at VDC range posts enforce these rules. The DFO rangepost offices are a significant distance away from these forests: the range post as-sociated with the monitoring of Sugabhanjyang resides approximately 14 km tothe south-west of Milan, and the DFO range post for Latauli and Kaswang islocated approximately 16 km south-east of the village of Latauli. At each rangepost, approximately ten guards are stationed. These guards are responsible forpatrolling – largely on foot – a hilly, almost mountainous area that extends over100 km2. Their task is daunting and their effectiveness appears to be quite limited.It is not surprising that their efforts, while weak everywhere, appear to be moreeffective in geographic locations more easily accessible from their range post loca-tions. Villagers report relatively few interactions with forest guards, but when theydo occur they tend to be more frequent in areas along the motorable road in thewestern side of the study site. At one point during our fieldwork, we witnessed theDFO enforcing the no encroachment rule. Guards destroyed the home of a villagerwho had encroached upon land in the western side of Sugubhanjyang near theroad, and they hauled the building material away with a truck. This incident, whilereportedly rare, proves that there is rule enforcement in areas reasonably close tothe motorable road through Milan.
The map in Figure 2 shows the road crossing the Shakti River and going throughthe eastern village of Shaktikhor. This map is deceiving, for crossing this river ina vehicle at any time of the year is quite difficult. The convergence of the two
238 CH. M. SCHWEIK
rivers at this juncture leads to a process called ‘the backwater effect’ (Bruijnzeeland Bremmer, 1989: 64), where tremendous quantities of boulders and rocks aredeposited in the Shakti riverbed. Motorable crossing is very difficult even in thedry season. The result, confirmed by villager reports, is that the monitoring ofthe Latauli and Kaswang Forests by DFO guards is even less frequent than inSugubhanjyang.
Forest use by villagers in all communities is quite similar. Villagers harvesttimber for construction and for tools, fodder, leaf litter and grasses for livestockand other agriculture purposes, and fuelwood for cooking and heating purposes. Wewitnessed extensive foraging activities during our weeks in the field. People fromall villages reported that timber extraction, fuelwood gathering, and tree loppingare the major forces in what they see as a rapid depletion of their forest resources.Tree lopping is especially prevalent, which, as Metz (1990: 285) notes, significantlyreduces the opportunity for species to regenerate.
While the formal DFO rules appear to be well understood, in many respectsthey are not followed: what we heard and witnessed in the field proved that theserules were consistently being broken. The slash and burning of forest land foragriculture perhaps is the most extreme DFO rule violation and this practice isprevalent especially in higher locations in the hills. This aspect of the human-forest dynamics is described in much more detail in Schweik, Adhikari, and Pandit(1997). However, foraging-related violations also occur frequently. The villagersfrom Milan, more wealthy (relatively) and ethnically diverse, report that they har-vest not only from the Sugabhanjyang Forest in their VDC, but they also lop treesfor fodder and gather grasses from the eastern Latauli Forest in the neighboringVDC. This is a direct violation of the formal DFO-established rules in the region.Interestingly, the villagers from the western communities of Shaktikor and Lataulido not seem to mind, as no complaints have been registered to the DFO range post.Even more puzzling, the villagers from the western communities of Shaktikhor, andLatauli, on the other hand, report that they forage only in the Latauli Forest. Theyexplain that most of the year the Kair River flows too wide for them to access theKaswang Forest to the south, and that no consideration is even given to harvestingin the Sugabhanjyang Forest in the neighboring VDC. Research team memberswho lived in the villages, held numerous discussions with villagers about foragingbehavior, and monitored forest-harvesting activities for nearly six weeks, confirmthis behavior (Shrestha, 1996). Each side takes a ‘that’s just the way it is’ mentalitywhen asked about these foraging patterns. It appears, then, that an unwritten socialnorm exists across communities that effectively permits western (Milan) villagerforaging in the eastern Latauli Forest but does not allow the converse to occur. Thisadds additional foraging pressure on the Latauli Forest – a forest already heavilyused by the Shaktikhor, Latauli, and Chherwang villagers.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 239
3.1. A FOCUS ON IMPORTANT PRODUCTSPECIES
If we are interested in understanding deforestation practices in a foraging com-munity, it is most helpful to focus analysis on those species the communities findmost important for their livelihood. My point here is simple but, I think, important:In any setting where foraging levels are high, the severity of the deforestation willmanifest itselffirst in the distribution (or lack thereof) of the particular speciesthat contributes most to villagers’ daily subsistence requirements. When asked, thepeople of all three villages mentioned thatShorea robustawas by far the mostcritical tree species for supplying timber, fodder, and fuelwood needs.
3.2. HYPOTHESES RELATED TO PATTERNS IN THE DISTRIBUTION OFSHOREA
ROBUSTA
We now can develop hypotheses related toShorea robustapatterns given what weknow about the forest governance structure and monitoring capacity, forest productuse, geographic locations of households, and community relationships. Three rivalhypotheses exist.
Hypothesis 1: There is little or no evidence of human overconsumption of Shorearobusta. The forest is regenerating at a rate greater than, or equal to, what isextracted. The first possible pattern is one of a ‘sustained’ forest ecosystem whereforests are able to regenerate at a rate faster, or equal to, the rate of what humans areremoving. The pattern ofShorea robustain any of the forests would be no differentthan what would be found in a comparable forest in a similar ecological settingthat has not been subjected to human harvesting – what is sometimes referred toas a reference forest. Each particular species follows its own ‘naturally induced’distribution over the topography. Figure 3 describes this landscape. The likelihoodthat this type of pattern exists in Sugubhanjyang or Latauli is doubtful, however,given that villagers from all communities, the very people who know these forestsbest, report that these two forests have been significantly degraded over the past20 yr.
Hypothesis 2: Shorea robusta are being removed at a rate faster than the forestcan regenerate. The pattern of depletion will reflect an open-access situation and aprocess of optimal foraging. The second possible pattern that might exist is onethat reflects an open-access situation where human decision making and harvestingis driven simply by optimal foraging strategies. Optimal foraging theory depictshuman foragers as actors who maximize their net rate of return of energy per unitof foraging time (Smith, 1983). While a number of alternative theories on foragingdecision making exist (ibid.: 627), they all characterize the forager as a person whostrives to minimize his or her search time and effort (Hayden, 1981; Winterhalder,1993). If humans harvest important product species at a faster rate than it regener-ates naturally, optimal foraging predicts that lower numbers of these species will
240 CH. M. SCHWEIK
Figure 3.Expected patterns in a ‘sustained forest’ scenario.
be found in locations easily accessed by humans (e.g., a short distance away fromthe village, near a path, or at a low elevation). In this setting, then, we would expectthe number ofShorea robustatrees to be higher in number in those areas furtheraway from villages and at higher altitudes where it is more difficult to traverse.Consequently, we would expect Kaswang to exhibit species distributions reflectingno or very low foraging activities, given that it is well protected from foraging bythe river systems and little human settlements exist within or near it. We would alsoexpect the northern part of Sugubhanjyang to be relatively untouched by humans,given that it is high up in the hills and no villages exist to the north. Alternatively,the southern half of the Sugubhanjyang Forest near Milan should be relatively hardhit in terms of foraging pressure, as well as in the eastern side of the Latauli Forestwhere it is completely surrounded by settlements. But optimal foraging wouldpredict that the forest hardest hit in terms ofShorea robustaextraction would bethe western side of the Latauli Forest – the part in the center of the map that sits inbetween the three villages of Milan, Shaktikhor, and Latauli. A graphic depictionof the expected foraging patterns in an open-access, optimal foraging situation isprovided in Figure 4.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 241
Figure 4.Expected patterns as a result of open access and optimal foraging.
Hypothesis 3: Shorea robusta are being removed at a rate faster than the forestcan regenerate. The pattern of depletion will reflect a process of optimal foragingaltered by the geographic configuration of effectively enforced institutions.Smith(1983) reports that empirical studies testing optimal foraging theory have revealedsome instances where human foragers are selective in their utilization of avail-able resources. Other studies have revealed foragers who exhibit much less con-cern. Smith also states that there is little agreement in the anthropological com-munity over these foraging differences (ibid.: 628–29). While not stated specific-ally, Smith’s discussion alludes to the importance of community relationships andthe important role institutional arrangements play in the influence of human for-aging patterns and their efforts for natural resource preservation.2
Ostrom (1990) extends Smith’s argument by emphasizing the role institutionalarrangements play in altering the incentives humans face in their decision-makingcontext. Institutions in this context refer to the property rights and rules-in-use thatgovern the harvesting of a particular species or particular areas (what we mightrefer to as management units) within a forest. The forests in this particular case are,
2 For example, Smith (1983: 632) describes the role that ‘exclusive control’ plays in the conser-vation of natural resources. Feit (1973) describes rotational hunting by the Waswanipi Cree peopleas a method in which the size of animal population can be controlled.
242 CH. M. SCHWEIK
to a significant degree, open access, leading to the expectation that optimal foragingpatterns will be prevalent. But the possibility exists that foraging decision makingover the years may have been altered by what I refer to as the past and present ‘insti-tutional landscape’ configurations (Schweik, 1997). These institutional landscapes,albeit weak (see Schweiket al., 1997, for more detail), still may have altered humanforaging behavior to some limited degree. In such an instance, the pattern in thelandscape would reflect a new optimal foraging calculus, where the decision toharvest or not to harvest at a particular location includes consideration of rules,rule penalties, and the likelihood of getting caught.
In the site description presented earlier, there exist two primary institutions thatappear to be somewhat influential in driving human decision making away fromwhat optimal foraging might predict: the monitoring practices along the road inMilan and the established social norms that exist between the Milan and Shaktikhorcommunities. In this case, DFO guard monitoring appears to be relatively inef-fective, with the possible exception of forested areas adjacent to the road throughMilan. This more-prevalent forest monitoring of DFO guards in the west couldadd more incentive for the Milan villagers to harvest up into higher regions ofSugubhanjyang and across the river in Latauliin locations that are not visible fromthe road. This, in conjunction with the interesting social dynamic we discovered –the unwritten or accepted rule that allows villagers of Milan to harvest in the LatauliForest but not vice versa – places added pressure on the eastern side of the LatauliForest. Thus, in a setting where both optimal foraging and these institutional-induced incentives are present, we would expect a landscape produced that reflectsmore of a continuous degradation of the forest – a depletion trend – as one movesfrom the west to east (Figure 5).
4. Hypothesis Testing with Traditional Forest Condition Measures
Can the traditional ‘aggregated’ analyses of forest plot conditions provide supportfor one of the three hypotheses indicated above? Figures 6–11 provide aggregateplot analyses for each of the three forests. Figures 6 and 7 present a comparison ofthe mean diameter at breast height (DBH) and the mean height ofShorea robustaand four other species deemed highly valuable by the villagers in the region:Nycth-anthes arbortristis(Parijat),Adina cordifolia (Karma),Lagerstroemia parviflora(Botdhainyero), andTerminalia tomoentosa(Saj). While there is some fluctuationin mean DBH between forests for particular species, nothing strikingly different isidentified in this comparison.
Figures 8–11 provide a comparison of absolute density, frequency, and dom-inance and species importance values of these species across the three forests.3
Across all indicators but mean height, Kaswang reflects a much higher presence of
3 Density provides a measure of the number of species present in a forest. It is determined bycounting the number of individual species and then dividing this by the total area of plots. Frequency
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 243
Figure 5.Patterns as a result of optimal foraging combined with geographic institutional influences.
Figure 6.Mean DBH for preferred forest product species.
Shorea robustacompared to the Sugabhanjyang and Latauli Forests. This supportsthe contention that Kaswang is subject to significantly less foraging as villagerssuggested in the field. We would expect aShorea robustaclimax forest, left re-
provides a measure of how widely a species is distributed within a forested area. It is calculatedby taking the number of plots in which a species occurs and dividing this by the number of plotssampled. Dominance provides a measure of the standing biomass a particular species contributes toa forest composition. Dominance is calculated by taking the total basal area of a species and dividingit by the area sampled. Finally, the importance values of each species reports the summation of therelativedensity, dominance, and frequency together divided by three.
244 CH. M. SCHWEIK
Figure 7.Mean height for preferred forest product species.
Figure 8.Absolute density (sp/ha) of important product species.
latively untouched by humans, to exhibit high values in these indicators for theShorea robustaspecies. But while it is clear that Sugabhanjyang and Latauli arecomprised of differing levels of vegetation than Kaswang, this is about the onlydefinitive conclusion we can make. We cannot easily identify from this one time-
Figure 9.Absolute frequency of important tree species.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 245
Figure 10.Dominance of important tree species.
Figure 11.Importance values for preferred tree species in the forests of Shaktikhor study area.
point aggregate data whether Latauli or Sugabanjyang follow patterns of optimalforaging in Figure 3b or optimal foraging coupled with institutional influences inFigure 3c. The lower measures found in Latauli and Sugabhanjyang could be aresult of purely biophysical differences such as topography. In other words, Lataulicould havealwaysexhibited lessShorea robustaindividuals than its neighboringforests.
5. Further Testing of the Hypotheses: Three Forest Plot Event-CountModels
The argument made earlier is that by giving extra consideration to spatial rela-tionships and testing for these factors we can improve our understanding of forestchange in instances when baseline forest condition data is unavailable. Three nes-ted models will be used to test the hypotheses articulated above using multivariateregression. Model 1 is associated with Hypothesis 1. It contains abiotic and bioticfactors considered important forShorea robustagrowth where human foragingdisturbance is minimal (Figure 3). Model 2 tests Hypothesis 2: patterns of humanoptimal foraging will be found in the distribution ofShorea robustaover the land-
246 CH. M. SCHWEIK
scape (Figure 4). Model 2 requires the control for abiotic and biotic variables ofModel 1 but has additional parameters included to capture the influence of humanoptimal foraging. Similarly, Model 3 tests Hypothesis 3: the geography of rules andenforcement of these rules (Figure 5) shift optimal foraging patterns to the easternside of Figure 2. Model 3 requires the control of all variables from Models 1 and2 plus a parameter capturing the institutional pressures. The three models, variableoperationalization, and expected signs are summarized in Table II.
5.1. THE DEPENDENT VARIABLE: A MEASURE OFSHOREA ROBUSTA
ABUNDANCE
Spatial analysis requires forest plots to be the unit of analysis rather than aggregateforest measures. We argued earlier that a focus on important forest product speciesis a useful endeavor for identifying deforestation in cross-sectional data sets. Giventhe extreme importance ofShorea robustato the villagers in these communities, thedependent variable for each of the models is a count of the number of this type oftree in a plot. A count provides a simple but useful measure of species abundance.
5.2. THE INDEPENDENT VARIABLES: FACTORS THAT INFLUENCE WHERE
SHOREA ROBUSTA EXISTS
5.2.1. Model 1 Variables: Abiotic and Biotic FactorsEach forest plot contains physiographic characteristics that influence the capacityfor particular species to grow in its environment. These abiotic characteristics in-clude plot steepness, aspect, elevation, and soil type and condition and can playa tremendous role in the number and type of vegetation that grows in a particularplot (Spurr and Barnes, 1992; Schreieret al., 1994).
Plot steepness is operationalized as a continuous variable measured in degrees.Figure 2 reveals the hilly topography of the site. In areas of extreme topographicvariation, we would expect that as slope in degrees goes up, the number ofShorearobustatrees in a plot will go down.
Aspect captures the direction a plot faces. In one study of forests in the middlemountain region of Nepal, Schreier and colleagues (1994: 148) assume high el-evation north-facing slopes to be moist and cool, and low elevation south-facingslopes to be hotter and dryer. But the sub-tropical forests studied here reside inthe southern Siwalik hill region of Nepal, and slope-sunlight differences may beless pronounced. It is not clear what role, if any, aspect plays in determining whereShorea robustaexist. It is included in Model 1 to determine whether it has aneffect and is operationalized as a categorical variable following the assumptionsof Schreieret al. (1994). A zero represents a north-facing slope; a one representseither a north-west or north-east facing slope; a two represents an east or west-facing slope; a three represents a south-west or south-east-facing slope, and a fourrepresents a completely south-facing slope.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 247
TABLE II
Three imbedded models, variable operationalization, and expected relationships
Steepness of plot. Steepness in degrees. (–) As steepness goes up,number ofShorea robustatrees goes down.
Aspect or orienta-tion of plot.
Degree of southness (0–4). A zero represents anorth-facing slope; a onerepresents either a north-west or north-east-facingslope; a two represents aneast or west-facing slope;a three represents a south-west or south-east-facingslope; and a four repres-ents a completely south-facing slope.
Unknown relationship; butpotentially important andshould be included in themodel.
Depth of A and Bsoil horizons.
Measured in centimeters. Unknown relationship,but potentially importantbased on work by Burtonet al. (1989).
Hypothesis2 variables(humanoptimalforaging)
Number of house-holds within a 1 kmradius of plot.
Count. (–) As number of house-holds increase, the num-ber ofShorea robustatreesshould decrease.
Average distance toplot of 1 km house-holds.
Average distance (inmeters) of householdswithin a 1 km radius ofplot.
(+) As average distanceincreases, the number ofShorea robustatrees inplot should also increase.
Plot elevation. Measured in meters. (+) As elevation increases,Shorea robustaalso in-creases.
Hypothesis3 variables(institutionalpressures)
X UTM coordinate. Measured in meters. UTMeasting coordinate. Sub-tracted the average to es-tablish a 0,0 coordinate inthe middle of the map.
(–) As one moves east, thenumber ofShorea robustatrees are expected to de-crease.
248 CH. M. SCHWEIK
Elevation is another parameter thought to influence where particular speciesgrow (Spurr and Barnes, 1992).Shorea robustais known to reside at elevationsas high as 1250 m (Storrs and Storrs, 1990). The highest plot taken in our sampleis 830 m. Therefore, theoretically, elevation is not needed in the model to capturealtitudinal effects onShorea robustabecause there should be none.
Soil nutrients, moisture, and physical composition also affect the character andgrowth of vegetation (Spurr and Barnes, 1992; Burtonet al., 1989). Four soilhorizons are typically analyzed: ‘O’ (humus or ground litter layer), ‘A’ (between 0–20 cm), ‘B’ (20–50 cm), and ‘C’ (>50 cm). Burton, Shah, and Schreier (1989: 398)studied soil conditions in degraded and undisturbed forests only a few kilometerssouth of this study site and report variation in A horizon samples but few differ-ences in C horizon samples. During our fieldwork, we collected A and B horizondepth in each plot. This is included in Model 1 to capture its potential influence onShorea robustagrowth. Just what relationship to expect between these soil depthsand existence ofShorea robustais unknown. Other estimates of soil condition,such as soil color and texture, were also collected, but exhibit little variation acrossplots.
Several biotic parameters were considered, but ultimately not included in Model1. The proximity of neighboringShorea robustaseed trees often determines whethera tree will grow in a particular plot. The seed of aShorea robustatree is winged(Storrs and Storrs, 1990) and these seeds can travel great distances by wind. There-fore, given that these are allShorea robustaclimax forests (see Figures 6–11), itis assumed that each forest plot has an equal likelihood of havingShorea robustaseed trees somewhere in its vicinity and that this factor need not be specified inModel 1.
Animal foraging may affect the fate of many seedlings. In contrast, species of nointerest to animals may continue to survive or even thrive. While animal foragingis probably an important parameter, the grazing of livestock is closely related to thelocation of households and therefore its influence will be captured through optimalforaging variables specified later. The influence of other animals is assumed to bea random event.
Competition from other species is another potential parameter for Model 1. Ameasure of competing biomass was included in earlier runs of this model. However,after further consideration it was determined that this variable follows more of asimultaneous relationship with the count ofShorea robustathan a causal factor. Forexample, the abundance of rival species may be the result of opportunities providedbecauseShorea robustawas harvested. For this reason, competing biomass is notincluded in Model 1.
5.2.2. Model 2 Variables: Human Optimal Foraging PressuresNow we turn to the challenge of how to best operationalize independent vari-ables that capture optimal foraging pressure on a plot. Significance of such vari-
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 249
ables would lend support to Hypothesis 2 (the pattern described in Figure 4). Fiveoperationalization options exist.
The ideal method to capture plot foraging efforts would be to measure thedistance and steepness of the trails leading to each plot from village centers orindividual household locations. This approach proved to be impractical becausethe trails on the current map do not accurately reflect what we witnessed in thefield. The topographic map in Figure 2 reveals only a few trails; yet from fieldexperience, we know there exists an elaborate series of trails within each forest.This operationalization approach was rejected.
A second plausible method would be to calculate a straight-line distance fromeach plot to the center of a village or villages. Using GIS functionality, this is a rel-atively straightforward task,if one can define village centers. This is extremely dif-ficult in this case, for households are scattered throughout the landscape (Figure 2).This approach was also rejected.
The third and fourth methods to quantify optimal foraging pressure involve de-veloping a count of the number of households within a certain distance from eachplot and developing an aggregate measure of how far these households are fromeach plot. A relatively accurate depiction of household locations existed on the1995 topographic map interpreted from recent aerial photographs. These householdpoint locations were digitized and some additional household points were addedfrom our field experience. While in some instances foragers may travel beyond onekilometer distance, the assumption is reasonable given what we witnessed in thefield and the hilly terrain. The Arc-InfoTM GIS ‘pointdistance’ function calculatedthe distance between each forest plot and each household point falling within a onekilometer search radius. An average distance for all houses within the one kilometercircle of the plot could then be calculated. Plots with a larger number of householdsin the one kilometer circle are expected to have low counts ofShorea robustatrees.We would also expect that the shorter the household average distance, the moreforaging pressure the plot is subjected to and the lessShorea robustatrees will befound.
The fifth and perhaps easiest method to capture a component of optimal for-aging pressure, relates back to the discussion of plot elevation. It was stated earlierthat there is no theoretical justification to include elevation in the model for purelybiophysical reasons. But elevation could be important from a human foraging per-spective because it captures the altitudinal harvesting effort required to get to andfrom a particular plot from household locations. Figure 2 shows that the majorityof the villagers in this region live in the lowlands near riverbeds. Optimal for-aging would predict that the villagers would tend to avoid making a trek fromthe riverbed to high altitude locations for tree products if at all possible. Fromthis discussion, the viable optimal foraging parameters for Model 2 are: (1) thenumber of households within a one kilometer radius capturing household pressure;(2) the average distance of households within the one kilometer radius to the plot,
250 CH. M. SCHWEIK
capturing a distance component of foraging effort; and (3) plot elevation, capturingthe altitudinal component of foraging effort.
5.2.3. Model 3 Variables: Institutional InfluencesThe question now turns to how to develop a variable that captures the effect ofreasonably well-enforced rules in the western side of Figure 2 and the more limitedenforcement of rules in the eastern side on optimal foraging patterns to operation-alize Hypothesis 3. One method for capturing this influence would be to assign acategorical variable to each plot that is our own assessment in the field of the like-lihood (e.g., high, medium, or low) that this plot is well monitored by DFO guards.This type of operationalization, however, is subjective and would lose preciousdegrees of freedom in statistical implementation. For these reasons, this methodwas rejected.
However, given that the predicted pattern is one whereShorea robustadepletesas one moves farther east, another technique is a possibility. Geographers haveapplied coordinate systems as independent variables – what is commonly referredto as trend surface models – to capture trends across landscapes. We can thereforeutilize the UTM plot coordinates collected by DGPS. Average X and Y coordinateswere calculated and subtracted from each point to establish a 0,0 origin-near thepoint where the Shakti and the Kayar Rivers converge. The X and Y coordinatesrepresent the distance, in meters, in a east-west and north-south direction from thisorigin. Hypothesis 3 anticipates a trend moving from west to east (Figure 5), soonly the X UTM coordinate of a plot is required. If the hypothesized eastern trendexists, X UTM is expected to exhibit a negative sign signifying depletion in theeastern direction. Model 3 includes this variable and controls for all other variablesspecified for Models 1 and 2.
6. Statistical Methods and Results
The nested models representing the rival hypotheses will be estimated and com-pared using multiple regression. The dependent variable, number ofShorea robustain a plot, takes on values from zero to some positive integer. Traditional OrdinaryLeast Squares (OLS) regression could be applied to estimate the influence of theindependent variables on this event count, but it has been shown that such anapproach yields inefficient, inconsistent, and biased estimates (King, 1988; Long,1997). Further, the OLS assumption of normally distributed residuals is incorrectwhen counts of biological phenomena are being estimated (Ludwig and Reynolds,1988). Scatterplots were made of each independent variable versus the dependentvariable and the results confirm nonlinear relationships. Several nonlinear countdata models (e.g., Poisson, negative binomial) are potentially more appropriate andthe choice depends on the distributional assumption made regarding the presenceor absence ofShorea robusta.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 251
Ludwig and Reynolds (1988) report that counts of species usually follow oneof three types of spatial arrangements: random, clustered, or uniform. In the caseof a random dispersal of species, each plot has an equal chance of hosting aShorearobusta individual. In such random patterns, the variance will be very close tothe mean in value and the Poisson distribution is appropriate (Ruser, 1991; Long,1997). The second pattern, a clustered pattern, is commonly found in biologicalstudies. Clustering will result and a large number of plots where noShorea robustaindividuals exist will be identified. The variance in a clustering pattern will begreater than the mean. In these instances of overdispersion, the negative binomialdistributional assumption is more appropriate. The third pattern often identifiedis a uniform pattern, where almost every plot exhibits the same number ofShorearobustaindividuals. In these spatial patterns, the variance will be less than the mean(Ludwig and Reynolds, 1988).
A ‘reference forest’ is required to identify the ‘natural’ distribution of theShorearobusta. A reference forest is a forest that (1) is an adequate representation ofthe other forests of interest and (2) is generally undisturbed by human activity.Kaswang satisfies the above two conditions. The Nepali foresters identified allthree forests asShorea robustaclimax forests. Moreover, Kaswang’s natural pro-tection by the river systems suggests that it is the least impacted by human activ-ities. The aggregate measures in Figures 6–11 confirm these suspicions. For thesereasons, the 31 forest plots sampled in the Kaswang Forest are treatedseparatelyas the reference forest to determine a natural distribution ofShorea robusta. Themultiple regression models then utilize the other 66 forest plot data from only theSugabhanjyang and Latauli Forests.
A variance-to-mean ratio or index of dispersion test (Ludwig and Reynolds,1988) helps determine the appropriate distributional assumption of theShorearobustacount for the Kaswang Forest (Table III). The value for the Chi-squaredstatistic (df 30) is larger than the critical value at the 0.01 probability level, im-plying thatShorea robustain natural settings follows a clumped pattern (varianceis greater than the mean). In such cases of overdispersion, the negative binomialdistribution is appropriate (King, 1989a,b; Long, 1997). Three negative binomialregression models, each representing the three rival hypotheses, will be compared.4
6.1. STATISTICAL RESULTS
Several variables are of unclear theoretical importance (e.g., aspect, A and B hori-zon depth, specific operationalization of optimal foraging variables). Consequently,
4 The question arises whether the dependent variable should be treated as a truncated or non-truncated variable and whether a negative binomial or a zero-inflated negative binomial is required(see Long, 1997: chap. 8). The dependent variable is a count of all trees with DBH >10 cm. Thenegative binomial regression model assumes that each plot has a positive probability of producinga tree >10 DBH. Given that these areShorea robustaclimax forests withShorea robustatrees ap-pearing everywhere, and that we are accounting for abiotic factors in the model, this is a reasonableassumption.
252 CH. M. SCHWEIK
TABLE III
Chi-square test of the index of dispersion ofShorearobustatrees in the Kaswang forest
Average number of individuals per plot 5.968
Number of plots 31
Variance 12.644
Index of dispersion (variance/mean ratio) 2.119
X2 statistic [X2 = ID(N-1)] 63.562∗
∗ Significant at the 99% level of confidence.
five models, not three, are actually estimated. The approach follows ideas posedby Leamer (1983, 1985) where model construction includes ‘focus’ and ‘doubt-ful’ variables. ‘Focus’ variables are those known to be theoretically important orof particular interest to the researcher. ‘Doubtful’ variables are those of possibleimportance but unclear based on prior work. All model results are provided inTable IV so that the reader can make his or her own comparative judgements.Discussions of Models 1A, 1B, 2A, and 2B will concentrate primarily on whetherinclusion of doubtful variables captures theoretical concepts correctly and shouldremain in the models. Interpretations of coefficients and model comparisons willbe left for the discussion of the full model, Model 3.
Models 1A and 1B represent two alternative specifications for the abiotic andbiotic factors argued in Hypothesis 1. In Model 1A, plot steepness is found tohave a negative influence on the existence ofShorea robustaspecies. The expec-ted relationship holds, and it is not surprising that this parameter is statisticallysignificant. It is difficult for larger trees to effectively root in steep terrain. Oneof the doubtful variables, plot aspect, is found to be not statistically significant.This is not surprising. We witnessedShorea robustaon all types of terrain facingall types of directions. Similarly, no statistically significant relationship appears toexist between another doubtful variable, depth of the A and B horizon, and thenumber ofShorea robustatrees in a plot.Shorea robusta, being the climax speciesof these forests, may be robust in its ability to grow in a variety of soils. From thisanalysis the doubtful variables, aspect and A and B horizon depth, do not appearto be important factors related toShorea robustagrowth. Consequently, a moreparsimonious Model 1B is estimated to test Hypothesis 1.5
Model 2A adds the three optimal foraging parameters and represents Hypo-thesis 2. In this model, elevation is statistically significant at the 99 percent level ofconfidence, and exhibits the expected sign. Given thatShorea robustais known to
5 Model 1B is surprisingly simple. But recall that many other abiotic and biotic factors were con-sidered for inclusion in the model and rejected (e.g., elevation, insect or animal damage, competingbiomass, etc.).
OP
TIM
AL
FO
RA
GIN
G,IN
ST
ITU
TIO
NS
AN
DF
OR
ES
TC
HA
NG
E253
TABLE IV
Negative binomial coefficients for three foraging models (Dependent variable – number ofShorea robustatrees in forest plots)
Independent Model 1A Model 1B Model 2A Model 2B Model 3
Numbers in parentheses are standard errors.∗∗∗ Significant at the 99 percent level of confidence;∗∗ Significant at the 95 percent level of confidence;∗ Significant at the 90 percent level of confidence.
254 CH. M. SCHWEIK
grow in elevations much higher than any plot in this study, the elevation parameterundoubtedly captures an effort component of optimal foraging. Average distanceof one-kilometer households to the plot has the expected sign and is significant atthe 90 percent level of confidence. This parameter measures the trekking distanceeffort required to move from a household to the plot and back. The third optimalforaging variable, the number of households within one kilometer of plot, is notstatistically significant and exhibits the opposite sign than is theoretically expec-ted. This variable was intended to capture the amount of household pressure ona particular plot but is probably an inadequate measure. Since it is calculated asa straight-line distance, it most likely does not adequately take into account theintricate trail network that exists. For this reason, this third variable is removedand Model 2B is estimated. This revised model represents the parameters requiredfor Hypothesis 2 (Figure 4) and will be revisited in later statistical tests betweenmodels.
Model 3 adds institutional influences (Hypothesis 3, Figure 5). Note that theother two models (1B and 2B) are imbedded and the X UTM coordinate is added tocapture the anticipated influence of a foraging shift to the eastern side of Figure 2.Plot steepness, elevation, and the X UTM coordinate are all statistically significantat the 99 percent confidence level in Model 3 and exhibit theoretically expectedsigns. One optimal foraging parameter, average household distance within onekilometer, becomes statistically insignificant in this model.
With Model 3 representing the full nested model, we can interpret coefficients.Caution is required given that the results from negative binomial regression cannotbe interpreted in the same manner as they would be if they were produced byan OLS regression. One of the most intuitive ways of interpreting these resultsis by creating the incident rate ratio (IRR). IRRs can be easily interpreted as apercentage of growth or decline in the dependent variable due to a one-unit changein the independent variable, controlling for everything else. Model 3 coefficientscan be interpreted as follows: Holding all else constant, a one-degree increase insteepness will result in a 3.87 (100∗[1-IRR] or 100∗[1–0.9613]) percent decrease inthe expected number ofShorea robustatrees. Similarly, every one-meter increase inelevation increases the expected number ofShorea robustatrees by 0.4% (100∗[1–1.004]). Lastly, the negative coefficient for the X UTM parameter suggests thatfor every one-meter shift east, the expected number ofShorea robustatrees willdecrease by 0.05% (100∗[1–0.9995]). In other words, holding abiotic, biotic, andother optimal foraging parameters constant, as one moves east on the map, thenumber ofShorea robustatrees found in plots decreases. The percentage decreaseis relatively small because one meter on the map is a fine unit.
To verify that no multicollinearity problems exist, pairwise correlation coeffi-cients were estimated for all parameters in Model 3 (Table V). Only the X UTM andthe average household distance variables show any potential signs of being related,and this relationship is not very strong. Given that they both are calculated fromthe GIS grid, it could be that a small component of the X UTM coordinate effect
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 255
TABLE V
Pairwise correlation coefficients for Model 3 parameters
Steepness Avg distance Elevation X UTM
of 1 km coordinate
households
Steepness 1.0000
Avg distance of –0.1840 1.000
1 km households –0.1840 1.000
Elevation 0.0184 0.0880 1.000
X UTM coordinate –0.0022 –0.6388 0.1331 1.000
is picking up a portion of the effects of the traditional optimal foraging process.However, that alone does not explain why the X UTM coordinate is so stronglyrelated – much more so than the average household one-kilometer distance variable– to the count ofShorea robustatrees.
Now to the key question: which hypothesis is confirmed? Which model bestdescribes the geographic pattern ofShorea robusta? Since Models 1B (Hypothesis1) and 2B (Hypothesis 2) are nested in Model 3 (Hypothesis 3), a likelihood ratiotest can be used to determine, quantitatively, which hypothesis is supported (seeLong, 1997: 93–97). First, a test was run to determine whether the data supportHypothesis 1 over Hypotheses 2 or 3. This tests whether the parameters for for-aging and institutional effects (average household distance, elevation, and X UTMcoordinate) are simultaneously zero. The results of the test confirm that Models2 and 3 improve explanatory power (LR X2 = 13.37, df = 3, p <0.01). Somepattern related to optimal foraging exists in the plot data. Next, a test can be madebetween parameters associated with Hypotheses 2 and 3. The test is that the XUTM coordinate’s coefficient is equal to zero. The results of this test confirm thatthe X UTM parameter also improves explanatory power (LR X2 = 6.51, df = 1, p<0.05). The coefficient in support of Pattern 3 is supported statistically.
7. Discussion
7.1. SUBSTANTIVE FINDINGS
Earlier, three rival hypotheses were presented. Hypothesis 1 suggested that thedistribution ofShorea robustashould be one of regrowth with no signs of humandisturbance (Figure 3). Hypothesis 2 suggested that the distribution ofShorea ro-bustashould exhibit patterns of human disturbance and would be best predictedby optimal foraging theory (Figure 4). Hypothesis 3 posed a different scenario,
256 CH. M. SCHWEIK
suggesting that the distribution ofShorea robustawould reflect optimal foragingaltered by the geographic pattern of rule enforcement in the region (Figure 5). Thestatistical results above support Hypothesis 3.
Of the several abiotic and biotic factors thought to influence the existence ofShorea robustain this study site, only plot steepness is statistically significant. Thesoil A and B horizon depth provided little explanatory power. Other soil measures,such as soil color and texture (to get at concepts such as soil moisture), were appliedin earlier analyses not shown and were also found to have little influence. Thiseither means that other soil measures are required (e.g, soil nutrients) or thatShorearobustacan grow in a variety of soil conditions in this region.
The importance of optimal foraging parameters associated with Hypothesis 2suggests that overharvesting in patterns anticipated by optimal foraging are occur-ring in the region. The rate of harvesting does appear to be outpacing the rate ofShorea robustaregrowth in areas within reasonable foraging access. While twoparameters pick up some of this influence in Model 2B, it is elevation that provesto be a very important variable in determining the number ofShorea robustatreesin a plot. Because this tree species is known to grow in elevations much higher thanthe highest plot of 830 m, the best explanation for why this variable is so importantstatistically is the altitudinal effort component of optimal foraging. It takes greateffort to trek up to high elevations and bring harvestedShorea robustatrees down.
Hypothesis 3, the institutional-related influences of optimal foraging, is sup-ported by the significance of the X UTM parameter in Model 3. Experience fromthe field led us to the hypothesis that more effective western forest monitoringand sanctioning, along with social hierarchical structure, lead villagers to foragefurther east than they would if the institutional structure were not there. The XUTM parameter identifies a depleting trend as one moves west to east toward theeastern side of the Latauli Forest and supports this contention. This suggests thatthe eastern side of the Latauli Forest in Figure 2 has been subject to higher levelsof Shorea robustaharvesting than the other forested areas.
The reader could argue that the identified trend might not be institutionallydetermined but rather be a result of some other still unidentified process. If onetakes this view, the question then turns to what better explanation of the trend isthere? In this analysis, every attempt has been made to control for abiotic, biotic,and optimal foraging parameters that explainShorea robustagrowth. I can think ofonly one other plausible explanation.
The rival explanation brings us back to the contention that differences in soilcondition are not adequately measured in this study and this may be the unknownprocess driving the trend. However, when considering this explanation, we run intothe classic ‘which came first, the chicken or egg’ dilemma. In their study ofShorearobustaforests only a few kilometers away from our study site, Burton, Shah, andSchreier (1989) report a relatively uniform soil base (e.g., C horizon) in the region.Their site is, in a biophysical sense, quite representative of this study area andtherefore, it is unlikely that differences in C horizon cause the trend. The authors
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 257
also report differences between degraded and nondegraded forest A and B soilhorizons in the region. It may be inadequately measured soil condition differences(e.g., nutrient content) in plot A and B horizons that explain this trend. But thestudy by Burtonet al. (1989) suggests that the opposite relationship exists. Theyreport differences in soil condition measures that are related to where forests havebeen degraded by human action. In other words, their work suggests that whereShorea robustais depleted, soil quality will be diminished and not the other wayaround. This suggests that missing soil condition variables probably do not explainthe identified trend. We are left with the institutional component providing the bestexplanation for the trend phenomenon.
7.2. METHODOLOGICAL IMPLICATIONS
While ecologists and biologists have made tremendous advances in the study ofthe spatial distribution of various plant species, to my knowledge, this is the firstanalysis of its kind that applies recent technological advances of DGPS, GIS, anda spatial statistical technique to this effort. This study provides an example of howthe inclusion of a spatial influential variable in a regression model may assist inunderstanding the human dimensions of forest change when longitudinal data isnonexistent. The findings support the earlier claim that a plot level of analysis mayreveal findings that would not be discovered at the forest level of analysis.
Moreover, this study may also be the first of its kind to apply an institutionalanalysis to the study of the distribution of a particular species over space. In anyforaging setting, the first signs of forest depletion will be changes in geographicpattern of particularly important product species as humans base their decisionsand actions on the attributes of the physiographic, institutional, and communitynorms related to use of that forested area. After taking into consideration the naturaldistribution of a particular species, and accounting for physiographic influencesthat encourage or discourage growth, the analyst can study the existing pattern toreveal human response to past and present institutional arrangements. In this case,such an analysis provides evidence that monitoring and social norms produce shiftsin foraging patterns away from traditional optimal foraging. Such evidence couldnot be discovered with an analysis of aggregated cross-sectional forest conditiondata.
The study raises the important problem of how we link spatial statistical analysisto the study of the geographic component of institutions. Rules and levels of theirenforcement have geographic properties. In forest and watershed management, animportant policy question is how the spatial configuration of institutions changesthe spatial configuration of human action, which then leads to changes in the geo-graphic properties of the natural resource. Granted, the X coordinate in this studyis not the ideal parameter to capture the ‘institutional pressure’ concept, but forstatistical purposes in this study, it was the best parameter available. The adventof GIS, GPS, and spatial statistical procedures brings forward new methodological
258 CH. M. SCHWEIK
and measurement challenges to the social sciences. One challenge is how to movebeyond what has been done here? How can we improve our methods for under-standing the spatial influence of rules on human behavior? A second importantchallenge we now face is how to measure the degree to which rules are followedand enforced that moves beyond a simple binary variable toward more of a continu-ous variable. Finally, a third and more problematic issue is how we can successfullyinventory sets of potentially complex rule systems over broad landscapes. It mayrequire empirical sampling of rules over geographic space.
Overcoming these challenges is important if we wish to understand the geo-graphic ramifications of institutional or policy design. It is certain that when therule and monitoring mechanisms were designed for this area of southern Nepal,it was not the intention to cause a harvesting shift into eastern areas deeper andhigher into a watershed. Technological and statistical tools, such as the ones usedhere, provide the opportunity to devise methods that help us verify the performanceand geographic implications of our institutional configurations in natural resourcemanagement-something that has not been easily conducted before. I hope this workinspires others to explore the utility of such methods.
Acknowledgements
I am very appreciative of the support received from the Ford Foundation, Dr. JohnAmbler, and Dr. Ujjiwal Pradhan, as well as the financial and intellectual supportprovided by Elinor Ostrom at the Workshop in Political Theory and Policy Ana-lysis, Indiana University, Bloomington. The Forests, Trees and People Programmeof the Food and Agriculture Organization was also quite helpful through its supportof the IFRI research program. Special thanks go to Rajendra Shrestha, Bharat ManiSharma, Mukunda Karmacharya, Vaskar Thapa, and Sudil Gopal Acharya for theirefforts in data collection. I am also grateful to K. N. Pandit, K. R. Adhikari, A. K.Shukla, Ganesh Shivakoti, and the Institute of Agriculture and Animal Science inRampur, Chitwan, for their assistance in the field. I am indebted to Dusty Becker,Erling Berge, Brenda Bushouse, Clark Gibson, Joby Jerrells, Robert and JoanneSchweik, John Williams, and two anonymous reviewers for extremely helpful com-ments on earlier drafts. Thanks as well to Robin Humphrey and Julie England whoprovide ongoing database and technical support with amazingly good cheer, andto Patty Dalecki, for her outstanding editorial assistance. Most of all, I am deeplygrateful for the openness and warmth given by residents of the Shaktikhor VDC inChitwan, Nepal, during our days in the field.
References
Aldhous, P.: 1993, ‘Tropical deforestation: Not just a problem in Amazonia’,Science259, 1,390.
OPTIMAL FORAGING, INSTITUTIONS AND FOREST CHANGE 259
Angelsen, A.: 1995, ‘Shifting cultivation and ‘deforestation’: A study from Indonesia’,WorldDevelopment23(10), 1713–1729.
Ascher, W.: 1995,Communities and Sustainable Forestry in Developing Countries,ICS Press, SanFrancisco, California.
Bruijnzeel, L. A. and Bremmer, C. N.: 1989, ‘Highland-lowland interactions in the GangesBrahmaputra river basin: A review of published literature’, ICIMOD Occasional Paper No. 11.ICIMOD, Kathmandu, Nepal.
Burton, S., Shah, P. B. and Schreier, H.: 1989, ‘Soil degradation from converting forest land intoagriculture in the Chitwan district of Nepal’,Mountain Research and Development9(4), 393–404.
Feit, H. A.: 1973, ‘The Ethno-Ecology of the Waswanipi Cree, or How Hunters can Handle theirResources’, in: Cox, B. (ed.),Cultural Ecology, McClelland and Stewart, Toronto.
Hayden, B.: 1981, ‘Subsistence and Ecological Adaptations of Modern Hunter-Gatherers’, in: Hard-ing, R. S. O. and Teleki, G. (eds),Omnivorous Primates, Columbia University Press, NewYork.
IFRI (International Forestry Resources and Institutions) Research Program: 1994,IFRI Data Col-lection Forms of the Shaktikhor Site, Nepal, Workshop in Political Theory and Policy Analysis,Indiana University, Bloomington.
Keller, M., Clark, D. A., Clark, D. B., Weitz, A. M. and Veldkamp, E.: 1996, ‘If a tree falls in theforest...’,Science273, 201.
King, G.: 1988, ‘Statistical models for political science event counts: bias in conventional proced-ures and evidence for the exponential Poisson regression model’,American Journal of PoliticalScience32, 838–863.
King, G.: 1989a, ‘Variance specification in event count models: from restrictive assumptions to ageneralized estimator’,American Journal of Political Science33, 762–784.
King, G.: 1989b, ‘Event count models for international relations: generalizations and applications’,International Studies Quarterly33, 123–147.
Leamer, E. E.: 1983, ‘Let’s take the con out of econometrics’,American Economic Review73(1),31–43.
Leamer, E. E.: 1985, ‘Sensitivity analysis would help’,American Economic Review75(3), 308–313.Long, J. S.: 1997,Regression Models for Categorical and Limited Dependent Variables, Sage
Publications, Thousand Oaks, California.Lovejoy, T. E.: 1980, ‘A Projection of Species Extinctions’, in: Barney, G. O. (ed.),The Global 2000
Report to the President: Entering the 21st Century, Council on Environmental Quality, U.S.Government Printing Office, Washington, D.C.
Ludwig, J. A. and Reynolds, J. F.: 1988,Statistical Ecology. Wiley, New York.McKean, Margaret A.: 1992, ‘Management of Traditional Common Lands (Iriaichi ) in Japan’, in:
Bromley, D. W.et al. (eds),Making the Commons Work: Theory, Practice, and Policy, ICS Press,San Francisco, California.
Metz, J.: 1990, ‘Forest-product use in upland Nepal’,The Geographic Review80(3), 279–287.Morrow, C. E. and Hull, R. W.: 1996, ‘Donor-initiated common pool resource institutions: The case
of the Yanesha forestry cooperative’,World Development24(10), 1641–1657.Myers, N.: 1988, ‘Tropical Forests and their Species: Going, Going...’, in: Wilson, E.O. and Peter,
F.M. (eds),Biodiversity, National Academy Press, Washington, D.C.Norton, B. J. (ed.): 1986,The Preservation of Species, Princeton University Press, Princeton, N.J.Ostrom, E.: 1990,Governing the Commons: The Evolution of Institutions for Collective Action,
Cambridge University Press, New York.Ostrom, E. and Wertime, M. B.: 1995, ‘IFRI research strategy’, Working paper, Workshop in Political
Theory and Policy Analysis, Indiana University, Bloomington.Pace, S.et al.: 1995,The Global Positioning System: Assessing National Policies, Rand Corporation,
Santa Monica, California.
260 CH. M. SCHWEIK
Pickett, S. T. A. and Candenasso, M. L.: 1995, ‘Landscape ecology: spatial heterogeneity inecological systems’,Science269, 331–334.
Reid, W. V. and Miller, K. R.: 1989,Keeping Options Alive: The Scientific Basis for ConservingBiodiversity,World Resources Institute, Washington, D.C.
Repetto, R.: 1988,The Forest for the Trees? Government Policies and the Misuse of Forest Resources,World Resources Institute, Washington, D.C.
Richards, J. F. and Tucker, R. P. (eds): 1988,World Deforestation in the Twentieth Century, DukeUniversity Press, Durham, N.C.
Ruser, J. W.: 1991, ‘Workers’ compensation and occupational injuries and illnesses’,Journal ofLabor Economics9(4), 325–350.
Schreier, H., Brown, S., Schmidt, M., Shah, P., Shrestha, Nakarmi, G., Subba, K. and Wymann,S.: 1994, ‘Gaining forests but losing ground: A GIS evaluation in a himalayan watershed’,Environmental Management18(1), 139–150.
Schweik, C.: 1997, ‘The Spatial Analysis of Natural Resources in East Chitwan, Nepal: ConceptualIssues and a Multi-Scale Research Program’, in: Shivakoti, G.et al. (eds),People and Participa-tion in Sustainable Development: Understanding the Dynamics of Natural Resource Systems,Proceedings of an International Conference held at the Institute of Agriculture and AnimalScience, Rampur, Chitwan, Nepal, March 17–21, 1996, pp. 219–234.
Schweik, C. M., Adhikari, K. and Pandit, K. N.: 1997, ‘Land-cover change and forest institutions:A comparison of two sub-basins in the southern Siwalik hills of Nepal’,Mountain Research andDevelopment17(2), 99–116.
Shrestha, R.: 1996, IFRI Team leader, Personal communication (June).Smith, E. A.: 1983, ‘Anthropological applications of optimal foraging theory: a critical review’,
Current Anthropology24(5), 625–52.Spurr, S. H. and Barnes, B. V.: 1992,Forest Ecology,Krieger, Malabar, Fla.Storrs, A. and Storrs, J.: 1990,Trees and Shrubs of Nepal and the Himalayas, Pilgrims Book House,
Kathmandu.Task Force on Global Biodiversity, Committee on International Science: 1989,Loss of Biological Di-
versity: A Global Crisis Requiring International Solutions,National Science Board, Washington,D.C.
Thomson, J. T., Feeny, D. and Oakerson, R. J.: 1992, ‘Institutional dynamics: the evolution anddissolution of common-property resource management’, in: Bromley, D. W.et al. (eds),Makingthe Commons Work: Theory, Practice, and Policy,ICS Press, San Francisco, California.
Umans, L.: 1993, ‘The unsustainable flow of himalayan fir timber’,Mountain Research andDevelopment13(1), 73–88.
Winterhalder, B.: 1993, ‘Work, resources and population in foraging societies’,Man28, 321–340.
