Wu Group Pnas20130

6 pages

Please download to get full document.

View again

of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Wu Group Pnas20130
  Nonlinear effects of group size on collective actionand resource outcomes Wu Yang a,1 , Wei Liu a , Andrés Viña a , Mao-Ning Tuanmu a , Guangming He a , Thomas Dietz a,b , and Jianguo Liu a,1 a Center for Systems Integration and Sustainability, Department of Fisheries and Wildlife, Michigan State University, East Lansing, MI 48823;and  b Environmental Science and Policy Program, Department of Sociology and Animal Studies Program, Michigan State University, East Lansing, MI 48824Edited by Bonnie J. McCay, Rutgers, State University of New Jersey, New Brunswick, NJ, and approved May 21, 2013 (received for review January 26, 2013) Fordecades,scholarshavebeentryingtodeterminewhethersmallorlarge groups are more likely to cooperate for collective action andsuccessfully manage common-pool resources. Using data gatheredfromtheWolongNatureReservesince1995,weexaminedtheeffectsof group size (i.e., number of households monitoring a single forestparcel) on both collective action (forest monitoring) and resourceoutcomes (changes in forest cover) while controlling for potentialconfounding factors. Our results demonstrate that group size hasnonlinear effects on both collective action and resource outcomes,withintermediategroupsizecontributingthemostmonitoringeffortand leading to the biggest forest cover gain. We also show howopposing effects of group size directly and indirectly affect collectiveaction and resource outcomes, leading to the overall nonlinearrelationship. Our  󿬁 ndings suggest why previous studies have ob-served differing and even contradictory group-size effects, and thushelp guide further research and governance of the commons. The 󿬁 ndings also suggest that it should be possible to improve collectiveaction and resource outcomes by altering factors that lead to thenonlinear group-size effect, including punishing free riding, enhanc-ing overall and within-group enforcement, improving social capitalacrossgroupsandamonggroupmembers,andallowingself-selectionduring the group formation process so members with good socialrelationships can form groups autonomously. casual inference  |  commons governance  |  ecosystem services  | biodiversity conservation  |  sustainability G roups are basic units for collective action and may achieveoutcomes that individual efforts cannot (1). However, thethreat of free riding implies that the optimal amount of collectiveaction does not always occur, and has led to a substantial literaturetrying to understand what factors facilitate or block the emergenceof collective action. Because collective action is needed to managemany common-pool resources, understanding the mechanisms thatshapecollectiveactionandresourceoutcomesisacriticalchallengefor sustainability (2, 3).From Pareto in 1906 (4) and especially since the in 󿬂 uential work by Olson in 1965 (5), group size has been hypothesized asa crucial factor affecting collective action and resource outcomes.(We note that Olson used an unusual de 󿬁 nition of   “ group size ” :the potential number of group members. Here we follow con- ventional practice and consider the actual number of partic-ipants.) However, the debate on group-size effect continues withsome researchers arguing that it is linear and negative (5 – 7),others arguing for linear and positive (8 – 11), and still othersinsisting it is curvilinear (12 – 14), ambiguous (1, 15 – 17), or non-signi 󿬁 cant (18 – 20). Even in the most recent work (8, 15, 19, 21 – 24), a consensus on the nature of the effect or even its existencestill remains elusive.Previous literature indicates that there are two hypotheticalopposing forces through which group size affects collective actionand resource outcomes (Fig. 1). Group members play differentroles in collective action, ranging from free riders (i.e., members who enjoy group bene 󿬁 ts without paying for the costs) and con-ditionalcooperators(i.e.,memberswhowillcontributemorewhenothers contribute more) to altruists (i.e., members who contributeregardless of others ’  behaviors), as well as various roles mixingthese strategies (25). Group size can have diverse effects. On theone hand, members tend to free ride as the group becomes larger(5, 26). As group size increases, transaction costs (e.g., commu-nication costs, costs of monitoring to maintain a necessary level of excludability) may rise sharply (1, 7, 13 – 15); thus, the larger thegroup, the more dif  󿬁 cult to detect and reduce free riding. If thecommon good has any degree of rivalry, average individual payoff  will shrink as group size increases, which further aggravates freeriding (15 – 17). On the other hand, small groups often lack theresources (e.g., labor, time, funds) that large groups can deploy (7, 13, 14, 27). When available resources are limited, it is dif  󿬁 cultto devote additional resources to collective action (1, 15). Takingadvantage of more resources, large groups may enhance enforce-mentthroughmonitoringandpunishmenttoreducefreeridersandthus improve collective action and resource outcomes (13, 14, 20,21, 24, 28). Ostrom scrutinized previous evidence and pointed outthe problem of focusing on group size itself without consideringfactors that in 󿬂 uence or are in 󿬂 uenced by group size (7). Ostromthen suggested further research to focus on the hypothesized cur- vilinear effects of group size (7). A few previous studies qualitatively described the curvilinear ornonlinear effects of group size (12, 26, 29), and some claimedanonlinearrelationshipbysimplyplottingcollectiveactionagainstgroup size without controlling other factors (13, 14). However,none has provided a quantitative analysis of   󿬁 eld evidence whilecontrolling potential confounding factors, as suggested by Ostrom(7). Furthermore, there is little empirical examination of themechanisms of nonlinear group-size effects, which is essential toguide commons governance.To  󿬁 ll these knowledge gaps, we used empirical data from ourlong-term studies (30 – 44) in Wolong Nature Reserve, SichuanProvince, China (N 30°45 ′  –  31°25 ′ , E 102°52 ′  –  103°24 ′ ) (Fig. 2).Wolong Nature Reserve is home to  ∼ 10% of the total wild giantpanda (  Ailuropoda melanoleuca ) population, and home to  ∼ 4,900localhumanresidentsdistributedin ∼ 1,200households.Inresponseto degradation of forest and panda habitat because of human ac-tivities since the 1970s (31), the Reserve implemented the NaturalForest Conservation Program (NFCP) in 2001. NFCP is a nation- wide conservation program that aims to conserve and restore nat-ural forests through logging bans, afforestation, and monitoring,using a payments-for-ecosystem-services scheme to motivate con-servation behavior (45). Of the total  ∼ 120,500 ha in the NFCPmonitoring area in Wolong,  ∼ 40,100 ha were assigned to  ∼ 1,100rural households and the remaining areas were monitored by thestaff of the reserve ’ s administrative bureau. Meanwhile, the bureau Author contributions: W.Y., W.L., and J.L. designed research; W.Y., W.L., A.V., and G.H.performed research; W.Y., A.V., M.-N.T., and T.D. analyzed data; and W.Y., W.L., A.V., M.-N.T.,G.H., T.D., and J.L. wrote the paper.The authors declare no con 󿬂 ict of interest.This article is a PNAS Direct Submission. 1 To whom correspondence may be addressed. E-mail: yangwu1201@gmail.com or liuji@ msu.edu.This article contains supporting information at www.pnas.org/lookup/suppl/doi:10.1073/ pnas.1301733110/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1301733110 PNAS Early Edition  |  1 of 6       S      U      S      T      A      I      N      A      B      I      L      I      T      Y      S      C      I      E      N      C      E  set two timber checkpoints at the two ends of the only main roadcrossingthereserve(Fig.2).Thecommon-poolresourceinquestionin the Reserve is the forest (an essential component of the pandahabitat) assigned to households. Because logging is largely the ac-tion of local residents ( SI Appendix , Section 2.4.1), collective action(i.e., forest monitoring) has the potential to reduce illegal loggingand improve resource outcomes (i.e., changes in forest cover).The bureau administering the NFCP has assigned the forestparcels to household groups of various sizes ranging from 1 to 16( SI Appendix , Table S2). Parcels distant from households wereassigned to large groups with slightly higher payments ( SI Ap- pendix , Table S2). Households could not choose which parcel tomonitor or in which household groups to participate. Our anal- yses indicate that the distance from a household to its monitoredparcel and NFCP payment do not affect the group-size effects ( SI  Appendix , Section 2.4.3). Thus, the current distribution of groupsize is suitable for examining the group-size effects and mecha-nisms. Each assigned household group decides autonomously onits monitoring strategies (e.g., monitoring frequency, duration,and whether to subdivide to monitor in turns). The bureau eval-uates the monitoring performance based on  󿬁 eld assessments of illegal activities (e.g., logging) and rewards people who reportillegal activities (in cash). All households within a group share thesame monitoring responsibility and suffer the same payment de-duction when any illegal activities are detected by the bureau intheir comonitored parcel. However, the households are exemptfrom penalties if they report lawbreakers, in which case the cor-responding lawbreakers are punished instead.To understand the group-size effects and the underpinningmechanisms, we combined data on characteristics of households,household groups, and monitored parcels ( SI Appendix , Section 1).We acknowledge that con 󿬂 icts with regard to monitoring mightoccurwithinahousehold,butbecausethepolicyisdesignedtotreathouseholds — not individuals — as monitoring units, the commonpractice of treating households asthe unit ofanalysis is appropriatehere. We measured household monitoring efforts by the totalamount of labor input (one unit of labor input is de 󿬁 ned as onelaborer working for 1 d) ( SI Appendix , Section 2.1) through surveys.Wemeasuredresourceoutcomes aschangesinforest coverderivedfrom previously published forest-cover maps ( SI Appendix , Section1.1.1). We also measured factors that might explain the mecha-nisms,includingfree riders(i.e.,households thatdidnotparticipatein monitoring), the level of within-group enforcement (i.e., strongenforcement if there are punishment measures for free-ridingmembers within the group; otherwise, weak enforcement), and within-group division (i.e., whether groups divide into subgroups toconduct monitoring in turns) ( SI Appendix , Section 2). Some othercontextual factors shown in previous studies to affect group size,collective action, or resource outcomes were used as control vari-ables ( SI Appendix , Section 2.3). Results Our results show that group size has a nonlinear effect on themonitoring efforts per household, with an intermediate groupsize contributing the most (Fig. 3  A  and Table 1). These resultsare consistent whether or not we include the households whomonitored parcels individually (i.e., group size of one) and whenusing different combinations of control variables ( SI Appendix ,Table S13). The effect peaks at a size of eight or nine house-holds, where a household spends 9.2 labor units per year moni-toring its forest parcel. Our results also indicate that some otherfactors besides group size matter substantially. The level of socialties to local leaders has a signi 󿬁 cantly negative effect on perhousehold monitoring efforts (Table 1). When all other variablesare at their mean values, households with strong social ties tolocal leaders on average input 54% less labor units than house-holds with weak social ties to local leaders. Our experience in theReserve helps explain this effect. The staff members in the ad-ministrative bureau who are in charge of combatting illegallogging activities are hired from outside the Reserve, and anyonecan report illegal logging and receive a cash reward from theadministrative bureau. We are also not aware of a single case in which staff members turned a  “ blind eye ”  to illegal logging sohouseholds with strong ties could avoid monitoring or sanctions.Rather, additional analyses ( SI Appendix , Section 2.4.2) revealthat, compared with households with weak social ties to localleaders, households with strong social ties often have more socialrelationships, power, knowledge, and experience. Our extensive 󿬁 eldwork experience at the site indicates that these social tiesprovide social capital and reputation that discourages othersfrom conducting illegal activities in their monitoring parcels, andthus reduce the need for them to spend efforts on formal mon-itoring. The distance between each household and the main roadhas a positive effect on a household ’ s monitoring efforts, with Fig. 1.  Hypothetical effects of free riding, within-group enforcement, andgroup size on collective action and resource outcomes. Both free riding andwithin-group enforcement are hypothesized to be positively related togroup size. However, free riding is hypothesized to be negatively related towithin-group enforcement. The combined effects of free riding and within-group enforcement on collective action and resource outcomes are notexpected to be additive because of interactions between within-group en-forcement and free riding. The net effect of group size is determined by thedynamics (e.g., strength and variation with group size) of free riding andwithin-group enforcement, which may form a nonlinear pattern. Fig. 2.  Map of the location, main road, forest cover in 2007, and householdmonitoring parcels of Wolong Nature Reserve in Sichuan Province, China. 2 of 6  |  www.pnas.org/cgi/doi/10.1073/pnas.1301733110 Yang et al.  distant households doing more monitoring (Table 1). The aver-age household that lives 1 km further from the main road onaverage spends 33% more labor units in forest monitoring. Ad-ditional analyses ( SI Appendix , Section 2.3) suggest that house-holds far from the main road are closer to the parcels they monitor (Spearman ’ s  ρ  =  –  0.201,  P   <  0.05).Our results demonstrate that group size also has a nonlineareffect on changes in forest cover, with an intermediate group sizeleading to the biggest gain (Fig. 3  B  and Table 2). These resultsare consistent whether we include the parcels monitored by single households (i.e., group size of one) or not ( SI Appendix ,Section 2.5.2). The effect peaks at a size of nine households where the forest cover increases 15.8% in comparison with thereference level in 2001. The effects of slope, wetness, initialforest cover in 2001, and spatial error correlation are also sig-ni 󿬁 cant (Table 2).We accounted for as many as possible alternative explanationsof the observed nonlinear group-size effects based on systematicquantitative and qualitative analyses. No factor other than groupsize seems to account for the observed nonlinear effects. First,correlation tests ( SI Appendix , Table S2) show that except for thetwo criteria used for household group assignment (see details in SI Appendix , Section 1.2) by the administrative bureau (i.e.,distance between each household and its assigned parcel andreceived NFCP payment), no other factors were signi 󿬁 cantly associated with group size and thus are implausible as possiblealternative explanations for the group-size effects. We used twoadditional approaches to ensure that the observed nonlineareffects were not caused by the two criteria used for householdgroup assignments ( SI Appendix , Section 2.4.3). We examinedthe associations between the two criteria used for householdgroup assignment and household monitoring efforts, and we A B Fig. 3.  The nonlinear group-size effects on collective action and forest outcomes. This  󿬁 gure shows the predicted monitoring effort (  A ) and forest-coverchange ( B ) from 2001 to 2007 under different group sizes (i.e., number of households monitoring a single forest parcel). The graphs show the net effects ofgroup size on per household monitoring effort and on change in forest cover, while controlling the other variables in Tables 1 and 2. The blue line is thepredicted 󿬁 t based on group size, and the orange dots are the actual observations. One dot may represent several overlapping observations. Except for linearand quadratic terms of group size, all other independent variables were controlled as their mean values ( SI Appendix  , Tables S1 and S3). In  B  our conclusionstill holds as the nonlinear effect is still signi 󿬁 cant even when excluding the parcels with group size of one, or the two parcels with group sizes of 15 and 16(see details in  SI Appendix  , Section 2.5.2). However, for  A  and  B , the observations do not visually 󿬁 t the predicted lines in the same way as the observations inordinary least-squares regressions (54) because these models are not ordinary least-squares regressions (see details in  SI Appendix  , Section 2.5). Table 1. Coef 󿬁 cients of the Tobit model for the nonlinear effect of group size on collective action Variable Coef 󿬁 cients (robust SE) Marginal effectsIntercept 8.921*** (2.360)  — Quadratic term of group size  − 0.128** (0.041)  — Group size 1.331** (0.408) 0.767Social ties to local leaders (binary: 0 for weak social ties; 1 for strong social ties)  − 5.377** (1.920)  − 3.012Distance between each household and the main road 2.787* (1.216) 1.749Additional controls Not signi 󿬁 cant ( SI Appendix, Table S9 )  — Unit of analysis is the household. Dependent variable is total labor input for monitoring per year. Additional controls include household size, number ofhousehold laborers, education of adults, household income, and percentage of agricultural income ( SI Appendix  , Table S9). Log pseudolikelihood is  – 390.962.Total number of observations is 156. Independent variables were mean centered before entering the model. * P   <  0.05, ** P   <  0.01, *** P   <  0.001. Yang et al. PNAS Early Edition  |  3 of 6       S      U      S      T      A      I      N      A      B      I      L      I      T      Y      S      C      I      E      N      C      E  estimated two-step Tobit models of monitoring effort. Usingeither approach, all hypothesized alternatives to group size werelinearly associated with household monitoring efforts, and thuscould not lead to the observed nonlinear effects.Our path analysis (Table 3) con 󿬁 rms that group size has effectsthrough the two opposing forces (Fig. 1). If the balance betweenpositive and negative effects shifts with group size, it can yield theobserved nonlinear pattern. On the one hand, group size hasa signi 󿬁 cantly positive effect on the probability of a household freeriding (  P   <  0.01) (Table 3). With all other relevant factors con-trolled at their mean values, an increase of group size by onehousehold increases the free-riding probability by 15%. On theother hand, group size has a signi 󿬁 cantly positive effect on within-group enforcement (  P   <  0.01), which signi 󿬁 cantly reduces freeriding (  P   <  0.01) (Table 3). Again, controlling all other relevant variables at their mean values, an increase in group size by onehouseholdstrengthenswithin-groupenforcementby10%,whereasa shift from weak to strong within-group enforcement reduces freeriding by 52%. Additional analyses ( SI Appendix , Section 2.4.4)suggest that as groups become larger, a group member would facehigher pressure of deteriorating social relationships with the othermembers in each group, which enhances within-group enforce-ment and thus reduces free riding. This result is consistent with thesigni 󿬁 cant effect of social ties on household monitoring efforts(Table 1), indicating that social capital plays an important role inaffecting conservation behaviors of households. It follows thatcollective action might be easier to maintain when social rela-tionships among group members are improved or members withgood social relationships can form their groups autonomously. Discussion The coexistence of two opposing forces may also explain why previous studies found different group-size effects. If, as we ar-gue, the net effect of group size is determined by the dynamics(e.g., strength and variation with group size) of the two opposingforces, the optimum point of the net effect (or the necessary range of group size to observe a nonlinear effect) would be de-pendent on the context (14). The range of group size in our study area may appear to be small. However, the nonlinear pattern weobserved means that such a range is large enough to exhibit thenonlinear effect in our context. One of the reasons we  󿬁 nd sucheffects with only moderate variation in group size may be be-cause our study area is a  󿬂 agship nature reserve for giant pandas. As a result, the local administrative bureau has relatively abun-dant resources to allocate payments for household groups tomonitor parcels and evaluate their performance biannually  Table 2. Coef 󿬁 cients of the spatial autoregressive error model for the nonlinear effect of groupsize on resource outcomes Variable Coef 󿬁 cients (SE)Intercept 0.146*** (0.015)Quadratic term of group size  − 1.056E-03* (4.800E-04)Group size 7.205E-03* (3.643E-03)Slope 0.339** (0.121)Wetness 0.048*** (0.012)Initial forest cover in 2001  − 0.269*** (0.030)Additional controls Not signi 󿬁 cant ( SI Appendix  , Table S16) λ  (Coef 󿬁 cient of spatial error correlation) 0.561***Moran ’ s I 0.021 Unit of analysis is the forest parcel. Dependent variable is the percent of forest-cover change from 2001 to2007. Additional controls include parcel size, parcel size per household, elevation, distance between each parceland the nearest household, and distance between each parcel and the main road ( SI Appendix  , Table S16). Totalnumber of observations is 151. Log likelihood is 170.281. Independent variables were mean centered beforeentering the model. Detailed discussion of the spatial autoregressive models are in  SI Appendix  , Section 2.5.2.* P   <  0.05, ** P   <  0.01, *** P   <  0.001. Table 3. Path analysis of the two opposing forces through which group size affects collective action Path analysisUnstandardizedcoef 󿬁 cient (SE)Dependent variable: Free rider (binary: 0 for a household that does not free ride; 1 for a household that free rides)Group size 0.146** (0.051)Within-group enforcement (binary: 0 for weak enforcement; 1 for strong enforcement)  − 0.522** (0.184)Dependent variable: Within-group enforcementGroup size 0.103** (0.038)Within-group division (binary: 0 for no within-group division; 1 for within-group division) 0.376 (0.266)Group size  ×  Within-group division  − 0.050 (0.061)Dependent variable: Group sizeSocial ties to local leaders (binary: 0 for weak social ties; 1 for strong social ties) 0.052 (0.651)Distance to main road (log)  − 0.067 (0.136)Number of laborers  − 0.051 (0.350)Household size 0.027 (0.243)Education of adults 0.016 (0.117)Household income (log)  − 0.093 (0.311)Percentage of agricultural income 1.839 (0.946) Unit of analysis is the household, but both characteristics of households and their assigned groups are considered. Continuous independent variables aremean centered. All goodness-of- 󿬁 t indices show that the model 󿬁 t is respectably high ( SI Appendix  , Table S5). Total number of observations is 113 households.** P   <  0.01. 4 of 6  |  www.pnas.org/cgi/doi/10.1073/pnas.1301733110 Yang et al.
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks