Information Networks in Amenity Transition Communities

19 Pages·2013·0.8 MB·English

HumEcol(2013)41:885–903
DOI10.1007/s10745-013-9595-7
Information Networks in Amenity Transition Communities:
A Comparative Case Study
JordanW.Smith
Publishedonline:7August2013
#SpringerScience+BusinessMediaNewYork2013
Abstract Amenity transition, a major socio demographic upontheirabilitytoprofitfromthedevelopmentorextraction
trend in areas rich in natural resources, is characterized by of local natural resources (England and Brown 2003). The
economic and population growth asa result of retirement in- accessibility of natural resources was greater in small cities
migration, increased rates of second home ownership, and when compared to urbanized areas, and often resulted in ho-
increases in the number of industries that do not need to be mogenous resource-dependent economies. Economic and so-
proximatetoaspecificgeographiclocation.Amenitytransition cial stability in small, resource-dependent communities was
is also characterized by increased intra-community conflict often short-lived however, as extractive boom periods were
between long-term residents and in-migrants. This research oftenfollowedbybustperiodsafteralocalresourcehadbeen
analyzes whether the population growth accompanying ame- exhausted(Freudenburg1992).Bustcycleswerecharacterized
nitytransitionisassociatedwithvariationsinthestructureand byseverelyreducedsocialandeconomicwellbeingaswellas
characteristicsofintra-communityinformationalnetworks,as largeratesofoutmigrationandpopulationdecline(Smithetal.
sociological theory would suggest. Methodologically, this is 2001).Forthelastthreedecades ofthelastcenturyhowever,
accomplishedthroughacomparativeanalysisofthestructure communities rich in natural amenities experienced renewed
andcharacteristicsofinformationalnetworksinthreecommu- economic and population growth as a result of retirement
nities undergoing amenity transition. The analyses suggest in-migration,rapidincreasesinratesofsecondhomeowner-
population density is not related to either the structure of ship,andnewindustriesthatdidnotneedtobegeographically
informational networks or the concentration of trust/distrust tetheredtoacentrallocationofeconomicproduction(Johnson
within them. When considered in conjunction with previous 1999).Between 1970 and1996,forexample,averagepop-
empiricalwork,thesefindingssuggesttheconflictsassociated ulation growth in amenity rich rural counties was 120 %
with amenity transition are more likely to arise because of greater than rural counties lacking amenities such as
conflictingvaluesystemsandideologiesasopposedtosocial favorable climates, lakes or rivers, and topographical varia-
structuralchangesinthecommunitiesthemselves. tion(McGranahan1999).
The growth of amenity-based communities is not only
Keywords Affiliationnetworks .Bipartitenetworks . being experienced in the United States. Small communities
Two-modenetworks in other developed countries such as Australia (Holmes
2002, 2006), the United Kingdom (Wilson 2001, 2006),
andItaly(Saraceno,1994)havealsoexperiencedshiftsaway
Introduction from traditional land uses and industrial clustering towards
morediversifiedmodesofeconomicproduction.Thispattern
TheGrowthofAmenityTransitionCommunities ofdevelopment,whichhasbeentermed“ruralrestructuring”
bysomescholars(Nelson2001;Woods2003)affectsnotjust
For the majority of the 20th century, the economic and social thepopulationdensitiesandeconomicactivitieswhichoccur
stabilityofmanysmallcommunitiesinAmericawasdependent in small communities, it also affects their patterns of intra-
personal social interaction (Gosnell and Abrams 2011). For
example,asubstantialamountorresearchhasfocusedonthe
J.W.Smith(*)
differencesandconflictsbetweenlong-timeresidents,whoare
NorthCarolinaStateUniversity,Raleigh,NC,USA
e-mail:[email protected] believedtobemoreconcernedwithlocalcommunityplanning
886 HumEcol(2013)41:885–903
efforts,andshort-termresidentsorin-migrants,whoaretyp- populousandmorediverse,individualsbecomelesslikelyto
ically portrayed as having a more transitory orientation, and establish intimate and trusting reciprocal relationships with
subsequently are less concerned about local planning efforts others(Durkheim1933).Consequently,theyaremorelikely
(Brunsonetal.1997;Salamon2003a,2003b). to be socially isolated and less willing to cooperate with
others.Thisis,yetagain,anotherstructuralbarrierthatmight
ImplicationsofAmenityTransitiononCommunityPlanning alsoaccompanyamenityin-migration.
Efforts Inthisstudy,Isetouttoexplicitlytestwhetherthesetwo
factors, a decrease in the density of informational networks
For community planners, amenity-migration can lead to in- and a decreased level of trust placed in others and social
creaseddifficultiesincomingtoconsensusonlocaldevelop- institutions, correlate with amenity in-migration and popula-
mentissuesasin-migrantsclashwithlong-termresidentsover tiondensitylevelsinamenityrichcommunities.Iaccomplish
howthecommunityanditssurroundinglandscapeshouldbe thisthroughamixed-methodscomparativecasestudyofthree
developed (Hurley and Walker 2004; Travis 2007). While small communities located inwestern NorthCarolina, USA.
communityplanningeffortsmightbehamperedbyconflicting First, I provide a deeper review of the relationship between
valuesheldbetweenlong-termresidentsandin-migrants,the amenityin-migration,populationdensity,anddecreasedcon-
abilitytosuccessfullyimplementplanningeffortsmayalsobe centrationsofinformationaldensity.Ithenfollowwithamore
impactedbymorestructural1 issuesrelatedtolarge-scalein- theoreticallyinformedreviewabouttherelationshipbetween
migration.Greaterpopulationgrowthinamenityrichcommu- population density and trust before explicitly forming the
nities inherently implies greater population densities that, as hypothesestobetested.
sociologicaltheorysuggests,leadstotwokeyfactorslikelyto
hamperplanningefforts.Firstisadecreaseintheconcentra-
tionofintra-communityinformationexchangeandsecondis RelatedLiterature
anincreasingdistrustinothersandinsocialinstitutions.
Thedecreasedconcentrationofintra-communityinforma- Communityandenvironmentalplannersrealizethatthesuc-
tionexchangereferstocommunityresidentsnolongerhaving cessful formulation of policy, development of plans, and
torelyonsingleoutlets,suchasweeklycommunitynewspa- implementation of management approaches, often hinges
pers, church services, or civic organizations to disseminate upon how local residents or stakeholders interact with one
information about local community issues (Flora and Flora another (Freeman 2001; Jacobs 1961). All too often, plan-
2008).Anin-migrationofdissimilarothersleadstotheintro- ners cite breakdowns in communication and information
duction of alternative modes of communication that often disseminationaskeyfactorslimitingtheirabilitytoachieve
challengethedominantmodesofinformationdissemination, desired objectives (Maier 2001). Planning, after all, is a
namely kinship ties, in small cities (Hofferth and Iceland processthatoccurswithin,andisdependentuponarelational
1998).Consequently,informationalexchangeinamenitytran- structure of individuals and organizations. Consequently,
sition communities is likely to be more decentralized and sociological theory can offer insights into how the actual
involveadiversearrayofinstitutionssuchascivic,religious, social structure of communities themselves, and inherently
educational, and interest groups as well as individuals with thestructureoftheirinformationnetworks,caninfluencethe
whomindividualshaveless-frequentcontact.Ifthisproposi- capabilities of local planning agencies. More dynamically,
tionistrue,itsuggestsplanningefforts inamenity transition sociologicaltheorycanofferkeyinsightsintohowthecapa-
communities are likely to face more structural barriers than bilitiesofplanningagenciesarelikelytochangeasaresultof
thosegeneratedbyconflictingvaluesystemsheldbetweenin- shiftsinthesocialstructureofthecommunitiesinwhichthey
migrantsandlongtermresidents. operate.
Asecondfactorofdecreasedtrustinsocialinstitutionsis
alsolikelytoinfluenceplanningefforts.Recentresearchhas PopulationDensityandPatternsofSocialInteraction
foundtheleveloftrustindividualsplaceinothersaswellas
the level of trust they place in social institutions is signifi- A vast amount of sociological theory has been devoted to
cantly higher in small rural areas relative to their urban understandingvariationsincommunitystructure.Theearli-
counterparts (Beaudoin and Thorson 2004). A long history estwritingoncommunitystructurecanbetracedatleastas
of sociological theory suggests that as places become more far back as 500 B.C. (Sorokin 1979). Nearly all theories of
socialstructure arerootedintheassumptionthatanypopu-
lation larger than a small group consists of individuals
1The word “structural” throughout this article simply refers to the
connectedtooneanotherviasomeformofsocialinteraction
relational nature of all social interactions, whether they be between
(Simmel 1955). Therelationalnatureofhumaninteractionis
two individuals, or between and individual and an organization
(Entwisleetal.2007). integraltotheflowofinformationbetweenindividuals,groups,
HumEcol(2013)41:885–903 887
communities, and larger aggregations (Granovetter 1973). linear relationship between the population density of indi-
However,interactionsbetweenindividualsarenotcompletely viduals’ communities of residence and those individuals’
random across entire populations. Rather, the frequency and number of self-reported friendship/kinship ties (Kasarda
intensityofanyindividuals’interactionsareafunctionofthe andJanowitz1974).Withinthis“linear-development”hypoth-
number of potential others they are physically proximate to esis, population size and density are exogenous factors be-
(Wilkinson1991).Asaconsequenceofthisbasicassumption, lieved to influence individual social behavior. Following
numeroussocialtheoristshaveconjecturedthatthecharacter- Toennies,itwasbelievedurbanizationwouldleadtoareduced
isticsofanycommunityfundamentallyrestuponthefrequen- concentration of informational networks. In other words, the
cyandqualityofinteractionsamongitsmembers. presence of friendship and kinship ties were expected to de-
TheGermansocialtheoristFerdinandToenniespresented cline as the probability of developing non-intimate ties in-
thefirsttheoreticaltypologyofcommunitystructure(1887). creased. The results of this research, however, revealed no
Toenniesarguedtheorganizationofanycommunitycouldbe relationship between population density/urbanization and
located on a continuum ranging from the fully communal the dissolution of friendship and kinship ties. The sec-
(gemeinschaft) to the fully individualistic (gesellschaft). ond notable study, conducted several years later using a
FundamentaltoToennies’argumentwasthebeliefthatcom- largersampleofresidentswithinGreatBritain,resurrectedthe
munitiestendedtoevolvefrombeingcomposedofindivid- linear-development hypothesis (Sampson 1988). Again, no
ualsprimarilyconcernedaboutthecommunityasawholeto empirical evidence was found for a positive relationship
individuals principally occupied with personal fulfillment. between population densities and individuals’ self-reported
This evolution occurred as communities transitioned form friendship/kinshipties.
sparse,ruralsettlementpatternstomore-dense,urbanagglom- ThesepreviousempiricalresultsseemtosuggestToennies’
erations.Toenniesalsoconjecturedthatnumerousothercom- conceptualization,whilecommonlyassumed,doesnotaccurate-
munity characteristics, aside from communal orientation, ly reflect reality. Consequently, the formal hypothesis under
changedaspopulationsbecomemoreurbanized.Specifically, examinationinthisstudyisdevelopedfrompreviousempirical
more urban populations tended to be characterized by more work and postulates no relationship between a community’s
industrial diversification, more variable belief systems, more populationdensityandthedensityofitsresidents’informational
dispersedsocialties,moreinfrequentsocialinteractions,anda networks.
greaterneedforrulestoovercomedistrust(Brint2001).
Oncetheearlytypologyandconceptualizationslaiddown PopulationDensityandTrust/DistrustinInformational
byToennieswereadoptedintheUnitedStates,alonglistof Sources
research sought to document degrees and consequences of
thetransitionofcommunitiesformruraltourban(Wilkinson BeforefullydisregardingtheutilityofToennies’conceptuali-
1991). A large majority of this research was fueled by the zation,itisimportanttonotehesuggestednumerouscommu-
Morrill Act of 1890 which tasked land-grant universities nitycharacteristicscouldvaryrelativetolevelofurbanization.
with disseminating information among the geographically Toenniessuggestedthatnotonlydidthequantityofsocialties
dispersed rural areas of the country (Field et al. 2002). For varyrelativetopopulationdensities,buttheirqualitativecom-
socialscientistsatthetime,whatwas(andarguablystillis) ponents did as well.2 The strength ofindividuals’ tiesto one
of interest is the arrangement and dispersion of people and anotheraswellastheirtiestoformalandinformalassociation
activities on the landscape. “Dispersion is of sociological groupsarenotsolelydeterminedbythefrequencyofinterac-
importancebecauseofitspresumedeffectsontheinteraction tions. Rather, the strength of a social tie is defined by a
of people” (Wilkinson 1991:52). Despite increased interest combination of both quantitative (e.g., amount of time, fre-
amongsocialscientists,nearlyallresearchconductedduring quencyofinteraction)andqualitative(e.g.,emotionalintensity,
this time relied upon descriptive case studies of individual feelingsofreciprocity)measures(Granovetter1973).
communities(SandersandLewis1976). This argument is succinctly stated in the early work
In the latter half of the twentieth century, empirical re- of Pahl (1966) who pointed out that attempting to classify
search on the relationship between population density and entire geographic regions as more communally oriented
patternsofsocialinteractiongainedmoreanalyticaltraction, (gemeinschaft)ormoreindividualistic (gesellschaft)resulted
particularlywiththeintroductionandwidespreaddispersion inagrossoversimplificationoftheheterogeneityofcommu-
of the network concept (White et al. 1976). Two notable nitiesandthesocialrelationshipsthatoccurwithinthem.Pahl
studies conducted during this time explicitly drew upon
Toennies’ conceptualizations and can subsequently be used
2However,Toenniesarguedthatboththequantitativeandqualitative
to inform the hypothesis being tested in this research. The
dimensionsofsocialtiesdidnotco-varyWhileseveralstudiessupport
first of these studies utilized survey data collected from Toennies’ argument (e.g., Putnam 1993), consistent and definitive
nearly2,200residentsofEnglandtoexamineahypothesized supportislacking(Brint2001).
888 HumEcol(2013)41:885–903
notedthepopulationdensityofanareawasmuchlessimpor- characteristics. First, each community had to have been eco-
tant than the class and life-style characteristics of specific nomically dependent upon natural resource based industries
individuals and groups. There is obviously heterogeneity in withinthepasthalf-century.4 Second,allcommunitieshadto
thepatternsofintra-communitysocialinteractionsthatcannot be located within a single geographic region experiencing
beaccountedforthroughasimplemeasureofpopulationden- amenity-basedpopulationandeconomicgrowth.Thiscriterion
sityalone.Henotesthatbothgemeinschaftandgesellschaft-type wasestablishedinanattempttocontrolforpotentialregional
relationshipscanbefoundamongdifferentgroupsinthesame variationsintheavailabilityofpotentialinformationalsources
place. Consequently, it is logical for sociologists to place a (i.e.,thenumberoflocalnewspapers,civicgroups,etc.).The
greater emphasis on the qualitative and socio-psychological thirdanddistinguishingcriterionwasthatthesetofcommuni-
definitionsofcommunity,suchasidentity(Bell,1992)andtrust tieshadtovaryintheirpopulationsizesandassociatedpopu-
(Greideretal.1991). lation densities. This third criterion, coupled with the fact the
Arguablythemostcentralqualitativemeasuredefiningthe analytical methods used were literally replicated within each
socialtieswithinacommunityistheextenttowhichothersare case, allowed me to make comparative inference across the
trusted ordistrusted (Coleman1990). Relative topopulation studycommunitiesregardingthepotentialrelationshipbetween
density, interpersonal trust has been relatively unstudied. populationandinformationalnetworkdensities(Yin2009).
Perhaps this is unsurprising, given examinations of rural to
urban variations in the nature of interpersonal relationships CommunitySelectionandDescription
hasbeenrelegatedtothemarginsofdisciplinessuchasrural
sociology(HofferthandIceland1998).Whatisknown,how- Following the three selection criteria outlined above, I se-
ever,suggeststhatpopulationdensityisrelatedtothequality lected three small cities (Waynesville,Franklin,and Spruce
of interpersonal ties. Data from an analysis of the 1985 Pine)locatedwithinextremewesternNorthCarolina,USA.
General Social Survey, which is administered to a random The region has a long history of natural resource depen-
sample of Americans, revealed the social networks of resi- dence, primarily within the forestry and mining industries
dentsinnon-metropolitanareasofthecountryarecomprised (Salstrom 1997). However, in recent decades the natural
of “more intense”, closer, relationships relative to residents resource-based industries have steadily declined and been
living in metropolitan areas3 (Beggs et al. 1996). Relatedly, supplanted my amenity-oriented growth and development
empirical evidence suggests feelings of reciprocity (particu- (Case etal.2008;Kirketal.2012;Shultz2011).Thethree
larly among kinship ties) are stronger among individuals cities of Waynesville, Franklin, and Spruce Pine vary sub-
living in rural areas relative to their urban counterparts (Lee stantiallyintheirpopulationsizeanddensities(Table1).The
etal.1994).Morerecentevidencealsosuggestsactualrecip- largest city of Waynesville had a 2010 population of 9,869
rocal behaviors, as opposed to subjective feelings, are also andadensityofabout489people/km2whilethesmallestcity
higher among individuals living in more sparsely populated of Spruce Pine had a 2010 population of only 2,175 and a
areas (Hofferth and Iceland 1998). While this previous re- moresparsedensityofabout215people/km2.Alsonotethat
search doesn’t explicitly gauge the degree of trust present all three cities have experienced population growth, and
withinindividuals’networks,theevidencedoessuggestrela- concurrent increases in population densities, since 2000.
tionalnetworksinmoreruralareasarecharacterizedbyhigher BothWaynesvilleandSprucePinehavealsobeenexperienc-
degreesoftrustinothersandinorganizations.Therefore,the ingpopulationgrowthsinceatleast1990.
formalhypothesisexaminedinthisstudyproposesthathigher
levels of trust will be negatively related to a community’s SecondaryDataCollection
populationdensity.Thereciprocalisalsohypothesized.
The first method used in my examination of these three
communitiesinvolveddrawinguponsecondarydatasources
Methods todeterminewhetherornottherewasarelationshipbetween
changes inpopulation density andthetransition away from
ComparativeCaseStudies naturalresourcebasedindustriesandtowardamenitybased
industries.Inshort,Iwantedtoestablishaclearlinkbetween
To answer the research questions posed above, I employed a amenity-led growth and changes in the spatial densities of
comparativemultiplecase-studydesignwhereinspecificcases each population. I accomplished this by compiling second-
(i.e., communities) were selected based upon several key arydataonemploymentinnaturalresourcebasedindustries
3Thismeasureofintensityisconstructedfromindividuals’responses 4Idefinedeconomicdependenceashavingatleast10%employment
to questions about whether people in their network are “especially innaturalresourcebasedindustries(farming,forestry,fishing,hunting,
close”,“somewhatclose”,or“totalstrangers”(Marsden1987). andmining)atonepointintimefrom1970to2010.
HumEcol(2013)41:885–903 889
Table1 Populations,areas,anddensitiesofstudycities
City Population 10-yearpopulation Annualgrowthrate Area Density 10-year
change forpreviousdecadea (km2) (pop./km2) densitychange
Waynesville,NC
2010 9,869 637 0.67 20.2 488.56 31.53
2000 9,232 2474 3.17 20.2 457.03
1990 6,758 20.2
Franklin,NC
2010 3,845 355 0.97 10.1 380.69 35.15
2000 3,490 −588 −1.54 10.1 345.54
1990 4,078 10.1
SprucePine,NC
2010 2,175 145 0.69 10.1 215.35 14.36
2000 2,030 20 0.10 10.1 200.99
1990 2,010 10.1
Datasource:2010,2000,and1990USCensuses(www.census.gov)
(cid:4)(cid:4)(cid:1) (cid:3) (cid:5) (cid:5)
a Annualgrowthrate¼ ttn0 1=10 −1 (cid:2)100.Wheret0=originalpopulationandtn=population10yearslater
aswellasamenity-basedindustriesandanalyzingitscorre- were: Employment in farming; employment in agricultural
lationwithchangesinpopulationdensitiesovertime. services, forestry, fishing, and hunting; and employment in
Note that this preliminary analysis draws upon data mining.Isummedemploymentcountsacrossthelatterthree
aggregated at the county level and is not exclusive to the measuresanddivideditbytotalemploymentinallindustries
three study communities that form the core focus of this to create the proportional employment in natural resource
research. The limited availability of census tract level data basedindustriesmeasure.
onemploymentbyindustry5requiredmetomakethetrade-off
between either just assuming each of these cities was AmenityDependence
transitioning toward and amenity-based economy, or using
moregrosslevelsofdatatoillustratethat,atleast,thecounties Likenaturalresourcedependence,thereareavarietyofways
whichcontainthesecitieshaveexperiencednotableshiftsin to measure amenity dependence. McGranahan (1999) sug-
their social and economic composition over the past several gesttheenvironmentalcharacteristics(e.g.,topography,nat-
decades. uralwaterbodies)ofacity’ssurroundinglandscapearegood
indicators.JohnsonandBeale(2002)alternativelysuggesta
NaturalResourceDependence community’s recreational infrastructure (e.g., per capita re-
ceiptsfromhotels/motels)serveasbettermeasures.Andstill
Thereareavarietyofwaystogaugenaturalresourcedepen- others(Winkleretal.2012;Winkler2010)suggestanalter-
dence; typically it involves either the proportion of a nativecompositemeasureofseasonalhome-ownershiprates,
locality’s total earned wages coming from natural resource proportion of the population that has recently in-migrated
relatedindustriesortheproportionofalocality’stotalnum- frommetropolitanareas,andtheproportionofhousingunits
ber of jobs being within those industries (Stedman et al. valuedatmorethan$200,000US.Whileallofthesemeasures
2007). In this research, I opt for the proportional employ- arevaluable, and suitable to specific researchquestions, they
ment approach, which has successfully been utilized in a arealleitherrelativelystaticinnature(e.g.,anarea’senviron-
variety of different contexts (Freudenburg and Gramling mentalconditionsorrecreationalinfrastructurechangeslowly
1994; Nord 1994; Smith et al. 2012). Three county-level overtime)ortheyrelyonsecondarydatathatareonlyreported
employment measures were compiled from the US Bureau atlargetimeintervalssuchasdecennialCensuses(e.g.,mea-
of Economic Analysis (www.bea.gov). These measures sures of owner-occupied housing rates). Consequently, they
are not particularly suited to illustrating changes in amenity
dependenceovertime.
5Thisisespeciallytrueinsmallcitieswherepotentialsourcesofdata
In this study, I utilize a simple annual indicator, the
arecensoredtoavoiddisclosurepertainingtoaspecificorganizationsor
individuals(USCensusBureau2012a). proportion of an area’s total employment in service-based
890 HumEcol(2013)41:885–903
industries, as an indirect measure of amenity transition. generated from representative samples drawn from the sys-
Similar approaches have been used elsewhere (Hoffman temofinterest(Freeman2004).Dataiscollectedonsampled
2008) using the logic that increases in amenity migrants individuals’ relationships with other individuals, groups, or
and second-home ownership will result in an increase in events(i.e.,thedatarepresentegocentricnetworks7)and,if
theneedforserviceindustryjobs(PowerandBarrett 2001; the sample is representative of the population of interest,
ViasandCarruthers2005;Winkleretal.2007).Idividedthe inference can be made about the characteristics of relation-
employmentcountforservicesindustries6byacounty’stotal shipswithinthatpopulation(Frank2011;Marsden2011).
employment in all industries to generate the proportional If a researcher decides to use survey methods to collect
employmentinservices-basedindustriesmeasure. networkdata,theymustmakeseveralcriticaldecisionsprior
For both the natural resource and service sector depen- to data collection. First, they must decide whether relation-
dencemeasures,dataareannualfromtheyear1971to2007. ships of interest exist between a single set of units (e.g.,
For each industry, I matched the Standard Industrial individuals)ormultiplesets(e.g.,therelationshipsbetween
Classification (SIC) codes (which were used to identify in- individualsandgroups).Isthenetworkofinterestone-mode
dustriesfrom1969to2000)totheNorthAmericanIndustrial ortwo-mode?Second,theymustdefinetheboundaryofthe
ClassificationSystem(NAICS)codes(whichhavebeeninuse populationforwhichtheywanttodrawinference(Laumann
since2001)tocompilethemostcomprehensivedataset. etal.1989).Thirdandfinally,theyshoulddefinethetypesof
ties, for which data can be collected, that are important for
Population answeringtheirresearchquestions.
Regarding the first point, my research question is
Themeasuresofpopulationcomefromtwosources.Forthe concernedwiththeaggregatecharacteristicsofinformational
years1980,1990,and2000Iusedactualpopulationcounts networkswithinacommunity.Thisincludesboththeflowof
for each city as reported by the decennial US Census. For informationbetweenindividualactorsandtheflowofinfor-
intermediateyears,Iusedthemid-yearpopulationestimates mation betweenindividualsandorganizationssuchaslocal
publishedbytheUSCensusBureau(2012b). religious, civic, and educational groups. Consequently, my
research design called for soliciting information about ties
SecondaryDataAnalysis betweentwosetsofunits,individualsandpotentialsources
ofinformation.Thenetworksaretwo-mode.
Secondary data analysis simply involves calculation of a Toaddressthesecondpointofsolvingtheboundaryspec-
correlation matrix comprised of the proportion of natural ificationproblem,Ichosetolimitthepopulationsofinterestto
resourceemployment,theproportionofemploymentinser- allfull-timehomeownersresidingwithinthecitylimitsofthe
vice related industries, and the population density for each three study communities. Here, as in other research using
county. surveysamplestogeneratenetworkdataandinferenceabout
aggregate social groups (e.g., Marsden 1990), the boundary
SocialNetworkDataCollection specificationisrelativelyeasytosolvegiventhe“boundary”
of the network coincides with the entire population residing
Foralargeproportionofsocialscienceresearch,thecollec- withinagivenarea.
tionofnetworkdataisimpractical.Recordsofrelationships Thethirdandfinalpointcallsforaddressingthetypesof
between individuals and organizations frequently do not ties within the network that are of interest. I chose not to
exist,oriftheydo,theymightnotcontaininformationabout measure and treat all informational ties equally. Rather, in
the characteristics of the relationship that are of interest. concordance with the study’s theoretical underpinnings, I
More commonly however, the social system for which the wantedtogaugewhethertheinformationthatwasprovided
researcher would like to make inference to is too large or byasourcewaseithertrustedordistrusted.
diverse to reliably collect data on all relevant relationships.
Under these conditions, social scientists often utilize data SurveyMethodology
Iusedsurveysdeliveredandreturnedviamailtocollectdata
6Prior to 2001, the SIC system aggregates all service industry jobs.
ontheinformationalnetworkswithinthethreestudycitiesof
Since2001,thenewerNAICScodesseparateserviceindustryjobsinto
Waynesville, Franklin, and Spruce Pine. Within each city, I
the following categories: arts, entertainment, and recreation related
employment; employment in accommodation andfoodservices; pro-
fessionalandtechnicalemployment;managementjobs;employmentin
administration, waste, and remediation services; educational service
related employment; and employment in health care services. I sum 7Asopposedtorepresenting“whole-networks”bycollectingdatafor
thecountsacrossalltheseindustriestomatchthemwiththepre-2001 eachrelationshipwithinaboundedsocialgroup,andwhichcanbeused
SIC-baseddata. todescribethestructureofthatgroup.
HumEcol(2013)41:885–903 891
drewrandomsamples of300 full-time residenthomeowners Everett 1997). For my data, the degree centrality for each
frompublictaxrecords(i.e.,900totalhomeownerssampled). informationalsourcesimplyreferstothenumberofsampled
respondents who reported using that source. Given the data
InformationalSourcesandTrust are generated from a representative sample of each city’s
population, degreecentrality can alsobe normalizedrelative
The questionnaires, which I mailed in the summer of 2011, tosamplesizeandexpressedastheproportionofeachsample
includedaquestionthatstated“…Iwouldliketoknowwhere whouseeachparticularinformationalsource.
youobtaininformationaboutlocalcommunityissuesandhow
muchyoutrustinformationfromthosesources.”Respondents Closeness Closenesscentralityisameasureofhow“far”an
werethengivenalistof13potentialinformationalsourcesto individualnodeistoallothernodeswithinthenetwork.Itis
whichtheyindicated whether they either“alwaystrusted that calculatedforeachnodeandisdefinedastheinverseofthe
source”, “trusted the source only sometimes”, or “always average length of the shortest paths to/from all other nodes
distrustedthesource.”Respondentswerealsogiventheoption withinagraph(Freeman1979).Fortwo-modedata,theraw
ofindicatingthattheydonotgetinformationfromthatpartic- centralitymeasuremustbenormalizedtoaccountforthefact
ular source. Initial analysis of this data involved treating it as that each node cannot be a geodesic distance of one from
two-modeindividual-by-sourcenetworkdataandutilizingtra- nodes of an alternative type (i.e., informational sources or
ditionalsocialnetworkanalysismeasures(Latapyetal.2008). individuals are not directly connected). This normalization
involvesdividingtherawclosenessmeasureinton+2(n –1)
j i
SocialNetworkDataAnalysis where n is the number of nodes of a particular type
i
(individuals) andn is thenumberofnodesof thealternative
j
In the lexicon of social network analysis, the survey data type(informationalsources).
collected involve two modes, individuals and informational
sources, and thus can be analyzed as an affiliation network. Betweenness Betweenness, the third measure of central-
Theanalysisofthestructureofeachcommunity’sinformation- ity, is a sum ofalltheshortestpathsbetweennodesthatgo
al network involves the calculation of node-level measures throughagivennode.Here,betweennesscentralityiscalculat-
(density, degree centrality, closeness centrality, betweenness edforeachinformationalsourceandthuscanbeinterpretedas
centrality,andeigenvectorcentrality)andgraph(network)level the number of shortest paths between individuals which
measures (degree centralization, closeness centralization, and go throughthatparticularsource. Following Borgatti and
betweenness centralization) (Hanneman and Riddle 2011). Everett (1997), I normalize betweenness centrality mea-
All matrix manipulation was completed with R (www.r- sures to account for the two-mode nature of the data.
project.org) and all network analysis was completed with the Normalization involves dividing the raw betweenness
userwrittenigraphpackage(www.igraph.sourceforge.net). centrality by
Density ð0:5(cid:2)n ðn −1ÞÞþð0:5(cid:2)ð13−1Þð13−2ÞÞ
1 i
Densityissimplythenumberoftiespresentwithinanetwork þðð13−1Þðni−1ÞÞ
(Borgatti and Everett 1997). Density values are typically
normalized by dividing the total number of observed ties where n is the size of each sample.
i
bythemaximum numberpossible,whichinthisanalysisis
2(n)(13).Densityvaluesnear0wouldbeexpectedin“loosely Eigenvector The final measure of centrality I consider is
connected”networkswhereasdensityvaluesnear1representa eigenvector centrality, which is defined as the principal ei-
“highlyconnected”network.Notethatdensityvaluesarecom- genvector of the adjacency matrix of a graph (Bonacich
puted prior to transforming the two-mode data to one-mode 1991). Eigenvector centrality assumes a node’s centrality
dataofinformationalsources.8 withinthenetworkisafunctionofthecentralityofallother
nodes connected to it. Here, the eigenvector centrality of
Centrality each informational source is the sum of the centralities of
all the individuals who use that source. I normalize the
Degree Degreecentralityisacount,calculatedforeachnode, eigenvector centrality measure to account for the two-
ofthenumberofedgesincidentuponthatnode(Borgattiand mode nature of the data. This normalization involves
dividing raw eigenvector scores by the square root of
one half, the maximum score theoretically obtainable for
8The density value calculated from source-by-source matrix would
any single node in a two-mode graph (Borgatti and
undoubtedly be 1 given each unique pair of informational sources is
likelytobeusedbyatleastonerespondent. Everett 1997).
892 HumEcol(2013)41:885–903
Centralization collectedfromarepresentativesampleofindividualswhouse
those informational sources can be drawn upon and
The next analysis of information network structure deter- operationalizedasaproxymeasurefortherelativesimilarity
mined the extent to which each network has a central infor- betweeninformationalsources.
mational source that individuals utilize, its centralization More formally, the two-mode individual-by-source ma-
(Freeman 1979). Centralization measures are simply graph trix A is transformed into a one-mode source-by-source
(network)levelmeasuresbaseduponthecentralityscoresof matrix B by multiplying it by its transpose (B=A’A). The
each node within the network. Consequently, there are three source-by-source matrix simply represents the count of in-
networkcentralizationmeasures(degreecentralization,close- dividuals within each sample who use each distinct pair of
ness centralization, and betweenness centralization) based informationalsources.Aftercreatingtheone-modematrices
uponthecalculationofeachtypeofcentrality.Broadly,these foreachstudycommunity,Inormalizethedatatoaccountfor
measures express the degree of variance in a network as a different sample sizes. This is accomplished simply by di-
percentage of the hypothetical perfectly centralized network vidingthecountvalueforeachpairofinformationalsources,
(i.e., a star) of the same size. Centralization measures near b ,b ,…,b ,bythemaximumcountpossible(n).10 The
11 12 mn
0representnovariabilityacrossallnodes’centralitymeasures subsequentmatrixB*isanon-binarymatrixrepresentingthe
whilemeasuresnear1representa highdegreeofvariability. similarity, or overlap, between each pair of informational
Eachcentralizationmeasureisnormalizedfollowingthefor- sources.Theanalysisculminatedbyillustratingthedensityof
mulaprovidedbyBorgattiandEverett(1997). informationaloverlapamongpairsofinformationalsources.11
InformationalOverlap ProbabilityofAlwaysDistrustingorTrustingInformational
Sources
AfteranalyzingnetworklevelcentralizationmeasuresIfeltit
prudent to examine the coverage overlap between informa- All of the network analysis thus far has been based off of
tional sources (i.e., the number of individuals who use any individuals’ use of information, irrespective of individuals’
uniquepairofsources).Thisanalysisservesasimilarfunction personalappraisalsofthatinformation.Thefinalstageinthe
asthecentralizationmeasuresinthatitillustratestheconcen- analysisinvolvedcalculatingtheprobabilitythatindividuals
trationofinformationprovisionamongpairsofsources.The will either “always distrust” or “always trust” the informa-
analysis of informational overlap involved converting the tiontheyindicateusing.Iaccomplishthisbycomparingthe
individual-by-source survey data into a source-by-source af- whole two-mode informational network (used in previous
filiation matrix (Borgatti and Everett 1997). When data on analyses) with the two-mode networks comprised of only
relationshipsofinterestarenotavailable,analystscaninferthe those informational sources that are “always distrusted” or
presenceandstrengthofthoseunobservedrelationshipsifdata “always trusted.” I accomplish this by generating the re-
are available on shared “affiliations” between that unit of ducedtwo-modenetworks,matchingindividualrespondents
analysis. In short, co-affiliations between units of interest to the two-mode whole information network, and multiply-
can be used as indicators of unobserved ties (Borgatti and ingtheresultingmatricesbythetransposeofthetwo-mode
Halgin 2011). This study is concerned with the structure of wholeinformationnetwork.12Theresultisa1-modesource-
informationalnetworks(i.e.,therelationshipsbetweendiffer- by-source matrix characterizing the number of individuals
ent sources of information within a community). However, who either “always trusted” or “always distrusted” each
data collection on the relational nature of informational uniquepairofinformationalsourcesused.Afternormalizing
sources is highly difficult to collect.9 Alternatively, the data this matrix (dividing by the diagonal of the 1-mode whole
information network matrix),the resultisa transitionproba-
bility matrix characterizing the probability that individuals
9Itwouldalsorequiretheanalysttomakearbitrarydistinctionsastowhat
will either always distrust or trust each unique pair of infor-
isandisnotapotentialinformationalsource.Forexample,shouldonly
mationalsourcesused.Thisanalysisculminatesbyillustrating
formalentitiessuchascivicandreligiousgroupsbeanalyzedeventhough
avastmajorityofinformationflowsthroughinterpersonalconnections?
Theapproachadoptedinthispapercircumventsthislimitationbysimul-
taneouslyanalyzingformalandinformalsourcesofinformation.While
the large majority of affiliation network studies evaluate individuals’ 10Imaketheassumptionthatallinformationalsourcesareavailableto
membership in formal groups such as clubs (e.g., McPherson and allrespondents.
Smith-Lovin 1986, 1987), a substantial literature also focuses on less 11While this one-mode matrix can subsequently be analyzed using
well-bounded patterns of co-affiliation such as participation in online typical network analytic techniques (so long as they don’t require
forumsormessageboards(e.g.,Allatta2003).AsFeld(1981:1017)notes binary data), I felt it would be unfruitful given the relatively small
“thenatureofrelationstoobjectsvary;yettheyareabstractlysimilarin numberofpotentialinformationalsources.
thattheymaybeconsideredrelationstogenerallydefinedfoci;andthey 12This process is identical to analysis of network mobility where
thereforehavesimilarimplicationsforgroupstructure.” affiliationalmembershipiscomparedacrossyears.
HumEcol(2013)41:885–903 893
a
the density oftheprobabilitiesamong pairs ofinformational
sources.
Results
Amenity-ledPopulationGrowth
First, I investigated the relationships between population
growthandeitherdeclinesinnaturalresourcedependenceor
increases in amenity-based activities. The longitudinal data
sets for population levels and employment in both natural
resource and amenity-based industries are shown in Fig. 1.
Three notable trends appear in the figure. First, all three
counties have experienced steady and dramatic increases in
population(andconcurrentlypopulationdensities)from1971
b
to2007.Second,allthreecountieshaveexperiencedsteady,if
not notable, declines in natural resource based employment.
NaturalresourceemploymentinMaconCounty,forexample,
droppedfromahighofnearly10%inthemid1970stobeing
nearlynonexistentin2007.Finally,serviceindustryemploy-
ment appears to be positively related to population growth;
this would be expected given the previous literature on the
changing nature of economic activity in amenity transition
communities.
Next,Icalculatethecorrelationmatricesforthethreevari-
ables of interest; these matrices are shown in Table 2. The
resultsconfirmthethreevisualtrendsseeninFig.1.First,for
two of the three counties (Haywood and Macon) there was a
highly significant and negative relationship between the pro-
portions of natural resource related employment and local
population densities. For the third county (Mitchell) the rela- c
tionship was negative, however not statistically significant.
Second, for all three counties a significant and positive rela-
tionshipwasseenintheproportionofservicebasedjobsand
localpopulationdensities.Acrosstheboard,thelevelofcorre-
lationwasveryhigh(.8793–.9651).Finally,Inoticedanega-
tive relationship between the proportion of natural resource
based jobs and the proportion of jobs in the service industry
fortwooutofthethreecounties(HaywoodandMacon)
Collectively, these results illustrate that all three counties
are,orhave,transitionedfromanaturalresourcebasedecon-
omy to a service based economy as the data show steady
declines in natural resource related jobs and dramatic in-
creasesinthetypesofemploymentmostfrequentlyassociated
withamenity-basedcommunities.Furthermore,andmoreim-
portantly,theseresultsillustratethatamenitytransitionimplies
Fig. 1 Changes in population size and amenity-related industries in
not just dramatic shifts in the dominant modes of economic
three study communities a Haywood County (Waynesville), North
activity,butshiftsinlocalpopulationdensitiesaswell.These
CarolinabMaconCounty(Franklin),NorthCarolinacMitchellCounty
preliminaryfindingssuggestamenitytransitioncommunities (SprucePine),NorthCarolina
mightbeundergoingsomefundamentalsocialchangesrelated
to how their residents interact and communicate with one proposition by looking at variations in community informa-
another. The next stage of the analysis explicitly tests this tionnetworksacrossthethreestudycommunities.
894 HumEcol(2013)41:885–903
Table2 Correlationmatricesfor
eachcounty HaywoodCounty,NC(Waynesville)
Populationdensity Proportionofnaturalresource
relatedemployment
Proportionofnaturalresourcerelatedemployment −0.8416***
Proportionofemploymentinserviceindustries 0.9472*** −0.9340***
MaconCounty,NC(Franklin)
Populationdensity Proportionofnaturalresource
relatedemployment
Proportionofnaturalresourcerelatedemployment −0.8740***
Proportionofemploymentinserviceindustries 0.8793*** −0.8740***
MitchellCounty,NC(SprucePine)
Populationdensity Proportionofnaturalresource
relatedemployment
Proportionofnaturalresourcerelatedemployment −0.0094
Proportionofemploymentinserviceindustries 0.9651*** −0.0539
***p≤0.001
CommunityCharacteristics thesamples’datawerenotsignificantlydifferentwhencom-
paredto2010Censusdataonowner-occupiedhousingunits.
Of the 900 mailed questionnaires, 40 were undeliverable
because the potential respondent no longer resided at the AggregateMeasuresofInformationalUse
addressreportedinthepropertytaxrecords,orbecausethey
haddied sincetherecordswereupdated.Fortheremaining Across all three study communities, respondents indicated
deliveredquestionnaires,420werereturnedandcompleted; using just over half of the 13 informational sources asked
thistabulatesouttoa48.8%responserate. about(Waynesville:M=7.16,SD=3.32;Franklin:M=6.67,
I began the analysis with an initial comparison of the SD=3.77; Spruce Pine: M=7.10, SD=3.80). However, the
samples’ sociodemographic characteristics (Table 3). For datarevealalargeamountofvariabilityintheextentoftrust
allthreesamples,respondentstendedtobenearmiddleage thatindividualsplaceineachpotentialinformationalsource
and predominantly male (as would be expected surveying (Fig.2).Byandlarge,residentsineachofthethreecommu-
individuals listed on property tax records). On average, re- nitiestendtotrusttheinformationtheyreceivefromimme-
spondents in Waynesville and Franklin reported household diate family members, churches, close friends, and local
incomesofbetween$50,000and$75,000peryear;income newspapers more than information coming from other
levels were slightly less ($35,000 to $50,000 per year) in sources.Theleasttrustedinformation,againacrossallthree
Spruce Pine. Respondents were also predominantly white. study communities, comes from elected officials, national
Acrossallthesociodemographiccharacteristicsqueriedabout, televisionnews,onlinenewssources,andcoworkers.
Table3 Sociodemographiccharacteristicsofeachstudycommunity
Waynesville,NC Franklin,NC SprucePine,NC
n=159 n=158 n=103
Medianagea 53 55 51
Percentfemale 31 32 26
Modalhouseholdincomecategoryb $50,000–$74,999 $50,000–$74,999 $35,000–$75,000
Percentwhitec 99 96 97
aNotsignificantlydifferentthan2010Censusdataonowner-occupiedhousingunits(Waynesville,χ2(p)=0.092;Franklin,χ2(p)=0.076;Spruce
Pine,χ2(p)=0.051)
bNotdifferentthan2010Censusdataonowner-occupiedhousingunits
cNotsignificantlydifferentthan2010Censusdataonowner-occupiedhousingunits(Waynesville,t(10053)=0.333,p>.05;Franklin,t(4001)=
0.116,p>.05;SprucePine,t(2276)=0.124,p>.05