Methods
We conducted a meta-analysis of peer-reviewed journal
articles published from 1 January 1990 through March
2011 in English that quantified the attitudes of stakeholders
who had experienced direct conflict with carnivores,
elephants, primates, or ungulates. We defined an attitude
as “a disposition or tendency to respond with some
degree of favor, or not, to a psychological object” (Fishbein
& Ajzen 2010). To qualify, attitude measures had
to be evaluative and quantified on a scale. Studies detailing
attitudes of individuals not having direct experience
with HWC were excluded because the general public
can have more positive attitudes toward wildlife when
not directly affected (Williams et al. 2002; Mart´ın-L´opez
et al. 2008), although, in some cases, negative attitudes
are displayed by people not having contact with a species
(Treves et al. 2013).We restricted our references to those
published in scientific journals (Calver & King 1999).
Although inclusion of gray literature in meta-analyses is
sometimes recommended to prevent publication bias for
significant results (Rosenthal 1979), this was not an issue
in our review because attitudeswere recorded as percentage
of respondents having positive, neutral, or negative
attitudes.
We searched Web of science for terms listed in Supporting
Information. We then located additional publications
by examining the reference list of each publication.
Finally, we refined the publications to include only
those published after 1990 because studies conducted
before 1990 were few and commonly applied outdated
methods. We then examined the selected publications
in detail and extracted and compiled 8 variables in an
Excel spreadsheet. The variables extracted were defined
by their availability across all publications and their relevance
to our research questions. The variables are defined
in Table 1.
Data Analyses
The attitudes reported in percentages in each publication
were extracted and converted to a binary variable
as either positive or nonpositive. A binary variable was
necessary because some publications reported 2 category
responses (e.g., yes or no) to attitude questions. Where a
middle value of an attitude scale was used,we categorized
it as either a positive or nonpositive value depending on
the context of the question. For example, for questions,
such as would you like the population of species x to
increase, stay the same, or decrease? We combined “stay
the same” and “increase” because we considered “stay
the same” to be more aligned with a positive rather than
nonpositive attitude. For cases where the middle value
was not obviously aligned with a positive attitude, responses
were categorized as nonpositive. We think it is
more robust to have a false negative than a false positive
because assuming people are more positive than they are
would be more detrimental to a species.
We assigned responses for each individual participating
in a survey to a positive or nonpositive attitude category
using the following computation: if 20% of a survey
sample of 300 individuals reported positive attitudes, 60
individuals were coded as positive and 240 nonpositive.
To derive a similar individual record for the damage
variable, we converted the percentage of respondents
experiencing damage into a probability of experiencing
damage per individual. For example, if 40% of a sample
experienced damage, then the probability of each
individual experiencing damage was 0.4. We assigned a
probability to each individual rather than a definitive yes
or no because information on individual respondents was
typically unavailable.
Not all publications reported what proportion of the
sample experienced damage from individual species.We
therefore compiled 2 types of data sets, a smaller one
which did not report a damage proportion and a larger
one that did. For most analyses, we used the 2 data sets
combined to create one large data set without a damage
variable (whole data set [WD]). However, since we were
also interested in the effect of experiencing damage on
attitudes, we used the smaller data set (damage data set
[DD]) to examine this.
We conducted 2 types of multivariate analyses. First,
we used classification and regression trees (CART)
(Breiman et al. 1993) to produce importance plots and
cost sequence plots (Supporting Information). Second,
we used logistic regressions to calculate Wald statistic
and odds ratios. For both analyses, we used Statistica
11 (StatSoft 2012). Due to the exploratory approach
of the CART procedure and subsequent risk of overfitting
the data, we randomly split the data set into a test
sample of 30% of all records and a train sample of the
remaining 70% of the data. We compared the results
of these 2 subsets to check the validity of our tests.
Analysis of the damage extent variable was conducted
using one-way ANOVA with Fisher least signficant difference
(LSD) post hoc tests. As described above and
in Table 1, we used 2 data sets WD and DD and thus
conducted 2 analyses (CART and logistic regression) on
each. We also conducted 2 scales of analysis, the first on
primary variables (column 1 in Table 1) and the second
on secondary variables (column 3 in Table 1). Secondary
variables formed subcategories of primary variables. For
example, the primary variable stakeholder comprised
4 secondary variables: commercial farmers, communal
farmers, urban residents, and others. For most analyses
we report on the WD only, while analyses of the DD
are reported when examining the effect of experiencing
damage on an individual’s attitudes. We defined tolerance
as “the proportion of individuals who have a positive
attitude toward a species group despite suffering damage
by that species group” and computed a tolerance
to damage index (TDI) as follows: TDI = proportion of
individuals suffering damage – (1 – proportion of individuals
with positive attitudes), where the proportion
of individuals suffering damage is the proportion of the
respondents in a study who experienced some damage
from a species and 1 – proportion positive is the proportion
of individuals in a study whose responses were
nonpositive.
A tolerance value of 0 indicates neutrality (i.e., proportion
of respondents with a positive attitude is proportional
to the proportion of respondents experiencing
damage). A negative value indicates low tolerance, and
a positive value indicates high tolerance. Because we
could not match damage data to individual attitudes, we
calculated this index with publication level data and thus
could not incorporate sample sizes of each study into this
index.
We identified 508 publications related to the topic
of HWC, which was refined down to 54 publications
that met the criteria for inclusion in the meta-analysis
(Supporting Information). When coded, this produced a
data set of 83,820 individual responses for the WD and
28,436 individual responses for the DD. The 54 publications
covered 22 countries and 43 different species
(Supporting Information). Twenty-two (41%) of the studies
were conducted in developed nations and 32 (59%)
in developing nations. One publication was conducted
in both developed and developing countries (Supporting
Information).
The number of publications which surveyed people’s
attitudes toward different carnivore species (64) was
more than twice the number of publications which
surveyed people’s attitudes toward different ungulate
species (30), 9 times more than the number of publicationswhich
surveyed people’s attitudes toward elephants
(7), and 16 times more than the number of publications
which surveyed people’s attitudes toward primates (4)
(Supporting Information). Considering the total number
of respondents surveyed, 81% were surveyed on their
attitudes toward carnivores, 14% were surveyed on their
attitudes toward ungulates, 3% were surveyed on
their attitudes toward elephants, and 2% were surveyed
on their attitudes toward primates. Attitudes of respondents
were solicited for 22% of carnivore species (International
Union for Conservation of Nature [IUCN] total
= 285 spp.), 9% of ungulate species (IUCN total = 329
spp.), and 1% of primate species (IUCN total = 414 spp.)
listed on the IUCN Red List (2008). The percentage for
elephants was 3500% because there are only 2 species.
MethodsWe conducted a meta-analysis of peer-reviewed journalarticles published from 1 January 1990 through March2011 in English that quantified the attitudes of stakeholderswho had experienced direct conflict with carnivores,elephants, primates, or ungulates. We defined an attitudeas “a disposition or tendency to respond with somedegree of favor, or not, to a psychological object” (Fishbein& Ajzen 2010). To qualify, attitude measures hadto be evaluative and quantified on a scale. Studies detailingattitudes of individuals not having direct experiencewith HWC were excluded because the general publiccan have more positive attitudes toward wildlife whennot directly affected (Williams et al. 2002; Mart´ın-L´opezet al. 2008), although, in some cases, negative attitudesare displayed by people not having contact with a species(Treves et al. 2013).We restricted our references to thosepublished in scientific journals (Calver & King 1999).Although inclusion of gray literature in meta-analyses issometimes recommended to prevent publication bias forsignificant results (Rosenthal 1979), this was not an issuein our review because attitudeswere recorded as percentageof respondents having positive, neutral, or negativeattitudes.We searched Web of science for terms listed in SupportingInformation. We then located additional publicationsby examining the reference list of each publication.Finally, we refined the publications to include onlythose published after 1990 because studies conductedbefore 1990 were few and commonly applied outdatedmethods. We then examined the selected publicationsin detail and extracted and compiled 8 variables in anExcel spreadsheet. The variables extracted were definedby their availability across all publications and their relevanceto our research questions. The variables are definedin Table 1.Data AnalysesThe attitudes reported in percentages in each publicationwere extracted and converted to a binary variableas either positive or nonpositive. A binary variable wasnecessary because some publications reported 2 categoryresponses (e.g., yes or no) to attitude questions. Where amiddle value of an attitude scale was used,we categorizedit as either a positive or nonpositive value depending onthe context of the question. For example, for questions,such as would you like the population of species x toincrease, stay the same, or decrease? We combined “staythe same” and “increase” because we considered “staythe same” to be more aligned with a positive rather thannonpositive attitude. For cases where the middle valuewas not obviously aligned with a positive attitude, responseswere categorized as nonpositive. We think it ismore robust to have a false negative than a false positivebecause assuming people are more positive than they arewould be more detrimental to a species.We assigned responses for each individual participatingin a survey to a positive or nonpositive attitude categoryusing the following computation: if 20% of a surveysample of 300 individuals reported positive attitudes, 60individuals were coded as positive and 240 nonpositive.To derive a similar individual record for the damagevariable, we converted the percentage of respondentsexperiencing damage into a probability of experiencingdamage per individual. For example, if 40% of a sampleexperienced damage, then the probability of eachindividual experiencing damage was 0.4. We assigned aprobability to each individual rather than a definitive yesor no because information on individual respondents wastypically unavailable.Not all publications reported what proportion of thesample experienced damage from individual species.Wetherefore compiled 2 types of data sets, a smaller onewhich did not report a damage proportion and a largerone that did. For most analyses, we used the 2 data setscombined to create one large data set without a damagevariable (whole data set [WD]). However, since we werealso interested in the effect of experiencing damage onattitudes, we used the smaller data set (damage data set[DD]) to examine this.We conducted 2 types of multivariate analyses. First,we used classification and regression trees (CART)(Breiman et al. 1993) to produce importance plots andcost sequence plots (Supporting Information). Second,we used logistic regressions to calculate Wald statisticand odds ratios. For both analyses, we used Statistica11 (StatSoft 2012). Due to the exploratory approachof the CART procedure and subsequent risk of overfittingthe data, we randomly split the data set into a testsample of 30% of all records and a train sample of theremaining 70% of the data. We compared the resultsof these 2 subsets to check the validity of our tests.Analysis of the damage extent variable was conductedusing one-way ANOVA with Fisher least signficant difference(LSD) post hoc tests. As described above andin Table 1, we used 2 data sets WD and DD and thusconducted 2 analyses (CART and logistic regression) oneach. We also conducted 2 scales of analysis, the first onprimary variables (column 1 in Table 1) and the secondon secondary variables (column 3 in Table 1). Secondaryvariables formed subcategories of primary variables. Forexample, the primary variable stakeholder comprised4 secondary variables: commercial farmers, communalfarmers, urban residents, and others. For most analyseswe report on the WD only, while analyses of the DDare reported when examining the effect of experiencingdamage on an individual’s attitudes. We defined toleranceas “the proportion of individuals who have a positiveattitude toward a species group despite suffering damageby that species group” and computed a toleranceto damage index (TDI) as follows: TDI = proportion ofindividuals suffering damage – (1 – proportion of individualswith positive attitudes), where the proportionof individuals suffering damage is the proportion of therespondents in a study who experienced some damagefrom a species and 1 – proportion positive is the proportionof individuals in a study whose responses werenonpositive.A tolerance value of 0 indicates neutrality (i.e., proportionof respondents with a positive attitude is proportionalto the proportion of respondents experiencingdamage). A negative value indicates low tolerance, anda positive value indicates high tolerance. Because wecould not match damage data to individual attitudes, wecalculated this index with publication level data and thuscould not incorporate sample sizes of each study into thisindex.We identified 508 publications related to the topicof HWC, which was refined down to 54 publicationsthat met the criteria for inclusion in the meta-analysis(Supporting Information). When coded, this produced adata set of 83,820 individual responses for the WD and28,436 individual responses for the DD. The 54 publicationscovered 22 countries and 43 different species(Supporting Information). Twenty-two (41%) of the studieswere conducted in developed nations and 32 (59%)in developing nations. One publication was conductedin both developed and developing countries (SupportingInformation).
The number of publications which surveyed people’s
attitudes toward different carnivore species (64) was
more than twice the number of publications which
surveyed people’s attitudes toward different ungulate
species (30), 9 times more than the number of publicationswhich
surveyed people’s attitudes toward elephants
(7), and 16 times more than the number of publications
which surveyed people’s attitudes toward primates (4)
(Supporting Information). Considering the total number
of respondents surveyed, 81% were surveyed on their
attitudes toward carnivores, 14% were surveyed on their
attitudes toward ungulates, 3% were surveyed on
their attitudes toward elephants, and 2% were surveyed
on their attitudes toward primates. Attitudes of respondents
were solicited for 22% of carnivore species (International
Union for Conservation of Nature [IUCN] total
= 285 spp.), 9% of ungulate species (IUCN total = 329
spp.), and 1% of primate species (IUCN total = 414 spp.)
listed on the IUCN Red List (2008). The percentage for
elephants was 3500% because there are only 2 species.
การแปล กรุณารอสักครู่..
