Professional Documents
Culture Documents
A Keyword Approach To Finding Common Ground in Community-Based Definitions OfHuman Well-Being
A Keyword Approach To Finding Common Ground in Community-Based Definitions OfHuman Well-Being
Human Well-Being
Author(s): R. S. Fulford, I. Krauss, S. Yee and M. Russell
Source: Human Ecology, Vol. 45, No. 6 (DECEMBER 2017), pp. 809-821
Published by: Springer
Stable URL: https://www.jstor.org/stable/45154066
Accessed: 20-04-2022 16:45 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Springer is collaborating with JSTOR to digitize, preserve and extend access to Human
Ecology
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
HumEcol (2017) 45:809-821 CrossMark
DOI 1 0. 1 007/s 1 0745-0 1 7-9940-3
Abstract Ecosystem-based management involves the integra- community type yielded significant differences in five of the
tion of ecosystem services and their human beneficiaries into eight domains examined. Community type differences follow-
decision making. This can occur at multiple scales; addressing ed an urban to rural trend with urban communities focusing on
global issues such as climate change down to local problems Education and Living Standards, and more rural communities
such as flood protection and maintaining water quality. At the focused on Social Cohesion and Leisure Time. Across all eight
local scale it can be challenging to achieve a consistent and domains multivariate analysis suggested communities were
sustainable outcome across multiple communities, particularly distributed along two largely orthogonal gradients; one be-
when they differ in resource availability and management pri- tween Living Standards and Leisure Time and or Connection
orities. A key requirement for consistent decision support at the to Nature, and a second between Safety and Security and Social
community level is to identify common community objectives, Priorities (Education/Health/Culture/Social Cohesion). Overall
as these can form the basis for readily transferable indices of these findings demonstrate the use of automated keyword anal-
ecosystem benefit and human well-being. We used a keyword- ysis for obtaining information from community strategic plan-
based approach to look for common terminology in community ning documents. Moreover, the results indicate measures and
fundamental objectives as a basis for transferable indices of perceptions of well-being at the local scale differ by community
human well-being and then compared those commonalities to type. This information could be used in management of eco-
community demographics, location, and type. Analysis cen- system services and development of indices of community sus-
tered on strategic planning documents readily available from tainability that are applicable to multiple communities with
coastal communities in the conterminous United States. We similar demographics, regional location, and type.
examined strategic planning documents based on eight do-
mains of human well-being, and found that Living Standards Keywords Ecosystem services • Well-being • Decision
and Safety and Security were the most commonly addressed support • Fundamental objectives • United States
domains, and Health and Cultural Fulfillment were the least. In
comparing communities, regional differences were observed in
only one well-being domain, Safety and Security, while Introduction
Electronic supplementary material The online version of this article There has been a systemic shift away from managing single
(https://d0i.0rg/l 0. 1 007/s 1 0745-0 1 7-9940-3) contains supplementary issues or systems in isolation towards the concept of
material, which is available to authorized users.
ecosystem-based management (Bennett et al. 2015;
O'Higgins et al. 2014). Ecosystem-based management has at
IS! R. S. Fulford
its core the implicit understanding that human-environment
Fulford.Richard @ epa.gov
interactions are not simply stressors to be understood, but
complex interactions that affect ecosystem dynamics and
1 Gulf Ecology Division, US Environmental Protection Agency, 1 drive management outcomes (Bennett et al. 2015; Larson
Sabine Island Road, Gulf Breeze, FL 32561, USA
et al. 2013). Large-scale human induced stressors, such as
2 E3436 Valley Lane, Kewaunee, WI 54216, USA air pollution (Eshleman and Sabo 2016; Reid and Aherne
Í) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
810 Hum Ecol (2017) 45:809-821
2016), eutrophication of public waterways (Ascott et al 2016; sociological (Hsieh and Shannon 2005) and lexicographic re-
Wu et al. 2016), and climate change (Hernandez-Montilla search (Ball 1994; Berber Sardinha 1996). In most cases, this
et al. 2016; Yanez- Arancibia et al. 2013) have received con- involves the identification of keywords across documents of a
siderable attention, but smaller scale perturbations, particular- common type or corpus (Lam 2007) and subsequent analysis
ly at the local level (e.g., coastal land use) are usually treated of keyword relative frequency (Kilgariff 1996b). Unlike a
as a separate problem to be managed on a case by case basis typical corpus, strategic planning documents are not standard-
(Adams et al. 2016; Kozak et al. 2016). Local decision mak- ized with respect to organization, content, or scope; thus, a
ing occurs at smaller scales, but can have a cumulative impact comparison of message will require identification of meaning-
comparable to issues like air pollution when many communi- ful keywords prior to analysis. However, when this require-
ties are tackling the same problems with similar solutions. ment can be met, a keyword-based analysis has provided im-
This provides an opportunity for decision guidance and sup- portant insights on similarities and differences both among
port at the community level to have a large impact on human and within multiple document types (Berber Sardinha 1996;
well-being. The challenge in scaling up decision guidance Lam 2007; Thomas and Wilson 1996).
from one community to many is to understand commonalities We applied a summative content analytical approach
in community goals and priorities that could form the basis for (Hsieh and Shannon 2005) based on pre-selected keywords
consistent and comparable decision support. We currently lack to examine a set of strategic planning documents at the county
a practical way to identify this common ground and make it level from coastal counties throughout the conterminous
useful for ecosystem-based management. United States to identify community fundamental objectives
A key component of decision support is the identification of similar to those obtained from associated in-person communi-
measurable outcomes that have broad stakeholder support. ty engagement (Fulford et al. 201 6). We limit the scope of our
Human well-being has been identified as a qualitative frame- analysis to coastal counties both to provide a narrowed context
work useful for sustaining quality of life (Costanza et al. 2007). for the analysis and because coastal communities are subject
More recently, quantitative measures of the many facets of to several common issues (e.g., sea level rise, coastal habitat
human well-being have been developed that offer a viable path loss, population increases) amenable to a unifying theme for
for measuring success of community decision making by eval- decision making. These strategic planning documents are
uating the contribution of decisions to the domains of human most often produced at the county level offering the most data
well-being (Smith et al. 2013b). Yet, communities differ in for comparison. Our objective is to validate an analytical ap-
their well-being priorities (Fulford et al. 2015) and measuring proach, use that approach to generate results for a core set of
changes in human well-being is complicated by these differ- communities, and then compare the outcomes between com-
ences. We seek common ground in community interpretation munities as a function of geography; as well as social, envi-
of human well-being that is useful for setting goals and mea- ronmental, and economic differences. The outcome will be
suring success of ecosystem-based management. insight into the common themes of human well-being among
Many communities have invested significantly in the com- coastal communities useful for making sustainable environ-
pilation of strategic planning documents intended to guide cur- mental decisions, and a practical tool for identifying these
rent and future decision making (Ex. http://www.co.benton.wa. commonalities in community goals.
us/; http://www.co.la-crosse.wi.us/comprehensiveplan/). These
documents usually involve a thorough examination of the
community including solicitation of stakeholder input and Methods
evaluation of community resources. It may be possible to
describe stakeholder priorities and community fundamental Communities were selected for analysis based on a definition
objectives, and ultimately measures of success toward of coastal county used by NOAA, which includes 662 counties
sustainability goals, from an examination of these documents nationwide (http://www.census.gov/geo/landview/lv6help/
using content analysis (Hsieh and Shannon 2005). Yet, these coastal_cty.pdf). Within this set of all coastal counties, the
documents are not uniformly structured, use inconsistent lan- sample set was further reduced to those counties with an
guage to describe similar things, and may be written to meet available strategic planing document. We then further edited
different goals (e.g., long-term planning vs. problem solving). the list of included counties to three regional areas, which
Thus, the challenge for making use of information in commu- allowed for a numerically balanced geographic comparison.
nity strategic planning documents for identifying commonali- Keyword analysis of strategic planning documents proceeded
ties is to identify and test an analytical framework that is well- in three steps. First, the keyword list was created and validated
suited to identifying common themes in language and can pro- through comparison of outcomes from a test set of strategic
vide consistent analysis of a wide variety of document types. planning documents. Initial keyword selection was based on
Text-based analysis and comparison of documents based the eight domains of the human well-being index (Table 1)
on word frequency is a well-established technique in (HWBI; Smith et al. 2013b). The HWBI framework is a useful
5) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 811
Table 1 Summary of the eight domains of the Human Well-being as well as example keywords used for each domain. The full keyword list
Index (HWBI; Smith et al. 2013b) used for this study. Domain labels is available online in Supplementary data
are followed by specific examples of service metrics for each domain,
Safety and Security Being and feeling safe; exposure to natural hazards Safe, hazard, toxic, financial, disaster
Connection to Nature Spiritual or emotional connection to nature Nature, trail, hike, hunt, fish, boat, biodiversity,
natural beauty, outdoors
Cultural Fulfillment Relative importance of culture and cultural activities Heritage, culture, religion, festival, art
to well-being
Social Cohesion Participation in community and democracy; importance Town, community, satisfaction, historic, sense of place,
of sense of place; satisfaction with community pride, quality of life, vote, volunteer
*Keyword examples are summarized and do not include multiple versions of a word used in the analysis. A full description of the keyword list is
available online in the Supplementaiy material
organization tool as it links community decision-making to by a development group (n = 4) based on their expert knowl-
elements of human well-being based on both stakeholder input edge of the domains in the HWBI. This group served as a
and objective analysis of community characteristics (Smith starting point for connecting words and phrases used by com-
et al. 2013a; Summers et al. 2014). The keyword list also munities to categories of human well-being as described by
included the development of 'near' and 'exclude' words the HWBI. While this effort proceeded in an ad hoc manner,
intended to parse out generic word use (e.g., 'nature trial' vs. the approach closely followed practical recommendations on
'the nature of polities'). Second, the keyword list was refined development of a code book for qualitative text analysis from
and finalized based on observed outcomes from a validation MacQueen et al. (1998). The development group created the
analysis. Finally, the refined keyword list was used to analyze a initial keyword list through a line by line examination of the
new set of strategic planning documents obtained for 97 com- index community planning document (Table 2). Word selec-
munities in the coastal region of the conterminous United tion was based on previous keyword development experience
States (Fig. 1). of the expert group and a consensus that chosen words clearly
The keyword list was initially created based on an exami- reflected a community value or priority rather than a narrative
nation of a single representative community planning docu- or an external expert opinion. The goal was to allow the words
ment (Table 2). Keywords were extracted from this document of stakeholders to drive the development of keywords rather
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
812
Table 2 Strategic planning documents used for the validation of the keyword list used in this analysis. Document for Moss Point, MS is at the civic
level and was used for initial assembly of the keyword list but not for validation
♦Representative strategic planning document used for initial development of keyword list
than scientific jargon (MacQueen et al 1998). This initial list workshops were similarly based on the HWBI framework
was then culled by collective discussion among the develop- and a comparison of keyword-based results to the results of
ment group to create a test keyword list, which was then sub- direct stakeholder input (Fulford et al 2016). These work-
jected to validation. shop results involved both a group and an individual ranking
Validation of the test keyword list involved the comparison of the relative importance of the eight domains of HWBI
of manual (i.e., by a person) evaluation of selected strategic (Smith et al 2013b). The validation comparison of automat-
planning documents to an automated (i.e., keyword-based) ed results to both manual and workshops results were exam-
evaluation of the same documents (Creswell and Miller ined by the validator group and the keyword list was further
2000). The test keyword list already described was evaluated amended to improve agreement between the manual/
by comparison of keyword and manual reads of five strategic workshop and automated reads. In addition, the possibility
planning documents not previously used for development of of a keyword hit not matching any of the domains of HWBI
the keyword list (Table 2). The automated read proceeded as was considered with the inclusion of an 'Other' domain
described below with the test keyword list. The manual read during validation. The Other domain was ultimately not
was completed by a selected group of five validators, each of used for analysis, as the eight domains are highly inclusive
which read one or two of the five documents. The validators and little information was placed in this category. For the
were asked to extract any statement that they considered a purposes of amending the keyword list (See Supplementary
match to a domain of the HWBI. The validators worked inde- material), the manual/workshop results were considered
pendently and were not given the keyword list prior to com- most representative of HWBI domain categories.
pleting their analysis of a document. Once the set of keywords was labelled as final by the valida-
Once the manual read was complete the validators met as tion group, a separate set of community planning documents
a group to review and evaluate the results of the manual (See Supplementary material) was selected and examined using
read. The results were reviewed for disparity between indi- the automated keyword search combined with the final keyword
vidual manual reads of the same document for any domain list. This set of community planning documents was selected
and questionable matches were discussed and reconsidered from a systematic search of all counties in three coastal regions
for inclusion by the group. The final results of the two in the conterminous United States (Oregon/Washington state,
manual reads were combined for each document and com- Minnesota/Wisconsin, and Louisiana/Florida). These regions
pared to the results of the automated read (Chew and were selected to allow for a broad geographic comparison along-
Eysenbach 2010). The manual and automated reads were side the focal comparison of demographics and community type.
compared and evaluated for agreement based on four met- The documents were found on county websites; if none were
rics: relative importance of the eight domains for each doc- available, the county was excluded. Plans were downloadable
ument measured as proportion of total matches per read, in .pdf format, either uploaded as images or text. When uploaded
number of statements per document found in both read as images Adobe Pro's Optical Character Recognition (adobe,
types, number of statements per document found in manual com) was used to change the images into text format. In each
but not automated read, and vice versa. The latter three case it was necessary to convert the .pdf file into a text file (.txt)
metrics represent indicators of exact Tine by line' matching before keyword analysis. Formatting issues and unnecessary
of outcomes from keyword and manual examination of the information (e.g., historical narrative) were manually corrected
same document in that the same line is chosen (or not) by to minimize file conversion errors. Since the objective was to
both methods. Additionally, validation included the compar- evaluate priorities for future community actions, narrative
ison of automated reads to the results of stakeholder work- elements, such as tables, figures, background history, table of
shops held in the same communities ( n = 4). These contents, and appendices, were all manually removed.
€) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 813
The reformatted documents were then analyzed with an domain. The hits were totaled at the domain level for each
original text search algorithm written in the software package separate planning document. To take into account varying
R (https://cran.r-project.org/). The R-script was used to extract lengths of planning documents, indices were normalized rela-
statements containing the selected keywords. Each document tive to the total hits per document. Data from all planning
was transformed into one continuous text string and then sep- documents were then combined, summarized, and compared
arated into lines by periods, semi-colons, colons, tabs, ques- by region (WA/OR, MN/WI, and LA/FL), as well as by popu-
tion marks, and exclamation points. The R-script then lation size, education, median income, and community type
searched the separated lines individually for select keywords. based on an existing community classification system
Selection was further parameterized by "near" and "exclude" (Fulford et al. 2015) (Table 3). Comparison of relative hits by
words identified during the validation process. Phrases were domain were made with analysis of contingency tables (Zar
only selected as hits if they included a keyword and a near 2010). In addition, the combined dataset was analyzed with a
word (Example for HWBI domain Connection to Nature: key- principal components analysis in order to examine broader pat-
word - environment; near word - improve; exclude word - terns involving multiple independent variables.
business). If one of the exclude words was in the phrase it was
not counted as a hit. Each HWBI domain had a unique set of
keywords chosen to characterize that domain and each key- Results
word had multiple near and exclude words to further refine
domain-specific hits. Near and exclude words were initially Keyword List Validation Keyword selection proceeded as
identified during the examination of the manual reads, but described with an initial list of words selected from a detailed
were further refined during the keyword analysis based on review of the planning document for Moss Point, MS
examination of keyword results. (Table 2). This document was selected by the keyword devel-
Two different outputs of keyword importance were provided opment group as representative of strategic planning efforts to
by the R-script: a count of total hits per domain and a complete be targeted for the general analysis and contained a maximum
list of every statement categorized as a hit. During the valida- comprehensive list of focal areas under consideration. Review
tion phase, the statement list was manually reviewed to verify of this document resulted in a list of 122 keywords each hav-
the R-script was pulling phrases that fit the intent of the key- ing between one and 15 near and exclude words respectively.
words. If a keyword was pulling hits established as 'false' upon This initial list was reviewed by the panel and adapted into the
manual review, the near and exclude words were altered to test keyword list for validation.
maximize positive hits. Duplicate hits (i.e., same phrase con- Validation of the keyword list proceeded first with the
taining multiple keywords and chosen twice for the same do- paired analysis of six test planning documents (Table 2).
main) were removed via the R-script for each indicator and Overall the largest difference in hits between manual and
Table 3 Summary of community classification (CC) used for structure and data can be found in (Fulford et al. 2016). For each
comparison of keyword-based analysis of community planning community classification group (CC1-CC8) the count of communities
documents. The CC groups used were developed as a part of an earlier in the original analysis of all US coastal counties is given for reference
study of Human well-being (Fulford et al 201 5) and full details of the CC
CCI: Urban Urban/suburban 223 Economically balanced; All communities >5000 pp./mi2
CC2: Rural workers Rural/older/manufacturing 129 Contains all communities with high dependence
on industry; higher median age.
CC3: Rural farms Rural/dependence on local natural resources 1 33 Mix of dependence on natural resources and service
in larger rural communities
CC4: Rural diverse Rural/farming/high ethnic diversity 3 Subset of CG6 with extreme dependence on natural resources
and high ethnic diversity score
CC5: Rural balance Rural/natural resources mixed with industry 45 Mixed economic dependence on natural resources
and industry
CC6: Rural natural res. 15 High dependence on natural resources low on industry
CC7: Older suburban High median age; dependencies similar to 46 All in Florida; unique suburban communities characterized by
CGI but communities smaller in pop. Size. urban economically, but rural
in demographics.
CC8: Suburban industry Small to rural towns with high economic 25 Mostly in Great Lakes region
dependence on industry
Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
814 Hum Ecol (2017) 45:809-821
keyword reads were observed in the Living Standard domain Table 4 Summary of line by line matches between keyword and
manual reads of test documents organized by HWBI domain. Test
(Figure 2)ģ Positive differences indicate higher frequency of hits documents are described in Table 2
for the automated read. This domain had a difference exceeding
20% for three documents and exceeding 30% for one (i.e., Domain Keyword only Human only Both Total
Wayne County, PA). Maximum normalized difference was be-
Connection to Nature 91 27 4 122
low 1 5% for all other domains and documents. Living Standards
Cultural Fulfillment 65 37 8 110
also had the most overall hits among the eight domains with an
Education 122 21 14 157
average of 38 hits per document. Another notable difference was
Health 69 39 14 122
that for Living Standards the difference was positive for the
manual read meaning the manual results were higher than the
Leisure Time 162 44 19 225
keyword results. With the exception of the Wayne County doc-
Living Standards 175 281 43 499
Safety and Security 146 95 31 272
ument the majority had fewer hits for the manual read. The
greatest disagreement in results between documents occurred
Social Cohesion 121 113 17 251
in the Social Cohesion and Safety and Security domains where
three documents were positive for manual read and three were
negative. For a specific document Wayne County showed the higher than between the two human reads. These occurred twice
most disagreement with the other documents showing differ- in health, then once in each Education, Safety and Security , and
ences in sign, magnitude or both for five of the eight domains. Social Cohesion. Keyword list finalization involved selection of
An exact line by line examination of the outcome of both read the list that generated maximum matches between human and
types indicated lower agreement between human and automatic automated reads with a primary focus on proportional importance
reads when attempting to make a specific match (Table 4). In of domains and a secondary focus on line matches.
most cases, there were less than 10% matching phrases per do- The second stage of keyword validation involved the com-
main and document between human and automatic reads. The parison of the automated read to data from stakeholder engage-
highest agreement across all test documents was in the Health ment (Fulford et al. 20 1 6). Several differences were observed in
and Safety and Security domains (11%) and the lowest was for the outcomes of these two methods when they were applied to
Connection to Nature (3%). Of three planning documents with the same communities (Figure 3). The workshop method gen-
two human reads, there were three instances (i.e. domain- erated more diverse findings that nonetheless consistently re-
document combination) where human to human agreement was ported high importance in the domains of Education and
noticeably higher than human to automatic reads. Two of these Social Cohesion. In contrast, the keyword method was always
occurred in the Living Standards domain while the other one was dominated by the Living Standards domain, which is the prima-
in Leisure Time. In contrast, there were five instances where line ry economic domain of the HWBI. However, if one removes the
matches between human and automatic reads were noticeably Living Standards domain from the keyword results, the relative
Fig. 2 Summary of normalized differences between keyword and perfect agreement, while positive and negative results indicate more or
manual reads of selected test documents used for validation of the less hits respectively in the keyword analysis for a particular domain.
keyword list. See methods for more details. The Other category was Labels indicate county name followed by state abbreviation. Jackson
included to allow for phrases selected that did not fit one of eight County, MS includes separate analyses at county and neighborhood scale
domain categories, but it was not used for analysis. Zero indicates
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 815
Fig. 3 Summary comparison of combined results from stakeholder (hashed; WS rank), results from a group voting exercise during the
engagement and keyword analysis of strategic planning documents in workshop (solid; WS dot vote), and results of the relative proportion of
four focal communities. Community name is given in each panel. Three hits from the keyword analysis (diamond; Keyword). See Fulford et al
bars are results from an individual ranking exercise during the workshop 2016 for details on stakeholder workshops
importance of the remaining domains is highly congruent with (5. 1 5) were consistently the least mentioned domains. By contrast,
workshop results (Figure 3). Therefore, the findings of the key- Living Standards (23.82), Safety and Security (15.52), and Leisure
word analysis can be thought of as hierarchical with a primary Time (15.12) were mentioned most often. In the middle were
dominance for Living Standards as a priority in strategic plan- Education (8.09), Connection to Nature (11.3), and Social
ning, but the domains of secondary importance (i.e., domains Cohesion (14.06). Living Standards consistently had the highest
other than Living Standards) were more similar to workshop median value across all documents analyzed (Table 5; Figure 4).
outcomes. There is, thus, strong support for the complementar- Similar patterns among domains are seen when data are
ity of the two methods and the validity of the final keyword list. organized by region, but significant differences were evident
between regional community groups only for the Safety and
Community Comparison of Keyword Analysis Keyword- Security domain (Table 6). Of the three regions (WA/OR, MN/
based analysis of 97 planning documents indicated important dif- WI, and LA/FL) analyzed, each had Living Standards as the
ferences among domains in a community's stated priorities based most represented domains. The middle category included
on normalized keyword hits for each HWBI domain (Table 5, Safety and Security , Social Cohesion , Leisure Time , and
Figure 4). Cultural Fulfillment (median value: 3.64) and Health Connection to Nature , while the low group included Cultural
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
816 Hum Ecol (2017) 45:809-821
The domain Living Standards is again either the most or sec-
ond most represented domain in for each classification group,
while Cultural Fulfillment and Health remain the least repre-
sented domains similar to the overall results. Yet, among the
five of eight classification groups with at least 5 documents
(Groups 1, 2, 3, 5, and 7; Table 3) there also were some
important differences. Community classification (CC) Group
1 stands out for Connection to Nature with the lowest score of
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 817
Table 7 Summary of mean normalized keyword hits for the domains of system groups. The p-value is for a Chi-square goodness of fit test with
human well-being index organized by coastal community classification four degrees of freedom based on comparison of CC categories with at
(CC) groups. See Table 2 for description of community classification least five documents. Values in bold are significant at a = 0.05
affected by population size. Median age was split into two CF/SC/ED gradient with CC group 5 clustered at the SS end.
categories (>40 and <40) (Table 9) and the older group scored For the regional comparison (Figure 5a) the WA/OR group
higher for Connection to Nature and Leisure Time , while the was more variable along the SS - CF/SC/ED gradient but
younger group scored higher for Education. more clustered towards LT/CN on the other gradient. In con-
When variance in all variables, including both community trast, the MN/WI group was more variable along the LS-LT/
characteristics and keyword-based priorities by domain, are CN gradient but generally clustered more towards LT/CN and
analyzed together the individual patterns are easily observed CF/SC which is consistent with the significantly lower score
(Figure 5), but there are interesting pairings among variables for SS compared to the other two regions. The LA/FL group
suggesting some useful linkages for interpretation. The prin- was the most variable with regards to both gradients and dem-
cipal components analysis demonstrated clear negative rela- onstrated minimal pattern. Overall the multivariate results are
tionships between prioritization of Leisure Time (LT) and consistent with univariate results but show a pattern that is
Connection to Nature (CN) with prioritization of Living consistent across domains.
Table 8 Summary of mean normalized keyword hits for the domains of (U.S. Census Data; accessed 14 September 2016). Values were
human well-being index organized by proportion of adult citizens with normalized separately for HSD and BD
either a high school diploma (HSD) or a bachelor's degree (BD) in 2000
High school degree High school degree or High school degree Bachelor's degree Bachelor's degree or Bachelor's degree
or higher >85% higher 80-85% or higher 80%> or higher >20% higher 1 5-20% or higher 1 5%>
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
818 Hum Ecol (2017) 45:809-821
Table 9 Summary of mean normalized keyword hits for the domains of human well-being index organized by median income level, median age, and
population size in 2000 (U.S. Census Data; accessed 14 September 2016). Values normalized separately for income, age and population size
Count 4 50 54 54 54 38 35 35
Connection 12.09 11.59 12.23 12.65 11.16 12.38 11.41 11.88
Cultural Fulfillment 5.13 3.29 5.47 4.30 4.51 4.46 3.81 4.94
Education 7.65 8.38 9.91 7.54 9.62 7.47 8.61 9.76
Health 4.76 5.35 5.97 5.00 5.70 4.80 5.47 5.81
Leisure Time 16.75 14.94 13.41 16.60 13.49 16.95 13.98 14.05
Living Standards 23.28 24.24 25.17 23.33 25.11 23.70 24.30 24.70
Safety and Security 16.95 18.09 12.61 17.09 15.50 17.06 18.02 13.73
Social Cohesion 13.40 14.11 15.23 13.50 14.92 13.19 14.40 15.12
this analysis was the importance of quality of life metrics (i.e. communities and community types the consistently dominant
Living Standards, Safety) to stakeholder priorities. Across well-being domains, in terms of total number of keyword hits,
were Living Standards followed by Safety and Security
followed closely by Social Cohesion, Connection to Nature,
and Leisure Time. An interest in quality of life is not surpris-
ing. The consistent low scores for Education and Health were
surprising and suggest these are not community-level priori-
ties, but may be important at a different scale (e.g., personal/
family). For instance, even in cases where an action may di-
rectly benefit human health (e.g., investment in hospitals) the
community-scale priority for the action may not be directly
tied to improving community health, but rather to ancillary
benefits more aligned with community-scale priorities such
as job creation, reductions in burden on public services, or
community reputation. These differences can be important to
setting measures of success at the appropriate scale. It is also
important to understand if measures of success should differ
among community types (Bagstad and Shammin 2012).
A key objective of this study was to associate common
themes in well-being to communities of a common type to
make the information useful for decision support in multiple
communities. The dominant delineations for stakeholder pri-
orities at the community level were between multi-state re-
gions and community classification (CC) groups. For CC
groups, five could be meaningfully compared. The CC groups
differed most in Leisure time and Living Standards , which
tracks a trajectory between these two domains of wellbeing
mentioned earlier. It can be categorized as an economic tra-
jectory in contrast to the more social trajectory between Safety
and the cluster of Education , Health , Culture and Social
Fig. 5 Plot summarizing results of a principal components analysis of
normalized hits per domain of the human well-being index by individual Cohesion. Generally, the CC groups tended to vary between
document. Only PCI (x axis; loading 0.29) and PC2 (y axis; loading 0.20) groups along the economic trajectory and more within group
are shown. Documents are labeled for geographic region (a) or along the social trajectory. The most striking differences were
community classification group (b). Regions are Washington/Oregon
between the urban CC group (Group 1 ) and the most rural CC
(WA/OR), Minnesota/Wisconsin (MN/WI), or Louisiana/Florida (LA/
FL). Typology groups CCI - CC8 are described in Table 2. Note that group (Group 5), which had either the highest or lowest score
that panel (b) is zoomed in for clarity and has a smaller scale for seven out of eight well-being domains and were directly
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 819
antithetical for three domains ( Education , Leisure Time, and Keyword analysis is an effective technique for describing
Living Standards). The urban communities placed a higher community priorities for human well-being from large docu-
emphasis on Education and Living Standards , while the more ments such as the planning documents considered here, but
rural communities emphasized Leisure Time , Health , and the technique requires careful consideration of assumptions.
Safety. The differences observed in more specific demograph- Keyword analysis as a form of context analysis (Hsieh and
ic groups such as median age and income are important but Shannon 2005) has been commonly used to compare docu-
were generally ancillary to differences observed among the ments and corpora for similarity (Berber Sardinha 1996;
CC groups. The broader categories, such as CC, are not inde- Peirsman et al. 2010), to look at word frequency in lexico-
pendent from these specific categories (e.g., demographics) as graphic studies (Kilgariff 1996a), and use of automated tools
these data were also used as a part of the CC group score is a well-developed analytical technique (Ball 1994).
calculation (Fulford et al 2015). However, a lack of pattern However, these approaches typically involve analysis of word
in these strategic planning documents among demographic clusters without pre-defined meaning (Berber Sardinha 1999).
groups for most domains of wellbeing suggests specific de- The planning document analysis approach is similar in execu-
mographics issues are more individualistic in nature and may tion, but is dependent on the use of specific words groups
not drive community decision making other than to crystalize developed a priori with pre-assigned meaning based on the
individual interests.
HWBI framework. This makes the analysis more dependent
The regional comparison made here is limited to three on manual vetting of word lists, such as the validation proce-
coastal regions (WA/OR, WI/MN, and LA/FL), but this al- dure described earlier (Ball 1994; Creswell and Miller 2000).
lows for a useful comparison of spatial differences to differ- Therefore, keyword analysis can be useful, but its limits must
ences in other categories. For individual domains of be carefully considered in drawing conclusions.
wellbeing, communities within regions were most similar in Validation results indicated keyword outcomes here were
the less heavily cited categories like Connection to Nature, generally consistent with both a manual read of the same doc-
Health, and Education showcasing key regional differences ument, as well as direct engagement of stakeholders in the
in stated well-being priorities. The combined differences same community. Clarifying community priorities from doc-
among regions were largely along the social trajectory al- ument analysis is limited by the scope of the document, as
though the LA/FL regional group was highly variable and well as the level to which the document reflects community
hard to separate from the other two. input rather than the input of elected officials or hired external
The value of understanding these differences among groups is experts. Yet, these issues can largely be minimized by appro-
to identify the domains of human well-being for which the CC priate document selection, so the process of reviewing and
group or geographic delineations are the most informative for selecting documents for analysis should be reported for max-
identifying community priorities that may drive decision making imum value of the results. The documents for this study were
across multiple communities. These most informative differences selected from a random group of available documents from
lie either on a gradient from an emphasis on Safety and Security each region but also filtered to remove documents not based
on one end to an emphasis on 'community' domains centered on significant stakeholder engagement. The keyword results
around Social Cohesion on the other end or a gradient from were not biased by document length or organization, which
Living Standards to Leisure Time/Connection to Nature. This suggests a wide variety of documents can be potentially se-
generalization is supported by both the categorical results, as well lected for analysis. Exact line-by-line matching of results be-
as the multivariate analysis of all domains together. These two tween automated and manual reads was less consistent, but it
largely orthogonal gradients are also consistent with an urban to seems the two methods get to the same interpretation even
rural gradient in that it is directly related to population size, and with some variability in exact phrases aligned with particular
demographics as 'ruralness' tends to be related to an increased keywords. Manual reads of the same document were incon-
emphasis on social connectivity (Bagstad and Shammin 2012; sistent with each other, which created a lot of the discrepancy
Smith and Clay 2010). Communities that are more urban, more and suggests an objective keyword approach should generate
diverse, or less dependent on local natural resources also tend to more consistent results than interpretation of a document by
have a higher median income and a higher overall education multiple individuals.
level, which is consistent with, but not dependent on increasing The low level of line by line agreement between manual
urbanization. They also seem to prioritize Living Standards and and automated reads of the same documents is interesting in
Education ; and reduce priorities for Leisure Time and and of itself, as the manual reads did not target particular
Connection to Nature. The inverse relationship between educa- words but extracted whole phrases viewed by the validation
tional attainment in a community and their interest in Education group as consistent with one of the domains of human well-
was surprising but fairly consistent, which suggests it is a real being. This approach resulted in a closer match in terms of
trend among groups, insomuch as these characteristics influence total number of hits but differences in the actual phrases se-
organized goal-setting. lected in each case. Some domains such as Health and Safety
Ô Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
820 Hum Ecol (2017) 45:80^-821
had higher agreement suggesting the language used for these urban to rural gradient. Therefore, future research should fo-
domains is more consistent and identifiable. The intent was cus on improving the understanding of how community type
not to focus on the manual read, but use it as an independent may inform differences in the relative importance of the do-
check on the automated outcome, and this is why the keyword mains of human well-being that can be used to both develop
list was not used by the validation group. The numerical and assess decision options at the community level.
agreement is the more important outcome for this analysis,
Acknowledgements We would like to thank all those who aided in data
and line agreement was mainly used by the validation group
collection and analysis, as well as those involved in the development of
to hone the keyword list. That said, the line agreement infor-
the Human well-being index used in this study. In particular, we acknowl-
mation may be used in the future for a more detailed exami- edge Linda Harwell, Lisa Smith, Leah Oliver, Matt Harwell, and Virginia
nation of how these documents can be interpreted in compar- Hansen for their assistance in development and validation of the keyword
ison to other forms of stakeholder engagement. list used in this study. This paper was greatly improved by the comments
of two anonymous reviewers and we also thank them for their efforts. The
Keyword analysis of strategic planning documents shows
views expressed in this paper are those of the authors and do not neces-
great promise as a contributing method for clarifying the long- sarily reflect the views or policies of the U.S. Environmental Protection
term priorities of human stakeholders and for delineating com- Agency.
specific documents were used for the analysis and this is a Linguistic Computing 9: 295-302.
Bennett, E. M., Cramer, W., Begossi, A., Cundill, G., Diaz, S., Egoh, B.
topic for future study. These findings strongly suggest that
N., Geijzendorffer, I. R., Krug, C. B., Lavorel, S., Lazos, E., Lebel,
content analysis combined with a CC group-based compari- L., Martin-Lopez, B., Meyfroidt, P., Mooney, H. A., Nel, J. L.,
son can be very informative regarding differences in the rela- Pascual, U., Payet, K., Harguindeguy, N. P., Peterson, G. D.,
tive importance of community-scale priorities such as Social Prieur-Richard, A. H. N., Reyers, B., Roebeling, P., Seppelt, R.,
Solan, M., Tschakert, P., Tscharntke, T., Turner, B. L., Verbürg, P.
Cohesion , Living Standards , Leisure Time , and Safety.
H., Viglizzo, E. F., White, P. C. L., and Woodward, G. (2015).
Beyond the specifics, it is evident that communities differ Linking biodiversity, ecosystem services, and human well-being:
in how they rank and prioritize the domains of human well- three challenges for designing research for sustainability. Current
being and these differences are predictable and based on com- Opinion in Environmental Sustainability 14: 76-85.
Berber Sardinha, A. P. (1996). Review of WordSmith tools. Computers
munity type. This conclusion supports the value of communi-
and Text 12:19-21.
ty delineations for informing the decision process. However, it Berber Sardinha, A. P. (1999). Using keywords in text analysis: practical
also indicates that measures of success can only be partially acpects. In DIRECT papers, Working paper 42 ISSN 1413-442x:
generalized and the very definition of human well-being may CEPRIL, PUC-SP, Brazil, and AELSU, Liverpool University, England.
Chew, C., and Eysenbach, G. (2010). Pandemics in the age of twitter:
differ among community types, as has been pointed out in the
Content analysis of tweets during the 2009 H INI outbreak. Plos
past (Bagstad and Shammin 2012; Moller et al. 2015). Such One 5: el41 1 8.
differences must be kept in mind when comparing the objec- Costanza, R., Fisher, B., Ali, S., Beer, C., Bond, L., Boumans, R.,
tive well-being across communities, particularly along the Danigelis, N. L., Dickinson, J., Elliott, C., Farley, J., Gayer, D. E.,
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms
Hum Ecol (2017) 45:809-821 821
Glenn, L. M., Hudspeth, T., Mahoney, D., McCahill, L., Mcintosh, Implications for natural resources management and planning in the
B., Reed, B., Rizvi, S. A. T., Rizzo, D. M., Simpatico, T., and Snapp, Great Barrier Reef, Australia. Journal of Environmental
R. (2007). Quality of life: An approach integrating opportunities, Management 117: 226-234.
human needs, and subjective well-being. Ecological Economics MacQueen, K. M., McLellan, E., Kay, K., and Milstein, B. (1998).
61: 267-276. Codebook development for team-based qualitative analysis.
Creswell, J. W., and Miller, D. L. (2000). Determining validity in quali- Cultural Anthropology Methods 10: 31-36.
tative inquiry. Theory into practice 39: 124-130. Moller, V., Roberts, B., and Zani, D. (2015). The Personal Wellbeing
Eshleman, K. N., and Sabo, R. D. (2016). Declining nitrate-N yields in Index in the South African IsiXhosa Translation: A Qualitative
the Upper Potomac River Basin: What is really driving progress Focus Group Study. Social Indicators Research 124: 835-862.
under the Chesapeake Bay restoration? Atmospheric Environment O'Higgins, T., Cooper, P., Roth, E., Newton, A., Farmer, A., Goulding, I.
146: 280-289. C., and Tett, P. (2014). Temporal constraints on ecosystem manage-
Fulford, R. S., Smith, L. M., Harwell, M., Dantin, D., Russell, M., and ment: definitions and examples from Europe's regional seas.
Harvey, J. (2015). Human well-being differs by community type: Ecology and Society 19.
Toward reference points in a human well-being indicator useful for Peirsman, Y., Geeraerts, D., and Speelman, D. (2010). The automatic
decision support. Ecological Indicators 56: 194-204. identification of lexical variation between language varieties.
Fulford, R. S., Russell, M., Harvey, J., and Harwell, M. C. (2016). Natural Language Engineering 16: 469-491.
Sustainability at the community level: searching for common Reid, H., and Aherne, J. (2016). Staggering reductions in atmospheric
ground as a part of a national strategy for decision support. Gulf nitrogen dioxide across Canada in response to legislated transporta-
Breeze, FL: US EPA EPA/600/R-16/178. tion emissions reductions. Atmospheric Environment 146: 252-260.
Hernandez-Montilla, M. C., Martinez-Morales, M. A., Vanegas, G. P., Smith, C. L., and Clay, P. M. (2010). Measuring Subjective and Objective
and de Jong, B. H. J. (2016). Assessment of Hammocks (Petenes) Well-being: Analyses from Five Marine Commercial Fisheries.
Resilience to Sea Level Rise Due to Climate Change in Mexico. Human Organization 69: 158-168.
Plos One 1 1 . Smith, L. M., Case, J. L., Harwell, L. C., Smith, H. M., and Summers, J.
Hsieh, H. F., and Shannon, S. E. (2005). Three approaches to qualitative K. (2013a). Development of relative importance values as contribu-
content analysis. Qualitative Health Research 15: 1277-1288. tion weights for evaluating human wellbeing: An ecosystem ser-
Kilgariff, A. (1996a). Comparing word frequencies across corpora: vices example. Human Ecology 41: 631-641.
Why Chi-square doesn't work and an improved LOB-Brown Smith, L. M., Case, J. L., Smith, H. M., Harwell, L. C., and Summers, J.
comparison. In ALLC-ACH Conference, (ed.), A.o.L.a.L. K. (2013b). Relating ecoystem services to domains of human well-
Computing, Bergen, Norway. being: Foundation for a US index. Ecological Indicators 28: 79-90.
Kilgariff, A. (1996b). Using word frequency lists to measure corpus Summers, J. K., Smith, L. M., Harwell, L. C., Case, J. L., Wade, C. M.,
homogenity and similarity between corpora. In COLING workshop Straub, K. R., and Smith, H. M. (2014). An Index of Human Well-
on very large corpora. Being for the US: ATRIO Approach. Sustainability 6: 3915-3935.
Kozak, J. P., Bennett, M. G., Piazza, B. P., and Remo, J. W. F. (2016). Thomas, J., and Wilson, A. (1996). Methodologies for studying a corpus of
Towards dynamic flow regime management for floodplain restora- doctor-patient interaction. In Thomas J., and Short M. (eds.), Using
tion in the Atchafalaya River Basin, Louisiana. Environmental corpora for language research, Longman, London UK, pp. 92-109.
Science & Policy 64: 118-128. Wu, M. L., Wang, Y. S., Wang, Y. T., Sun, F. L., Sun, C. C., Cheng, H.,
Lam, P. Y. W. (2007). A corpus-driven lexico-grammatical analysis of and Dong, J. D. (2016). Seasonal and spatial variations of water
English tourism industry texts and the study of its pedagogic impli- quality and trophic status in Daya Bay, South China Sea. Marine
cations in English for Specific Purposes. In Hidalgo E., Quereda L., Pollution Bulletin 112: 341-348.
and Santana J. (eds.), Corpora in the Foreign Language Classroom, Yanez-Arancibia, A., Day, J. W., and Reyes, E. (2013). Understanding the
pp. 71-89. Coastal Ecosystem-Based Management Approach in the Gulf of
Larson, S., De Freitas, D. M., and Hicks, C. C. (2013). Sense of place as a Mexico. Journal of Coastal Research: 244-262.
determinant of people's attitudes towards the environment: Zar, J. H. (2010). Biostatistical analysis, Prentice Hall, New Jersey, USA.
£) Springer
This content downloaded from 132.174.250.76 on Wed, 20 Apr 2022 16:45:43 UTC
All use subject to https://about.jstor.org/terms