Ecosystem services is an interdisciplinary field examining the relationship between human welfare and environmental management. In recent years, the field has exploded from a niche literature into an organizing principle for research and management, influencing institutions such as university departments and government agencies. However, in some cases the terminology is overused such that the original intent to address a “lack of appreciation of societal dependence on natural ecosystems” (Daily 1997:xv) has been obscured. Although numerous studies self-identify as ecosystem services investigations, they do not always clearly link to human welfare.
The perspective of “final” ecosystem services can be seen as a means of refocusing environmental study on direct links to human health and well-being. Final services are select, valued endpoints of ecological production functions (Boyd and Banzhaf 2007, Ringold et al. 2009, 2013, Boyd and Krupnick 2013, Landers and Nahlik 2013, Boyd et al. 2016). A point to be stressed is that final services rely on extraordinarily important intermediate ecosystem functions and processes. The final model is not the only way to structure an ecosystem services inquiry, but it does carry the advantage of promoting attention on measurable environmental quantities of direct interest to the public, while preserving the role of underlying ecological complexity. Final services, once identified, may then serve as the variables for communicating environmental condition, or judging trade-offs between environmental outcomes in a form transparent for public review.
It should be noted that there is active debate regarding ecosystem services frameworks. In particular, the final services approach on measurable quantities differs from categorizing ecosystem services as provisioning, supporting, regulating, and cultural, as adopted by the Millennium Ecosystem Assessment (2005). Although we recognize differences in approach, we also assert that the goal of identifying natural resources that people care about is a broad arena that transcends the debate. A goal similar to ours occurs whenever investigators seek to directly link environmental measurements with human value, a common aspiration occurring under various names. For example, Keeler et al. (2012:18260) discussed an ecosystem services framework linking environmental changes “to changes in the provision of ecosystem goods and services that directly affect human well-being;” Wainger and Mazzotta (2011:712) described a similar ecosystem services framework hinging on “outcomes that matter to people,” and Haines-Young and Potschin (2010) framed ecosystem services as a cascade starting with biophysical processes and ultimately yielding human value via final products. It isn’t even necessary to use ecosystem services terminology. Schiller et al. (2001) described publicly relevant common-language indicators, and Braden et al. (2015:450) discussed the need to “systematically collect data at interfaces linking humans to the natural environment” in the context of sustainability. Scientists need not invoke specialized terms to take part in research describing the relationship between humans and nature. Likewise, people don’t need to be familiar with scientific jargon to appreciate the natural world. Studies have shown that people ascribe importance to the environment, despite unfamiliarity with the ecosystem services phrase (Böck et al. 2015, Nature Conservancy 2010).
With a perspective on final services, linking data, or any other similar term, the key question soon becomes: what are these select environmental attributes? Multiple original data methods can be used to identify them, such as expert opinion (Ringold et al. 2009, 2013), focus groups and/or semistructured interviews with laypersons (Weber and Ringold 2015), and survey research. Our investigation takes a different tack, relying instead on the vast arena of text available online.
To extract information from sampled texts, we used content analysis, a commonly used quantitative method to analyze qualitative data (Bernard 2011). Content analysis of secondary sources is increasingly common in natural resource management in which primary data may be cost prohibitive or incomparable across regions or time (Houston et al. 2010:392). Examples include media representations of human-wildlife conflict (e.g., Siemer et al. 2007, Muter et al. 2009, Houston et al. 2010, Alexander and Quinn 2011, Alessi et al. 2013), environmental risks or issues (e.g., Gunter and Harris 1998, Brossard et al. 2004, Gunter 2005, Tilt and Xiao 2010, Lopera and Moreno 2014, Nelson et al. 2014), and management and planning (Proctor 1998, Bengston et al. 2004, Norton 2008).
Our study builds on previous research directly engaging diverse members of the public (Weber and Ringold 2015; see also http://conference.ifas.ufl.edu/aces14/posters/Weber,%20Matthew%20-%20Oregon_FEGS_Poster_corrected.pdf). A valuable outcome of this prior work was a provisional codebook organizing the breadth of river and stream attributes and associated motivations for valuing them as manifest in layperson narratives. Whereas statistical analysis of transcripts is limited because of issues such as “groupthink” (Janis 1972), content analysis on a random sample of individual texts allows relatively straightforward hypothesis testing. Taken together, the chain of research comprises a classic mixed methods sequence having a preliminary exploratory phase to develop hypotheses, with follow-up statistical testing (Creswell 2014). Specifically, we address three topics: (1) Document the range and relative frequency of attributes and motivations used in texts to represent stream ecosystems and human interest in them; (2) Test whether attributes and motivations vary by publication source or date, and; (3) Test associations between defined categories of attributes and motivations.
Strategic sampling was a challenging aspect of the study. Overall, we wanted to compare and contrast different “voices” on rivers and streams, at a minimum, environmentalist perspectives vs. mainstream media. While drawing from different sources, we also wanted to control for the influence of time by using a consistent sampling window because newsworthy river-related phenomena can be episodic, e.g., flooding. An additional factor affecting sampling was that we found significant resources were needed to extract the information we sought in a reliable manner from a given article. This meant that we could not set a final sample size at the outset, but rather had to plan adaptively according to how quickly available project resources were used.
For environmentalist texts, we decided to analyze the National Geographic blog, Water Currents (http://voices.nationalgeographic.com/blog/water-currents/). Although not blatantly environmentalist, the National Geographic Society mission statement includes a reference to helping protect the diverse creatures that share our world, which stands out from the organizing principles of mainstream news sources. Blog contributors are affiliated with groups such as National Geographic itself, National Geographic’s Freshwater Initiative, and the Environmental Defense Fund. The blog is hosted by Sandra Postel, Freshwater Fellow with National Geographic and Director and Founder of the Global Water Policy Project. Blog content is almost exclusively devoted to rivers and streams. Blog articles were manually filtered to include only those dealing with domestic U.S. rivers and to exclude articles with only a passing reference to rivers and streams (defined as two or fewer paragraphs within an article). Initially, we had intended to census sample the blog; however, due to a high effort needed to code numerous concepts reliably, we did not have the capacity to sample the most recent posts, and the sample frame was restricted to November 11, 2009 (the date of the first post) to February 15, 2012, which included 28 blog posts. This interval became the boundaries for the sampling window applied to other sources.
In determining a mainstream news source, we sought something with high circulation and accessible indexing. The New York Times (NYT) and the Wall Street Journal (WSJ) have consistently high rankings for daily print and digital circulation and are both available through an online database (ProQuest 2017). Candidate content from NYT was determined through querying for subject index terms “rivers” and “creeks and streams.” Content was considered acceptable if it contained at least some discussion of U.S. rivers and/or streams. Within the sampling window there were 71 articles meeting our criteria, of which 40 were randomly selected for analysis.
After preliminary statistical analysis, we found significantly diminished influence on relative code frequencies from the last few blog and NYT articles analyzed. We thus decided to include several WSJ articles as a third source. This allowed us to explore whether the WSJ’s reputation as a conservative, business-interest periodical would show a dramatic contrast with other sources. Given the small sample size planned for WSJ, we used a slightly different sampling approach, stratifying across years 2010 and 2011 (instead of the full sample interval used for the blog and the NYT) to maintain the ability to test the influence of time. There were 52 articles meeting sampling criteria, of which 8 were randomly selected for analysis.
The codebook is hierarchical with two main categories: physical attributes of rivers and streams, and motivations for interest in rivers and streams. Attributes are further divided into water, fish, wildlife, vegetation, channel, and human. Note in particular the last category: not all measurable attributes of streams are ecological, and we purposefully included human attributes to allow the contrasting of the incidence of ecological features vs. human modifications. Motivations are divided into consumptive use, nonconsumptive use, and not use contingent. These categories facilitate distinguishing the traditional extractive uses from the recreational and ecological motivations, and parallel the components of “total economic value” (Millennium Ecosystem Assessment 2005, National Research Council 2004:46). The not use contingent category is important for representing so-called nonuse values. For example, people may value the existence of a species or natural area without the necessity of actually seeing or otherwise using those resources (e.g., Krutilla 1976, Freeman 2003, Johnston et al. 2003). Codes are applied at most once per paragraph, although we expected that multiple different codes would occur within a given paragraph, with repeated codes possible for different paragraphs within a single article.
An important distinction between various content analysis methodologies is the use of human coders vs. automated approaches. Each method carries trade-offs, with efficacy of computer aids varying by project (Krippendorff 2013). A general limitation of automated coding is that it requires the researcher to anticipate how concepts will be presented in texts, whereas human coding does not require such omniscience. Some research questions, such as occurrences of given keywords, are easily automated. Specialized software is also available for more detailed questions, for example Bengston et al. (2004) studied the longitudinal frequency of three forest value orientations in national newspapers using automated paragraph scoring based on preprogrammed word combinations. Our content analysis study is unusual in that the coding schema is complex. The codebook was designed and vetted to include a mechanism to code essentially every measurable feature of rivers and streams as well as every encountered motivation. Furthermore, each code can be expressed multiple ways. For example, concern over “water contact health risk” could be stated in a variety of word combinations. Accordingly, we ascertained that the effort needed to tailor and test an algorithm specific to our study, although theoretically possible, would not be feasible within our budget and timeline. This limited us to a smaller sample size than typical for an automated study, but with a trade-off of richly coded data.
Two student coders were used, allowing meaning to emerge from texts without subconscious coding biases of the principal investigators (Neuendorf 2002). A pilot phase was a crucial element of the study, particularly for coders to practice abstracting from their own reactions and instead cueing to author intent. Several refinements to the provisional codebook occurred during the pilot phase, mainly adjusting coding rules to encourage reliable coding, but also the consolidation of some codes and the emergence of a few new themes. The final codebook contains 54 codes total, including NOCODE, denoting no attribute or motivation codes applied to a paragraph.
Initially, we hoped each coder would code separate texts to increase the sample of analyzed articles, but during the pilot stage, intercoder reliability (Fleiss’ Kappa; Fleiss 1971, Krippendorff 2013) for many codes indicated a need for additional consistency checks. Although some of our codes were relatively unambiguous (e.g., BRD for any reference to birds), several motivation codes required interpretation of nuance. For example, our code “supposed to be” (SUPP) necessitated that a paragraph contains a judgment regarding a preferred “natural” state. There also arose passages that were ambiguously written even for coding rules that seemed deterministic. We could have dropped the codes that tended not to meet minimum the intercoder reliability statistics, yet loss of data richness would sacrifice our study objectives of providing a comprehensive view of recurring attributes and motivations manifest in texts. We thus developed a redundancy coding and resolution strategy to maximize the quality of information extracted. In team-based qualitative research, collaborators often discuss disagreements until consensus is reached (e.g., Gerbert et al. 1999, Mackey et al. 2004). However, redundant coding and reconciliation procedures appear to be rare in content analysis. In the first step of our redundant coding design, coders independently coded content for a group of articles and then relayed their work to a principal investigator. The principal investigator then merged the two files for agreement analysis, a process facilitated by the “Coding Analysis Toolkit” (Shulman 2017). A list of coding disagreements was thereby generated and sent back to coders for their independent review because we wanted our coders to maintain autonomy throughout the reconciliation process. Coder revisions were then merged and again checked for agreement. In instances in which coders still did not agree, the principal investigators made a decision, thereby generating a final coded article. Paragraphs without attribute or motivation codes were labeled NOCODE retroactively to allow the total number of paragraphs to be represented in the final analysis. Coding was conducted using ATLAS.ti software (Scientific Software Development 2015).
Data generated by coding were reformatted and reduced to conduct analyses inaccessible to a solely qualitative approach (Riffe et al. 2014). Code counts per article were normalized by the number of paragraphs to control for article length. The year of each article was also coded, as well as the season (spring: March to May; summer, June to Aug.; fall, Sept. to Nov.; winter, Dec. to Feb.). To document frequencies of attributes and motivations manifested in source texts, we utilized simple descriptive statistics expressed in bar charts and box plots. To investigate the influence of article source, publication year, and publication season on normalized code frequencies, we conducted a series of nonparametric multivariate analyses of variance (MANOVA) tests (Anderson 2001). As an added data visualization aid for interpreting MANOVA results, we used nonmetric multidimensional scaling (NMS) plots (Andsager and Powers 1999, McCune and Grace 2002). Separate two-dimensional NMS plots were planned for attribute and motivation code frequencies, with each article providing a datapoint on each plot. Plot axes are correlated with codes that account for the greatest combined difference between articles in terms of code frequencies. This technique is especially useful for visualizing clusters of similar datapoints and has been used in ecology to discern similar species distributions for various field sites (Clarke 1993). There may be as many vectors as there are variables, with vector length being a measure of that code’s importance in accounting for differences between datapoints.
Tests of association between attributes and motivations were conducted at the paragraph level to make use of finer data resolution. To support these tests, presence/absence for each code for each paragraph was obtained via ATLAS.ti’s export feature. Both Pearson chi-squared and Spearman rank correlation tests were conducted. Anticipating insufficient “n” for numerous cells in a full crosstabulation matrix, tests of association were planned for codes grouped into categories. In formulating hypotheses for how attributes and motivations might be linked, we recategorized attributes as follows: basic needs or harm, recreation, and remaining codes. Testing against the null of independence, our alternate hypothesis was that these categories would be associated with consumptive use, nonconsumptive use, and not use contingent motivations, respectively. The recategorization proved challenging because some attributes reasonably span multiple categories, for example mammals (MAMM) might be in the context of direct use (hunting), passive use (wildlife viewing), or nonuse (preserving for the future). However this difficulty is precisely the reason we wanted to conduct association tests. Attributes associated with consumptive and nonconsumptive use are somewhat predictable, yet attributes not associated with direct or indirect use are more difficult to hypothesize. To complete the recategorization, we took a conservative approach and only placed codes clearly related to basic needs or harm, or recreation, in those two respective categories. The third category, remaining codes, contained everything that could not be reasonably ascribed solely to a use-type motivation. All statistics were conducted with R software (R Core Team 2014), with multivariate analyses performed using the vegan package for R (Oksanen et al. 2015).
A foundational result of the study is the codebook itself, extensively tested to reliably extract the variety of measurable attributes of rivers and streams as well as human motivations for interest in those attributes. The full codebook with coding rules is attached as Appendix 1, with code shorthands and brief descriptions shown in Table 1. For attribute codes, the original and alternative code groupings for statistical tests of association are shown in Table 2. Note that two motivation codes listed in the codebook, medicinal uses (MED) and rights of species (RIGHT) were never actually used in our sample. The total number of utilized codes is then 52.
In all, the 76 sampled texts contained 1273 paragraphs and 2979 code occurrences (with 214 of those being NOCODE). To provide examples of how codes applied to actual text, selections for each utilized code are listed in Appendix 2. To provide further context on the codes, Appendix 2 also briefly describes subthemes arising under each code. To document the sampled texts themselves, and additional background beyond code frequencies, Appendix 3 lists a synopsis alongside each article title.
For reporting summary statistics, the NYT and WSJ are pooled because MANOVA results (described next) showed no discernible difference between these sources. Two bar charts, Figures 1 and 2, show normalized code frequencies for attributes and motivations, respectively. Box plots in Figures 3 and 4 show normalized code frequencies aggregated by category. Overall, code occurrences are split 54.4% attributes, 38.4% motivations, and 7.2% NOCODE.
Nonparametric MANOVA tests for differences in normalized code frequencies across articles are summarized in Tables 3 and 4. Article source, year, and season were tested, as well as source-year and source-season interactions. Initially MANOVAS were run on all the data, with separate runs for attribute and motivation code frequencies. These tests showed only source to be a significant variable (p-value < 0.05). Thus, three follow-up pairwise tests were run to isolate which sources differed from each other. Although the blog differed from both the NYT and the WSJ, no difference was found between the latter two.
The MANOVA results indicate significant differences between the blog and the NYT/WSJ, but particular code frequencies driving the difference are not specified. The NMS plots in Figures 5 and 6 provide visualizations of complex differences between individual articles. The NMS vectors are filtered to show only combined correlations over 0.3, to isolate codes that account for a larger degree of difference.
The outcome of chi-squared and Spearman rank correlation tests for the three hypothesized relationships are shown in Table 5. All expected associations were significant, with p-values well below 0.01 and with the correlation between remaining attributes and not use contingent motivations being the strongest of the three. Given the uncertainty of particular attributes associated with nonuse motivations, a simple contingency table of co-occurring remaining attributes and not use contingent motivations was queried (table not shown). Out of 19 attributes represented in the table, just 3 accounted for more than one third of the 368 co-occurrences: fish (FISH), water quality other (QUALO), and wildlife other (WLO). Given this strong link between fish and wildlife codes and not use contingent motivations, we were then curious how the fish and wildlife and vegetation attribute category correlated with each of the three motivation categories. Correlations were rather low for consumptive use and nonconsumptive use at 0.05 and 0.02, respectively, whereas the correlation with not use contingent was 0.45.
Simple bar charts and box plots provided in Figures 1 through 4 tell much of the story regarding relative frequency of different codes and code categories, as well as comparisons across article sources. Within attributes, water codes are most common, with fish, wildlife, and vegetation codes being a close second. Within motivations, consumptive use codes are the most frequent. The box plots show the blog and NYT/WSJ having similar rankings at the highest code category level, however, contrasts are seen when inspecting the finer resolution available on the bar charts. The blog is enriched in water quantity attributes, fish and wildlife and vegetation attributes, and not use contingent motivations. The NYT/WSJ has more water quality attributes, human attributes, and nonconsumptive use motivations. There are also notable differences at the code level. For example, viewing Figure 1, the blog focuses on water supply scarcity (WSS), whereas the NYT/WSJ barely mentions it.
The NMS plots are useful for summarizing results at the article level. Figure 5, dedicated to attributes, indicates some clustering of blog articles toward the lower half of axis 2, which is associated with water supply scarcity (WSS). The NYT articles tended toward the opposite direction, more closely associated with water supply health risk (WSHR) and water quality other (QUALO: general reference to water quality). Furthermore, there are a few blog articles, in the lower right quadrant, associated with aquatic life other (AQO: nonfish aquatic animals) and endangered species (ENDG). Property damage from flooding (FD) is an arm of the data shared by all sources. The Figure 1 bar chart substantiates source differences for WSS, QUALO, AQO, and ENDG; however average WSHR code frequencies are similar. It appears that just one or two NYT articles with especially high code frequencies are driving the WSHR vector to appear as a factor describing differences between articles.
Turning to motivations, Figure 6 indicates that blog and NYT articles share interest in agriculture (AG), with a particularly strong signature in the blog. The NYT/WSJ shows clustering in the directions of industry (IND) and contact recreation (CREC). Overuse (GREED) also appears as an important vector, but with few articles plotted nearby. Cross-checking Figure 6 with Figure 2, differences between sources for AG and CREC are corroborated, with some difference in IND, although IND remains the most prevalent motivation for both sources. Average code frequencies for GREED are similar, indicating this to be a factor more for describing differences between articles than for between sources.
The above discussion draws on data to visualizations to highlight overall results and other easily visible insights. However, our analysis also benefits from a foundation of careful sampling, allowing statistical testing. The MANOVA tests find significant differences between the blog and combined NYT/WSJ sources. Tests of association also yielded statistically significant results in a manner that would be impossible through visual inspection. Quantitative analysis was successful in documenting consistent patterns in the data, regarding influence of text source, and even predictable correlations between codes at the paragraph level across sources. Quantitative evidence of these patterns is especially compelling given that article topics varied widely, i.e., everything from flooding on the Mississippi, to a controversial art project that would drape shade over a river in Colorado.
Our findings mark a new threshold in final ecosystem services identification. Preceding work has refocused ecosystem services concepts in the direction of measurable specifics, along with a call for more research on identifying those final services (or other similarly named term) at the interface between biophysical science and social systems (Boyd and Banzhaf 2007, Fisher et al. 2009, Johnston and Russell 2013). Some investigators have continued to concentrate on theoretic issues or other top-down structures, such as Boyd and Krupnick (2013), Landers and Nahlik (2013), and Boyd et al. (2016). Others have directed efforts toward bottom-up data collections using social science methods to document valued attributes in a reproducible manner. Furthermore, other empirical efforts have been ongoing for decades to isolate and then value specific environmental resources in case study work (Mitchell and Carson 1989, Arrow et al. 1993, Bateman et al. 2003, Weber et al. 2016).
A complete joining of top-down and bottom-up insights relevant for final ecosystem services is beyond our scope and would constitute an enormous effort indeed. However, we attempted a limited summary of how our research compares and connects with other studies attempting to identify the breadth of river final services. In that vein of research, Ringold et al. (2009) established a pattern by outlining a matrix of beneficiaries on one axis and categories of measurable river quantities on the other. The matrix was derived via an expert workshop of diverse attendees and served to broaden ideas of river monitoring beyond traditional indices. There is much overlap in their expert assessment and our content analysis results, however, there are also many differences. Most meaningfully, there was no attempt to assign relative importance to different river attributes, and workshop results were stated to be a hypothesis requiring follow-up study. Further development of attribute-by-beneficiary matrices continued with Ringold et al. (2013) and most ambitiously with Landers and Nahlik (2013). However, these continued to be limited to expert judgments regarding what attributes different groups of people valued.
Schiller et al. (2001) and later Weber and Ringold (2014, 2015) executed qualitative data collections to isolate river attributes as evident in focus group and semistructured interview transcripts. The latter 2 efforts involved more than 200 human subjects in 2 separate U.S. geographies, with code frequencies reported in results, allowing a measure of relative priority. Taken together, the studies provided the initial working codebook for our work (as noted in the introduction), and thus cannot be considered truly independent. That said, there is an important contrast in methods. Our study relied on secondary data, which despite many advantages, did not allow the ability to follow up for clarification as during a live meeting. Thus, our study is more susceptible to coding intermediate ecosystem services, i.e., measurable attributes not important in and of themselves, but instead indicators of other, more relevant attributes. For example, one of our codes was sewage (SEW). Although people are certainly likely to have an immediate distaste for sewage in waterways, there are a host of potentially related concerns that may or may not have been actually written out, such as safety of water contact, safety of water supplies, and health of aquatic life.
Despite differences in data sources and methods, code rankings are similar across our study and Weber and Ringold (2014, 2015). There tends to be a focus on charismatic flora and fauna, and attributes that could cause direct harm to humans, such as poor drinking water quality, or flooding risk. However, there is also consistent concern for topics such as native species, evidencing widespread environmentalist perspectives to some degree. All three studies were designed to identify river attributes important to so-called nonusers, a classic category in resource economics (Krutilla 1976) and ecosystem services (Chan et al. 2012). Weber and Ringold (2014, 2015) specifically screened for diverse participants, including those with minimal recreational contact with rivers. Their results show consolidation of interest around a limited number of final services across socio-demographics, with just a few notable differences (e.g., between urban vs. rural participants). Their results do not show evidence of particularities for nonuser participants, e.g., codes only relevant to nonusers, or nonuser lack of interest in codes other groups were interested in. We isolated a nonconsumptive use motivation code category and statistically tested correlation with a remaining attributes code category. As noted in the results, fish (FISH) and wildlife other (WLO) were prevalent in that association, leading to the subsequent discovery that fish and wildlife and vegetation attributes were most closely correlated with not use contingent motivations. This provides evidence that fish and wildlife are not solely associated with extractive or recreational motivations. Our findings challenge the common supposition that people value fish, wildlife, and vegetation primarily for direct or indirect uses. Instead, sampled texts more commonly associate not use contingent motivations with these attributes.
There is an important question regarding how well the particular texts we sampled represent public opinion or even their readership. Significant editorial bias from publishing entities is possible and some criticize news sources as being sensationalistic in general. A counterargument is that news media is an integrated component of popular culture (Altheide and Schneider 2013). Furthermore, periodicals may themselves influence public perceptions of environmental issues or management (Siemer et al. 2007, Muter et al. 2009, Lopera and Moreno 2014). Although we are only able to analyze a relatively small number of texts, we have endeavored to include archetypal representations of different points of view. Texts from additional perspectives would be interesting to examine, such as agricultural, industrial, or tribal sources. Monograph texts, such as books on rivers and streams by high-profile authors are another possibility. With regard to closer qualitative analysis of data already in hand, more of the complex context surrounding how different codes were presented could be drawn out. For example, FD has a similar frequency for the two sources, but the blog emphasized the idea of using wetlands and similar strategies to prevent FD, whereas the NYT/WSJ focused more on occurrences of FD. As noted earlier, Appendix 3 briefly notes such subthemes.
A related question is how well our results would continue to represent popular interests in the future, or even the present day, given that the articles analyzed are several years old. Undoubtedly public perceptions and interests evolve over time, and we must leave the question unanswered as to how different results would be if a similar content analysis were performed in 5, 10, or 50 years. However, we suspect that although specific newsworthy details vary, such as the specific threat to clean drinking water, the location of flooding, or the particular invasive species, most of the same final ecosystem services will continue to be discussed. We did take pains to investigate this question statistically, specifically testing for influence of year and season. We found no effect with either, but the test was limited by the data only spanning two years.
To shed light on attributes of waterways people care about and the motivations for their interest, we conducted a content analysis of both environmentalist and mainstream texts. Our codebook was rigorously tested via a pilot phase and is itself an outcome of the study that may be useful as a basis for future empirical studies of final ecosystem services. For attributes, the most common codes were regarding water quantity and quality, fish, wildlife, and vegetation. The most common motivation codes were about consumptive use. Somewhat surprisingly, these high-level results were similar across sources, although statistical testing indicated that text source significantly influenced code frequencies overall. Closer inspection of the data via both code-level bar charts and article-level nonmetric multidimensional scaling plots showed that the selected environmentalist source more strongly emphasized water supply scarcity, aquatic life besides fish, endangered species, and motivations not contingent on use. The two mainstream news sources more frequently discussed water quality generally and recreational motivations. Influence of publication date on code frequencies was also tested, but was found insignificant. At a paragraph level, expected co-occurrences between categories of attribute and motivation codes were strong. Some attribute codes were more associated with motivations not contingent on use, rather than direct use or recreational enjoyment. One notable example of this was fish, the most common wildlife code for both sources. These results lend support to the importance of nonuse oriented ecological values in both environmentalist and mainstream news sources. Despite differences between sources, overall most of the interest in rivers consolidates around relatively few codes, a finding that triangulates with previous focus group and semistructured interview research.
We believe our methodology would serve for other kinds of social-ecological systems research, and indeed, the main appeal of our paper to other researchers may lie in the methods rather than the results. In the spectrum between qualitative and quantitative approaches, we found ourselves in the middle, due to the detail we wished to glean from texts, and our goal of statistical testing. We found few examples of content analysis matching our level of detail, and by necessity forged our own way forward. We ultimately succeeded in drawing on the advantages of both qualitative and quantitative traditions, thanks to an interdisciplinary team, with much effort toward careful sample design, manual coding, and the benefit of specialized visualization and statistical methods. Even so, our codebook detail pushes the limit of manual content analysis, and we recommend fewer codes for future work if possible. Vast research possibilities exist with online data, and we hope our method opens doors for quantitative analysis of qualitative data, for investigators having the necessary human resources but limited research dollars.
The authors are grateful to Claire Cvitanovich and Brian Wilson for their careful coding and to Susan Yee for helpful comments on an earlier draft of this paper. This manuscript has been subjected to the U.S. Environmental Protection Agency review and has been approved for publication. The views expressed in this paper are those of the authors and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.
Alessi, M. G., C. A. Miller, and E. E. Harper. 2013. Content analysis of three waterfowl hunting DVDs. Human Dimensions of Wildlife 18.2:152-158. http://dx.doi.org/10.1080/10871209.2013.745103
Alexander, S. M., and M. S. Quinn. 2011. Coyote (Canis latrans) interactions with humans and pets reported in the Canadian print media (1995-2010). Human Dimensions of Wildlife 16.5:345-359.
Altheide, D. L., and C. J. Schneider. 2013. Field notes and other data: accounting for ourselves. Pages 125-132 in D. L. Altheide and C. J. Schneider, editors. Qualitative media analysis. Sage, Thousand Oaks California, USA.
Anderson, Marti J. 2001. A new method for non-parametric multivariate analysis of variance. Austral Ecology 26(1):32-46.
Andsager, J. L., and A. Powers. 1999. Social or economic concerns: how news and women’s magazines framed breast cancer in the 1990s. Journalism and Mass Communication Quarterly 76(3):531-550. http://dx.doi.org/10.1177/107769909907600309
Arrow, K., R. Solow, P. R. Portney, E. E. Leamer, R. Radner, and H. Schuman. 1993. Report of the NOAA panel on contingent valuation. Federal Register 58(10):4601-4614.
Bateman, I. J., R. T. Carson, B. Day, W. M. Hanemann, N. Hanley, T. Hett, M. Jones-Lee, G. Loomes, S. Mourato, E. Özdemiroglu, D. W. Pierce, R. Sugden, and J. Swanson. 2002. Economic valuation with stated preference surveys: a manual. Edward Elgar, Northampton, Massachusetts, USA.
Bengston, D. N., T. J. Webb, and D. P. Fan. 2004. Shifting forest value orientations in the United States, 1980-2001: a computer content analysis. Environmental Values 13:373-392. http://dx.doi.org/10.3197/096327104323312734
Bernard, H. R. 2011. Research methods in anthropology. AltaMira, Lanham, Maryland, USA.
Böck, K., S. Muhar, A. Muhar, and R. Polt. 2015. The ecosystem services concept: gaps between science and practice in river landscape management. GAIA 24(1):32-40. http://dx.doi.org/10.14512/gaia.24.1.8
Boyd J. W., and S. Banzhaf. 2007. What are ecosystem services? The need for standardized environmental accounting units. Ecological Economics 63:616-626. http://dx.doi.org/10.1016/j.ecolecon.2007.01.002
Boyd, J., and A. Krupnick. 2013. Using ecological production theory to define and select environmental commodities for nonmarket valuation. Agricultural and Resource Economics Review 42(1):1-32. http://dx.doi.org/10.1017/S1068280500007590
Boyd, J., P. Ringold, A. Krupnick, R. Johnston, M. Weber, and K. Hall. 2016. Ecosystem services indicators: improving the linkage between biophysical and economic analyses. International Review of Environmental and Resource Economics 8(3-4).
Braden, J. B., D. G. Brown, D. R. Maidment, and S. T. Marquart-Pyatt. 2015. Populating the water world: exploring data aspirations of water experts. Society and Natural Resources 28:439-451. http://dx.doi.org/10.1080/08941920.2014.945060
Brossard, D., J. Shanahan, and K. McComas. 2004. Are issue-cycles culturally constructed? A comparison of French and American coverage of global climate change. Mass Communication and Society 7(3):359-377. http://dx.doi.org/10.1207/s15327825mcs0703_6
Chan, K. M. A, T. Satterfield, and J. Goldstein. 2012. Rethinking ecosystem services to better address and navigate cultural values. Ecological Economics 74:8-18. http://dx.doi.org/10.1016/j.ecolecon.2011.11.011
Clarke, K. R. 1993. Non-parametric multivariate analyses of changes in community structure. Australian Journal of Ecology 18:117-143. http://dx.doi.org/10.1111/j.1442-9993.1993.tb00438.x
Creswell, J. W. 2014. Research design: qualitative, quantitative, and mixed methods approaches. Sage, Thousand Oaks, California, USA.
Daily, G. C. 1997. Nature’s services: societal dependence on natural ecosystems. Island, Washington, D.C., USA.
Fisher, B., R. K. Turner, and P. Morling. 2009. Defining and classifying ecosystem services for decision making. Ecological Economics 68(3):643-653. http://dx.doi.org/10.1016/j.ecolecon.2008.09.014
Fleiss, J. L. 1971. Measuring nominal scale agreement among many raters. Psychological Bulletin 76:378-382. http://dx.doi.org/10.1037/h0031619
Freeman, III, A. M. 2003. The measurement of environmental and resource values. Resources for the Future, Washington, D.C., USA.
Gerbert, B., N. Caspers, A. Bronstone, J. Moe, and P. Abercrombie. 1999. A qualitative analysis of how physicians with expertise in domestic violence approach the identification of victims. Annals of Internal Medicine 131(8):578-584. http://dx.doi.org/10.7326/0003-4819-131-8-199910190-00005
Gunter, V. J., and C. K. Harris. 1998. Noisy winter: the DDT controversy in the years before Silent Spring. Rural Sociology 63(2):179-198. http://dx.doi.org/10.1111/j.1549-0831.1998.tb00670.x
Gunter, V. J. 2005. News media and technological risks: the case of pesticides after Silent Spring. Sociological Quarterly 46(4):671-698. http://dx.doi.org/10.1111/j.1533-8525.2005.00031.x
Haines-Young, R., and M. Potschin. 2010. The links between biodiversity, ecosystem services and human well-being. Pages 110-139 in D. G. Raffaelli and C. L. J. Frid, editors. Ecosystem ecology: a new synthesis. Cambridge University Press, Cambridge, UK. [online] URL: https://www.nottingham.ac.uk/cem/pdf/Haines-Young&Potschin_2010.pdf
Houston, M. J., J. T. Bruskotter, and D. Fan. 2010. Attitudes toward wolves in the United States and Canada: a content analysis of the print news media, 1999-2008. Human Dimensions of Wildlife 15(5):389-403. http://dx.doi.org/10.1080/10871209.2010.507563
Janis, I. L. 1972. Victims of groupthink: a psychological study of foreign-policy decisions and fiascoes. Houghton Mifflin, Boston, Massachusetts, USA.
Johnston, R. J., E. Y. Besedin, and R. F. Wardwell. 2003. Modeling relationships between use and nonuse values for surface water quality: a meta-analysis. Water Resources Research 39(12). http://dx.doi.org/10.1029/2003WR002649
Johnston, R. J., and M. Russell. 2011. An operational structure for clarity in ecosystem service values. Ecological Economics 70(12):2243-2249. http://dx.doi.org/10.1016/j.ecolecon.2011.07.003
Keeler, B. L., S. Polasky, K. A. Brauman, K. A. Johnson, J. C. Finlay, A. O’Neill, K. Kovacs, and B. Dalzel. 2012. Linking water quality and well-being for improved assessment and valuation of ecosystem services. Proceedings of the National Academy of Sciences 109(45):18619-18624. http://dx.doi.org/10.1073/pnas.1215991109
Krippendorff, K. 2013. Content analysis: an introduction to its methodology. Third Edition. Sage, Thousand Oaks, California.
Krutilla, J. V. 1967. Conservation reconsidered. American Economic Review 57:777â€“786. [online] URL: http://www.rff.org/files/sharepoint/News/Features/Documents/071003%20Krutilla-ConservationReconsidered.pdf
Landers, D. H., and A. M. Nahlik. 2013. Final ecosystem goods and services classification system (FEGS-CS). United States Environmental Protection Agency, Washington, D.C., USA. [online] URL: https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=257922
Lopera, E., and C. Moreno. 2014. The uncertainties of climate change in Spanish daily newspapers: content analysis of press coverage from 2000 to 2010. Journal of Science Communication 1:1-18.
Mackey, R .A., M. A. Diemer, and B. A. O’Brien. 2004. Relational factors in understanding satisfaction in the lasting relationships of same-sex and heterosexual couples. Journal of Homosexuality 47(1):111-136. http://dx.doi.org/10.1300/J082v47n01_07
McCune, B., and J. B. Grace. 2002. Analysis of ecological communities. MjM Software Design, Gleneden Beach, Oregon, USA.
Millennium Ecosystem Assessment. 2005. Ecosystems and human well-being: synthesis. Island, Washington, D.C., USA. [online] URL: http://millenniumassessment.org/documents/document.356.aspx.pdf
Mitchell, R. C., and R. T. Carson. 1989. Using surveys to value public goods: the contingent valuation method. Resources for the Future, Washington, D.C., USA.
Muter, B. A., M. L. Gore, and S. J. Riley. 2009. From victim to perpetrator: evolution of risk frames related to human-cormorant conflict in the Great Lakes. Human Dimensions of Wildlife 14(5):366-379. http://dx.doi.org/10.1080/10871200903045210
National Research Council. 2004. Valuing ecosystem services: toward better environmental decision-making. National Academies Press, Washington, D.C., USA. http://dx.doi.org/10.17226/11139
Nature Conservancy. 2010. Communicating ecosystem services. Nature Conservancy, Arlington, Virginia, USA. [online] URL: https://www.conservationgateway.org/ConservationPractices/EcosystemServices/CommunicatingEcosystemServices/Pages/communicating-ecosystem-s.aspx
Nelson, P., N. Krogman, L. Johnston, and C. C. St. Clair. 2014. Dead ducks and dirty oil: media representations and environmental solutions. Society and Natural Resources 28:345-359. http://dx.doi.org/10.1080/08941920.2014.948241
Neuendorf, K. A. 2002. The content analysis guidebook. Sage, Thousand Oaks, California, USA.
Norton, R. K. 2008. Using content analysis to evaluate local master plans and zoning codes. Land Use Policy 25(3):432-454. http://dx.doi.org/10.1016/j.landusepol.2007.10.006
Oksanen, J., F. G. Blanchet, R. Kindt, P. Legendre, P. R. Minchin, R. B. O’ Hara, G. L. Simpson, P. Solymos, M. H. H. Stevens, and H. Wagner. 2015. Vegan: community ecology package. R package version 2.2-1. R Foundation for Statistical Computing, Vienna, Austria. [online] URL: http://CRAN.R-project.org/package=vegan
Proctor, J. D. 1998. Environmental values and popular conflict over environmental management: a comparative analysis of public comments on the Clinton Forest Plan. Environmental Management 22(3):347-358. http://dx.doi.org/10.1007/s002679900110
ProQuest. 2017. U.S. major dailies. ProQuest, Ann Arbor, Michigan, USA. [online] URL: http://www.proquest.com/products-services/US-Major-Dailies.html
R Core Team. 2014. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. [online] URL: http://www.R-project.org/
Riffe, D., S. Lacy, and F. G. Fico. Analyzing media messages: using quantitative content analysis in research. Routledge, Abingdon, UK.
Ringold P. L., J. W. Boyd, D. Landers, and M. A. Weber. 2009. Report from the workshop on indicators of final ecosystem services for streams. EPA/600/R-09/137. U.S. Environmental Protection Agency, Washington, D.C., USA.
Ringold P. L, J. W. Boyd, D. H. Landers, and M. A. Weber. 2013. What data should we collect? A framework for identifying indicators of ecosystem contributions to human well-being. Frontiers in Ecology and the Environment 11(2):98-105. http://dx.doi.org/10.1890/110156
Schiller, A., C. T. Hunsaker, M. A. Kane, A. K. Wolfe, V. H. Dale, G. W. Suter, C. S. Russell, G. Pion, N. H. Jensen, and V. C. Konar. 2001. Communicating ecological indicators to decision makers and the public. Conservation Ecology 5(1):19. http://dx.doi.org/10.5751/ES-00247-050119
Scientific Software Development. 2015. ATLAS.ti. Version 7.5.6. Scientific Software Development, Berlin, Germany.
Shulman, S. W. 2017. Coding analysis toolkit. Texifter. [online] URL: http://cat.texifter.com/
Siemer, W. F., D. J. Decker, and J. Shanahan. 2007. Media frames for black bear management stories during issue emergence in New York. Human Dimensions of Wildlife 12(2):89-100. http://dx.doi.org/10.1080/10871200701195415
Tilt, B., and Q. Xiao. 2010. Media coverage of environmental pollution in the People’s Republic of China: responsibility, cover-up and state control. Media, Culture and Society 32(2):225-245. http://dx.doi.org/10.1177/0163443709355608
Wainger, L., and M. Mazzotta. 2011. Realizing the potential of ecosystem services: a framework for relating ecological changes to economic benefits. Environmental Management 48(4):710-733. http://dx.doi.org/10.1007/s00267-011-9726-0
Weber, M. A., T. Meixner, and J. C. Stromberg. 2016. Valuing instream-related services of wastewater. Ecosystem Services 21:59-71. http://dx.doi.org/10.1016/j.ecoser.2016.07.016
Weber, M. A., and P. L. Ringold. 2015. Priority river metrics for residents of an urbanized arid watershed. Landscape and Urban Planning 133:37-52. http://dx.doi.org/10.1016/j.landurbplan.2014.09.006