SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

This article offers preliminary insights and a possible empirical model for managing the conceptual, methodological, and technological challenges entailed in developmental analysis of link-mediated relations. We offer a “mid-range” approach to making sense of linking practices, midway between close rhetorical/ethnographic analysis of links and large-scale link mapping. We suggest that systematic human coding and interpretation of linked-to producer types affords a more concrete and specific basis for hypothesizing about linking strategies than machine mapping, while providing a more robust attempt to generalize across the universe of candidate Web sites than ethnographic analysis. To illustrate this two-pronged approach to link analysis, we examine the linking practices exhibited on Web sites produced by U.S. Congressional candidates during the 2002 campaign season, focusing on the extent and development of links from candidate Web sites to other types of political Web sites during the three months prior to the November, 2002 election.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

We view the link as the essence of the Web. Links represent relationships between producers of Web materials, and they can be viewed as structuring connections between sites for Web users. Through links, Web producers transform independently produced sites into contiguous elements of a common Web sphere. The notion of a Web sphere, bounded temporally and by a shared object-orientation, offers a unit of analysis that enables examination of both the structure and substance of hyperlink networks.

This essay offers preliminary insights and a possible empirical model for managing the conceptual, methodological, and technological challenges entailed in developmental analysis of link-mediated relations. We propose a “mid-range” approach to making sense of linking practices, midway between close rhetorical/ethnographic analysis of links and large-scale link mapping. We suggest that systematic human coding and interpretation of linked-to producer types affords a more concrete and specific basis for hypothesizing about linking strategies than machine mapping, while providing a more robust attempt to generalize across the universe of candidate Web sites than ethnographic analysis.

We propose two complementary methods and illustrate the benefits of combining them for analysis purposes. The first method involves aggregating data from three independent cross-sectional samples of candidate Web sites coded at different points during the three month campaign period. By aggregating these data, a larger and more representative sample is produced that allows fuller characterization of overall linking strategies by different types of candidates. The second method entails analyzing a panel of candidate Web sites three times during the campaign period in order to assess any changes in linking practices over time. We suggest that the combination of these methods illuminates both the linking practices employed by the candidates, and the shifts in these practices during the campaign period.

To illustrate this two-pronged approach to link analysis, we examine the linking practices of a set of comparable Web producers during a specific time period, in an attempt to understand the nature and shape of the Web sphere that was collaboratively produced, and the factors that were associated with these practices. More specifically, we examine the linking practices exhibited on Web sites produced by U.S. Congressional candidates during the 2002 campaign season. We focus on the extent and development of links from candidate Web sites to other types of political Web sites during the three months prior to the November, 2002 election.

This study is part of a larger project analyzing online campaign practices in the U.S. 2002 elections, funded by the Pew Charitable Trusts. The approach to link analysis described in this article is enabled by the Election 2002 Web Archive, commissioned and made available on the Web by the Library of Congress Minerva Project in collaboration with http://WebArchivist.org and the Internet Archive, with additional funding from the Pew Charitable Trusts supporting a research-based interface to the archive at http://PoliticalWeb.info. We view archives of Web materials as a critical foundation for social research on Web-based phenomena, particularly for retrospective and/or developmental analyses, and suggest that ideally Web archives should be made publicly accessible for the purposes of verification and cumulative knowledge production.

Approaches to Hyperlink Analysis

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

In tracing developments in Web studies since the mid-1990s, we observe that strategic practices of hyperlinking, and their role in the co-production of a Web sphere, remain understudied. An examination of the origin of the hyperlink concept, and the literature analyzing the purpose and meaning of hyperlinks, illustrates the importance of this concept to a full-scale Web sphere analysis. The term “hypertext” was coined by Ted Nelson in the 1960s, in an Association of Computing Machinery conference presentation where he described his work on developing early text processing software (Keep, McLaughlin, & Parker, 1993). In his book Dream Machines (1978), Nelson describes the interconnectedness of ideas in types of links. In his typology, referential links displayed the destination on the originating page. Note links display linked destination in a separate screen (similar to a pop-up window) where non-linked information can be read. Keep, McLaughlin and Parker (1993) compare this kind of link to footnotes or endnotes in a paper document. Expansion links supply the destination information in expandable outline form at the source. Finally, command links run some external function, such as starting another software program. Nelson's typology of ideas interconnected by links was foundational to Tim Berners-Lee's (2000) vision for the World Wide Web. December (1996) suggests that in the early days of investigation of Internet communication, there was a lack of theoretical integration; he proposes some common units of analysis to guide future investigations, as the lack of commonalities, he claims, can lead to poor definitions, and an inability to make comparisons. December provides a detailed list of potential units of analysis, and his proposal for consistency in the conceptualization of units of analysis for Internet studies is still relevant. However, the definition of a hyperlink, what comprises its source and destination nodes, and its meaning are still ambiguous in many studies of the Web, perhaps because researchers approach hyperlink analysis from a variety of disciplines.

Paccagnella (1997) notes the challenge to sociological research posed by online social spaces. He mentions the ease of data collection, but also the need for longitudinal analysis. Like December (1996), Paccagnella notes the wide-ranging approaches to Web studies, but rather locates this as an asset to the field in that it allows for a wide range of perspectives. He claims that the Web has an interpretive flexibility, in the sense that phenomena on the Web are often short-lived, which he sees as a bolster for social constructivism. He suggests a methodological solution for investigating on-line communities that leads to an intimate understanding of a single case and incorporates a comparative and longitudinal component. Paccagnella recognizes a certain divergence from anthropology when conducting ethnography in virtual spaces, but offers solutions, such as comparative and longitudinal techniques, to make the connection to anthropological methods of studying community and artifacts stronger.

Mitra's (1999) study of intertextuality online characterizes hypertext as inherently dialogical, i.e. site producers engage in a form of dialogue by linking between texts. Mitra (1999) argues that sense-making communities develop around interlinked texts, providing interpretive lenses. The connections between these discourses, that is, the hyperlinks created between texts, provide meaning for a community. Similarly, Christine Hine (2000) examines the methodological issues of an ethnographic approach to Web studies. Her work attends to the context and situatedness of Web sites. She questions the aims, and strategies, and the resulting identity-construction processes of Web site producers as they are mediated through hyperlinks. Hine (2000) reflects upon an ethnographic approach to Web studies as a way to ask questions about the potential individual uses to which the technology can be put. Her case study of the American court case of a British nanny, Louise Woodward, and its life on the Web, illustrates the connections made between sites, but also the connections made between on-line and off-line locations associated with the case.

Following the suggestions of Pacagnella, Howard (2002) proposes “network ethnography” as an approach which blends ethnographic methods of participant observation with social network analysis to add a rich qualitative analysis of human interaction while also tracing the interaction broadly over a host of new media technologies. Howard claims that methodological advances have not matched technological advances, resulting in studies that employ a kind of virtual ethnography, but lack justification of methods. The network ethnography approach allows a way to enter into a community while providing a context for observations within the complex linked system of technology that supported the community (Howard, 2002 p.522).

Garton, Haythornthwaite, and Wellman (1997) focus on online social networks based on links. In particular, they identify the relationship between two linked points as the unit of social network analysis. In contrast with studies that investigate the structure within which social relationships can be forged, these authors focus on “exchanges that create and sustain work and social relationships” (Garton et al., 1997, par.9).

Jackson (1997) points out that some researchers tend to view the Web as a repository for information that users are constantly negotiating. Instead, Jackson focuses on the associations created within a collection of information. Expanding on this notion of relationship negotiation through links, Burbules and Callister (2000) argue that hyperlinks are more than a way of organizing information; rather, they structure associative relations. Based on concepts from classic rhetorical studies, they propose a typology of over a dozen forms of associative relations that can be constructed through hyperlinks.

Although the method of hyperlink mapping that Rogers and Marres (2000, 2002) propose as a way to track and analyze issue debates on the Web differs significantly from rhetorical analysis, their argument is similar to Burbules and Callister's. “Debate-scaping” is based on the assumption that “hyperlinking behavior is non-random” (Rogers & Marres, 2000). Rather, Rogers and Marres (2000) contend that hyperlinks evidence socio-epistemic networks in which actors on the Web acknowledge other actors by linking to their sites. Adamic and Adar (2001) also consider links as more than associations between bits of information; rather, links convey some kind of social relationship between the producers of the links' nodes, although not necessarily a personal relationship. Through these link structures that are created between producers of sites, Adamic and Adar (2001) claim that the relationships between producers can be inferred from hyperlinks in addition to associations between information and activities presented in the linked texts.

Park, Barnett, and Nam (2002) examine the structure of a hyperlinked network to assess the credibility of Web sites that constitute the network. Park (in press) suggests that social patterns may be found and analyzed on the Web through the hyperlinks between Web sites. He proposes a method of identifying and characterizing emergent structures of links which he calls hyperlink network analysis, which he and colleagues develop further in a subsequent article (Park, Kim, & Barnett, in press). Park's approach builds on Jackson's (1997) suggestion to apply social network analysis to hyperlinked Web sites. Park notes that in prior research, scholars have focused on the individual creating connections, and observes a recent shift to consider the linked Web site as a type of actor. Like Rogers and Marres (Rogers & Marres, 2000), Park argues that hyperlinks are inscriptions of communicative and strategic choices on the part of site producers. Proponents of this view suggest that hyperlink networks can serve a communicative social function for the producer and users.

We concur with many of the studies cited above which emphasize that hyperlinks are not merely connectives between texts, but rather mediators of a wide range of associative relations between producers of Web materials. However, the development of such linking practices over time has been largely unaddressed. Although methods for establishing the type of linked-to sites in the large-scale link studies cited above are not fully explicated, they appear to rely primarily on inference from the domain name, e.g.org versus .gov or .com. We consider domain names to be a blunt and often unreliable indicator of the type of entity that has produced a site, and propose employing systematic human interpretation of several types of information in establishing the site producer type.

Web Sphere Analysis

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

We have previously proposed the notion of a Web sphere as a useful unit of analysis (Foot & Schneider, 2002). We conceptualize a Web sphere as not simply a collection of Web sites, but as a hyperlinked set of dynamically defined digital resources spanning multiple Web sites deemed relevant or related to a central theme or “object.” The boundaries of a Web sphere are delimited by a shared object orientation and a temporal framework. Web sphere analysis is an analytic strategy that includes relations between producers and users of Web materials, as potentiated and mediated by the structural and feature elements of Web sites, hypertexts, and the links between them.

The most crucial element in this definition of a Web sphere is the dynamic nature of the sites to be included. This dynamism comes from two sources. First, the researchers involved in identifying the boundaries of the sphere are likely to continuously find new sites to be included within it. Second, the notion of defining a Web sphere is recursive in that pages that are referenced by other included sites as well as pages that reference included sites are considered as part of the sphere under evaluation. Thus, as a Web sphere is archived and analyzed over time, its boundaries are dynamically shaped by both the researchers' identification strategies and changes in the sites themselves.

As we describe in more detail below, to establish the Web sphere to be analyzed in this study, we focused in an ongoing process of identifying all of the sites produced by candidates running for office in the Senate or House over the course of four months leading up to the elections. Our focus on candidate Web sites reflects the nature of the candidate-centered political culture in the United States, in which parties and civic and advocacy organizations play a less critical role in citizen mobilization and information production than in other democracies. For our broader study of the political Web sphere created in relation to the 2002 election, we identified hundreds of sites produced by five other types of political actors. These other site producer types were government bodies, political parties, civic/advocacy groups, news organizations, and individual citizens.

In this article we present a two-pronged approach to analyzing practices of outlinking, i.e. patterns of links created on one type of political Web site to other types of political sites, based on the Web sphere concept. This approach facilitates knowledge about the paths and connections afforded to Web users by site producers through the production of links. The approach employs structured observations of a substantial number of cases, however, it differs from the machine-based methods of link-mapping and hyperlink network analysis described above in two ways. First, our method of coding outlinks requires human interpretation of the linked-to site producer type, and generates a set of Web pages on which outlinks have been identified which can then be used for various forms of additional qualitative analysis. Second, this approach entails a developmental lens that enables the analysis of shifts in outlinking strategies over time.

Political Structure and Web Practices

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

Our analysis of the linking practices of candidates in the 2002 US elections is informed by the relatively brief history of politics on the Internet. As outlined in numerous sources (Abbate, 1999; Castells, 1996) the history of the Internet begins with the advent of the ARPANET in the United States in the mid-1960s. The first overtly political uses of the Internet are usually traced back to Usenet, which was first introduced in 1979. By 1986, communities with explicit political agendas had adopted email and bulletin board systems and were using these Internet applications intensively. The overarching consensus of both users and contemporaneous scholars was that computer networking technology had the potential to dramatically alter the nature and shape of political discourse, and of democracy itself, by engaging and energizing new participants in the political process (Abramson, Arterton, & Orren, 1988; Downing, 1989; Dulio, Goff, & Thurber, 1999; Garramone, Harris, & Anderson, 1986; Glass, 1996; Hacker, Howl, Scott, & Steiner, 1996; Margolis & Resnick, 2000; Meadow, 1986; Mickunas & Pilotta, 1998; Myers, 1993; O'Sullivan, 1995; Schneider, 1996). Although computer-mediated political discourse, as it has been studied in the literature cited above, is beyond the scope of this essay, the linking practices of campaigns may be viewed as forms of implicit, networked political discourse.

By the 1996 and 1998 campaigns in the U.S., Internet-based politics had moved from the arcane and utopian (Diamond, McKay, & Silverman, 1993; Hacker et al., 1996; Myers, 1993) closer to the mainstream (Bimber, 1998; D'Alessio, 1997; Dulio et al., 1999; Kaid & Bystrom, 1998; Kamarck, 1999).This movement is suggested by the emergence on the Web of mainstream political candidates and party organizations. For example, in 1996, both major party candidates (Clinton and Dole) had general election Web sites, as well as presidential candidates from minor parties, including Perot and Nader. In addition, several candidates in the presidential primaries, including Buchanan and Forbes, maintained Web sites during the first phase of the campaign as well. At least one-third of Senate candidates maintained Web sites during the 1996 campaign (Kamarck, 1999), a figure which increased to more than two-thirds of Senate and open-seat House candidates in 1998 (Dulio et al., 1999). In the 2000 campaign, we estimate that 72% of Senate candidates and 53% of House candidates maintained campaign sites - figures that increase to 76% and 68%, respectively, when considering challengers and candidates in open seats (Stromer-Galley, Foot, Schneider, & Larsen, 2001). This finding is consistent with Puopolo (2001), who found that 88% of major party Senate candidates maintained Web sites. Additionally, a wide range of political actors, including for-profit political portals (Schneider, 2000), civic and advocacy groups, political parties and press organizations established Web presences with political content that formed an interlinked complex of sites comprising the political Web sphere. These sites provided online structures that facilitated various forms of political action (Schneider & Foot, 2002) in which citizens and other political actors could engage.

Some analysts have suggested that as the Internet developed between 1996 and 2000, emerging as an important component of many social, economic and political sectors, the impact of the Web on politics and democratic practice would be less “revolutionary” than first predicted (and hoped for) by the optimists and utopianists (Margolis & Resnick, 2000; Norris, 2001). This argument, most clearly articulated by the title of Margolis and Resnick's 2000 book, Politics as Usual, suggests that politics and political practice on the Internet will closely resemble politics offline, and that traditional factors affecting the distribution of political resources will affect the way that political actors use the Web. This ‘normalization’ hypothesis (Margolis & Resnick, 2000) was reinforced by the failure of commercial models of Web-based political information distribution in the United States during the 2000 election campaign. The normalization hypothesis would suggest that factors such as incumbency, type of political party, level of race competitiveness and office sought would be useful in explaining distinct types of practices and strategies on candidate Web sites, as these measures are indicators of the distribution of political resources (Goldenberg & Traugott, 1984; Kingdon, 1966; Westlye, 1991). To the extent that these factors endogenous to the political structure provide explanatory power with respect to any Web practices, the contention that political factors determine the types of online structure provided by political actors is bolstered and supported. The analytical approach to campaign linking strategies proposed in this article enables assessment of factors both endogenous and exogenous to political structure.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

Since candidates and their campaign organizations, rather than parties or civic or advocacy organizations, are the primary actors in the U.S. electoral process, we were particularly interested in the linking strategies employed on their Web sites. The aim of this study was to analyze outlinking strategies on Web sites produced by candidates seeking election on November 7, 2002 to one of 435 seats in the United States House of Representatives or one of 34 seats in the United States Senate. Using publicly available sources, including Web sites produced by state governments, political portals (e.g. politics1.com) and press organizations, we identified all candidates for these races, including primary candidates from those states with primary elections after August 2, 2002. Using the same sources, as well as additional searching on Web search engines such as Google, we identified as many candidate Web sites as we could. We defined a candidate Web site as one that explicitly advances a named individual for a specific office, and is not identified as produced by or sponsored by anyone other than that candidate or his or her official campaign organization. This definition excluded Web sites or Web pages within sites produced by political parties, and business or personal Web sites produced by candidates but not explicitly advancing the named individual for a specific office. As primary elections and other events unfolded during the campaign season, we adjusted our universe of candidates to include only those who were going to appear on an election ballot on or prior to November 7, 2002. This analysis was recalibrated each week during the study period. We consistently found that about 63% of all House and about 73% of all Senate candidates during the study period maintained Web sites that fit our criteria.

Given the lack of resources to examine all candidate Web sites, we faced several challenges in developing a method to sample candidate sites for analysis. We were interested in examining the development of candidate Web sites over time and in understanding the range of the phenomena across the breadth of candidate Web sites. Additionally, we wanted to ensure that our approach was sensitive to the fact that the universe of candidate sites was changing during the study period, as more Web sites were launched or identified during the campaign. These objectives led us to develop a two-pronged strategy for sampling candidate Web sites, combining panel or time-series methods with independent cross-sectional sampling approaches. In this paper, we report data drawn from the panel of sites to estimate changes in linking practices during the campaign, and report data drawn from the cross-sectional samples to estimate the presence of linking practices on candidate Web sites.

Candidate Web sites were examined during the 12-weeks between August and October, 2002 that coincided with the peak campaigning period in the United States. We selected a panel of sites from among those identified at the start of the study period. Sites in the panel were analyzed three separate times, in August, September and October, during the campaign season for the midterm federal elections. This approach allowed us to measure changes in candidate Web site linking practices during the course of the campaigns. Given the necessity of analyzing individual candidate Web sites multiple times, however, we were limited to analyzing a relatively small sample of sites in the panel, which restricted our ability to make robust comparisons among different types of candidates. In order to make assessments among different types of candidates, as well as to allow for the possibility of newly identified sites entering the universe from which the samples were drawn, we created independent monthly cross-sectional samples of sites and analyzed them in coordination with sites in the panel. Both the panel and cross-sectional samples were drawn to ensure robust analysis of candidates from races at all levels of competitiveness, as well as to facilitate analysis of Web sites produced by candidates running against each other. The sites examined were weighted to reflect the competitiveness levels across the universe of candidate sites; in the tables below, Ns are reported for unweighted cases, and descriptive statistics are reported for weighted cases.

Trained coders evaluated sites in the samples for the presence or absence of links to other types of political Web sites. Candidate sites were evaluated directly from the Web within one-week period for each month as they developed during the campaign season, and retrospectively from the Election 2002 Web Archive in order to confirm previous observations and examine additional cases. (This procedure was repeated during the other weeks of the month using different protocols with measures not presented in this paper). Coders were assigned to specific candidate Web sites randomly using specialized software developed specifically for this process by http://WebArchivist.org. The software provided opportunities for coders to retrieve operational instructions and coding protocols over the Web; coders could ask specific questions of coding supervisors via email or instant messaging. Coders were instructed to look carefully for links to other Web sites on the front page of the site, as well as second level pages of the site. Second level pages of the site were identified as those linked from the front page. Coders were instructed to examine third level pages of the site, defined as those linked from second level pages, only if the link to the page clearly suggested that the linked page would include Web resources or additional links. For example, if an “endorsement” page linked from the front page of the site included a link that referenced “Links to organizations endorsing candidate X,” coders were expected to visit that third-level page. Pages linked-to from the candidate site that were clearly identifiable as part of another Web site, either by the URL or by content appearing on the page, were not coded for additional links.

Our measures of links sought to identify the types of sites to which candidates were providing outlinks, rather than a count of the total number of outlinks provided. For example, the coding template included items such as “Does this candidate site link to a press site?” Coders were instructed to click on outlinks on the candidate site to determine the type of actor producing the linked-to site. This determination was made by an interpretive process after examining the outlinked page, the front page of the outlinked site, and an “about” page on the outlinked site for information about the site producer. This information was evaluated by the coder in view of a set of definitions of various types of political actors. Once the type of the actor producing the linked-to site was determined, a code of “present” for an outlink to that type of political site, along with the URL of the candidate site page on which the outlink was found, were recorded on the template, and the coder moved on to look for an outlink to another type of political site. (Recording the URL provides the opportunity for future, retrospective analyses of many kinds based on the archived materials.) Thus, candidate Web sites linking to one, say, press organization were not differentiated from candidate Web sites linking to multiple press organizations. If no outlinks to a particular type of political site were found, a code of “absent” was recorded for that type of outlink on that candidate site.

In this paper, we are reporting on measures that assessed the presence of links to four types of political actors: civic and advocacy groups, political parties, government agencies, and press organizations. Civic and advocacy groups were defined as national and state level non-profit organizations with a civic, voting or election emphasis (e.g., League of Women Voters), national and state level cause-oriented groups that may have a legislative/policy agenda (e.g. NARAL, Sierra Club, AFL-CIO), and ideological groups that represent themselves as ‘conservative,’‘liberal,’‘progressive,’ etc. Political party sites included national, state and local chapters of political parties. Press organizations were defined as any organization that produced news content and delivered it online; excluded from the definition were portal sites that delivered news produced by others. Government sites were defined as sites produced by U.S., State or local government entities, including those produced by individual officeholders.

Our coding was verified by having a sub-sample of sites coded by multiple coders to generate inter-coder reliability scores. We calculated overall percent agreement, as well as Krippendorff's alpha, for each link type. The overall agreement rate shows the overall percentage agreement of all coders across all sites. Krippendorff's alpha demonstrates the level of agreement between coders beyond that which can be ascribed to mere chance. All link types have a minimum alpha of .50 and a minimum overall agreement rate of 87% (see Table 1).

Table 1.  Inter-coder reliability scores.
Links to Site TypeAlpha% Agreement
Political Party sites.6492%
Government sites.5387%
Press sites.5387%
Civic/Advocacy Group sites.5087%

Data from these measures were combined with data about the candidates themselves-party type, incumbency status, level of competitiveness, and office sought-that we expected might illuminate our findings on linking practices, and on the strategic deployment of linking practices during the campaign. Party, incumbency and office sought were drawn from public sources such as press sites, political portals and government sites. Party type is classified as “major party” (Democratic or Republican), and “Third Party” (all other parties). Level of competitiveness was drawn from the National Journal, and based on the Cook's Political Report as reported on National Journal's Policy Central Web site. Candidates in races identified as “toss-up” are classified as “highly competitive.” Candidates in races identified as “leaning” or “likely” to be won by one or the other major party were classified as “medium” in competitiveness. Candidates in races identified as “strong” Republican or Democratic were classified as “low” in competitiveness.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

We first examine, using the combined cross-sectional samples, the extent to which candidate Web sites provided links to other political Web sites. We also consider the types of sites to which they linked, and the candidate characteristics associated with these strategies. This approach essentially reports a snapshot, taken over a three-month period leading up to the 2002 elections, of the portion of the political Web sphere consisting of the candidate Web sites.

In the discussion and tables that follow, we present percentages of Web sites produced by candidates with the characteristics discussed above (party type, office sought, incumbency status, and level of competitiveness) that exhibit specific linking practices. The linking practices we observed were (1) presence of outlinks to any of the four types of political Web sites discussed in the previous section (civic/advocacy group, political party, press, or governmental); (2) number of types of political Web sites for which outlinks were provided; and (3) presence of outlinks to specific types of political Web sites.

We show that three-quarters (76%) of candidate Web sites provided outlinks to any type of site. Of these campaign sites with outlinks, or outlinking sites, nearly all (84%) linked to at least one of the types of political sites examined: civic/advocacy groups, government bodies, press organizations or political parties (see Table 2). A logistic regression, using party type (major party or third party), office sought (House or Senate), level of competitiveness (low, medium or high), and incumbency status (incumbent or challenger) as regressors failed to predict the presence of outlinks to political Web sites among any of the sites examined. This indicates that these factors are of little value in suggesting what kinds of candidates are most likely to have Web sites that provide outlinks to other political Web sites.

Table 2.  Frequencies of outlinks on outlinking candidate Web sites.
Candidate Characteristics% sites with outlinksMean # of outlink typesStd. DevSites Examined*
  1. * indicates statistical significance at .05 or lower for independent variable in logistic regression

All Candidates84%1.671.18535
Party Type    
Major Party83%1.581.15413
Third Party91%2.001.21122
Office Sought    
House85%1.731.19458
Senate81%1.291.0177
Incumbency    
Incumbent84%1.661.17149
Challenger84%1.681.18386
Race Competitiveness    
Low83%1.721.22339
Medium88%1.611.11129
High87%1.551.0567

An analysis of the number of types of political Web sites to which candidate Web sites outlinked indicates that 25% of the sites examined provided outlinks to three or four types of sites, and 35% provided outlinks to one of the four types of political Web sites. The mean number of site types linked to by all candidates was 1.67. A multiple regression using the same factors as above as regressors explained only 4.4% of the variance in the number of types of outlinked sites, indicating that these characteristics of candidates are not associated with the number of types of political Web sites to which candidate sites provide outlinks.

An analysis of the presence of outlinks to particular types of political Web sites, and the characteristics of candidates providing links, is most illuminating of potential underlying patterns related to measures of political structure when examining links to civic/advocacy groups. We found that sites produced by government bodies were most likely to be linked to by outlinking campaign sites (51%), followed by political party (43%), civic/advocacy group (38%) and press (32%) sites (see Table 3). However, the four characteristics of candidates examined provide explanatory power only when examining the presence of links to civic/advocacy groups. A logistic regression using the four candidate characteristics correctly predicts the presence or absence of a link to civic/advocacy groups in 67% of our sampled sites. Our model was more successful in predicting the absence of a link (90% correct) than the presence of a link (32% correct). Incumbency was not a statistically significant factor in the equation. Third party candidates, those seeking House seats, and those in non-competitive races were more likely to provide links to civic/advocacy groups than major party candidates, those seeking Senate seats, or those in medium or highly competitive races. Logistic regressions using the same four regressors fail to predict the presence of a link to political parties, press organizations or governmental bodies. This indicates that these factors are of little explanatory value in suggesting what kinds of candidates are most likely to have Web sites that provide outlinks to these types of political Web sites.

Table 3.  Percentages of outlinking candidate sites with outlinks to specific site types.
Characteristics of CandidatesPercent of Candidate Sites with Outlink to Type of Site
 Civic/Advocacy GroupPolitical PartyPressGovernment
  1. * indicates statistical significance at .05 or lower for independent variable in logistic regression

All Candidates38%43%32%51%
Party Type*   
Major Party32%36%31%56%
Third Party62%69%31%37%
Office Sought*   
House42%45%32%53%
Senate21%39%26%44%
Incumbency    
Incumbent34%39%27%63%
Challenger41%46%33%47%
Race Competitiveness*   
Low43%47%29%51%
Medium33%36%35%55%
High25%42%36%51%

We next examined data collected using the panel study to ascertain the degree, if any, to which candidate Web sites evolved political Web site linking strategies during the campaign. As discussed above, the panel study assessed a single sample of candidate Web sites each month during the study period. This approach essentially reports a series of snapshots, taken during different points in the three-month period, showing a portion of the political Web sphere consisting of the candidate Web sites.

We calculated a change in outlinking pattern as a shift from presence to absence, or a shift from absence to presence, for each of the four types of outlinked political Web sites examined. As shown in Table 4, just over 10% of candidate Web sites had changes in linking practices during the course of the campaign. Very few sites added or deleted outlinks to either political party or government sites. Just over 4% of the candidate Web sites observed changed linking practices to civic/advocacy sites, and just over 8% of the sites changed practices related to press sites. Logistic regressions using the same four regressors as used above failed to predict the presence of a change in linking practices to civic/advocacy groups, political parties, press organizations, governmental bodies, or to any political Web sites. Again, this failure indicates that these factors are of little explanatory value in suggesting what kinds of candidates are most likely to change outlinking practices during the campaign.

Table 4.  Percentage of sites with changes in outlinking practices to specific site types by candidate and race characteristics.
 Civic/AdvocacyPolitical PartyPressGovernmentAnyN
All Candidates4.2%1.0%8.3%2.1%10.4%96
Party Type      
Major Party4.8%1.2%8.3%2.4%10.7%84
Third Party0.0%0.0%8.3%0.0%8.3%12
Office Sought      
House5.8%1.4%10.1%2.9%13%69
Senate0.0%0.0%3.7%0.0%3.7%27
Incumbency      
Incumbent3.6%0.0%7.1%0.0%7.1%28
Challenger4.4%1.5%8.8%3.0%11.7%68
Race Competitiveness      
Low0.0%0%0%0%0%22
Medium2.9%2.9%11.8%5.8%14.7%34
High7.5%0%10%0%12.5%40

Implications

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References

We draw three sets of implications from this analysis. First, we comment on our findings with respect to the nature of Web practices by electoral candidates, and the lack of explanatory power provided by factors endogenous to the political context of a campaign. Second, we discuss the value of our two-pronged approach to analyzing linking practices on a large set of Web sites over time. Finally, we identify some challenges associated with coding hyperlinks, and suggest some operational techniques that address these challenges. We concur with the observation by Rogers and Marres (2000) that “to link is to recognize…. Similarly, non-linking is a form of non-recognition, or, more radically, is an act of silencing through non-recognition” (pp. 16–17). As discussed above, we find that linking practices vary among candidate Web sites, and we contend that the presence/absence of links to other site types indicates recognition/non-recognition by campaigns of other political actors. While 84% of the sites included a link to at least one type of political Web site, the observation of variance in the number of types of political Web sites suggests the presence of different Web practices. Further evidence of this finding is provided by the presence of variance in the types of political Web sites to which outlinks were provided. Finally, our findings provide a hint of strategic linking practices with the observation of shifts during the campaign period in the types of outlinks provided.

Most importantly, our inability to explain the types of Web practice by factors endogenous to the political structure, such as party type, incumbency, level of race competitiveness and office sought, while potentially related to our relatively small sample sizes, could also suggest looking elsewhere for explanatory variables. It is possible that Web practices such as linking are still relatively novel for campaign site producers, and thus the lack of discernible patterns due to political-structural variables may be due to varying forms of experimentation. Furthermore, analyzing the number of outlinks on campaign sites, rather than the types of outlinks, may produce results that correspond more predictably with political-structural variables. However, while other measures associated with political resources, such as amount of campaign spending, candidate likelihood of winning, and prior political experience of a candidate, may be useful in explaining Web practices, we suggest that factors exogenous to the political structure should also be examined. These factors could include characteristics of the candidate or the candidate's organization, such as the presence in the campaign organization of a strong Internet advocate or the values and experience of the campaign's Internet team, or characteristics of the constituency or District, such as the level and type of Web access or the amount of Web development (e.g. the number of organizations with Web sites). Should future research indicate that factors associated with the Internet are related to specific Web practices, it would support the contention that Web practices are not shaped by traditional political factors, providing an alternative to the Politics as Usual (Margolis & Resnick, 2000) view of the Internet's impact on behavior of political actors.

Our conclusions concerning the nature of campaign linking practices are bolstered, we believe, by our methodological choices. Employing the Web sphere as a unit of analysis enables a useful site identification strategy. When resource constraints impose limitations on the number of observations that can be completed on sites within the Web sphere, analyses of the Web practices of a set of actors over a period of time will be stronger when a combination of cross-sectional and panel methods are used. Cross-sectional methods involving larger samples provide opportunities to make generalizations about subsets of cases, while panel studies support analyzes of change over time within specific cases. By combining methods, scholars will be able to examine more fully the nature of practices enacted by Web producers within a Web sphere during a specified period of time. In addition to drawing attention to the need for developmental studies of linking, the two-pronged approach we've proposed and modeled in this article fills a gap between large-scale link mapping and fine-grained ethnographic analyses of linking practices.

This research also identified several challenges related to measuring the presence of hyperlinks and interpreting the identity of site producers. Determining what constitutes a link is paramount. This study, for example, took a user's perspective in defining a link by including the criterion of functionality in its definition; after all, from a user's perspective, two sites are not linked by a broken link, despite the producer's intentions. Although a broken link may be repaired at any future point in time, there is no guarantee that it will be repaired, nor is there any way of determining whether the lack of functionality is accidental or intentional on the part of either the outlinking site producer or the linked-to site producer. Thus, the user's experience of the site's functionality at the moment of observation becomes a decisive criterion.

In cases where researchers are interested in identifying links to pages created by another Web producer, a clear operationalization strategy defining both the parameters of the actor's Web site and indicators of an outlinked site is necessary. In the present case, for example, our operational instructions accounted for the potential that the initial “site” could be delivered on multiple servers with distinct URLs, and instructed analysts to mark site boundaries subjectively based on information provided on “about us” and similar pages. Characterizing the outlinked site by type of producer requires clear definitions of potential actors with operational instructions to identify the relevant information. Additionally, researchers must specify precisely how links are to be identified, and provide procedures for navigating the object site to search for links. While some producers provide pages specified as “links” pages, others embed links within text and documents. Our solution to this issue was to limit the search for links to front pages, second-level pages linked from the front page, and third-level pages identified as “Links pages” linked from second-level pages.

Our experience in this research has deepened our belief that robust Web sphere analysis benefits from robust Web archives. Web archives that support scholarly research should be constructed with the research goals firmly in place. This suggests that scholars contemplating research work closely with librarians and archivists at the earliest stages of archive design and collection. If research questions are not considered in the development of collection policies, scholars are likely to find their analysis strategies frustrated. For example, broadening collection policies to include outlinked pages supports analyses of outlinks such as those presented here.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Approaches to Hyperlink Analysis
  5. Web Sphere Analysis
  6. Political Structure and Web Practices
  7. Methods
  8. Results
  9. Implications
  10. References
  • Abbate, J. (1999). Inventing the Internet. Cambridge , MA : MIT Press.
  • Abramson, J. B., Arterton, F. C., & Orren, G. R. (1988). The electronic commonwealth: The impact of new media technologies on democratic politics. New York : Basic Books, Inc.
  • Adamic, L. A., & Adar, E. (2001, May 15). You are what you link. Paper presented at the The 10th International World Wide Web Conference.
  • Berners-Lee, T. (2000). Weaving the web: The original design and ultimate destiny of the World Wide Web. New York : HarperCollins.
  • Bimber, B. (1998). The Internet and political mobilization: Research note on the 1996 election season. Social Science Computer Review, 16(4), 391401.
  • Burbules, N. C., & Callister, T. A. (2000). Watch IT: The risks and promises of information technologies for education. Boulder , CO : Westview Press.
  • Castells, M. (1996). The rise of the network society. Cambridge , MA : Blackwell.
  • D'Alessio, D. (1997). Use of the World Wide Web in the 1996 US election. Electoral Studies, 16(4), 489500.
  • December, J. (1996). Units of analysis for internet communication. Journal of Computer Mediated Communication, 1(4). Available: http://www.ascusc.org/jcmc/vol1/issue4/december.html.
  • Diamond, E., McKay, M., & Silverman, R. (1993). Pop goes politics: New media, interactive formats, and the 1992 campaign. American Behavioral Scientist, 37(2), 257261.
  • Downing, J. (1989). Computers for political change: Peacenet and public data access. Journal of Communication, 39(3), 154162.
  • Dulio, D. A., Goff, D. L., & Thurber, J. A. (1999). Untangled web: Internet use during the 1998 election. PS: Political Science and Politics, 32(1), 5358.
  • Foot, K. A., & Schneider, S. M. (2002). Online action in campaign 2000: An exploratory analysis of the U.S. political Web sphere. Journal of Broadcasting & Electronic Media, 46(2), 222244.
  • Garramone, G. M., Harris, A. C., & Anderson, R. (1986). Uses of political computer bulletin boards. Journal of Broadcasting and Electronic Media, 30(3), 325339.
  • Garton, L., Haythornthwaite, C., & Wellman, B. (1997). Studying online social networks. Journal of Computer Mediated Communication, 3(1). Available: http://www.ascusc.org/jcmc/vol3/issue1/garton.html.
  • Glass, A. J. (1996). On-line elections: The internet's impact on the political process. The Harvard International Journal of Press/Politics, 1(4), 140146.
  • Goldenberg, E. N., & Traugott, M. W. (1984). Campaigning for Congress. Washington D.C. : Congressional Quarterly Press.
  • Hacker, K. L., Howl, L., Scott, M., & Steiner, R. (1996). Uses of computer-mediated political communication in the 1992 presidential campaign: A content analysis of the Bush, Clinton and Perot computer lists. Communication Research Reports, 13(2), 138146.
  • Hine, C. (2000). Virtual ethnography. Thousand Oaks , CA : Sage.
  • Howard, P. (2002). Network ethnography and hypermedia organization: New organizations, new media, new myths. New Media & Society, 4(4), 550574.
  • Jackson, M. (1997). Assessing the structure of the communication on the World Wide Web. Journal of Computer-Mediated Communication, 3(1). Available: http://www.ascusc.org/jcmc/vol3/issue1/jackson.html.
  • Kaid, L. L., & Bystrom, D. G. (1998). The electronic election: Perspectives on the 1996 campaign communication. Mahwah , NJ : Lawrence Erlbaum.
  • Kamarck, E. C. (1999). Campaigning on the Internet in the elections of 1998. In Kamarck, E. C., & Nye, J. J. S. (Eds.), Democracy.com?: Governance in the network world (pp. 99123). Hollis , NH : Hollis Publishing.
  • Keep, C., McLaughlin, T., & Parker, R. (1993). The electronic labyrinth.
  • Kingdon, J. W. (1966). Candidates for office: Beliefs and strategies. New York : Random House.
  • Margolis, M., & Resnick, D. (2000). Politics as usual: The cyberspace revolution. Thousand Oaks , CA : Sage.
  • Meadow, R. G. (1986). The electronic machine: New technologies in political campaigns. Election Politics, 3(3), 2631.
  • Mickunas, A., & Pilotta, J. (Eds.). (1998). Technocracy vs. democracy: Issues in the politics of communication. Cresskill , NJ : Hampton Press.
  • Mitra, A. (1999). Characteristics of the WWW text: Tracing discursive strategies. Journal of Computer-Mediated Communication, 5(1). Available: http://www.ascusc.org/jcmc/vol5/issue1/mitra.html.
  • Myers, D. D. (1993). New technology and the 1992 Clinton presidential campaign. American Behavioral Scientist, 37(2), 181187.
  • Nelson, T. H. (1978). Dream machines: New freedoms through computer screens-a minority report. South Bend , IN : The Distributors.
  • Norris, P. (2001). Digital divide: Civic engagement, information poverty, and the Internet worldwide. New York : Cambridge University Press.
  • O'Sullivan, P. B. (1995). Computer networks and political participation: Santa Monica's teledemocracy project. Journal of Applied Communication Research, 23(2), 93107.
  • Paccagnella, L. (1997). Getting the seat of your pants dirty: Strategies for ethnographic research on virtual communities. Journal of Computer-Mediated Communication, 3(1). Available: http://www.ascusc.org/jcmc/vol3/issue1/paccagnella.html.
  • Park, H. W. (in press). What is hyperlink network analysis?: New method for the study of social structure on the Web. Connections.
  • Park, H. W., Barnett, G. A., & Nam, I.-Y. (2002). Hyperlink-affiliation network structure of top Web sites: Examining affiliates with hyperlink in Korea. Journal of the American Society for Information Science, 53(7), 592601.
  • Park, H. W., Kim, C.-S., & Barnett, G. (in press). Socio-communicational structure among political actors on the Web: A network analysis of Korean political parties' and assemblymen's Web sites in 2000–2001. New Media & Society.
  • Puopolo, S. T. (2001). The Web and U.S. senatorial campaigns 2000. American Behavioral Scientist, 44(12), 20302047.
  • Rogers, R., & Marres, N. (2000). Landscaping climate change: A mapping technique for understanding science & technology debates on the World Wide Web. Public Understanding of Science, 9(2), 141163.
  • Rogers, R., & Marres, N. (2002). French scandals on the Web, and on the streets: A small experiment in stretching the limits of reported reality. Asian Journal of Social Science, 30(2), 339353.
  • Schneider, S. M. (1996). Creating a democratic public sphere through political discussion: A case study of abortion conversation on the Internet. Social Science Computer Review, 14(4), 373393.
  • Schneider, S. M., & Foot, K. A. (2002). Online structure for political action: Exploring presidential Web sites from the 2000 American election. Javnost (The Public), 9(2), 4360.
  • Stromer-Galley, J., Foot, K. A., Schneider, S. M., & Larsen, E. (2001). What citizens want, where they went, and what they got online in the U.S. election 2000. In Coleman, S. (Ed.), Elections in the age of the Internet: Lessons from the United States (pp. 2635). London : Hansard Society.
  • Westlye, M. C. (1991). Senate elections and campaign intensity. Baltimore : Johns Hopkins University Press.