Community Networks: Where Offline Communities Meet Online


  • Andrea Kavanaugh,

    Corresponding author
    1. Senior Research Scientist and Associate Director of the Center for Human-Computer Interaction in the Department of Computer Science at Virginia Tech. She is interested in the use and social impacts of information technology and computer networking.
    • Address: Dept of Computer Science, 660 McBryde Hall (0106), Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: The Pennsylvania State University, 307H Information Sciences and Technology Building, University Park, PA 16802-6823 USA

      Address: 0330C Information Sciences & Technology Building, The Pennsylvania State University, State College, PA 16802 USA

      Address: 3160 Togeson Hall, Center for Human Computer Interaction, Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: Wheeling Jesuit University, Center for Educational Technology, 316 Washington Ave., Wheeling, WV, 26003-6243 USA

    Search for more papers by this author
  • John M. Carroll,

    Corresponding author
    1. The Edward M. Frymoyer Chair Professor of Information Sciences and Technology at the School of Information Sciences and Technology, Pennsylvania State University. His research focuses on methods and theory in human-computer interaction, particularly as they apply to networking tools and systems for collaborative learning and problem solving, and the development of community.
    • Address: Dept of Computer Science, 660 McBryde Hall (0106), Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: The Pennsylvania State University, 307H Information Sciences and Technology Building, University Park, PA 16802-6823 USA

      Address: 0330C Information Sciences & Technology Building, The Pennsylvania State University, State College, PA 16802 USA

      Address: 3160 Togeson Hall, Center for Human Computer Interaction, Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: Wheeling Jesuit University, Center for Educational Technology, 316 Washington Ave., Wheeling, WV, 26003-6243 USA

    Search for more papers by this author
  • Mary Beth Rosson,

    Corresponding author
    1. Professor of Information Sciences and Technology at the School of Information Sciences and Technology, Pennsylvania State University. Her research interests include scenario-based and participatory design methods, computer-supported collaborative work, and end user development.
    • Address: Dept of Computer Science, 660 McBryde Hall (0106), Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: The Pennsylvania State University, 307H Information Sciences and Technology Building, University Park, PA 16802-6823 USA

      Address: 0330C Information Sciences & Technology Building, The Pennsylvania State University, State College, PA 16802 USA

      Address: 3160 Togeson Hall, Center for Human Computer Interaction, Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: Wheeling Jesuit University, Center for Educational Technology, 316 Washington Ave., Wheeling, WV, 26003-6243 USA

    Search for more papers by this author
  • Than Than Zin,

    Corresponding author
    1. Research Associate in the Center for Human-Computer Interaction, Department of Computer Science at Virginia Tech. Dr. Zin's research focuses on the use of technology for educational purposes in K-12 education.
    • Address: Dept of Computer Science, 660 McBryde Hall (0106), Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: The Pennsylvania State University, 307H Information Sciences and Technology Building, University Park, PA 16802-6823 USA

      Address: 0330C Information Sciences & Technology Building, The Pennsylvania State University, State College, PA 16802 USA

      Address: 3160 Togeson Hall, Center for Human Computer Interaction, Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: Wheeling Jesuit University, Center for Educational Technology, 316 Washington Ave., Wheeling, WV, 26003-6243 USA

    Search for more papers by this author
  • Debbie Denise Reese

    Corresponding author
    1. An educational researcher at Wheeling Jesuit University's Center for Educational Technologies. The Center houses the United States National Aeronautics and Space Administration (NASA) Classroom of the Future (COTF). Reese's research concentration is metaphor: developing a model for the design, development, implementation, and evaluation of metaphor-enhanced learning objects that increase learner understanding of complex concepts.
    • Address: Dept of Computer Science, 660 McBryde Hall (0106), Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: The Pennsylvania State University, 307H Information Sciences and Technology Building, University Park, PA 16802-6823 USA

      Address: 0330C Information Sciences & Technology Building, The Pennsylvania State University, State College, PA 16802 USA

      Address: 3160 Togeson Hall, Center for Human Computer Interaction, Virginia Tech, Blacksburg, VA 24061-0106 USA

      Address: Wheeling Jesuit University, Center for Educational Technology, 316 Washington Ave., Wheeling, WV, 26003-6243 USA

    Search for more papers by this author


This study explores the design and practice of the Blacksburg Electronic Village (BEV), a mature networked community. We describe findings from longitudinal survey data on the use and social impact of community computer networking. The survey data show that increased involvement with people, issues and community since going online is explained by education, extroversion and age. Using path models, we show that a person's sense of belonging and collective efficacy, group memberships, activism and social use of the Internet act as mediating variables. These findings extend evidence in support of the argument that Internet use can strengthen social contact, community engagement and attachment. Conversely, it underlines concern about the impact of computer networking on people with lower levels of education, extroversion, efficacy, and community belonging. We suggest design strategies and innovative tools for non-experts that might increase social interaction and improve usability for disadvantaged and underrepresented individuals and groups.


Community computing is computer networking among and between residents, organizations, government and businesses in a geographically bounded setting for local purposes and activities. These range from the trivial to the profound: checking your bank account and the dry cleaner's hours, to staying in touch with your friends, family, and your church group, and emailing the School Board about inadequate bus routes. Users come from all walks of life, especially in community networks where there is high Internet penetration and use.

In geographic communities, people typically get to know each other in face-to-face settings, and then maintain contact via communication technologies, such as telephone and email. When geographic communities have high Internet penetration, people, groups and organizations readily turn to email, listservs, and the World Wide Web to stay in touch and exchange information. Local non-profit groups, voluntary associations and governmental organizations typically use Internet tools to keep members informed and maintain group interaction. Such social activity that involves groups of people interacting online is call online community (Preece & Maloney-Krichmar, 2003). In highly networked communities, the people who interact online typically get to know each other originally from a face-to-face context or, if they haven't met already, they expect that they could encounter each other face-to-face in the future. For example, most online group interaction occurs within existing formal or informal groups (e.g., the soccer team, book group, or church).

Thus, a key distinction between online communities in community network contexts and online communities across dispersed populations is that people are already part of each other's social networks, or they expect that others could become part of their social network at any time. The fact that members of groups interacting online typically already know each other in networked communities mitigates against some of the problems of social presence online. Social presence theory explains how different communication technologies convey the presence of participants (Short, Williams, & Christie, 1976). When there is no visual image of participants, it is often necessary to include verbal cues (e.g., emoticons) and verbal clarifications (e.g., IMHO) to convey to readers a sender's intended tone and meaning. While there is still a need for such verbal cues online in networked geographic communities, the fact that most people already know each other at least as acquaintances, provides a lot of background information about personality and manner from prior face-to-face interactions. There is also a lesser tendency to flame each other, since participants are likely to meet face-to-face again soon.

Social Network Theory and Online Community

This cross-over between the offline and online worlds for the same social network members has important implications for other aspects of online behavior, such as governance rules, people's roles, and norms of reciprocity.

Social participation in geographic communities takes many forms, including individual and collective, and formal and informal participation. We interact with members of our social networks (friends, family and acquaintances) one-on-one and in groups; we participate in various formal organizations (e.g., school, workplace, voluntary associations) and informal groups (e.g., babysitting circle, carpool, and book group). Our participation in social networks and voluntary associations is dynamic and negotiated, typically based on an exchange of costs and benefits. We invest time and energy in relationships with individuals and groups, and we expect some return in terms of direct or indirect benefits. As rational actors, people seek to minimize the costs or negative effects of interactions and to maximize benefits and utility. Norms of reciprocity or balanced exchange of social and material resources is a fundamental form of human interaction and the central premise of social exchange theory (Emerson, 1976; Pfeffer & Salancik, 1978). Reciprocity is also critical to social network maintenance and resource flows (Berkowitz, 1982; Wellman, Carrington, & Hall, 1988).

Social network exchange theory provides a comprehensive explanation of participation and leads to the development of explanations concerning both formal and informal participation (Cook, 1992; Edwards & Booth, 1973; Emerson, 1972; Fischer, 1977). With roots in earlier theories developed in cultural anthropology, neoclassical economics, and psychology, the school of thought on social exchange focuses on how interaction patterns are shaped by power relationships between individuals, and the resulting efforts to achieve balance in exchange relations. Norms of reciprocity are important to the vitality of both physical and virtual communities (see, e.g., Putnam, 2000; Wellman & Berkowitz, 1988).

Lack of reciprocity can become a ‘social dilemma’ in online communities, especially when the online community is composed of people who are geographically dispersed (Axelrod, 1984; Kollock, 1998). When there is little expectation that participants will encounter each other face-to-face, there is more temptation for people to take resources (help, information, support) from the group and not give back (Walther, 1994). Conversely, in community network settings, when people interact in local groups online, there is much less of this kind of ‘free-rider’ problem since the chance of meeting people from the online community in person is extremely high. Related to this aspect of sociability online are governance rules for the group and roles of participants.

Local voluntary associations and non-profit groups have a basis in the physical community and in their existing face-to-face interactions. They have established rules of governance for the group and roles for participants (leaders, members, committees, etc.). Offline community rules and roles migrate naturally into the online community (Preece & Maloney-Krichmar, 2003). For example, a formal organization, such as a church, that has a sizable membership and hierarchical structure with clear entry and exit strategies for membership, is likely to set up rules of governance online that reflect its offline rules. If it has a listserv, it will probably be established as a moderated list and possibly separate lists for committee and special concerns. On the other hand, an informal group, like a book group, does not have many members or a hierarchical structure and is more likely to establish an unmoderated list or simple group email. The roles of online group members in a networked community typically reflect their roles offline. The organizational leader(s) tend to send the bulk of information and communication to members online. People's expectations of others are largely based on the roles they perceive them to be playing (expert, assistant, nay-sayer, skeptic, etc.).

Social networks plus norms of reciprocity and trustworthiness that arise from them comprise what Putnam (2000) and others call social capital. Coleman (1988) originated the concept of social capital, calling it a common set of expectations, a set of shared values, and a sense of trust among people. Putnam, in a personal communication with Rice (Rice, Katz, Acord, Dasgupta, & Kalpana, 2004, p. 23), clarified that “social capital does not require face-to-face, but face-to-face interaction has more tie density and more reciprocity.” That is, in face-to-face interactions, people are more likely to have friends or acquaintances in common (tie density) and more exchange of aid than in computer-mediated interactions. Wellman, Quan-Haase, Boase, Chen, Hampton, Diaz, and Miyata (2003) note this definition of Putnam's has two components: social contact and civic engagement. Social capital in their view is best sustained by community involvement.

Wellman et al. (2003) add a third component: feeling of community, that is, emotional attachment to a community. Community attachment is multidimensional. We become attached to community through institutional ties, social activity, local intimates, and affective response to (feelings about) place. The longer we live in a community the more opportunities there are to develop and cultivate such ties and associations. Prior studies have also identified life cycle stage and social class as determinants of the extent of attachment (Fischer, 1977; Rothenbuhler, 1984).

Finally, people who belong to more than one group or organization create weak social ties between groups (Granovetter, 1973), or what Putnam (2000) refers to as ‘bridging’ social capital. Simmel ([1908] 1971) is credited with the classic insight that, in essence, intergroup networks simultaneously connect persons and institutions (Wolff, 1950). Two persons may be connected through an interpersonal tie, but a single person may also connect two groups when he or she is a member of both. Such joint memberships form group-to-group ties that indirectly connect all persons in each separate group and that facilitate information flow across separate groups throughout a community, and strengthen bridging (across groups) as opposed to bonding (within groups) forms of social capital. Evidence from our earlier work (Kavanaugh, Reese, Carroll, & Rosson, 2003) shows that people who act as bridges are more involved in local community and use the Internet to sustain and facilitate their involvement.

The focus of the present investigation is whether Internet-based technologies make a difference in the extent to which people become involved and participate in local social life. To address this, we test the flow (cause-effect) and strength of relationships among multiple measures of social participation, social networks and communication behavior that explain perceived increases in overall social involvement among respondents since getting on the Internet (also referred to as “going online”). In short, what are the characteristics, behavior and interests of people who experience increased levels of social participation and community involvement since going online?

The Blacksburg Electronic Village

The town of Blacksburg (population about 43,000 in 2000) benefits from the presence of Virginia Polytechnic Institute & State University (also known as Virginia Tech), a land grant university. The university was the instigator of the community network project, the Blacksburg Electronic Village or the BEV (, in the early 1990s, and recruited the town and the local phone company as project partners. In Fall 2001, 87.7 % of Blacksburg residents and 78.9 % of Montgomery County residents reported having Internet access (Kavanaugh et al., 2003). This is among the world's highest Internet densities. Over 150 community groups and more than 450 local businesses (>75 %) maintained web sites. All of the 20 Montgomery County schools (encompassing Blacksburg) had T1 Ethernet or Token Ring local area networking since 1996, compared to about 65% nationally; all of the middle and high schools had T1 Internet connectivity since 1995.

The BEV has hosted standard Internet content since its inception in 1993 and has managed local services such as web space, listservs, and email accounts, as well as specific community-oriented services like information and web-based forums for local town and county government, social services, public education, libraries, and health care. It has also provided some support for the commercial sector, by linking from its ‘Village Mall’ listing to merchant websites hosted elsewhere. Many community-oriented initiatives are maintained, including community newsgroups, organization lists, a senior citizens' nostalgia archive, and video streaming of Town Council meetings. Some larger local organizations, such as the public schools, manage their own infrastructure, online content and services. Free public access and a variety of ongoing training classes have been available through the local libraries and town recreation center.

As the Internet diffused throughout the geographic and networked community of Blacksburg, Virginia and environs, the proportion of total users who used the Internet for political purposes declined, and the proportion who used it for social purposes continually increased (Kavanaugh & Patterson, 2001). This is largely due to the tendency of early adopters to have higher levels of education and civic engagement, and to use the Internet in pursuit of their interests in local issues and civic concerns. Later adopters are more interested in using the Internet for entertainment, shopping and maintaining social ties than for political and civic engagement (Kohut, 1999). We know from other studies that with the exception of political purposes, most people who use the Internet use it for many different activities and purposes, not for just a single purpose (Kraut, Kiesler, Bonka, Cummings, Helgeson, & Crawford, 2002; Madden & Rainie, 2003). Maintaining or cultivating social network ties (whether local or distant) through email and online discussion formats has been one of the most popular uses of Internet (see, e.g., Patterson & Kavanaugh, 1994; Madden & Rainie, 2003).

In a geographic community such as Blacksburg and surrounding Montgomery County where Internet use has been well established for a long period (over a decade) and almost 90% of the Blacksburg population (and almost 80% of the County population outside Blacksburg) is online, we have an opportunity to investigate what Sproull and Kiesler (1991) call second-order effects. That is, beyond the initial deployment and diffusion of innovative technology there are changes that people begin to make in their expectations of what is possible, how to communicate with each other, and ways to coordinate collective action. Further, they adjust their behavior to technological change, or they sometimes change the technology to adapt it to their behavior. For example, community groups expect that if they put information on a web site, the majority of their membership will have access to it. In Blacksburg, many groups report they do not struggle as much to find leaders and other volunteer help because near-universal email usage means there is less work for leaders to manage information and communication exchange and coordinate activity among members and other leaders (Kavanaugh & Schmitz, 2004). As a result, people are more willing to take on the chores of running the organization and coordinating volunteer help. While Blacksburg and environs are ahead of the curve in terms of diffusion and usage, the secondary effects observed there provide indications of what is likely to occur in other similar communities throughout the rural United States.

Survey Methods

Survey Households and Instrument

Our survey sample pool encompassed all households in Blacksburg and Montgomery County. We used a random sample of 1250 households that we had purchased one year earlier from Survey Sample, Incorporated (SSI) and from which we had already identified and eliminated about 300 out-of-date or otherwise undeliverable addresses. SSI had drawn up the sample based on zip code areas we had provided, and using telephone directories, Department of Motor Vehicles databases, and post office notifications (deaths, address changes), among other records. We sent a letter of invitation to the 870 valid household addresses by postal mail asking them to participate in our study and, if interested, to complete a short (10 item) non-anonymous questionnaire asking their name, household location (Blacksburg or Montgomery county), highest level of education attained in the household, number and ages of household members, years lived in the community, occupation, Internet use, and whether the respondent owned or rented their home (as a proxy for income level). From this invitation, we received 188 responses of completed questionnaires (22%). We created a stratified sample of 100 households, based on location, education level, and Internet use. We wanted to have a sample that would be fairly representative of the combined population of Blacksburg and Montgomery County (77,500 comprising 27,000 households in 2001) according to the key factors of the study. We drew half of our stratified sample from Blacksburg in order to have equal representation from the two locations; the population in Blacksburg (about 43,000 in 2000, according to census data) and the population in surrounding Montgomery County (roughly 34,000) are not exactly evenly split. But drawing half of our stratified sample from each location allowed us to investigate usage patterns and impacts outside the university town of Blacksburg, where the population is a mix of better-educated and disadvantaged, low income residents. An equal number of Blacksburg and Montgomery households returned the original 10-item household survey (91 versus 90). We selected households along the factors of education and Internet use that represented the whole population.

We adapted our survey instrument from questionnaires previously administered by the BEV research group (Kavanaugh, Cohill, & Patterson, 2000; Patterson & Kavanaugh, 1994) and by the HomeNet study (Kraut, Scherlis, Mukhopadhyay, Manning, & Kiesler, 1996; Kraut et al., 2002). We hand-delivered the surveys to each of the 100 participating households throughout the Fall of 2001 in order to begin a process of getting to know our respondents, and to improve retention over the two rounds (two years) of the survey administration. About twenty of these households were selected for a series of in-depth interviews and/or logging of computer usage. We asked each member of the household to complete a separate survey and mail it back within several weeks to us in a prepaid envelope. We paid participants a participant fee of $25 for each completed survey. We had a slightly modified survey completed by youth (aged 10-16) and a short survey for small children (up to 10 years old) that was completed by an adult in the household. We repeated this procedure one year later, in Fall 2002, for a second survey round with the same households and individuals and the same survey questions.

While our sampling is at the household level, our unit of analysis in this article is the individual. The same individuals completed the same survey (rounds one and two) one year apart. There were some minor changes in the participants between rounds one and two. Six households (19 individuals) dropped out of the study or never completed the second round of the survey and were replaced by the same number of households (six) with similar values on the stratification measures: location (Blacksburg or Montgomery County), highest household education level, and whether there was a household member who used the Internet. However, the total number of individuals in the replacement set of six households was six (6), which gave us a smaller total number of individual respondents in round two (143) compared with round one (156). This small change in the individuals in the second round does weaken slightly the comparability of results in the two rounds, and may account for some discrepancies in significant findings between the exploratory and confirmatory analyses. We named our study ‘Experiences of People, Internet and Community’ (EPIC).

Survey Variables and Constructs

The survey questions were organized into sections that pertained to the main study variables and constructs; these were: community involvement and attachment, interests and activities, Internet use and experience, social circles, significant life changes, psychological attributes, and demographics.

Exploratory Data Analysis

In exploratory data analysis we analyzed the first round of survey data by aggregating variables related to common constructs (such as, community involvement, types of activities and interests, extroversion, and Internet use and experience) and ran correlations on the variables for each construct. We conducted reliability tests to obtain constructs comprised of one factor, with a Cronbach alpha greater than 0.7, indicating high reliability that these questions all measure a single construct (DeVellis, 1991). While we developed constructs for all the main items in the survey, we examine in this paper only those that bear on social participation. In confirmatory analyses, we analyzed round two survey data to test for the same constructs. Table 1 contains reliability indices (Cronbach alphas) for each construct we analyzed and tested to create the Social Participation path model (see Figure 1). Table 1 also contains example survey questions and/or definitions for variables or constructs.

Table 1. Survey variables and constructs: Rounds 1 and 2
Variable or ConstructAlpha1Example Survey Question and/or Definition of Variable/Construct
 Round 1Round 2 
  1. a NA = Not applicable. Alpha value is not applicable because the variable is one item measure

  2. b Same measures from round 1 were used for round 2 data

EducationNAaNAWhat was the last grade of school you completed? (low=1, less than 8th grade; high=7, completed graduate school)
Extroversion.86bNASpend time with friends; I am not alone
AgeNANAAge calculated from self-reported Date of Birth
Age SquaredNANASquared of Age
Participation:.88.86Activism, Affiliation, Belonging (see Results Section)
Collective Efficacy.86.88Thirteen-item measure on respondents' attitudes and beliefs regarding the potential of the community as a whole to solve problems or to achieve goals (see Table 2)
Table 2. Collective efficacy scale
Our community can present itself in ways that increase tourism.
We can greatly improve the roads in Blacksburg and Montgomery County, even when there is opposition within the community.
I am convinced that we can improve the quality of life in the community, even when resources are limited or become scarce.
Our community can greatly improve the quality of education in Montgomery County without help from the Commonwealth of Virginia.
As a community, we can handle mistakes and setbacks without getting discouraged.
Our community can cooperate in the face of difficulties to improve the quality of community facilities.
I am confident that we can be united in the community vision we present to outsiders.
Despite our differences, we can commit ourselves to common community goals.
The people of our community can continue to work together, even when it requires a great deal of effort.
We can resolve crises in the community without any negative aftereffects.
Our community can enact fair laws, even when there is disagreement among people.
I am confident that our community can create adequate resources to develop new jobs despite changes in the economy.
Our community can greatly improve services for senior citizens in Blacksburg and Montgomery County without help from the Commonwealth of Virginia.
Figure 1.

Social participation path model
**p<.01, *p<.05, ?p=.06
Age represents two measures of age: linear age and curvilinear age (or age squared). Only one representation for age is shown in the model to simplify the visual presentation.

In addition to demographic data and some psychological attributes (e.g., extroversion), we examined in this article the variables and constructs of Participation (and components of participation derived from factor analysis, including Activism, Affiliation and Belonging discussed under Results), Collective Efficacy, Internet use for social purposes (Online Social), and Overall Involvement since getting on the Internet. We also compared the construct Affiliation (an agreement scale on joining many organizations, and belonging to many groups) with the actual number of groups and organizations that the respondents listed in which they participate (our variable Membership). We later coded these specific groups that respondents listed into four major categories, based on categories established in other studies that fit well with our research (Edwards & Booth, 1973; Kasarda & Janowitz, 1974; Putnam, 2000). These are: 1) civic/political, 2) religious/charitable, 3) social, and 4) educational/professional. Some studies have found that membership in multiple voluntary associations (as opposed to just one or none) is an important predictor of community involvement (e.g., Edwards & Booth, 1973; Kavanaugh et al., 2003; Putnam, 2000).

Group Membership

Within the survey, we used a matrix format to investigate respondents' level of involvement and the group's mode(s) of communication with members for each organization each respondent had listed. The matrix provided four types of involvement (attendee, member, leader, and financial contributor) and asked respondents to check all that applied. For communication, we had listed seven modes (face-to-face, telephone or postal mail, person-to-person email, email discussion or listserv, newsgroup, chat room, and online bulletin board or discussion board), and asked them to check all that applied. We later coded these specific groups into four major categories, based on categories established in other studies that aligned with our research (Edwards & Booth, 1973; Kasarda & Janowitz, 1974; Putnam, 2000). These are: 1) civic/political, 2) religious/charitable, 3) social, and 4) educational/professional. Respondents were asked to indicate level of involvement for each organizational affiliation (leadership, attendee, contribute money, and membership). We coded level of involvement into active (leadership, attendee) and passive (contribute money, am a member). The survey listed seven modes of communication for each organization and asked respondents to select those that were employed by each of their organizational affiliations. We coded the seven modes of organizational communication into two categories: traditional (i.e., face-to-face, telephone/postal mail) and electronic (i.e., one-to-one email, email discussion/listserv, newsgroups, chat room, online bulletin board). We ran t-tests across rounds one and two for each of the four types of groups (civic, social, etc.) to investigate differences in group communication, and levels and types of involvement in the group. Members of each type of organization were defined as subjects who belong to the same type of organization in both rounds. That is, subjects who belonged to the organization in either round (but not both) were not used in the t-test analysis. This methodology allows us to control for changes that might be due to different types of people entering or leaving the same type of group.

Collective Efficacy

We measured collective efficacy using a 13-item scale we developed based on Bandura's (2000) concept of collective efficacy. Collective efficacy is a measure of a respondent's belief that the community as a whole can overcome obstacles to solve common problems. The construct of Collective Efficacy forms a single construct with high reliability, as well as three dimensions derived from factor analysis (Carroll & Reese, 2003).

Extroversion was also measured by Likert-scale items of agreement regarding self-reported psychological and behavioral attributes, such as: being talkative, outgoing, enthusiastic, emotionally stable, and handling stress well (e.g., I feel that there is no one I can share my most private worries and fears with). We did not repeat the questions regarding extroversion in round two, as we did not expect this psychological attribute to change from round one to two. The drawback was that for the handful of respondents who were recruited only in round two to replace similar households that dropped out after round one, we do not have a measure of extroversion.

Internet Use and Outcomes

We measured Internet Use variables with questions about respondents' ratings on frequency scales of Internet use for various purposes (civic, political, health, commerce, social, educational/professional). We formed the Social Internet Use construct from responses to frequency ratings on the six questions: I use the Internet to communicate with friends (local, non-local), family (local, non-local), co-workers (about non-work issues), and to meet new people. The construct had high reliability (.79, .80) in rounds one and two, respectively.

The dependent (outcome) variables investigated in this study were comprised of nine questions about changes in the respondent's level of involvement since getting on the Internet(Table 3). These questions are adapted from a set of questions originally developed and used by the Georgia Tech Visualization and Usability Center in 1996 (GTVUC, 2005).

Table 3. Dependent variable: Since getting on the Internet
Since getting on the Internet…
… are you more, less or equally involved in local issues that interest you?
… are you more, less or equally involved in national issues that interest you?
… do you feel less or equally connected with people inside your community?
… do you feel more, less or equally connected with people outside your community?
… do you feel more, less or equally connected with a diversity of people inside your community?
… do you feel more, less or equally connected with a diversity of people outside your community?
… have you become more, less or equally involved inside the local community?
… have you become more, less or equally involved outside the local community?
… have you attended more, less, or an equal number of meetings and events of local organizations that interest you?

These nine questions all form one construct we call ‘Overall Involvement’ with a high reliability (Cronbach alpha =.833 and .869 in rounds one and two, respectively).

Confirmatory Data Analysis

Confirmatory data analysis goes beyond hypothesis formation, descriptive statistics and correlations, and tests particular hypotheses in the study (Kidder, 1981). The t-test is a common form of confirmatory data analysis that compares means of two samples, or the same sample at two different times. One can also calculate the effects of a set of variables on an outcome variable by regressing the dependent variable on the set of independent variables. In this type of analysis each indicator variable is treated as an exogenous variable. That is, the researcher makes no claims about the causes of the relations among the independent variables. Given a theoretical formulation of how variables are related, a researcher can hypothesize a model of the causal effects among those variables. Variables for which the model does not provide a causal explanation remain exogenous. Endogenous variables are those determined by the exogenous variables or by other endogenous variables. The model is referred to as a path model (Pedhazer, 1997). A path analysis consists of a series of regression equations, one for each endogenous variable in the model. The standardized regression coefficients become the path coefficients in the model. A researcher can then use the path coefficients to decompose the total effects of variables on each other into direct effects and mediated, or indirect effects.

We developed a path model initially using exploratory (round one) data to explain changes in involvement in local issues of interest since getting on the Internet. We then tested the same model using round two data to provide confirmatory analyses. Although the Pedhazur (1997) discussion assumes that all relations within a model are linear, we have allowed for a curvilinear relationship between the exogenous variable of age and the endogenous variables. For more background on the survey questions and constructs and the statistical analyses, please see the project web site (



The demographics are roughly the same between rounds one and two because these are the same respondents with the minor exceptions noted above(Table 4). Descriptive statistics of respondents show that education and income are relatively high, as are Internet use and experience. The average education (based on highest in household) is college graduate; median household income is $35,000-$50,000. This is not surprising given that Blacksburg is a university town and home of the BEV. The average length of residence (an indicator of community attachment) is 18 years.

Table 4. Demographics and basic indicators of survey respondents
VariablesMeansStandard DeviationNumber of RespondentsPercent
  1. Note: All variables have been coded so that higher numbers indicate more of the variable.

  2. 1Dichotomous Variable (0/1)

  3. 2Converted from DOB; range 16-80 years old

  4. 3Frequency from Race Categories

  5. 4Seven Categories Elementary to Completed Graduate work

  6. 5Frequency from Relationship Status categories

  7. 6Six categories from Under $25,000 to Over $100,000

  8. 7Range 1-79 years

  9. 8Number of Local Organizational or Group memberships; Range: 0-9

Location (Blacksburg)1    1561434845.5
Gender (male)1     1424642.3
Age24647.4515.0314.39 143  
Race (Caucasian)3      91.690.2
Household education45.045.061.491.47    
Marital status (married)5      6666.4
Household income (median)63.5641.51.57145134  
Years lived in a community71818.6716.4117.01156143  
Number of local groups82.342.132.051.9    

For the total sample, eighty-three percent (83.1%) of respondents report having access to the Internet in round one, and 84% in two. Within Blacksburg, 87.7% of respondents report using the Internet in round one (78.9% in Montgomery County), rising to 90.8% in Blacksburg in round two. (Montgomery County falls slightly to 78.2%, possibly due to the minor changes in households between rounds one and two.) These are very high penetration rates compared to national levels: 58% of US individuals in Spring 2002 (Lenhart, Horrigan, Rainie, Allen, Boyce, Madden, & O'Grady, 2003).

Among Internet users, over 90% have Internet access at home, although a slightly lower percentage (89.8%) reports using the Internet from home. This is due to the fact that not all adult members of a networked household go online and there is also some ‘second-hand’ or indirect use (Dunlap, Schafer, Carroll, & Reese, 2003).

The fact that we derived our survey from established instruments allows us to ground the characteristics of our sample, findings, and possible generalizations through comparison to both national averages (see Pew, and the HomeNet study population. In our study (EPIC), Internet users tend to be slightly older than the national average, wealthier, and better educated than the average American Internet user. For example, in 2001 (EPIC round one time frame) across the nation 29% of Internet users were 17-29 years old. Only 17% of the EPIC Internet users fell into this age range at that time (round 1). While only 18% of the nation's Internet users fell into the 54-64 age range, 27% of the EPIC users were 54-64 years old. Across the United States, Internet usage was pretty evenly split by gender. There were 10% more EPIC women Internet users than men. EPIC Internet users are more affluent than the typical American Internet user. Across the nation, 18% of Internet users are in the $50,000-$75,000 income bracket. Thirty-one percent of EPIC Internet users reported an annual income within this range. In round one of our study, 58% of respondents used the Internet at work, corresponding to 64 out of 100 households in the study. This is high compared with 28% of households nationally and 32% of households in the state of Virginia in 2000. EPIC Internet users were also more educated than the typical United States Internet user. Across the nation, 71% of Internet users reported some college, college degree or graduate degree compared to 88% of EPIC Internet users. Thus, the average EPIC Internet user was older, more affluent, and more educated than the typical American Internet user.

Participation and Membership

The Participation construct is statistically derived from multiple variables, most of which are Likert ratings for level of agreement or frequency scales. Activism, Membership, and Belonging are the three components derived from the construct Participation(Table 6). Further construct reliability testing revealed that the construct Belonging is comprised of the factor analysis components we named Social and Attachment. We named the construct Belonging, as it measures a sense of belonging to the general community (Attachment) and belonging to a group of friends (Social). It seems reasonable that belonging to a group of friends should combine with high reliability with a sense of belonging to the general community, since earlier studies show that people's attachment to a community is greater when they have strong ties to friends, neighbors or family in that place. In our round one data, for example, Pearson correlation tests showed that the construct Social (belonging to a group of friends) is highly correlated with Community Attachment (.716, p<.01) which is consistent with studies showing that strong social ties tend to increase a person's attachment to community.

Table 6. Survey constructs from round I and round II
 Alpha1Examples of variables in Participation Construct and its components
  1. 1The values of alpha were rounded to two decimal places

ConstructRound 1Round 2 
Affiliation.70.54Join many organizations; join many groups
Activism.89.89Have ideas to improve local community; work to bring local change; active & involved
Belonging:.73.69Social, Attachment (below)
Social.69.69Spend time with friends; not alone; have group of friends; help friends/ neighbors in need
Attachment.67.67Belong in local community; feel part of local community

Factor analysis (principal axis factoring) of round one data showed that the Participation construct is comprised of four components we named Activism, Membership, Social, and Attachment(Table 7).

Table 7. Factor analysis of participation construct from round 1
  1. Rotation Method: Varimax with Kaiser Normalization

Work to bring about local change.758.241.109.185
Have ideas to improve community.700.191.080.078
Active & involved in community.689.301.378.225
Know people who know what's going on.655.158.187.225
Spend time helping solve local problems.650.246.305.023
Participate in local activities.505.435.426.168
Spend time with my friends.094.761.156.084
Do not spend time alone.187.492.025.209
Have a group of friends.187.487.222.136
Help friends or neighbors in need.246.457.018-.031
Join many organizations.122.209.854.101
Join many groups.371.040.686.078
Belong in the local community.
Feel part of the local community.258.422.121.508

Activism is a composite of several survey items about how active and involved the respondent is in the local community compared with others they know, how frequently they have ideas to improve their local community, and how frequently they work to bring about change in their local community, how frequently they get together with others who know what's going on in the local community. Belonging is a composite of multiple survey items, including how much time a respondent spends with friends, whether they are frequently alone, and Likert-scale items of agreement on help a respondent provides to friends or neighbors in need.

Membership was a more reliable measure of respondents' participation in local groups and organizations than the construct Affiliation. In round two, Affiliation not only had a lower reliability than round one (Cronbach's alpha in round two was only .54 compared with .70 in round one, Table 7). We also found that even though a respondent may have listed only one organization to which s/he belonged, for example, s/he indicated high agreement with Likert-type questions such as ‘I join many groups,’ or ‘I join many organizations,’ as described earlier. Therefore, a simple sum of the number of groups or organizations listed by the respondent (what we refer to as Membership) provided a more reliable measure for this variable. Group or organizational membership is positively correlated with social class (measured by education and income), length of residence (years lived in the community) and home ownership, all of which are also considered to be indicators of community involvement and attachment.

In order to investigate differences in participation by type of group, we ran paired t-tests across rounds one and two for each type of group (civic, social, religious, professional) to investigate differences in type of group communication, and in levels and types of involvement in the group. The social type of groups showed significant differences between rounds one and two(Table 8) on several measures of communication (traditional versus electronic, and level and types of involvement in the group).

Table 8. Differences in social type of groups over time (N=137)
VariablesMean Values for Rounds 1 and 2 and (SD) p Valid N
  1. **p<.01, *p<.05

Traditional Communication2.46(1.04).02*40
Electronic Communication0.99(0.69).05*40
Level of Involvement2.38(0.94).01**40
Active Involvement1.04(0.63).000**40

Social groups had significant increases in several measures between round one and two that were not present in the other three types groups (civic, educational/professional or religious/charitable). Both their traditional forms of group communication (face-to-face, telephone) and their electronic forms of communication (email, listserve and Web) increased significantly over time. Their level of involvement also increased over time (that is, the measure of participants' involvement in terms of attending, becoming a member or leader, and contributing financially). Social groups were also the only group to show significant increases over time in active types of participation (attending or becoming a leader).

Internet Use

For the questions comprising the section on Internet use, we obtained eight constructs: commerce, education/work, leisure, health, general Internet, political, social, and civic life. The constructs were statistically derived from multiple variables, each a Likert rating for the frequency of particular online activity during the past six months. The Social Internet Use construct (?=.7878, N=129) is comprised of five variables measuring frequency of Internet-based communication with friends and family, and with co-workers about non-work issues(Table 9).

Table 9. Internet use and outcome variables
 Alpha1Examples of variables in construct
ConstructRound 1Round 2 
  1. 1The values of alpha were rounded to two decimal places

Social Internet Use0.790.80Five-item measure on frequency of using Internet for social purposes (e.g., I use the Internet to communicate with friends in the local area)
Overall Involvement0.830.87Nine-item measure on level of involvement since going online (from Table 3)

Each respondent received a score for Overall Involvement that was the average of the nine scores from the 9-item scale (shown in Table 3). Most respondents on questions of Involvement since getting online reported that they were equally involved; that is, getting on the Internet did not affect their level of involvement in community, issues of interest, and people. Among those who reported changes in involvement, however, a higher percentage reported they were more involved than reported they were less involved since going online.

Social Participation Path Model

We tested various constructs and variables to create a path model, based on previous studies, theoretical arguments and our own statistical testing (correlations among study variables and regression analyses). As noted earlier, path analysis is a method for studying direct and indirect effects of variables hypothesized as causes of variables treated as effects. It sheds light on the tangibility of the causal models that a researcher formulates based on knowledge and theoretical considerations. In path analysis, more than one regression analysis may be called for. In the present study, six regression analyses were run to draw the path diagram (Figure 1).

Notes for Figure 1:

  • 1In round 1, the path from Age to Belonging was significant (p<.05) and the path coefficient was -.785; the path from Age Squared to Belonging was not significant. In round two, both Age and Age Squared had significant paths (p<.01) to Belonging and the path coefficients were -1.405 and 1.181, respectively.
  • 2Both Age and Age squared had significant paths (p<.05) to Collective Efficacy in round 1 and the path coefficients were -.879 and .857 respectively. Both Age and Age squared had significant paths (p<.05) to Membership in round 1 and path coefficients were -.891 and 1.087, respectively.
  • 3For summation of direct and indirect effects in the path model see the project web site:

At each stage of the path analysis, an endogenous (dependent) variable is regressed on the variables that are hypothesized to affect it. The standardized regression coefficients s calculated are the path coefficients for the paths leading from the particular set of variables (predictors) to the dependent variable under consideration. The path diagram visually displays the hypothesized pattern of causal relations among a set of variables. A path coefficient indicates the direct effect of a variable (from which an arrow is drawn) hypothesized as a cause of a variable taken as an effect (to which the arrow is pointed). For example, there are three arrows pointed at the construct Belonging in Figure 1. Hence, the three causes (hypothesized and data supported) of the Belonging construct are the variables from which these three arrows are drawn, namely, Extroversion, Aget, and Membership. The numbers on the arrows are the values of corresponding path coefficients or standardized regression coefficients s.

The amount of variance in dependent variables (the R square value) explained by the set of corresponding predictor variables are described as percentages for each construct in the model. (Round two variances are shown in parentheses.) For example, Education, Extroversion and Age explain 16% and 15% of the variance in Collective Efficacy in rounds one and two, respectively (although Age and Age Squared are significant in round one only). Dotted lines between variables in the diagram are the paths that were significant either in round one or two with path coefficients shown along the corresponding path (and round two path coefficients in parentheses). Only the respondents who use the Internet are included in the path analyses.

Education, Extroversion and Age indirectly predict changes in Overall Involvement since going online. Collective Efficacy, Membership and Belonging are significant mediating variables explaining variance in Activism and in Social Internet Use (using the Internet for social purposes). Activism and Social Internet Use are significant in explaining variance in the dependent variable Overall Involvement since going online. Age has a curvilinear, quadratic relationship with Belonging; that is, people in the middle of the age range spend less time with friends and neighbors, and feel less a part of the community than either younger or older people. In examining this further, we conducted a one-way ANOVA on the entire EPIC sample in round one which confirmed that younger people (age 16-34, mean = 3.57, N = 37) and older people (age 65-80, mean = 3.55, N = 16) are more likely to be part of a group of friends and feel they are part of the general community than those in the middle range (age = 35-64, mean = 3.21, N=103), F (2,153) = 4.8, p<.01. Tukey post hoc analysis identified only the young and middle groups as significantly different (mean difference = .35, p<.05).

The results of the series of regressions in the path model, including dependent variables and their corresponding set of predictors, F values and probability for Type I error are shown in Table 10. The significant F (p<.05) values tell us what the corresponding set of predictor variables can predict about the dependent variable at better than chance levels.

Table 10. Regression results for the social effect model
PredictorsDependent variables and Percent of variance explained (R2)aF values & Probability statistics (p)
  Round 1Round 2
  1. a Variance explained in round 1 & round 2 are presented in two rows

  2. b Overall Involvement variable was regressed on Collective Efficacy, Membership, Belonging, Activism, and Social Internet Use. Valid N = 126 in round 1 and 119 in round 2 for Overall Involvement

1. Extroversion   
2. AgeBelongingF (5,122) = 12.428F (5,108) = 10.088
3. Age squared33.7%p<.01p<.01
4. Collective Efficacy31.8%  
5. Membership   
1. Extroversion   
2. Collective EfficacyMembershipF (5,122) = 4.974F (5,108) = 2.384
3. Education16.9%p<.01p<.05
4. Age9.9%  
5. Age squared   
1. Extroversion   
2. EducationCollective EfficacyF (4,123) = 6.019F (4,109) = 4.89
3. Age16.4%p<.01p<.01
4. Age Squared15.2%  
1. BelongingActivismF (3,124) = 50.38F (3,116) = 31.179
2. Collective Efficacy54.9%p<.01p<.01
3. Membership44.6%  
1. ActivismSocial Internet UseF (3,123) = 5.783F (3,116) = 7.253
2. Belonging12.4%p<.01p<.01
3. Membership15.8%  
1. ActivismOverall InvolvementbF (4,122) = 12.173F (4,114) = 20.7
2. Belonging28.5%p<.01p<.01
3. Membership42.1%  
4. Social Internet Use   

A linear combination of Belonging and Membership explain about 50% of the variance in Activism. As noted earlier, the Activism construct is comprised of how frequently the respondent gets together with others who know what's going on in the local community, and several questions about how active and involved they are to bring about change in the local community. Membership is significant and Belonging approaches significance in explaining variance in Social Internet Use in round two.

Belonging and Membership indirectly affect Overall Involvement, mediated by Activism and Social Internet Use. A combination of Activism and Social Internet Use explains 29% and 42% of the variance in the outcome variable, Overall Involvement since getting on the Internet, in rounds one and two, respectively. Both paths from Activism and Social Internet Use to Overall Involvement are significant in both rounds.

Socio-technical Implications

We have marshaled evidence that Internet use can strengthen social contacts, community engagement and attachment for people with relatively high levels of education, extroversion, sense of community belonging, community collective efficacy, group memberships, activism and social use of the Internet. But these results have a darker side with respect to the potential impact of computer networking on people with lower levels of education, extroversion, efficacy, and community belonging. Over time, will not these patterns aggravate the digital divide?

We know from studies noted earlier that even populations with low socio-economic status can develop higher levels of efficacy through positive experiences and reinforcement. These positive experiences tend to be external, such as local government's concern and active recruitment of feedback from underrepresented groups over sustained periods of time, or the public school's investment in assistance to students with special needs, continuing education and remedial academic support. While such responses from community organizations tend to be the exception rather than the rule, they are demonstrated to be effective in raising collective efficacy and fostering optimism in targeted populations.

Two mutually-reinforcing design strategies that could address this challenge are (1) strongly participative approaches to defining and implementing community-oriented technologies, and (2) public access to the Internet in general, and to community-oriented networks in particular, emphasizing active participation, and not just browsing and buying. Communities with local community computer networking initiatives (such as Blacksburg and surrounding Montgomery County) have an innate goal to involve citizens at the grassroots level in the design and development of online content, tools and infrastructure.

Participatory design is the direct and autonomous involvement of users, or future users, in the technical design of functionality and capabilities (Carroll, Chin, Rosson, & Neale, 2001; Muller & Kuhn, 1993; Preece & Maloney-Krichmar, 2003; Schuler & Namioka, 1993). Such participation ensures that a broad range of interests, needs, and values are addressed in the design work. We had developed a Java-based toolkit for collaborative component-based software (Isenhour, Rosson, & Carroll, 2001) and a long-term approach to participatory design (Carroll et al., 2001). Each represented a technical innovation in its own area, and each guided and leveraged the other. In working with end users on requirements analysis and understanding their needs for technology design, we were able to develop tools that met their purposes. Even several years later, public school teachers learning about these developments referred to them as the ‘answer to our prayers.’ However, participatory design is rarely used for people with disadvantages in education, such as below poverty line groups, or others who typically have low computer literacy (e.g., the elderly). In Blacksburg, a large group of senior citizens became active early on in using the Internet through the BEV project, and benefited from the training and support project staff offered (Casalegno, 2000). Through innovative projects that were designed by seniors, such as the Nostalgia Project (Carroll, Rosson, VanMetre, Kengeri, Kelso, & Darshani, 1999), the elderly enhanced their software skills and concepts, and engaged in social interaction with one another other and with younger people.

For people with lower education, learning to use information technology effectively is still a daunting task. Moreover, learning to use the Internet often stops with learning how to search for and purchase consumer products. Just placing technology in publicly accessible locations, such as public libraries, municipal buildings, or even street kiosks can help lower the barrier to participation. In the early phase of the BEV project, townspeople made extensive use of four networked computers in the public library. New users were able to get help with starting an account and performing basic tasks.

Contemporary innovations in tool designs are making networked information easier to retrieve, edit and customize. They can both accelerate and broaden learning. Tools such as wikis, for example, are more transparent and have greater usability than web pages that must be edited with a full knowledge of HTML. Web logs (or blogs) are easier to use for group interaction online and provide more exchange among participants than either listservs or newsgroups. These innovations in tools for non-experts and others on the horizon may help to extend the community benefits shown in the social participation model to a broader population.


It is important to bear in mind that most of the variables and constructs presented and tested in this longitudinal study have relevance to the localism that defines community computing (e.g., Collective Efficacy, Belonging, Activism, Social Internet Use, Group Membership, and Overall Involvement since getting on the Internet). In community computing contexts, online interaction often occurs among people who belong to corresponding social networks and formal and informal groups that exist offline, and therefore, who also see each other in face-to-face settings. Being members of each other's social networks provides built-in norms of reciprocity, rules, and roles that carry over into the online community.

People with two or more group memberships (forming ‘bridging’ or weak social ties across groups) are more active and involved in the community, and have more social ties and stronger community attachment. Social groups also showed significant differences between survey rounds one and two. Unlike the civic, educational/professional or religious/charitable groups, social groups had significant increases in several measures between rounds one and two. Both their traditional forms of group communication (face-to-face, telephone) and their electronic forms of communication (email, listserv and Web) increased over time. Respondents' involvement in the group also increased significantly over time (that is, the measure of participants' participation in terms of attending, becoming a member or leader, and contributing financially). Social groups were also the only group to show significant increases over time in active types of participation (attending or becoming a leader). All of these differences emphasize the social aspects in the community that are reinforced and strengthened not only by more face-to-face interaction but also by more online interaction.

The path model shows that Education, Extroversion and Age are significant exogenous variables explaining involvement in community, issues and people since getting on the Internet. Education, Extroversion and Age are among the few variables in the study that do not pertain to localism or geographic community. Overall Involvement comprises local and non-local forms of involvement (e.g., involvement in local and global issues, local and non-local people, a diversity of people inside and outside the local community). However, all but one of the mediating variables in the model contribute significantly to the variance in the Overall Involvement outcome: Collective Efficacy, Membership, Belonging, and Activism. The construct Social Use of the Internet measures both local and non-local friends, and family, as well as co-workers (who would typically be local).

The results of the path model emphasize the social nature of increases in overall involvement. That is, the mediating variables of Collective Efficacy, Membership, Belonging, Activism and Social Internet Use all pertain to social aspects of community life. Even Activism (the least social for the mediating variables) contains a social item: ‘I frequently get together with people who know what's going on in my local community.’ Moreover, all of these constructs and variables that are mediating variables are grounded in offline community: believing that community members can work together to solve problems despite obstacles (Collective Efficacy); belonging to multiple groups or organizations (Membership); having a sense of belonging to a group of friends, feeling a part of the general community (Belonging); Using the Internet to communicate with friends, family, coworkers (about non-work issues); and being active and involved in the local community (Activism). These offline relationships have built-in norms of reciprocity, governance rules, and participant roles that transfer to their online interactions. It is harder to shirk responsibility online when we expect to see other members face-to-face in the near future.

One main concern that our findings raise is that the positive social impacts of information and communication technology are associated with higher levels of education and extroversion and with life cycle (i.e., 35-64 years of age). These demographics and psychological attributes predict participation in community life, local groups, sense of belonging and collective efficacy, all of which lead to higher levels of activism and social uses of the Internet, and ultimately to increases in overall involvement both within and beyond the geographic community once people go online. There is a clear concern that people with lower levels of education who are more introverted, younger or older, and who do not participate in local groups, will not contribute to the general pool of social capital and collective action. This lack of participation may be due to lack of motivation-and motivations might change with changes in circumstances or with specific critical incidents. It may also be due to an overall lack of self-efficacy and/or collective efficacy or something else altogether. These are areas for future work.


This research is supported, in part, by the National Science Foundation, IIS 0080864. We thank our collaborators Albert Bandura, Ann Bishop, Daniel Dunlap, Philip Isenhour, Robert Kraut, Wendy Schafer and Steven Winters.