Author Note.

This research operates under the aegis of Carnegie Mellon’s Human Computer Interaction Institute and has been supported in part by grants from Apple Computer, AT&T, Bell Atlantic, Bellcore, CNET, Hewlett Packard, Intel, Carnegie Mellon Universi ty’s Information Networking Institute, Interval, Lotus Corporation, The Markle Foundation, the National Science Foundation (Grant #IRI-9408271), the NPD Group, Nippon Telegraph and Telephone Corporation, Panasonic, the US Postal Service, and US West.

Please direct correspondence to Vicki Lundmark (412-268-7505, lundmark+@andrew.cmu.edu) or Sara Kiesler (412-268-2848, kiesler+@andrew.cmu.edu) at Social and Decision Sciences, 319 Porter Hall, Carnegie Mellon University, 5000 Forbes Ave., Pittsbur gh, PA 15213; fax (412) 268-6938.

In a field study of household Internet usage, 89 percent of 93 families but only 49 percent of individuals asked the help desk for assistance. Characterizing individual requests for external technical support as a form of "voice," we find th ey not only stem from commitment ("loyalty") but they raise the level of commitment as well as lower technical barriers. Help requests also reflect a division of labor within households. A committed family member becomes the computer "gur u," the help desk caller, and the person whose Internet usage predicts other members’ usage. Our findings extend Hirschman’s (1970) theory to show how voice affects commitment and how group-level analyses can add to our understanding of people’s res ponses to technology. They also imply that people who access computer support lines may represent a skewed sample of usability problems.

When your washing machine stops working, you needn’t wonder if your clothing, home wiring, or skill at measuring soap is at fault; it’s the appliance. When your computer stops working, diagnosis is more uncertain; the problem could be related to hardw are, software, connectivity, or to a myriad of actions that you took or failed to take. Despite the widespread dissemination of computer technology in workplaces and households, computers still pose mysterious technical barriers to their users. The popu larity of computer advice columns and training courses and the growth of the usability engineering industry attest to the complexities people encounter when they use computer technology. Even skilled technical workers encounter usability problems when th ey try to learn new systems and programs (e.g., Barley, 1988; Orlikowski, 1996).

Vendors and service providers have established support lines and help desks to answer people’s questions, but requesting outside help may require a degree of competence and commitment that many people do not have. If people who lack these attributes f ail to ask for help, they lose not just immediate solutions to their difficulties but also the chance to increase their competence through interaction with experts. Here we explore this argument by examining requests for external computer support in a sa mple of families in their first year of using the Internet and comparing those who asked for help with those who did not. Our analyses draw from a popular theory of responses to organizational dissatisfaction (Hirschman, 1970) in an important technologic al domain, home computing, and extend this theory to understand patterns and consequences of requesting external technical support in families. In brief, Hirschman’s theory predicts that people’s commitment to a product or service increases the likelihoo d they will exert "voice"—complain or request help. We argue that in the technological domain, technical competence and family role as well as commitment will predict whether people request external support, and that these requests, in turn, re inforce and lead to future increases in commitment and competence.

The data for our analyses derive from a longitudinal field study of 93 Pittsburgh families whom we provided in 1995 and 1996 with a computer and free access to the Internet. The initial goal of "HomeNet" was to examine residential Intern et use and its social impact in an environment in which economic and technical barriers to Internet access would be reduced. However, our data showed economic barriers to be far more malleable than technical barriers. That is, whereas household income d id not predict Internet use, most families participating in the study could neither set up their computer nor connect to the Internet without help. Even though their computers were made "Internet ready" and at least two family members had atten ded a training session, over 70% of families required personal assistance before they could connect to the Internet (Kraut, Scherlis, Mukhopadhyay, Manning, & Kiesler, 1996). However, many participants who encountered technical problems during the ye ar never requested help. What would cause people not to use free technical support? We found that low pre-trial self-reports on a measure of computer competence predicted both low average Internet usage and premature dropping of Internet service in the first year. Perhaps those who lacked competence also lacked sufficient confidence to articulate their problems or to manage a technician’s help. We were led by these questions to conduct the current analyses, in which we systematically investigated the attributes that led some people and not others to request help and the cognitive and behavioral consequences of their doing so.

Our analyses were guided by an extensive literature evaluating the social process of complaining and requesting help, much of it influenced by A. O. Hirschman’s (1970) monograph, Exit, Voice, and Loyalty: Responses to Decline in Firms, O rganizations, and States. Hirschman argued that consumers might force improvements in poor product or service quality through "voice" (i.e., complaining or requesting help) or "exit" (i.e., abandoning the product or service). Alt hough exit and voice can both occur (Graham and Keely, 1992; Hirschman, 1981, p. 212), in Hirschman’s theory, voice operates (at least for a time) as a substitute for exit and should be negatively associated with exit most of the time. This argument has been applied to consumer-firm relations (e.g., Kolarska & Aldrich, 1980; Mante & Forrester, 1993), employee relations (e.g., Hodson, 1997; Leck & Saunders, 1992; Lee & Jablin, 1992), citizen responses to government services (e.g., Lyons &a mp; Lowery, 1986, 1989), migration processes (e.g., Brubaker, 1990), economic gender relations (e.g., Hobson, 1990; Hoyman, 1987), and marital or dating relationships (e.g., Castles & Seddon, 1988; Drigotas, Whitney, & Rusbult, 1995). What makes the theory so appealing is its usefulness for "understanding how individuals may act when things are not going well" (Withey and Cooper, 1989, p. 52). The voice-exit framework has obvious relevance to computer technology, whose complexity frequ ently leads to things "not going well."

Hirschman introduced the idea of loyalty to explain why a person would exert voice rather than exit (1970, p. 78). Loyalty is a form of psychological commitment, a concept that has been operationalized in two ways. First, people can be attitudinally committed to a product or service by virtue of their valuing it highly or feeling attached to it. Second, people can be behaviorally committed to a product or service by virtue of their using it frequently or in preference to its alternatives (Graham &am p; Keely, 1992). Hirschman argued that commitment to a product or service can raise the cost of exit by changing its meaning from a rational response to dissatisfaction to one of desertion (1970, p. 98). Hence whereas requesting help or complaining is b y definition an expression of dissatisfaction, ironically, requesting help tends to reflect commitment. In several studies, voice was negatively correlated or uncorrelated with product dissatisfaction (Iverson and Currivan, 1997; Leck & Saunders, 199 2). Instead both attitudinal and behavioral commitment were found to be positively correlated with voice and negatively correlated with exit (e.g., Iverson and Currivan, 1997; Leck & Saunders, 1992; Withey & Cooper, 1989).

The literature on Hirschman’s theory suggests that the level of technical problem or difficulty experienced with technology may not predict requests for external technical support. An additional complication in the technological domain is that people exploring new software or services may call help desks or support lines not to fix something broken, but to perform a new task they have chosen to explore, e.g., learning how to create their own Web page. In the latter situations, they may consider their communications with a help desk to be a matter of gaining know-how or adding value to the product or service rather than one of repairing product or service deficiencies. For these reasons, we did not predict that subjective dissatisfaction or the serio usness of computer problems would predict requests to the help desk in this study. However, we attempted to categorize participants’ computer problems from the logs maintained by the help desk, and we compared the difficulties of more and less technicall y competent and loyal persons.

When people use a complex technology such as a computer connected to the Internet, they have to care enough about using it to take on the costs of requesting external support, including the risk that these requests will not result in fixing their probl em. Calling a help desk or support line also can incur costs such as having to wait for return calls, feeling inadequate to describe the problem, feeling embarrassed by one’s mistakes, obtaining confusing or insufficient help, and not being able to judge what should constitute satisfactory service. Hence, Hirschman’s theory about the relationships between commitment and voice would seem to hold true for computer technology products and services. That is, people who have more commitm ent to using their technology should be more likely to request technical support.

It also seems likely that people’s initial technical skill and confidence will be associated with the likelihood of their requesting support. Competence should reduce the uncertainty and anxiety surrounding requests for help and increase the benefits of receiving help because know-how can accrue best to people who begin with more understanding. In the context of HomeNet, technical competence as well as commitment should describe those who have exerted voice by requesting help desk support. These der ivations from the voice-exit literature led us to formulate the following initial hypotheses.

H1: Commitment, operationalized as favorable computer and Internet attitudes and as more frequent use of the Internet, and pre-existing technical competence will predict requests to the help desk.

H2: Commitment and pre-existing technical competence will negatively predict stopping Internet service.

Because of their cost and protean functionality, home computers and online Internet services are likely to be viewed by many people as a shared household or family resource much like the family washing machine, refrigerator, or cable service. In t he HomeNet study, the investigators provided a computer and Internet service to the entire family; in almost every case, multiple members of the household used the computer and tried Internet service. When more than one member shared the computer, family members would be able to provide each other with help and encouragement and help one another call the HomeNet help desk if they had difficulty solving problems themselves. However, we expected that just as families organize divisions of labor for other domestic tasks such as paying bills or getting the car repaired, they would exhibit role specialization in helping one another and obtaining outside computer support. It seemed reasonable to expect family members to turn to the most committed and technic ally competent family member for computer help and to rely on this person to mediate outside technical help; that family member, in turn, would become increasingly more proficient in managing outside help so the division of labor would be self-reinforcing .

In sum, we expected that in many households a committed and technically competent family member would take the dual roles of household computer guru and the boundary spanner who would obtain outside technical support. We also expected consequences of this role specialization both for the specialized member and for the others in the member’s family. Voice in the context of computer help involves an exchange of information and, perhaps, know-how. That is, not only can requests for help improve the &qu ot;product" as Hirschman envisioned the process, but also it can improve the knowledge of the person asking for help. Hence we predicted that the frequency with which a family member exerted voice would positively influence his or her own loyalty an d competence. We also expected that there might be a secondary and positive "spillover" of that family member’s outcomes for other family members. That is, we hypothesized that the frequency with which a competent family member requested exter nal technical help might positively influence the commitment and competence of others in the family. We posed these ideas in the following hypotheses.

H3: Within households, help desk requests will be concentrated rather than distributed across many family members.

H4: Technical competence and commitment will predict those in the household who more frequently contact the help desk.

H5: Those in the household who contact the help desk will increase their commitment and gain further technical competence.

H6: Those who do not contact the help desk in the household will gain commitment and technical competence if someone else in the household contacts the help desk.

The HomeNet study began with a sample of 44 families from four neighborhoods in Pittsburgh. People in these families began using a computer and the Internet at home in early spring, 1995. A second sample of 49 families began approximately a year later. Within these 93 families, 237 members signed consent forms, were given email accounts on the Internet, and logged on at least once. Babies, children younger than 10, and uninterested members of the household are not in the sample. The participan ts were followed over time; their Internet usage was monitored continuously, and they completed questionnaires before, during, and after one year.

Each year’s sample was drawn from four school or neighborhood groups so that the participants would have some pre-existing communication and information interests in common to get them started using email. The first sample consisted of families with t eenagers in four area high schools. The common entry point to these families was the journalism programs at the four high schools. The journalism students were asked if their families would like to join the study. Anyone in the household over the age o f 10 could have an Internet account and participate in the study, providing that at least one adult and one teenager in the household agreed to join the study. For the second sample, members of the boards of directors of four community development organi zations in the Pittsburgh area were invited to add their families to the study. The community activists were not required to have a second family member join the study but nearly all of them did.

Except for the larger family size and greater number of teenagers in the first sample, the demographic characteristics of the two samples were similar. Their median household income at the start of the study was $42,500, somewhat higher than the natio nal average but lower than the income of households connected to the Internet nationally (Anderson, Bikson, Law, and Mitchell, 1995). About 25% of the households were minority (mainly African-American). Approximately 60% percent of the participating per sons were 19 years old or older and 51% were female.

None of the groups approached about the study declined the invitation, and over 90 percent of the families contacted within each group agreed to participate. Because the recruitment plan excluded households or persons with active Internet connections, the data represent people’s first experiences with Internet use, and for all but a few of the households, their first experience with a powerful home computer. Twenty-five percent had used a computer at work.

Families received a free extra telephone line and free access to the Internet in their homes in exchange for permitting the researchers to automatically track their Internet usage and services, for answering periodic mail questionnaires, and for agreei ng to an in-home interview. The equipment provided to participant households included Macintosh computers with modems and software installed for ready Internet access. The families used Carnegie Mellon University’s proprietary software for electronic ma il, MacMail II, Netscape Navigator for web browsing, and ClarisWorks Office. Each family also received a family Web page with links to some of their pre-existing interests and hobbies. At least two family members also received a morning’s training in th e use of the computer, email, and the Web.

The families were also invited to telephone a dedicated help desk line or send email to the help desk should they encounter any difficulties. Help desk staff were Carnegie Mellon undergraduates in computer science or information systems with computer expertise and training in teaching basic computer skills who were outfitted with beepers so that they could return calls during their work hours. The help desk staff used a program called Timbuktu for diagnosing problems, which allowed them to view the c omputer display of the participant remotely. When necessary, a help desk staff person visited participants’ homes to fix problems. The project’s technical director, who also answered help requests and made home repair visits, was responsible for hiring and supervising the help desk staff.

Our analyses focus on Internet service, whose use by each individual we could measure reliably, rather than general use of the computer, whose use we could not measure reliably. The data on which our analyses are based consist of automated Interne t usage statistics, questionnaires, help desk logs, and 25 home interviews.

Help desk requests. Help desk staff completed a form for each telephone conversation they handled and saved an archive of email messages received and responded to by the HomeNet help desk. From these request logs, we tallied all the telephone or email contacts each person and family made with the help desk within the first 52 weeks of their household’s start date. We then counted the frequency of requests made by each person. We also tallied the number of requests within each household and co mputed for each person the number of other family members who requested help. Because help desk requests were skewed, we used logged transformations of the data in our analyses.

Help desk requests can be considered an operational definition of voice in the pursuit of Internet service, commitment to which was expected to predict voice. However rather than include in our operationalization only requests explicitly concerning th e Internet, we included requests for any technical computer and Internet support. That is, we counted complaints about household phone lines to the computer, computer equipment including printers, and software, as well as Internet service. We had two re asons for doing so. First, many people could not diagnose whether their technical problems concerned the Internet or the computer. Second, it was not possible to use the Internet if the computer did not work, so all help requests could affect Internet s ervice.

Dropped Internet service. Computer-generated usage data were collected by setting automatic probes on each person’s Internet account that identified each time the person logged on or off and the total time they were connected using email, the W eb, or other Internet services. To increase reliability, the computer-generated usage data were summed each week. From these automated logs of Internet connect hours, we created an operationalization of "exit," a dichotomous variable based on the last week that participants used the Internet. Because participants did not need to log on to the Internet every week to be considered active, we needed to correct for right-censoring of the data. Therefore, the "exit" or dropped service variable was based upon a family’s first 48 weeks of project involvement. (That is, anyone who had logged on in Week 48 but not between Weeks 49-52 was judged still active because being coded as having dropped service required at least 5 weeks of inactiv ity.) We created two companion variables. Early dropouts were those whose last login fell within the first 26 weeks of the trial. Late dropouts were those whose last login fell within weeks 27-48.

Behavioral commitment: Connect hours. This variable is an average based upon the total hours each week that participants were connected to the Internet during the period when they were still active. In other words, this variable reflects usage only up until the time a participant dropped service. We also created two companion variables to distinguish average weekly connect hours in the first and last 26 weeks of the year. The latter variable necessarily reduced the sample size because it did not include anyone who was an early dropout.

Attitudinal commitment: Computer value. All participants completed a pretest questionnaire before they received their computers and Internet service, and a posttest questionnaire after a year had gone by. (The first sample of families received the posttest 1 1/2 years later than the pretest; the second sample received the posttest 1 year later.) On these questionnaires, a 12 item scale using 5-point Likert ratings asked participants, "How useful do you think a home computer is for each o f the following purposes?" Participants rated the value of the computer for getting paid work done, getting school work done, getting personal work done (e.g., bills, taxes, writing), shopping, entertainment, having fun, keeping in touch with friend s, keeping in touch with family, keeping in touch with coworkers or schoolmates, learning more about interests and hobbies, learning new things, and providing information to others (Cronbach’s alpha = .85). Based on our pretesting, the scale did not spec ifically distinguish "computer" from "Internet" functions because many participants confused the terminology (e.g., some thought the Web but not email was an Internet function).

Computer competence. A 5 item scale using 5-point Likert ratings on the pretest and posttest asked participants how much they agreed with the following statements: I am very skilled at using computers; I use computers almost every day; I am afr aid of using a computer; using computers is fun; I don’t know much about using computers (Cronbach’s alpha = .85).

Session difficulty was the average response to a question that was repeated on short questionnaires sent three times after the study began. Participants were asked to recall the last time they used a computer, to describe what they were doing d uring the session, and to rate on a scale from 1 (none or little) to 5 (a lot) the overall level of difficulty they had during that session.

The two sample groups described above, the families who started in 1995 (44 families, 136 participants) and those who started in 1996 (49 families, 101 participants), differed significantly on two of the key variables in our analyses. The 1995 sample used the Internet twice as much on average as the 1996 sample (p < .01) and requested help more frequently (p < .01), and this difference remained even when we restricted the analysis to adults. However, the samples did not show different patterns in the inferential statistical analyses we conducted to test our hypotheses. Therefore, we report findings for the combined samples. To analyze data from the combined sample, we examined the pretest and posttest questionnaires and the first 52 weeks of usage data and help desk logs for each family.

All but a few participating families experienced difficulties using their computer. These problems were discovered through logs of calls to the help desk and all 25 home interviews. (During the interviews, we asked each participant in the family to sit at the computer and show us "how you use your computer on a typical day, starting with turning it on. " There were no interviews in which we did not observe a major usability problem.) Table 1 illustrates the variety of problems particip ants experienced and the causes eventually uncovered by help desk staff or by in-home interviewers. Problems noted early in the trial included difficulties installing phone service, configuring the telecommunication software, and receiving busy signals u pon dial-up. Because participants were inexperienced with terminology and concepts such as client-server, mice, keyboards, scroll bars, radio buttons, menus, and software, often they had no idea how to diagnose the source of their problems. For example, one participant thought her internal hard drive had disappeared when her granddaughter accidentally deleted the label on the drive icon. During the home interview, she said she didn’t call the help desk because she was embarrassed about breaking her co mputer. Another participant who had trouble with his modem didn’t realize it was part of the system necessary to connect with the Internet.

Table 1. Examples of participants’ computer and Internet problems reported to the help desk

After the first 6 months, problems declined markedly but did not cease. Problems logged later included difficulties downloading programs from the Web, computers "freezing," and minor but irritating problems such as mice that no longer worked . For example, participants had trouble finding people and sites on the Web, using email and their "address book," making downloaded software work, or recovering from errors when they deleted crucial system files. Problems often had emotional as well as practical ramifications. In the home interviews, respondents who were demonstrating their use of the Internet typically expressed frustration or resignation when their problems were revealed. Only one responded with humor: a grandfather rema rked that the error message that greeted his family each time any of them logged on had become "an old friend."

To further inform our thinking about the problems participants encountered, we devised two qualitative coding schemes to categorize problems from the written material in the help desk logs and the email messages. In the first scheme, we divided the pr oblems into groups such as hardware, software, or Internet problems, but this method proved unreliable (e.g., "my computer freezes" might be a problem caused by software and remote servers). In the second scheme, we divided the help requests in to three ordinal categories: novice questions (such as "how do I set up my computer," or "I can’t get connected"), intermediate questions (such as "how do I find so-and-so’s email address"), and expert questions (such as &qu ot;I’m trying to make this Web page, and want to know what you can tell me about such-and-such new software"). This coding scheme resulted in low reliability coefficients (.51 for telephone requests and .78 for email requests). However, email reque sts were coded reliably as more sophisticated than telephone requests.

Participants’ problem category did not predict whether they called the help desk. We also found correlations lower than r = .20 between calling the help desk and questionnaire responses to an item asking about the level of difficulty participants had in their last session with a computer. These observations accord with the research cited earlier in which the experience of dissatisfaction was not positively associated with voice. In the technological environment, the lack of association between r equests for technical support and dissatisfaction could be due in large part to the mix of people who call the help desk. The callers are not only dissatisfied participants but also highly satisfied participants who are helping other family members with their problems or obtaining "training" for themselves to do interesting new computer or Internet tasks. In any case, because of these preliminary analyses, we dropped the measures of dissatisfaction from further analyses.

Tables 2 and 3 provide descriptive statistics on the key variables used in our hypothesis tests. Table 2 shows that approximately half of the individuals in the sample made no contact with the help desk. About 25% made 1-3 requests for help, and anot her 25% made more than 4 requests. (The highest rates were reached by two individuals who made more than 30 help requests each within the first 52 weeks of participation.) Making requests by telephone was the preferred method by about 2-to-1; 103 partic ipants contacted the help desk by telephone whereas only 47 did by computer using email. Table 3 shows that help desk requests declined in the second half of the year, that behavioral commitment (Internet connect hours) remained stable at nearly 3 hours a week per person (not including those who dropped service), and that both competence and attitudinal commitment (computer value) increased from before the study to the end of the first year.

Bivariate correlations we performed on these data showed that the demographic characteristics of participants measured in the study had no direct association with their frequency of contacting the help desk but did have several associations with other key variables. Therefore, demographic variables were included as control variables in the analyses used to test our hypotheses. In accordance with hypothesis 1, help desk requests were positively associated with pretest computer competence (r = .14, p < .05), Internet connect hours (r = .36, p < .01), and pretest computer value (r = .13, p < .05).

However, having other family members make more help requests was negatively associated with a person’s own help requests (r = -.16, p < .05). We predicted this outcome as a result of role specialization. More surprising, having other family members make more help requests overall was associated modestly but significantly with increased likelihood of dropping Internet service in the second 6 months (r = .14, p < .05), and help desk requests by other family members in the first 6 months al so predicted dropping service in the second 6 months (r = .20, p < .01). These data suggest two possibilities. One possibility is that when someone else called the help desk, it discouraged others’ continued use. For instance, one could imagine in s ome families that the most expert family member monopolizes the computer or acquires so much expertise that others are scared off. However, another possibility is that when a family member was very discouraged and ready to quit, someone else called the h elp desk more frequently. To pursue these ideas we included the number of requests for help made by others in the family in all individual-level analyses and later examined patterns of help desk requests within families.

Table 2. Proportion of participants who made requests for help desk support and who prematurely stopped Internet service during the first year of service (individual level)

1Only 13 of this group requested help by email only.

Table 3. Mean frequency of help desk requests, commitment, and computer competence (individual level)

To review, we measured the number of times a participant consulted with the HomeNet help desk for assistance. Table 4 shows a series of regressions testing hypothesis 1, that behavioral and attitudinal commitment to computers and the Internet and technical competence will predict the frequency of requests to the help desk. Since the frequency of help requests and connect hours were skewed variables, the analysis uses the log of the number of requests and the log of connect hours. The table prese nts results separately for three periods—the first 6-month period, the second 6-month period, and the whole year.

The model in the first column for each period regresses the demographic control variables on help requests and shows that demographic variables alone did not predict the number of requests. The model in the second column for each period includes the h ypothesized predictors of help desk requests, that is, pretest computer value, pretest computer competence, and the number of others in the family who requested help from the help desk. This analysis shows that pretest computer competence predicted help requests in weeks 27-52, pretest computer value predicted help requests during the year marginally, and help requests by other family members was a highly significant negative predictor of a person’s own help requests during weeks 1-26. The last finding, that a family member was less likely to contact the help desk if someone else in the family did, bears on our role specialization hypothesis, discussed later.

Table 4. Standardized regression coefficients predicting requests for help desk support (logged)

1The analysis in the middle panel excludes the 30 participants who dropped out early (before Week 27).

Behavioral commitment, that is, connect hours, shown in the third column for each period, was by far the most important predictor of help desk requests. That is, the number of actual Internet connect hours was a highly significant predictor of a person’s help requests. This finding supports Hirschman’s theory that commitment or loyalty increases the probability of voice.

We predicted in hypothesis 2 that commitment and technical competence would negatively predict prematurely stopping Internet service. We defined stopping service as a dichotomous variable, occurring if a participant discontinued use of the Interne t during the first year of service. This analysis, a logistic regression predicting which of the 20% of participants who dropped service in the first 48 weeks, is shown in Table 5. Because the measure of dropping service was related to the measurement o f connect hours (r = -.23, p < .01), we used connect hours in the first 6 months to predict not only stopping Internet service in the first 6 months, but also stopping Internet service in the second 6 months and over the year. Hence the middle set of columns is a lagged model testing whether earlier low connect hours (i.e., low behavioral commitment) predicts dropping service later. The models show that this prediction was supported. However, pretest computer competence and computer value did not si gnificantly predict dropping Internet service.

Other significant influences on dropping service aside from connect hours were race (marginally), income, and the number of help requests made by the participant. But the number of help requests was not found to be important when connect hours in the first 6 months was included as a variable. Hence, these results suggest that, as with our findings for help requests, the predictor variable of most importance in dropping service was behavioral commitment—actual use of the Internet.

Table 5. Odds ratios for logistic regressions predicting dropping Internet service in the first year

.

Computer competence, computer values, and connect hours were correlated positively within families. Families were led by the most competent and committed family member. Excluding the 18 families with only one study participant (leaving 75 familie s in the analysis), the more competence the most computer competent family member possessed, the higher was the average competence in the rest of the family (r = .51, p < .01). Computer values correlated positively within families but to a lesser exte nt than computer competence. The more value the most attitudinally committed family member placed upon the computer and the Internet, the higher was the average computer value in the rest of the family (r = .29, p < .05). Further, the higher the usag e of the highest Internet user in the family, the higher was the average usage of the rest of the family (r = .40, p < .01). These findings point to the reason why help desk requests (which predominated among highly competent and committed participant s) predicted other family members’ dropping service. That is, given the correlations cited above, it seems far more plausible that problems that led ultimately to another family member’s dropping Internet service caused increased help desk calls by the m ost competent and committed member rather than vice versa.

We predicted in hypothesis 3 that contacts with the help desk would be concentrated within households, and, in hypothesis 4, that the person or persons with more computer competence and commitment within households would take the roles of guru and boun dary spanner. In support of these hypotheses, help desk requests were highly concentrated. Help requests by one family member were negatively associated with help requests by others (r = -.16, p < .05). In the majority of families, one person made a lmost all contacts with the help desk. In families with 2 participants (n = 34), the mean number who requested help from the help desk was 1.2; in families with 3 participants (n = 23), the mean number who requested help from the help desk was 1.4; in fa milies with 4 participants (n = 10), the mean number who requested help from the help desk was 1.6; in families with 5 participants (n = 6), the mean number who requested help from the help desk was 1.5; and in the 2 families with 6 participants the mean number who requested help at all was 2. These data support our argument that the boundary spanning role would be concentrated in households. Calling the help desk was adopted as a role by one or two members of these households.

Table 6 shows the distribution of role behavior in more detail. We divided each household with more than one participant into three categories: (a) the family member who made the most requests, (b) the family member who made the next most number of he lp requests (if any), and (c) every other participant in the family. The data in this table indicate that the dominant role of requesting external support, on average, was taken by the family member with the highest commitment (both behavioral and attitu dinal) within the family (hypothesis 4). This finding only partially supports the predictions in hypothesis 4 because initial computer competence was not significantly different between top help requesters and all remaining family members.

Table 6. Mean help desk requests, pretest computer competence, and commitment by within-family status as requester of help desk support

The last rows of Table 6 address the question of whether the boundary spanner who made calls to the help desk was also the person within the family who functioned as the computer guru. These data come from a questionnaire administered to a subset of p articipants in May 1997. We asked this group if anyone in the household gave them computer help. The table shows the number of family members that these participants enumerated. Most family members asked only one other family member for help, and the p erson who was most likely to call the help desk was much less likely to receive help from other family members. These data suggest that the roles of guru and boundary spanner were likely to be taken by the same person.

A somewhat different perspective on the data can be gained by evaluating behavioral commitment (Internet connect hours) as a dependent variable. Hirschman’s theory implies that voice can lead to an improved product or service; in the current situ ation, help desk requests should lower technical barriers to using the computer and Internet services. However, there may be other psychological effects. We argued that asking for external support not only provides technical solutions but may result in enhanced feelings of effectiveness and more competence, which may in turn increase future commitment. In Table 7, we show an analysis exploring this argument. The equations of most interest are shown in the middle panel, predicting connect hours in the second 6 months. Note first that teenagers and children (persons age 10-18) in the study had much higher usage (Kraut et al., 1996). Also, connect hours in the first 6 months were a strong predictor of connect hours in the second 6 months. Finally, the frequency of help desk requests significantly predicted connect hours in the second 6 months, even controlling for connect hours in the first 6 months. This finding supports hypothesis 5, that help desk requests will increase behavioral commitment for t hose in the household who made these requests. Help desk requests by other family members had no direct effects on a given family member’s commitment.

Table 7. Standardized coefficients predicting average weekly Internet connect hours (logged)

1The analysis in the middle panel controls for average weekly connect hours during the prior period (Weeks 1 to 26.)

The argument that requests for external technical help can have personal impact is further explored in Table 8, where we examine the effects of help desk requests on computer competence and attitudes. Although these analyses are weakened by the fact t hat some participants failed to complete a posttest questionnaire or left the study (e.g., went to college), the data suggest that the strongest impact on posttest competence was pretest competence, pretest computer values (attitudinal commitment), and ac tual connect hours (behavioral commitment). The strongest impact on posttest computer values were pretest computer values and connect hours. When these factors were included, help desk requests exerted no independent effect on competence or computer val ues. Help requests made in the family did not improve other family members’ posttest competence or commitment as we proposed in hypothesis 6. Our investigations of demographic effects on competence and attitude changes, however, found some evidence of "catch up" effects among females and, significantly, among minorities, whose mean competence and attitude scores at the posttest equaled or nearly equaled the scores of males and whites.

Table 8. Standardized coefficients predicting posttest computer competence

Note. This analysis includes all participants who completed a posttest. Due to the participants who did not complete a posttest (e.g., moved), there are 41 missing cases in the computer competence analysis (20 were dropouts and 21 were not), and ther e are 56 missing cases in the computer value analysis (21 were dropouts and 35 were not). In a few cases, e.g., when participants joined the study late after reaching the age of 10, their pretest measure was taken late or had to be imputed from the means for their race, gender, and age group.

+p<.10 *p<.05 **p<.01 ***p<.001

Our analyses suggest that existing theory and research on voice, exit, and loyalty have some application in the technological domain. However the psychology of voice, we think, differs when the product, service, or organization has a technical compone nt. Unlike most consumer products and services, computers require technical competence and know-how. Insufficient initial competence can discourage people from obtaining experience, which leads to commitment and asking for help, and can cause them to gi ve up using computers and to drop online services.

Our data suggest competence, commitment, and help requests are closely tied. Competence predicts future commitment, and commitment predicts help requests. Help requests appear to be inextricably connected with commitment rather than a behavior that o ccurs independent of commitment. We think this is especially true within households for several reasons: (1) More experience with the home computer and the Internet leads to a person’s encountering more problems and to the person’s initiating new challe nges or goals that need technical support; (2) more experience leads to a person’s taking responsibility for other family members’ problems, and (3) experience increases technical competence, which in turn increases a person’s confidence in managing commu nications with the help desk staff. From our home interviews, we surmised that for both participants and the help desk staff, talking with one another was like encountering an alien culture (see also Sproull, Kiesler, & Zubrow, 1984). Our interviews suggest that computer competence gives people sufficient grounding to use computers and the Internet, and that their experience along with their skill permits them to diagnose symptoms, articulate questions, and have the confidence to call and interact w ith support staff.

A second major finding of our study is that voice within households reflected a socio-technical structure consisting of a computer guru (or two) who also played the role of boundary spanner in communicating with the help desk. The household role of te chnical expert or guru, while informal, seemingly mirrors the centralization of computer expertise into information systems or internal IT (Information Technology) support departments in work organizations. Of course, centralization of technical support has its good and bad points. One of its bad points is that it tends to result in technical dependency by those who, without the expert, might solve problems on their own. In this vein, we found that the help desk interactions of those who took on the r ole of expert did not always positively aid the future commitment of others in the household and indeed there was a tendency in the data suggesting increased help desk requests by the guru did not stop inexpert family members from dropping service. It is possible that the guru wrested control of the machine and monopolized the computer, discouraging those with the least confidence and commitment from continuing.

In spite of these problems, the major influence of the most technical family member seems to have been highly positive. Commitment and competence were positively correlated within households with the commitment and competence of the most technical fam ily member, suggesting that the enthusiasm, modeling, or direct help of this family "guru" encouraged others in the family to use the computer and Internet more. It might be argued that a common causal factor (such as prior experience with tech nology) could have influenced everyone in the family; however, our time series (lagged) results tend to refute that argument.

Our study has several limitations that could be addressed in future research. First, we did not observe and document how participants tried to solve particular computer problems in their homes, such as trying the help desk rather than working out the problem on their own or asking a family member for help. An ethnographic study involving observations of family members in their homes, a study requiring participants to keep diaries, or a "beeper study" would allow better linking of the ki nds of complaints or dissatisfaction that lead to different problem solving strategies, including requests for external support.

Another limitation of the study is that we did not monitor and document the results of help desk calls, for instance, whether these calls led to satisfaction or to more complaints. For instance, we do not know if repeat calls occurred because particip ants’ problems were not resolved by the first request. More frequent requests for external support (exerted by those with more competence and commitment) might be a consequence of higher standards for service or the pursuit of an unsolved problem or both . Computer support services often organize staff into groups, with the best staff assigned to speak with those who escalate their complaints (Berselli, 1997). Our data suggest that this kind of organization makes sense, in that those who contacted the h elp desk more frequently also were likely to be more sophisticated users.

One implication of our results is that the existing individual-level theory of voice, exit, and loyalty probably predicts well for the person with highest commitment and competence in the family or household but less well for others in the family u nit. Modifying the theory to predict at the group level may require that we consider voice as a role or responsibility that can accrue to one group member. The most technical person’s taking responsibility for requesting help from the help desk may rend er others’ calling unnecessary and less likely. With each call, the most technical member would have gained confidence and familiarity with the help desk staff, while the others in the family would have felt more removed from this kind of interaction, in creasing their costs of contacting external support services.

Our analysis has practical importance for policymakers and for suppliers of Internet and computer services. For policymakers, our study indicates that simply providing people the wherewithal to access the Internet will not assure that people actually use this access. Computer expertise and confidence still seems critical to using the Internet. For computer companies, our data suggest that support or help-line data cannot be relied upon as good evidence of the distribution of usability success or usa bility problems among customers. The problems of those who make demands of help desks and support services may differ both in quantity and quality from those who do not make these requests. Despite the existence of numerous humorous deprecatory stories about customers’ calls to support staff, in actuality these callers are likely to be more technically sophisticated and more interested in the answers than those who do not call. Furthermore, our data suggest that not calling may lead people to limit the ir technical experience or commitment and curtail their acquisition of computer skills.

References

Barley, S. R. (1988). The social construction of a machine: Ritual, superstition, magical thinking and other pragmatic responses to running a CT scanner (pp. 497-539). In M. Lock & D. R. Gordon (eds.), Biomedicine Examined, NY: Kluwer Acad emic Publishers.

Berselli, B., (1997) When ‘Help gives way to ‘HELP!’ Top-level techies pick up the line. Washington Post, A1, A7.

Brubaker, R. (1990). Frontier theses: Exit, voice, and loyalty in East Germany. Migration World Magazine, 18, 12-17.

Castles, F. C.. & Seddon, E. (1988). Towards an organizational model of marital instability. Australian Journal of Social Issues, 23(2), 113-127.

Drigotas, S. M.., Whitney, G. A., & Rusbult, C. E.. (1995). On the peculiarities of loyalty: A diary study of responses to dissatisfaction in everyday life. Personality and Social Psychology Bulletin, 21, 596-609.

Graham, J. W., & Keeley, M. (1992). Hirschman’s loyalty construct. Employee Responsibilities and Rights Journal, 5, (191-200

Hirschman, A. O. (1970). Exit, voice, and loyalty: Response to decline in firms, organizations, and states. Cambridge, MA: Harvard University Press.

Hirschman, A. O. (1981). Essays in trespassing: Economics to politics and beyond. Cambridge, MA: Cambridge University Press.

Hobson, B. (1990). No exit, no voice: Women’s economic dependency and the welfare state. Acta Sociologica, 33(39), 235-250.

Hodson, R. (1997). Individual voice on the shop floor: The role of unions. Social Forces, 75, 1183-1212.

Hoyman, M. (1987). Female participation in the informal economy: A neglected issue. Annals of the American Academy of Political and Social Science, Sept., 64-82.

Iverson, R. D., & Currivan, D. B. (1997). Union participation, job satisfaction, and employee turnover: An event history analysis of the exit-voice model. Paper presented at the 1997 American Sociological Association annual meetings, Toronto, A ugust.

Kolarska, L. & Aldrich, H. (1980). Exit, voice, and silence: Consumers’ and managers’ responses to organizational decline. Organization Studies, 1, 41-58.

Kraut, R., Scherlis, W., Mukhopadhyay, T., Manning, J., & Kiesler, S. (1996). The HomeNet field trial of residential Internet services. Communications of the ACM, 39, 55-65.

Leck, J. D., & Saunders, D. M. (1992). Hirschman’s loyalty: Attitude or behavior? Employee Responsibilities and Rights Journal, 5, 219-230.

Lee, J. & Jablin, F. M. (1992). A cross-cultural investigation of exit, voice, loyalty and neglect as responses to dissatisfying job conditions. Journal of Business Communication, 29(3), 203-228.

Lyons, W. E., & Lowery, D. (1986). The organization of political space and citizen responses to dissatisfaction in urban communities: An integrative model. Journal of Politics, 48, 321-346.

Lyons, W. E., & Lowery, D. (1989). Citizen responses to dissatisfaction in urban communities: A partial test of a general model. Journal of Politics, 51, 841-868.

Maute, M. F.. & Forrester, W. R. (1993). The structure and determinants of consumer complaint intentions and behavior. Journal of Economic Psychology, 14(2), 219-247.

Orlikowski, W. J. (1996). Improvising organizational transformation over time: A situated change perspective. Information Systems Research, 7, 63-92.

Sproull, L., Kiesler, S., & Zubrow, D. (1984). Encountering an alien culture. Journal of Social Issues, 40, 31-48.

Withey, M. I. J., & Cooper, W. H. (1989). Predicting exit, voice, loyalty, and neglect. Administrative Science Quarterly, 34, 521-539.

Footnotes