The Kindness of Strangers: The Usefulness of Electronic Weak Ties for Technical Advice

People use weak ties—relationships with acquaintances or strangers—to seek help unavailable from friends or colleagues. Yet in the absence of personal relationships or the expectation of direct reciprocity, help from weak ties might not be forthcoming or could be of low quality. We examined the practice of distant employees (strangers) exchanging technical advice through a large organizational computer network. A survey of advice seekers and those who replied was conducted to test hypotheses about the viability and usefulness of such electronic weak tie exchanges. Theories of organizational motivation suggest that positive regard for the larger organization can substitute for direct incentives or personal relationships in motivating people to help others. Theories of weak ties suggest that the usefulness of this help may depend on the number of ties, the diversity of ties, or the resources of help providers. We hypothesized that, in an organizational context, the firm-specific resources and organizational motivation of people who provide advice will predict the usefulness of advice. We investigated these theories in a study of employees of a global computer manufacturer. We collected survey and observational data on the relationships between information seekers and information providers; the number, diversity, resources, and motivations of information providers, and subjective ratings of the usefulness of the advice (from both parties in the exchange) and whether or not the advice solved information seekers' problems. We found that information providers gave useful advice and solved the problems of information seekers, despite their lack of a personal connection with the seekers. The data support the main hypotheses and provide some support for resource and diversity explanations of weak tie influence. We discuss how this organization's culture sustained useful information exchange through weak ties. (Organizational Behauior, Communication, Computer Networks; Social Networks; Helping; Altruism; Organizational Citizenship; Information Exchange; Information Systems) In geographically dispersed organizations, employees cannot always get useful advice from their local colleagues. If expertise is not available locally, simply finding out who has it may be difficult. Inducing those who have it to share it may be even more difficult. People in organizations usually prefer to exchange help through strong collegial ties, which develop with physical proximity (e.g., Allen 1977, Monge et al. 1985, Kraut et al. 1988), group membership (Zurcher 1965, Crane 1971), a history of prior relationships (Krackhardt 1992, 1994a), and demographic similarity (Wagner et al. 1984, Zenger and Lawrence 1989). Depending upon unknown employees at distant locations for technical advice requires depending upon the kindness of strangers. Computer networks, which are being used by growing numbers of organizations, make it relatively easy and inexpensive to ask distant acquaintances for advice via e-mail. They also make it possible to ask strangers for advice. A person can post a query of the form, "Does anybody know... ?" to a large electronic distribution list, electronic conference, or computer bulletin board without knowing who might read it. People who read the query can reply without having .to know the person who posted it. Computer networks can link people in the absence of acquaintance, physical proximity, group membership, a history of prior relation1047-7039/96/0702/0119/$01.25 Copyright © 1996. Institute for Operations Research and the Management Sciences ORGANIZATION SCIENCE/VOL 7, No. 2, March-April 1996 119 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers ships, and demographic similarity. This paper explores the process of giving and receiving technical advice over an organizational computer network. We draw upon theories of weak ties and prosocial motivation to suggest how this process can lead to useful advice and we use data from one multinational firm to illustrate the process. Theoretical Framework and Predictions A theory of "the strength of weak ties" proposed by Granovetter (1973) suggests that relative strangers could offer an advantage over friends and colleagues in obtaining useful information. Granovetter (1973, 1982) argues that strong-tie relationships occur among people who are similar in many respects; similar people are likely to know the same things and are unlikely to know dissimilar things. When information is unavailable through strong ties, people may obtain it through weak ties: relationships characterized by absent or infrequent contact, lack of emotional closeness, and no history of reciprocal services. Weak ties serve as information bridges across cliques of strong ties and can offer people access to resources that are not found in their strong-tie relationships. Subsequent research has given some support to this "strength-of-weak-ties" hypothesis (Granovetter 1982, Brown and Reingen 1987, Stevenson and Gilly 1991, Weenig and Midden 1991). Weak tie theorists have proposed three arguments for why weak ties are useful. One argument is simply that weak ties comprise more numerous potential helpers than strong ties do (Friedkin 1982). Statistically, if weak ties are more numerous than strong ties, then calling on weak ties increases the probability that at least one contact will have useful information. If an employee broadcasts a request for technical information on a computer network, many people will see the request. Numerous replies increase the probability of finding one correct answer. If the problem is additive, that is, the solution is made up of many parts, numerous replies could increase the total usefulness of contributions. This argument leads to our first hypothesis: Advice from more people will be more useful than advice from fewer people. There are some reasons for arguing that numerous weak ties might not result in more useful advice, however. A bigger sample of weak ties is likely to increase bad advice as well as good advice. Very poor advice might be especially costly, causing confusion, uncertainty, or "information overload."Numerous weak ties, therefore, will be useful only to the extent that the benefits of more good advice outweigh the costs of more bad advice. Perhaps it is not the number of people giving advice, per se, that makes weak ties useful but the range or diversity of those ties. Burt (1983) points out that if all of a person's weak-tie contacts are themselves members of the same strong-tie group, then the expertise offered by those ties will be redundant. He suggests that it is the extent to which weak-tie contacts tap diverse groups that makes weak ties useful. A diverse range of ties increases the probability of finding a useful answer if expertise is heterogeneously distributed across groups but homogeneously distributed within groups. Group diversity also could increase alternative solutions (people offer different answers) or provide pieces to a multi-part solution (people offer partial advice that can be combined into a solution). This argument leads to our second hypothesis: Advice from more diverse ties will be more useful than advice from less diverse ties. Useful weak ties might draw upon not merely diverse resources, but superior resources. Lin, Ensel, and Vaughn (1981) argue that those using weak ties will solicit help from people having desirable resources: wealth, status, prestige, power, or access to others (see also Lin, Vaughn, and Ensel 1981, Lin 1982, DiMaggio and Mohr 1985, Lin and Dumin 1986, Marsden and Hurlbert 1988). Investigations evaluating this idea have been concerned primarily with job mobility. Referrals to attractive jobs come from people who have more seniority, higher job status, or more desirable employers. Useful weak-tie contacts therefore tend to have resources that are superior to those of seekers. In this view, the number and range of ties are important only insofar as they tap superior resources. In the context of technical advice, superior resources are ones that increase the probability of a contact's offering correct or otherwise useful advice. In an organization, a contact's personal resources and social resources derived from his or her organizational position, department, or location might be differentially helpful (e.g., Krackhardt 1992). Personal resources important in technical help might be the contact's technical expertise, industry experience, and firm experience. The more technical expertise and experience a contact has, the more likely that contact will provide useful technical advice. Resources relevant to technical help that depend on job or position are the contact's physical proximity to the technical center of the organization and the contact's hierarchical status. Physical proximity to other experts has been shown to be important for the exchange of technical. information (Allen 1977, 120 ORGANIZATION SCIENCE/VOI. 7, No. 2, March-April 1996 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers Kraut et al. 1990, Finholt 1993, Lave 1991). Thus, proximity to the technical center of the organization can be viewed as a useful resource. The relationship of hierarchical position to useful technical help is problematic. Hierarchical position is an important social resource for organizational power (Krackhardt 1990). However, power may not be relevant to technical advice. An organizational cliche is that people whose technical knowledge is outdated are promoted into management. In this view, a contact's hierarchical status would be negatively related to useful technical knpwledge. Alternatively, even though managers may no longer have useful technical information themselves, they may know who does have it. In this case, a contact's hierarchical status would be positively related to useful technical knowledge. Because we had no way of gauging which organizational resources would be most important for giving useful advice, we posed the following general resource hypothesis: Advice from people with more resources will be more useful than advice from people with fewer resources. The Problem of Motivation The usefulness of computer network help from strangers is problematic. The help seeker has no direct way of assessing the provider's reliability, expertise, possible strategic motives for misinformation, or knowledge about the seeker's situation. The seeker also has no control over the provider's incentives. The provider has littie information about the seeker and therefore may misunderstand the request for help or advice, use inappropriate assumptions in generating a response, or formulate that response using language or concepts not shared by the seeker. These difficulties should increase with the weakness of the tie, i.e., with the physical and social distance of the information provider from the seeker. Some theorists suggest that if help is offered in the absence of direct reciprocity, it may not be very useful (e.g.. Thorn and Connolly 1987). In this view strangers who could offer high quality help will find it too costly do so. Only those who have "nothing better to do" may offer assistance, which is likely to be of poor quality. Even when weak ties are potentially helpful, the motivation of strangers to help may be poor. People provide help to people they know, people they like, people who are similar, and people who have helped them (e.g., Festinger et al. 1950, Fulton et al. 1977, Kelley and Thibaut 1978, Dovidio 1984, Williamson and Clark 1989, Amato 1990, Heimer 1992, Krackhardt 1992). In personal relationships, benefactors themselves benefit from providing help, either through increasing the beneficiary's obligation to reciprocate or through receiving the beneficiary's esteem or both. Generalized requests for help over a computer network do not meet the requirements of personal connection. Why would someone respond to a request for help from a stranger when the likelihood of direct personal benefit is low? Friendship and similarity are unlikely explanations. Personal friendships are uncommon across the geographic distances spanned by computer networks (Feldman 1987). Computer networks offer few cues to make demographic similarity salient to a potential benefactor (Sproull and Kiesler 1986). Also, computer networks do not provide a very rich medium for proffering esteem and gratitude (Daft and Lengel 1986). Theories of prosocial motivation suggest two alternative processes that could lead people to provide useful technical help to strangers, even when this help is personally costly. First, some theorists have posited that people are not only pragmatic but also expressive of feelings, values, and self-identities (Bandura 1986, Schlenker 1985, Shamir 1991). If technical expertise is important in self-identity, experts can gain personal benefits from helping strangers on a computer network with technical problems. Helping others can increase self-esteem, personal identification with the organization, self-respect, respect from others, and feelings of commitment (e.g., Orr 1989). This reasoning suggests that personal benefits can lead experts to offer technical help even in the absence of personal acquaintance, similarity, or the likelihood of direct reciprocity. However, the usefulness of advice from experts motivated by personal benefit is questionable since such advice may be provided idiosyncratically or without close attention to the requirements of information seekers. A second theoretical argument is that instead of direct personal benefit, help on a computer network is founded on organizational citizenship (Bateman and Organ 1983, Brief and Motowidlo 1986) and norms of generalized reciprocity (Mauss 1967, Berkowitz and Daniels 1964, Titmuss 1971). Faced with a request for help, those who are organizationally motivated would be concerned with such things as how much they are needed, how they can be useful to others, and how their advice might solve organizational problems. People who have a strong organizational orientation are likely to be sensitive to the needs of help seekers and to adjust their advice to the requirements of those asking for help. Accordingly, we offer a fourth hypothesis: Advice from people who are more organizationally motivated will be more useful than advice from people who are less organizationally motivated. ORGANIZATION SCIENCE/VOL 7, No, 2, March-April 1996 121 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers Generalized reciprocity emerges when people have positive regard for the social system in which requests for help are embedded and show respect for it through offering help. Their regard may have an indirect basis in personal experience. For instance, they may have been helped by others on the computer network in the past or they may expect that someone on the network would help them in the future if they had a question. Alternatively, their regard might stem from a more abstract view of the computer network as an organizational resource and worthy institution. In either case offering help is unrelated to direct reciprocity and more related to maintaining the social institution of the network as an organizational resource. Hence in opposition to the prediction developed from an economic rational theory of exchange (e.g., Thorn and Connolly 1987) that help providers will be unmotivated people who have nothing better to do and not much to offer, we offer a fifth hypothesis: On average, information providers will represent a pool of people whose resources for helping are at least as good as and perhaps better than those of information seekers. Methodological Perspective Previous research on weak ties has been conducted mainly through studies using retrospective accounts of the search for a successful outcome. These studies do not provide base rate data on all the weak ties that seekers of help tried, and do not include ties that proved useless (see Nohria 1992, for one exception). Lacking within-search comparisons of individuals' useful and useless contacts, we cannot estimate the true association between weak-tie connections and the outcome of seeking help through weak ties. Further, retrospective accounts often produce "good stories" (Ross 1989). Retrospective survey research on the search process may be biased in favor of stories that inflate the usefulness of weak ties. Previous research also has not addressed the liabilities of weak-tie searches. Although weak ties potentially draw on resources that are more numerous, diverse, or better than strong ties do, they also could generate wasted effort and useless information. By collecting data from information seekers and each of their subsequent information providers, our study provides a way to estimate the overall value of search. Method Research Setting The study was conducted at Tandem Computers Incorporated, a Fortune 500 computer manufacturer whose headquarters and main technical organization are located in Silicon Valley. Tandem employs over 11,000 people worldwide in the manufacture and sales of its products. Virtually all employees use a corporate computer network, which allows employees to send and receive mail messages from computer terminals on their desks. The system is used extensively: employees from all levels of the company and from all locations feel free to (and in fact do) send messages to the president, to managers not in their direct chain of command, or to people outside their subunits; tens of thousands of messages are sent and received each day. The e-mail system at Tandem organizes messages into first-class, second-class, and third-class mail. First-class mail is for person-to-person messages and for work-related distribution lists (e.g., All_ Sales_Reps). Second-class mail is for work-related broadcast messages that go to the entire organization, including announcements from headquarters, industry news, and requests for information. Third-class mail is for extracurricular broadcast messages such as restaurant reviews or "want ads." The focus of this research is broadcast requests for information appearing on second-class mail. About 30% of the second class messages on the network contained such requests. An example of a query and responses found in second class mail is given in Appendix A. Study Design and Data Collection The study used an event-driven survey methodology, whose triggering event was a request for information broadcast in second-class mail. During the six-week research period, 82 employees broadcast one or more questions and announced they would make the replies available in public reply files on the network. When each question appeared, we sent survey questionnaires electronically to the information seeker who broadcast the question and to the information providers who replied to the question.' Information seekers received two surveys: the first asked them about themselves and requested that they keep all replies to their question; the second survey, sent one week later, asked them to evaluate each reply they had received. The survey sent to information providers asked them for information about themselves and about why they replied to the information seeker. We also captured the text of each question and all replies that had been placed in public reply files. We used Tandem's online organizational database to gather data on respondents' geographic location and hierarchical level. For purposes of comparison we also collected data from an additional 67 employees who broadcast re122 ORGANIZATION SCIENCE/VOI. 7, No. 2, March-April 1996 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers quests for information but did not create public reply files. We captured the text of their questions and sent them both surveys but we did not collect data from their providers because their providers' replies had not been made public. Independent and Dependent Variables Strength of Ties and Relationship Variables. We measured the strength of ties by asking information providers how well they knew the information seeker on a 10-point scale (1 = "don't know at all" and 10 = "know very well"). We measured network history by asking how many questions they had posed and replies they had given in the past year on second class mail. We evaluated the demographic similarity of seekers to providers using difference scores on headquarters location (1 = seeker or provider at headquarters; 0 = both or neither at headquarters), managerial position (1 = seeker or provider is a manager; 0 = both or neither), hierarchical level (difference in the number of levels from the CEO), firm experience (difference in years of experience), and industry experience (difference in years of experience). Our measures of firm and industry experience similarity are roughly equivalent to demographic similarity variables used in prior research (Zenger and Lawrence 1989). Resources. We used years of firm experience and years of experience in the computer industry as measures of resources for both seekers and providers. We estimated the expertise of information providers by asking them on a 10-point scale: "How informed are you on the subject matter of this question?"(1 = novice and 10 = expert). We also used three measures of resources associated with a person's organizational position. One was location at headquarters, the site of most engineering and product development (0 = no; 1 = yes). Another was whether or not the person held a managerial position. A third was hierarchical level, coded as lower when the employee reported through fewer levels to the CEO. We used two estimates of the range or diversity of the social groups to which information providers be-, longed. Diversity is a group-level variable, a characteristic of the group of information providers who reply to the information seeker's request for advice. The measure of group diversity should be relevant to the usefulness of technical advice. Because Tandem has field offices all over the world, we operationalized diversity for each information seeker's set of providers as the number of different countries where information providers were based. Since the firm's products and clients differ across countries, and since field offices in each country have staff who were educated in their native land, the more countries represented in a set of replies, the more those replies would be expected to reflect differences in people's resources for technical advice. We also operationalized diversity for each information seeker's providers as the number of different hierarchical levels of information providers. People at different levels often have different perspectives on a firm's approach to problems: for example, on whether a machine is worth fixing, or whether one should rely on experts at headquarters. The more hierarchical levels represented in a group of replies, the more perspectives the information seeker receives.^ Provider Motivation. We measured providers' motivations by asking them to allocate 100 points among several reasons that they might have had for replying to the information seeker. We listed four reasons associated with personal benefits and four reasons associated with organizational motivation. Usefulness and Content of Aid. To measure the usefulness of advice, we asked information seekers to "please 'award' $0 to $25 to each answer based on how helpful it was to you." We asked information providers to award themselves $0 to $25 according to how useful they thought their reply was to the information seeker. We also asked information seekers to impose a "fine" of as much as $25 on any answer that wasted their time. And we asked information seekers an open-ended question, "What have you done or what do you intend to do as a result of each reply?" We coded the responses to indicate whether one or more of the replies solved the seeker's problem (solved problem = 1, did not solve problem = 0). A coder blind to the research hypotheses analyzed the texts of replies to determine their content using the following categories: technical content, story of personal experience, pointer to a person, document, database, or customer, inclusion of a document or e-mail from others, request for clarification, and expression of interest in seeing replies from others. The first author coded a subset of the responses to check for reliability. Scott's Pi for inter-coder agreement ranged from 0.75 to 1.0, with the exception of the story category, which had a Pi of 0.45, and was dropped from analysis. Analyses The data are grouped, with each group consisting of ,data about: (a) an information seeker, (b) the question broadcast by the seeker, (c) the information providers ORGANIZATION SCTENCE/VOI, 7, No, 2, March-April 1996 123 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers who responded to that information seeker, and (d) the replies given by the information providers. We tested hypotheses involving individual characteristics, such as the resources of information providers, at the individual level controlling for group (information seeker). This procedure controls for dependence among providers and replies to the same question. We tested hypotheses involving group characteristics, such as diversity of contact resources, at the group level. Results Response Rates and Sample Characteristics We collected 100% of the 82 question texts and sent two questionnaires sequentially to the 82 information seekers who posted them, 55 of whom completed both (67% response rate). These information seekers reported receiving 429 replies to their broadcast questions, an average of 7.8 replies per question. They gave us the names of 365 repliers to whom we sent surveys. We received 295 completed surveys from the repliers (80% response rate) and obtained the text of 263 replies. We obtained online organizational employee data for 92% of information seekers and providers. For analyses combining data from different sources (seeker and provider characteristics, and question and reply characteristics), we used a core sample of information exchanges with no missing categories of data. These exchanges include the seeker's question, two completed questionnaires from the information seeker, a text file of public replies to the question, and completed questionnaires from the repliers. Full data were available for 48 of the 82 questions broadcasted, leaving an effective response rate of 58%: 48 information seekers, their 48 questions, their 263 information providers, and the 263 public replies they gave. To explore possible effects of missing data, we compared the similarity of our research sample to (a) information seekers who kept their replies private, (b) public replies archived over the previous year, and (c) the general population of employees in the firm. We had data on the office location, hierarchical level, industry experience, firm experience, and self-reported experience with second-class e-mail of both public and private information seekers. They differed only in their reported experience with second-class e-mail; information seekers in our research sample had replied to others more, p < 0.05. Because we did not survey private information providers, we could not compare them with public information providers. However, insofar as we could determine, private and public transactions did not differ; there were no differences between . the two in number of replies, usefulness of replies, or whether replies solved the information seeker's problem. Our research sample was similar to the population of employees who use second class mail. Sample employees are predominantly male professionals, proportionally the same as those who use second-class mail generally and the same as in the company as a whole. However, significantly more of our sample and of second-class e-mail users are in sales and field support positions than in the company as a whole, and more are located in smaller sites, in offices more distant from headquarters, and outside the United States. Questions Asked and Information Received Information seekers asked primarily technical questions that averaged nearly half a page (mean = 12.1 lines of text, s.d. = 8.7). They did not pose their questions lightly: 91% of information seekers reported querying at least one other information source before broadcasting their question. Information providers, on average, estimated they spent 9 minutes (s.d. = 4.1) on their replies. Replies averaged about three quarters of a page (mean = 19.1 lines, s.d. = 81.3). Fifty-four percent of the answers contained technical information, and 53% contained a referral to another source of information, such as a specific person or computer file. Information seekers valued the usefulness of the average reply they received at $11.30 (s.d. = $8.70; min. = $0, max. = $25). (See Appendix A for an example of values given by one information seeker to the replies she received.) The value of the best reply was highly correlated with the value of the mean reply (r[48] = 0.79, p < 0.0001). This correlation suggests a "halo" response effect by information seekers, or that replies were additive. Information seekers used fines so infrequently—only 8.4% of replies were assigned any fine—that we did not use fines in the analysis. On average, information providers valued their own replies at $13.20 (s.d. = $8.00; min. = $0, max. = $25). Half of the information seekers (49%) said that the replies they received solved their problem. The mean usefulness of information seekers' replies was positively but not significantly correlated with whether or not information seekers said their problem was solved (/= 0.18). The value of the best reply also was not significantly correlated with whether or not the problem was solved (r = 0.20). Since we cannot evaluate the comparative validity of these measures, we used both the dollar value of usefulness and whether or not the problem was solved as dependent variables in our 124 ORGANIZATION SCIENCE/VO1. 7, No. 2, March-April 1996 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers analyses. Usefulness can be used in both individuallevel and group-level analyses whereas problem solution can be used only in group-level analyses. If weak ties for technical help actually do provide useful advice, this usefulness should be reflected in the technical content of replies. To address the question of how the content of advice is related to its usefulness, we regressed the usefulness of information on the presence or absence of technical content and referrals to other sources of information, controlling for information seeker and the resources of providers. This analysis was performed at the individual level, so that the content and usefulness of each reply could be matched.^ The model is significant (f[46,159] = 2.60, p < 0.0001, R^ = 0.429) as is the improvement over a model with only provider resources and information seeker (A/?^ = 0.034, F[2, 159] = 5.50, p < 0.005). The unstandardized coefficients for technical content and referrals are 2.74 (std. err. = 1.27, p = 0.03) and 3.36 (std. err. = 1.27,/? = 0.009), respectively. Strength of Ties and Basis of Relationship As we had expected, people did not have a personal connection to the person they helped. Information providers did not know their seekers; 81% of the providers said they did not know the seekers at all; an additional 10% said they were barely acquainted. Acquaintance was uncorrelated with the number of replies (r = -0.11), with the usefulness of replies (r = 0.07), or with the solution of the seekers' problems (r = 0.05). A history of posting or answering questions on secondclass e-mail was uncorrelated with acquaintance (rs = 0) nor did this history predict the number of replies (rs = 0) or their usefulness (r = 0.06, r = 0.03, respectively) or whether or not seekers' problems were solved (r = -0.07, r = 0.03). Providers did not help based on the friendliness, social content, or tone of the questions they saw. We used three "sociability" measures: whether the question included a greeting, a closing, or named the company. The presence of a personal opening or closing or both in the question was negatively correlated with the number of replies, r[48] = -0.30,/) < 0.05. The social content of questions did not predict the usefulness of replies or whether or not the problem was solved. Information providers generally did not help based on their similarity to the person needing help. On only one measure of similarity, whether both or neither person worked at a headquarters location, were information providers more similar to their "own" information seeker than they were to the mean of all the information seekers. There was no similarity effect for managerial status, hierarchical level, firm experience, and industry experience. Group level correlations between the usefulness of replies and the similarity of information seekers to their information providers indicated that similarity of managerial status and similarity of firm tenure were negatively related to the usefulness of replies (r = -0.12, p < 0.07 and r = -0.12, p < 0.05, respectively) and similarity of industry tenure was related positively (r = 0.13, p < 0.05). However, adding similarity variables to a regression of usefulness on information provider did not improve the model (R^ without similarity variables = 0.34; with similarity variables = 0.37, change F = 1.64, n.s.). The similarity of information providers to their information seeker also was not correlated with whether or not the seeker's problem was solved; none of the individual correlations was significant and a logistic regression of problem solution on information provider and similarity variables was not significant. In sum, the data reported above do not describe a system of direct social exchange whereby people give useful help to those with whom they have a personal relationship, to those who are similar, or to those who have helped them. Neither acquaintance, a history of reciprocity using the network to exchange advice, nor similarity was very important in predicting the incidence or usefulness of replies. The Usefulness of Information We examined three hypotheses about the usefulness of weak-tie information: the number of replies will predict the usefulness of replies; the range or diversity of groups from which providers come will predict the usefulness of replies; the resources of information providers will predict the usefulness of replies. (See Table 1 for a correlation matrix of all variables used in these analyses.) In each analysis we used hierarchical regression to estimate the contribution of separate groups of variables to the overall model. At the individual level we can test hypotheses about how the number of replies and the resources of information providers predict usefulness. At the group level we can test hypotheses about how the number of replies and resource diversity predict problem solutions, as well as usefulness. Number of Replies. We regressed the usefulness of replies on the number of replies controlling for information providers' resources. This regression is significant at p < 0.01 and the coefficient for the number of replies is negative. (See Table 2.) This result does not ORGANIZATION SCIENCE/VOI. 7, No. 2, March-April 1996 125 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers support the hypothesis that the number of weak ties statistically increases the likelihood of obtaining good advice. We explored this relationship further by investigating if the number of replies was positively related to the most useful reply, reasoning that more replies might offer more opportunity to find one truly useful answer. Regressions of the most useful reply on the number of replies were carried out at the group level, since each information seeker can have only one best reply (or several replies receiving a tie score for most usefulness). The coefficients are positive but not statistically significant, with and without controls for resources (bivariate r = 0.22, p = 0.14). The number of replies is more strongly related to the least useful reply (bivariate r = — 0.30, p < 0.04). This relatively big drop in the usefulness of the worst reply when there were more replies could explain why mean usefulness declined with more replies. The same result was obtained in analyses using information providers' own assessments of the usefulness of their replies which, unlike seekers' assessments, are not affected by their evaluations of other replies. The number of replies also did not predict whether or not information seekers' problems were solved. In sum, for these information seekers, more replies did not improve the net benefit of weak ties. Information Provider Resources. As Table 2 shows, resources of information providers were positively related to seekers' usefulness ratings.'' At the individual level, controlling for the number of replies, there are significant coefficients at the 0.05 level for the resources of being a manager, hierarchical level, and expertise. In a similar analysis controlling for information seeker instead of number of replies, the only significant coefficient is the resource of expertise. These analyses generally support our hypothesis that weak ties for technical advice in an organization are useful to the degree they tap people with superior resources. At the group level we used mean usefulness of replies as the dependent variable (Table 3) to test effects of groups of replies rather than of individual replies. A model with resources, controlling for number of replies, is significant at the 0.06 level. It contains significant coefficients at the 0.10 level for the resource of being a manager and at the 0.05 level for the resource of firm experience. Diversity. We used the number of different countries and the number of different hierarchical levels represented in a group of replies as proxies for diversity. We regressed mean usefulness on number of replies, resources, and diversity. (See second column of Table 3.) Diversity contributes to the prediction of usefulness but in a direction opposite to the hypothesis. In the combined model the resources of being a manager and firm experience, but fewer levels of the hierarchy represented in replies, contribute to usefulness. Adding diversity to a model containing number of replies and resources increases the R^ by 0.09 (p < 0.10). This analysis suggests that advice from a less hierarchically diverse (rather than more diverse) set of ties increases usefulness. Solving the Problem A group-level model regressing solving the problem on resources, controlling for number of replies, is significant ( A"̂ = 17.25, p = 0.02). (See Table 4.) Within this model the coefficient for working at the technical center of the organization is significant at the 0.05 level. In contrast to the findings for mean usefulness, the diversity of ties contributes positively to the model when problem solution is the dependent variable. Generally, the coefficients are positive in this analysis. In the full model, the resources of working at the technical center ip < 0.01) and being an expert (p < 0.10), but not being a manager ip < 0.05), and having more countries represented in the replies (p < 0.10) predict solving the problem. Diversity alone, controlling for number of replies, does not predict problem solution. The full model with resources and diversity, controlling for number of replies, is an improvement over a model without diversity (A;̂ '̂ = 4.85, p < 0.10). Effects of Motivation When asked, "Why did you answer this question?" information providers gave reasons of personal benefit as well as reasons related to general organizational benefit, but they gave more of the latter than the former (paired-f [262] = 8.75, p < 0.01). That is, they favored reasons such as "Answering questions like this is part of being a good company citizen" over "I enjoy helping people" or "I enjoy solving, problems." (See Table 5.) We found little evidence that providers' personal history of giving or receiving help over the network or their acquaintance with information seekers predicted the reasons they gave for helping.^ Table 6 shows individual level analyses regressing usefulness of advice on personal benefits and organizational motivations for answering questions. The regressions are shown with and without controls for resources. Column 1 shows that the personal benefit of earning respect (p < 0.05) and the organizational mo126 ORGANIZATION SCIENCE/VOI. 7, No. 2, March-April 1996 DAVID CONSTANT, LEE SPROULL AND SARA KIESLER The Kindness of Strangers -0.43 -0,13 0.04 -0.09