Printer Friendly

Power, social influence, and sense making: effects of network centrality and proximity on employee perceptions.

reviewers.

1 A central theoretical debate in this body of work is whether the key sources of influence are an individual's direct interaction partners or, instead, structurally equivalent others who occupy similar roles and thus serve as the appropriate referents (Burt, 1982, 1987). As discussed in more detail below, we did not address this distinction in any depth because the data set did not lend itself to positional analysis.

2 Closeness centrality scores were also computed. Correlations In their seminal article, Salancik and Pfeffer (1978) posited that attitudes and perceptions derive from the social context in which they are formulated. Arguing that finding meaning in a job environment is an information-processing activity, they suggested that people develop attitudes as a function of the information available to them through their social relationships and adapt their beliefs to the reality of their own situation. While social-information-processing (SIP) theory has spawned a wealth of research and theorizing (see Blau and Katerberg, 1982; Thomas and Griffin, 1983; Zalesny and Ford, 1990, for reviews), a common critique is that it "has not articulated the mechanisms by which social information flows to and from individuals" (Contractor and Eisenberg, 1990: 7).

This paper joins a growing body of organizational research (e.g., Dean and Brass, 1985; Hartman and Johnson, 1989; Rice, Schmitz, and Torobin, 1990; Burkhardt, 1991; Rice and Aydin, 1991)in arguing that social network theory and methods provide the necessary tools for elucidating key social-information-processing mechanisms. Instead of comparing the effects of cohesion and structural equivalence (Burt, 1982, 1987) on attitude similarity, however, this research explores the relationship between network interaction and perceptions of work-related conditions. In particular, we propose that the effects of two different network-mediated substantive processes need to be untangled: (1) systemic power effects by which individuals' locations in their organization's informal hierarchy shape their access to and control over resources and thus affect positive or negative evaluations of workplace features and (2) localized social influence processes that produce attitude convergence among socially proximate pairs of individuals. The former emphasizes power differentials associated with differences in network centrality, as they affect people's organizational experiences along general dimensions; the latter focuses on specific social influence transmitted through particular relationships. We propose that both types of mechanisms need to be investigated jointly in order to arrive at an improved understanding of the ways in which social networks affect information processing.

In investigating either mechanism, a critical question that has been ignored concerns what types of social relationships or interaction networks are pertinent (Salancik and Pfeffer, 1978; Marsden and Friedkin, 1993). Perceptions are shaped by the opinions of salient or relevant others (Rice, 1993). Salience or relevance, however, may be based on different criteria, including interpersonal similarity and closeness or, instead, power differentials and deference, which are associated with different types of informal networks. Further, the possibility that different network mechanisms may be important for different types of network relationships has not been previously investigated.

This research applies a social network perspective to the study of job-related perceptions, in particular, perceptions of conditions reported as facilitating or inhibiting innovation and creativity in an advertising firm. Following Marsden and Friedkin (1993), we attempt to elucidate the substantive processes underlying claims that network interaction patterns affect perceptions, while simultaneously considering the impact of personal characteristics and formal positions on SIP. As such, the study addresses two central questions: (1) Are there empirically distinguishable SIP processes undergirding the impact of network interaction on individuals' perceptions?, and (2) Do network factors exercise an effect on perceptions above and beyond that accounted for by individual attributes and formal positions?

SOCIAL NETWORKS AND SOCIAL INFORMATION PROCESSING

Social-information-processing theory was developed to explain attitudes, perceptions, and beliefs about organizational phenomena. Noting that individual attributes and objective job characteristics have generally failed to explain much of the variance in people's reactions to workplace features, Salancik and Pfeffer (1978) argued that attitudes and perceptions are socially constructed. SIP theory posits that the social environment provides cues that make certain dimensions of the workplace more salient and more important or desirable than others, furnishes information on other people's evaluations of those dimensions, and regulates direct evaluation of the work setting along positive or negative dimensions. Social information processing plays a particularly important role in shaping perceptions under conditions of uncertainty or ambiguity: When judgments are problematic, people are more likely to arrive at socially derived interpretations of events (Festinger, 1954). The earliest stream of research cited in support of SIP compared the utility of individual characteristics (e.g., sex, education, and personality)and structural variables indicative of social context (e.g., subunit affiliation and hierarchical level) in explaining individual job attitudes (Herman and Hulin, 1972; Herman, Dunham, and Hulin, 1975; O'Reilly and Roberts, 1975; Pfeffer, 1981). Results of these studies strongly indicated that "affective responses to work are predominantly associated with organizational characteristics rather than individual ones" (O'Reilly and Roberts, 1975: 148-149).

Network theorists, however, have long argued that structure is best viewed as patterned, repeated interaction; thus, variables based on actual interaction patterns provide a more adequate operationalization of individuals' social constraints, organizational experiences, or social context (e.g., Weick, 1979; Wellman, 1983; Monge and Eisenberg, 1987). Evidence for this assertion dates back to small-group laboratory studies (see Shaw, 1964, for a review) indicating that centrality in communication networks is strongly related to satisfaction with group activity. A large number of field reports also suggest that network centrality is associated with a variety of positive evaluations of job and workplace features, including job satisfaction (Rice and Mitchell, 1973; Roberts and O'Reilly, 1979; Dean and Brass, 1985), perceptions of ability to take risks (Cancian, 1967), access to information (Coleman, Katz, and Menzel, 1966; Brass, 1984), feelings of belonging or acceptance (Miller, 1980), and organizational commitment (see Hartman and Johnson, 1989, for a review).

Effects of Network Interaction on Perceptions

The substantive processes underlying the relationship between network centrality and attitudes or perceptions, however, are usually left unspecified. Although it is assumed that individuals are embedded in social structures that influence their interpretations of organizational reality and regulate their access to or control over valued resources, we propose that more than one substantive process may be operational (Marsden and Friedkin, 1993). As Gartrell (1987: 59) noted, "it is one thing to say that networks have an effect on social evaluation processes and quite another to say precisely what the effects are, how they are produced, and more generally, what networks add to the explanation of social evaluation phenomena." In this paper we consider two distinguishable mechanisms or processes that may account for observed network interaction effects: Network centrality may influence individuals' perceptions by defining their status or position in the broader social context, or its effects may instead be largely attributable to social influence transmitted through specific interactions. As discussed below, this distinction also corresponds to differences in level of analysis: Social influences on perceptions can stem either from the larger structural context in which individuals are embedded or from their more immediate social environment (Salancik and Pfeffer, 1978).

One set of explanations for the effect of network centrality on affective responses to the workplace is based on theories that emphasize the relationship between situational opportunities and constraints, on the one hand, and attitudes and perceptions, on the other (Salancik and Pfeffer, 1978). The structural context of network relationships, i.e., to whom one is connected via direct and indirect network links, determines access to valued resources; actors who are centrally located within organization-wide webs of interaction have greater control over relevant resources and enjoy a broad array of benefits and opportunities unavailable to those on the periphery of the network (Burt, 1982; Brass, 1992; Ibarra, 1993). Thus, central actors' views of their general situation will be more favorable than those of the more peripheral actors because their situation is objectively more advantageous. The theoretical emphasis is on resource control, power, informal hierarchy, and status differences, and the empirical focus is on explaining the favorability of perceptions rather than their specific content. Further, no direct interaction is necessary for the underlying substantive process to be operational: Actors who are similarly central need not be in direct contact with each other in order to be subject to similar opportunities or constraints and thus have similarly positive or negative perceptions.

An alternate explanation for the effect of centrality and, more broadly, network interaction on attitudes and perceptions is based on communication theory. From this perspective, perceptions are socially constructed in the course of direct interaction with others, and network links are the channels through which organizational culture and norms are communicated (Rogers and Kincaid, 1981). By contrast to the systemic power effect discussed above, the substantive processes are contagion and social comparison, which are predicated on direct interaction. Specific network contacts provide opportunities for comparing and interpreting perceptions, which in turn influence information saliency and subsequent perceptions (Erickson, 1982; Rice and Aydin, 1991).(1) Favorable perceptions, therefore, are viewed as developed or reinforced in direct interaction with people who have favorable views; the greater an individual's centrality, the more likely he or she is to be in contact with others who perceive workplace features in favorable terms, hence the positive correlation between centrality and attitudes. This perspective is consistent with a recent body of empirical research that indicates that people develop shared attitudes and norms through exposure to proximate others in a social network (Wellman, 1983; Dean and Brass, 1985; Hartman and Johnson, 1989; Rice and Aydin, 1991).

Within an organizational context, these two different mechanisms may have contradictory effects on people's perceptions, since information available through specific network ties may easily be incongruent with the reality of an individual's situation. Figure 1 provides an illustration. At a very high level of network centrality (e.g., actors A, B, and C), centrality and proximity effects are virtually impossible to distinguish. Individuals who are highly central tend also to interact with others who are highly central, making it difficult to determine whether their perceptions are due to an advantageous location in the informal hierarchy or, instead, to specific social influences from their highly central network contacts. People on the periphery of a social network (e.g., actors X, Y, and Z), however, are far removed from the social or normative influence of the core of the network. They may have similar views that reflect their low power and integration, or they may instead hold vastly divergent opinions that can only be explained as a function of social interaction within their local subcultures.

Compelling explanations for network interaction effects, therefore, need to specify the substantive mechanisms by which they are produced. Support for a social influence explanation requires evidence of a significant effect due to proximity, i.e., direct contact, while holding constant any effects due to the location of the focal actor with respect to the center of a network. By the same token, the more structural interpretation, based on power and informal hierarchy, could not be supported without ruling out the alternate explanation that the relationship between network centrality and work attitudes is a spurious effect of the tendency for people to have perceptions that are similar to those of the people with whom they interact. The following hypothesis expresses the assumption that both social influence and power will affect individuals' perceptions:

Hypothesis 1: Network interaction affects individuals' perceptions through two mechanisms: localized social influence based on network proximity and systemic power based on network centrality.

Relevant Network Types

A second ambiguity pertaining to the relationship between network interaction and perceptions derives from the fact that multiple types of interaction networks coexist within organizations. Few SIP studies, however, have examined more than one kind of network. Thus, empirical evidence and theoretical development with respect to varied sources of social influence are lacking (Marsden and Friedkin, 1993). Further, Rice (1993) has argued that the operationalization of the "source other" (i.e., those who provide expectations of how the individual should behave or believe or those who serve as role models) is an area of ambiguity, and he criticizes prevalent approaches as lacking conceptual justifications for deciding which sources are influential. While Rice has advocated the use of self-reports by respondents of the importance of each network contact, he does not explicitly note that importance may be based on different criteria, such as emotional closeness or deference and professional credibility. Thus, network interaction effects may depend on characteristics of the relationships (Erickson, 1988), which differ for different types of network ties.

Network types are distinguished by their relationship content, which indicates the kind of relationships linking proximate actors and the sources of power that accrue to more central actors. Social network theory distinguishes between the instrumental network links that arise in the course of work-role performance and expressive network relations that primarily provide friendship and social support (Tichy, Tushman, and Fombrun, 1974; Lincoln and Miller, 1979; Fombrun, 1982). Empirical research indicating that network type may affect the amount of social influence transmitted is indirect but suggestive. Friendship ties tend to be stronger, more intimate links, tend to connect people who are similar on a variety of personal characteristics (Marsden, 1988), and involve more frequent interaction (Krackhardt and Porter, 1986; Krackhardt and Stern, 1988; Krackhardt, 1990). Instrumental links, by contrast, tend to be weaker ties linking people who differ in personal characteristics and/or in their positions in the vertical and horizontal division of labor or in access to scarce resources (Laumann, Galaskiewicz, and Marsden, 1978; Lincoln, 1982; Lin, 1982).

This contrast has several implications for the transmittal of social influence. First, according to social comparison theory, only people who are similar or have convergent interests are useful comparison points. Friendship ties tend to develop between people who are similar on a variety of personal characteristics, including gender, race, age, and religion (Marsden, 1988; Ibarra, 1992) and are also highly affected by propinquity (Festinger, Schacter, and Back, 1950), such that dense friendship networks tend to develop within organizational subunits or departments (Krackhardt and Stern, 1988). Consequently, friendship ties are more likely than instrumental ties to link people who are similar with respect to both personal characteristics and organizational affiliations and who are thus more likely to have consistent interests. Second, friendship ties also tend to be characterized by more frequent interaction than other types of ties (Granovetter, 1973; Krackhardt and Porter, 1986), providing greater repetition of information and increasing the opportunity for the transmission of social cues (Salancik and Pfeffer, 1978). Finally, due to their strength and concomitant pressures for conformity, expressive links carry greater potential for persuasion and influence (Rogers and Kincaid, 1981; Granovetter, 1982; Krackhardt, 1992). Information obtained from friends thus may be more credible or relevant, more easily or frequently available, and more persuasive or influential (Brass, 1992). These arguments lead to the following hypothesis:

Hypothesis 2a: Expressive network proximity will be a stronger predictor of work-related perceptions than instrumental network proximity.

Recent research also indicates that substantially different centrality effects may obtain depending on whether instrumental or expressive networks are examined (Krackhardt, 1990; Ibarra, 1993). As noted above, power and resource control are hypothesized to be the operative mechanism accounting for centrality effects that are not attributable to the social influence carried by proximate others. The relevant distinction between expressive and instrumental networks therefore pertains to differences in structural features of the two types of networks and, thus, the types of resources available through the network. Because expressive relations tend to be strong, symmetrical (i.e., reciprocated) ties, they tend to cluster in dense, interconnected cliques (Krackhardt, 1992) that are better suited for social support than for access to information and resources (Granovetter, 1973). Instrumental relations, by contrast, are more likely to be weak, asymmetrical ties that serve as connections to disparate parts of the social system and that are critical for instrumental action. Further, asymmetries in network ties reflect discrepancies in social power. Thus, while dissimilar people may not be very useful for making comparisons, chains of asymmetric relationships are extremely useful for access to resources (Lin, 1982). It follows, then, that instrumental network centrality will explain work-related perceptions better than friendship network centrality:

Hypothesis 2b: Instrumental network centrality will be a stronger predictor of work-related perceptions than expressive network centrality.

Alternative Explanations: Individual Attributes and Formal Position

Also critical to theory development is the ability to rule out alternative explanations for social information processing, without which the value-added of network explanations cannot be determined (Marsden and Friedkin, 1993). In exploring the effects of network interaction on perceptions of organizational conditions, therefore, it is important to consider other factors that operate to give certain actors greater power and privilege and that shape communication patterns. As noted above, much research to date has not compared the relative contributions of individual attributes, formal organizational roles, and network variables in accounting for differences in attitudes and perceptions.

Attributes including education, past experience, and tenure in the firm often serve as indicators of status and may potentially affect an individual's experiences and perceptions (Pfeffer, 1981). In addition, tenure may act as a proxy for cohort; groups entering an organization at about the same time are likely to share similar views (Krackhardt and Kilduff, 1989). Gender is often associated with access to opportunity (Kanter, 1977) and is thus likely to have a significant effect on an individual's interpersonal and organizational reality. It is also well established that shared membership in a collective of any kind is associated with similarity of perceptions (Hinings et al., 1974; Salancik and Pfeffer, 1978; Berger and Cummings, 1979; Oldham and Hackman, 1981). Further, subunit affiliation may also act as a surrogate for power: According to resource dependence theory (Pfeffer and Salancik, 1978), power accrues to departments that are most instrumental in bringing in or providing resources that are highly valued by the total organization. Formal rank is also associated with power and access to opportunity (Astley and Sachdeva, 1984); a large literature suggests that position in the formal hierarchy affects people's perspectives on their organizations.

Based on previous research (Herman and Hulin, 1972; Herman, Dunham, and Hulin, 1975; O'Reilly and Roberts, 1975; Pfeffer, 1981), we expect structural variables to have greater explanatory power than individual attributes. Further, although SIP theorists recognize that formal positions play an important role in shaping perceptions, network theorists (see Wellman, 1983, for a review) argue that non-network variables can only serve as proxies for the interaction patterns directly measured by network indicators. We hypothesize, therefore, that perceptions of organizational experiences are best explained by network factors:

Hypothesis 3: The rank order of independent variables, from the least to the most effect on work-related perceptions will be as follows: individual attributes (i.e., education, gender, tenure, and past experience), formal positions (i.e., rank and subunit membership), and centrality and proximity.

Types of Perceptions

The social-information-processing literature has also not been very specific with respect to the types of work-related perceptions that are expected to be subject to proximity and centrality effects. Some research findings suggest that network effects tend to be contingent on the types of perceptions examined (Hartman and Johnson, 1989; Krackhardt and Kilduff, 1989), but empirical evidence in this domain is particularly scant and somewhat contradictory. Krackhardt and Kilduff (1989), for example, found that social influence derived from friendship network interaction had stronger effects on individuals' job satisfaction than on their commitment to the organization. They interpreted these results by suggesting that "whereas commitment represents a global response to the organization as a whole, satisfaction is a specific response to one's job". By contrast, Hartman and Johnson (1989) reported that relational proximity mechanisms were more strongly associated with organizational commitment than with role ambiguity, which they viewed as a more behaviorally based variable. Both studies, however, share in common the grouping of perceptions into two broad types: perceptions pertaining to one's job or specific situation (e.g., job satisfaction, role ambiguity) and perceptions of more general organizational conditions (e.g., organizational commitment), a distinction that corresponds to the two types of substantive processes discussed above: localized social influence and systemic power.

Although it has been strongly recommended that dependent variables should be chosen explicitly to test the boundaries of SIP (Zalesny and Ford, 1990), the use of instruments and concepts that are defined a priori, and thus may not be meaningful to organizational members, has been frequently criticized (Alderfer and Brown, 1972; Blackburn, 1982). In SIP research this is particularly important, because phenomena that are not salient or are deemed trivial by individuals may not stimulate them to search for comparative information (Kilduff, 1990; Meyer, 1991). As discussed in more detail in the method section, the work-related perceptions measured in this research were not chosen a priori to capture any particular theoretical dimensions but, rather, were allowed to emerge from field research. The central interest was to tap salient features of organizational members' social context, so that social-information-processing effects could be observed (Rice, 1993).

The Network-Effects Model

The lack of empirical evidence to date for the conceptual arguments proposed above can be partially accounted for by the lack of suitable analytic methods, without which it is difficult to assess the relative importance of different forms of social influence (Marsden and Friedkin, 1993). Although they have been infrequently used in organizational research, network-effects models (Erbring and Young, 1979; Doreian, 1981) can take into account the extent to which similarity of perceptions between an individual and proximate others is a significant determinant of the focal person's perceptions, while simultaneously estimating the effects of variables operationalized at the individual level, including network centrality, individual attributes, and organizational positions.

Beyond the ability to consider the effects of centrality and proximity jointly, this approach has several advantages. First, social comparison research typically does not shed light on social factors that may account for people's particular perceptions; emphasis is placed on the degree to which people agree with each other, not what they will agree on (Erickson, 1988). Network-effects models, by contrast, provide estimates of the amount of pressure exerted by proximate others, while providing information about the actual content of perceptions. Further, currently available methods for gauging the effects of proximity on attitude similarity are essentially correlational (see Huber and Schultz, 1976) and cannot provide estimates of the relative significance of different variables in accounting for differences in attitudes and perceptions. Given that non-network variables, including departmental affiliations and formal rank, have been shown to exert powerful effects on attitudes and perceptions (O'Reilly and Roberts, 1975; Pfeffer, 1981), models that do not allow comparison of the relative significance of different types of variables cannot provide compelling evidence of the value-added of network explanations of social evaluation processes over and above explanations based on variables that network researchers have argued are poor proxies for actual interaction patterns (Wellman, 1983). The network-effects model, as used in this research, is discussed in more detail below.

METHOD

Research Site

This research was conducted as part of a larger study of informal networks in a New England advertising and public relations agency. At the time of data collection, the firm had 94 full-time employees. The research was conducted in two phases. The goal of the first phase was to better understand the research context through unstructured interviews with representatives of the various organizational groups. In this phase, the network boundaries were identified on site, and the data were gathered that were used to construct the dependent variable scales. The data reported here were collected in a second phase through structured interviews, during which participants were asked to complete a survey instrument containing sociometric questions, items concerning their perceptions of the organization, and a background information sheet.

Participants

Network analysis requires an appropriate determination of the boundaries of the network under study, as errors can distort the overall configuration of actors in a system (Laumann, Marsden, and Prensky, 1983). For this reason, in defining the network population, all professional staff (N = 74) were included a priori and an iterative boundary definition method was used to determine what secretarial staff to include in the study. The decision rule was to enlarge the network population to include any secretary who was nominated by more than two of the initial respondents in answer to the sociometric questions described below; two or fewer nominations was deemed to indicate that a person was marginal to the network, and he or she was excluded from the study.

Once the network has been identified, one must collect data from all members because, at present, no generally accepted techniques have been developed for sampling within a network (Rogers and Kincaid, 1981). The final response rate was 97.5 percent of the network population, which included 98.6 percent (73 of 74 members) of the originally identified professional population and six of the seven members of the secretarial/clerical staff identified by the iterative method. As the additional employees did not nominate anyone else more than twice, no further participants were included in the study. In all cases, participation was voluntary, and respondents were assured that their responses would be kept confidential.

Measures

Network indices. Following Krackhardt (1990) instrumental and expressive networks were measured with two sociometric questions. Respondents were asked to name the people in their firm (1) who are "important sources of professional advice, whom you approach if you have a work-related problem or when you want advice on a decision you have to make," and (2) "who are very good friends of yours, people who you see socially outside of work." Answers to these two questions provided the raw data used to derive centrality and proximity indicators. In an effort to limit measurement error (Holland and Leinhart, 1973), respondents were not restricted to a fixed number of nominations. Ten blanks were provided after each question but respondents were instructed to employ as many spaces as needed, resulting in several individuals creating additional blanks. As a recognition aid, respondents were also provided with the firm's one-page telephone directory, listing all members.

Network centrality. Centrality was operationalized as an "aggregate prominence" (Knoke and Burt, 1983) measure, which indexes individual centrality as a function of the centrality of those to whom one is connected through direct and indirect links (Bonacich, 1987). Rather than allowing all relationships of equal proximity to contribute equally to an actor's centrality, as in Freeman's (1978) "closeness" measure, this formulation assumes that centrality is increased positively by connections to others who are highly central and assigns the highest level of centrality to the actors with the closest relations (that is, direct or short indirect links) with many central actors (Bonacich, 1987).(2) Aggregate prominence was computed as follows (Burt, 1988):

|Mathematical Expression Omitted~

where |C.sub.j~ = centrality of actor j, |C.sub.i~ = centrality of person i, N = population size, |z.sub.ij~ = distance between actors i and j, which varies from 0 to 1 as the number of indirect links between them decreases, and g = the maximum eigenvalue of the aggregate relation matrix. Centrality scores range from 0 to 1, higher values indicating greater centrality.

In computing the centrality scores, relationships were not symmetrized, preserving the distinction between being the source versus the object of a relation (Burt, 1982), for two reasons. First, many instrumental relationships, such as advice relations, are inherently asymmetrical, and even intimate relationships such as friendship ties may be asymmetrical, since conceptions of closeness vary across individuals (Krackhardt, 1990; Marsden, 1990). Reciprocation rates for the advice and friendship networks of .32 and .46, respectively, are consistent with the distinction between instrumental and expressive relations. Second, the arbitrary symmetrization of relationships could potentially obscure power or status differentials; when asymmetry is preserved, Burt (1982) argued, the scores denote prestige or power, since |z.sub.ij~ may be interpreted as the extent to which events or resources controlled by the focal actor are of interest to all others in the network.

Network proximity. A nagging methodological problem in network research on attitudes and perceptions is that, because of the very nature of the phenomena of interest, independence of observations, a key assumption of standard linear regression models, is untenable. Actors in a network communicate with one another and thereby influence each other's attitudes. As a result, network autocorrelation obtains, such that an individual's perceptions (i.e., values on the dependent variables) are not independent of those of the other actors in the study population. To the extent that such influence processes are not included in the specification of the model, the use of OLS regression techniques results in correlated error terms and hence in potentially biased interpretations about the effects of the independent variables. Krackhardt (1988) has demonstrated the degeneration of significance tests of OLS parameter estimates in the face of such autocorrelation.

In organizational research, the most frequently used technique for handling network autocorrelation is the Quadratic Assignment Procedure (QAP). QAP abandons the parametric approach inherent in the estimation of regression parameters and, instead, tests whether dyadic ties are significantly related to dyadic similarities in attitudes or perceptions (Krackhardt, 1988). An alternative method of taking social influence processes into account is the network-effects model, also known as the spatial-effects model, the mixed regressive-autoregressive model, and an endogenous feedback model of social influence (Erbring and Young, 1979; Cliff and Ord, 1981; Odland, 1988). We chose this method because it is the only currently available method that directly models the social influence effect, taking into account the network relationships between individual respondents that result in nonindependent values for the dependent variable (Doreian, 1981, 1989; Doreian, Teuter, and Wang, 1984). An additional advantage is that the method also yields parameter estimates of the effects of the independent variables, as in OLS regression. The network effects model is, following Doreian (1982):

Y = |Rho~WY + BX + |Epsilon~.

The model is an adaptation of the linear regression model that takes into account a set of independent variables but also includes a matrix of ties connecting actors in a network (W). In this research, the set of independent variables (X) includes individual attributes, formal positions, and centrality in the advice or friendship network. As in OLS regression, the B's are the parameter estimates for the effect of these variables on the dependent variable (Y). For each focal individual, or ego, the term consisting of rho, W, and Y estimates the effect of the values of the dependent variable on ego's value for the dependent variable for all alters who are proximate to ego. Following Odland (1988: 52), the usual OLS assumption concerning error terms, homoscedasticity, and independence apply.

The weighting matrix, W, may be computed on the basis of relational or positional, i.e., cohesion versus structural equivalence, criteria (Doreian, 1989; Marsden and Friedkin, 1993). Analyses using a structural equivalence criterion for estimating rho were attempted, but these yielded meaningless results and are not reported.(3) Thus, W is the matrix of ties sent from actors to other actors in either the advice or friendship network: a 1 indicates the presence of a direct relationship between a pair of actors, while a 0 indicates the absence of a direct relationship. The weighting factor is the strength of the relationship, as operationalized by relational adjacency. The social influence of direct ties is assumed to be stronger than that of indirect ties, which are taken into account by simultaneously estimating the product of W and Y across all actors in the social setting.(4)

When W is multiplied by the vector of scores on the dependent variable, Y, this term serves to weight the amount of influence of alter's level of Y on ego's. Rho is the parameter estimate of the average effect of others' values of the dependent variable on one's own level of the variable.(5) A positive value indicates similarity or conforming pressures between an actor's views and those of his or her proximate alters; a negative value indicates dissimilarity. A nonsignificant rho indicates that, statistically, the respondent's perceptions may be considered to be unaffected by those of his or her alters. In order to estimate the model, the W matrix is row-normalized, so that each non-zero entry is the proportion of all nominations that person makes. In the case of binary ties, this means that each non-zero entry is the inverse of the number of nominations made by the individual. By using such a normalization, the rho parameter has bounds of -1 to 1.

Because the endogenous variable appears as both explanatory variable and outcome, this model cannot be solved numerically (Doreian, 1980, 1981, 1982; Cliff and Ord, 1981; Doreian, Teuter, and Wang, 1984). To estimate the model, iterative maximum likelihood techniques are used, as developed by Doreian (1982) and implemented by Friedkin (1992). Results may be interpreted in both of two ways: (1) the regression coefficients for the exogenous variables can be interpreted as usual, with a control for the social influence (proximity) effect, or (2) the rho coefficient may be interpreted as an estimate of the degree and direction of the social influence (proximity) effect on the focal actor's score on the dependent variable. The latter interpretation, therefore, differs from QAP approaches in that it estimates the magnitude of the effect of beliefs held by relationally proximate alters on ego's perceptions.

Individual Characteristics and Formal Positions

Several of the individual characteristics and formal positions used in this study as alternative explanations to network-derived sources of social influence have particular relevance to the research site studied. First, firm size, as measured by number of employees and billings (advertising revenues) had doubled in the three years prior to this study. This growth was accompanied by an influx of advertising professionals from large, established advertising firms or corporations, who tended to have views that contrasted with those of the local, home-grown talent. As a result, perceptions and organizational experiences were expected to differ as a function of tenure and past work experience. Second, the organizational division of labor was not independent of gender: Women were predominantly found in either low-status support departments or in the lower ranks of higher prestige, male-dominated departments. Thus, it was expected that gender would also affect employees' organizational experiences.

The organization of the firm, which was typical of most advertising firms, also suggests that subunit membership might be an important indicator of social context. The agency was organized into groups responsible for attracting and managing clients (Account Services and Public Relations), groups in which the core work of the firm, the creation of advertising campaigns, is accomplished (Creative Services and Operations), and various support departments, including Media and Accounting. These groups were not only distinguished by function but also by important differences in sources of power and status differentials. Client-management groups tend to be the most powerful because they control an agency's sole source of revenues, the clients. Their elite status and concomitant power are enhanced by the many uncertainties that characterize the client relationship: In advertising there is typically no specified period of time during which agencies are assured of employment and it is not unusual for clients to terminate relationships with agencies abruptly (Comanor, Kover, and Smiley, 1981). The Creative staff, however, has the capacity to win national advertising awards, a critical form of recognition that is instrumental in attracting and retaining clients.

Also as in most advertising agencies, the groups differed in their values and beliefs. The primary allegiance of Account Service personnel is to the agency, and their job is customer satisfaction; the Creative staff, by contrast, judges its performance against that of its peers in the profession and is primarily interested in opportunities for unrestricted creativity. Consequently, Account Service often feels that the Creative staff is unwilling to consider its client's needs, while the artists and writers claim that Account Service will promise clients anything they want, without regard to feasibility or creative quality. In this research site, both were often in dispute with support groups, including Media and Accounting, who tend to be viewed as "paper pushers," fulfilling no essential function.

The non-network independent variables were operationalized as follows. Gender is a dummy variable indicating being male (0 = female, 1 = male). Education is coded as a three-level variable (0 = no college degree; 1 = college degree; 2 = graduate degree). Tenure in the organization is measured in years. Prestige of past work experience is also a dummy variable, a value of 1 indicating previous experience in Fortune 500 or Advertising 100 firms. Although the firm had no formally published organization chart, senior management identified five distinct levels of formal authority that were independent of whether individuals held professional or administrative jobs. Senior management included the president and vice presidents. Those who reported directly to them were considered "middle management," a group that included all the firm's "star" artists and writers who were not vice presidents. The third layer of the firm consisted of supervisors and lower-level professionals. This strata differed from the next, the entry-level employees, in years of experience, tenure in the firm, and job responsibilities. At the lowest level was the secretarial/administrative staff. Respondent rank was coded as follows: 5 = senior management; 4 = mid-management; 3 = supervisors and lower-level professionals; 2 = entry-level; 1 = secretarial.

Since the organization was composed of nine small departments, each respondent was assigned a value corresponding to membership in one of the three larger groups discussed above. These groups were defined by aggregating departments that were located on the same floor of the facility, were in constant contact with each other, held similar positions in the workflow, and were afforded the same level of status. Department 3 includes the Public Relations and Account Service departments, the two groups responsible for client relationships and therefore the most critical boundary spanners. Department 2 includes departments in which the core work of the organization was conducted: the Creative, Research, Production, and Traffic departments. Finally, Department 1 includes members of the Media and Accounting departments, groups that provide support services to the core and boundary spanners and were generally regarded as having low status. Department was thus coded as two dummy variables: Department 2 and Department 3, with Department 1 as the reference category.

Dependent Variables: Perceptions of Organizational Conditions

Starting with the premise that an advertising agency's reputation and livelihood depend on creativity and innovation, a preliminary phase of the research consisted of open-ended interviews with representatives of the various organizational groups. The interviews were designed to elicit employees' views with respect to factors perceived to encourage or inhibit innovation and creativity in the firm. Five dimensions emerged from the interviews: encouragement of risk taking, information access, individual acceptance, job autonomy, and interdepartmental conflict. Five dependent variables were derived from this raw interview data using an empathic or organic method of constructing a questionnaire (Alderfer and Brown, 1972). An empathic questionnaire, in contrast to a standardized instrument, is developed on-site to reflect salient organizational issues and to help formulate questionnaire items that use the organization's language (Alderfer and Brown, 1972). The resultant five scales used were intended to measure aspects of the workplace that are salient and relevant for study respondents. The scale items and alpha reliability coefficients are reported in the Appendix. Individuals' perceptions on each of the five dimensions were computed as the mean of all items in the scale, measured from 1 = strongly disagree to 6 = strongly agree. Negatively stated items were reverse-coded to maintain consistent direction.

Data Analysis

Data analysis consists of a series of nested regression models. The first model regresses each of the five perception scales on individual attributes and formal organizational positions, providing a baseline against which the various network measures can be compared. Advice and friendship centrality indices are then added separately to the baseline model for each of the five dependent variables, in order to assess whether and what type of centrality adds unique variance explained beyond that explained by the non-network variables. The third step incorporates the baseline model (without centrality) into the network-effects model. Since network-effects models do not provide estimates of variance explained, this step can only assess the presence of a significant proximity effect. The effects of proximities in instrumental and expressive networks, controlling for the effects of non-network variables, are gauged separately. In the fourth and final step, advice and friendship centrality indices are added, respectively, to the advice and friendship network-effects models in order to assess whether proximity and/or centrality play a significant role in the presence of controls for the other.

RESULTS

Table 1 provides the means and standard deviations for all variables. Table 2 reports the intercorrelations among all variables. All the non-network variables were significantly correlated with at least one of the five dependent variables. Relative to friendship centrality, advice centrality evidenced higher correlations with the dependent variables. Advice centrality, however, was also much more highly correlated with rank than friendship network centrality and thus may be more likely than friendship centrality to duplicate the effects of formal position (Krackhardt, 1990).
Table 1

Means and Standard Deviations
Variables                       Mean     S.D.

Empathic scales
Risk taking                     4.41      .99
Acceptance                      4.54     1.13
Information access              3.68     1.08
Interdepartmental conflict      3.38     1.04
Autonomy                        4.92      .87

Individual characteristics
Gender                           .43      .50
Tenure                          4.85     4.32
Prestige                         .23      .42
Education                       1.89      .64

Formal position
Rank                            3.08     1.23
Department 1                     .32      .47
Department 2                     .30      .46
Department 3                     .38      .49

Centrality
Advice                           .25      .31
Friendship                       .42      .33


Table 3 provides the results of three regression analyses for each of the five dependent variables. The models in column I include only non-network variables; the models in columns TABULAR DATA OMITTED II and III include advice and friendship centrality, respectively. Consistent with hypothesis 3, in no case do any of the individual attributes provide a significant contribution to variance explained. Column I results indicate, however, that formal positions tend to have an important impact on responses to all but the interdepartmental conflict scale: Individuals who are higher in rank and/or affiliated with at least one of the high-status departments tend to report greater encouragement of risk taking, acceptance, autonomy, and access to information than individuals lower in rank or affiliated with the reference category, Department 1.

Columns II and III report the results of adding the advice and friendship centrality, respectively, to the baseline models. Two patterns are evident. First, friendship centrality does not appear to be a significant predictor of responses to any of the empathic scales. Second, advice centrality produces significant increases in variance explained (ranging from .04 to .15) with respect to three of the five scales: risk taking, acceptance, and information access. Thus hypothesis 2b was supported. Not surprisingly, because information exchange is reflected in the advice network (Brass, 1992), advice centrality contributed most significantly to perceptions of information access. Further, the effects of rank, one of the most significant predictors in the models in column I, disappear (and in one case, change signs) when advice centrality is added to the risk-taking, acceptance, and information access models (column II). This finding suggests that the portion of variance accounted for by formal rank in the models in column I is captured by advice centrality and is consistent with the hierarchical interpretation of centrality suggested earlier: Advice centrality may be a better indicator of the "real" hierarchy than formal rank or friendship centrality. Departmental affiliation, however, appears to have significant independent effects.

TABULAR DATA OMITTED

By contrast, advice centrality had no significant explanatory power in accounting for variance in responses to the interdepartmental conflict and autonomy scales. Perceptions of job autonomy appear to be exclusively regulated by the formal division of labor: Higher-rank employees and/or Creative staff members probably do have an objectively higher degree of job autonomy. While none of the variables TABULAR DATA OMITTED appear to be good predictors of responses to the interdepartmental conflict scale, it should be noted that there is a significant effect for department membership, but it is between departments 2 and 3.

Table 4 reports the results of the network-effects models, using indicators derived from the matrix of advice relationships. Table 5 reports the results of similar models, using network indices derived from the friendship matrix. For each of the five dependent variables, the results of two regression equations are reported. Column I reports the estimates of the network proximity effect, without centrality in the models. The effects of centrality and proximity, each controlling for the other, are then assessed in the models in column II. In both sets of analyses, regression coefficients for the individual attributes (i.e., education, gender, prestige, and tenure) remained stable and insignificant and are not reported.

Consistent with hypothesis 2b, results reported in Table 4 suggest that the operative SIP mechanism with respect to the advice network is centrality. Proximity (the rho effect) in the advice network was not significant in any of the five models, indicating that focal actors' perceptions tended to be unaffected by the perceptions of those with whom they had direct contact in the advice network, whether or not the effects of advice centrality were held constant. By contrast, controlling for proximity in the advice network, advice centrality remains significant wherever it was significant in TABULAR DATA OMITTED the OLS models, i.e., in predicting responses to the risk-taking, information access, and acceptance scales. Also as in the OLS results, the significant effects of rank disappear when advice centrality is added (column II) to the risk-taking, acceptance, and autonomy models and change signs in the information-access model. This finding is discussed in more detail in the next section.

Results reported in Table 5 suggest that friendship network proximity is not a very good predictor of responses to four of the five perception scales. Consistent with hypothesis 2a, however, where interaction in the friendship network affects perceptions it does so through proximity and not through centrality: Perceptions of interdepartmental conflict were significantly affected by the perceptions of people cited as friends, while friendship network centrality evidenced no significant effects on perceived conflict.

DISCUSSION AND CONCLUSIONS

This study provides support for the assertion that informal interaction networks, in channeling social influences as well as control of valued resources, have a significant impact on job-related perceptions, over and above the effects of traditionally emphasized sources of influence such as formal position and departmental affiliation. The results of this study provide support for the notion that two independent mechanisms underlie the effects of network interaction on perceptions: instrumental network centrality and friendship network proximity. The former, however, was more strongly and consistently supported than the latter, which was only operative with respect to one set of perceptions. Findings also indicate strong support for a resource control or power-based interpretation of centrality effects. Advice network centrality appears to shape organizational experiences by regulating access to information, resources, and legitimacy, and not by exposing focal actors to the views of those they seek out for advice. Overall, results underscore the importance of distinguishing between instrumental and expressive networks and are consistent with the notion that different types of perceptions may be affected by different types of network mechanisms, such that competing explanations "may not be so much competing as contingent on the nature of the processes examined" (Hartman and Johnson, 1989: 543).

A power-based interpretation of centrality is also consistent with the diminished magnitude of the regression coefficients for rank in models that included advice centrality (column 2, Tables 3 and 4), as well as the negative effect of rank in the information access model that included both advice proximity and advice centrality (column 2, Table 4). This suggests that the social processes underlying the observed relationship between rank and favorable attitudes are better captured by advice centrality. These appear to be status-hierarchy effects--that department membership tends to retain an independent effect when advice centrality is taken into account suggests a clear distinction between hierarchical sources of power (advice centrality and rank) and other structural power bases. Because rank measures formal position in the organization's hierarchy, while advice centrality measures prominence in the organization's informal hierarchy, the variance associated with formal rank net of advice centrality is irrelevant to explanations of people's perceptions. Further, in the models explaining perceptions of information access, rank did not just diminish in magnitude, its effect became negative. A speculative interpretation (which we were unable to test directly) is that high-ranking individuals who are peripheral to the advice network may perceive themselves to be particularly "out of the loop" when it comes to access to information.

That centrality appeared to govern responses to the risk-taking, information access, and acceptance scales to a much greater extent than the second group of scales (autonomy and interdepartmental relations), where no significant centrality effects were obtained, also merits further discussion. There are two possible explanations for these findings. Consistent with SIP theory assertions that individuals are particularly vulnerable to social influence under conditions of uncertainty or ambiguity, the risk-taking, information access, and acceptance scales may be viewed as tapping more abstract, subjective, or uncertain phenomena, rather than aspects of the job that are experienced more directly as a function of hierarchical position and job characteristics, such as autonomy and conflict. Alternatively, the autonomy and conflict scales may be viewed as exceptions to broader trends that can be explained by unique characteristics of the study site. Means and standard deviations reported in Table 1 indicate that responses to the autonomy scale were the most consistently favorable. It is reasonable to expect that centrality will affect only perceptions of workplace features on which there is a wider range of opinion. Responses to the conflict scale, by contrast, appear to be largely a function of departmental affiliation, with the largest perceptual difference occurring between the Creative and Account Service departments.

An unexpected finding was the weak support for the localized social influence or proximity effect hypothesized to operate via the friendship network: Perceptions tapped by four of the five empathic scales were unaffected by friendship network proximity. One possible explanation for the lack of support for this hypothesis may rest in characteristics of the firm. While friendship relations usually form within groups (Krackhardt and Stern, 1988), in smaller organizations friendship links may be more likely to cross departmental boundaries, thus exposing people to a wider (as opposed to smaller) range of perceptions from their friends in different subunits. It is also possible that the weak friendship proximity effects reflect a research setting in which instrumental and expressive networks differ and where job-related information flows primarily through the former. In other settings, for example, voluntary organizations, there may be less of a distinction between the two kinds of networks and, hence, a greater likelihood of observing proximity effects. Alternately, the significant, convergent effect on perceptions of interdepartmental conflict (and not the other types of perceptions) suggests that it may exemplify a category of perceptions that is more susceptible to social influences. Future research may explore whether certain types of perceptions are more likely to be transmitted through friendship proximity mechanisms.

The most important limitation of this study derives from a research design based on one organization, which raises questions regarding generalizability to other settings. While the fine details of the findings may not generalize far beyond the current research setting (e.g., perceptions of autonomy and interdepartmental conflict tend to be unaffected by centrality), the more general interpretations and implications of the findings are likely to have wider applicability. Three broad findings that should generalize to other settings are that network mechanisms do not operate independent of relationship content in their effects on individuals' perceptions, that at least two substantive processes underlie the relationship between network interaction patterns and perceptions, and that the informal hierarchy in which people are embedded is a critical element of the social context. In addition, because the research was cross-sectional, it may have obscured causal relationships. Rather than centrality producing more favorable perceptions of organizational conditions, it is possible that communicating the "right" beliefs facilitates access to inner circles, thus increasing network centrality. Similarly, people may become friends precisely because they share similar views (Erickson, 1982).

Implications for Theory and Research

This research has important implications for both SIP and social network theory and research. Although entirely consistent with the central tenets of social-information-processing theory, the findings highlight the conceptual value-added of clarifying several key areas of ambiguity, and novel methodologies for operationalizing those ideas are suggested. In the domain of social network theory and research, beyond pointing to different types of network-mediated substantive processes (i.e., systemic power and localized social influence), these findings indicate the benefit of distinguishing between instrumental and expressive informal networks.

The differential effects obtained for network indicators based on advice versus friendship relations suggest that the instrumental-expressive distinction may be pertinent to a broad range of network research, particularly research concerned with cognitive processes and information processing. That proximity results were only obtained with respect to the friendship network, for example, is consistent with the relatively weak proximity results reported in previous research, which has tended to focus on instrumental networks (e.g., Rice and Aydin, 1991). Moreover, although concerned with social influence, SIP research has focused primarily on exposure to information, ignoring the role of power discrepancies and asymmetric relationships in shaping perceptions. As a result, non-network variables such as rank and subunit membership that better capture hierarchical aspects of social structure may play a more critical role (Marsden and Friedkin, 1993).

A second research implication derives from the now-cumulative findings that network effects tend to be contingent on the types of perceptions examined (Hartman and Johnson, 1989; Krackhardt and Kilduff, 1989). While a broad range of perceptions has been investigated in SIP research, each individual study has focused on a fairly limited subset of that range (Hartman and Johnson, 1989 is an exception), to allow more precise specification of the boundaries of SIP (Zalesny and Ford, 1990). Further, research to date, including this study, has not chosen dependent variables on the basis of dimensions previously speculated to be associated with greater or lesser SIP or with different SIP mechanisms. The need to better designate the specific conditions under which one or more of the mechanisms discussed above will affect attitudes and perceptions, however, may at times conflict with the desire to use dependent variables that are clearly grounded in the realities of the respondents.

People's attitudes and perceptions do not develop in a vacuum. They are powerfully molded by social situations. But there is no single social process in which interpersonal interaction is embedded. Much of what occurs in organizations is only remotely related to the social exchanges prescribed by the formal organizational chart. And, like formal structure, informal structure is complex and multilayered: People are involved in multiple, dynamic, and overlapping webs of relationships. Friendship ties may produce similarity in ways of viewing the world, but information and opinions are also routed by informal power hierarchies that structure interaction between people who see the world very differently. Our findings provide some insight into the intricacies involved in the sense-making process and suggest new and productive ways of exploring how social networks shape our points of view.

Funding for this study was generously supplied by the Organizational Behavior Department of Yale University and the Harvard Business School Division of Research. Earlier versions of thus paper were presented at the Sunbelt International Social Network Conference, San Diego, 1990 and the Academy of Management Annual Meetings, Miami, 1991. The authors are grateful to Paul DiMaggio, David Krackhardt, Peter Marsden, and Ron Rice for helpful suggestions. We also profited greatly from comments provided by Marshall Meyer and this journal's
COPYRIGHT 1993 Published for Cornell University, Johnson Graduate School.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1993 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:includes appendix
Author:Ibarra, Herminia; Andrews, Steven B.
Publication:Administrative Science Quarterly
Date:Jun 1, 1993
Words:9249
Previous Article:The context of interunit influence attempts.
Next Article:Affect and managerial performance: a test of the sadder-but-wiser vs. happier-and-smarter hypotheses.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |