C H A P T E R T W E L V E Influence Networks in a College Town T-COMMUNITIES, CHILDREN, AND THE HORIZON OF OBSERVABILITY How do stage 4 neighbor networks and the neighborhood community norms and values and social control they produce relate to tertiary street networks? In the previous chapter, I discussed Friedkin’s (1983) work, which found that academics separated by two steps (i.e., they discussed their current research in face-to-face communication with the same third alter) were more likely than those who didn’t to be aware of each other’s research. I argue that Friedkin’ s research provides a useful baseline to study neighbor networks and collective efficacy. It seems reasonable to suppose that any neighborly relation at least as intensive as discussing current research in face-to-face communication might be expected to create awareness at least two steps out. Recall that, in the college town, any two typical households with children knew a common other household with children (with one more intermediary household with children required a minority of times), being separated by an average of 2.38 steps. Thus, this relation of knowing each other is near the edge of Friedkin’s “horizon of observability” for households with children. In contrast, the additional two steps that were required to connect the typical households without children in this community (4.41 steps separated average households of this sort) suggest that they are far beyond any “horizon of observability,” more than doubling it. Furthermore, for the three of four intermediary households typically required to connect the average pair of households without children in this community, at least one typically had children living in it. Thus, while the short path lengths among households with children, near the horizon of observability, may have made them aware of each other as a community, the long path lengths 162 INFLUENCE NETWORKS IN A COLLEGE TOWN among households without children would have made them very unlikely to be aware of each other as a community, and, when they were aware of a community, this community would have involved households with children. Two households with children who merely know each other may not have a very intense relation. However, one relation we interviewed about, neighbors’ monitoring of each other’s children and their evaluations of other neighbors’ monitoring each other’s children in spontaneous playgroups, seems likely to be at least as important to neighborhood families as academic’s discussions of their current research is to academics. For the exhaustive census, 6,198 such child-monitoring relations were reported among 669 households with children. This relation forms a giant connected component1 with 518 members. Seventy-seven percent of all households with children are in this giant component. The remaining households with children are either isolates (66 households that have no child-monitoring relation with another household) or members of trivially sized components. The trivial components include 13 dyads, five triads, four four-household components, two fivehousehold components, and three six-household components. Within this giant component encompassing over three-fourths of all households with children in the neighborhood, the median path length between households is two and the mean is 2.35. Thus, 77 percent of the households with children in this neighborhood existed in a community where the typical household watched over the children of someone who watched over the children of other typical households (with one more step of “watched over the children of” in some households). Furthermore, these networks were several times as dense as those Friedkin (1983) studied, being far more connected by two-step paths than were Friedkin’s academics. On average, Friedkin’s academics were connected by two distinct two-step paths of sharing research, while in contrast the neighborhood average was 13.72. Thus, a typical household with children had watched over the children of 14 other households, each of whom who watched over the children of the same other typical household with children in the neighborhood.2 Thus, it seems likely that a very large portion of the neighborhood is mutually observable through this relation. 163 C H A P T E R T W E LV E In sum, the length of this relation, trusting each other to watch over children in spontaneous playgroups, is near the edge of Friedkin’s horizon of observability, and its density is much greater than that of the networks Friedkin studied. In addition, this relation is arguably at least as sociologically substantive as two academics discussing their research. If parents are at least as interested in the values of those who monitor their children as academics are interested in the research of their colleagues, then it seems reasonable to assume that they may have a good sense of the values, at least relating to children, of households monitoring children throughout this neighborhood. Clearly, in this case, the behavior of neighborhood children and those who monitor them is observable to most of the other households with children. The network of relations also appeared to create a community value. Every single person who was part of the neighborhood-sized, but closed, community of households watching over each other’s children in spontaneous playgroups felt it was a good neighborhood for children. While others who were not part of this network also believed it was a good neighborhood for children, 65 of the 151 households with children not included in this giant component did not assess the neighborhood as good for children (44 of these were isolates). The likelihood that 44 of the 65 households who believed it was not a good neighborhood for children might be isolates was statistically insignificant (p < .05). Furthermore, the likelihood that none of the 65 would be in the giant component was statistically infinitesimal. It simply could not have resulted from chance. For those in the giant component, the neighbor network clearly created the social closure necessary to affirm this belief for everyone. Thus, to be part of this community of neighbors watching over each other’s children was equivalent to believing it was a good neighborhood for children. T-COMMUNITIES AND SOCIAL CONTROL An interesting, although unhappy, example of this observability in action presented itself for us quite fortuitously. During the Halloween immediately following the exhaustive census an incident occurred in the t-community we had been intensively studying. Several parents 164 INFLUENCE NETWORKS IN A COLLEGE TOWN claimed to have seen a teenager exposing and sexually gratifying himself while watching children “trick or treat.” The parents did not recognize the teenager and believed he did not live in the neighborhood. A police report was filed that provided details of the incident, as well as which parents claimed to have observed the incident, and a date. This incident was intentionally not reported in any local papers, even in the police beats. It was not announced at any of the local schools. In fact, school officials were unaware that it had taken place. This news spread through the t-community, however. The details were modified somewhat as the event was described by one neighbor to another. In one version or another, the event became part of the collective memory of at least some of the neighborhood’s residents. It became such a powerful part of the neighborhood’s informal collective memory that, during the following Halloween, parents organized meetings a few weeks in advance to actively monitor their children while they were going house to house. While this semiorganized response did not occur the subsequent years, parents testified to increased vigilance. Since we sampled this neighborhood, the t-community and surrounding areas, three years later, I had an opportunity to follow up on the story. I could map the degree to which the news had spread through the neighborhood. I could verify the accuracy of the details. I could, more importantly, determine who was and who was not aware of the story. Had it spread primarily among parents, or was it more widely known? Had families who had moved into the neighborhood after that Halloween become aware of it? What were the geographic parameters that guided and constrained its spread? When interviewing residents during the data collection three years later, we asked if they had heard about an incident occurring at Halloween “some years earlier.” While details of the their description varied tremendously, of the 824 families with children we surveyed, 294 mentioned the incident in some form. Only 17 of the households without children mentioned the incident. Of the families with children, 161 had moved into the neighborhood after the incident had occurred three years earlier. Thirty-five of these new families with children mentioned it. Altogether, the families interviewed lived in three elementary school districts, with 380 families in one district, 259 families in another, and 184 families in a third. Of these, 161 families in the first school catch165 C H A P T E R T W E LV E ment, 46 families in the second school catchment, and 86 families in the third referred to the incident. The most telling fact, however, was that 432 families of the interviewed families lived in the t-community in which it had occurred, and all 294 who referred to it lived in that tcommunity. Thus, the story—the observed behavior and the observation of parental response, the valuation of behavior by the neighborhood community—had persisted, had been introduced to new residents, had spread almost exclusively among parents, had spread throughout all three school catchments, but had remained entirely within the t-community. NEIGHBOR INFLUENCE AND T-COMMUNITY CULTURE Having noted the remarkable closure and short range of these neighborly relations, I now explore their utility. What does it matter whether one lives in a neighborhood that is closed and dense with short paths? I argue that it matters because one’s behavior is observable throughout the neighborhood, as are one’s reactions to the behaviors of others. Thus, one’s norms and values are readily transmitted, and the norms and values of others are readily accessible. I have argued that neighbor networks are the mechanism translating individuals and their social norms and values into neighborhood communities. In this section, I explore this mechanism by focusing on four particular neighborhood values: whether neighbors share one’s values, whether people are the best part of the neighborhood one lives in, whether one lives in a safe neighborhood, and whether the neighborhood one lives in is “good for children.” I explore whether one’s values are similar to those of neighbors. Table 12.1 shows that there are correlations of .96, .86, .76, and .75, respectively, between one’s own opinions and those of one’s immediate neighbors about whether the neighborhood is safe, whether people are the best part of the neighborhood, whether neighbors share one’s values, or whether the neighborhood is a good one for families with children. Thus, in one of the typical 10-step interview chains conducted in the first set of 68 Los Angeles samples, these would be the correlations of the fifth interviewee with the fourth and the sixth interviewees, who 166 INFLUENCE NETWORKS IN A COLLEGE TOWN TABLE 12.1 Similarity in Residents’ Beliefs about Neighborhood, by Neighbor Network Steps Separating Them 1 step 2 steps 3 steps away in the away in the away in the interview interview interview chain chain chain 4 or more steps away in the interview chain Safe neighborhood .96*** .92*** .88*** .84*** People are best part of this neighborhood .86*** .74*** .64*** .55*** Neighbors share my values .76*** .58*** .45*** .34** Good neighborhood for families with children .75*** .57*** .42*** .32** ** p < .01 *** p < .001. were each one step away. These correlations drop to .92, .74, .58, and .56 between one’s opinions and one’s neighbors’ opinions who are two steps in the interview chain away. Again, in one of the typical 10-step interview chains, these would be the correlations of the fifth interviewee with the third and the seventh interviewees, who were each two steps away. From the perspective of the fifth interviewee in the typical 10step interview chains, three steps away would be the second and the eighth interviewees, while four or more steps away would be the first, ninth, and tenth interviewees. Clearly, one’s perception of one’s neighborhood is very similar to one’s immediate neighbors’ perception, and this similarity of perception concatenates, with only somewhat diminished impact, across neighbor networks, step by step. Especially noteworthy is that, at any distance in the neighbor network, chains of respondents shared a correlation of at least .84 about whether or not it was a safe neighborhood. Not only are one’s perceptions of one’s neighborhood values similar to those of one’s neighbors, but they relate directly to one’s interactions with neighbors. Table 12.2 shows the high correlation between individual-level neighborly relations and one’s general perception of one’s neighborhood. For example, the majority of the variation in whether or not a person believed it was a safe neighborhood could be accounted 167 C H A P T E R T W E LV E TABLE 12.2 Neighborly Interactions and Perception of Neighborhood, All Households Safe neighborhood People are best Neighbors part of this share my neighborhood values Good neighborhood for families with children If I felt unsafe, there is a neighbor I would call before I called the police .90*** .36*** .26*** .49*** There is a neighbor I have given my house keys to .78*** .69*** .51*** .36*** *** p < .001. for by whether she had at least one neighbor whom she would call if she felt unsafe before she called the police, or if there was a neighbor someone had given keys to so that he could let in a service person or one’s children if they were locked out. These neighbor-level activities related powerfully to impressions of the neighborhood, and the converse is true as well. If we focus on households with children, similar powerful correlations appear. The majority of the variation in whether or not one believes it is a good neighborhood for children or believes neighbors share one’s values can be accounted for by whether or not there is a neighbor who has monitored one’s children in a spontaneous playgroup and whether or not there is a neighbor one would not be concerned about disciplining one’s child. Again, these neighbor-level activities related powerfully to impressions of the neighborhood, and the converse is true as well. In short, neighbors influence each other’s beliefs both by their actions and by their interactions. The beliefs and values foundational to neighborhood effects, such as the working trust necessary for the development of collective efficacy, emerge from these networked interactions. This suggests that neighbor networks also affect how successful 168 INFLUENCE NETWORKS IN A COLLEGE TOWN TABLE 12.3 Neighborly Interactions and Perception of Neighborhood, Households with Children Only Good People neighborhood Neighbors are best for families share my part of this Safe with children values neighborhood neighborhood There is a neighbor whom I would not be concerned about disciplining my child .84*** .82*** .79*** .76*** There is a neighbor who has monitored my child in a spontaneous play group .80*** .70*** .28*** .54*** *** p < .001. neighborhood-oriented individuals may be in generating efficacious neighborhood communities. The college town follow-up survey provided an opportunity to directly test the role of influence and to apply Friedkin’s influence model both to determine which geographic feature best proxies the actual influence structure in the neighborhood and to show that consensus in values follows as a result. I was able to do this because I had self-reports of various beliefs about the neighborhood from the same individuals at two different time periods. Thus, I had data on the residents’ change, if any, in their beliefs and values. While there were numerous beliefs and values I could model, I illustrate with two dichotomous beliefs. The first was whether or not the respondent believed that “People are the best part of this neighborhood.” This belief speaks directly to the perceived importance of neighbors. Furthermore, it proved closer to equivalent numbers of affirmations and repudiations than any other statement, thus offering maximal variance to explore. Of the 213 respondents who were interviewed at both time periods, 93 clearly agreed with this statement at time 1. This 169 C H A P T E R T W E LV E number increased to 106 at time 2. While many more respondents agreed with this statement in part, issuing replies such as “They’re one of the best parts” or “There are many good parts in this neighborhood,” for this test I only count those who unequivocally agreed. The other belief I tested was whether or not the respondent believed that “My neighbors share my values.” Of the 213 individuals interviewed in both samples, 153 agreed with this statement at time 1. An almost identical number of people (not necessarily the same people), 154, agreed with this statement at time 2. My question was what these same respondents would say at time 2 and, if they altered their response, what network structure guided the influence at work upon them. Recall that, in Friedkin’s model, one’s beliefs at a particular time, yit−1, are a function of one’s beliefs at an earlier time, yi1, one’s susceptibility (i.e., one’s weighting of one’s own beliefs relative to the beliefs of others), ai, and the social structure guiding the flow of interpersonal influences, w. yit = Ai ∑j wij yjt−1 + (1 − ai) yi1. To proxy the influence structure, I used two different weight matrices. The first weight matrix, w1, I created by the following method. fbij = 1 if and only if person i shares a face block with individual j 0 otherwise w1ij = 1/∑j fbij Thus, the value in the ijth cell was 0 if individuals i and j did not share a face block, which was of course the most common situation, and, if they did share a face block, it was 1 divided by the number of other individuals i shared the face block with. Thus, this model assumed that all those who shared a face block with individual i had equal influence upon individual i and that no one else did. This, of course, is quite an oversimplification, but it was a convenient one to measure. Since the first weight matrix might be expected to form pockets of consensus the size of a face block, the second weight matrix I used expanded the horizon somewhat. It was similar to the first weight matrix except that in this case I assumed that all those who shared a face block with individual i or those who lived on a Face block separated from individual i’s face block by one and only one tertiary intersection 170 INFLUENCE NETWORKS IN A COLLEGE TOWN had equal influence upon individual i and again that no one else did. Thus, tiij = 1 if person i shares a face block with individual j 1 if individual j lives on a face block separated from individual i’s face block by exactly one tertiary intersection 0 otherwise w2ij = 1/∑jtiij While still a dramatic oversimplification, this weighting allows for overlapping spheres of influence. Thus, in this model, influence is less likely to form distinct pockets. Perhaps the most difficult aspect in using this model was estimating how susceptible a respondent would be, or how much a respondent would value her own opinion relative to her neighbors’ opinions. Given the noted relationship between neighborhood effects and residential longevity and given the apparently reasonable intuition that the longer one has lived in a neighborhood the more established one’s opinion of it will be, I used the number of years someone had lived in the neighborhood. I tried to scale the time in which residents had lived in the neighborhood many ways. The most successful in producing results proved to be the natural logarithm of 1 more than the number of years the person had lived in their current residence (I added 1 to avoid the undefined natural logarithm of zero). Having done this, I divided the resulting scores by the maximum value to scale susceptibility between 0 and 1. This scale, however, placed the highest value on those who had lived in the neighborhood the longest and the lowest value on those who had lived there the shortest. My intuitions were, however, that those who had lived there the longest would be least susceptible and that those who had lived there the shortest would be most susceptible. To fix my scale, therefore, I subtracted this result from 1 (so that 1 becomes 0, 0 becomes 1, etc.) to reverse the direction of the scale. Using this method, I produced estimates for each individual’s response at time 2. I was thus able to determine how accurate my estimate was using either method. Of course, perhaps the most widely argued for estimate of how each individual would respond at time 2 was how they responded at time 1. This is equivalent to setting respondents’ susceptibility to 0 (i.e., 171 C H A P T E R T W E LV E FIGURE 12.1 “My neighbors share my values” at times 1 and 2 they cannot be influenced but will maintain their opinions). I use this as a baseline. One hundred and fifty of the 213 respondents gave the same response at time 2 to the query whether neighbors shared their values.3 This evidences a remarkable stability in belief structures within the neighborhood. The influence models, however, made better predictions. Using shared face blocks as a weight matrix estimated the correct response for 175 of the 213 respondents, and using separated by no more than one tertiary intersection produced the correct response for 183 of the 213 respondents. With reference to the query whether neighbors were the best part of the neighborhood, 186 of the 213 respondents gave the same response at time 2 (86 of the 93 individuals who said yes the first time said yes 172 INFLUENCE NETWORKS IN A COLLEGE TOWN FIGURE 12.2 Comparison of models’ predictive power for “My neighbors share my values” again, and 100 of the 120 individuals who said no the first time said no the second time as well). As before, this evidences a remarkable stability in belief structures within the neighborhood. The influence models, however, provided better estimates. Using shared face blocks as a weight matrix in the influence model estimated the correct response for 204 of the 213 individuals at time 2. Finally, using separated by no more than one tertiary intersection as the weight matrix in an influence model produced a near perfect result. It estimated the correct response for all but two of the 213 individuals at time 2. 173 C H A P T E R T W E LV E FIGURE 12.3 “People are the best part of this neighborhood” at times 1 and 2 What is remarkable here is that, while it is well known that people tend to hold persistent beliefs across time, the influence models were able to make not only better, but nearly perfect predictions for their beliefs. Thus, in this case, a better approximation of how a respondent would reply at time 2 than merely what they said at time 1 would take into account the influence network structure, other’s beliefs, and their susceptibility to those beliefs. Clearly, in this case at least, structured influence networks matter in the determination of residents’ beliefs about their fellow neighbors’ values and utility. This could not bear more directly on the concepts of social capital and collective efficacy. One might be concerned that these findings result not so much from influence as from self-selection out of the sample. While only 213 re174 INFLUENCE NETWORKS IN A COLLEGE TOWN FIGURE 12.4 Comparison of models’ predictive power for “People are the best part of this neighborhood” spondents were interviewed at both time periods, this was primarily due to budgetary limits; we simply did not attempt to interview everyone the second time. Furthermore, we have no clear information on how many residents who had been present at time 1 were around to be interviewed at time 2. We do know, however, that those who were interviewed at both time periods did not differ in a statistically significant way from those who were not, either in number of neighbors identified, or by the presence of children in their household, or by race, or by their responses to the prompts “People are the best part of this 175 C H A P T E R T W E LV E neighborhood” or “My neighbors share my values.” If selection occurred, it happened independently of these factors. One final point is worth highlighting. While residents’ beliefs and values conformed to those who shared their tertiary street network, not all neighbors in this study conformed to each other. Recall, from the discussion of the data in chapter 6, that there were three distinct tcommunities in the second college sample (and three distinct elementary school catchments as well). Residents’ beliefs and values conformed to those who shared their particular tertiary street network, but the three t-communities were converging to different values. Even in the short period of three years, it was clear that this process could lead to distinct neighborhood cultures. In the next chapter, I discuss an example where residential stability does in fact lead to extremely consistent norms and values shared among residents and a very distinctive neighborhood culture. MAIN POINTS IN REVIEW In this chapter, I explored the relationship between influence networks and neighborhood-level outcomes in an insular setting, a college town. (I will explore the relationship between influence networks and neighborhood-level outcomes in a distinct insular setting, a gang barrio, in the next chapter.) Exploring the college town in this chapter, I begin by analyzing one particular neighboring relation, trusting each other to watch over children in spontaneous playgroups, and show that it is both dense and short enough to be within the horizon of observability, allowing the behavior of neighborhood children and those who monitor them to be observable to most of the other households with children in the t-community. I provided an example of a particular criminal incident, where the observation of the illicit behavior, the parental response, and the valuation of these behaviors by the neighborhood community was “observed” through influence networks by the residents throughout the t-community but nowhere else. Shared tertiary streets, but not shared elementary school catchments, circumscribed neighborhood collective memory and produced collective efficacy for children. 176 INFLUENCE NETWORKS IN A COLLEGE TOWN I then showed that neighbors influence each other’s beliefs both by their actions and by their interactions. One’s perceptions of one’s neighborhood’s values are both similar to those of one’s neighbors and directly related to one’s interactions with one’s neighbors. The beliefs and values foundational to neighborhood effects, such as the working trust necessary for the development of collective efficacy, emerge from these networked interactions. The structure of influence networks, which was heavily determined by the structure of the tertiary street network, powerfully affected residents’ beliefs about their fellow neighbors’ values and utility. The norms and values that emerged within one t-community, while internally consistent, differed from those that emerged in neighboring t-communities. 177