Each participant named 25 members of her personal network, resulting in a total of 700 named alters. Relatives, which included blood relatives and in-laws (but not spouses), comprised 37.6% of the 700 alters. On average, women named 9.4 relatives. For alters who were not relatives, the interviewer asked how the participant met the person and categorized their response. The most common alter types were individuals met at a shelter (17.3% of all 700 alters; average of 4.3 per woman), met through someone else (e.g., “a friend of a friend”; 13.6% of all alters; average of 3.4 per woman), met on the street (6% of all alters; average of 1.5 per woman), and service providers (4.4% of all alters; average of 1.1 per woman). Nearly 8% of alters were linked some way with the participants’ substance use, either met through their drinking (4.6% of all alters), drug use or selling (1.6% of all alters), or rehab/recovery efforts (2.1% of all alters). Women also mentioned alters met at jobs and school, as neighbors, and through religious and non-religious organizations; each of these categories represented less than 3% of the 700 named alters. On average, women had 5.5 different types of alters in their network (SD = 1.40), which provides some indication of network diversity.
shows the order in which different types of alters were named (e.g., alters 1–5, 6–10, 11–15, 16–20, 21–25). For presentation clarity, only the alter types comprising at least 5% of alters at any of the five naming positions are shown. There was a linear inverse association between emotional closeness and the order in which alters were named, likely a function of the instructions women were given for naming alters (i.e., to begin with the people who are most important and then “work outwards”): the average emotional closeness rating for the first five alters was 4.41 (on a scale from 1–5), but declined to 2.95 for the last five alters named. As shown in , relatives were more likely to be named at the beginning than toward the end of the task, comprising 65% of the group named in positions 1–5, but only 30% of the group named in positions 16–20. Service providers and individuals met in a shelter were more likely to be named as the task went on, comprising 1% and 8% (respectively) of the group named in positions 1–5, but 8% and 23% (respectively) of the group named in positions 16–20. The prompt that occurred after naming the 20th alter corresponded with a increase in alters who were met on the street or through somebody else.
Relationship Quality with Alters
To explore the relationship between various types of alters and the characteristics of their relationships with the women, we conducted several correspondence analyses on multiway data formed by stacking multiple two-way contingency tables (Weller & Romney, 1990
). We examined the graphs for patterns in the relationship between alter types and other categorical variables (Watts, 1997
). Note that duration of the relationship (standardized as number of months) was not included in the correspondence analysis because it is measured on an interval level and thus is not appropriate for this categorical analysis technique. shows results from the first correspondence analysis, which is a graphical display of the first two non-trivial factors in the association between alter types and four ordinal variables measuring relationship characteristics: frequency of contact, emotional closeness, tangible support, and disagreements. The variance explained by these first two non-trivial factors is 64.5%.
Figure 2a. Types of Alters by Relationship Characteristics
Each relationship category and type of alter are displayed in as labeled points, with each point representing either a column or row in the “Type of Alter by Relationship Characteristics” contingency table. The graph is interpreted by noting the points that are either close together or far apart, clusters of points, and distributions of points along a dimension. The objective of the analysis is to generate insight into patterns of associations among the variables that can be tested for statistical significance with other techniques. indicates that frequency of contact and emotional closeness are the key relationship characteristics associated with alter type. There is an obvious dimension of higher to lower frequency of contact from the upper left hand corner to the lower right hand side of the graph. Alters who are most associated with high contact are those met through rehab/recovery programs, met in homeless shelters, and (to a lesser extent) service providers. Alters who are most associated with high emotional closeness dimension are relatives and those met through religious organizations. Most of the alter types are clustered in the quadrant that represents low emotional closeness and low frequency of contact. The high and low tangible support and disagreement categories are clustered almost on top of each other in the center of the graph, indicating that they are not strongly associated with the alter types in this analysis.
Correspondence analysis is an exploratory, descriptive technique that cannot determine statistically significant associations and thus we conducted bivariate ordered logistic regression analyses to test the associations between alter type and the relationship quality and length variables (and thus confirm the significance of the associations noted in the correspondence analysis). Each alter type was compared with all other alters to determine if they were significantly more or less likely to be rated as having higher quality relationships, more contact and longer relationships. We used the SAS procedure PROC GENMOD with a multinomial distribution and a cumulative logit link option to test the null hypothesis that one type of alter was not more likely to be rated higher or lower on these ordinal response options than alters who were in some other category. Below we highlight group differences that are statistically significant at p < .05.
On average, women had contact with the named alters once a week (M = 5.00). Confirming results from the correspondence analysis, women had more frequent contact with alters met through a rehab/recovery program (M = 7.13), alters met in a shelter (M = 7.68), and service providers (M = 6.48) compared to other alters. Women had relatively less contact with relatives (M = 4.01) and alters met through somebody else (M = 3.89), school (M = 2.53), or using or selling drugs (M = 3.09). Overall, participants felt “somewhat” emotionally close to their network members (M = 3.53). Closeness was higher with relatives (M = 4.19) and individuals met through religious organizations (M = 4.25), but lower with individuals met at a shelter (M = 3.21), through somebody else (M = 3.07), on the street (M = 2.71), through alcohol use (M = 2.88), or as a service provider (M = 2.29). The overall frequency of tangible support (M = 1.93) and disagreements (M = 1.76) from network members was “almost never.” Alters met through using alcohol provided particularly low levels of tangible support (M = 1.34) and disagreements (M = 1.16) compared to other alters. Finally, women knew the non-relative members of their network for an average of 3.91 months. Longer-term relationships were those with alters met through somebody else (M = 6.47), as a neighbor (M = 13.89), or at school (M = 11.28), whereas shorter-term relationships were with alters met at a shelter (M = 1.05) or through the participants’ alcohol use (M = 1.54).
Perceived Risk Behavior of Alters
Women were asked their perceptions of whether each alter was likely to have drunk to intoxication, used drugs, or engaged in risky sex during the past 6 months. In addition, they were asked to identify those alters whom they knew had ever been to jail or prison. On average, women had 11.71 members of their network who could be considered “low risk” – that is, perceived by the women as abstaining from drug use and heavy drinking, not engaging in risky sex, and not having a history of incarceration. Nonetheless, the vast majority of women named at least one alter who was perceived as “extremely likely” or “somewhat likely” to drink to intoxication (93%), use drugs (96%), or have risky sex (93%). More than one-half of the women (57%) named at least one alter who had been incarcerated. About one-half of women reported that 8 or more alters, nearly one-third of their personal network, were “extremely likely” or “somewhat likely” to drink to intoxication (54%), use drugs (46%), or have risky sex (57%) during the past 6 months. Eighteen percent of women named 8 or more alters with a history of incarceration. Among the 700 named alters, about one-third were perceived as “extremely likely” or “somewhat likely” to drink to intoxication (32%), use drugs (31%), and have risky sex (31%).
We conducted another correspondence analysis similar to that described in the previous section, with the goal of understanding perceptions of risky behaviors and incarceration history as being normative among certain types of alters. is the graphical output of a the first two non-trivial factors from a correspondence analysis on the “Types of Alters by Participant Rated Likelihood to Engage in Risky Behavior and Incarceration History” multiway data created from stacked two-way contingency tables. These first two factors explain 73.8% of the variance and the graph shows a clear division between two clusters of variables. On the left hand side, the four non-risky alter attributes are closely clustered together. Surrounding them are categories of alters that appear to be strongly associated with an absence of risky behaviors. The risky behavior attributes appear along the right hand side of the graph with a set of alter categories clustering around the drinking, drugs, and risky sex variables. The incarceration variable is somewhat removed from the rest of the risk variables. The only two alter-type variables that do not appear strongly associated with either the risky or non-risky clusters are variables representing alters met through drinking and alters met through rehab/recovery. In addition to the risky/non-risky clustering, there appears to be a dimension of incarceration stretching from the upper right hand to the lower left hand of the graph and a dimension that is mostly driven by drinking and non-drinking alter behaviors stretching from the lower right hand side to the upper left hand side. If two lines were drawn to represent these dimensions, most alters would fall into either the drinking and incarceration quadrant or the non-drinking, no incarceration quadrant. Alters who were met through drinking and rehab are the only alters in the upper left hand quadrant representing non-drinking and incarceration.
Findings from a bivariate ordered logistic regression confirmed that certain alter types were consistently rated as more or less likely to engage in risky behaviors across the behavior types. Overall, alters were rated as somewhat unlikely to engage in heavy drinking (M = 1.84), drug use (M = 1.84), and risky sex (M = 1.81), and 14.6% of all alters were believed to have a history of incarceration. Alter types perceived as being less likely to engage in each of these behaviors compared to other alters included relatives (M = 1.68, 1.54, 1.43; 6.1%, respectively), service providers (M= 1.19, 1.10, 1.21; 3.2%, respectively), and alters met through religious organizations (M = 1.0, 1.0, 1.2; 0%, respectively). In contrast, alters met through somebody else (M = 2.47, 2.61, 2.54; 30.5%, respectively) were perceived as more likely to engage in each of these behaviors. For other alter types, the associations with risk behavior were less consistent: alters met on the street were perceived as more likely to engage in drug use (M = 2.34) and risky sex (M = 2.85), and have been incarcerated (28.6%), than other alter types; however, they were not significantly more or less likely to engage in heavy drinking. Alters met through the participants’ alcohol use were perceived as less likely to use drugs (M = 1.23), but more likely to have an incarceration history (34.4%). Alters met through a job were perceived as more likely to engage in heavy drinking (M = 2.60). Finally, those met as a neighbor were perceived as more likely to engage in heavy drinking (M = 2.78) and risky sex (M = 2.83), where as those met through using/selling drugs were perceived as more likely to engage in drug use (M = 2.64) and risky sex (M = 2.82).
Women’s Engagement in Risk Behavior with Alters
Fifty-seven percent of the women drank alcohol with at least one person in their network during the past 6 months and 11% drank with roughly one-third (i.e., 8 or more) of their network members. The average number of drinking partners was 3.0. In the case of drugs, 39% of the women used with at least one person and 21% used with roughly one-third of their network members during the past 6 months. The average number of drug-using partners was 3.32. Overall, 11% of the named alters had used alcohol and 13% had used drugs with the participant during the past 6 months. In terms of sex partners, all women named at least one sex partner during the network elicitation task, with an average of 3.7 sex partners named by each woman. One-quarter of women had engaged in substance use with a sex partner in the past 6 months.
shows the first two non-trivial factors resulting graph from a correspondence analysis of the “Type of Alter by Risk Behavior With Participant” multiway contingency data. The factors represented in the graph depict 86.2% of the variance. There is a clear division between the risk and non-risk relationships, with the risk relationships (drug use partner, drinking partner, sex partner) falling on the right-hand side of the graph and the non-risk relationships (not a drug use partner, not a drinking partner, not a sex partner) falling on the left-hand side of the graph. The non-risk relationships are clustered with a group of alter types that includes relatives and service providers, as well as people met through homeless shelters, rehab/recovery programs, religious organizations, and the participants’ alcohol use. Among the risk relationship variables, there is a clear contrast between the “drug use partner” and “drinking partner” variables and the “sex partner” variable suggesting a dimension of types of risk from substance abuse to risky sexual behavior. The alters met through the participants’ drug use are located highest on the substance use end of this dimension and the alters categorized as “Other” fall furthest along the sexual activity end of this risk behavior dimension. Most of the remaining alter types cluster between the substance abuse and sex partner ends of this dimension.
Results of a chi-square test of independence between the alter type variables and the dichotomous risk behavior variables for the most part confirmed that the alter types that appeared to be associated with risk or non-risk behaviors in were significantly related. Overall, 48.1% of non-relative male alters were sex partners, 11.4% of alters drank alcohol with the participant, and 13.3% of alters used drugs with the participant. Alters classified in the “other” category (100%) were significantly more likely than other alters to be a sex partner, whereas service providers (5.9%) were significantly less likely than other alters to be a sex partner. Women were more likely to use alcohol and drugs with alters met through somebody else (24.2% and 29.5%, respectively) and met on the street (21.4% and 45.2%, respectively) compared to other alters. Women were also more likely to drink with alters met as neighbors (38.9%), and to use drugs with alters met through using/selling drugs (36.4%) compared to other alters. Relatives and alters met at a shelter were less likely than other alters to use alcohol or drugs with the participant (relatives: 6.1% in each case; met in shelter: 5.8% and 7.4%, respectively), and those met through alcohol use and as service providers were less likely than other alters to use drugs with the participant (0% in each case).