Neighborhood Poverty as a Predictor of Intimate Partner Violence Among White, Black, and Hispanic Couples in the United States

Neighborhood Poverty as a Predictor of Intimate Partner Violence Among White, Black, and Hispanic Couples in the United States

Neighborhood Poverty as a Predictor of Intimate Partner Violence Among White, Black, and Hispanic Couples in the United States: A Multilevel Analysis ...

170KB Sizes 0 Downloads 18 Views

Neighborhood Poverty as a Predictor of Intimate Partner Violence Among White, Black, and Hispanic Couples in the United States: A Multilevel Analysis CAROL B. CUNRADI, PhD, MPH, RAUL CAETANO, MD, PhD, CATHERINE CLARK, PhD, AND JOHN SCHAFER, PhD

PURPOSE: This study assessed the contribution of neighborhood poverty, measured at the census tract level, to the risk of male-to-female and female-to-male partner violence (MFPV, FMPV) among white, black, and Hispanic couples in the United States. METHODS: As part of the 1995 National Alcohol Survey, a representative sample of married/cohabiting couples was obtained through a multistage, multicluster household probability sampling frame. The outcome variables, MFPV and FMPV, were measured through the Conflict Tactics Scale, Form R. Sociodemographic, psychosocial, and alcohol consumption covariates that were statistically significant through bivariate analysis were retained as individual-level predictors. Neighborhood poverty, indicating residence in a census tract where greater than 20% of the population lived below the Federal poverty line, was assessed by appending 1990 Census data to the primary data set. Multilevel logistic regression models were constructed, with separate analyses performed for each outcome (MFPV, FMPV) among the white, black, and Hispanic couples. RESULTS: Couples residing in impoverished neighborhoods are at increased risk for both MFPV and FMPV. The association between residence in an impoverished neighborhood and MFPV was statistically significant for black couples (Odds Ratio [OR] 2.87; 95% Confidence Interval [CI] 1.36, 6.07). The association between residence in an impoverished neighborhood and FMPV was statistically significant for black couples (OR ⫽ 2.35; 95% CI 1.18, 4.71) and white couples (OR ⫽ 4.12; 95% CI 1.94, 8.75). CONCLUSIONS: Characteristics of the socioenvironment, such as neighborhood poverty, are associated with the risk of partner violence, particularly among black couples. Policies aimed at reducing community poverty may contribute to effective partner violence prevention strategies. Ann Epidemiol 2000;10:297–308.  2000 Elsevier Science Inc. All rights reserved. KEY WORDS:

Poverty Areas, Domestic Violence.

INTRODUCTION Spousal violence research over the past 30 years has identified a complex array of individual-, household-, and societallevel factors associated with its occurrence (1–4). While some researchers have analyzed the role of macro-level correlates (e.g., indices of gender inequality at the state level, societal norms approving marital aggression) (5–7), most studies have focused on the sociodemographic (e.g., age, income, education level) and psychological characteristics (e.g., level of marital discord, verbal aggression) of the perpetrator and/or victim of partner violence without explicitly

From the Alcohol Research Group, Berkeley, CA (C.B.C., C.C.); Division of Public Health Biology and Epidemiology, University of California, Berkeley, CA (C.B.C.); University of Texas, School of Public Health, Houston, TX (R.C.); MPH Program at University of Texas, Southwestern Medical Center, Dallas, TX (R.C.); Department of Psychology, University of Cincinnati, Cincinnati, OH (J.S.). Address requests for reprints to: Carol Cunradi, Prevention Research Center, 2150 Shattuck Avenue, Suite 900, Berkeley, CA 94704. Received July 14, 1999; accepted April 4, 2000.  2000 Elsevier Science Inc. All rights reserved. 655 Avenue of the Americas, New York, NY 10010

modeling the role of socioenvironmental or ecologic factors. For example, Sorenson et al., in their multivariate analysis of data from the National Survey on Families and Households, found that urban dwellers were 40% more likely to report physical violence in their marriage in the previous year than suburban dwellers (8). In their study, there were no a priori assumptions concerning the possible association between characteristics of the urban environment (e.g., population density, levels of crime) and the occurrence of partner violence. Rather, residence in an urban (or suburban/rural) area was treated as an individual sociodemographic characteristic. A growing body of epidemiologic research over the last 10 years has investigated the association between socioenvironmental factors and various health outcomes and behaviors (9–24). In particular, community socioeconomic characteristics, such as levels of neighborhood poverty and area unemployment, have been linked to numerous individuallevel health outcomes. For example, residence in a socioeconomically disadvantaged area (typically measured at the census tract level) has been shown to be associated with increased risk of having a low birthweight infant (19); alco1047-2797/00/$–see front matter PII S1047-2797(00)00052-1

298

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

Selected Abbreviations and Acronyms MFPV ⫽ Male-to-female partner violence FMPV ⫽ Female-to-male partner violence NAS ⫽ National Alcohol Survey PSU ⫽ Primary Sampling Unit SRS ⫽ Simple random sample

AEP Vol. 10, No. 5 July 2000: 297–308

of neighborhood poverty (measured at the census tract level) to the risk of partner violence among a national sample of white, black and Hispanic couples.

METHODS Sampling

hol-related problems among black men (22); initiation of sexual activity during adolescence (17); household criminal victimization (14, 15); prevalence of coronary heart disease and coronary risk factors (23); and all-cause mortality (9, 10). The publication of these studies has been paralleled by articles and monographs highlighting the utility and methodology of contextual models (25–29) that seek to incorporate community or aggregate-level characteristics into a multilevel framework of individual- and group-level predictors of individual risk (25). While there is considerable evidence that individualand household-level indicators of low socioeconomic status (measured through unemployment, blue-collar occupational status, and level of education and income) are associated with intimate partner violence (8, 30–34), few studies have explored whether couples residing in impoverished neighborhoods are at greater risk for partner violence than couples residing in more affluent areas. A recent ecologic study by Miles-Doan and Kelly that examined rates of police-reported partner violence within 131 census tracts in Duval County, Florida found that median rates of partner violence were nine times higher in concentrated poverty tracts than in non-poverty tracts (35). O’Campo and colleagues (36) analyzed neighborhood- and individual-level predictors of male partner violence during the childbearing year using data from 157 women residing within 76 census tracts in Baltimore, Maryland. Their findings indicate that women residing in census tracts in the lowest quartile of per capita income were four times more likely to report partner violence than women residing in census tracts in the highest quartile of per capita income. An association of similar magnitude was found for women residing in census tracts characterized by high vs. low unemployment rates. These effects were independent of individual-level income. Although they were confined to discrete geographic areas, these two studies provide limited evidence that residence in an impoverished neighborhood may be associated with the occurrence of intimate partner violence. The present study, based on a national probability sample of married and cohabiting couples in the 48 contiguous states, seeks to test the hypothesis that couples who reside in impoverished neighborhoods are at increased risk for partner violence, after controlling for individual- and household-level factors, compared to couples residing in non-impoverished neighborhoods. Specifically, this study aims to assess the contribution

Subjects were selected through a multistage area household probability sampling procedure from individuals 18 years of age or older living in households in the 48 contiguous states. Only married or cohabiting couples were included for the final stage of selection. The sample had 100 PSUs (primary sampling units based on counties or groups of counties), and included over-samples of black and Hispanic couples. There were 1925 eligible couples selected for the survey; 1635 couples participated for a response rate of 85%. As the aim of this study was to examine the association between neighborhood poverty and partner violence among white (n ⫽ 555), black (n ⫽ 358) and Hispanic (n ⫽ 527) couples, the data analyzed in this paper are based on interviews with the couples who self-identified as belonging to those racial/ ethnic groups. The final sample was thus comprised of 1440 couples. Data Collection Main respondents were interviewed as part of the 1995 National Alcohol Survey (NAS). Detailed information regarding the NAS is provided elsewhere (37, 38). One-hour, face-to-face interviews were conducted in respondents’ homes with standardized questionnaires. Spouses or cohabiting partners were interviewed separately and were administered a brief version of the main respondent questionnaire that averaged approximately 20 minutes. Hispanic respondents were given a choice of being interviewed in English or Spanish. Separate independent variables were created for male and female respondents; joint household variables were created for income, number of children, and relationship length (described below). Each couple’s residential address was geocoded, a process that links each address with the census tract number for their state and county. Measurements Intimate Partner Violence. Participants were asked about the occurrence of 11 violent behaviors during the past year that they may have perpetrated against their partners, or that their partners may have perpetrated against them, after being read the following statement: “No matter how well a couple gets along, there are times when they disagree. Couples get annoyed with each other, or just have spats or fights because they’re in a bad mood or tired. They also use many different ways to settle their differences. The following list are things you and your (husband/wife/partner) might

AEP Vol. 10, No. 5 July 2000: 297–308

have done when you had an argument or disagreement.” These items were adopted from the Conflict Tactics Scale, Form R (39), and include: threw something; pushed, grabbed, or shoved; slapped; kicked, bit, or hit; hit or tried to hit with something; beat up; choked; burned or scalded; forced sex; threatened with a knife or gun; used a knife or gun. Due to survey time constraints, no frequency data were collected. Given this limitation, intimate partner violence was operationalized as a dichotomy, with separate variables created for male-to-female (MFPV) and female-to-male partner violence (FMPV). Violence was considered to have occurred if at least one partner reported a violent incident in the past year, regardless of whether the incident was corroborated by the other partner. Thus, if either partner reported the occurrence of a violent incident, the partner violence variable (depending on the gender of the perpetrator) was coded “1” if neither reported an incident, the variable was coded “0.” Individual- and Household-Level Variables Sociodemographic Variables. Racial/Ethnic Identification. Respondents who selected “black of Hispanic origin (Latino, Mexican, Central or South American, or any other Hispanic origin)” and “white of Hispanic origin (Latino, Mexican, Central or South American, or any other Hispanic origin)” were classified as Hispanic. Respondents who selected the category “black, not of Hispanic origin” were classified as black. Subjects who selected “white, not of Hispanic origin” were classified as white. Age. Years of age was recorded for each respondent. Due to near collinearity between measurements of husbands’ and wives’ ages (r ⫽ .95), the mean years of age for each couple was calculated and entered into the model as a continuous variable. The absolute difference in years of age between spouses was also entered into the model. Gender. Each respondent was classified as male or female. Income. Total household income was initially recorded into 12 categories: 1) $4000 or less; 2) $4001 to $6000; 3) $6001 to $8000; 4) $8001 to $10,000; 5) $10,001 to $15,000; 6) $15,001 to $20,000; 7) $20,001 to $30,000; 8) $30,001 to $40,000; 9) $40,001 to $60,000; 10) $60,001 to 80,000; 11) $80,001 to $100,000; and 12) more than $100,000. A representative value for each couple was then set at the midpoint of their stated category, except for those reporting incomes greater than $100,000, which was set at $100,000. Household income was then analyzed as a continuous variable. The resulting beta coefficient from the logistic model could then be interpreted as the percent increase or decrease in partner violence per each additional $1000 in household income. Employment status. Male respondents were categorized into 4 employment categories: retired; unemployed; other (disabled, student, employed part-time); and employed (reference group). Female respondents were categorized into 5 employment categories: home-

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

299

maker; retired; unemployed; other; and employed (reference group). Marital status. Couples were classified as either married or cohabiting (reference group). Number of children. The number of children aged 17 years and under residing with the couple was measured as reported. Relationship length. The number of years each couple has lived together was measured as reported by the partner who was the NAS main respondent. Education. Each respondent was asked about the number of completed years of education. Measurements of education between partners were found to be correlated (r ⫽ .67). Therefore, the mean of each couple’s years of education, along with the absolute difference in their years of education, were entered into the model. Psychosocial variables. Childhood violence victimization. Respondents were asked whether or not they had experienced any of the following acts at the hands of their parent or care giver during their childhood or adolescence: hit with something; beaten up; choked, burned or scalded; threatened with a knife or gun; had a knife or gun used against them. A dichotomous variable was created based on a positive response to any of the questions. Those reporting no history of childhood violence served as the reference group. Approval of marital aggression. Each respondent was asked to rate their approval on a 4-point Likert scale to the following behaviors: (1) a husband is acting in a verbally aggressive or verbally abusive way toward his wife; (2) a wife is acting in a verbally aggressive or verbally abusive way toward her husband; (3) a husband is behaving in a physically violent way to his wife; (4) a wife is behaving in a physically violent way to her husband. Those responding “Always approve,” “Sometimes approve,” or “Sometimes disapprove” to any of the behaviors were counted as approving. Those responding “Always disapprove” to all of the behaviors were counted as disapproving (reference group). Due to high correlation between those approving verbal and physical abuse, these categories were combined into a single “approval of marital aggression” category. Impulsivity. Respondents were asked to rate their responses to the following 3 statements on a 4-point Likert scale: (1) I often act on the spur of the moment without stopping to think; (2) You might say I act impulsively; and (3) Many of my actions seem to be hasty. The mean of the responses to these 3 statements formed the impulsivity measure, with higher scores reflecting higher levels of impulsivity. Drinking. Alcohol volume. Alcohol consumption was assessed by combining the self-reported frequency and quantity (reported in number of 4-ounce glasses of wine, 1-ounce shots of distilled spirits, and 12-ounce cans of beer they drank, each of which contains approximately 12 grams of absolute alcohol) to achieve the average total number of drinks consumed weekly. The drinking volume for each spouse was included in the models as a continuous independent variable. Alcohol-related problems. Respondents who were not abstainers were asked about 11 alcohol dependence

300

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

symptoms and 15 drinking-related social consequences they may have experienced in the 12 months prior to the interview. These questions were based on previous survey research measuring alcohol-related problems (40). Alcohol dependence symptoms (specific alcohol-related behaviors, experiences and feelings) are associated with alcohol addiction and physical dependence symptoms, such as withdrawal. Drinking-related social consequences refer to those social, financial, or health-related problems that may have resulted from alcohol consumption. Due to survey time constraints, frequency data were not collected. In addition, the distribution of alcohol dependence symptoms and drinkingrelated social consequences in this population is highly skewed, with most people (80%–95%) reporting no alcoholrelated problems in the previous 12 months [see (41) for alcohol problem prevalence, and a detailed description of alcohol problem survey questions]. For these reasons, as well as the significant correlations (p ⬍0.01) between alcohol dependence symptoms and drinking-related social consequences, these measures were combined into a dichotomous “alcohol-related problems” variable (presence of any alcohol problems vs. no problems) for each spouse. Neighborhood-Level Variables Socioeconomic Variables. Data obtained from the 1990 census (STF3A) (42) were appended to each couple’s geocoded record to describe four census tract-level socioeconomic characteristics: undereducation, unemployment, working-class composition, and poverty. Krieger’s censusbased methodology (18) was employed to define census tract-level undereducation, working-class composition, and poverty. Undereducation. Based on File P58, the percent of the population age 25 years and older with a high school diploma was subtracted from the total population of those 25 years and older to produce the percent without a high school diploma. Tracts where at least 25% of those 25 years and older did not have a high school diploma were characterized as “undereducated.” Unemployment. Based on file P65d, the percent of the population age 16 years and older who were in the labor force but reported being unemployed was used to determine the tract unemployment level. Working-class composition. Based on File P76, the percent of employed persons age 16 years and older who reported being employed in the following occupational categories was used to calculate the overall percent of each tract employed in working class occupations: sales; administrative support (including clerical); private household service; service occupations except protective and household; precision production, craft and repair; machine operators, assemblers, and inspectors; transportation/material moving; handlers, equipment cleaners, helpers and laborers. As Krieger points out, “these occupations . . . disproportionately contain people who can be considered working class (i.e., employees

AEP Vol. 10, No. 5 July 2000: 297–308

who do not own their workplace, are not self-employed, and generally occupy subordinate positions at work” (18, p. 704). The percent of employed persons age 16 years and older who reported being employed in the following occupational categories was used to calculate the overall percent of each tract employed in non-working class occupations: executive, administrative, and managerial; professional specialty; technicians and related support; protective service; farming, forestry, and fishing (including farm owners and managers). Tracts were categorized as working class in which 66% or more of the employed persons age 16 years and older were classified as being in working class occupations. Poverty. Based on File P98, the percent of each tract’s population below the poverty line was used to create a dichotomous “poverty” variable, which was coded as “1” if 20% or more of the tract’s population was below the poverty line, and “0” if less than 20% of the population was below the poverty line. In 1990, the Federal government defined the poverty line at $12,700 annual income for a family of four (43). Correlations between these census tract variables were found to be high. For example, the correlation at the tractlevel between the percent of high school graduates and the percent in poverty was found to be ⫺.71. Similarly, the correlation between the percent of high school graduates and the percent employed in working class occupations was ⫺.77. The percent unemployed was correlated with the percent of high school graduates at ⫺.70, and with the percent in poverty at .77, respectively. The correlations between the percent in working class occupations and the percent in poverty (r ⫽ .47), and the percent unemployed (r ⫽ .50) were also significant. As the poverty variable appeared to capture the characteristics that describe impoverished neighborhoods (undereducated populations, high unemployment, working class composition), this variable alone was retained for the multilevel analysis as a neighborhood-level predictor of partner violence. Statistical Analysis Weighting. The data were weighted to adjust for the probability of selection into the sample and non-response rates. Post-stratification weights were calculated to adjust the sample to known population distributions on certain demographic variables (ethnicity of the household informant, metropolitan status, and region of the country) (38). Data analysis. Individual- and household-level covariates that were significantly associated with partner violence from previous bivariate and multivariate analyses of the sample (41, 44) were retained for the multilevel models to which a dichotomous census tract-level independent covariate, neighborhood poverty, was added. Separate racial/ethnic group-specific logistic regression models were constructed, with male-to-female and female-to-male partner violence (MFPV, FMPV) as the dependent variables.

AEP Vol. 10, No. 5 July 2000: 297–308

Analysis of data obtained through a multistage, multicluster sample design presents a number of unique problems. First, data sampled from clusters are typically characterized by some degree of homogeneity, which tends to increase the variance of the sample. In other words, people sampled from the same cluster tend to be more alike than people randomly sampled from the population. This measure of homogeneity can be measured by rho (␳), the intraclass correlation coefficient (45, p. 161). Kalton explains that “clustering leads to a loss of precision compared with an SRS (simple random sample) of the same size whenever the cluster intraclass correlation coefficient ␳ is positive, as is almost always the case” (46, p. 77). Analyzing the data under the assumptions of simple random sampling, without taking the complex sample design into account, may result in misestimation (usually underestimation) of the sampling error (46, p. 75; 38, p. 41). The SUDAAN computer software program is specifically designed for the analysis of stratified, multistage cluster samples, and makes use of the first-order Taylor series approximation method to estimate the standard errors (47, p. 1–5). Second, appending census tract-level variables derived from the 1990 Census introduces another level of intraclass correlation to the analysis. This can be represented by the following equation: PV ⫽ ␣ ⫹ ␤(NPOV) ⫹ b(x) ⫹ e ⫹ ⑀ where PV ⫽ partner violence (dependent variable), NPOV ⫽ neighborhood poverty (measured at the census tract level), x ⫽ individual covariate(s), e ⫽ individual error, and ⑀ ⫽ group error While the SUDAAN program can adjust for the intraclass correlation at the strata and PSU level, it does not adjust for clustering within census tracts. To test for higherorder random effects, each racial/ethnic group-specific logistic regression model was initially analyzed in the multilevel computer software program HLM for Windows (48), which is designed for data having an hierarchical structure (i.e., individuals nested within census tracts). The analysis of each nonlinear hierarchical model produced a chi-square statistic for the final estimation of variance components, which indicates the proportion of variance explained by higher order variation. This chi-square statistic is the deviance explained in the model by assuming normally distributed random effects for higher order terms in the equation. It is compared with the deviance accounted for by the nonhierarchical simple fixed effects model. The test of random effects was non-significant for all of the models (data not shown). Although the intraclass correlation at the census tract-

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

301

level is explicitly analyzed within HLM, lower levels (i.e., strata and PSU level) of intraclass correlation resulting from the multistage cluster sampling frame are not accounted for within this program. In addition, the HLM program, in its analysis of nonlinear hierarchical models, cannot weight the data to adjust for unequal probability of selection into the sample, non-response, or post-stratification. For these reasons, the final analysis was performed using the SUDAAN version 7.0 PROC LOGISTIC module. Interaction. It was hypothesized that the association of certain genderspecific covariates with partner violence might vary depending upon the presence of a third variable (e.g., the spouse’s corresponding characteristic). Thus, within each racial/ethnic-specific model, interaction effects were tested by including cross product terms representing the joint occurrence of the following: male and female alcohol-related problems, male and female history of childhood violence, and male and female approval of marital aggression. To avoid Type II errors, the significance level for these tests was set at ␣ ⫽ 0.20 (49, p.188).

RESULTS Bivariate Association Between Neighborhood Poverty and Partner Violence The study found significant differences across racial/ethnic groups in the proportion of couples living in impoverished neighborhoods (␹2 ⫽ 44.25, 2df, p ⬍ 0.0001). Approximately half of the black and Hispanic couples (47% and 54%, respectively) resided in impoverished neighborhoods as measured through the 1990 Census. Only 11% of white couples, however, resided in such neighborhoods. The pairwise association between the occurrence of partner violence and residence in an impoverished neighborhood was then assessed. The results (Table 1) indicate that both male-tofemale partner violence and female-to-male partner violence were significantly associated with residence in a poverty area among black couples only. Predictors of Male-to-Female Partner Violence The results of the racial/ethnic group-specific multivariable logistic regression analyses revealed that couples who resided in impoverished neighborhoods were at greater risk for MFPV than couples who did not reside in impoverished neighborhoods (Table 2). The strength of these associations, however, varied across racial/ethnic group-specific models as described below. White Couples. White couples residing in impoverished neighborhoods had a 70% higher risk of MFPV compared to those couples residing in non-poverty areas, but the association was not statistically significant. Further, most household- and individual-level socioeconomic characteristics

302

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

AEP Vol. 10, No. 5 July 2000: 297–308

TABLE 1. Bivariate associations between poverty, area residence, and partner violence White (n ⫽ 555)

MFPV No MFPV ␹2 FMPV No FMPV ␹2

Black (n ⫽ 358)

Hispanic (n ⫽ 527)

Poverty area (n ⫽ 63)

Non-poverty area (n ⫽ 492)

Poverty area (n ⫽ 167)

Non-poverty area (n ⫽ 191)

Poverty area (n ⫽ 261)

Non-poverty area (n ⫽ 266)

10% 90%

12% 88%

34% 66%

13% 87%

20% 80%

14% 86%

15% 85%

41% 59%

21% 79%

23% 77%

12.87a

0.19 21% 79%

3.49

12.35a

1.55

19% 81% 1.61

p ⬍ 0.001 Percentages are rounded to the nearest tenth.

a

(income, education, employment status) were not linked with an increased risk of MFPV. Rather, a number of individual-level female characteristics were found to be predictive. For example, couples in which the female reported being a homemaker had a lower risk for MFPV compared to white couples in which the female reported being employed (p ⬍ 0.01). In addition, white couples in which the female approved of marital aggression had a fourfold risk of

MFPV compared to couples in which the female disapproved of marital aggression (p ⬍ 0.01). White couples who reported female alcohol-related problems had a twofold risk for MFPV compared to couples who did not report female alcohol-related problems (0.05 ⭐ p ⬍ 0.10). Interaction between female and male approval of marital aggression was tested and found to be non-significant (p ⬎ 0.20); the term was subsequently dropped from the model. Interaction terms

TABLE 2. Predictors of male-to-female partner violence among married and cohabiting couples: Odds Ratios (95% Confidence Intervals) Covariate Neighborhood poverty Household income (per $1000) Married vs. cohabiting No. children ⬍ 17 at home Number years lived with partner Couple mean age (years) Couple age difference Couple mean education (years) Couple education difference Female employment Homemaker Retired Unemployed Other Female childhood violence Female approval marital aggression Female alcohol related problems Female alcohol volume Female impulsivity Male employment Retired Unemployed Other Male childhood violence Male approval marital aggression Male alcohol related problems Male alcohol volume Male impulsivity Interactions Male/female alcohol related problems Male/female approval marital aggression a

White

Black

1.68 0.99 0.90 1.18 0.99 0.96 0.91 1.03 1.12

(0.47, (0.97, (0.34, (0.81, (0.95, (0.91, (0.78, (0.87, (0.93,

6.00) 1.008) 2.37) 1.73) 1.04) 1.005)e 1.07) 1.21) 1.36)

3.09 0.99 0.39 1.06 0.95 1.01 1.07 0.83 1.07

(1.35, (0.98, (0.13, (0.81, (0.89, (0.94, (0.99, (0.65, (0.82,

0.23 1.21 1.01 0.60 1.61 4.88 2.57 1.03 0.99

(0.08, (0.11, (0.19, (0.10, (0.87, (1.57, (0.98, (0.99, (0.62,

0.63)b 12.92) 5.40) 3.79) 2.97) 15.15)b 6.77)d 1.06) 1.58)

0.92 0.01 1.45 0.76 4.00 1.98 4.42 1.02 1.20

0.39 1.61 1.26 1.63 1.86 1.88 0.99 1.33

(0.05, (0.32, (0.15, (0.77, (0.71, (0.87, (0.97, (0.88,

3.21) 7.95) 10.27) 3.42) 4.85) 4.07) 1.01) 2.02)

0.25 0.13 0.37 1.40 5.37 7.19 0.99 0.96

— —

p ⬍ 0.05; b p ⬍ 0.01; c p ⬍ 0.001; d 0.05 ⭐ p ⬍ 0.10; e p ⫽ 0.0717; f p ⫽ 0.1347

7.04)b 1.02) 1.20) 1.40) 1.01) 1.07) 1.16)d 1.06) 1.41)

Hispanic 1.34 0.97 0.53 0.97 1.04 0.94 1.03 1.05 0.92

(0.71, (0.95, (0.28, (0.79, (0.99, (0.89, (0.97, (0.89, (0.75,

(0.27, 3.22) (0.001, 0.12)c (0.38, 5.56) (0.11, 5.36) (1.69, 9.47)b (0.67, 5.87) (1.34, 8.73)a (0.98, 1.05) (0.83, 1.73)

0.93 0.02 0.65 1.82 1.62 0.42 0.17 0.90 1.79

(0.47, 1.85) (0.004, 0.10)c (0.24, 1.79) (0.51, 6.52) (0.84, 3.09) (0.11, 1.64) (0.02, 1.50) (0.79, 1.03) (1.12, 2.86)a

(0.01, (0.03, (0.05, (0.54, (2.64, (3.05, (0.98, (0.50,

2.02 2.17 1.12 1.57 1.16 1.48 1.01 1.37

(0.31, (0.96, (0.45, (0.81, (0.56, (0.61, (0.99, (0.86,

— —

4.98) 0.60)b 2.86) 3.66) 10.91)c 16.95)c 1.004) 1.84)

2.54) 0.99)a 0.98)a 1.20) 1.08) 0.99)a 1.09) 1.24) 1.13)

13.25) 4.89)e 2.81) 3.04) 2.41) 3.59) 1.03) 2.17)

8.47 (0.83, 86.90)e 6.87 (0.55, 86.25)f

AEP Vol. 10, No. 5 July 2000: 297–308

for white couples reporting the presence of female and male childhood violence victimization and the presence of both male and female alcohol-related problems were found to be of marginal significance (p ⫽ 0.17 and p ⫽ 0.19, respectively). As these p-values are likely to represent the result of random variation, both of these interaction terms were dropped from the model of MFPV among white couples. Black Couples. Black couples who resided in poverty areas were at a threefold risk for MFPV compared to black couples not residing in poor areas (p ⬍ 0.01). Most household- and individual-level socioeconomic characteristics were not strongly associated with the outcome. Two employment categories, however, were associated with a decreased risk of MFPV. For example, black couples in which the female reported her employment status as “retired” were at decreased risk for MFPV compared to black couples in which the female reported being employed. Interestingly, black couples in which the male reported an employment status of “unemployed” were at reduced risk (p ⬍ 0.01) of MFPV compared to black couples in which the male reported being employed. As noted in a previous analysis (41), female and male alcohol-related problems were both significant predictors of MFPV among black couples. Black couples who reported female alcohol-related problems had a threefold risk of MFPV compared to black couples not reporting female alcohol-related problems (p ⬍ 0.01). Those reporting male alcohol-related problems had a sevenfold risk of MFPV compared to black couples not reporting male alcohol-related problems (p ⬍ 0.001). Two psychosocial variables were significant predictors of MFPV among black couples. For example, black couples in which the female reported childhood violence victimization had a fourfold risk of MFPV compared to black couples in which the female did not report childhood violence (p ⬍ 0.01). Additionally, black couples in which the male approved of marital aggression had a fivefold risk of MFPV compared to couples in which the male disapproved of marital aggression (p ⬍ 0.001). Possible interaction between male and female history of childhood violence victimization, male and female approval of marital aggression, and male and female alcohol-related problems were explored through cross-product terms, and were found to be nonsignificant (p ⬎ 0.20). These terms were then dropped from the final model of MFPV among black couples. Hispanic Couples. Among Hispanic couples, household income rather than neighborhood poverty was found to be a significant predictor of MFPV. For example, each $1000 increase in reported household income was associated with a 3% decrease in risk of MFPV (p ⬍ 0.05). Hispanic couples in which the male reported being unemployed had a twofold risk of MFPV compared to couples in which the male reported being employed (0.10 ⬍ p ⬍ 0.05). As with black

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

303

couples, Hispanic couples in which the female reported being retired were at significantly less risk for MFPV compared to couples in which the female reported being employed (p ⬍ 0.001). Age also conferred a protective effect; each additional year of the couple’s mean age was associated with a 6% reduction in risk of MFPV (p ⬍ 0.05). Hispanic couples who reported being married were at approximately half the risk for MFPV compared to couples who reported being unmarried but cohabiting (p ⬍ 0.05). The female’s mean impulsivity score was also associated with the occurrence of MFPV among Hispanic couples (p ⬍ 0.05). Interaction between female and male childhood history of violence victimization was explored and found to be nonsignificant (p ⬎ 0.20). There was evidence of interaction between female and male approval of marital aggression, indicating that the association among Hispanic couples between a spouse approving of marital aggression and the occurrence of MFPV was dependent upon whether or not the other spouse approved of marital aggression. Hispanic couples in which both partners approved of marital aggression faced a nearly sevenfold risk for MFPV compared to couples not reporting joint approval of aggression (p ⫽ 0.13). In addition, there was evidence of a significant interaction between male and female alcohol-related problems, indicating that the association among Hispanic couples between a spouse reporting the presence of alcohol-related problems and the occurrence of MFPV depended upon whether or not the other spouse reported having alcoholrelated problems. Hispanic couples reporting the presence of both male and female alcohol-related problems were over eight times at risk for MFPV compared to couples not reporting the presence of joint alcohol problems (0.05 ⭐ p ⬍ 0.10) Predictors of Female-to-Male Partner Violence Residence in an impoverished neighborhood was associated with the occurrence of female-to-male partner violence among white, black and Hispanic couples (Table 3). Here again, the strength of the association varied across racial/ ethnic-specific models, as described below. White Couples. White couples residing in an impoverished neighborhood had a near fourfold risk of FMPV compared to couples not residing in poor neighborhoods (p ⬍ 0.001). Numerous household- and individual-level sociodemographic characteristics were also predictive of FMPV. For example, white couples in which the male reported being unemployed were at a threefold risk for FMPV compared to white couples in which the male reported being employed (p ⫽ 0.06). Conversely, white couples in which the female reported being unemployed were at decreased risk for FMPV compared to those couples in which the female reported being employed (p ⫽ 0.048). Although household income was not associated with the occurrence of FMPV among white couples, mean years of educational

304

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

AEP Vol. 10, No. 5 July 2000: 297–308

TABLE 3. Predictors of female-to-male partner violence among married and cohabiting couples: Odds Ratios (95% Confidence Intervals) Covariate Neighborhood poverty Household income (per $1000) Married vs. cohabiting No. children ⬍ 17 at home Number years lived with partner Couple mean age (years) Couple age difference Couple mean education (years) Couple education difference Female employment Homemaker Retired Unemployed Other Female childhood violence Female approval marital aggression Female alcohol related problems Female alcohol volume Female impulsivity Male employment Retired Unemployed Other Male childhood violence Male approval marital aggression Male alcohol related problems Male alcohol volume Male impulsivity Interactions Male/female alcohol related problems Male/female childhood violence a

White

Black c

Hispanic a

3.92 1.01 1.89 0.99 0.98 0.95 0.94 0.81 0.92

(1.83, (0.99, (0.59, (0.70, (0.94, (0.90, (0.85, (0.71, (0.73,

8.40) 1.03) 6.03) 1.39) 1.02) 1.0001)e 1.05) 0.94)b 1.17)

2.32 (1.10, 4.86) 1.002 (0.98, 1.02) 0.51 (0.21, 1.26) 1.19 (0.94, 1.50) 0.97 (0.92, 1.04) 0.95 (0.90, 1.01) 1.09 (1.004, 1.18)a 0.81 (0.67, 0.98)a 1.04 (0.80, 1.35)

1.42 0.98 0.75 0.98 1.01 0.91 0.97 1.21 0.98

(0.85, (0.96, (0.39, (0.81, (0.96, (0.86, (0.86, (1.06, (0.86,

1.27 0.56 0.12 0.64 1.81 4.51 13.42 1.02 1.16

(0.54, (0.06, (0.01, (0.21, (0.96, (1.48, (2.55, (0.98, (0.82,

3.03) 4.97) 0.98)a 1.92) 3.43)e 13.77)b 70.56) 1.05) 1.65)

0.85 0.58 0.70 0.66 2.61 2.24 5.07 1.02 1.14

(0.31, (0.05, (0.20, (0.14, (1.24, (0.86, (1.93, (0.96, (0.76,

2.38) 7.06) 2.50) 3.22) 5.52)a 5.86) 13.33)b 1.07) 1.69)

0.84 0.04 2.03 0.14 4.91 1.77 2.09 0.89 1.32

(0.44, 1.59) (0.005, 0.30)b (0.36, 11.55) (0.02, 1.04)d (2.03, 11.88) (0.59, 5.36) (0.65, 6.71) (0.77, 1.02) (0.82, 2.13)

1.67 3.64 0.26 0.89 1.39 1.44 1.01 1.23

(0.16, (0.95, (0.05, (0.45, (0.70, (0.60, (0.99, (0.86,

17.10) 13.98)d 1.29) 1.76) 2.79) 3.47) 1.02) 1.78)

0.53 1.06 1.65 2.03 2.20 2.44 0.99 0.76

(0.09, (0.23, (0.48, (0.88, (1.14, (1.04, (0.97, (0.44,

3.22) 4.79) 5.60) 4.69) 4.26)a 5.74)a 1.01) 1.31)

2.79 1.86 2.87 2.52 1.19 1.69 1.02 2.07

(0.26, (0.64, (0.83, (1.19, (0.50, (0.80, (0.99, (1.48,

0.15 (0.02, 1.33)e —

— —

2.39) 1.004) 1.44) 1.17) 1.08) 0.97)b 1.05) 1.38)b 1.13)

30.42) 5.40) 9.92) 5.37) 2.82) 3.60) 1.05) 2.88)c

— 0.26 (0.09, 0.72)b

p ⬍ .05; b p ⬍ .01; c p ⬍ .001; d 0.05 ⭐ p ⬍ 0.10; e p ⫽ 0.0891.

attainment was associated with a protective effect. For example, each mean year of education was associated with a 19% decrease in risk of FMPV among white couples (p ⬍ 0.01). In addition, each mean year of couple age was associated with a 5% decrease in risk (p ⫽ 0.051). No evidence of interaction was found for joint male and female approval of marital aggression, or for joint male and female reports of history of childhood violence victimization. These terms were then dropped from the model of FMPV among white couples. Examination of the terms representing main effects revealed that white couples in which the female approved of marital aggression were at fourfold the risk for FMPV compared to couples in which the female disapproved of marital aggression (p ⬍ 0.01). White couples in which the female reported childhood violence victimization were at nearly double the risk for FMPV compared to couples in which the female did not report childhood violence victimization (0.05 ⭐ p ⬍ 0.10). The test for interaction between joint alcohol-related problems and FMPV indicated that white couples in which both partners reported alcohol-related problems were at reduced risk for FMPV

compared to couples in which neither partner reported alcohol-related problems (0.05 ⭐ p ⬍ 0.10). Black Couples. Among black couples, those residing in poor neighborhoods were at a twofold risk for FMPV compared to those not residing in poor neighborhoods (p ⬍ 0.05). Each mean year of education was associated with a 21% decrease in risk of FMPV. Although mean years of age was not a significant predictor of FMPV among black couples, each year of difference in age between spouses was associated with an 8% increase in the risk of FMPV (p ⬍ 0.05). Interaction terms for joint male and female alcoholrelated problems, male and female reports of childhood violence victimization, and male and female approval of marital aggression were tested in the model of FMPV among black couples. No evidence of interaction was observed for any of these variables and they were subsequently dropped from the final model. Significant main effects were then found for many of these individual covariates. For example, black couples in which the female reported having alcohol-related problems had a fivefold risk for FMPV compared to black couples in which the female did not report having alcohol-

AEP Vol. 10, No. 5 July 2000: 297–308

related problems (p ⬍ 0.01). Similarly, black couples in which the male reported having alcohol-related problems were at double the risk for FMPV compared to black couples in which the male did not report having alcohol-related problems (p ⬍ 0.05). Two psychosocial variables that were significantly associated with the occurrence of MFPV among black couples were also associated with the occurrence of FMPV. For example, black couples in which the female reported childhood violence victimization were at double the risk for FMPV compared to black couples in which the female did not report childhood victimization. Likewise, male approval of marital aggression was associated with a twofold risk of FMPV compared with couples in which the male disapproved of marital aggression. Hispanic Couples. Hispanic couples who resided in impoverished neighborhoods were at 40% greater risk for FMPV compared to couples who did not reside in such areas, but the association was not statistically significant. Hispanic couples in which the female reported being retired were at reduced risk for FMPV compared to couples in which the wife reported being employed (p ⬍ 0.01). The couple’s mean age was inversely associated with the risk of FMPV; each mean year of age was associated with a 9% decrease in risk. In contrast, each mean year of education among Hispanic couples was associated with a 21% increase in the risk of FMPV. Among Hispanic couples, mean male impulsivity score was a significant predictor of FMPV (p ⬍ 0.001). The test for interaction between joint female and male reports of childhood violence victimization and FMPV among Hispanic couples also indicated a negative effect. Hispanic couples in which both partners reported childhood violence were at significantly less risk for FMPV compared to couples in which neither reported childhood violence (p ⬍ 0.01).

DISCUSSION The Role of Neighborhood Poverty As noted in a previous analysis of this study population (44), the 12-month prevalence of interpersonal violence among married and cohabiting couples varied significantly by race/ethnicity, with black couples reporting the highest rates, followed by Hispanics and whites, respectively. The current analysis indicates that the contribution of neighborhood poverty to the risk of partner violence also varies by race/ethnicity. Although a positive association was observed within each racial/ethnic group-specific model between residence in an impoverished neighborhood and the occurrence of partner violence, the association was statistically significant only for black couples as a predictor of MFPV and FMPV, and for white couples as a predictor of FMPV. These findings suggest that characteristics of the socioenviron-

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

305

ment, such as neighborhood poverty, are associated with the occurrence of interpersonal violence, but the strength of the association may vary across communities. A number of sociodemographic factors, including race/ethnicity, household socioeconomic status, and the gender of the perpetrator, appear to play a mediating role. For example, while Hispanic couples in this study were most likely to report low levels of household income (54% had household incomes less than or equal to $20,000, compared to 39% of black couples and 17% of white couples), as well as reside in impoverished neighborhoods (50% vs. 47% of blacks and 11% of whites), neighborhood poverty was not a significant predictor of partner violence among Hispanic couples. Instead, household income was inversely associated with the risk of MFPV among Hispanics. Thus, for Hispanic couples, the effect of neighborhood poverty may be mitigated by the influence of other protective socioenvironmental factors within their communities. Conversely, white couples in this study were most likely to report high levels of household income (52% had annual household incomes greater than $40,000, compared to 28% of blacks and 14% of Hispanics), and were the least likely to reside in an impoverished neighborhood. Among white couples, neither household income nor neighborhood poverty was significantly associated with MFPV, but residence in an impoverished neighborhood was highly associated with the occurrence of FMPV. This finding suggests that for white couples, the role of neighborhood poverty as a predictor of partner violence may be differential and dependent upon which spouse perpetrates the violence. The interpretation of this result is also complicated by the fact that no data are available for this sample on which partner in the couple initiated the violence. Thus, it is unknown whether white women who reside in impoverished neighborhoods are at greater risk than non-poverty area residents to initiate violence against their mates, or whether white women who reside in such neighborhoods are at greater risk for responding to their partners’ verbal or physical provocations through physical aggression. Among black couples, residence in an impoverished neighborhood was significantly associated with the occurrence of intimate partner violence regardless of which spouse perpetrated the violence. Based on 1990 census tract data, nearly half the black couples in this sample (47%) resided in impoverished neighborhoods. Wilson characterizes these inner-city neighborhoods as depressed communities which have disproportionate numbers of “individuals who lack training and skills and either experience long-term unemployment or are not members of the labor force, individuals who are engaged in street crime and other forms of aberrant behavior, and families that experience long-term spells of poverty and/ or welfare dependency” (50, p. 8). These communities are also characterized by high social isolation (50, p. 60), and low

306

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

community social organization (51), factors that have been linked to community child maltreatment rates (52). Many of the findings were consistent with previously identified cross-sectional predictors of intimate partner violence (8, 53–55). For example, age and household income were inversely associated with the risk of MPFV among Hispanic couples; female approval of marital aggression was positively associated with both FMPV and MFPV among white couples; female history of childhood violence was associated with both FMPV and MFPV among black couples; and education was inversely associated with the risk of FMPV among both white and black couples. Other results, however, are somewhat unique within the partner violence literature and warrant further investigation. For example, among Hispanic couples, each mean year of education was associated with a 21% increase in the risk of FMPV. Black couples in which the male reported being unemployed were at decreased risk for MFPV compared to couples in which the male was employed. Perhaps the most unusual findings were the results of the interaction test for the association between FMPV and the joint occurrence of male and female alcohol-related problems among white couples. The test results indicated a significant protective association. Similarly, the results of the test for interaction for the association between FMPV and the joint reporting of female and male childhood violence among Hispanic couples indicated a highly protective effect. Future research will be necessary to determine whether or not these findings are artifactual. If these results are replicated, it will be necessary to disentangle why the joint occurrence of these factors is associated with a decreased risk of partner violence. Study Strengths and Limitations This study has a number of limitations. First, the respondents were interviewed during 1995, but the measurements of community poverty were based on data obtained five years earlier for the 1990 census. Implicit in the analysis are the assumptions that 1) neighborhoods that were impoverished in 1990 remained essentially unchanged through 1995, and 2) couples resided within the same neighborhoods between 1990 and 1995. Krieger suggests that contextual analyses that link census tract or block-group indicators to residential addresses use census data that fall within a 5-year time frame of the closest census (18). This would help minimize bias resulting from the use of census-based socioeconomic descriptors of neighborhoods that may become inaccurate over longer periods of time due to migration and population growth (56). A second limitation of the study concerns the generalizability of the findings in relation to the dichotomized measurement of partner violence. As described under the Methods section, participants were asked about the occurrence of 11 violent behaviors during the past year that they may

AEP Vol. 10, No. 5 July 2000: 297–308

have perpetrated against their partners, or that their partners may have perpetrated against them. Separate dichotomous variables were created for male-to-female (MFPV) and female-to-male partner violence (FMPV), which were scored positively for any violent episode regardless of severity and potential for producing injury. Most of the violent acts that occurred, however, fell within the “minor assault” section of the Conflict Tactics Scale; episodes of severe violence among the study’s couples were rare (44). Thus, the association between residence in an impoverished neighborhood and the occurrence of partner violence may only be valid for the most commonly occurring, least injurious forms of partner violence (i.e., throwing something; pushing, shoving, grabbing; slapping; and hitting). Predictors of the most injurious and/or lethal forms of spousal violence, such as sexual assault, assault with objects, and the use of knives or guns, may not be well represented within this analysis. Future research aimed at modeling the most severe forms of partner violence may need to be based on clinical populations rather than household probability samples (57). Temporal ambiguity, which is intrinsic to cross-sectional survey research, is a third limitation of this study. The results presented herein do not indicate, for example, whether female approval of marital aggression among white couples leads to the increased occurrence of partner violence, or whether women who have experienced partner violence are subsequently more likely to voice such opinions. The association between partner violence and a covariate measuring an event that clearly predates its occurrence, such as childhood history of parent-perpetrated violence, may even be biased if respondents who are currently experiencing or perpetrating spousal violence are more likely to recall having been victimized as children than those in non-violent relationships. Longitudinal studies that prospectively follow couples from the outset of their relationship will be necessary to help identify those factors that may put couples at risk for spousal violence throughout the life course (33, 58–60). Similarly, longitudinal study designs will be necessary to gauge the impact of broad socioeconomic upturns or downturns on the occurrence of partner violence. Data from South Korea, for example, link that country’s recent economic downturn and rising unemployment rate with a surge in domestic violence and family abandonment (61). Despite its limitations, this study draws upon a number of strengths that are unique in the intimate partner violence literature. For example, while many of the previous national partner violence surveys relied upon information from only one informant per couple (2, 3, 62), this study was based on separate interviews with both members of the pair. This allowed for the construction of two distinct dependent variables based on the gender of the perpetrator, regardless of which partner reported the violence, or whether their report was corroborated by the other partner. In addition, prior research has shown an important correspondence between

AEP Vol. 10, No. 5 July 2000: 297–308

couple-level characteristics and intimate partner violence (32). The present study incorporates the individual-level characteristics of each spouse, couple- and household-level covariates, and a neighborhood-level socioeconomic indicator into a contextual analysis of intimate partner violence. Policy Implications Given the high proportion of black couples residing in impoverished communities, and the elevated risk of partner violence among black couples (63), these findings may hold important public health and domestic violence prevention implications. For example, federal, state, and local government policies aimed at reducing inner-city community poverty may contribute to decreasing the risk of partner violence among black couples. Policies that foster full employment and reduce income inequalities, as well as renewed funding efforts for urban schools and other public institutions, may be an effective primary prevention strategy. As Lynch and colleagues point out, “economic policy is public health policy” (64). In addition, impoverished neighborhoods may be targeted by public health and non-profit agencies for partner violence prevention campaigns and intervention services, such as counseling programs and battered women’s shelters. CONCLUSION The results of the analysis lend support to the hypothesis that characteristics of the socioenvironment, such as neighborhood poverty, are associated with the occurrence of intimate partner violence. These findings are consistent with the results of the multilevel analysis reported by O’Campo et al. (36), although that study was based on a sample consisting of pregnant, mostly African-American women residing within the city of Baltimore, Maryland. The present study is the first contextual analysis of intimate partner violence to find an association between residence in an impoverished neighborhood and the risk of spousal violence among a nationally representative sample of married and cohabiting white, black and Hispanic couples. These findings suggest that a complex array of individual-, household-, and socioenvironmental-level factors are associated with the occurrence of partner abuse; these associations may vary in magnitude between racial/ethnic groups and by perpetrator gender. Although the results of this study need to be replicated through additional research, these findings may hold significant public health and partner violence prevention implications. Future prevention efforts that are based on a contextual perspective of partner violence may be more effective than strategies which do not account for the impact of the socioenvironment on individual behavior. This study was supported by an Intimate Partner Violence Prevention Research Grant (grant no. 914053) from the Centers for Disease Control and Prevention and by the National Institute of Alcohol Abuse and Alcoholism (grants no. RO1AA10908 and R37AA10908). The authors gratefully acknowledge the technical assistance of Paul Gruenewald, Ph.D.

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

307

An earlier version of this paper was presented at a poster session of the American College of Epidemiology Annual Scientific Sessions, San Francisco, CA, September 1998.

REFERENCES 1. Fagan J, Browne A. Violence between spouses and intimates: Physical aggression between women and men in intimate relationships. In: Reiss J, Albert J., Roth JA, eds. Understanding and Preventing Violence. Washington, DC: National Academy Press, 1994:115–292 2. Straus MA, Gelles RJ. Physical violence in American families: Risk factors and adaptations to violence in 8,145 families. New Brunswick, NJ: Transaction, 1990. 3. Straus MA, Gelles RJ, Steinmetz SK. Behind closed doors: Violence in the American family. Garden City, NJ: Anchor Books, 1980. 4. Stark E, Flitcraft AH. Spouse abuse. In: Rosenberg ML, Fenley MA, eds. Violence in America: A Public Health Approach. New York: Oxford University Press, 1991:123–157. 5. Straus MA, Gelles RJ. Societal change and change in family violence from 1975 to 1985 as revealed by two national surveys. J Marriage Family. 1986;48:465–479. 6. Straus MA. State-to-state differences in social inequality and social bonds in relation to assaults on wives in the United States. J Comp Family Studies. 1994;25:7–24. 7. Yllo KA, Straus MA. Patriarchy and violence against wives: the impact of structural and normative factors. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick, NJ: Transaction, 1990:383–399. 8. Sorenson S, Upchurch D, Shen H. Violence and injury in marital arguments: Risk patterns and gender differences. Am J Public Health. 1996;86:35–40. 9. Haan M, Kaplan GA, Camacho T. Poverty and health: prospective evidence from the Alameda County study. Am J Epidemiol. 1987; 125:989–998. 10. Anderson RT, Sorlie P, Backlund E, Johnson N, Kaplan GA. Mortality effects of community socioeconomic status. Epidemiol. 1996;8:42–47. 11. Ben-Shlomo Y, White IR, Marmot M. Does the variation in the socioeconomic characteristics of an area affect mortality? BMJ. 1996; 312:1013–1014. 12. Kaplan GA. People and places: Contrasting perspectives on the association between social class and health. Intl J Health Services. 1996; 26:507–519. 13. Kaplan GA, Pamuk ER, Lynch JW, Cohen RD, Balfour JL. Inequality in income and mortality in the United States: Analysis of mortality and potential pathways. BMJ. 1996;312:999–1003. 14. Miethe TD, McDowall D. Contextual effects in models of criminal victimization. Soc Forces. 1993;71:741–759. 15. Smith DA, Jarjoura GR. Household characteristics, neighborhood composition and victimization risk. Soc Forces. 1989;68:621–640. 16. Sampson RJ, Raudenbush SW, Earls F. Neighborhoods and violent crime: A multilevel study of collective efficacy. Science. 1997;277: 918–924. 17. Brewster KL, Billy JOG, Grady WR. Social context and adolescent behavior: the impact of community on the transition to sexual activity. Soc Forces. 1993;71:713–740. 18. Krieger N. Overcoming the absence of socioeconomic data in medical records: Validation and application of a census-based methodology. Am J Public Health. 1992;92:703–710. 19. O’Campo P, Xue X, Wang M-C, Caughy MOB. Neighborhood risk factors for low birthweight in Baltimore: a mulitlevel analysis. Am J Public Health. 1997;87:1113–1118. 20. Reijnveld SA, Schene AH. Higher prevalence of mental disorders in socioeconomically deprived urban areas in the Netherlands: Community or personal disadvantage? J Epidemiol Comm Health. 1998;52:2–7.

308

Cunradi et al. POVERTY AS A PREDICTOR OF PARTNER VIOLENCE

21. Wang J, Siegal HA, Falck RS, Carlson RG. Needle transfer among injection drug users: A multilevel analysis. Am J Drug Alcohol Abuse. 1998;24:225–237. 22. Jones-Webb R, Snowden L, Herd D, Short B, Hanna P. Alcoholrelated problems among black, Hispanic and white men: The contribution of neighborhood poverty. J Stud Alcohol. 1997;58:539–545. 23. Diez-Roux AV, Nieto FJ, Muntaner C, Tyroler HA, Comstock GW, Shahar E. Neighborhood environments and coronary heart disease: a multilevel analysis. Am J Epidemiol 1997;146:48–63. 24. Robert SA. Community-level socioeconomic status effects on adult health. J Health Social Behavior. 1998;39(March):18–37. 25. Von Korff M, Koepsell T, Curry S, Diehr P. Multi-level analysis in epidemiologic research on health behaviors and outcomes. Am J Epidemiol. 1992;135:1077–1082. 26. Diez-Roux AV. Bringing context back into epidemiology: Variables and fallacies in multilevel analysis. Am J Public Health. 1998;88: 216–222. 27. Hox JJ, Kreft IGG. Multilevel analysis methods. Soc Methods Res. 1994;22:283–299. 28. Iversen GR. Contextual Analysis. Newbury Park, CA: Sage, 1991. 29. Duncan C, Jones K, Moon G. Context, composition and heterogeneity: Using multilevel models in health research. Soc Sci Med. 1998;46: 97–117. 30. Kaufman Kantor G, Straus MA. The ‘drunken bum’ theory of wife beating. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick, NJ: Transaction, 1990:203–224. 31. Aldarondo E, Sugarman DB. Risk marker analysis of the cessation and persistence of wife assault. J Consult Clinical Psych. 1996;64:1010– 1019. 32. Hotaling GT, Sugarman DB. An analysis of risk markers in husband to wife violence: The current state of knowledge. Violence Victims. 1986;1:101–124. 33. Magdol L, Moffitt TE, Caspi A, Newman DL, Fagan J, Silva PA. Gender differences in partner violence in a birth cohort of 21-year-olds: Bridging the gap between clinical and epidemiological approaches. J Consult Clinical Psych. 1997;65:68–78. 34. O’Brien JE. Violence in divorce prone families. J Marriage Family. 1971;33:692–698. 35. Miles-Doan R, Kelly S. Geographic concentration of violence between intimate partners. Public Health Rep. 1997;112:135–141. 36. O’Campo P, Gielen AC, Faden RR, Xue X, Kass N, Wang M-C. Violence by male partners against women during the childbearing year: A contextual analysis. Am J Public Health. 1995;85:1092–1097. 37. Caetano R, Clark CL. Trends in alcohol-related problems among whites, blacks, and Hispanics: 1984–1995. Alcohol Clin Exp Res. 1998;22: 534–538. 38. Porcellini L, Lombard CS. 1995 National Alcohol Survey (NAS): Sampling, Weighting and Sampling Error Methodology. Philadelphia, PA: Institute for Survey Research, Temple University, 1997. 39. Straus MA. Measuring intrafamily conflict and violence: The Conflict Tactics (CT) Scales. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick: Transaction, 1990:29–47. 40. Hilton ME. A note on measuring drinking problems in the 1984 National Alcohol Survey. In: Clark WB, Hilton ME, eds. Alcohol in America: Drinking Practices and Problems. Albany, NY: State University of New York Press, 1991:51–72. 41. Cunradi CB, Caetano R, Clark CL, Schafer J. Alcohol-related problems and intimate partner violence among white, black and Hispanic couples in the U.S. Alcohol Clin Exp Res. 1999;23:1492–1501.

AEP Vol. 10, No. 5 July 2000: 297–308

42. U.S. Bureau of the Census. Summary tape file 3 on CD-Rom technical documentation. Washington, DC: US Department of Commerce, Economics, and Statistics, 1992. 43. Federal Register. Annual update of the poverty income guidelines, 1990. 44. Caetano R, Cunradi CB, Schafer J, Clark CL. Intimate partner violence and drinking among white, black and Hispanic couples in the U.S. Journal of Substance Abuse, in press. 45. Kish L. Survey Sampling. New York: John Wiley & Sons, 1965:148–181. 46. Kalton G. Introduction to Survey Sampling. Newbury Park, CA: Sage, 1983. 47. Shah BV, Barnwell BG, Bieler GS. SUDAAN User’s Manual, Release 7.0. Research Triangle Park, NC: Research Triangle Institute, 1996. 48. Bryk A, Raudenbush S, Congdon R. HLM: Hierarchical linear and nonlinear modeling with the HLM/2 and HLM/3L programs. Chicago, IL: Scientific Software International, 1996. 49. Selvin S. Statistical Analysis of Epidemiologic Data. New York: Oxford University Press, 1991. 50. Wilson WJ. The Truly Disadvantaged: The Inner City, The Underclass, and Public Policy. Chicago, IL: University of Chicago Press, 1987. 51. Kassarda JD, Janowitz M. Community attachment in mass society. American Sociological Rev. 1974;39:253–302. 52. Coulton CJ, Korbin JE, Su M, Chow J. Community level factors and child maltreatment rates. Child Develop. 1995;66:1262–1276. 53. Cazenave NA, Straus MA. Race, class, network embeddedness, and family violence: a search for potent support systems. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick, NJ: Transaction, 1990:321–339. 54. Straus MA, Smith C. Violence in Hispanic families in the United States: incidence rates and structural interpretations. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick, NJ: Transaction Books, 1990:341–363. 55. Suitor JJ, Pillemer K, Straus MA. Marital violence in a life course perspective. In: Straus MA, Gelles RJ, eds. Physical Violence in American Families. New Brunswick: Transaction, 1990:305–317. 56. White MJ. American Neighborhoods and Residential Differentiation. New York: Russell Sage Foundation, 1987. 57. Straus M. Injury and frequency of assault and the “representative sample fallacy” in measuring wife beating and child abuse. In: Straus M, Gelles R, eds. Physical Violence in American Families. New Brunswick: Transaction, 1990:75–91. 58. Murphy CM, O’Leary KD. Psychological aggression predicts physical aggression in early marriage. J Consult Clin Psychol. 1989;57:579–582. 59. O’Leary KD, Malone J, Tyree A. Physical aggression in early marriage: Prerelationship and relationship effects. J Consult Clin Psychol. 1994;62:594–602. 60. Quigley BM, Leonard KE. Desistance of husband aggression in the early years of marriage. Violence Victims. 1996;11:355–370. 61. Wiltrout K. Women Bear Brunt of Korea’s Fall. San Francisco Chronicle, September 3, 1998:A15. 62. Kaufman Kantor G, Jasinksi JL, Aldarondo E. Sociocultural status and incidence of marital violence in Hispanic families. Violence Victims. 1994;9:207–222. 63. Hamptom RL, Gelles RJ. Violence toward black women in a nationally representative sample of black families. J Comp Family Studies. 1994; 25:105–119. 64. Lynch J, Kaplan G, Salonen J. Why do poor people behave poorly? Variation in adult health behaviours and psychosocial characteristics by stages of the socioeconomic lifecourse. Soc Sci Med. 1997;44: 809–819.