Heterogeneity Among Violence-Exposed Women: Applying Person-Oriented Research Methods

Heterogeneity Among Violence-Exposed WomenApplying Person-Oriented Research MethodsPaula S. NuriusSchool of Social Work and Harborview Injury Prevention and Research Center affiliate, University of Washington

Rebecca J. MacySchool of Social Work, University of North Carolina

Variability of experience and outcomes among violence-exposed people poseconsiderable challenges toward developing effective prevention and treatmentprotocols. To address these needs, the authors present an approach to researchand a class of methodologies referred to as person oriented. Person-orientedtools support assessment of meaningful patterns among people that distin-guish one group from another, subgroups for whom different interventions areindicated. The authors review the conceptual base of person-oriented meth-ods, outline their distinction from more familiar variable-oriented methods,present descriptions of selected methods as well as empirical applications ofperson-oriented methods germane to violence exposure, and conclude withdiscussion of implications for future research and translation between researchand practice. The authors focus on violence against women as a population,drawing on stress and coping theory as a theoretical framework. However,person-oriented methods hold utility for investigating diversity among violence-exposed people’s experiences and needs across populations and theoreticalfoundations.

Keywords: person-oriented; variable-oriented; domestic violence; sexualassault

This special issue focuses on emerging frameworks in the study of vio-lence, particularly in relation to how experiences of violent victimiza-

tion can manifest in various ways in terms of trauma, revictimization, andconversely, resilience. Our contribution acknowledges growing evidence ofthe variability of experience and outcomes among violence-exposed people,

Journal of InterpersonalViolence

Volume XX Number XXMonth XXXX xx-xx

© 2008 Sage Publications10.1177/0886260507312297

http://jiv.sagepub.comhosted at

http://online.sagepub.com

1

the complexity inherent in capturing differences in changes over time, and thechallenges this variability and complexity pose for developing promisingand effective prevention and treatment protocols.

To address this challenge, we present a class of methodologies—referredto as pattern-centered or person-oriented research—that expand importantavenues for discerning substructure and coherence embedded in heterogene-ity among victims of violence, as well as the dynamic relationships leadingto and stemming from violence exposure. Briefly, variable-oriented methodspursue relationships among variables based on least squares approaches, inwhich researchers ascertain how much variance in the dependent variable isexplained through manipulations (or changes) in the independent variableor variables (such as correlations or regression). In general, the results ofvariable-oriented methods are models of the relationships between thedependent and independent variable(s), which characterize the relation-ships between the variables of interest for the sample overall. Quantitativeperson-oriented methods, on which we focus, pursue relationships at thepersonal level rather than at the variable level (such as patterns of charac-teristics that one subgroup holds in common and that distinguish it fromother subgroups). Whereas both approaches use variables in the sense ofmanipulating quantitative data, person-centered strategies test for intraindi-vidual and interindividual differences among the variables of interest. Theresults of person-oriented statistical methods are models of the relation-ships among the variable(s) of interest, which show the distinct configura-tions of heterogeneity within a sample.

Following illustration of heterogeneity among victims of violence againstwomen, we present the conceptual base underlying person-oriented analyses,drawing on traumatic stress and coping theory to demonstrate one set ofmechanisms that account for heterogeneity among victims. We then reviewdistinctions between person-oriented research and more familiar variable-oriented approaches, including description of selected methods. This is fol-lowed by empirical examples of the use of person-oriented analyses germaneto heterogeneity in violence against women. We conclude with implicationsfor future research and the translational articulation of research and practice.

We focus throughout the article on violence against women for heuristicpurposes, which allows us to coherently and parsimoniously interweavesubstantive examples with methodological explanations. However, the fun-damental thrust for the need to empirically manage heterogeneity and com-plexity and to make fuller use of currently available analysis tools appliesacross populations affected by violence.

2 Journal of Interpersonal Violence

Heterogeneity Among Victims of Violence Against Women

Research spanning more than three decades has produced a robust bodyof findings establishing the magnitude of interpersonal violence againstwomen. Estimates suggest that more than a quarter of women in the UnitedStates will be physically abused in an intimate relationship (Elliott, Mok, &Briere, 2004; Tjaden & Thoennes, 2000) and that one in six women willexperience completed or attempted rape at some point in their lives (Tjaden& Thoennes, 2006), with rates of sexual violence considerably higher instudies of a range of sexual assault experiences (e.g., forced fondling, forcedoral sex; Koss, 1993). Similarly, research findings are dense with docu-mentation of intimate partner violence (IPV) and sexual assault as stressorsthat negatively affect women’s health and well-being (Campbell et al.,2002; Golding, 1999; Logan, Walker, Cole, & Luekefeld, 2002; Plichta, 2004;Resick, 1993), with similar consequences among women who experiencedstalking, sex trafficking, and torture (Arcel, 2002; Davis, Coker, & Sanderson,2002; Farley, 2004). The negative sequelae of violent victimization spanmultiple aspects of women’s biopsychosocial functioning, including poorerphysical health, mental health, economic well-being, more impaired socialrelationships, and risk of future revictimization among abused women rel-ative to nonabused women (Bogat, Levendosky, Theran, von Eye, & Davidson,2003; Campbell & Lewandowski, 1997; Eby, Campbell, Sullivan, & Davidson,1995; Kessler, Molnar, Feurer, & Appelbaum, 2001).

There is indication of a dose–response relationship between violenceexposure and outcomes wherein women who report high levels of exposure,severity, and chronicity typically fare most poorly in outcomes (Campbell,2002; Golding, 1999; S. J. Woods, 2000). However, findings also indicatethat a singular linear trend (as the level of exposure increases, the negativityof outcomes uniformly increases) does not appropriately represent variabilityamong victims’ experiences and outcomes (see, e.g., Carlson, McNutt, Choi,& Rose, 2002; Messman-Moore & Long, 2002). Not all victims exhibit neg-ative biopsychosocial outcomes (Campbell, Sullivan, & Davidson, 1995;Johnson, 1995), nor is the same violence inherently appraised and experi-enced in the same way by all victims (Lindhorst, Nurius, & Macy, 2005;Smith, Smith, & Earp, 1999). As Briere and Jordan (2004) conclude fromtheir review of findings, there is considerable heterogeneity in how violenceis experienced, how women cope with the traumatic consequences of vio-lence victimization, and how these consequences manifest in victims’ healthand well-being. Consistent with the broader trauma research, effects of and

Nurius, Macy / Heterogeneity and Person-Oriented Methods 3

responses to abuse differ across victims; exposure to abuse alone does not inand of itself lead to development of serious trauma outcomes (Arias, 1999;Bonanno, 2004; Updegraff & Taylor, 2000), nor does exposure to similarforms of exposure result in universal needs for treatment or other supports.

The bulk of research on the relationships among women’s violence expo-sure, trauma, health, and well-being has been undertaken through variable-oriented methods that focus on the relationships among variables but arelimited in their ability to yield statements about the women themselves. AsBogart, Levendosky, and von Eye (2005) state, “researchers often write aboutthese analyses ‘as if’ they say something about individuals (e.g., women suf-fering from depression are more likely to have impaired parenting), but theyare really statements about variables (e.g., depression is related to parenting)”(p. 50). Variable-oriented analyses provide valuable tools toward investigatingcritical questions that have established violence against women as a seri-ous and complex social justice and public health issue. However, variable-oriented analyses do not generally show the mix of relationships found withinsamples, such as the variety of ways that behaviors or characteristics inter-correlate for different subgroups (Bergman, 2001; Mitchell & Plunkett,2000). Sole reliance on variable-oriented analyses involves a potential dangerin obscuring the diverse nature of samples and populations, thereby fosteringmisleading, overly generalized conclusions, because these statistical approachescharacterize the relationships among variables of interest for the sample over-all (von Eye & Bergman, 2003).

Thus, a challenge facing violence researchers is how to discern multi-ple patterns and/or trends of risk and resilience in relation to violence. What tools substantially advance our capacity to capture not only the co-occurrence of multiple factors that synergistically combine to affect risk andresilience but also the ways that these factors coalesce in meaningfully dis-tinct combinations to fuel differing pathways? This challenge is echoedacross multiple fields and is particularly evident in the perspectives ofdevelopmental science and developmental contextualism (Bergman, Cairns,Nilsson, & Nystedt, 2000). We do not argue for total relativism or that eachindividual’s experience needs be treated as a wholly unique phenomenon.Rather, theory and working evidence suggest that although there is consid-erable variety in how human beings experience and respond to violent vic-timization, using person-oriented approaches, such heterogeneity can bestatistically and meaningfully modeled by a limited quantity of numericalresults (Bergman, 2001; Messman-Moore & Long, 2002; von Eye & Bogat,2006). In turn, such empirical results will inform the development of tai-lored theory about and interventions for groups of violence survivors whoare similar in important ways. We argue that the need is great to proactively


investigate, document, and understand subgroup patterns that reflect thediversity of violence-affected people.

Conceptual Base for Person-Oriented Methods

Variable-oriented research is based on the premise that samples and theircorresponding population are homogeneous, or at least sufficiently homoge-neous, so that summary statistics and characterizations (such as means oroverall trends) provide useful information about the relationships among thevariables of interest. Variation is, of course, recognized in variable-orientedanalyses, such as variance and standard deviation statistics, but generally thepurpose is to characterize the relationships between the variables of interestfor the overall sample. In contrast, person-oriented research is based on thepremise that meaningful subgroups exist and that the statistical results thatbest describe the overall sample may contradict those of subgroups or indi-viduals in important ways (Bergman, 2001; von Eye & Bogat, 2006).

In addition, Bergman and Trost (2006) characterize the person-orientedapproach as grounded in a holistic, transactional, metatheoretical paradigm,viewing the individual as an active agent in the person–environment dynamic(also see Bergman & Magnusson, 1997). This tradition stresses the impor-tance of regarding the individual as a “functioning whole”—capturing thejoint contribution of multilevel components with the recognition that peopleand behaviors are nested within community, social, and environmental con-texts that can yield distinct differences.

Person-oriented analysis is based on the following premises (von Eye &Bergman, 2003; von Eye & Bogat, 2006): (a) the organization and dynam-ics of behavior are at least partly specific to the individual; (b) the com-plexity of behavior and development necessitates taking multiple factorsand their interrelations into account; (c) there should be a predictability andstructure to intraindividual constancy and change and to interindividual dif-ferences in this regard; (d) many processes occur in a consistent way, whichtake form in patterns of individual factors; (e) the meaning of the involvedfactors is at least partly shaped by interactions among the factors; and (f) somepatterns occur more frequently than others, thus constituting relativelycommon types (i.e., patterns that are sufficiently represented in a sample orpopulation to comprise identifiable groups). Thus, person-oriented analytictools help ascertain whether there are common types of patterned interrela-tionships that constitute an empirical structure of heterogeneity; that is,clusters of theoretically meaningful characteristics that are shared withinsubgroups and that distinguish subgroups from one another.


Within the social sciences, person-oriented theorizing and methods haveperhaps gained most ground in the study of human development. This arenais inherently dynamic across varying points of time, settings or conditions,populations, and combinations therein. That is, in the study of human devel-opment, researchers hypothesize about variety in developmental trends dueto gender, environment, culture, and so on, as well as combinations amongthese factors. Within this field of study, ordinary least squares statisticalmodels of the relationships between the dependent and independent vari-able(s) are often not sufficient approximations of phenomena (e.g., theremay be different trends or trajectories for different groups, or the trends maybe nonlinear), interactions are more common than rare, distinct subgroupschallenge the validity of summary statistics, and variables often have moresalient meaning as parts of constellations and patterns than by themselvesalone (Bergman, 2000).

For the most part, research on the relationship of violent victimization towomen’s health and well-being has been largely cross-sectional, thoughthere are some noteworthy exceptions (see, e.g., El-Bassel, Gilbert, Wu, Go,& Hill, 2005; Tolman & Wang, 2005; Zlotnick, Johnson, & Kohn, 2006).Little empirical work has been dedicated to investigating the presence ofsubgroups based on differing patterns of theorized relationships, variabilityin their developmental progressions relative to violence exposure, or uniquefeatures in how victimization affects and is affected by other aspects of func-tioning or the environment (Williams, 2003). This neglect leaves open ques-tions such as: Why do some women experience negative biopsychosocialconsequences after victimization and others do not? Why do some womenexperience repeated violence and trauma over the course of their lives andothers do not? What are the important intrapersonal and interpersonal fac-tors that promote a resilient life trajectory following violent victimization?Although an oversimplification, one might think in terms of unexplainedvariance in variable-oriented analysis, which is often a considerable propor-tion of the total variance. What remains unexplained is attributable to manypossible factors, among them being the unassessed yet meaningful variabilityamong significantly distinct subgroups.

The concepts underlying person-oriented methods are familiar across sev-eral arms of research and are not inherently contradictory to variable-orientedresearch. There are multiple variable-oriented analytic tools that aim to detecttransactional dynamics and approximate more holistic assessments. Structuralequation modeling and hierarchical lineal modeling as they are currentlyapplied in social science research are examples of powerful variable-orientedanalytic tools in this regard (see, e.g., Kline, 2005; Raudenbush, 2001).


However, if one has reason to believe that unidentified subgroups exist withina population, such as partner violence and sexual assault victims, variable-oriented analysis is likely to obscure the presence of such groups as well asefforts to obtain descriptive or predictive details about their differences. Inshort, variable-oriented and person-oriented approaches allow researchers topursue different and complementary kinds of questions (Bandee-Roche,Miglioretti, Zeger &, Rathouz, 1997; Bergman, 2001; Land, 2001).

Contrasting Person-Oriented to Variable-OrientedAnalytic Approaches

Concern with overlooking important forms and consequences of diver-sity has spawned interest in the application of person-oriented analyses.Practically speaking, no one study can capture the totality of any popula-tion or phenomenon of interest. Selection of a delimited number of compo-nents or variables is necessary, and this selection will be sculpted by theoryregarding the specific processes or outcomes of interest. In our own researchat the interface of violence and trauma, we have found it critical to under-stand the variability in how victims experience, respond to, cope with, andprogress in the aftermath of violent victimization. The multivariate hetero-geneity in how a person experiences and reacts to violence is difficult tomeasure directly (Bandee-Roche et al., 1997; Hagenaars & Halman, 1989).This is partly related to the measurement tools at hand but also becausesome constructs (such as vulnerability or resilience) are often not ade-quately represented by a single observed measure and people differ in thecombinations of the measures that are needed to adequately represent theconstruct. The analytic tools to investigate this heterogeneity have not yetbeen extensively applied in violence research. In this part of the discussion,we provide more detail about ways that person-oriented analyses are dis-tinct from and complementary to variable-oriented analyses using violenceagainst women and variability in stress responses to violence exposure assubstantive illustration.

Variable-oriented, ordinary least squares statistical methods have providedthe backbone of advances in violence research to date, including such find-ings as the incidence and prevalence rates, the different types, characteris-tics and directionality of violence expression, the aggregate relationship ofviolence exposure to a wide range of physical and health outcomes, andvariables antecedent to and/or associated with the perpetration and experi-ence of violence. Fitting slopes through data arrays, for example, provides


important insights as to the characteristics of general linear trends as well asa foundation for subsequent analysis. On the downside, this analytic approachtends to best represent those individuals closest to the slope line, “washingout” distinctiveness more distant from that referent point. Variable-orientedstatistical strategies can be used in pursuing person-oriented research aimsonce subgroups have been established, but variable-oriented tools such ascorrelation, regression, analysis of variance, and hierarchical linear model-ing are not designed for detecting differing configurations and patterns ofvariables within a sample (Macy, in press).

Person-oriented statistical tools, in contrast, are well equipped to show dif-fering types of interconnected patterns. Like using factor analysis to test forfactor structure across several variables, person-oriented analyses detect simi-larities among analyzed people, based on whether they have distinct profilesof relationships among the variables of interest rather than on whether they arehigh or low on any one variable or set of variables. Figure 1 provides a gen-eral illustration of the function of person-oriented analytic approach relative to variable-oriented tools, albeit limited by portrayal of only two variables.Within this two dimensional distribution, we show five subgroups of individ-uals arrayed along the regression line. Although we are figurally unable to gobeyond two dimensions, one can think of the person-oriented approach assearching for coherent clusters within multidimensional space defined by morenumerous variables than two dimensions.

Person-oriented methods are based on varying statistical approaches toclustering or classifying groups of people based on the comparability of theirvalued or locations on an array of variables. Variable selection is driven by


Figure 1Bivariate Example of Subgroup Clusters

0

5

10

15

20

25

30

0 5 10 15 20 25 30

Variable B

Var

iab

le A

theory that is used to interpret the meaning of the subgroups and their dif-fering relationships to other variables of interest. There is a wide array of spe-cific analysis methods available (for overviews, see Bergman, von Eye, &Magnusson, 2006; Everitt, Landau, & Morven, 2001). Choices of whichmethod to use will be guided by factors such as the metric of the availabledata (e.g., categorical vs. continuous), whether cross-sectional or longitudi-nal distinctions are sought, what functional form the relationships may have(e.g., linear, curvilinear), the availability of specific analysis software, thetraining of the user, standards of the discipline or substantive area of inquiry,and conceptual or methodological priorities of the study at hand.

The result of some person-oriented analysis, for example, is a latent vari-able of multivariate profiles of subgroups within a sample that are more sim-ilar to one another than to others in the sample (Muthen, 2002). Latentvariables are not direct measures but are statistically inferred from multipledirect measures whose interrelationships are assessed simultaneously. Theseanalyses allow a researcher to represent theorized and unmeasured con-structs that underlie the combined relationships of the measured variables aswell as the position of individuals within multidimensional space. In ourfigure, one can think of adjusting the circles to capture the greatest compa-rability across all the indicators reflecting coherence within groups and dis-tinctiveness between groups.

Investigation of differences may be cross-sectional determinations of sub-groups as shown in Figure 1 or more anchored in trends or developmentalpathways over time. For example, von Eye and Bogat (2006) report findingsfrom a study in which children were observed from birth through 3 years ofage to determine the effects of domestic violence on partners and the devel-opment of their young children. Using variable-oriented methods, they founda main effect for violence on children’s negative mood over time but did notfind anticipated effects related to the gender of the child or interaction ofgender with violence status (victimized vs. nonvictimized mothers). Visualinspection of data curves at the individual level spawned a person-orientedanalysis on child withdrawal development that uncovered four distinct sub-groups. Figure 2, adapted from von Eye and Bogat (2006), provides a depic-tion of differing curves. These subgroups revealed effects of violence status(violence exposed vs. no violence exposure) in the children’s mood statesover time that were not evident through the variable-oriented analysis.

We present this figure to illustrate that for some questions, it is differencesin experiences over time or across contexts that are most relevant. Thus, theability to discern and track the shape of these differences is crucial to under-standing heterogeneity in dynamic processes. The result of this line of analysis


was that the variable-oriented approach allowed meaningful, overall results, butthat the important unexplained heterogeneity remained. In this case, the abilityto identify and track subgroups of children along differing developmental path-ways contrasted with the overall picture, providing nuanced findings with directclinical relevance for subsets of vulnerable children.

To further illustrate differences in variable-oriented and person-orientedmethods and the types of information they provide, von Eye, Bogat, andRhodes (2006) undertook data analyses using both approaches on the samedata set, a sample of youth from the National Cross-Site Evaluation of HighRisk Youth Programs. Specifically, they report the results of a mixed effect,repeated measures ANOVA as the variable-oriented method, and ConfiguralFrequency Analysis (CFA) as the person-oriented method. The substantivefocus was on the relationship over time between parental attitudes andgender as it relates to youth alcohol consumption. The ANOVA identifiedan overall trend, yet accounted for little variance. The CFA explained whythis was the case, by identifying groups of respondents whose behavior wasdifferent from the overall trend. Although findings from these two approacheswere quite different, the authors view them as complementary and collectivelycontribute more than either could individually. For example, two of the maineffects from the ANOVA support the current literature: Boys drink morealcohol than girls, and more negative parental attitudes toward drinking are


Figure 2Example of Differing Trajectories over Time for

Withdrawal Pattern Subgroups

Source: Adapted from von Eye and Bogat, 2006.

associated with less adolescent drinking. Such findings have important gen-eral implications for interventions that target alcohol consumption reduc-tion. The CFA identified groups of girls and boys who behaved counter tothe general trend, findings that also carry important implications for tai-lored or adaptive interventions, particularly for some higher risk youth.

In summary, in contrast to variable-oriented methods, person-oriented sta-tistical approaches do not determine a single statistical model for a sample.Instead, person-oriented approaches enable the testing of distributional diver-sity within a sample. For researchers concerned with violent victimization andits traumatic consequences, for example, this translates into investigation ofdiversity among biopsychosocial factors manifest through identifiable sub-groups that hold significant risk and protective potential for trauma develop-ment, negative physical and mental health sequelae, impairments in socialfunctioning, and revictimization. In variable-oriented research, respondentswhose data are dissimilar relative to overall trends may be treated, statisticallyor conceptually, as outliers. Person-oriented research spotlights questions as towhether dissimilar points or trends may be importantly and meaningfully dif-ferent, meriting more systematic study.

Latent Variable Approaches

Although it is beyond the scope of this article to provide a comprehensivedescription of person-oriented methodological and statistical approaches, webriefly describe statistical modeling approaches that can be used to investi-gate the statistical relationships among sample participants using a person-oriented analytic strategy. As Bergman and Trost (2006) point out, this caninclude a wide set of methods, with differing researchers using somewhat dif-ferent criteria as to what is included. Overviews are provided by Bergman(2000), Bergman et al. (2006), and Bergman and Trost (2006), and Curranand Willoughby (2003), with respect to trajectory models. In the next section,we also provide examples of person-oriented investigations using a range ofmethods.

Here, we focus on methods using latent variables (for an overview oflatent variables and their use in social sciences, see Bollen, 2002). Althoughlatent variable person-oriented research tools are not wholly new, specifi-cation of formalized propositions, refinement of these tools, and the avail-ability of latent variable software to credibly accomplish testing for underlyingsubgroups have gained momentum over the past two decades (see, e.g.,Curran & Willoughby 2003; Muthen, 2002; Muthen & Muthen 1998-2004)


and are now far more accessible to and used by applied researchers. Asnoted, a latent measurement approach carries the advantage of being ableto go beyond directly measured constructs to capture patterns at the inter-face of multiply measured phenomena—determining underlying factors nototherwise attainable.

This simultaneous multivariate capacity holds particular promise in light ofthe growing call among violence researchers for examination of violence-related factors in combination (Arata, 2002; Messman-Moore & Long, 2002).Risks such as revictimization are complex and dynamic and thereby challeng-ing to directly measure. There are likely multiple configurations of risk andprotective factors through which violence-exposed women become vulnerableor resilient to revictimization and other future risks. Theorizing within the vio-lence arena is rapidly developing to match these lines of inquiry, providing aconceptual foundation to guide appropriate use of methodological tools such asperson-oriented analyses. Because of the conceptual power of latent-variable,person-oriented strategies and the relative availability of software packagesand statistical training, we focus on these approaches for illustration. Four suchapproaches include latent class analysis (LCA), latent profile analysis (LPA),latent transitional analysis (LTA), and latent growth analysis. First, we describeLCA, LPA, and LTA.

LCA and LPA modeling approaches are appropriate for person-orientedanalyses with cross-sectional research designs; LCA is used with categori-cal data, and LPA is used with interval and ratio-level data. In addition, withlongitudinal, repeated measures research designs, person-centered analysescan be conducted using LTA modeling approaches (Graham, Collins, Wugalter,Chung, & Hansen, 1991; Muthen, 2002). Using LTA, one can investigateparticipants’ group membership at each measured time and the extent towhich there is change among groups at specific times.

In brief, all three of these approaches use maximum likelihood, iterativelyrunning and comparing models, to probabilistically determine (a) distinctsubgroups on variables of interest and (b) an undergirding latent variablecomposed of the variables of interest (Everitt et al., 2001; Muthen, 2002). Inaddition, the results of these analyzes have these functions: (a) identify andshow the number of statistically well-defined groups of participants detected,if any; (b) identify the factor constellations that make the groups distinctive;(c) show the classifications of participants into groups; which in turn (d) showthe prevalence and size of these groups within the sample (Macy, in press;Macy & Chapman, 2006; Muthen, 2002). The analysis also determines several kinds of results, including model fit statistics, which will help theresearcher to assess the statistical merit of a given model as well as the meritof using person-centered approaches in a given sample (Muthen, 2002).


However, researchers should also carefully consider the theoretical and sub-stantive utility of a model, for unlike LCA and LPA, the LTA analysis alsoshows how sample participants may shift from one group to another at timepoints under investigation.

Here, we provide substantive illustration of such statistical findings. Forexample, LCA could be used by a researcher who is interested in investi-gating the heterogeneity of co-occurring DSM-IV-TR mental health disor-ders diagnoses (i.e., yes/no; does the participant meet diagnostic criteria formajor depression, posttraumatic stress disorder, substance abuse disorders,and so on?) among a sample of child sexual abuse survivors who are nowyoung adults. This kind of analysis could help shed light on different clus-ters of co-occurring mental health disorders among adult survivors of child-hood abuse. Alternatively, if this same researcher had interval-level scalesfor mental health symptomatology for a variety of disorders, LPA could beused to investigate the heterogeneity of mental illness symptom combina-tions. This kind of analysis could help investigate the complex patterns inwhich trauma consequences of childhood abuse can manifest in adultmental illness symptoms, as well as the extent to which these patterns areconsistent with (or not consistent with) current DSM diagnostic criteria.Extending this example, if the researcher had measured trauma and mentalhealth problems among this same group of child abuse survivors during keydevelopmental periods, the researcher could investigate changes in patternsof trauma and mental health disorder symptoms across these importanttimes. Such an investigation may shed light on how the negative sequelaeof child abuse may alter during children’s development across adolescenceand into adulthood.

Although latent growth models (LGM), such as latent growth analysis,also rely on latent variables to determine statistical models, this analyticsapproach developed within a structural equation modeling (SEM) context.As a result, the way in which latent variables are used is different than thestatistical approaches described above, and the analytic basis and proce-dures for conducting LGM is much like SEM. Specifically, the growthmodel is a latent or an unobserved trajectory, which is inferred from a setof repeated, observed measures (Curran & Willoughby, 2003; Duncan,Duncan, Strycker, Li, & Alpert, 1999; Kline, 2005; Muthen, 2002). By test-ing a hypothesized model of change over time, the LGM analysis investi-gates the heterogeneity by investigating the overall longitudinal trajectoryfor a given sample as well as individual differences in these trajectoriesamong the participants within the sample. (For a detailed discussion of LGManalysis, procedures, models, and software, see Duncan et al., 1999). Thus,LGMs are important longitudinal analytic tools for violence researchers,


because of their utility for determining the diverse and dynamic nature ofthe consequences of violence in survivors’ lives over time.

Again, extending the mental health example from above, with repeatedmeasures of trauma and mental health symptoms among a group of childabuse survivors, a researcher could investigate the longitudinal trends intrauma and mental health disorder symptoms across children’s lives, through-out adolescence and into young adulthood. Such an investigation would showthe overall trauma and mental health trends for the sample as well as the indi-vidual variability within these trends. Results from this analysis may show, forexample, a group of participants with clearly negative mental health trendswith ever increasing severity over time, a second group of children exhibitingresilience over time with improving mental health, and a third group showinginitial resilience with worsening mental health over time.

Because the primary aim of this article is to outline distinctions and thepotential utility of person-oriented methods for violence and traumaresearchers, we have not included formulas or data requirements here. Werefer readers to other sources for specific guidance on how to use these sta-tistical analyses (Cleland, Rothschild, & Haslam, 2000; Curran & Willoughby,2003; Duncan et al., 1999; Everitt et al., 2001; Hagenaars & Halman, 1989;Kline, 2005). In addition, we recommend to readers Gibson (1959) whodeveloped the seminal work in LCA and LPA, as well as Muthen (2001,2002) for a contemporary discussion of these statistical analyses and theirpotential applications.

A note about general growth mixture models. There is yet another person-oriented, latent-variable statistical analysis worth noting here: general growthmixture models (GGMM; Muthen, 2001). In brief, GGMM identifies a LGMwhile identifying groups within the sample with distinct configurations andgrowth trajectories (Jedidi, Jagpal, & DeSarbo, 1997). There has been debateregarding the application of these particular statistical models (see Bauer &Curran, 2003, 2004; Muthen, 2003). Conventions regarding the use and inter-pretation of GGMM are currently under discussion and development. As withany emergent methodologies, this discourse exemplifies the iterative processof tool development, use, and refinement. The referenced sources provide afoundation for this discussion.

Illustrating Person-Oriented Methods in Violence Against Women

In this section, we provide examples of several applications to illustratethe types of questions for which person-centered analysis can advance our


understanding of variability in violence experiences. Some of these exam-ples use the latent variable statistical approaches that we highlighted above.However, with the aim of exposing readers to a variety of person-orientedapproaches, we also include examples of other statistical and methodolog-ical approaches, which can also be considered person centered, includingcluster analysis (Everitt et al., 2001).

A theory-based assessment of subgroups among sexual assault victimswas undertaken on the basis of factors, including prior victimization, alcoholconsumption, relationship expectations of the assailant, and precautionaryhabits hypothesized to contextualize women’s vulnerability to assault (Macy,Nurius, & Norris, 2007a). LPA established four subgroups, each character-ized by distinct clusters of the presence or absence of these risk factors andsupported by three tests of statistical fit. Group-difference tests relative toassault characteristics reinforced the interpretation of the subgroups’ coher-ence and meaning. Subsequent analysis (Macy, Nurius, & Norris, 2007b)tested hypotheses as to subgroup differences in situational coping with theassault experience. Multivariate analysis of variance confirmed that the fourprofile groups significantly differed both on forms of behavioral responding,primary and secondary appraisals, and affective states theorized to mediatebetween stress exposure and behavioral resistance.

In both studies, prior variable-oriented research combined with theory toprovide the substantive foundation on which beliefs about subgroups waspredicated. These studies also combined the use of variable-oriented analysiswith person-centered analysis as complementary tools. Collectively, theseresults illuminate both (a) the existence of distinct subgroups among acquain-tance assault victims that have implications for their coping and outcomes aswell as (b) evidence that these groups do indeed manifest hypothesized differ-ences as a function of preassault characteristics—differences that help disen-tangle prior findings of muted assault prevention and treatment effectivenessfor women with victimization histories (Blackwell, Lynn, Vanderhoff, &Gidycz, 2004; Rozee & Koss, 2001; Yeater & O’Donohue, 1999).

Kohl and Macy (in press) used a person-centered approach in an inves-tigation of female caregivers who reported partner violence and whosechildren were involved in child welfare. With the goal of assessing the sam-ple for profiles of combined needs and resources, this investigation appliedLCA to the following categorical variables, childhood history of abuseand/or neglect (yes/no), adult depression (yes/no, met diagnostic criteria formajor depression), substance use (yes/no as to current alcohol or drug useproblems), arrest history (yes/no, has caregiver previously been arrested),stress (yes/no, as to high stress on the family), and social support (low vs.high support). The LCA analysis determined four groups reflecting distinct


multivariate profiles of needs and resources among victimized caregiverswho significantly varied in their level of baseline and subsequent violenceseverity for both the women and their children. These results, in conjunc-tion with other findings about partner violence among child welfare involvedcaregivers, can contribute to development of guidelines for tailored inter-ventions within this vulnerable population.

Hughes and Huth-Bocks (2007) applied a hierarchical agglomerative clus-ter analysis method (Hair, Anderson, Tatham, & Black, 1998) to test for sub-group structure on the basis of parenting distress among African Americanbattered women. Squared Euclidean distance was used as the proximity mea-sures as it takes into account pattern, level, and scatter characteristics of thedata. Ward’s method was the clustering algorithm, because it maximizeswithin-group homogeneity. Both hierarchical cluster techniques (to suggest thenumber of clusters) and iterative techniques (to “fine tune” group membershipsas relative fit into one group or another is assessed) were used to improve qual-ity of the cluster solution. This analysis distinguished six clusters or subgroupswith distinct configurations of three forms of parenting stress, supported by val-idating variable-oriented analysis. These clusters were associated with differentparenting behaviors, levels of maternal psychological distress, and behavioraland emotional problems of the women’s children. This sample was fairlyhomogeneous with respect to the severity of reported domestic violence. Thecluster analysis revealed variation in ways that these women responded to thestresses of similar levels of violence exposure, reflecting different combinationsof strengths and vulnerabilities for themselves and their families.

A growing impetus for the use of person-oriented tools and their combinationwith variable-oriented methods is the need to chart distinctions in trajectories—longitudinal trends over time and circumstance. Anderson, Saunders,Yoshihama,Bybee, and Sullivan (2003), for example, target cumulative adversity hypothe-ses regarding differences in postseparation stressors among battered womenleading over time to widening gaps in levels of women’s depression. Usinggrowth curve analysis, women exposed to greater amounts of violence andsecondary stressors after shelter exit were found to experience comparativelyhigher levels of depression that did not alleviate or increase over time. From adevelopmental, longitudinal perspective, Oxford et al. (2005) applied LPA,establishing three subgroups of adolescent mothers—problem prone, psycho-logically vulnerable, and normative. These groups were then compared usingmultivariate analysis of variance as to their psychosocial functioning in earlyadulthood. Finding confirmed the existence of different pathways throughadolescent parenthood that are associated with later adult functioning, includ-ing criminal victimization and IPV in addition to physical and mental health,substance use, life stress, and socioeconomic factors.


We previously noted that person-oriented goals can also be pursued usingvariable-oriented tools, for example, investigating differences on the basis of apriori constructed groups, in contrast to the use of tools to test for the presenceof subgroups. Casey and Nurius (2005) report one such example, creatingthree subgroups within a general population survey on the basis of distinctionsin prior victimization histories, specifically single victimization, multiple vic-timizations committed by one perpetrator, and multiple victimizations com-mitted by multiple perpetrators. Group-difference results indicated that singlyvictimized women differed from women with repeat victimizations with respectto characteristics of their initial assault as well as long-term psychological andhealth outcomes. Women who had multiple victimizations by multiple perpe-trators differed from the other two groups in current psychological distress,worse health, and more cumulative experiences of nonsexual trauma.

Bogat et al. (2003) similarly used an a priori categorization approach inseeking to determine, when predicting psychosocial outcomes for IPV, theimportance of capturing historical as well as current experience of IPV.Specifically, they tested the differences between a two-group classification(IPV or no IPV in the past 6 months) and two four-group classifications—one distinguishing perpetrator differences (never IPV, IPV current partneronly, IPV former partner only, IPV both current and former partners) andthe other distinguishing violence exposure relative to pregnancy (no IPV,IPV in current pregnancy, IPV prior to current pregnancy, IPV at all timeperiods). Contrasting these classification approaches allowed assessment oftheir relative contributions to explaining psychosocial outcomes as well asrevealing commonalities and distinctions among the subgroups. Approachessuch as this hold potential for better crystallizing distinctions such as episodicversus more continuous or chronic exposure to violence.

The ability to discern differing configurations of risk and protective fac-tor combinations provides new insights as to supports and interventionsresponsive to these differences and thus is more likely to be effective. Theresults of person-oriented analysis can help distinguish factors appropriatefor universal prevention or treatment services or policies from those char-acterizing subgroups that require more selective, targeted interventions onthe basis of their distinct profiles.

Linking Use of Person-Oriented Tools to Theoretical Foundations

Within person-oriented research, it is the configuration of variable valuesamong the set of theoretically specified variables that is sought and meaningful.


Theory plays an integral role in person-oriented analysis—informing boththe variables to be used to search for structure within heterogeneity and tointerpret findings. In this section, we identify parallels within one theoreticaldomain—coping processes reactive to violence as a traumatic stressor—asa further illustration of the kinds of research questions and conceptual foun-dations suited to person-oriented methods. There are multiple other theo-retical foundations that may be appropriate to investigating heterogeneityand nonlinearity in violence-related phenomena. Stress and coping theoryhas undergirded some recent investigations particularly germane to person-oriented analysis and thereby provides a timely base for illustration.

Violence research is increasingly formulating risk of revictimization andtraumatization within ecological frameworks that specify multiple levels ofcontributors (Grauerholz, 2000; Lindhorst et al., 2005; Messman-Moore &Long, 2003; Nurius & Norris, 1996; Riger, Raja, & Camacho, 2002). Riskand protective factors analyses are directing attention to processes throughwhich risk is carried or blunted relative to revictimization and related nega-tive sequelae of violent experiences. In addition to differential exposure toviolence, social and psychological vulnerability constitute pivotal pathways.Influenced by transactional stress and coping theory, the dynamic interfacebetween perpetrators’ behaviors and victims’ behaviors and vulnerabilitiesprovide a crucial organizing framework (e.g., Anderson et al., 2003; Coker,Watkins, Smith, & Brandt, 2003; Messman-Moore & Long, 2003).

Although research shows that differences in the types and severity ofviolent victimizing experiences do differentially affect women’s healthand well-being post assault (Coker et al., 2002; Staggs & Riger, 2005),unique combinations of person, situational, and environmental character-istics also likely mediate the impact of violent trauma and its manifesta-tions in women’s health and well-being (Grauerholz, 2000; Macy, 2007;Nurius, Norris, Macy, & Huang, 2004). Better empirical knowledge abouthow the variability among the combinations of person and environmentcharacteristics mediates the experience and impacts of trauma, which canbe illuminated by person-oriented statistical tools, can help advance theviolence field. Leveraging from prior variable-oriented findings, we nowask whether these factors take shape in the same way across victims.Building on our accumulated aggregate knowledge to date, we can moveforward to assess patterning in how such factors differentially combineacross victims to erode or scaffold cognitive, social, emotional, and phys-ical means of agency to resist this form of traumatic stress, as well as logicalimplications for tailoring service responses (cf. Macy, 2007; Messman-Moore & Long, 2003).


In a similar vein, the processes and outcomes of positive adaptation andresilience point to the multidimensional complexity of surviving (and possiblythriving in spite of) violence exposure and its negative sequelae (Bonanno,2004; Ryff & Singer, 2003). Recovery and adaptation from violence exposurerequires resources and assets from a range of physical, psychological, andsocial factors (Hobfoll, 1989; Updegraff & Taylor, 2000; Updegraff, Taylor,Kemeny, & Wyatt, 2002). Person-centered analyses offer the opportunity toexplore the protective functions that different resource configurations mayhold in this complex mix of interrelationships.

Accumulation of events appears associated with both resilience and vul-nerability. Without the ability to examine multiple configurations of riskand protective factors (i.e., patterns characterizing different subgroups), itis difficult to identify differing pathways associated with violence exposureand its aftermath (positive or negative) or to sort out the different combina-tions of events that lead to a particular outcome. We have pointed to the use of latent variable person-oriented methods as particularly advantageous(Hagenaars & Halman, 1989; Muthen 2002), using multiple indicators torepresent constructs that are conceptually defined but difficult to measuredirectly, such as resilience and vulnerability. Through more systematic useof person-oriented approaches, we will be better able to investigate and testfor such hypothesized constructs in complex and holistic ways, consistentwith recent theoretical developments in violence and trauma research(Macy, 2007; Messman-Moore & Long, 2003; Riger et al., 2002; Updegraff &Taylor, 2000).

Conclusion

Many of the themes raised in this special issue hold implications for thevariable and person-oriented distinctions in violence-related research; fac-tors such as the relevance of context, social location, culture, personal con-struction of meaning, and the differential press of environment, history, andplace on how we understand differential vulnerability, resilience, and theexposure to and effects of violence. These themes provide depth and dimen-sional diversity to thinking about person and environmental transactions,the complexity of multilevel theorizing that will guide future developments,and the need for accessible and flexible methodological tools that allow usto achieve increasingly holistic findings that balance the bigger picture withunderlying distinctiveness. As partners to variable-oriented approaches,person-oriented methods provide means relatively unrepresented in violence


research to manage complexity, heterogeneity, and nonlinearity—factorsdirectly akin to recent calls for assessment and service development respon-sive to such variability (Briere & Jordan, 2004).

Closely allied with this special issue is a host of issues related to bridgingthe realms of research with practice and policy toward preventing and ame-liorating violence and its effects. We argue that person-oriented researchtools can be helpful toward traversing gaps between the predominantlyvariable-oriented research in violence and systems of service working withviolence-exposed individuals, families, and communities. Thoughtful useof person-oriented methods can, for example, assist in defining often hid-den dimensional diversity among people vulnerable to or affected by vio-lence; diversity that is particularly difficult to illuminate when it involvescomplex patterns that stand largely out of reach with our mainstay variable-oriented tools. Fuller use of person-oriented analysis will allow us to makeimportant contributions to translational needs in violence research, whereinfindings on basic behavioral processes inform assessment, prevention,treatment, and delivery of services to diverse people.

The ability to detect clinically and theoretically salient subgroups, docu-menting distinctions in their characteristics and trajectories across time andplace, for example, is crucial to the development of empirically supportedservices (prevention, treatment, maintenance) with greater potential for effec-tiveness and sustainability. Conceptual frameworks are available to guide suchservice specification. The notion of adaptive interventions, for example, pro-vides a framework for identifying differing components, sequences, forms, ordosages and levels of treatment that are assigned to different individuals and/orwithin individuals over time as a function of their unique constellation ofneeds (Collins, Murphy, & Bierman, 2004). In addition to the interventionimpact, greater elucidation of dimensional diversity among people affected byviolence fuels research and service development “upstream,” such as preven-tive and health promotive programming introduced at earlier points in the lifespan to avoid violence experiences or blunt their negative effects (Lesesne &Kennedy, 2005). The health promotive contributions of person-oriented meth-ods are in fact a large part of what has fostered particularly robust attention tothese methodologies among developmental researchers (Bergman et al., 2000;Bergman et al., 2006).

In this article, we have argued against the inherent superiority of person-oriented research, because it is not our intent to present this class ofmethodologies and the conceptualizations that scaffold them as free of lim-itations or caveats. Every methodology will have data requirements, under-lying assumptions, organizing tenets, strengths, and weaknesses that must


be evaluated relative to the research questions and conditions at hand, as tothe match between problem and method. For a fuller discussion of featuresto evaluate, especially in relation to the latent variable statistical models wehave presented here, the reader is directed to Bauer and Curran (2003, 2004),Everitt et al. (2001), Kline (2005), McCutcheon (1987), and Muthen (2003).

Rather, we argue for a new plateau in violence research wherein violence-oriented and person-oriented methods are creatively and meaningfully inte-grated into research programs, capitalizing on their comparative strengthsand complementarity. Within this vein, we adapt recommendation fromBergman et al. (2006), noting value in the following sequence: (a) prelimi-nary variable-based analysis to identify operating factors in a field of study,to create measurement models, and to aid in the formulation of relevant patterns; (b) followed by predominantly descriptive (person) pattern-basedanalyses, including testing that the results are consistent with theoreticalexpectations; (c) predominantly model-based pattern analyses for formulatingand testing models of person-oriented theories; and (d) multimethod studies or analyses that collectively advance knowledge of general linear trends, sig-nificant subgroup variations, as well as nonlinear and dynamic processes overtime and circumstance.

The theory bases and research foci are rapidly gaining sophisticationacross numerous fronts in violence research. Advances in directions such asthe study of resilience (Jonzon & Lindblad, 2006), cumulative adversity(Appleyard, Egeland, van Dulmen, & Sroufe, 2005), the psychobiology ofstress responding (A. B. Woods et al., 2005), and persistent socioeconomicand racial/ethnic disparities in risk exposure and negative outcomes (House,2002) are all being linked with violence and are revealing new landscapes ofheterogeneity. Although person-oriented methods are by no means the onlysets of tools germane to meeting these challenges, they are valuable assets,requiring new commitments of training, application, and discourse in theviolence research community.

References

Anderson, D. K., Saunders, D. G., Yoshihama, M., Bybee, D. I., & Sullivan, C. M. (2003).Long-term trends in depression among women separated from abusive partners. ViolenceAgainst Women, 9, 807-838.

Appleyard, K., Egeland, B., van Dulmen, M. H. M., & Sroufe, L. A. (2005). When more is notbetter: The role of cumulative risk in child behavior outcomes. Journal of ChildPsychology and Psychiatry, 46(3), 235-245.

Arata, C. M. (2002). Child sexual abuse and sexual revictimization. Clinical Psychology:Science and Practice, 9(2), 135-164.


Arcel, L. T. (2002). Torture, cruel, inhuman, and degrading treatment of women: Psychologicalconsequences. Torture, 2, 2-16.

Arias, I. (1999). Women’s response to physical and psychological abuse. In X. B. Arriaga &S. Oskamp (Eds.), Violence in intimate relationships (pp. 139-161). Thousand Oaks, CA:Sage.

Bandee-Roche, K., Miglioretti, D. L., Zeger, S. L., & Rathouz, P. J. (1997). Latent variableregression for multiple discrete outcomes. Journal of the American Statistical Association,92(440), 1375-1386.

Bauer, D. J., & Curran, P. J. (2003). Distributional assumptions of growth mixture models:Implications for overextraction of latent trajectory classes. Psychological Methods, 8,338-363.

Bauer, D. J., & Curran, P. J. (2004). The integration of continuous and discrete latent variablemodels: Potential problems and promising opportunities. Psychological Methods, 9(1), 3-29.

Bergman, L. R. (2000). The application of a person-oriented approach: Types and clusters. InL. R. Bergman, R. B. Cairns, L. Nilsson, & L. Nystedt (Eds.), Developmental science andthe holistic approach (pp. 137-154). Mahwah, NJ: Lawrence Erlbaum.

Bergman, L. R. (2001). A person centered approach in research on adolescence: Somemethodological challenges. Journal of Adolescent Research, 16(1), 28-53.

Bergman, L. R., Cairns, R. B., Nilsson, L., & Nystedt, L. (Eds.). (2000). Developmentalscience and the holistic approach. Mahwah, NJ: Lawrence Erlbaum.

Bergman, L. R., & Magnusson, D. (1997). A person-oriented approach in research on devel-opmental psychopathology, Development and Psychopathology, 9, 291-319.

Bergman, L. R., & Trost, K. (2006). The person-oriented versus the variable-orientedapproach: Are they complementary, opposites, or exploring different worlds? Merrill-Palmer Quarterly, 52(3), 601-632.

Bergman, L. R., von Eye, A., & Magnusson, D. (2006). Person-oriented research strategies indevelopmental psychopathology. In D. Cicchetti & D. J. Cohen (Eds.), Developmental psy-chopathology: Theory and method (Vol. 1, pp. 850-888). New York: John Wiley.

Blackwell, L. M., Lynn, S. J., Vanderhoff, H., & Gidycz, C. (2004). Sexual assault revictim-ization: Toward effective risk-reduction programs. In L. J. Koenig, L. S. Doll, A. O’Learry,& W. Pequegnat (Eds.), From child sexual abuse to adult sexual risk, trauma, revictimiza-tion, and intervention (pp. 269-295). Washington, DC: American Psychological Association.

Bogat, G. A., Levendosky, A. A., Theran, S., von Eye, A., & Davidson, W. S. (2003).Predicting the psychosocial effects of interpersonal partner violence: How much does awoman’s history of IPV matter? Journal of Interpersonal Violence, 18(11), 1271-1291.

Bogat, G. A., Levendosky, A. A., & von Eye, A. (2005). The future of research on intimatepartner violence: Person-oriented and variable-oriented perspectives. American Journal ofCommunity Psychology, 36(1/2), 49-70.

Bollen, K. A. (2002). Latent variables in psychology and the social science. Annual Review ofPsychology, 53, 605-634.

Bonanno, G. A. (2004). Loss, trauma, and human resilience: Have we underestimated the humancapacity to thrive after extremely aversive events? American Psychologist, 59, 20-28.

Briere, J., & Jordan, C. E. (2004). Violence against women: Outcome complexity and implica-tions for assessment and treatment. Journal of Interpersonal Violence, 19(11), 1252-1276.

Campbell, J. C. (2002). Health consequences of intimate partner violence. Lancet, 359, 1331-1336.Campbell, J. C., & Lewandowski, L. (1997). Mental and physical health effects of intimate

partner violence on women and children. Psychiatric Clinics of North American, 20(2),353-374.


Campbell, J. C., Snow J. A., Dienemann, J., Kub, J., Schollenberger, J., O’Campo, P., et al.(2002). Intimate partner violence and physical health consequences. Archives of InternalMedicine, 162, 1157-1163.

Campbell, J. C., Sullivan, C. M., & Davidson, W. S. (1995). Women who use domestic violence shelters: Changes in depression over time. Psychology of Women Quarterly, 19,237-255.

Carlson, B. E., McNutt, L., Choi, D. Y., & Rose, I. M. (2002). Intimate partner abuse andmental health. Violence Against Women, 8(6), 720-745.

Casey, E., & Nurius, P. S. (2005). Trauma exposure and sexual revictimization risk:Comparisons across single, multiple incident, and multiple perpetrator victimizations.Violence Against Women, 11(4), 505-530.

Cleland, C. M., Rothschild, L., & Haslam, N. (2000). Detecting latent taxa: Monte Carlo com-parison of taxometric, mixture model, and clustering procedures. Psychological Reports,87(1), 37-47.

Coker, A. L., Davis, K., Arias, I., Desai, S., Sanderson, M. Brandt, H., et al. (2002). Physicaland mental health effects of intimate partner violence for men and women. AmericanJournal of Preventive Medicine, 23(4), 260-268.

Coker, A. L., Watkins, K. W., Smith, P. H., & Brandt, H. M. (2003). Social support reduces theimpact of partner violence on health: Application of structural equation models. PreventiveMedicine, 37, 259-267.

Collins, L. M., Murphy, S. A., & Bierman, K. L. (2004). A conceptual framework for adaptivepreventive interventions. Prevention Science, 5(3), 185-196.

Curran, P. J., & Willoughby, M. T. (2003). Implications of latent trajectory models for thestudy of developmental psychopathology. Development and Psychopathology, 15, 581-612.

Davis, K. E., Coker, A. L., & Sanderson, M. (2002). Physical and mental health effects ofbeing staked for men and women. Violence and Victims, 17, 429-443.

Duncan, T. E., Duncan, S. C., Strycker, L. A., Li, F., & Alpert, A. (1999). An introduction tolatent variable growth curve modeling: Concepts, issues, and applications. Mahwah, NJ:Lawrence Erlbaum.

Eby, K. K., Campbell , J. C., Sullivan, C. M., & Davidson, W. S. (1995). Health effects ofexperiences of sexual violence for women with abusive partners. Health Care for WomenInternational, 16, 563-576.

El-Bassel, N., Gilbert, L., Wu, E., Go, H., & Hill, J. (2005). Relationship between drug abuseand intimate partner violence: A longitudinal study among women receiving methadone.American Journal of Public Health, 95(31), 465-470.

Elliott, D. M., Mok, D., & Briere, J. (2004). Adult sexual assault: Prevalence, symptomatology,and sex differences in the general population. Journal of Traumatic Stress, 17, 203-211.

Everitt, B. S., Landau, S., & Morven, L. (2001). Cluster analysis (4th ed.). New York: OxfordUniversity Press.

Farley, M. (Ed.). (2004). Prostitution, trafficking, and traumatic stress. Binghamton, NY:Haworth.

Gibson, W. A. (1959). Three multivariate models: Factor analysis, latent structure analysis, andlatent profile analysis. Psychometrika, 24, 229-252.

Golding, J. M. (1999). Intimate partner violence as a risk factor for mental disorders: A meta-analysis. Journal of Family Violence, 14(2), 99-132.

Graham, J. W., Collins, L. M., Wugalter, S. E., Chung, N. K., & Hansen, W. B. (1991).Modeling transitions in latent stage-sequential process: A substance use preventionexample. Journal of Consulting and Clinical Psychology, 59(1), 48-57.


Grauerholz, L. (2000). An ecological approach to understanding sexual revictimization:Linking personal, interpersonal, and sociocultural factors and processes. Child Maltreatment,5, 5-17.

Hagenaars, J. A., & Halman, L. C. (1989). Searching for idea types: The potentialities of latentclass analysis. European Sociological Review, 5(1), 81-96.

Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1998). Multivariate data analysis.Upper Saddle River, NJ: Prentice Hall.

Hobfoll, S. E. (1989). Conservation of resources. American Psychologist, 44(3), 513-524.House, J. S. (2002). Understanding social factors and inequalities in health: 20th century

progress and 21st century prospects. Journal of Health and Social Behavior, 43(2), 125-142.Hughes, H. M., & Huth-Bocks, A. C. (2007). Variations in parenting stress in African-

American battered women: Implications for children’s adjustment and family intervention.European Psychologist, 12(1), 62-71.

Jedidi, K., Jagpal, H. S., & DeSarbo, W. S. (1997). STEMM: A general finite mixture struc-tural equation model. Journal of Classification, 14(1), 23-50.

Johnson, H. (1995). Risk factors associated with non-lethal violence against women by mari-tal partners. In C. Block & R. Block (Eds.), Trends, risks, and interventions in lethal vio-lence: Proceedings of the Third Annual Spring Symposium of the Homicide ResearchWorking Group (pp. 152-168). Washington, DC: National Institute of Justice.

Jonzon, E., & Lindblad, F. (2006). Risk factors and protective factors in relation to subjectivehealth among adult female victims of child sexual abuse. Child Abuse & Neglect, 30, 127-143.

Kessler, R. C., Molnar, B. E., Feurer, I. D., & Appelbaum, M. (2001). Patterns and mentalhealth predictors of domestic violence in the United States: Results from the NationalComorbidity Survey. International Journal of Law and Psychiatry, 24, 487-508.

Kline, R. B. (2005). Principles and practice of structural equation modeling (2nd ed.). NewYork: Guilford.

Kohl, P. L., & Macy, R. J. (in press). Profiles of victimized women among the child welfarepopulation: implications for targeted child welfare policy and practices. Journal of FamilyViolence.

Koss, M. P. (1993). Detecting the scope of rape: A review of prevalence research methods.Journal of Interpersonal Violence, 8(2), 198-222.

Land, K. C. (2001). Introduction to the special issue on finite mixture models. SociologicalMethods & Research, 29(3), 275-281.

Lesesne, C. A., & Kennedy, C. (2005). Starting early: Promoting the mental health of womenand girls throughout the life span. Journal of Women’s Health, 14(9), 754-763.

Lindhorst, T., Nurius, P. S., & Macy, R. J. (2005). Contextualized assessment with batteredwomen: Strategic safety planning to cope with multiple harms. Journal of Social WorkEducation, 41(2), 331-352.

Logan, T. K., Walker, R., Cole, J., & Leukefeld, C. (2002). Victimization and substance abuseamong women: Contributing factors, interventions, and implications. Review of GeneralPsychology, 6(4), 325-397.

Macy, R. J. (2007). A coping theory framework for understanding and preventing revictimiza-tion. Aggression and Violent Behavior, 12, 177-192.

Macy, R. J. (in press). A research agenda for sexual revictimization: Priority areas and innov-ative statistical methods. Violence Against Women.

Macy, R. J., & Chapman, M. (2006, January). Using latent class analysis to move beyond aggre-gate findings: A conceptual and practical workshop using Mplus Software. Workshop pre-sented at the Society for Social Work and Research Annual Conference, San Antonio, TX.


Macy, R. J., Nurius, P. S., & Norris, J. (2007a). Latent profiles among sexual assault victims:Highlighting implications for defensive coping and resistance. Journal of InterpersonalViolence, 22(5), 543-565.

Macy, R. J., Nurius, P. S., & Norris, J. (2007b). Utility of person-centered analyses for under-standing sexual aggression victims and their assault experiences. Journal of InterpersonalViolence, 22(5), 520-542.

McCutcheon, A. L. (1987). Latent class analysis. Newbury Park, CA: Sage.McFarlane, J., Parker, B., Soeken, K., Silva, C., & Reel, S. (1998). Safety behaviors of abused

women after an intervention during pregnancy. Journal of Obstetric, Gynecologic, andNeonatal Nursing, 27, 64-69.

Messman-Moore, T. L., & Long, P. J. (2002). Alcohol and substance use disorders as predic-tors of child to adult sexual revictimization in a sample of community women. Violenceand Victims, 17(3), 319-340.

Messman-Moore, T. L., & Long, P. J. (2003). The role of childhood sexual abuse sequelae inthe sexual revictimization of women: An empirical review and theoretical reformulation.Clinical Psychology, 23, 537-571.

Mitchell, C. M., & Plunkett, M. (2000). The latent structure of substance use among AmericanIndian adolescents: An example using categorical variables. American Journal ofCommunity Psychology, 28(1), 105-125.

Muthen, B. O. (2001). Second-generation structural equation modeling with a combination ofcategorical and continuous latent variables: New opportunities for latent class-latentgrowth modeling. In L. M. Collins & A. G. Sayer (Eds.), New methods for the analysis ofchange (pp. 289-322). Washington DC: American Psychological Association.

Muthen, B. O. (2002). Beyond SEM: General latent variable modeling. Behaviormetrika,29(1), 81-117.

Muthen, B. O. (2003). Statistical and substantive checking in growth mixture modeling:Comment on Bauer and Curran (2003). Psychological Methods, 8(3), 369-377.

Muthen, L., & Muthen, B. (1998-2004). Mplus user’s guide. Los Angeles, CA: Muthen &Muthen.

Nurius, P. S., & Norris, J. (1996). A cognitive ecological model of response to sexual coercionin dating. Journal of Psychology and Human Sexuality, 8, 117-139.

Nurius, P. S., Norris, J., Macy, R. J., & Huang, B. (2004). Women’s situational coping withacquaintance sexual assault: Applying an appraisal-based model. Violence Against Women,10, 450-478.

Oxford, M. L., Gilchrist, L. D., Lohr, M. J., Gillmore, M. R., Morrison, D. M., & Spieker,S. J. (2005). Life course heterogeneity in the transition from adolescence to adulthoodamong adolescent mothers. Journal of Research on Adolescence, 15(4), 479-504.

Plichta, S. B. (2004). Intimate partner violence and physical health consequences-policy andpractice implications. Journal of Interpersonal Violence, 19(11), 1296-1323.

Raudenbush, S. W. (2001). Toward a coherent framework for comparing trajectories of indi-vidual change. In L. M. Collins & A. G. Sayer (Eds.), New methods for the analysis ofchange (pp. 33-64). Washington, DC: American Psychological Association.

Resick, P. A. (1993). The psychological impact of rape. Journal of Interpersonal Violence,8(2), 223-255.

Riger, S., Raja, S., & Camacho, J. (2002). The radiating impact of intimate partner violence.Journal of Interpersonal Violence, 17(2), 184-205.

Rozee, P. D., & Koss, M. P. (2001). Rape: A century of resistance. Psychology of WomenQuarterly, 25, 295-311.


Ryff, C. D., & Singer, B. (2003), Ironies of the human condition: Well-being and health on theway to mortality. In L. G. Aspinwall & U. M. Staudinger (Eds.), A psychology of humanstrengths: Fundamental questions and future directions for a positive psychology (pp. 271-287). Washington DC: American Psychological Association.

Smith, P. H., Smith, J. B., & Earp, J. A. L. (1999). Beyond the measurement trap a recon-ceptualization and measurement of woman battering. Psychology of Women Quarterly, 23,177-193.

Staggs, S., & Riger, S. (2005). Effects of intimate partner violence on low-income women’shealth and employment. American Journal of Community Psychology, 36(1/2), 133-145.

Tjaden, P., & Thoennes, N. (2000). Prevalence and consequences of male-to-female andfemale-to-male intimate violence as measured by the National Violence Against WomenSurvey. Violence Against Women, 6, 142-161.

Tjaden, P., & Thoennes, N. (2006). Extent, nature and consequences of rape victimization:Findings from the national violence against women survey. Washington DC: NationalInstitute of Justice.

Tolman, R. M., & Wang, H. (2005) Domestic violence and women’s employment: Fixedeffects models of three waves of women’s employment study data. American Journal ofCommunity Psychology, 36(1/2), 147-158.

Updegraff, J. A., & Taylor, S. E. (2000). From vulnerability to growth: Positive and negativeeffects of stressful life events. In J. H. Harvey & E. D. Miller (Eds.), Loss and trauma:General and close relationship perspectives (pp. 3-28). New York: Brunner-Routledge.

Updegraff, J. A., Taylor, S. E., Kemeny, M. E., & Wyatt, G. E. (2002). Positive and negativeeffects of HIV infection in women with low socioeconomic resources. Personality andSocial Psychology Bulletin, 28(3), 382-394.

von Eye, A., & Bergman, L. R. (2003). Research strategies in developmental psy-chopathology: Dimensional identity and the person-oriented approach. Development andPsychopathology, 15, 553-580.

von Eye, A., & Bogat, G. A. (2006). Person-oriented and variable-oriented research: Concepts,results, and development. Merrill-Palmer Quarterly, 52(3), 390-420.

von Eye, A., Bogat, G. A., & Rhodes, J. E. (2006). Variable-oriented and person-oriented per-spectives of analysis: The example of alcohol consumption in adolescence. Journal ofAdolescence, 29(6), 981-1004.

Williams, L. M. (2003). Understanding child abuse and violence against women: A life courseperspective. Journal of Interpersonal Violence, 18(4), 441-451.

Woods, A. B., Page, G. G., O’Campo, P., Pugh, L. C., Ford, D., & Campbell, J. (2005). Themediation effect of posttraumatic stress disorder symptoms on the relationship of intimatepartner violence and IFN-v levels. American Journal of Community Psychology, 36(1/2),159-175.

Woods, S. J. (2000). Prevalence and patterns of posttraumatic stress disorder in abused andpostabused women. Issues in Mental Health Nursing, 21(3), 309-324.

Yeater, E. A., & O’Donohue, W. (1999). Sexual assault prevention programs: Current issues,future directions, and the potential efficacy of interventions with women. ClinicalPsychology Review, 19(7), 739-771.

Yoshihama, M., & Horrocks, J. (2002). Posttraumatic stress symptoms and victimizationamong Japanese American women. Journal of Consulting and Clinical Psychology, 70(1),205-215.


Zlotnick, C., Johnson, D., & Kohn, R. (2006). Intimate partner violence and long-term psy-chosocial functioning in a national sample of American women. Journal of InterpersonalViolence, 21(2), 262-275.

Paula S. Nurius is professor and director of the Prevention Research Training Program at theUniversity of Washington School of Social Work. Her research addresses coping responses torelationship violence, risk and protective factors impinging on protective action, and preven-tive interventions.

Rebecca J. Macy, PhD, ACSW, LCSW, is an assistant professor at the School of Social Workat the University of North Carolina at Chapel Hill. Her research focuses on the health conse-quences of violent victimization and revictimization.


Heterogeneity Among Violence-Exposed Women: Applying Person-Oriented Research Methods

Documents