Close Relationships

A Sourcebook

Clyde Hendrick, Susan S. Hendrick, Clyde A. Hendrick, Susan S. Hendrick

• Publisher: SAGE Publications, Inc
• Year: 2000
• Print ISBN: 9780761916062
• eBook ISBN: 9781452236117
• Edition: 1
• Topic: Psychology
• Subtopic: Social Psychology
• Index: Psychology

"The authors in the volume extend the reach of their comprehensive reviews into theoretically driven and innovating explorations. The scope of coverage across and within chapters is striking. The developmentalist, the methodologist, the feminist, the contextualist, and the cross culturalist alike will find satisfaction in reading the chapters."

-Catherine A. Surra, The University of Texas at Austin

The science of close relationships is relatively new and complex. Close Relationships: A Sourcebook represents the growing maturity of this multidisciplinary enterprise. The volume offers 26 chapters organized into four thematic areas: relationship methods, forms, processes, and threats, as well as a foreword and an epilogue. The volume provides a panoramic view of close relationship research as it enters the 21st century, offering highlights from current literature, original research, practical applications, and projections for future research. Relationship Methods includes both qualitative and quantitative chapters. Relationship Forms includes many of the stages, types, and roles that characterize intimate relationships. In a developmental fashion, chapters address social networks, children?s friendships, adolescent relationships, adult friendships, and friendships in later life. Chapters on multicultural and multiracial relationships and gay, lesbian, and bisexual relationships illustrate the variety of relationship forms that the science of close relationships must consider. The alignments and realignments of traditional family structure are considered in terms of contemporary marriage, divorce and single parenting, and remarried families. Relationship Processes includes chapters on emotion, attachment, romantic love, sexuality, intimacy, communication, conflict, social support, and relational maintenance. The important topic of gender concludes the section. The shadow side of human nature is explored in the Relationship Threats section, with chapters on infidelity and jealousy, physical and sexual aggression, depression, and loss and bereavement. A foreword by Ellen Berscheid sets the stage for this broad-ranging collection of chapters. Steve Duck and Linda Acitelli conclude with an epilogue that provides a new beginning for the science of close relationships.

Part I

---

RELATIONSHIP METHODS

Contents

---

Introduction to Nonindependence

An Illustration of the Pitfalls of Ignoring Interdependence

What This Chapter Does and Does Not Do

Quantitative Methods for Dyadic Research

Distinguishability of Dyad Members

Types of Independent Variables

Hypothetical Research Example

Setting Up the Data

Exploring the Data

Testing the Average Effects of Within-Dyads Independent Variables (distinguishable dyads only)

Testing the Effects of Between-Dyads Independent Variables

Testing the Effects of Mixed Independent Variables

Quantitative Methods for Multiple Partner Designs: Friendship Groups and Families

The SRM With Friendship Networks and Other Nondistinguishable Groups

The SRM With Families

Conclusion

Quantitative Methods in Close Relationships Research

---

Deborah A. Kashy
Maurice J. Levesque

Data analysis in the field of close relationships presents unique and important challenges to researchers. The vast majority of available statistical techniques assume that the data we collect are independent from individual to individual, but such independence rarely exists in relationships research. In relationships, individuals are interdependent rather than independent. It is not uncommon for relationship researchers to avoid the independence problem by collecting data from only one person involved in a relationship or to use hypothetical scenarios. For these methods, researchers do not need to be concerned with issues of nonindependence. However, because interdependence is perhaps the defining feature of close relationships (Kelley et al., 1983; Kelley & Thibaut, 1978), avoiding interdependent data restricts researchers from seeking answers to many important questions about close relationships. Researchers are becoming increasingly aware of this problem, and recent relationships research more often acknowledges the importance of interdependence by collecting data from both members of dyadic relationships (e.g., Bradbury, 1998).

► Introduction to Nonindependence

The fact that data derived from individuals who are engaged in close relationships are not independent has important implications for how the data should be analyzed and the questions researchers can ask of those data. There are two major issues involved in the analysis of nonindependent data. The first issue is that of bias in hypothesis testing. Speaking somewhat generally, if a statistical technique that assumes independence (e.g., analysis of variance [ANOVA], regression) is used with nonindependent data, then the alpha level associated with the inferential statistics generated will not accurately reflect the true probability of making a Type I error. As Kenny and his colleagues (Kenny, 1995; Kenny, Kashy, & Bolger, 1998) have shown, in some instances the statistical tests will be overly liberal (too many false positives), and in other instances the tests will be overly conservative (too many false negatives).

The second issue concerns the types of questions that we, as relationship researchers, can address. In particular, one of the most important advantages of gathering nonindependent data (data from both/all partners involved in a relationship) is that researchers can examine not only how a person’s characteristics affect his or her own behavior but also how that person’s characteristics affect his or her partner’s behavior. These interdependent effects often are implied by theories regarding close relationships. For example, although perceptions of equity in relationships can be conceptualized purely in terms of one person’s inputs and outcomes, it seems appropriate with respect to the theory to consider how perceptions of equity are influenced by a partner’s inputs and outcomes.

Consider the following examples of relationship theory-based research that reveal the recent movement toward questions of interdependence. First, does the level of attachment-related anxiety exhibited by one partner influence the other’s behavior? Attachment research has demonstrated that individuals are influenced by their partners’ styles as well as by their own (Carnelley, Pietromonaco, & Jaffe, 1996; Simpson, 1990). Theories of adult attachment also articulate the role of partner behavior in activating attachment-relevant schemata (Berman, Marcus, & Berman, 1994). Second, does a person’s criticism lead his or her partner to be defensive and withdraw? Gottman’s (1979, 1994b) work on marital conflict has long been devoted to identifying patterns of interdependent interaction during conflict. Identifying the cycle of dysfunctional conflict has been possible only because researchers have collected data from both partners to allow for analyses that recognize the nonindependence of married persons’ behavior. Finally, treating the data from couples as nonindependent allows for the analysis of data addressing long-standing issues related to couple similarity (Kenny & Acitelli, 1994) and for the examination of new questions regarding the influence of positive illusions on well-being and relationship satisfaction (Murray, Holmes, & Griffin, 1996). These are just a few examples of the questions now being examined by relationship researchers. Clearly, interdependence between related individuals should not simply be considered a statistical annoyance. Rather, it should be embraced as an opportunity to ask old questions in new ways, to ask new questions, and to test theoretical propositions that are explicitly about interdependence.

An Illustration of the Pitfalls of Ignoring Interdependence

To highlight a few of the problems that can arise from ignoring interdependence, consider a hypothetical study designed to examine the use of criticism in distressed and nondistressed heterosexual dating couples. In this study, both members of the dating couple are recruited, and each couple is classified as distressed or nondistressed. The amount of criticism each person uses is assessed by self-report.

Because data are collected from both members of the couple, this study avoids one common but problematic method of coping with interdependence: collecting information from only one member of a relationship. However, even when data are collected from both persons, the researcher might be tempted to analyze the data from males and females separately to avoid the analytic complications presented by interdependence. Although these approaches do not violate the statistical assumption of independence, the researcher who employs these tactics sacrifices considerable information by not taking full advantage of techniques for analyzing nonindependent data. In this case, the researcher who fails to consider the interdependence of the dating partners misses an opportunity to examine how the use of criticism by one person affects the partner’s use of criticism and whether the strength of that association depends on the status of the relationship.

Even by collecting information from both partners and realizing the nonindependence of the male and female responses, the researcher might adopt yet another problematic approach. Specifically, he or she might collapse across dyad members by computing the average criticism score for each dyad. Again, the researcher loses valuable information with such an approach, and in some instances, the resulting statistics might be misleading. For example, Gonzalez and Griffin (1997) discussed how the correlations between averaged variables that result from this approach can be quite different from those obtained using each individual’s scores.

Imagine now that the researcher collects data from both members of the couple but is unaware of or ignores the nonindependence. Furthermore, as one might expect, criticism is strongly reciprocal, so there is a positive correlation between men’s and women’s use of criticism. Because the interdependence is not considered, the researcher chooses to analyze the data treating person, not couple, as the unit of analysis (i.e., the sample size in these analyses is based on the number of individuals, not on the number of couples) and uses between-subjects data analytic techniques. The researcher first tests whether there are sex differences in the use of criticism. Although the means suggest that men use criticism more frequently than do women, the statistical test fails to obtain significance and the researcher concludes that men and women do not differ in the frequency of criticism. The researcher then tests whether there are differences in the use of criticism as a function of whether the individual is a member of a distressed or nondistressed couple. This test finds a statistically significant difference suggesting greater criticism in distressed couples.

Nonindependence compromises the validity of both of these statistical tests. The sex difference analysis is problematic because sex is an independent variable that varies within a couple (each couple has a man and a woman), and ignoring nonindependence actually increases the likelihood of a Type II error (failure to reject a false null hypothesis) (Kenny, 1995). Contrary to the researcher’s conclusions, it might be that the sex difference really is statistically reliable. On the other hand, because distress is an independent variable that varies between couples (i.e., some couples are distressed and others are not), the test of distress is an overly liberal test. That is, this test is more likely to reflect a Type I error than the p value derived from the analysis suggests. Thus, by ignoring the nonindependence, the researcher risks missing a significant sex difference and might conclude that distressed couples use more criticism than do nondistressed couples when there is no such statistically significant difference. As Kenny and his colleagues (Kenny, 1995; Kenny et al., 1998) illustrate, the error depends on the nature of the nonindependence (i.e., positive or negative correlation between partners on the dependent measure) and on the nature of independent variable. Although the errors that result from employing inappropriate analyses will not always be so dramatic, researchers who ignore interdependence risk such errors unnecessarily because, as we show, statistical techniques are readily available for analyzing interdependent data.

What This Chapter Does and Does Not Do

In this chapter, we focus our attention on data analytic techniques that are appropriate for studying friendship dyads, heterosexual and homosexual romantic couples, marital dyads, and families. Because the majority of research in close relationships tends to consider two-person relationships, most of the analyses we describe are for dyadic data. Kashy and Kenny (2000) expanded a number of these analyses to groups with more than two members. We provide an overview of one approach to the analysis of friendship networks that also can be applied (with some modifications) to family data. Our presentation is generally descriptive, so we do not present exact formulas. Instead, we provide citations throughout the chapter to sources that detail the computations for the various tests we describe.

Although not covered in this chapter, one of the newest methodological and data analytic advances in the study of close relationships is multilevel modeling (also known as hierarchical linear modeling). This approach can be applied to the analysis of designs involving repeated measures and data in which there are two or more levels of analysis (e.g., individuals nested within groups). Multilevel modeling is particularly important for the analysis of research that uses the social interaction diary methodology. For example, one recent use of this analysis strategy examined the influence of attachment style on perceptions of a variety of relationships over time (Pietromonaco & Barrett, 1997). Details regarding the methodology and analysis of hierarchical linear models can be found in Bryk and Raudenbush (1992), Gable and Reis (1999), and Kenny et al. (1998).

► Quantitative Methods for Dyadic Research

To begin a discussion of dyadic data analysis, one first must be aware that the appropriate analysis depends on a number of factors. All of the methods we illustrate assume that the dependent or outcome measures are interval or ratio data. The analysis of designs with nominal or ordinal dependent measures often requires other techniques that are detailed in a number of sources (Bakeman 1991; Bakeman & Quera, 1995; Gottman & Roy, 1990; Kashy & Snyder, 1995; Wickens 1993). In dyadic designs, analysis strategy choice is determined both by characteristics of the dyad and by characteristics of the independent variables.

Distinguishability of Dyad Members

One important question in dyadic research and data analysis is whether the two dyad members can be distinguished from one another by some variable. In heterosexual dating relationships, dyad members are distinguishable because of their gender; each couple has one man and one woman. Similarly, in non-twin sibling dyads, the two siblings can be distinguished by birth order. However, there are many instances in which there is no such natural distinction. Same-sex friendship pairs, homosexual romantic partners, and twins all are examples of what we call indistinguishable dyads. The distinguishability issue is critical in a discussion of quantitative methods for relationship data because the data analytic techniques appropriate for distinguishable dyads might not be appropriate for indistinguishable dyads.

Types of Independent Variables

There are three types of independent or predictor variables in dyadic research, and the appropriate data analytic approach depends on the type of variable being studied. Within-dyads independent variables are those that vary across the two dyad members but do not vary, on average, from dyad to dyad. In heterosexual dating couples, gender is a categorical within-dyads variable because each couple has both a man and a woman, but the “average” gender score (averaging over the two dyad members) is constant across couples. A continuous within-dyads predictor variable might be the proportion of child care tasks completed by married couples who have children. Again, the proportion differs between members of a dyad, but across all dyads the proportions sum to 100%. Note that whenever there is a categorical within-dyads predictor variable, the dyad members are distinguishable with respect to that variable.

Between-dyads independent variables are those that vary from dyad to dyad, but within a dyad both members have the same score. A manipulated between-dyads variable might be stress level in a study in which some couples are placed in stressful situations and others are not. A common between-dyads variable in marital conflict research is the categorization of couples as distressed or nondistressed. A continuous between-dyads variable might be the length of acquaintance for friendship dyads.

The final type of independent variable is a mixed variable. Mixed independent variables vary both within and between dyads. In a study of the effects of attachment avoidance on dyadic conflict, avoidance would be a continuous mixed predictor because dyad members will tend to differ on avoidance, and the average level of avoidance will be high in some couples and low in others. Gender would be a categorical mixed variable in a study that included both heterosexual and homosexual dating couples. Although mixed variables present a number of data analytic challenges, they also present some of the best opportunities to study the interdependence between individuals who are involved in close relationships.

Hypothetical Research Example

To assist us in our presentation, consider an example based loosely on Simpson, Rholes, and Phillips’s (1996) study of the effects of attachment, stress, and gender on problem solving in dating couples. In their study, the attachment orientations of both members of heterosexual dating couples were assessed, and couples then were assigned to discuss either a major or minor relationship problem. Following the videotaped dyadic interactions, participants indicated their perceptions of their partners and of their relationships. The videotapes were coded for, among other things, signs of stress and anxiety. In this study, gender is a within-dyads predictor variable, the importance of the relationship problem is a between-dyads predictor variable, and anxious/ambivalent and avoidant attachment orientations are two mixed predictor variables. Behavioral manifestations of stress and anxiety are outcome measures.

Although we retain the basics of Simpson et al. (1996), we need to make some modifications to the design that will enable us to illustrate all of the designs, models, and analysis issues we will discuss. Consider the following variation of the Simpson et al. study. Researchers are interested in the factors that might predict the tendency for married individuals to engage in destructive conflict resolution behaviors. They predict that gender, avoidant attachment style, length of relationship, and each individual’s assessment of relationship equity all influence both the tendency to use criticism during discussions of relationship problems and general relationship satisfaction. In addition, they predict that the severity of the problem discussed will affect the use of criticism and postdiscussion relationship satisfaction. The researchers recruit a sample of married couples and categorize one person in the dyad as more distressed and the other as le...

Table of contents

• Cover page
• Title
• Copyright
• Contents
• Foreword: Back to the Future and Forward to the Past
• Preface
• Part I: RELATIONSHIP METHODS
• Part II: RELATIONSHIP FORMS
• Part III: RELATIONSHIP PROCESSES
• Part IV: RELATIONSHIP THREATS
• References
• Name Index
• Subject Index
• About the Editors
• List of Contributors