The Structure of Aggression in Conflict-Prone Couples: Validation of a Measure of the Forms and Functions of Intimate Partner Aggression (FFIPA)

Abstract

Objective

Intimate partner aggression (IPA) is a complex construct comprised of the means and the motivations by which a person harms their intimate partner. Existing measures only assess forms of IPA perpetration while neglecting to measure the motivations for aggressing. The present study sought to fill this lacuna by adapting and validating an existing measure of the forms and functions of adolescent peer aggression to assess IPA perpetration in adults. This new measure – the Forms and Functions of Intimate Partner Aggression (FFIPA) - comprises four latent dimensions of IPA (i.e., overt, relational, proactive, and reactive).

Method

Participants were 341 heavy drinking heterosexual couples (N = 682) with a recent history of psychological and/or physical IPA recruited from two metropolitan cities in the United States.

Results

The FFIPA demonstrated good model fit and internal validity. Unique patterns of convergent and criterion-related validity supported the four dimensions of the FFIPA. Results also indicated women perpetrated significantly more overt and relational aggression than men.

Conclusions

Findings support the FFIPA as a valid measure of the forms and functions of IPA perpetration. More importantly, as the only instrument that parses the forms and functions of IPA perpetration, the FFIPA delineates the unique motivations of an aggressive partner separately from the form of their aggressive behavior(s). Further replication is needed to generalize this measure to non-conflictual and other types of intimate relationships.

Intimate partner aggression (IPA) is a pervasive public health problem that includes use of myriad forms of aggression by a current or former intimate partner and precipitates multiple negative health impacts (Devries et al., 2013; Smith et al., 2017). A variety of IPA measures exist that capture the myriad forms of IPA (Thompson, Basile, Hertz, & Sitterle, 2006). However, despite extensive theory regarding the motivations for IPA perpetration (e.g., Holtzworth-Munroe & Stuart, 1994; Johnson, 2001), few measures assess the function of IPA (for exceptions, see Ross & Babcock, 2009) and no measures assess both the form and the function of IPA. The current study addresses these limitations by utilizing a well-validated conceptual framework – the Forms and Functions of Aggression Model (Little, Jones, Henrich, & Hawley, 2003) – to assess the forms and functions of IPA. Specifically, we sought to adapt and validate the Forms and Functions of Aggression Measure that assesses peer aggression in adolescent populations (Little et al., 2003) for use with adult heterosexual couples at risk for IPA.

Conceptualization of Aggression

Human aggression is a multifaceted behavioral construct that has been defined broadly as the intent to harm (Berkowitz, 1993). Initially, researchers categorized aggressive behaviors simply by the shape that the behavior took in its expression, primarily positive (overt) or negative (covert/relational) (Rosenzweig, 1941). Shortly thereafter, conceptual definitions of verbal and physical aggression (Buss, 1961; see also Berkowitz, 1994; Buss & Perry, 1992) and direct/indirect aggression (Feshbach, 1969; see also Baron & Richardson, 1994; Björkqvist, 1994) were also advanced. Despite these related categorizations of aggression by form, a two-term categorization system (overt versus relational) has been popularly adopted that best captures these related constructs (Little et al., 2003). Overt (direct) aggression is generally defined as verbal and physical behaviors that are directed at victims (Coie & Dodge, 1998). Relational (indirect) aggression is generally defined as behaviors that attack the social standing and inclusion of the victim (Crick & Grotpeter, 1995).

In addition to conceptualizations of the form in which aggression is expressed, Geen (1968; see also Bandura, 1973; Dodge & Coie, 1987) further classified aggression into two broad categories – premeditated and impulsive – based on intention of the perpetrator. Despite the litany of terms that emerged in this literature (e.g., provocative, premeditated, offensive, retaliatory, instrumental, hostile, impulsive, defensive, etc.), contemporary research has typically characterized these behaviors as reactive and proactive aggression, as these best capture these numerous, redundant terms (Babcock, Tharp, Sharp, Heppner, & Stanford, 2014). Reactive aggression is generally defined as an angry, defensive behavior in response to some provocation (Dollard, Doob, Miller, Mowrer, & Sears, 1939). Proactive aggression is generally defined as a self-serving, premeditated, goal-directed behavior in which aggression is a means to an end other than to simply harm (Bandura, 1973; Bushman & Anderson, 2001).

Together, these two-factor typologies of the forms (i.e., overt and relational) and the functions (i.e., proactive and reactive) of aggression has received robust empirical support (e.g., Björkqvist, Lagerspetz, & Kaukiainen, 1992; Poulin & Boivin, 2000; Raine et al., 2006) and largely guided research on aggression. Thus, it is not surprising that a large body of instruments have been developed to assess aggression either by its forms (e.g., Aggression Questionnaire; BAQ; Buss & Perry, 1992; Self-Report of Aggression and Social Behavior Measure; Morales & Crick, 1998; the Direct and Indirect Aggression Scales; Björkqvist et al., 1992) or functions (e.g., Modified Overt Aggression Scale; Kay, Wolkenfeld & Murrill, 1988; Instrument for Reactive and Proactive Aggression-Self-Report; Polman, Orobio de Castro, Thomaes, & van Aken, 2009; Reactive and Proactive Aggression Questionnaire; RPQ; Raine et al., 2006). A perhaps unintended consequence of this practice is that research rarely considers the multiple forms and motives that may characterize any one act of aggression (Bushman & Anderson, 2001; Warburton & Anderson, 2001). As such, it is important to consider the interaction of the form and function of a behavior simultaneously.

Forms and Functions of Aggression Measurement

Given this need for operational clarity among such numerous, but related types of aggression, researchers have aimed to address this lack of consensus by developing more parsimonious measures of aggression that aggregate all the various forms and functions (Forms and Functions of Aggression Measure, Little et al., 2003; Peer-Conflict Scale, Marsee et al., 2011). Following a review of the literature, the Forms and Functions of Aggression Measure (Little et al., 2003) has been subjected to the most empirical scrutiny. What distinguishes this developmentally-based, multidimensional measure from other measures of the forms and/or functions of aggression is its grounding in a framework that treats aggressive behavior as an intersection between both the form and function of aggression while quantifying the unique influence of the form and the function separately in a way that was not possible before. This measure defines the forms of aggression as overt (i.e., verbal and physical behaviors directed at victims) and relational (i.e., acts intended to damage social standing; e.g., destruction of relationships, social exclusion/ostracism, gossip, etc.) and defines the functions of aggression as reactive (i.e., hostile behavior that occurs as an angry, defensive response to provocation) and proactive (i.e., self-serving, deliberate behavior in anticipation of a reinforcing outcome). Importantly, this measure makes possible the examination of common and unique correlates of the forms and functions of aggression to understand more holistically any aggressive act.

The Forms and Functions of Aggression Measure has been validated as a robust measure of the forms (i.e., overt, relational) and functions (i.e., proactive, reactive) of aggression in many samples including middle class adolescents (Little et al., 2003), ethnic minority adolescents (Williford & Boulton, 2013), high-risk adolescents (Lee, Penney, Odgers, & Moretti, 2010), and socioeconomically diverse adolescents (Sijtsema et al., 2010) and has demonstrated strong construct validity and internal consistency (Fite, Stauffacher, Ostrov, & Colder, 2008; Little et al., 2003). This measure has also demonstrated strong concurrent validity, having been examined in association with adolescent temperamental reactivity and self-regulation (Dane & Marini, 2014), psychopathic traits including impulsivity and callous-unemotional traits (Orue, Calvete, & Gamez-Guadix, 2016), antisocial behavior (Little et al., 2003) and conduct problems (Fite, Stoppelbein, Gaertner, Greening, & Elledge, 2011). It has also demonstrated good predictive validity, accurately modeling the development of adolescent aggression (Ojanen & Kiefer, 2013; Sijtsema et al., 2010) and adult relational aggression (Schmidt & Jankowski, 2014).

A New Approach to the Measurement of Intimate Partner Aggression

In accordance with the general aggression literature, aggression within an intimate relationship has been measured in myriad ways depending on the theoretical basis of specific measures (Wilson, Mouilso, Gentile, Calhoun, & Zeichner, 2015). Most commonly, measures of IPA are typically defined by the form of IPA measured (e.g., physical, psychological, or sexual aggression) (e.g., Conflict Tactics Scale-2, Straus, Hamby, Boney-McCoy, & Sugarman, 1996; Sexual Strategies Scale, Strang, Peterson, Hill, & Heiman, 2013; Women’s Experience with Battering Scale, Smith, Earp, & DeVellis, 1995; Hurt/Insult/Threaten/Scream, Sherin, Sinacore, Li, Zitter, & Shakil, 1998). However, many of these operational definitions do not fully capture the well-established overt-relational categorization system. Most notably, the widely used Conflict Tactics Scale-2 assessment of psychological aggression exclusively references behaviors that are overt (e.g., shouting at one’s partner), but not relational, in form. Lately, measures have emerged that provide more gender-invariant, comprehensive delineations of forms of perpetration (e.g., Partner Victimization Scale, Hamby, 2016; Partner Violence Screen, Feldhaus et al., 1997; Multidimensional Emotional Abuse Scale, Murphy & Hoover, 1999). Thus, while most measures of IPA assess one form or another, no measure of IPA assesses the predominant functions of aggression as defined above (i.e., proactive, reactive), although some studies have assessed contextual factors that generally map onto reactive (e.g., high-arousal, Gottman et al., 1995) or proactive (e.g., “instrumental”, Babcock, Jacobson, Gottman, & Yerington, 2000) motivations. Most importantly, no published measure of IPA assesses concurrently the forms and functions of aggression. This is a weakness that, if corrected, has the potential to expand the conceptualization and measurement of IPA in a novel direction that may further elucidate the nature of conflictual intimate relationships. To address this weakness, the current study adapted the Forms and Functions of Aggression Measure for the measurement of IPA perpetration and sought to confirm the new measure’s factor structure and construct validity in a sample of heterosexual couples at risk for IPA.

The Present Study

The premise of the present study is that understanding IPA utilizing a forms and functions measure of aggression provides an ideal framework for defining and measuring the type of aggressive behavior utilized and understanding the perpetrator’s motive. In particular, despite extensive theory regarding the motivations for IPA perpetration (e.g., Holtzworth-Munroe & Stuart, 1994; Johnson, 2001), no existent measure of IPA clearly captures the motivations for aggressing (i.e., proactive, reactive) as postulated herein. Considering the motivation fueling conflict has been identified as an important gap in the literature, understanding motivation may provide important context for conceptualizing and preventing violence (Ross & Babcock, 2009). As such, the present study adapted Little et al.’s (2003) Forms and Functions of Aggression Measure to assess IPA (Forms and Functions of Intimate Partner Aggression; FFIPA) and had two primary aims: (1) Confirm the factor structure of the FFIPA; and (2) Examine the construct validity of the FFIPA. To support the construct validity of the FFIPA, we hypothesized that FFIPA constructs would correlate with other established measures of the functions of aggression and the forms of IPA perpetration. We also hypothesized that the FFIPA constructs would correlate with established risk factors for IPA perpetration, including trait anger (Birkley & Eckhardt, 2015), relationship dissatisfaction (Ulloa & Hammett, 2015), problematic drinking (Leonard, 2005), hostility (Birkley & Eckhardt, 2015), and a facet of impulsivity – negative urgency (Cyders & Smith, 2007) – that has been specifically associated with IPA perpetration (Leone, Crane, Parrott, & Eckhardt, 2016). In addition to these aims and hypotheses, this measure also permits an innovative exploration of possible gender differences in IPA perpetration. As such, potential gender differences in the use of the forms and functions of IPA were examined utilizing the FFIPA.

Method

This investigation utilized data collected as part of a larger study examining the effects of acute alcohol intoxication on IPA perpetration in heterosexual intimate partners. Although the focus of the current study does not examine effects of acute alcohol intoxication, couples were required to meet eligibility criteria for an alcohol administration study (see below). Measures pertinent to the current study were individually administered on a day which preceded the alcohol administration laboratory session. The present hypotheses are novel and have not been tested previously within the parent project.

Participants

Participants were 341 heterosexual couples who had been in their current relationship for at least one month prior to participation (N = 682). However, missing data for one member of six distinct couples produced a final sample of 674 individuals comprising 340 couples. Most participants self-identified as African American (43.1%) or Caucasian (44.7%), were 29 years old on average (SD = 8.48, range = 21–58), and had an average relationship length of 4.1 years (SD = 4.7 years, range = 1 month–38.5 years). Participants were compensated $10 (USD) per hour.

Couples were recruited from two metropolitan U.S. cities through advertisements placed in online/social media sites, community newspapers, and public transportation for a two-session study. Individual members of each couple were initially screened separately by telephone. To be eligible, couples had to be dating for at least 1 month and each partner had to be at least 21 years of age and identify English as their native language. Couples were excluded if either partner reported serious head injuries, a condition in which consuming alcohol is medically contraindicated, or a desire to seek treatment for alcohol use. In addition, at least one partner was required to meet two additional eligibility criteria. First, this individual had to report consumption of an average of at least five (for men) or four (for women) standard alcoholic beverages per occasion at least twice per month during the past year. Second, this individual had to be identified as perpetrating psychological or physical IPA toward their current partner via self- or partner-report on the CTS-2. Eligible couples were scheduled to complete a more comprehensive in-person laboratory assessment (Session 1), which served to confirm eligibility for a subsequent alcohol administration laboratory session (Session 2). This current study included all participants who completed Session 1 irrespective of their subsequent eligibility for or completion of Session 2. This study was approved by each university’s Institutional Review Board, and complied with the American Psychological Association’s ethical standards for the treatment of human subjects.

Materials

Demographic form

This form obtains information such as age, self-identified sexual orientation, race, relationship status, years of education, and yearly family income.

Forms and Functions of Intimate Partner Aggression (FFIPA; see Appendix A)

The original FFAM (Little et al., 2003) is a 36-item self-report measure that assesses the underlying functions (proactive/reactive) and the observed behavioral forms (overt/relational) of aggression via six subscales (i.e., overt, relational, proactive-overt, reactive-overt, proactive-relational, reactive-relational) that in turn are used to compute four latent constructs (i.e., reactive, proactive, overt, and relational aggression). Participants are instructed to indicate on a 4-point scale how well each item applies to them. Responses may range from 1 (Not at all true) to 4 (Completely true). “Pure” overt and relational aggression (see Figure 1) are assessed with items that measure only the dispositional, pure form variants whereas “pure” proactive and reactive aggression are assessed with items that measure both overt and relational forms of aggression together with their motivation; therefore, “pure” proactive and reactive aggression are unobserved, second order constructs that distill the unique motivation variance (e.g., reactive) from two subscales (e.g., reactive-relational and reactive-overt). Therefore, internal reliability coefficients cannot be computed for these unobserved variances attributable to motivation. Items for proactive aggression capture aggressive behaviors that are deliberate and self-serving without prior provocation (e.g., “I often start fights to get what I want”). Items for reactive aggression capture retaliatory, angry responses to provocation (e.g., “If others make me mad or upset, I often hurt them”). Items for overt aggression capture direct/visible verbal or physical aggression (e.g., “I’m the kind of person who hits and kicks others”) and have demonstrated good reliability, α = .83 - .85 (Lee et al., 2010). Items for relational aggression describe indirect/socially manipulative forms of aggression (e.g., “I’m the kind of person who spreads rumors about others”) and have demonstrated good reliability, α = .75 - .79 (Lee et al., 2010). Little et al. (2003) found alpha reliabilities ranging from .62 (“pure” relational aggression) to .84 (proactive overt aggression) in the original validation sample.

Figure 1.

Proposed factor structure of the FFIPA. Boxes denote parceled items.

The current study adapted the FFAM by changing item wordings to reflect aggressive behaviors directed at the participant’s partner. For example, Item 4 on the FFAM, “I’m the kind of person who puts others down” was modified to “I’m the kind of person who puts my partner down” on the FFIPA. Information on the FFIPA’s reliability in the current sample is reported in the Results section.

The Revised Conflict Tactics Scale – 2 (CTS-2; Straus et al., 1996)

The revised CTS is a 78-item self-report instrument that measures a range of behaviors that occur during disagreements within intimate relationships across five separate subscales, including physical assault, psychological aggression, injury, sexual coercion, and negotiation. Responses may range from 0 (never in the last year) to 6 (more than 20 times in the last year), and the frequency of behavior on each subscale is calculated by adding the midpoints of the score range for each item to form a total score. For example, if a participant indicates a response of “3–5” times in the past year, a score of “4” would be assigned. The current study utilized participants’ self-reported frequency of physical (twelve items, e.g., “I threw something at my partner that could hurt”) and psychological (eight items, e.g., “I destroyed something belonging to my partner”) IPA perpetration, as these CTS-2 subscales assess constructs most relevant to the forms of IPA perpetration assessed by the FFIPA. In the current study, these two subscales demonstrated good reliability (psychological IPA: α = .78; physical IPA: α = .80).

The Buss-Perry Aggression Questionnaire (BAQ; Buss & Perry, 1992)

The BAQ is a self-report questionnaire that measures dispositional aggression. This 29-item questionnaire contains four subscales: anger, physical aggression, verbal aggression, and hostility. Participants rate items on a 1 (extremely uncharacteristic of me) to 5 (extremely characteristic of me) scale, with higher scores reflecting increased propensity for aggression. The current study utilized the 8-item hostility subscale (e.g., “I am suspicious of overly friendly strangers”). The Cronbach’s alpha for the hostility subscale is good (α = .77) (Buss & Perry, 1992), which is consistent with the current sample (α = .82).

The Reactive–Proactive Questionnaire (RPQ; Raine et al., 2006)

The RPQ is a 23-item self-report instrument which measures reactive aggression (11 items, e.g., “How often have you…Reacted angrily when provoked by others?”) and proactive aggression (12 items, e.g., “How often have you…Used force to obtain money or things from others?”). Participants rate items on a 3-point scale [0 (never), 1 (sometimes), or 2 (often)], with higher scores indicating greater frequency of aggression. The internal reliability coefficients of the RPQ’s total score, reactive subscale, and proactive subscale have been strong 0.90, 0.81, and 0.84, respectively (Raine et al., 2006), which was consistent with the present sample (all α’s > .86).

The Investment Model Scale (IMS; Rusbult, Martz, & Agnew, 1998)

The IMS is a 40-item self-report instrument that measures relationship commitment at both facet- and global-levels of three observed indicators: relationship satisfaction, quality of alternatives, and investment size. Participants rate items on a 0 (do not agree at all) to 8 (completely agree) scale, with higher scores indicative of healthier and happier relationships. This investigation utilized the 5-item global satisfaction level as a measure of relationship satisfaction (e.g., “I feel satisfied with our relationship”). Higher scores indicated more relationship satisfaction. This subscale has demonstrated alphas above .80 (Rusbult, Martz, & Agnew, 1998), which was consistent with the present study (α = .90).

Trait Anger Scale (TAS; Spielberger, 1988)

The TAS is a 10-item, self-report, Likert-type (1 = almost never to 4 = almost always) scale on which participants report how angry they generally feel. Higher scores indicate the tendency to experience anger more frequently, with greater intensity, and for longer periods of time. Sample items include “I am quick tempered” and “When I get mad, I say nasty things.” Internal consistency reliabilities range from .81 to .91 (Spielberger, 1988), which is consistent with the current study (α = .86).

The UPPS-P (UPPS-P; Lynam, Smith, Whiteside, & Cyders, 2006; Whiteside & Lynam, 2001)

The UPPS-P is a 59-item self-report measure of five impulsivity-related traits including positive urgency, negative urgency, lack of premeditation, lack of perseverance, and sensation seeking. Participants rate items on a 1 (strongly disagree) to 5 (strongly agree) scale, with higher scores indicating greater impulsivity. This study utilized the 12 item negative urgency subscale, or the tendency to act rashly in response to negative affect (e.g., “In the heat of an argument, I will often say things that I later regret.”). Consistent with prior research (α = .89, Cyders, 2013) the negative urgency subscale demonstrated good reliability (α = .84).

The Alcohol Use Disorder Identification Test (AUDIT; Babor, Biddle-Higgins, Saunders, & Monteiro, 2001)

The AUDIT is a 10-item Likert-type scale that assesses hazardous patterns of alcohol consumption. Participants rate items on a 0 to 4 scale, with higher scores indicative of greater problematic drinking. The AUDIT has demonstrated high internal consistency across samples (Babor et al., 2001) with an alpha of .81 in the current sample.

Procedures

Upon arrival to Session 1, each member of the couple was separated into private testing rooms. After providing informed consent, participants completed a battery of questionnaires on a computer using MediaLab version 2014 software (Jarvis, 2014) which included other measures not pertinent to the current analysis. The experimenter provided instructions on how to operate the computer program and answered any questions during the session. After completion of the questionnaire battery, participants’ eligibility for the second session was determined. Eligible participants were scheduled for Session 2 on a subsequent day, whereas ineligible participants were debriefed, compensated, and released.

Results

Mean, standard deviation, and range for all study variables are reported in Table 1.

Table 1.

Means, standard deviations, ranges, skewness, kurtosis, and mode for study variables (N =680)

Correlates	M	SD	Range	Skew	Kurtosis	Mode
RPQ proactive PIM	0.084	0.194	0–2	4.287	25.000	0.0
RPQ reactive PIM	0.441	0.354	0–2	1.207	2.020	3.0
CTS-2 physical assault total score	1.987	6.651	0–72	6.551	52.774	0.0
CTS-2 psychological aggression total score	12.636	17.349	0–48	2.518	8.702	0.0
Relationship satisfaction PIM	6.709	1.373	0–8	−1.578	2.448	8.0
Impulsivity: Negative urgency PIM	1.964	0.606	1–4	0.514	−0.334	1.33
Trait anger PIM	1.622	0.521	1–4	1.356	2.483	1.20
Problematic drinking total score	7.350	4.818	0–28	1.809	4.158	5.0
Hostility PIM	1.975	0.753	1–5	0.718	−0.006	8.0

Note: PIM = per item mean (not overall mean; applicable to M, SD, and Range)

Aim 1: Confirming the factor structure of the FFIPA

In order to validate the factor structure of the FFIPA (see Figure 1), a confirmatory factor analysis (CFA) was conducted in Mplus (v.8.3) (Muthén & Muthén, 1998–2019) according to Little et al.’s (2003) design. This decision was grounded in a theoretical framework (Little et al., 2003) that required the utilization of a CFA to impose theoretical constraints on the factor structure of the FFIPA (e.g., parsing unique variance identifying motivations) that a data-driven (e.g., parallel/exploratory factor) analysis would not have been able to distinguish given the original FFAM was not a factor-analytically constructed measure. Model fit was evaluated with both goodness of fit and practical fit indices. Given that the chi-square goodness of fit test is overly sensitive in large samples and increases the risk of Type 1 error (Kline, 2016), additional practical fit indices were evaluated, including the Comparative Fit Index (CFI) (Bentler, 1990). A CFI ≥ .90 is considered adequate model fit by convention with values above .95 demonstrating a good fit to the data. Additionally, Root Mean Square Error of Approximation (RMSEA) (Steiger & Lind, 1980) was evaluated for which values ≤ .08 are acceptable and values ≤ .05 are preferred and indicate the model closely approximates the data. It should be noted that the thresholds for these model fit indices’ are not stringent cut-offs and should be examined in relation to the sample size, complexity of the model, and relative loadings/types of factors and indicators (Marsh, Hau, & Wen, 2004; Marsh, Hau, & Grayson, 2005).

Two items on the FFIPA demonstrated very low score distributions (Item 15 M = 1.012, SD = .154, Item 24 M = 1.014, SD = .149). To compute a more parsimonious model that reduced issues with multicollinearity and the likelihood of Type I and II error (Little, Cunningham, Shahar, & Widaman, 2002), item parcels were utilized (i.e., statistical mean of two individual items). Parcels were constructed using an item-to-construct balancing approach based on preliminary item-scale loadings. This resulted in three parcels per factor, and parcels were used as manifest indicators for the final model. The reported final model was constructed using maximum likelihood estimation with robust standard errors (MLR) utilizing standard SEM (Kline, 2016) in order to account for the nonnormality of the data. Missing data were handled using Full Information Maximum Likelihood (FIML; Arbuckle, 1996; Enders, 2001). After computing the final model, loadings and residuals were examined and were found to be uniformly significant, suggesting the model was appropriate for the data. Lastly, examining the modification indices indicated that no further modifications to the model would improve the fit while maintaining the proposed factor structure. Therefore, the reported final model approximated the data well.

Because couples consisted of heterosexual individuals distinguishable by gender, separate, equivalent models were preliminarily estimated and compared for men and women. A traditional multi-group CFA between genders to assess measurement invariance was not tenable due to statistical limitations resultant from our sample size and model complexity. Results indicated a similar factor structure for both men (χ² = 264.845, df = 130, p < 0.01; RMSEA = .056) and women (χ² = 273.175, df = 130, p < 0.01; RMSEA = .057). Therefore, a complex type of analysis (appropriate for clustered survey data) was utilized to compute the CFA for the entire sample (Wu & Kwok, 2012). This type of analysis (together with the MLR estimator) adjusts standard errors and a chi-square test of model fit that takes into account the violation of the independence assumption as may be found in cluster analysis (Muthén & Muthén, 1998–2019). A number of a priori restrictions were instituted in order to fit a model mirroring Little et al.’s (2003) design. Latent variables with observed indicators had their scales (variances and residual variances) fixed to 1.0. Correlations between latent variables representing the forms and functions of aggression were fixed at zero. The two forms of aggression were allowed to freely correlate. Initially, the two functions (i.e., proactive and reactive) of IPA were also allowed to freely correlate. However, initial models found a negligible covariance between the two functions, β = .054, p = 0.769. Therefore, this path was restricted to zero in the reported final model, mirroring Little et al.’s (2003) approach. Second order latent factors had their respective loadings on their first order latent factors equated.

The factor structure of the FFIPA was confirmed and mirrors the established factor structure of the original measure, consistent with Aim 1 (see Figure 2). Despite a significant chi-square test (χ² = 304.11, df = 130, p < 0.01), the model produced a RMSEA = .045, 90% CI [.038, .051], and a CFI = .910, which indicated acceptable model fit. Importantly, these indices were detected while taking into account the complexity of the model and dyadic nature of the data in light of the moderate sample size. Alternate models demonstrated significantly inferior fit to the data (utilizing a Satorra-Bentler corrected Δχ2) compared to the hypothesized model when collapsing the forms of IPA, Δχ² = 48.59, p < 0.01., the functions, Δχ² = 10.86, p < 0.01., and the forms and functions, Δχ² = 36.03, p < 0.01.

Figure 2.

Path model of the factor structure of the FFIPA.

Note. Confirmatory factor analysis diagram with standardized estimates of the FFIPA (mean structure not shown). Significant estimates, p ≤ .05, are noted with an *. Estimates fixed at 1.0 in the unstandardized model are noted with a ^. Equated factor loadings in the unstandardized model are noted with an _x.

In the factor structure of the original FFAM, reactive and proactive aggression are uncorrelated, whereas overt and relational IPA are significantly correlated. In order to examine and confirm the correspondent correlations on the FFIPA, the bivariate correlations between the latent factors were computed after the final factor structure of the measure had been established. Consistent with our expectations, the factor structure of the FFIPA replicates that of the FFAM. With the pure forms and pure functions of IPA restricted from correlating, overt IPA significantly and positively correlated with relational IPA, r = .80, p < .01. Reactive and proactive IPA were restricted from correlating based on negligible covariance in initial models, replicating the original FFAM.

Psychometric properties of the FFIPA

Alpha reliabilities for each observed subscale were as follows: overt (α = .74), reactive-overt (α = .80), proactive-overt (α = .78), relational (α = .66), reactive-relational (α = .70), and proactive-relational (α = .78). The subscales demonstrated acceptable internal consistency to be used in SEM techniques (Little, Lindenberger, & Nesselroade, 1999). Test-retest reliability was not evaluated in the current study.

Aim 2: Examining the construct validity of the FFIPA

It was hypothesized that (a) reactive IPA as measured by the FFIPA would be positively associated with reactive, but not proactive, aggression as measured by the RPQ, and (b) proactive IPA as measured by the FFIPA would be positively associated with proactive, but not reactive, aggression as measured by the RPQ. To test these hypotheses, univariate regressions between the constructs were examined after the final factor structure of the FFIPA had been established. The model consisted of the FFIPA construct predicting the RPQ constructs while allowing the RPQ subscales to co-vary. As hypothesized, reactive IPA from the FFIPA was significantly and positively associated with reactive aggression from the RPQ (β = .44, p < .01) but not with proactive aggression from the RPQ (β = .07, p = .51). Likewise, proactive IPA from the FFIPA was significantly and positively associated with proactive aggression from the RPQ (β = .39, p < .01) but not with reactive aggression from the RPQ (β = −0.16, p = .067).

In order to determine if the latent overt and relational IPA factors of the FFIPA are positively associated with the physical assault and psychological aggression subscales of the CTS-2 (as these corollary scales all capture IPA perpetration acts), univariate regressions between these constructs were examined after the final factor structure of the FFIPA had been established. The model consisted of the FFIPA constructs predicting the CTS-2 constructs and allowed the CTS-2 constructs to co-vary. As hypothesized, relational IPA from the FFIPA was significantly and positively associated with psychological IPA perpetration (β = .40, p < .01) and physical IPA perpetration (β = .24, p < .01) from the CTS-2. Overt IPA from the FFIPA was significantly and positively associated with psychological IPA perpetration (β = .59, p < .01), but not with physical IPA perpetration (β = .125, p = .14) from the CTS-2.

In order to test the hypothesis that each pure, latent construct on the FFIPA would be positively associated with the established IPA risk factors of trait anger, hostility, relationship dissatisfaction, impulsivity (negative urgency), and problematic drinking, univariate regressions between these FFIPA constructs were examined after the final factor structure of the FFIPA had been established. As hypothesized, each risk factor was significantly associated with overt and relational IPA on the FFIPA. In addition, trait anger and negative urgency were also significantly and positively associated with reactive IPA. No risk factors were significantly associated with proactive IPA.

Exploratory examination of gender differences

Potential gender differences in the forms and functions of IPA perpetration were also examined using the FFIPA. Each pure, latent FFIPA construct was regressed onto a categorical gender variable (male = 1; female = 0) once the final factor structure of the FFIPA had been established. Significant gender differences were found for overt (β = −.18, p < .01) and relational IPA (β = −.11, p < .01), with women reporting the use of significantly more overt IPA (M = 7.35, SD = 2.09) than men (M = 6.70, SD = 1.41) and relational IPA (M = 7.10, SD = 2.10) than men (M = 6.83, SD = 1.61). In contrast, analyses did not detect significant gender differences for reactive and proactive IPA perpetration.

Discussion

Despite advancements in IPA measurement, no instrument simultaneously captures both the motivations and forms of IPA. The FFIPA is the first self-report questionnaire that addresses this gap. Results support the FFIPA’s validity and factor structure for measuring the forms and functions of IPA in conflict-prone heterosexual couples. The primary strength of the FFIPA is its integrated and parsimonious measurement of both the forms and functions of IPA perpetration. It is also noteworthy that the FFIPA is the first self-report measure that demonstrates the validity of the Forms and Functions of Aggression Model (Little et al., 2003) outside of adolescence.

Internal Validity

The Forms and Functions of Aggression Model (FFAM; Little et al., 2003) is an empirically validated model of aggressive behavior that combines both the forms (i.e. overt, relational) and functions (i.e. reactive, proactive) of aggression. Consistent with Aim 1, the factor structure of the FFIPA replicated the well-established factor structure of the FFAM. Of note, though the pattern of parcel loadings onto factors was fairly equal across the forms and functions of aggression, the second-order, latent factor of pure reactive aggression had relatively weaker, albeit significant, equated loadings on its latent factor indicators of overt reactive and relational reactive aggression (see Figure 2). This suggests that the unique variance attributable to reactive aggression in the model is smaller than that attributable to proactive aggression and suggests that this instrument does not capture the unique construct of reactive aggression as well as it does proactive aggression. In further support of the FFIPA, the pattern of associations among the FFIPA factors was similar to the original FFAM (i.e., overt significantly correlating with relational, proactive uncorrelated with reactive). Results also indicate distinct patterns of association with risk factors for proactive and reactive IPA with reactive IPA uniquely correlating with trait anger and negative urgency whereas no significant associations were found between risk factors and proactive IPA. Additionally, results indicated comparable patterns of association between risk factors and overt and relational IPA (see Table 2). Taken altogether, these findings support the validity of the first self-report questionnaire to simultaneously assess the forms and functions of IPA perpetration.

Table 2.

Unique associations of correlates of IPA with FFIPA constructs (N =680).

Correlates	Forms of IPA		Functions of IPA
	Relational	Overt	Reactive	Proactive
Trait anger	0.469*	0.536*	0.366*	−0.111
Problematic Drinking	0.378*	0.251*	0.003	0.074
Hostility	0.418*	0.399*	0.179	−0.108
Relationship Satisfaction	−0.524*	−0.429*	−0.032	0.141
Impulsivity: Negative Urgency	0.388*	0.480*	0.503*	−0.176

Note. The table values are standardized regression estimates, estimated separately. The two forms of IPA are independent of the two functions of IPA.

^*p < .01.

Construct Validity

In accordance with Aim 2, the present study demonstrates the construct validity of the FFIPA via its correlation with related measures of the forms and functions aggression. As hypothesized, subscales which measure reactive and proactive IPA on the FFIPA were significantly and positively correlated with the corresponding subscales (but not with the non-corresponding subscales) on the RPQ. Associations were examined between the FFIPA’s overt and relational subscales and the physical and psychological IPA subscales from the CTS-2. As expected, all four of the associations were positive (range: β = .125 to .590); however, only the association between overt IPA and physical IPA was not significant (β = .125). As these findings were not wholly expected and are the first to compare the CTS-2 forms of IPA and the overt-relational forms of IPA, replication in various samples is necessary to fully understand these associations.

Evidence for the FFIPA’s construct validity is demonstrated by observed associations between established risk factors (i.e., trait anger, hostility, relationship dissatisfaction, impulsivity (negative urgency), and problematic drinking) for IPA and FFIPA factors. As hypothesized, each risk factor was significantly associated with overt and relational IPA. Furthermore, negative urgency and trait anger were significantly and positively associated with reactive IPA. This finding is consistent with established literature linking difficulties with negative affect regulation with reactive aggression (Marsee & Frick, 2007). In contrast to these findings, no risk factor correlated significantly with proactive IPA. This is likely due to the lack of measures of risk factors for proactive IPA (e.g., callous-unemotional traits, Marsee & Frick, 2007) in the current study. Taken altogether, these results support the use of the FFIPA to parse IPA into its unique forms and functions.

Gender Differences

The FFIPA provides new avenues of research for examining gender differences in aggression. Research utilizing the forms and functions model of aggression (Little et al., 2003) has found that adolescent males report higher levels of overt and relational aggression than females (Kistner et al., 2010; Little et al., 2003). In contrast, analysis of gender differences on the FFIPA indicates that women use significantly more overt and relational aggression than men. These data are consistent with major reviews which indicate that men and women perpetrate comparable levels of IPA (Jose & O’Leary, 2009) but that, dependent on sample and context, women may perpetrate higher levels of IPA than men (Thornton, Graham-Kevan, & Archer, 2016). Additionally, the FFIPA is the first measure that is able to assess gender differences in the functions of IPA. Given that we did not detect gender differences in these motivations, it remains unclear whether men or women’s aggression in conflict-prone relationships is more likely to be proactively or reactively motivated.

Limitations

Several limitations merit discussion. First, the FFIPA captures many types of behaviors that may be categorized as aggressive in their intentions from hitting to social exclusion of partners. However, the FFIPA is not exhaustive in its examination of every form and potential motivation for aggressing. For example, the FFIPA does not capture behaviors of a legal or financial nature, though this may be readily present for intimate partners in conflict (Hines, Douglas, & Berger, 2015; Weaver, Sanders, Campbell, & Schnabel, 2009). Consideration should also be given to other potential functions (e.g., aggression in self-defense) not captured by the current two-function taxonomic system. Second, only one measure of IPA (the CTS-2) was included in order to evaluate construct validity. Assessing associations between the FFIPA and other valid measures of IPA may further demonstrate construct validity. Third, more than half of the FFIPA items operationalize the behavior as occurring “often”, which in turn may decrease responding rates in those individuals who perpetrate IPA infrequently. This may be of particular relevance in future work that uses this measure to assess IPA in low-risk, low-conflict couples. Fourth, future work administering the FFIPA to couples would do well to estimate the model utilizing a multi-level/multi-group framework, especially for assessing measurement invariance. We were not able to use this modeling due to statistical limitations resultant from our sample size and model complexity. This approach would allow the further assessment of measurement invariance between the individual and dyad, genders, etc., further strengthening the validity and potential utility of this new instrument.

Conclusions

The present study supports the FFIPA as the first validated self-report measure that assesses both the forms and functions of IPA perpetration. This novel measure carries significant potential for future research on IPA perpetration and subsequent intervention development. The FFIPA will allow researchers to examine independently the unique functions of IPA separate from the form of behavior used for delivery. In doing so, research will be better positioned to advance the development of complex and multi-determined etiological models of IPA. Though latent modelling is recommended because it provides researchers the greatest insight using the pure forms and functions of aggression, scale scores for each of the six observed subscales may be computed as well (Lee, Penney, Odgers, & Moretti, 2010). There may also be benefit to examining the interactions of the forms and functions in predicting downstream outcomes of IPA (e.g., mental health issues) or other correlates (e.g., severity of violence). The FFIPA is well positioned for future research to consider these important avenues as the four observed subscales crossing form and function naturally provide interaction scales to begin these examinations. For these collective reasons, the FFIPA has the potential to sharpen our understanding of how different types of perpetrators aggress toward their intimate partners.

Public Significance Statement.

This study validated a new instrument that measures why and by what means heterosexual adults are aggressive toward their intimate partners. This instrument will allow researchers and practitioners to understand the function of aggressive behavior separate from its form.

APPENDIX A

FFIPA

Instructions: Below is a list of some things partners do while they are arguing. Please indicate how each statement applies to you on a scale from “1” (Not At All True) to “4” (Completely True).

		Not At All True			Completely True
	Overt
1.	I’m the kind of person who often fights with my partner.	1	2	3	4
2.	I’m the kind of person who hits, kicks, or punches my partner.	1	2	3	4
3.	I’m the kind of person who says mean things to my partner.	1	2	3	4
4.	I’m the kind of person who puts my partner down.	1	2	3	4
5.	I’m the kind of person who threatens my partner.	1	2	3	4
6.	I’m the kind of person who takes things from my partner.	1	2	3	4
	Reactive Overt
7.	When I’m hurt by my partner, I often fight back.	1	2	3	4
8.	When I’m threatened by my partner, I often threaten back.	1	2	3	4
9.	When I’m hurt by my partner, I often get back at him/her by saying means things to him/her.	1	2	3	4
10.	If my partner makes me upset or hurts me, I often put him/her down.	1	2	3	4
11.	If my partner has angered me, I often hit, kick or punch him/her.	1	2	3	4
12.	If my partner makes me mad or upset, I often hurt him/her.	1	2	3	4
	Proactive Overt
13.	I often start fights with my partner to get what I want.	1	2	3	4
14.	I often threaten my partner to get what I want.	1	2	3	4
15.	I often hit, kick, or punch my partner to get what I want.	1	2	3	4
16.	To get what I want, I often put my partner down.	1	2	3	4
17.	To get what I want, I often say mean things to my partner.	1	2	3	4
18.	To get what I want, I often hurt my partner.	1	2	3	4
	Relational
19.	I’m the kind of person who tells my friends to stop liking my partner.	1	2	3	4
20.	I’m the kind of person who tells my partner I won’t be in a relationship with him/her anymore.	1	2	3	4
21.	I’m the kind of person who keeps my partner from being with my group of friends.	1	2	3	4
22.	I’m the kind of person who says mean things about my partner.	1	2	3	4
23.	I’m the kind of person who ignores my partner or stops talking to him/her.	1	2	3	4
24.	I’m the kind of person who gossips or spreads rumors about my partner.	1	2	3	4
	Reactive Relational
25.	If my partner upsets or hurts me, I often tell my friends to stop liking him/her.	1	2	3	4
26.	If my partner has threatened me, I often say mean things about him/her.	1	2	3	4
27.	If my partner has hurt me, I often keep him/her from being with my group of friends.	1	2	3	4
28.	When I am angry at my partner, I often tell him/her I won’t be in a relationship with him/her anymore.	1	2	3	4
29.	When I am upset with my partner, I often ignore or stop talking to him/her.	1	2	3	4
30.	When I am mad at my partner, I often gossip or spread rumors about him/her.	1	2	3	4
	Proactive Relational
31.	I often tell my friends to stop liking my partner to get what I want.	1	2	3	4
32.	I often say mean things about my partner to my friends to get what I want.	1	2	3	4
33.	I often keep my partner from being with my group of friends to get what I want.	1	2	3	4
34.	To get what I want, I often tell my partner I won’t be in a relationship with him/her anymore.	1	2	3	4
35.	To get what I want, I often ignore or stop talking to my partner.	1	2	3	4
36.	To get what I want, I often gossip or spread rumors about my partner.	1	2	3	4