Couple relationship education (CRE) is a promising strategy when offered around the birth of a child, a period commonly associated with a decline in relationship satisfaction and intimacy (e.g., O’Brien & Peyton, 2002; Shapiro, Gottman, & Carrère, 2000). A recent meta-analysis showed that CRE for parents of a newborn can positively impact factors such as psychological well-being, couple communication, and relationship satisfaction (Pinquart & Teubert, 2010). Despite these encouraging findings, the impact of CRE on the population at large is limited by low rates of dropout among low-income and unmarried couples (e.g., Baucom, Chen, Perry, Revolorio, Reina, & Christensen, 2017; Wood, Moore, & Clarkwest, 2011). Thus, to increase the public health signiﬁcance of CRE, prevention scientists need to identify predictors of dropout in the demographic subgroup and develop and test theory-based strategies to increase participation.
Low-Income and Unmarried Couples and Relationship Problems
Low income and unmarried couples are underrepresented in CRE, which has traditionally been geared towards middle-class couples who are committed to each other—either engaged or already married (e.g., Hawkins, Blanchard, Baldwin, & Fawcett, 2008; Stanley, Amato, Johnson, & Markman, 2006). This is problematic because the financial stress associated with low income decreases relationship quality in families through its impact on couples’ interactions and parenting (e.g., Conger, Wallace, Sun, Simmons, McLoyd, & Brody, 2002). Financial stress has been linked to higher rates of conflict (e.g., Hardie & Lucas, 2010) and less warmth (e.g., Ross, O’Neal, Arnold, & Mancini, 2017), and well as to negative communication, not only within a couple (e.g., Cutrona, Russell, Abraham, Gardner, Melby Bryant, & Conger, 2003; Williamson, Karney, & Bradbury, 2013; Wilmarth, Nielsen, & Futris, 2014) but in parent-child relationships (e.g., Neppl, Senia, & Donnellan, 2016). CRE has the potential to equip couples with relationship skills that can mitigate the impact of economic hardship on relationship outcomes (e.g., Karademas & Roussi, 2016; Mitchell, Owen, Adelson, France, Inch, Bergen, & Lindel, 2015).
Low-income couples who raise children outside of marriage—so-called “fragile families” (Reichman, Teitler, Garfinkel, & McLanahan, 2001)—are at a particular high risk for relationship problems and family dissolution. They tend to be highly disadvantaged at the time of their child’s birth—younger, less healthy, less educated than other parents —and more than half split up within five years (McLanahan & Beck, 2010). Yet, at the time of their child’s birth, the majority of low-income couples express a desire to stay together and have positive views towards marriage, making it a promising intervention moment when fathers and mothers are typically highly motivated to work together to improve their relationship (McLanaham, Garfinkel, Mincy, & Donahuefragile, 2010).
In sum, low-income couples and particularly those who have children outside of marriage are at high risk for developing family problems and for family dissolution. CRE has the potential to prevent this negative course. Timing CRE around the birth of a child might reach low-income, unmarried couples when they are still romantically involved and are particularly hopeful about their futures together (e.g., Carlson, McLanahan, England, & Devaney, 2005). CRE that engages couples could help set a course that increases the likelihood not only of staying together, but of being happy doing so.
Implementations of CRE for Low-income and Unmarried Couples
Recognizing the unique challenges faced by low-income, unmarried parents and their potential impact on relationship and child outcomes, a variety of U.S. state and federal initiatives have allocated funding to provide free and low-cost CRE for this demographic (Hawkins & VanDenBerghe, 2014). For example, beginning in 2002, the Administration for Children and Families, a part of the U.S. Department of Health and Human Services, started funding CRE research and demonstration projects as part of their Healthy Marriage Initiative (HMI; National Healthy Marriage Resource Center, 2010) (Dion, 2005, for a review). Most of the demonstration projects were aimed at implementing CRE and did not publish program evaluations.
A recent meta-analysis has shown that CRE has an overall positive impact on low-income couples’ self-reports of relationship quality, communication, and aggression (Hawkins & Erickson, 2015), however, effects were weaker than those found in earlier meta-analyses with predominantly middle-to-higher income, well-educated, married or engaged couples (e.g. Blanchard et al., 2009; Fawcett et al., 2010; Hawkins et al., 2008; 2012; LucierGreer & Adler-Baeder, 2012; McAllister, Duncan, & Hawkins, 2012; Pinquart & Teubert, 2010). Moreover, the Hawkins and Erickson meta-analysis included effectiveness studies with mixed findings. For example, a multisite evaluation of the Building Strong Families Program project, which offered CRE to over 5,000 low-income, unmarried new parents, generally indicated no improvements in couples’ relationship outcomes for couples in the treatment group compared with those in the control group (BSF; Wood, McConnell, Quinn, Clarkwest, & Hsueh, 2010; Wood, Moore, Clarkwest, & Killewald, 2014). These sobering results have led some researchers to question the merit of CRE for low-income couples, at least in the manner that such programs are typically delivered (e.g. Johnson, 2013; Johnson, 2014; for a review see Cowan & Cowan, 2014).
Dropout in CRE for Low-Income Couples, Unmarried Couples
The null results in the Building Strong Families study might be partly due to the high dropout rate, which has been a key challenge in providing CRE for low-income, unmarried couples in community settings. In the Building Strong Families study for example, couples completed on average half of the planned sessions (Wood, Moore, Clarkwest, & Killewald, & Monahan, 2012). When the 12 BSF sites were analyzed separately, however, the site with the most favorable effects on relationship quality (Oklahoma City) was also the site with the lowest dropout rate (Devaney & Dion, 2010; Dion et al., 2008). Taken together, the results from the Building Strong Families study suggest that CRE engagement is typically poor in low-income, unmarried couples in real-life settings, consistent with the findings of similar programs (see review by Baucom, Chen, Perry, Revolorio, Reina, & Christensen, 2017); yet, when engagement is better, as at the Oklahoma City site, such programs can have positive effects.
Dropout is important for a number of reasons. Primarily because no intervention can be effective without sufficient “dosage”. Although it is not clear what dosage of CRE is needed to achieve meaningful effects, participants attending more sessions of a program have been found to experience more improvement (e.g., Barton, et al., 2015). From the standpoint of a program provider, steady and predictable attendance by couples is essential to ensure efficient use of program staff and other resources. From a research perspective, dropout constitutes a confounding factor when determining the effectiveness of a program. However, session attendance in CRE for low-socioeconomic status (SES) couples is not well understood and few studies report detailed information on session attendance (see Baucom, et al., 2017 for a review).
A Framework for Predictors of Dropout in CRE: The Health Belief Model
To understand how risk factors might predict enrollment and retention, we draw from a theoretical model that explain the circumstances under which people do or do not engage in preventive health behaviors—the health belief model (Hochbaum, 1958). The health belief model has been successfully used as a framework to predict health-related behaviors, including session attendance in both CRE and parenting education programs (e.g. Spoth, Redmond, & Shin, 2000; Sullivan, Pasch, Cornelius, & Cirigliano, 2004; Thornton and Calam 2011; Blair & Córdova, 2009; Winslow, Bonds, Wolchik, Sandler, & Braver, 2009). According to the health belief model, couples will attend more sessions of CRE if they (a) believe that they are susceptible to a potential problem (perceived threat); (b) perceive few barriers to taking the preventative action (perceived barriers); (c) believe that the preventative action will be effective in minimizing the risk (perceived benefits); and (d) are confident in their ability to effect change (perceived self-efficacy). Moreover, this model postulates that couples’ perception of threat, benefits, barriers, and self-efficacy can be influenced by modifying variables such as personal or family characteristics. We used the health belief model as a theoretical guide to review previous findings and select potential predictors of session attendance. Findings on predictors within the health belief model are reviewed below.
Socioeconomic Status. Low socioeconomic status is associated with a variety of destabilizing and disruptive factors that may increase barriers to session attendance. Perhaps as a result, session attendance in CRE and parenting education has often been lower in individuals with lower incomes (e.g., Chacko et al., 2016; Hawkins & Erickson, 2015; Petch, Halford, Creedy, & Gamble, 2012; Wood et al., 2011). Moreover, couples attended fewer CRE group sessions findings when the male partner is unemployed, (Dion et al., 2010). Individuals with lower education are also more likely to drop out of CRE and parenting education (e.g., Feinberg & Kan, 2008; Kavanagh & Bennet-Levy, 2010; Spoth, Goldberg, & Redmond, 1999; Wood et al., 2011). However, researchers have not always found this association of low income and education with dropout (e.g., Barton et al., 2015; Brown et al., 2012).
Demographic characteristics. Socioeconomic risk factors are disproportionately represented in African Americans (e.g., Williams, Priest, & Anderson, 2016) and might partly explain why African Americans have been found at a higher risk for dropout (e.g., Devaney & Dion, 2010; Wood et al., 2010). In addition, ethnic minority individuals experience more mistrust of providers and fears of stigmatization (Keller and McDade 2000), which might further increase barriers to session attendance. Perhaps by association with lower socioeconomic status, young age has also been linked to dropout in CRE and parenting interventions (e.g., Dion et al., 2010; Peters, Calam, & Harrington, 2005; Wagner, Spiker, Linn, & Hernandez, 2003; Korfmacher, Green, Staerkel, Peterson, Cook, & Roggman, 2008). Couples living in larger households with many adults, children or both may attend fewer CRE sessions because they may have demands competing for their attention or experience time and schedule constraints. Supporting this, couples with more children have been found to perceive higher barriers to PE, which in turn was linked to lower levels of enrollment (Spoth, Redmond, & Shin, 2000). The birth of a son, relative to a daughter, has been found to increase parental investment in the family, such as higher monetary investments in housing and goods, more time spent with the child by fathers, an increase in marital stability and relationship satisfaction, and higher odds of marriage in unmarried mothers (Lundberg, 2005). Perhaps as a result of increase parental investment in the family, mothers of girls were more likely to drop out (Pillhofer et al., 2016).
Relationship stability. Couples with low relationship stability may perceive fewer benefits to participating in a program designed to strengthen their relationship. An indicator of relationship stability—marriage—has been linked to high session attendance (e.g., Brown, Feinberg, & Kan, 2012; Feinberg & Kan, 2008). Among unmarried couples, other indicators of relationship stability have been associated with higher session attendance, such as living together instead of in living in separate households (Wood et al., 2011) and reporting a higher value for the institution of marriage (e.g., “Being married is one of the one or two most important things in life”) and for relationship stability (e.g., “How often have you thought your relationship might be in trouble”) (Busby et al. 2015). Further, commitment to the relationship in general has also been linked to higher session attendance (e.g., Dion et al., 2010; Barton, Beach, Hurt, Fincham, Stanley, Kogan, & Brody, 2015; Wood et al., 2011).
Family functioning. The health belief model states that people will act to prevent a negative health outcome if it is likely to afflict them (Rosenstock, 1966). Therefore, relationship and parenting problems might increase session attendance. Supporting this premise, low-income parents of young children who rated their parenting self-efficacy as lower attended more parenting education sessions (Garvey, Julion, Fogg, Kratovil, & Gross, 2006). Furthermore, couples in which the male partner reported a higher quality of relationship interaction (e.g. conflict management and expression of affection) were less likely to attend CRE sessions (Dion, et al., 2010). Findings concerning severe forms of relationship problems, such as intimate partner violence, are limited and inconsistent, with one study showing a trend of low-level violence predicting service completion in home-based child maltreatment prevention services (Damashek, Doughty, Ware, & Silovsky, 2010), whereas another finding no effect (Halford et al., 2006, Petch et al., 2012).
Personal functioning. Low levels of personal functioning—as indicated by symptoms of depression, a mental health diagnosis, interpersonal problems within and outside the family or poor time structure—could decrease session attendance through two pathways: (a) by increasing perceived barriers and (b) by decreasing perceived self-efficacy. People with depressive symptoms typically perceive more barriers to interacting with mental health professionals (Anderson et al. 2006; Kazdin, Holland, Crowley, & Breton, 1997). They also have lower self-efficacy (Bandura, 1997). Perhaps as a result, dropout in CRE and parenting education was higher among individuals experiencing symptoms of depression (e.g., Barton, et al., 2015) and among participants with more general symptoms of psychological distress such as restlessness, nervousness, and sadness (e.g., Dion et al., 2010). General distress with symptoms that include interpersonal problems within and outside the family, child-related rigidity, and unhappiness are further stressors that might interfere with program attendance. However, several studies did not find an association between psychological distress and dropout (e.g., Brown, et al., 2011; Petch et al., 2012). Another indicator of personal functioning is the mean the degree to which a person structures their time (Bond & Feather, 1988) which might be an indicator of self-efficacy and, on a more practical level, is required to attend CRE sessions and complete homework assignments.
Delivery complications and infant health. Pregnancy and delivery complications have been found to increase stress in parents (e.g., Janis, Callahan, Shelton, & Aubuchon-Endsley, 2016), which has been identified as a predictor of early intervention termination, and fewer intervention sessions attended in child mental health services among low income families (McKay, Pennington, & McCadam, 2001). Parental stress might also be increased when a parent has a fussy, hard to soothe infant, a situation that is more likely, for example, following a stressful delivery (Weerth & Buitelaar, 2007). Perhaps through the pathway of increased stress, mothers of infants with low birth weight are more likely to drop out of PE (Pillhofer, et al., 2015).
Preventive health care utilization. Prior health-related preventive behavior, like attending health appointments, could predict future health-related preventive behavior. Individuals who have shown health-related preventive behavior before might have a favorable combination of individual characteristics and environmental factors that increase perceived threats of health problems, reduce barriers, increase benefits, and/or increase self-efficacy and thereby makes health-related preventive behavior more likely. However, the association between general preventive health care utilization and session attendance in CRE has not been directly examined.
Participant in-session behavior. Behaviors that suggest participants in-session engagement (e.g., engaging in discussions) and alliance with the CRE coach (e.g., feeling comfortable with the coach) during the first session might predict further session attendance. In-session engagement could be a sign that participants perceive the topics covered in the program to be relevant to their needs, that there are few program-related barriers such as a misfit between participants reading skills and the reading-level of the practices. Participants who actively engage during sessions might do so because they believe that the training is helpful and because they trust in their ability to effect change in their relationship and parenting skills through participating in the program. Studies in the field of psychotherapy show that patients with stronger therapeutic alliance are less likely to drop out (e.g., Sharf, Primavera, & Diener, 2010), which might be true for CRE as well.
Overall, the research on CRE for low-income couples is still in its early stages. The literature review revealed a paucity of empirical literature on factors related to dropout. The range of factors examined in relation to dropout is generally limited in scope and variety and often focuses on variables of convenience. Furthermore, most of the predictors of dropout were studied in samples of married couples with higher socioeconomic status, and not in the couples that seem to be at a particularly high risk for dropout: Low-income and unmarried couples. Similarly, previous studies have failed to separately study characteristics of couples who dropout early versus late.
The Present Study
We explore predictors of dropout in a flexible delivery, home-based version of CRE, Couples Care for Parents, with a blended format of self-directed learning, telephone, and face-to-face contact in a sample of low-income, unmarried parents of a newborn. The study was funded by a demonstration grant from the Administration for Children and Families to examine the feasibility of implementation of this established program in a different sociodemographic subpopulation. We hypothesized that total session attendance would be lower and early dropout (i.e. dropout after the first session) would be higher in couples with low levels of sociodemographic risk (i.e., low income, unemployment, low education; H1), certain demographic characteristics (i.e., African American ethnicity, younger age, female infant, living in a household with more children, living with additional adults; H2), lower relationship stability (i.e., lower relationship commitment, non-cohabitation, fewer steps undertaken towards getting married; H3), higher family functioning (i.e., higher levels of parenting satisfaction, parenting efficacy, intimate relationship satisfaction, intimate relationship satisfaction, dysfunctional non-violent conflict resolution strategies, and lower levels of intimate partner psychological aggression and physical aggression; H4), poorer personal functioning (i.e., depression, psychiatric disorder, psychological distress, low time management skills; H5), poorer infant health and delivery complications (i.e., infant complications, APGAR scores, delivery complications; H6), lower levels of preventive care utilization (i.e., prenatal, prenatal class attendance, regular dental care utilization; H7), and lower levels of in-session engagement and alliance with the program coach during the first session (H8).
Recruitment took place at maternity wards in two hospitals in Suffolk County, NY. We screened a total of 11,014 couples for eligibility to participate one of two studies. To meet inclusion criteria for the current study, couples had to be (1) in a relationship with the other parent of the newborn, (2) unmarried, (3) be fluent in English, and (4) of low-income status. Low-income status was defined as falling below the self-sufficiency standard for Suffolk County in 2010, which was $4,066/month for two adults and one infant (Fiscal Policy Institute, 2010). In total, 1810 couple met the inclusion criteria for the current study and 708 couples signed up for intervention offered within the current study. Data analyses focused on the n = 467 couples who had (a) completed at least one session of the interventions (n = 548) and (b) had given birth at one of the two hospitals. These limits were imposed because we aimed to study dropout in couples who had started participating the intervention and because we were not given permission to access patient charts at one of the two hospitals. Sample characteristics are provided in Table 1.
After eligibility screening at the maternity wards, couples who met inclusion criteria and were interested in participating in the CRE program received two identical questionnaire packets (i.e. pre-intervention assessment packet), one for each partner. Couples were asked to complete the pre-intervention assessment packets and return them either by mail or in-person at the first session of the intervention. When study team members called to schedule the first intervention session, they reminded both members of a couple to complete and return their pre-intervention assessment packet. Participants were asked to complete a total of four assessments packets over the course of the study, as well as an additional protocol that is not of focus of this study. In total, participants were paid up to $175 per person for completing assessments, $15 of which was for the completion of the pre-intervention assessment packets that are of present focus. Participants were not paid for attending sessions of the intervention. The intervention was free for all participants. During the first session of the intervention, women participants were asked for consent for the research team to access their hospital charts. The coaches of the intervention rated participants in-session behavior after every session. Data analyzed in this study came from (a) the eligibility screening interview at the maternity ward, (b) the pre-assessment packet, (c) hospital charts, (d) coaches’ session ratings.
Couples Care for Parents is an evidence-based, dyad-focused program of couple relationship and parenting education designed to meet the needs of parents of newborns (Halford, Petch, & Creedy, 2010). The program includes the following topics: communication, conflict management, expectations about household chores, parenting, and infant care; infant care knowledge and skills; mutual partner support; stress management; balancing couple, individual, and family time; and caring, affection, and intimacy. For the present study, the program was adapted for U.S. participants and extended from six to seven sessions. The videos were re-shot with couples from the U.S., the language and examples in the workbook were modified to be more appropriate for U.S. couples, and some of the conflict resolution skill-building activities were specifically tailored to prevent intimate partner violence (see Halford, Heyman, Slep, Petch, & Creedy, 2009).
All sessions took place in the couples’ own homes. Session 1 and 4 were completed during home visits, the other sessions during telephone calls. Sessions were scheduled 1-3 weeks apart, with closer spacing at the beginning of the program. Home visits were approximately 1 hour, and phone sessions were 30-60 minutes. Each session included a watching a video, typically prior to the session. Video segments are 5-7-min long, shot in documentary style, and include a small didactic portion and demonstrations of the skills targeted in that session (e.g., communication, playing with a young baby, asking for support). After watching the video, participants complete individual and/or couple activities from their Couple CARE for Parents workbooks.
All telephone sessions and home visits were led by trained coaches who had either completed graduate training in social work or were currently enrolled in a clinical psychology doctoral program. The role of the coach was to (a) clarify any concepts with which the couple may be struggling with, (b) help the couple to identify and implement self-change objectives, and (c) act as a source of support and knowledge. The scheduling of sessions was flexible and allowed couples to schedule sessions on evenings and weekends. Project staff devoted significant attention to coordinating and rescheduling sessions with couples. The efficacy of earlier versions of Couple CARE as both a face-to-face (Halford, Sanders, & Behrens, 2001) and a flexibly-delivered intervention using only video and telephone support (Halford, Moore et al., 2004) has been demonstrated in randomized controlled trials.
Total session attendance and early dropout. The two outcome variables were (a) a count of the total number of sessions a couple had attended ranging from 1 (one session) to 7 (seven sessions) and (b) a dichotomous variable of early dropout status coded as 0 (attended more than one sessions) and 1 (attended exactly one session).
Socioeconomic status. Indicators of socioeconomic status assessed in the study included monthly household income in Dollars (females’ household if the couple was not cohabiting), unemployment status scored as 0 (not unemployed) and 1 (unemployed), and highest educational attainment ranging from 1 (did not graduate high school) to 6 (professional or doctoral degree).
Demographic characteristics. Participants responded to the question “What is your ethnic background?” by selecting one or more of seven response categories (African American, Caribbean American, Asian, Latino/Hispanic, Native American, Pacific Islander, White), their responses subsequently coded into seven separate variables (e.g., 0 = not African American, 1 = African American). Participants further reported their age (in years), the gender of their newborn with the categories 0 (male) and 1 (female), the number of children in the household (females’ household if the couple was not cohabiting), and additional adults in the household, apart from the couple scored as 0 (no additional adults) and 1 (additional adults) (females’ household if the couple was not cohabiting).
Relationship stability. Participants completed three measures of relationship stability that were part of the pre-assessment packet.
Commitment. Participants rated their commitment on the Broderick Commitment Scale (BCS; Beach & Broderick, 1983). The scale consists of a single item (i.e. “Please select the range of numbers from the scale below which corresponds to your commitment to your relationship”) that is rated on a 10-point scale ranging from 1 (1-10) to 10 (91-100) with higher scores indicating higher commitment. Commitment is defined as standing by one’s partner in times of adversity and accepting the other partners’ faults and/or problems. The BCS has been shown to reliably distinguishes between distressed and non-distressed couples, is correlated with other measures of commitment (Broderick & O’Leary, 1986), and predictive of women’s gains in marital therapy (Beach & Broderick, 1983).
Cohabitation.Participants indicated whether they lived in separate (0 = no cohabitation) or in the same households (1 = cohabitation).
Steps Toward Marriage. Participants completed the Steps Toward Marriage scale (STM). The STM is a study specific scale that was modelled after the Marital Status Inventory (Weiss & Cerreto, 1980) and measures the steps that unmarried couples have undertaken toward getting married. Participants responded to seven dichotomous items rated as 0 (no) and 1 (yes) that indicate. Each item was ranked based on its “difficulty” score from a Rasch model analysis, available upon request from the authors, resulting in the following order: 1 = I have thought about having a future together with my partner, 2 = We have talked about having a future together, 3 = We have seriously discussed getting married, 4 = We are engaged but have not set an exact date, 5 = We tell people that we are married but we are NOT legally married, 6 We are engaged and have set a date, 7 = We are legally married. Subsequently, using the highest ranked item per couple each couple was assigned a score on the single-item 7-point STM scale which higher scores reflecting progressively more steps toward marriage.
Family Functioning. Participants completed subscales of different measures of family functioning as part of the pre-assessment packet.
Parenting satisfaction and efficacy. Participants’ parenting satisfaction and perceived parenting efficacy was assessed with the Parenting Sense of Competence Scale (PSOC; Mash & Johnston, 1983). The 9-item Satisfaction subscale measures parenting frustration, anxiety, and motivation (e.g. “My talents and interest are in other areas, not in being a parent.”). The 8-item Efficacy subscale measures competence, problem-solving ability, and capability in the parenting role (e.g. “I honestly believe I have all the skills necessary to be a good parent.”). Each item has a 6-point response scale ranging from 1 (strongly disagree) to 6 (strongly agree), with higher scores reflecting more satisfaction or more efficacy. The measures two-factor structure has been replicated and associations with constructs in the nomological net of Satisfaction and Efficacy such as parenting behavior, parent wellbeing, and parental perceptions of child behavior have been reported (Johnston & Mash, 1989). In this study, Cronbach’s alphas for the Satisfaction subscale were α = .73 for men and α = .69 for women and for the Efficacy subscale α = .70 for men and α = .77 for women.
Relationship satisfaction. We measured relationship satisfaction with the Quality of Marriage Index (QMI; Norton, 1983), substituting references to “marriage” with “relationship.” The QMI comprises six broadly worded statements about relationship quality (e.g. “We have a good relationship”) which each participant rates on a 7-point scale ranging from 1 (very strong disagreement) to 7 (very strong agreement). The QMI has excellent convergent and discriminant validity (Heyman, Sayers, & Bellack, 1994) and excellent internal consistency for males and females in the present study (α = .95 for men and α = .95 for women).
Non-violent conflict resolution strategies. Non-violent strategies for resolving conflicts were assessed with three subscales from the Conflicts and Problem-Solving Scales (CPS; Kerig, 1996): Collaboration (8 items; e.g., “I compromise, meet partner half way, ‘split the difference”), Stalemate (7 items; e.g., “I complain, bicker without really getting anywhere”), and Avoidance-Capitulation (8 items; e.g. “I leave the room”). Items were rated on a 4-point scale ranging from 0 = (never) to 3 (often). High scores on Collaboration and low scores on Stalemate and Avoidance-Capitulation indicate better conflict solving strategies. The CPS has good test-retest reliability and convergent validity (Kerig, 1996) and, in the current study, acceptable internal consistency for Collaboration (α = .72 for men and α = .76 for women), Stalemate (α = .61 for men and α = .73 for women), and Avoidance-Capitulation (α = .73 for men and α = .70 for women).
Intimate partner psychological and physical aggression. Participants completed the Psychological Aggression (8 items; e.g., “called partner fat or ugly”) and Physical Aggression (12 items; e.g., “twisted arm”) subscales of the Revised Conflict Tactics Scale (CTS2; Straus, Hamby, Boney-McCoy, & Sugarman, 1996). The CTS2 has been widely used in studies of intimate partner violence (IPV; e.g., Nixon, Resick, & Nishith, 2004; Sivak, Taft, Goodman, & Dutton, 2013), and is associated with several factors in the nomological network of couple aggression (O’Leary, Slep & O’Leary, 2007). Each partner reported victimization and perpetration frequencies for different acts of aggression in the past 12 months on a 7-point scale ranging from 0 (this has never happened) to 6 (happened more than 20 times in the past year). When a respondent’s victimization score was different from her or his partner’s perpetration score of the same act, we used the higher of the two values to yield dual-informant total psychological and physical aggression scores for each person. The resulting total scores are indexes made of items describing different acts of aggression (e.g., “punching” or “hitting with an object”) that may not be present in the same couple (e.g., Shortt, Capaldi, Kim, & Owen, 2006) but nevertheless represent the same construct. Unlike items of scales, there is no assumption of item interrelatedness for items of indexes, reporting Cronbach’s alpha as a measure of internal consistency is therefore not recommended (Streiner, 2003).
Personal functioning. Participants completed three measures of personal functioning in the pre-assessment packet.
Psychological distress. Participants psychological distress was assessed using a 70-item version of the Child Abuse Potential Inventory (CAPI; Milner, 1986; Milner 2004) is a standardized and widely used self-report measure of personality traits, beliefs, and attitudes that differentiate parents who physically abuse children from those who do not. We used a 70-item version of the Physical Abuse scale that contains the subscales Rigidity (“A child should never talk back.”), Distress (“I often feel very upset.”), Unhappiness (“My life is happy.”, reverse scored), Problems with family (“My family fights a lot.”), and Problems with others (“People have caused me a lot of pain.”). Items are rated on a dichotomous scale of 0 (disagree) and 1 (agree). The average of all 70 items forms the Child Abuse Potential score that was used for analysis in the present study, with higher scores indicating higher abuse potential. Individuals with elevated CAPI scores are more likely to have experienced child abuse, poor social support, high physiological reactivity to stressful stimuli, and low self-esteem (Milner, Gold, Ayoub, & Jacewitz, 1984). This makes the CAPI score a good measure of general psychological distress. The CAPI abuse scale has good validity (Milner, 1994; Milner, et al., 1984; Robertson & Milner, 1983) and test-retest reliability (Merrill, Crouch, Thomsen, & Guimond, 2004). Internal consistency in the present study was high (α = .93 for men and α = .94 for women).
Depression. Depressive symptoms were measured using the Edinburgh Depression Scale (EDS; Cox, Holden, & Sagovsky, 1987), a 10-item self-report measure developed to screen for depression in postpartum women. It covers common symptoms of depression but excludes somatic aspects such as fatigue and appetite which can be associated with childbirth. It has been validated in men (Edmondson, Psychogiou, Vlachos, Netsi, & Ramchandani, 2010), and has good concurrent validity, sensitivity, specificity, and positive predictive value with a clinical diagnosis of major depression as the criterion (Eberhard-Gran, Eskild, Tambs, Opjordsmoen, & Samuelsen, 2001). Each item concerns symptoms experienced in the past week (e.g., “I have felt sad or miserable”) and is scored from 0 (yes, most of the time) to 4 (no, not at all). Item averages were calculated (α = .93 for men and α = .90 for women) with higher scores indicating more depressive symptoms.
Time structure.TheTime Structure Questionnaire (TSQ; Bond & Feather, 1983) measures the degree to which individuals perceive their use of time as structured (e.g., “Do you have a daily routine which you follow?”) and purposeful (e.g., “Could you tell how many useful hours you put in last week?”). The 26 items have a response scale ranging from 1 (never) to 7 (always). Supporting its concurrent validity, higher TSQ scores were found to correlate positively with constructs such as a sense of purpose in life, self-esteem, health, and optimism about the future, and negatively with depression, psychological distress, anxiety, neuroticism, symptoms and hopelessness (Bond, & Feather, 1988) An item average was calculated (α = .70 for both men and women), with higher scores reflecting a more structured time use.
Psychiatric diagnosis. Psychiatric disorder (e.g., depression and panic attacks) was read from the mothers’ hospital charts and coded as 0 (no) and 1 (yes).
Infant health and delivery complications. Three indicators of infant health and delivery complications were read from mothers’ hospital charts.
Infant medical complications.Infant medical complication(e.g., abnormality in fetal heart rhythm and neonatal jaundice)was a dichotomous variable that was read from the females’ hospital chart and scored as 0 (no) and 1 (yes).
APGAR scores. APGAR scores (Apgar, 1953) measured at one and five minutes after birth were also read from the hospital charts. The APGAR score measures infants Appearance (skin color), Pulse (heart rate), Grimace response (reflexes), Activity (muscle tone), and Respiration (breathing rate and effort) on scales ranging from 0 = (absent) to 2 (normal) with 1 indicating problems in the respective area (e.g. Activity: 1 = “Arms and legs flexed with little movement”, Respiration:1 =“Slow or irregular breathing”). Doctors, midwives or nurses combine these five factors into the Apgar score, which ranges between 0 and 10, with 10 being the healthiest score possible.
Delivery complication. Delivery complication (e.g., cord entanglement, Pre-eclampsia, and malposition of fetus) was read from the females’ hospital chart and scored as scored as 0 (no) and 1 (yes).
Preventive health care utilization. Participants previous use of preventive health care was assessed in two areas of preventive care: Prenatal and dental health care. The items were part of the pre-assessment packet.
Inadequate prenatal care. Inadequate prenatal care was read from mothers’ hospital charts and scored as scored as 0 (no) and 1 (yes).
Prenatal class attendance. Prenatal class attendance during previous or current pregnancy was assessed via participant self-report and scores as 0 (no) and 1 (yes).
Dental care. Dental care was averaged across a self- and a partner-report item (i.e. “How often do you visit a dentist for routine dental care?”, “How often does your partner visit a dentist for routine dental care?”). Items were rated on a 4-point-scale from 1 (twice per year or more) to 4 (less than once every two years).
Participant in-session behavior. After the first session, Couple CARE for Parents coaches rated participants’ in-session behavior on three study specific, rationally derived indexes. In the present study we analyzed two of the indexes: Engagement and Alliance. Cronbach’s alpha for the indexes are not reported (Streiner, 2003).
Engagement. The Engagement subscale consisted of a combination of three dichotomous items scores as 0 (no) and 1 (yes) (e.g., “participant’s self-change plan for this week is behaviorally specific.”) and one item with a 5-point scale ranging from 1 (not at all) to 5 (very much) (i.e., “The participant engaged in discussion throughout the session”). The items were averaged to form a composite index of engagement.
Alliance. The Alliance subscale consists of 10 items (e.g., “I worked together as a team with the female partner.” and “I created an environment that encouraged the female participant to open up.”) with 5-point scales ranging from 1 (not at all) to 5 (very much). All variables were averaged to form a composite index of alliance with higher scores indicating more engagement.
Missing Data. Missing data were prevalent. First, on average, 26% of data from the eligibility screening interview was missing because names and study ID number were not provided for several couples which made it impossible to merge the data with data from the other sources. Second, 29% of couples did not return their pre-assessment packets and several skipped items and whole measures. Third, variables read from the hospital charts had on average 41% of the values missing because the information on the charts was incomplete. Fourth, 55% of the coaches’ ratings for the first session were missing. Data on the two dependent variables (i.e., dropout and number of sessions) were complete, however. Descriptive statistics and the rate of missing data for all variables in the analysis are presented in Table S1 of the on-line supplement.
Two sets of regression models were estimated, one for the total number of session attended and the second for early dropout. We tested hypotheses H1.1 through H8.1 by regressing total session attendance on several indicators of (1) socioeconomic status, (2) demographic characteristics, (3) relationship stability, (4) family functioning, (5) personal functioning, (6) infant health and delivery complications, (7) preventive health care utilization, and (8) participant in-session behavior. We tested hypotheses H1.2 through H8.2 by regressing early dropout on the same set of predictors. Consistent with our bivariate hypotheses, each predictor was tested in its own model. In the first set of models, linear regression was used to estimate predictor-number of sessions associations. A substantial proportion of the couples attended only one session (n = 109, 23.3%) and a large proportion attended all seven sessions (n = 199, 42.6%), thus the distribution was bimodal and violated the assumptions of most estimators (Figure 1). We chose bias corrected bootstrapped SEs and CIs based on 10,000 replicates because this method has does not have distributional assumptions. In the second set of models, logistic regression was used to estimate predictor-dropout associations (a dichotomous variable). Continuous predictors were standardized by subtracting the mean from each value and dividing by the standard deviation. The regression analyses were conducted in Mplus (version 7.4; Muthén & Muthén, 1998-2012). We used robust full information maximum likelihood estimation (FIML), a direct model-based method for estimating parameters in a data set with missing data (Graham, Hofer, & MacKinnon, 1996; Olinsky, Chen, & Harlow, 2003). To account for the high number of tests in the analysis, we computed Benjamini-Hochberg False Discovery Rate (FDR; Benjamini & Hochberg, 1995) adjusted p-values within each domain of predictors using the function p.adjust in R package ‘stats’ (R Core Team, 2013; v.3.3.1).