Family Foundations

Study 1

Summary

Feinberg et al. (2008), Feinberg, Jones et al. (2014) and several related studies utilized a randomized controlled design in which heterosexual couples expecting their first child were randomly assigned to an intervention (n=89) or to a no-treatment literature control condition (n=80). Measures of co-parenting, parenting and child development were collected over five waves: pretest, posttest (when the child was about 6 months old), three-month follow-up (child was one year old), two-and-a-half-year follow-up (child was three years old), and a final follow-up (child was six to seven years old).

At wave 2 (posttest, when children were about six months), the study (Feinberg & Kan, 2008) reported a significant intervention effect for:

• Fathers' coparental support, parenting-based closeness, and parent-child dysfunctional interaction
• Mothers' coparental support, depressive symptoms, and anxiety
• Father-reported infant soothability
• Child duration of orienting

At wave 3 (six-month follow-up, when children were about one year old), intervention participants showed improved (Feinberg et al., 2009):

• Mothers' coparenting competition, coparenting triangulation, warmth to partner, parenting positivity, coparenting inclusion, and negative communication
• Fathers' coparenting competition, coparenting triangulation, warmth to partner, parenting positivity, coparenting warmth, and parenting negativity
• Observed child self-soothing

At wave 4 (2.5-year follow-up, when children were three years old), intervention participants showed program effects for (Feinberg et al., 2010):

• Parent-reported parental stress, parental efficacy, coparenting quality, parenting overreactivity, parenting laxness, and physical punishment
• Mother-reported child total behavior problems, externalizing problems, aggression, and social competence
• Mother-reported child internalizing problems and attention/hyperactivity (boys only)

At wave 5 (six-year follow-up, when children were 6-7.5 years old), intervention participants showed improvements in the following child outcomes (Feinberg, Jones et al., 2014 and additional analyses submitted to Blueprints):

• Teacher-reported internalizing problems
• Teacher-reported anxious/depressed, attention problems, aggressive behavior, and externalizing problems (boys only)

Pregnancy-related outcomes from mid-program showed a significant intervention effect (Feinberg, Roettger et al., 2014):

• Reduced levels of Caesarian birth

Evaluation Methodology

Design: A randomized controlled design was used to evaluate Family Foundations. Couples were randomly assigned to intervention (n=89) or to no-treatment control conditions (n=80), with the control condition consisting of receiving mailed literature on selecting quality childcare and developmental stages. Data from 4-5 couples were not utilized in analyses because of developmental difficulties, death of one of the parents, or congenital medical problems for the baby, resulting in a sample size of 164-165. Participants were primarily (81%) recruited from childbirth education programs at two hospitals located in small cities. All other couples were recruited from doctors' offices or health centers (8%), by newspaper ads or flyers (7%), by word of mouth (3%), or by unknown means (including radio advertisement, 1%). Presumably, all couples responding to recruitment were enrolled, though no further details on recruitment procedures were provided.

Data were collected on participants five times. The study gathered pretest data (Wave 1) on all 164-165 couples when mothers were pregnant. Posttest data collection (Wave 2) occurred after the intervention couples had completed the postnatal classes, when babies were about 6 months old. Of the eligible enrolled couples, 92% of mothers and 90% of fathers completed wave 2. The study administered a six-month follow-up (Wave 3), when babies were about one year old. For this wave, 93% of mothers and 88% of fathers participated. A two-and-a-half-year follow-up (Wave 4) was conducted when children were about three years old and included 137 families. Wave 5 took place when children were ages 6 to 7.5, or six to seven years after program conclusion. Ninety-eight families provided parent and/or teacher data on child development.

Additional analysis was conducted on a subsample (N=123) of mothers consenting to baseline cortisol measurement and completing posttest data collection (Feinberg, Roettger et al., 2014).

The investigators provided Blueprints with exact numbers of valid data for each outcome by each wave. Overall, they report sample sizes of 123 at the birth outcome assessment, 152 at wave 2, 156 at wave 3, 142 at wave 4, and 98 at wave 5. Among these samples of completers, the measures generally had little missing data. In a few instances, however, missing data among completers reached higher levels: 9.2% (138/152) in wave 2 for father's Infant Behavior Questionnaire Duration of Orienting, 22.4% (121/156) in wave 3 for mother and father's Dyadic Couple Communication, 24.6% (107/142) in wave 4 for Child Social Competence, and 27.6% (71/98) in wave 5 for teacher reports of Child Emotional Problems. The analytical sample for adverse birth outcomes (Feinberg et al., 2014) was 147.

Sample: Participants were 169 heterosexual couples who, at the time of recruitment, were expecting their first child and were living together. All participants were at least 18 years of age. The couples resided in rural areas, towns, and small cities. The majority of couples were married (82%) and the sample consisted mostly of non-Hispanic whites (90-91%), with the remaining participants of African American, Asian, Hispanic, or other ethnic descent. For the most part, participants were well-educated and middle class. The median annual family income was $65,000 and mothers attained an average of 15.06 years of education and fathers attained an average of 14.51 years. Mothers averaged 28 years old, and fathers averaged 30 years.

The subsample used to examine program effects on pregnancy-related outcomes was demographically similar to the overall sample. The subsample was 92.7% white and 85% married, and had a mean educational level of 15.1 years.

Measures: The study used the following mother, father, and relationship outcomes:

• Parental depression, taken from a subset of the Center for Epidemiological Studies Depression Scale, developed and validated by others. Cronbach's alphas were .84 for mothers and .66 for fathers across waves 1 and 2, and .86 for mothers and .83 for fathers at wave 4. This survey was administered to mothers and fathers at waves 1, 2, 3, and 4.
• Parental anxiety, taken from the Taylor Manifest Anxiety Scale, developed and validated by others. At baseline, alphas were .85 for mothers and .78 for fathers. This survey was administered to mothers and fathers at baseline and wave 2.
• Parent-reported efficacy, taken from the Parenting Sense of Competence scale, developed and validated by others. This survey was administered at waves 2, 3, and 4, with alpha coefficients of .84 for mothers and .83 for fathers.
• Parent-reported stress, taken from the Parenting Stress Index, developed and validated by others. This survey was administered at waves 2, 3, and 4, with alpha coefficients of .90 for mothers and .87 for fathers.
• Observed couple behaviors, taken from coded videotaped interactions between mother and father. Undergraduate and graduate students used coding systems created for the study or adapted from prior work. Subscales included negative communication and warmth to partner, with inter-rater intraclass correlations ranging from .63 to .88. Couple behaviors were observed at waves 1 and 3.
• Parent-reported relationship satisfaction, taken from the Quality of Marriage Index, developed and validated by others. This survey was administered at wave 4 and had alpha coefficients of .97 for mothers and .95 for fathers.

The study used the following parenting and coparenting indicators as outcomes:

• Parent-reported coparenting, taken from a measure developed for the study based in part on adaptation of prior measures. Subscales from this measure included coparental support, parenting-based closeness, and coparental undermining. Cronbach's alphas ranged from .72 to .83 for mothers and from .65 to .80 for fathers. This measure was collected at waves 2, 3, and 4.
• Parent-reported parent-child dysfunctional interaction, taken from a scale in the Parental Stress Index, developed and validated by others. Alphas were .79 for mothers and .77 for fathers. This measure was collected at wave 2.
• Observed coparenting, taken from coded videotaped interactions at wave 3 among mother, father, and child. Undergraduate and graduate students used coding systems created for the study or adapted from prior work. Subscales for coparenting included competition, triangulation, warmth, and inclusion, with inter-rater intraclass correlations ranging from .44 to .87.
• Observed parenting behaviors, taken from coded videotaped interactions at wave 3 among mother, father, and child. Undergraduate and graduate students used coding systems created for the study or adapted from prior work. Subscales for parenting behaviors included positivity, negativity, intrusiveness, and reactivity with inter-rater intraclass correlations ranging from .69 to .73.
• Parenting practices, taken from the Parenting Scale, developed and validated by others. This scale assesses discipline practices in parents of children from 18-48 months. The measure produced three outcomes: laxness, overreactivity, and physical punishment. Alpha coefficients for laxness and overreactivity ranged from .76 to .85. Physical punishment included a single item. The scale was administered at wave 4.
• Parent-reported partner psychological aggression, assessed at waves 1 and 4 with subsets of items from the Revised Conflict Tactics Scales. Behaviors reported by mother and father were combined and summed to create a frequency measure. The mothers' alpha was .65 and the fathers' was .68.
• Parent-reported parent-child physical aggression, assessed at wave 4 with the corporal punishment subscale of the Parent-Child Conflict Tactics Scales. Items in the subscale were summed to create a frequency score. Cronbach's alphas were .55 and .57 for mothers and fathers, respectively.

The following child behavioral and academic outcomes were used:

• Parent-reported infant regulation, taken from the Infant Behavior Questionnaire, developed and validated by others. This measure produced two outcomes for mothers and fathers, duration of orienting and infant soothability, and one outcome reported by mothers, child sleep habits. Alphas ranged from .75 to .86 for these indicators measured at wave 2.
• Observed child behaviors, taken from coded videotaped interactions at wave 3 of the mother, father, and child. Undergraduate and graduate students used coding systems created for the study or adapted from prior work. Subscales for child behaviors included self-soothing and sustained attention, with inter-rater correlations ranging from .67 to .87. The measure of child adjustment problems (alpha=.69) aggregated subscales of anger, activity, resistance to control, and sustained attention.
• Mother-reported child behavior problems, taken from the Child Behavior Checklist and the Head Start Competence Scale, administered at wave 4. The Checklist produced five outcomes, including three overall scores (total problems, externalizing problems, and internalizing problems) and two specific subscales (aggression and attention/hyperactivity). Two subscale outcomes were taken from the Competence Scale: social competence and emotional competence. Alpha coefficients for these indicators ranged from .78 to .90.
• Teacher-rated classroom participation, taken from the Total Classroom Participation, developed and validated by others. This outcome was collected at wave 5 (alpha=.95).
• Teacher-rated academic motivation, taken from the Academic Competence Evaluation Scales, developed and validated by others. This outcome was collected at wave 5 (alpha=.96).
• Teacher- and parent-reported conduct problems and emotional problems. These outcomes were collected at wave 5 from the Strengths and Difficulties Questionnaire, developed and validated by others. Teachers showed alpha coefficients of .72 (conduct) and .62 (emotional), and parents had alphas of .60 (conduct) and .59 (emotional).
• Teacher-reported child behavioral problems, taken from the Child Behavioral Checklist, developed and validated by others. The checklist provided five outcomes at wave 5, composed of three specific scales (anxious/depressed, attention problems, and aggressive behavior) and two broad-band indices (internalizing behavior and externalizing behavior). Alpha coefficients ranged from .55 to .92.

The study used the pregnancy-related outcomes listed below. The data were collected at wave 2 but reflect mid-program assessments when four of the eight classes had been offered. Few details were provided on these outcomes.

• Birth weight, in pounds
• Number of weeks born prior to due date
• Premature status, defined as born three or more weeks early
• Caesarian section
• Pregnancy complications rated from parent reports by research team pediatrician
• Number of days child in hospital after birth
• Number of days mother in hospital after birth

The study used the following measures as moderators:

• Parental education measured by total years of education at baseline.
• Parent-reported attachment insecurity in close relationships, measured at baseline with a 20-item subscale of the Relationships Scale Questionnaire, developed by others. Alphas were .80 for mothers and .79 for fathers.
• Parent gender and marital status indicating married and unmarried mothers or fathers at waves 2, 3, and 4.
• Child gender.
• Maternal cortisol levels at baseline. Trained research assistants collected saliva samples during home visits. The measure reflects residualized cortisol levels from regression models using time of day and gestation weeks to predict sample results.
• Observed parental negative communication level at baseline. Trained coders rated maternal and paternal behaviors in videotaped interactions and averaged scores across the two parents. Cronbach's alpha was .84 for this measure.
• Intimate partner violence at baseline. The overall violence scores multiplied frequency by severity, with frequency of violence created from summing mothers' and fathers' item frequency scores, and severity of violence produced from summing item severity scores. Cronbach's alphas were .77 and .81 for prevalence and .92 and .76 for frequency of mothers' and fathers' behaviors, respectively.
• Parent-reported partner psychological aggression at baseline, as described in the above measure of parent-reported psychological aggression.
• Parent-reported severity of parental physical aggression. This mutually exclusive categorical measure defined couples as perpetrating any severe aggression, perpetrating minor aggression only, or not perpetrating any aggression.

The study used the following mediating measure:

• Observed coparenting subscales of competition and positivity measured at wave 3, as described in the above outcome measure of observed coparenting.

Analysis: Multilevel, general linear model, ordinary least squares, logistic, and negative binomial regression models analyzed program effects. For outcomes available separately for mothers and fathers, multilevel models controlled for within-family dependency. For analyses looking at multiple waves of data, multilevel models nested waves of data within family, aggregated at the parent level (Feinberg et al., 2010). Some analyses used full-information maximum-likelihood techniques to accommodate missing data and allow for inclusion of the full eligible sample (Solmeyer et al., 2013; Kan and Feinberg, 2013a; Kan and Feinberg, 2013b). Condition status was assigned at the couple level. Analyses looking at maternal or paternal outcomes appropriately adjusted for clustering within couples, while other analyses were appropriately conducted at the couple level.

Additional analyses explored moderating effects with interaction terms for intervention by parent education and attachment insecurity (Feinberg and Kan, 2008), child gender (Feinberg et al., 2010; Feinberg, Jones et al., 2014), parent gender and marital status (Feinberg et al., 2010), baseline negative communication levels (Feinberg, Jones et al., 2014), and baseline cortisol level (Feinberg, Roettger et al., 2014). Mediation analyses used path analysis, a bias-corrected bootstrap test and the proportion-mediated measure to determine size and significance of mediators (Solmeyer et al., 2013).

The study used different covariates in analyses including parent age, education, and social desirability (Feinberg and Kan, 2008; Feinberg et al., 2009), respondent age, marital status, family income, respondent educational level, social desirability score, financial strain score, and maternal relationship attachment insecurity (Feinberg et al., 2010), child gender, child age, marital status, and baseline characteristics of parental education, family income, and negative communication (Feinberg, Jones et al., 2014), child gender, baseline maternal education, and marital status (Feinberg, Roettger et al., 2014), wave 1 maternal education and parent reports of social desirability (Solmeyer et al., 2013), mother's education (Kan and Feinberg, 2013a), and parent mean education (Kan and Feinberg, 2013b).

Baseline controls were not collected for coparenting, parenting, or child outcomes, as they would not have been appropriate before birth. Thus, most analyses did not control for baseline outcome levels. Analyses for parental depression and anxiety (wave 2) and couple behaviors (wave 3) controlled for baseline levels through the use of group-by-time interactions (Feinberg and Kan, 2008; Feinberg et al., 2009)

Analyses were run as intent-to-treat and included all parents completing data collection, regardless of level of program participation.

Outcomes

Implementation Fidelity: Though couples did not appear to participate fully in the program, the study showed high fidelity to program content. An observer from the project team attended and rated over 90% of intervention sessions for implementation fidelity, finding that an average of 95% of the curriculum content was delivered. Average attendance was 5.50 classes for mothers and 5.38 classes for fathers, with only 3% of mothers and 5% of fathers attending no sessions. Most couples (66% of mothers and 63% of fathers) attended five or more sessions.

Baseline Equivalence: As child outcomes could not have been collected at pre-birth baseline assessment, equivalence could not be tested for all measures. However, the study reported that "analyses indicated no significant differences between intervention couples and control group couples on a wide range of pretest variables, including age, income, education, marital status, weeks of gestation, mental health, and relationship quality" (Feinberg and Kan, 2008). In addition, the investigators provided Blueprints with comparisons of the intervention and control group on 35 parent variables at the time of birth and at waves 2-5. The results demonstrate group equivalence. Only 4 of 175 comparisons (5 times 35) showed a significance difference, and only one variable - father's insecure attachment - showed consistent significant differences across waves.

Differential Attrition: Attrition differed according to the wave of data collection, with participation declining across the study period. The studies reported partial information on differential attrition for individual waves (described below). However, the investigators provided Blueprints with a more detailed analysis of differential attrition that summarizes the pattern across five waves for 39 variables. The 195 tests (5 times 39) for differences between completers and dropouts showed no significant results at the time of birth and wave 2, one significant result at wave 3, five significant results at wave 4, and two significant results at wave 5. The wave 4 pattern of higher attrition among high-risk subjects was found in additional analyses to occur primarily in the control group, which may serve to attenuate differences across conditions.

Participation in posttest data collection was high, with 92% of mothers and 90% of fathers completing wave 2. The rates of participation were similar across control (91%) and intervention (89%) couples, and the study noted that "there were no differences in the association of pretest variables with continued participation across conditions" (Feinberg and Kan, 2008).

For wave 3, 88% of mothers and 93% of fathers of the original 165 couples completed data collection. At both baseline and wave 3, married couples were more likely to have videotaped interactions (Feinberg et al., 2009).

The study reported that 84.6% (N=137) of families provided data at wave 4. Regressions testing whether study participation between intervention and control groups was associated to background characteristics indicated that attriters in the control condition had lower maternal education levels than intervention attriters, though no other significant differences emerged. The study reported that wave 4 analyses using imputation techniques to estimate missing values for attriters supported the reported results for models only including wave 4 participants (Feinberg et al., 2010).

Wave 5 showed substantial attrition, due to the long follow-up period. The study predicted participation in wave 5 using several demographic and key study variables, finding only that family income was a significant predictor (Feinberg, Jones et al., 2014).

Analyses looking at moderation of program effects on mid-program pregnancy-related outcomes by cortisol levels used a subsample of mothers who consented to have cortisol measurements at baseline and completed data at wave 2. Of the 137 mothers who provided cortisol levels, 90% (N=123) completed posttest data. No analyses comparing participation rates were provided (Feinberg, Roettger et al., 2014).

Mid-Program Assessment: Pregnancy-related outcomes assessed mid-program (4 of 8 classes) indicated 1 main effect of the intervention (reduced levels of Caesarian birth), out of 6 outcomes. Birth weight, number of weeks born premature, premature status, the number of days the child was in the hospital, pregnancy complications, and the number of days the mother was in the hospital showed no direct program effects.

Posttest: At wave 2 (posttest when children were six months old), two of four child outcomes and 6 of 12 parent outcomes showed significant improvement. Mothers and fathers' coparental support, fathers' parent-child dysfunctional interaction and parenting-based closeness, mothers' depressive symptoms and anxiety, father-reported infant soothability, and child duration of orienting all showed significant differences. Mothers' and fathers' coparental undermining, mothers' parent-child dysfunctional interaction and parenting-based closeness, fathers' depressive symptoms and anxiety, mother-reported infant soothability, and child sleep habits did not show any significant program effect.

Follow-up: Of the 20 parenting, couple, coparenting, and child outcomes tested at wave 3, 13 showed significant improvements for the intervention participants. Mothers and fathers in the treatment group showed reduced competition and triangulation in coparenting, increased warmth to partner, and increased parenting positivity. Mothers, but not fathers, improved on coparenting inclusion and negative communication to partner, while fathers significantly increased coparenting warmth and reduced parenting negativity. Neither maternal nor paternal active coparenting cooperation showed a significant difference. For child outcomes, self-soothing improved significantly, but sustained attention did not.

Wave 4 analyses indicated a significant program effect on 10 of the 17 parent, interparental relationship, parenting, and child outcomes. Intervention participants improved parental stress, parental efficacy, coparenting quality, parenting over reactivity, parenting laxness, physical punishment, total behavior problems, child externalizing problems, child aggression, and child social competence. No significant effect emerged for parental depression, relationship satisfaction, child internalizing problems, child attention/hyperactivity, or child emotional competence (Feinberg et al., 2010), or for partner psychological aggression or parent-child physical aggression (Kan and Feinberg, 2013b). Additional analyses showed that among families with boys, the program had an effect on internalizing, attention/hyperactivity, and parent relationship satisfaction, and showed a stronger effect on total behavior problems, externalizing, and aggression for boys. For models assessing outcomes available at multiple waves (parental stress, efficacy, depression, and coparenting quality), there was no evidence that intervention and control conditions differed in rates of change across postintervention assessments.

Main effects analyses indicate that of the two parent-reported and nine teacher-reported child academic and behavioral outcomes measured at wave 5, one showed significant improvements for the intervention group: teacher-reported internalizing problems (Feinberg et al., 2014). In looking at program effects by child gender, boys but not girls showed significant improvement for anxious/depressed, attention problems, aggressive behavior, internalizing, and externalizing (additional analyses submitted to Blueprints). Parent- and teacher-reported conduct problems and emotional problems and teacher-reported classroom participation and academic participation showed no direct or gender moderated program effects.

Moderation: The study reported parental education, maternal and paternal attachment insecurity, the interaction of parent gender and marital status, baseline negative communication levels, baseline cortisol levels, baseline psychological partner aggression, baseline physical partner aggression, and baseline intimate partner violence as significant moderators of relationships between intervention status and different outcomes.

Results of moderation analyses on wave 2 parental and child outcomes indicated that intervention effects were greater for those with less education or higher levels of attachment insecurity. Parental education significantly moderated the intervention effect on maternal depression, mother report of coparental support, and child sleep habits, but did not moderate the other 13 outcomes. Of 16 moderation models determining whether the effects of the intervention depended on maternal or paternal attachment insecurity, seven showed significant results, though one effect was not interpreted as it showed a direction opposite to the other six. Father's insecurity moderated the effect of the intervention on maternal depression, mother's coparental support, coparental undermining, maternal dysfunctional interaction, and paternal dysfunctional interaction. Mother's insecurity moderated maternal depression.

Wave 3 models showed moderation of the treatment by baseline intimate partner violence perpetration for 6 of 12 comparisons. For mothers and fathers, parenting positivity, negativity toward daughters, and reactivity to distress indicated stronger program effects for couples with higher baseline levels of intimate partner violence. Parenting positivity, negativity toward sons, and parenting intrusiveness showed no moderation effects for mothers or fathers.

Models also tested for moderating effects of parent gender and marital status on parent and child outcomes collected at wave 4. One significant result indicated that among nonmarried mothers, intervention participants showed lower levels of depression. No other significant moderating effects of child gender or parent gender and marital status emerged. Additional results looking at moderation effects for wave 4 outcomes showed that the intervention had stronger effects on fathers' psychological partner aggression for couples with higher baseline levels of psychological partner aggression or physical partner aggression and on mothers' aggression toward the child for couples with higher baseline levels of psychological partner aggression. Parents' baseline negative communication moderated program effects on most parent- or teacher-reported child outcomes at wave 5. Parent-reported emotional problems and teacher-reported classroom participation, academic motivation, conduct problems, emotional problems, anxious/depressed, aggressive behavior, internalizing, and externalizing all indicated that the intervention worked best for children whose parents had negative communication at baseline.

Four of the 6 pregnancy-related outcomes indicated moderation, showing that the intervention improved outcomes only for women with high cortisol levels. This effect was observed for birth weight, number of weeks born premature, the number of days the child was in the hospital, and the number of days the mother was in the hospital. Caesarian birth and pregnancy complications showed no moderation effects from cortisol levels.

Mediation: In Solmeyer et al. (2014), coparenting competition mediated the effect of the intervention on wave 4 child adjustment problems for mother-son and father-child relationships, but not mother-daughter relationships. The proportion of total effects mediated by coparenting competition was 39% for mothers and sons and 55% for fathers. Coparenting positivity did not mediate program effects for mothers or fathers.

Study 2

Summary

Feinberg, Jones, Hostetler et al. (2016), Feinberg, Jones, Roettger et al. (2016), Jones et al. (2018), Feinberg and Jones (2018), and Feinberg and Gedaly (2021) conducted a randomized controlled trial with a sample of 399 cohabiting heterosexual couples expecting their first child. The intervention couples (n =221) and control couples (n = 178) were followed from the prenatal period to 10 months, two years and seven years after birth of the child. Child measures of soothability, nighttime waking, internalizing and externalizing came from parent reports. When the child was age 7, additional measures of child behavior were collected from direct observations and teacher reports.

Feinberg, Jones, Roettger et al. (2016), Feinberg, Jones, Hostetler et al. (2016), Jones et al. (2018), Feinberg and Jones (2018), and Feinberg and Gedaly (2021) found that the intervention group (compared to the control group) showed significantly:

• Greater observational family interaction and co-parenting at posttest and long-term follow-up
• Less parent-reported family violence at posttest
• Greater parent-reported child soothability, duration of orientation (i.e., attention span), and night-time sleeping at posttest
• Fewer parent-reported child internalizing behaviors at long-term follow-up
• Fewer parent-reported child nighttime wakings at long-term follow-up
• Higher observer-rated positive affect and lower negative affect at seven-year follow-up
• Lower observer-rated physical aggression toward objects at seven-year follow-up
• Greater observer-rated affection for mothers and fathers at seven-year follow-up
• Lower teacher-rated externalizing at seven-year follow-up

Evaluation Methodology

Design:

Recruitment: In three mid-Atlantic and one southern state, cohabiting heterosexual couples expecting their first child were recruited primarily through hospitals, childbirth education programs and Ob/Gyn clinics, but also through media advertisements and fliers. The study reported identifying 743 eligible couples, of which 399 agreed to participate. Separate cohorts of participants were recruited in succession across study sites between the fall of 2008-2012.

Assignment: The study randomly assigned the 399 couples to the treatment or control condition in a randomized block design, based on pretest data. The intervention sample included 221 couples who received the program at five hospital sites, and the control sample included 178 couples who received only mailed materials on childcare and child development.

Assessments/Attrition: The pretest came before the birth of the child, and the posttest came 10 months after the birth of the child. Of 399 participants, 312 (78%) completed the posttest. In addition, eight couples were excluded due to multiple births, developmental disorders, or severe parent or infant medical problems, leaving 304 (76%) participants in Feinberg, Jones, Hostetler et al. (2016). Feinberg, Jones, Roettger et al. (2016) excluded 45 other couples without pretest control measures, with low session attendance, or extreme propensity score. Their posttest sample included 259 participants (65%).

The long-term follow-up came two years after the birth of the child. In Jones et al. (2018) and Feinberg & Jones (2018), 302 of 399 (76%) randomized couples were included in the analytical sample. A total of 89 families (22%) were lost to follow-up. However, only 240 provided observational data (40% attrition).

Feinberg and Gedaly (2021) assessed the children at ages 7-8 in 2017-2020, but the COVID-19 pandemic stopped data collection before completion. Of the 399 randomized families, 200 (50%) contributed data. Additionally, of the 200 participating families, 157 fathers and 166 mothers from 168 families participated in the observational task procedure with their child (84% completion relative to the 200 participants and 42% completion relative to the 399 randomization participants). Of the 200 participating families, 173 gave consent to contact the child's teacher, and 125 teachers responded (62% completion relative to the 200 participants and 31% completion relative to the 399 randomization participants).

Sample:

The authors referred to the sample as well-educated, high functioning, and low risk. The average education level in the sample was 15.7 years. About 87% of the couples were married, and 81% of the participants were non-Hispanic white. The mean age of expectant mothers was 29.1 years and fathers was 31.1 years, and all were at least age 18.

Measures:

Feinberg, Jones, Hostetler et al. (2016) examined 26 outcomes in four groups: 1) 11 measures of positive and negative family interaction, as coded from videotapes by observers who were unaware of condition, 2) six parent-reported measures of family relations and their own adjustment, 3) four parent-reported measures of family violence, and 4) five parent-reported measures of child outcomes related to soothability, duration of orientation (i.e., attention span), and sleeping. Inter-rater reliabilities for the observational measures ranged from .66 to .85. The parent-reported measures had alphas of .75 or higher. Of the four groups of outcomes, only the last related to child behavior. However, because parents helped deliver the program, their ratings of child outcomes may not be independent.

Feinberg, Jones, Roettger et al. (2016) measured birth weight, maternal length of stay in hospital after birth (days), and newborn length of stay in hospital after birth (days) as outcome measures. These measures were obtained from parents up to 10 months after the birth of the child and depended on accurate recall.

Jones et al. (2018) and Feinberg and Jones (2018) included the following measures in their study: five observational family interaction tasks (coparenting triadic relationship quality, coparenting positivity aggregate, coparenting negativity aggregate, parenting positivity aggregate, and parenting negativity aggregate); two self-report parent measures of depressive symptoms and anxiety from the Center for Epidemiological Studies Depression Scale and the revised Penn State Worry Questionnaire; two parent-reported child measures of externalizing and internalizing scales from the Child Behavior Checklist; and two mother-reported number of night wakings and hours child sleeps during night from the Child Sleep Questionnaire. Two research assistant raters unaware of conditions coded the observational family interaction tasks. However, parents helped deliver the program and rated their children.

Feinberg and Gedaly (2021) used two types of measures. First, measures of observed child behavior came from videotapes of family interactions made during home visits. One set of codes measured general behavior: positive affect, activity, physical aggression towards objects, and negative affect (a composite of anger and sadness). Another set of codes measured behaviors towards each parent: affection, engagement, resisting control, hostility, and physical aggression. In a separate supplemental document provided to Blueprints, the authors noted that raters were unaware of condition assignment and that ICC values indicated acceptable inter-rater reliabilities. Second, two teacher-reported measures of externalizing (alpha = .79) and internalizing (alpha = .71) came from the Strengths and Difficulties Questionnaire. Teachers were not informed about the intervention or randomized assignment of participants, and presumably, provided fully independent ratings.

Analysis:

Feinberg, Jones, Hostetler et al. (2016) used linear and negative binomial regression models for family-level measures and multilevel regression models with a random intercept for outcomes available separately for each parent. When possible, baseline outcomes served as one of many control variables, but pre-birth parenting and child outcomes could not be measured.

Feinberg, Jones, Roettger et al. (2016) used separate regression models for each outcome to examine main and moderated effects of the intervention. Moderation was analyzed using two- and three-way linear and quadratic interactions.

Jones et al. (2018) used separate regression models to test the main effect of condition (intervention or control) for each outcome. For parent-specific outcomes nested within dyads, the study used multilevel regression models with a random intercept. Ordinary regression models were used for mother-reported child sleep and observed triadic relationship quality, which were outcomes available only at the family level.

Feinberg and Jones (2018) used structural equation models to test for mediation.

Feinberg and Gedaly (2021) used ordinary least squares regression with numerous covariates that had been utilized in prior studies of this program. The analysis could not use baseline child outcomes, as the baseline assessment occurred before birth of the child.

Intent-to-Treat: Feinberg, Jones, Hostetler et al. (2016) excluded eight families (2%) with multiple births, infant developmental disorders (e.g., autism, Down syndrome), or severe parent or infant medical problems (e.g., severe congenital defect, poor maternal health). Otherwise, the study used multiple imputation for missing data. Because the CONSORT diagram in Figure 1 shows that 304 participants were analyzed, rather than the full sample of 399, it appears that the multiple imputation was applied to missing pretest data rather than missing posttest data.

Feinberg, Jones, Roettger et al. (2016) violated the intent-to-treat criterion by excluding subjects who attended fewer than three of the five prenatal classes, which was deemed insufficient participation. However, the analysis also used propensity score weights to attempt to correct for selection bias created by dropping non-participants from the intervention group.

Jones et al. (2018) used ITT analyses and included multiple imputation procedures for missing data in the intervention and control groups. A total of 302 families had complete data available for analyses, which included 169 of 221 (76%) intervention group families and 133 of 178 (75%) control group families. Seven families from the intervention group and two families from the control group (a total of only 2%) were excluded from analyses due to multiple birth or child heath/development complications.

Feinberg and Jones (2018) used full information maximum likelihood estimation to include nearly all of the full sample. However, they excluded three families who did not attend any of the prenatal classes and eight families with multiple births, infant developmental disorders, or severe parent or infant medical problems.

Feinberg and Gedaly (2021) used all available data. The pandemic lockdown prevented complete data gathering. Although it is not certain, it appears that the stoppage did not relate to participation in the program.

Outcomes

Implementation Fidelity: The average attendance rate at the five prenatal classes was 87.9% among participants, with an average of 4.3 classes. The authors estimated that attending three of the five prenatal classes would represent minimally adequate dosage. Of the intervention mothers, 91.8% achieved that minimum attendance. Feinberg, Jones, Hostetler et al. (2016) reported that, based on ratings of videotaped classes, leaders delivered an average of 90% of the content as planned.

Jones et al. (2018) reported that more than half of intervention couples attended at least eight of the nine classes and only 4% of the families attended no sessions.

Baseline Equivalence:

Feinberg, Jones, Hostetler et al. (2016) assessed group differences on over 60 measured preintervention variables. The authors stated only that they "found no patterns of statistical differences between groups from this assessment."

Feinberg, Jones, Roettger et al. (2016) did not examine baseline equivalence in outcome measures, because the outcomes related to the subsequent birth of the child. Although the study did not provide significance tests for differences in demographic and other measures at baseline, the percentages appear similar (Table 1).

Jones et al. (2018) did not report significance tests for available baseline outcomes and sociodemographic measures. Feinberg and Jones (2018) noted that there were "no significant differences between intervention and control conditions" on 20 demographic and outcome measures listed in Table 1.

In a separate document provided to Blueprints that supplemented the Feinberg and Gedaly (2021) paper, the authors performed tests for baseline equivalence using the observational data (n = 168) and teacher data (n = 125) available at the long-term follow-up. These tests extended those for the full randomized sample by checking for distortions of equivalence due to attrition. In 16 tests for condition differences in sociodemographic and outcome variables using the observational data and in 16 tests for condition differences in sociodemographic and outcome variables using the teacher data, all were non-significant.

Differential Attrition:

Feinberg, Jones, Hostetler et al. (2016) used logistic regressions to predict posttest participation with interaction terms between condition and specific predictors. The predictors included baseline parent and couple characteristics representing stress, mental health, and relationship qualities. The authors stated that they "found no evidence of differential attrition."

Feinberg, Jones, Roettger et al. (2016) examined if baseline socioeconomic, mental health, and other demographic information predicted the likelihood to drop out before posttest. They reported only that there were "no significant condition differences in likelihood to drop out of the study before posttest."

Jones et al. (2018) and Feinberg and Jones (2018) tested for differential attrition as the effects of baseline measures on posttest participation and the interaction of condition and baseline measures on posttest participation. They stated that these analyses showed "no evidence of differential attrition between conditions" but provided no details.

In a separate document provided to Blueprints that supplemented the Feinberg and Gedaly (2021) paper, the authors performed additional tests for differential attrition. Logistic regression models examined the interaction effects of condition times each of 17 baseline sociodemographic and outcome variables in predicting attrition in the observational and teacher data. The interaction terms in all models were non-significant, which indicated no apparent attrition bias in the analysis samples.

Posttest:

Feinberg, Jones, Hostetler et al. (2016, Table 2) found significant and beneficial intervention effects for 12 of 21 risk and protective outcomes (effect sizes ranged from .20 to .79). One other outcome, the quality of marriage index, showed a significant iatrogenic effect. For child outcomes, the results showed significant and beneficial intervention effects for four of five parent-reported outcomes, including soothability, duration of orienting (i.e., attention span), night waking, and difficulty going back to sleep. Effect sizes ranged from .20-.27. For outcomes without a significant main effect, the authors presented some evidence of moderation that the intervention significantly benefited parents with high negative communication and high depression at baseline.

Feinberg, Jones, Roettger et al. (2016) found no main effects of the program on parent-reported birth outcomes. However, they found significant moderation effects on birth outcomes in 7 of 36 interaction terms (p < .05). The significant moderation indicated that the intervention reduced the harm of high economic strain, depression, and anxiety.

Long-Term:

Approximately two years post-intervention (when children were two years old), Jones et al. (2018) found significant positive effects for five of 11 outcomes. The intervention group, compared to the control group, showed greater co-parenting triadic relationship quality (d = .39) and lower co-parenting negativity (d = .38) and parenting negativity (d = .41) in observational family interaction tasks, and fewer parent-reported child internalizing behaviors (d = .19) and parent-reported child nighttime wakings (incidence ratio rate; IRR = .30). Moderation results found that intervention impact was moderated by pretest levels of observed couple negative communication in five of nine study outcomes (i.e., a larger overall intervention impact for higher risk families at baseline).

Feinberg and Gedaly (2021) reported results for children in second grade, nearly seven years after the program end. First, for the four observational outcome measures of general behavior, the intervention group was rated significantly better than the control group on three: positive affect (d = .53), physical aggression towards objects (d = -.36), and negative affect (d = -.52). Second, for the 10 outcome measures of behavior towards mothers and fathers, the intervention group was rated significantly higher than the control group on two: affection towards mother (d = .52) and affection towards father (d = .70). Third, for the two teacher-rated outcome measures, the intervention group was rated significantly better on one: externalizing (d = -.83). Fourth, some moderation tests suggested that the program reduced child hostility toward fathers among girls but not boys.

Mediation. Feinberg and Jones (2018) examined the indirect effects of the intervention on two outcomes - child externalizing and internalizing at age 2. The four mediators included the 10-month posttest scores on parent depression, parenting stress, parenting negativity, and co-parenting negativity. First, the authors examined each of the mediators separately. None of the indirect effects showed a significant mediated impact for either the externalizing or internalizing outcomes. Second, the authors examined mediators in combination. These results showed a significant indirect effect on externalizing and a marginally significant indirect effect on internalizing (p = .089). The authors argued that, although the impact of a single mediator was small, the impact of multiple mediators combined revealed the ability of the program to improve child outcomes.