What dyadic reparation is meant to do: An association with infant cortisol reactivity

Abstract

Background

The latency to reparation of interactive mismatches (interactive repair) is argued to regulate infant distress on a psychobiological level and maternal anxiety disorders might impair infant regulation.

Sampling & Methods

N = 46 dyads (n = 19 mothers with an anxiety disorder, n = 27 controls) were analyzed for associations between interactive repair and infant cortisol reactivity during the Face-to-Face-Still-Face 3–4 months postpartum. Missing cortisol values (n = 16) were imputed. Analyses were conducted on both the original and the pooled imputed data.

Results

Interactive repair during the reunion episode was associated with infant cortisol reactivity (original data: p < .01; pooled data: p < .01), but not maternal anxiety disorder (p > .23). Additional stepwise regression analyses found that latency to repair during play (p < .01), an interaction between distress during the first trimester of pregnancy and latency to repair during reunion (p < .01) and infant self-comforting behaviors during the reunion episode (p = .04) made independent contributions to cortisol reactivity in the final regression model.

Conclusions & Limitations

This is the first study demonstrating that interactive repair is related to infant psychobiological stress reactivity. The lack of a relation to maternal anxiety disorder may be due to the small sample size. However, this result emphasizes that infants respond to what they experience and not to the maternal diagnostic category.

Introduction

An exposure to high levels of stress during the early postnatal period is related to alterations in brain functioning [1]. A critical question is why some infants react to stressful experiences and why others do not [2]. During the first months of life, HPA-axis functioning is associated with the quality of the caregiver-infant interaction [3], which is argued to provide an external source of regulation [4]. Higher ratings of maternal sensitivity and dyadic coordination in infancy are associated with increased behavioral and physiological regulation [5, 6] and emotional resilience at older ages [7]. For example, the more sensitive the mother interacts with her three-month-old infant, the better is the infant’s cortisol recovery from an everyday stressor [8]. For the most part, existing studies utilize global ratings of maternal sensitivity. While sensitivity has proven to be critical for understanding attachment, it is recognized that sensitivity ratings are both multi-dimensional and global and do not specify what aspects of sensitivity are at work [9]. One putative mechanism underlying sensitivity might be interactive reparation [10, 11], which refers to the quality and form of the mutual regulation between infants and mothers. It is the capacity of both members of the dyad, infant and caregiver, to repair affective and behavioral uncoordinated (mismatching) states and consequently, stressful states change back to more positive and coordinated (matching) states [12]. The successful transformation of mismatching into positive matching states might account for an early form of implicit relational knowing that social interactions can be positive and repairable [13]. The establishment of this implicit knowledge of relationship might be of vital importance for infant emotional development. Recently, matching states were demonstrated to be associated to infant affective regulation [14]. Nevertheless, to our knowledge, there are no studies using micro-analytic and psychobiological measurements, which evaluate the hypothesis that short latencies to the reparation of micro-temporal affective mismatching states scaffold infant regulation whereas long latencies increase dysregulation and distress [15].

Infants have a repertoire of early self-regulatory behaviors (hand-to-mouth-movements, non-nutritive sucking) that are thought to regulate infants’ stressful experiences [16]. However, these self-directed regulatory behaviors (self-comforting behaviors) are limited in down-regulating heightened affective states [17]. Furthermore, these self-comforting behaviors decrease with age as infants engage in more complex regulatory strategies by shifting attention, using objects and engaging with the caregivers [18]. Certainly, by three months of age, if not earlier, infants signal the caregiver to resume interaction, if the dyad is in mismatching states (e.g., the caretaker is unresponsive) [9]. If these attempts fail to reestablish dyadic coordination, the infants experience negative affect and distress [19]. Chronic failure is likely to have negative effects on infant’s emotional, social and cognitive development [10, 19].

The developmental risk for infants of parents with affective disorders is well established [20] and might in part be mediated by an impaired mother-infant interaction. For example, dyads where mothers suffer from major depression show fewer positive matched states and longer latencies to reparation of mismatching states [21] during the Face-to-Face-Still-Face paradigm (FFSF) [22], an experimental paradigm in which infants experience a socio-emotional stressful event. Furthermore, infants of clinically depressed mothers use more self-comforting behaviors as a regulatory strategy compared to control infants, who engage in social monitoring to signal interactive reengagement [23].

Dyads with anxious mothers, though less well studied than dyads with depressed mothers, also show interactive difficulties. In a study on depressed mothers and their three-month-old infants [24], the authors found that mothers with high scores in the State-Trait-Anxiety-Inventory (STAI) [25] interacted less positively and more intrusively than controls. In another study [26] in which participants with co-morbid depression were excluded, mothers who were more anxious showed less sensitivity and emotional vocalizations in their interactions with their ten- to fourteen- month-old infants. A study on mothers with a generalized anxiety disorder according to DSM-IV criteria [27] suggested that mothers with this diagnosis were less responsive in the interaction with their 10-month old infant, especially when a ruminating style of thinking was induced [28]. At older ages (7–12 years), it was found that anxious mothers interact more intrusively and less warmly with their children, effects that were moderated by the child’s expression of anxiety and mediated by the mother’s experience of negative emotions [29]. Thus, the authors suggested that maternal anxiety is associated with a reduced tolerance of children’s negative emotions. It also is a frequent finding that anxious mothers demonstrate more insensitive behaviors compared to healthy controls [30] and as a consequence infants of anxious mothers might frequently lack sufficient regulatory scaffolding. Recently, excessive crying in infants was predicted by the mothers’ anxiety disorder prior to pregnancy [31]. This dysregulation might in part underlie the increased risk for the development of mental disorders in infants of anxious caregivers [20].

It has been argued that an early and chronic dysregulation of the HPA axis may account for developmental risks [32] observed in infants of mothers with anxiety disorders. However, there are only few studies on the influence of maternal anxiety on infant psychobiology. In a study on six-month-old infants of mothers with co-morbid depression and anxiety, the infants were found to express significantly increased cortisol reactivity in comparison to control subjects [33]. Another study demonstrated that a prepartum maternal anxiety disorder and global measures of maternal sensitivity at seven months postpartum independently predicted infant cortisol reactivity in the FFSF [34]. Furthermore, this research group demonstrated that the association between maternal sensitivity and infant distress was especially marked for infants of women who experienced a prepartum anxiety disorder [35]. However, anxiety disorders meeting DSM-criteria have not been investigated sufficiently with regard to their influence on early dyadic regulation on a micro-temporal level in combination with psychobiological measurements.

The FFSF is the prevailing method to investigate mother-infant interaction and the effects of distress on infants [36]. The experimental interruption of maternal engagement (still-face episode) is a socio-emotional stressor to the infant [9, 36]. Affective and behavioral responses to the still-face are striking and include a decrease in positive affect, an increase in negative affect and infant behaviors that are aimed at changing the mothers’ behavior and reducing stress, such as gaze and self-comforting behaviors [9, 36]. Infants also show signs of physiologic reactions of vagal tone [37, 38] and skin conductance [38, 39]. During the reunion episode, mother and infant are challenged to re-establish interactive coordination and mutual regulation following the stress of the still-face. This reunion episode is particularly informative regarding the regulatory quality of the interaction [40]; infants gaze more towards the mother and express more positive affect. Negative affect also decreases, though it may still be at higher levels than in the first play episode [40]. Although cardiac measures recover [41], it has been found that skin conductance remained high during the reunion episode [38]. In addition to these physiological markers of distress [37, 39], salivary cortisol concentrations were successfully used to quantify an increase in reactivity of the HPA-axis in response to the still-face [42, 43].

The primary purpose of this study was to examine the influence of micro-temporal interactive reparation and maternal anxiety disorders on infant cortisol reactivity. For this analysis, we concentrated on the challenging reunion episode of the FFSF. We expected shorter intervals to interactive reparation would be associated with lower infant cortisol reactivity. Furthermore, we expected that infants of mothers with a diagnosed anxiety disorder would have increased cortisol reactivity in comparison to the control group, and that the association between latency to repair and cortisol reactivity would be greater for dyads in the clinical group. Additionally, the associations of positive dyadic matching states, infant self-comforting behaviors, and maternal distress during pregnancy with cortisol reactivity were examined. We assumed coordinated states to be negatively, and infant self-comforting as well as distress during pregnancy would be positively related to infant cortisol reactivity. Finally, using regression analyses we evaluated the independent contributions of these variables to infant cortisol reactivity.

Method

Sample

This sample was part of a larger longitudinal study [44, 45]. Recruitment took place using flyers, newspaper advertisements and public birth announcements as well as by pregnancy screenings at the Heidelberg University Women’s Hospital between July 2006 and October 2010. In total, N = 122 women were recruited for the larger study. Mental health disorders were diagnosed according to DSM-IV criteria. For the clinical group, co-morbid acute axis I disorders, as well as acute suicidal tendencies were exclusion criteria. The controls needed to have no current or antecedent mental health problems. For the present analyses, excluded from the total sample were n = 14 dyads, who met diagnostic exclusion criteria; n = 37 dyads as recruited too late (entering the study later than age 4.5 months); the video recording failed on n = 2 dyads. In the remaining subsample, N = 47 mothers agreed to salivary cortisol sampling of their infants. Infant medication (e.g., cortisone) was an exclusion criterion (n = 1). Furthermore, prematurity (defined as a gestational age at birth below the completion of the 37^th week) and SGA (Small for Gestational Age as evaluated by OBGYNs and/or pediatricians) were infant exclusion criteria. However, there were no such cases in the final sample (N = 46). It consisted of n = 19 dyads with mothers who had an anxiety disorder (clinical group) and n = 27 dyads with mothers who had no clinical diagnosis (control group). In the clinical group, n = 15 women suffered from more than one anxiety disorder: n = 12 women were diagnosed with a panic disorder with or without agoraphobia or an agoraphobia without history of panic disorder; n = 9 women had a generalized anxiety disorder; n = 8 women had an obsessive-compulsive disorder; n = 6 women were diagnosed with a social phobia; and n = 6 mothers had a specific phobia; n = 1 woman suffered from a post-traumatic stress disorder; and n = 1 woman was diagnosed with an anxiety disorder not otherwise specified (NOS). All mothers had a prepartum onset of anxiety disorder and did not suffer from any somatic disease. Infants were born full term and had no congenital abnormalities. All APGAR scores were equal to or higher than seven. Maternal and infant demographic statistics are presented in Table 1.

Table 1.

Maternal and infant demographics and tests on comparability of subgroups.

	general	control	anxiety	t (p)	female	male	t (p)
maternal age (years) a M (SD)	32.4 (5.1)	33.2 (4.8)	31.3 (5.5)	1.27 (.21)	31.5 (4.9)	34.1 (5.3)	1.64 (.11)
gestation age (weeks) b M (SD)	39.6 (1.3)	39.7 (1.4)	39.3 (1.3)	0.96 (.35)	39.7 (1.3)	39.2 (1.4)	1.15 (.26)
APGAR (average) c M (SD)	9.4 (0.7)	9.4 (0.7)	9.4 (0.7)	0.44 (.66)	9.4 (0.8)	9.4 (0.5)	0.18 (.86)
infant age (months) d M (SD)	3.3 (0.4)	3.3 (0.3)	3.4 (0.4)	1.42 (.16)	3.7 (0.4)	3.3 (0.3)	0.99 (.33)
maternal education (frequencies)	general	control	anxiety	U (p)	female	male	U (p)
university degree	25	15	10		17	8
university entrance qualification	7	5	2	231.5 (.54)	4	3	232.0 (.99)
high secondary qualification	12	7	5		9	3
low secondary qualification	2	0	2		1	1
number of children (frequencies)	general	control	anxiety	U (p)	female	male	U (p)
one infant	28	14	14	197.0 (.13)	22	6	165.0 (.07)
two infants	13	9	4		6	7
three infants	5	4	1		3	2
marital status (frequencies)	general	control	anxiety	χ2 (p)	female	male	χ2 (p)
married	32	22	10	2.67e (.10)	21	11	0.74f (.39)
not married	10	4	6		8	2
infant gender (frequencies)	general	control	anxiety	χ2 (p)	female	male	χ2 (p)
female infants	31	19	12	0.26g (.61)	/	/	/
male infants	15	8	7		/	/	/

Notes.

^amin = 22.0; max = 43.0;

^bmin = 37.0; max = 41.9;

^cmin = 7.0; max = 10.0;

^dmin = 2.5; max = 4.3;

^e1 cell has expected count less than 5, minimum expected count is 3.81;

^f1 cell has expected count less than 5, minimum expected count is 3.10;

^g0 cells have expected count less than 5, minimum expected count is 6.20.

Procedure & Instruments

After arrival at the laboratory, mothers were informed about the study aims and procedures and completed a questionnaire assessing their socio-demographic status. Written informed consent was obtained. Mother-infant interaction was assessed between the third and fourth month postpartum in a video laboratory of the Heidelberg University Hospital using the FFSF. The infant was secured in a booster seat in front of the mother who was briefed using a standard text. One camera focused on the infant while another was focused on the mother. A single screen, simultaneously displaying the two different frontal views, was created by transmitting both recordings through a split screen generator.

The FFSF paradigm consists of three episodes: First, an initial face-to-face interaction in which the mothers are instructed to play with their infant as usual, but without the aid of toys and pacifiers. Next, the still-face episode in which the mothers turn their head aside while silently counting to ten and then look back at the infant, but do not make any gestures, facial expressions or vocalizations creating a prolonged state of interactional mismatch. Finally, the procedure ends with the reunion episode in which the mother resumes the face-to-face play with her infant. Each of the three FFSF episodes lasted two minutes and was ended by a tap from a research assistant from the adjoining room, which likewise served as initiation of subsequent episodes.

Salivary cortisol is a valid marker for infant stress reactivity in early infancy [8, 44], despite the weak circadian organization of the HPA-axis in early months [3, 46]. It was collected immediately prior to (C₁), immediately after (C₂) and 20 minutes after the FFSF paradigm (C₃). Infants sucked on a cotton pad until it was saturated. The saliva was then expressed and stored at −20 °C until analysis. To account for possible effects of circadian rhythm on cortisol reactivity, we attempted to have the visits to the laboratory between 10.00 and 11.00 AM (M = 10.9 AM, SD = 1.7h), though this was not feasible for every mother. Ten infants (21.7 % of study sample) were assessed after 11 AM (M = 13.4 AM, SD = 1.58 h). Consequently, time of day was considered as potential confounder. Moreover, since cortisol reactivity is strongly associated with daytime napping or feeding, mothers were instructed to keep their infants well rested and well fed on their usual routine in order not to confound cortisol assessment. Additionally, the time to and length of prior feeding and napping were considered as potential confounders.

Diagnosis of maternal Anxiety Disorder

Following the FFSF, the German version of the Structured Clinical Interview for DSM-IV-disorders (SKID) [47] was administered to the mothers. According to DSM-IV, anxiety disorders include generalized anxiety disorder, panic disorder with and without agoraphobia, agoraphobia without history of panic disorder, specific phobias, social phobia, obsessive-compulsive disorder, post-traumatic stress disorder and the anxiety disorder NOS.

Coding of mother-infant interactions

Infants’ and mothers’ behavior during the FFSF was coded by two trained and reliable coders using the German translation and revision of the micro-analytical Infant and Caregiver Engagement Phases (ICEP-R) [48]. The coders were blind to the hypotheses of the study and the maternal psychiatric status. The ICEP-R phases combine information from the infant’s and caregiver’s face, direction of gaze and vocalizations. The ICEP-R engagement phases for the infant are negative engagement (further divided into withdrawn and protest), object/environment engagement, social monitor and social positive engagement. The ICEP-R codes for the caregiver are negative engagement (further divided into withdrawn, hostile and intrusive), non-infant focused engagement, social monitor/no vocalizations or neutral vocalizations, social monitor/positive vocalizations and social positive engagement. Additionally for infants and co-occurring with the engagement codes, oral and manual self-comforting behaviors, distancing and autonomic stress indicators were coded. Oral self-comforting included: (1) the infants’ initiated skin contact between their own body parts and their mouth, (2) the infants’ initiated mouth contact to objects, or (3) sucking on the caregiver’s hand or fingers (self-initiated or not). Manual self-comforting behaviors are coded if the infants touch one hand by the other. Distancing and autonomic stress indicators occurred too rarely (distancing: M = 0.06 %; autonomic stress indicators: M = 0.15 % over the whole FFSF), to be included in the analyses.

We coded the video tapes using the Noldus Observer Video-Pro^® coding system with one sec. time intervals. 20% (n = 9 dyads) were randomly selected and coded by the two independent study coders. Coders were not aware of coding reliability videos. Interrater reliability was determined for the categorical engagement phases codes on a second-by-second-basis. It was computed using mean Cohen’s κ [49] (κ = .82 for the infant codes; κ = .73 for the maternal codes). This interrater reliability is similar to those reported in previous studies [21, 50].

Matching states are defined as the mother and infant simultaneously exhibiting the same affective-behavioral state [15]. We concentrated on one type of match, positive social match. We assumed this coordinated state to be a sign for positive interaction [51]. A positive social match was defined as: The mother is in positive engagement or social monitor/positive vocalizations and the Infant is in positive engagement or social monitor.

The primary independent measure, the latency to interactive repair, was calculated as the average time interval from positive social match offset to positive social match onset; that is the average mismatch duration in seconds (play episode: M = 10.46 sec., SD = 8.57 sec., min = 1.45 sec. max = 36.56 sec.; reunion episode: M = 9.45 sec., SD = 5.31 sec., min = 1.08 sec. max = 20.79 sec.). Additional measures were relative time durations (for descriptive results multiplied by 100 %) of positive social matching states and infant self-comforting behaviors. That is the sum of seconds dyads were in the positive social matching states divided by the time of the FFSF episode (play episode: M = 17.38 %, SD = 18.29 %, min = 0.00%, max = 68.60 %; reunion episode: M = 15.86 %, SD = 13.63 %, min = 0.00 %, max = 55.00 %) and the sum of seconds in which infants engage in either oral or manual self-comforting behaviors divided by the time of the FFSF episode (play episode: M = 12.99 %, SD = 21.43 %, min = 0.00%, max = 94.00 %; still-face episode: M = 15.30 %, SD = 24.89 %, min = 0.00 %, max = 85.90 %; reunion episode: M = 10.97 %, SD = 17.27 %, min = 0.00 %, max = 76.70 %).

Assessment of infant Cortisol Reactivity

Sampling, storage, transport and analysis of cortisol samples took place according to standard protocols [52]. The limit of detection of the used assay was 0.1 – 15.0 ng/ml. Intra-assay variances were 5.95 % Vol. for 2.6 µg/100ml, 1.59% Vol. for 17µg/100ml and 4.62% for 26.6µg/100ml. For two infants of the sample the C₁, for one infant the C₂ and for n = 13 infants the C₃ value was missing. Reasons for these missing values were too small amounts of saliva, interruption of assessment by breast-feeding or by infants falling asleep. Average salivary cortisol values in the C₁ (M = 1.29 ng/ml, SD = 1.41 ng/ml, min = 0.10 ng/ml, max = 7.10 ng/ml), the C₂ (M = 1.30 ng/ml, SD = 1.34 ng/ml, min = 0.10 ng/ml, max = 6.50 ng/ml) and the C₃ measurement (M = 1.05 ng/ml, SD = 1.00 ng/ml, min = 0.10 ng/ml, max = 3.90 ng/ml) were comparable to normative values [53]. Following analytic procedures [54], area under the curve with respect to increase (AUC_I) was calculated as an index for infant cortisol reactivity. This measure is the integral of the curve resulting out of the three cortisol measures (C₁, C₂, C₃) and denotes the time distance between measurements in contrast to statistical tests for repeated measures. AUC_I is calculated with reference to the first value (C₁) and therefore measures the change over time. The AUC_I mean (M = −5.06 ng/ml x min., SD = 19.31 ng/ml x min., min = −46.00 ng/ml x min., max = 39.20 ng/ml x min.) was negative in our sample. This indicates that cortisol levels decreased from the baseline (C₁) to the C₂ and C₃ assessments. Given this finding we separated infants, whose AUC_I lay one SE (3.53 ng / ml x min.) above zero to estimate the rate of responders. The procedure revealed n = 9 responders (30 %) in the sample of infants that had cortisol values for all three points of measurement (n = 30). For the analyses, all infants were considered, if they were responder or not. The AUC_I were screened for outlying values defined as any value deviating more than three interquartile ranges from the median. No outlying values were identified. AUC_I was checked for associations to potential confounding variables (infant and maternal age, marital status, financial concerns, gestational age, PDA, breastfeeding, number of infants, APGAR values, daytime of assessment, time distance to and length of prior feeding and napping, count and length of daytime naps and nighttime awakes, sleeping arrangement and childcare). No significant associations to confounders were found (all p > .19). Consequently, we excluded these variables as potential confounders.

Prenatal Emotional Stress Index

The Prenatal Emotional Stress Index (PESI) is a self-report instrument, which assesses the emotional distress during pregnancy separate for each trimester [55]. It consists of 33 items, 11 items per pregnancy trimester. The items assess anxiety, sadness, joy, distress and tension of the mother via a visual analogous scale ranging from zero to 100 %. The scale value is computed by summing the 11 items (2 items with reversed polarity) for each trimester and averaging the sum by the number of items resulting in a PESI for each trimester ranging from 0 to 100. Cronbach’s α revealed excellent reliability for our data (α = .91 for the first, α = .92 for the second and α = .93 for the third trimester). The correlations between the first and second trimester (r = .87, p < .01), between the second and third trimester (r = .86, p < .01) and between the first and third trimester (r = .68, p < .01) reveal a medium to high inter-scale consistency. Mean Scores were M = 32.85 (SD = 26.61, min = 0,00, max = 92.27) for the first, M = 29.17 (SD = 22.24, min = 0,00, max = 83.18) for the second and M = 30.63 (SD = 22.88, min = 0,00, max = 85.45) for the third trimester. Compared to the descriptive results of Möhler and colleagues [55], the PESI is slightly increased. Their general mean in the non-clinical sample was M = 26.52 (SD = 14.29) compared to the general mean for our sample, which was M = 30.88 (SD = 22.23, min = 2.42, max = 86.97).

Statistical Analyses

We used the Statistical Package for Social Sciences (IBM^® SPSS^® v. 22.0.0.0) for all the analyses conducted in this study. Power-estimations for the confirmative analysis were computed using G-Power v. 3.1.9.2 [56, 57]. Before carrying out the main analyses, we evaluated if the list-wise case-exclusions as described in the participants section were valid for our data set. This was done using Little’s MCAR-test [58]. The MCAR-test evaluates if the missing-completely-at-random-condition (MCAR) is fulfilled. If non-significant, differences between excluded cases and the remaining sample are unlikely. In addition, missing values are unlikely to depend on third variables. Consequently, the MCAR-test was repeated for the study sample prior to the multiple imputation procedure. Furthermore, differences related to maternal age, gestation age, APGAR values, infant age, maternal education, number of children, and marital status between controls and their clinical counterparts and between males and females were explored (via t-tests, U-tests and χ²-tests) to ensure comparability between the groups. Generalized linear modelling (with robust maximum likelihood estimation) was used, since the distributions of interactive variables were significantly skewed (p < .01 in Kolmogorov-Smirnov and Shapiro-Wilk test). Especially in small samples and between unequally sized groups, the general linear model may lack sufficient robustness against the violation of mathematical assumptions (e.g. normal distribution) and thus may lead to progressive statistical decisions [59]. Primary hypotheses were all tested in one model, avoiding the cumulation of α-errors. Variables were not centered. Thus, B-weights are not standardized. However, as estimator for effect sizes $w^2(\frac{{\chi }^2}{N})$ was computed for significant results. According to Cohen’s conventions [60], w² = .01 are small, w² = .09 are medium-sized and w² = .25 are large effects. The critical α-error for the analyses was α = .05. Empirical p-values are one-tailed for the directional hypotheses. The α-errors of the additional analyses were not adjusted. To evaluate the independent contribution of additional variables of interest (e.g. positive social matching states, self-comforting behaviors and distress during pregnancy) to infant cortisol reactivity, a stepwise backward regression was chosen since a forward regressions bears the risk of not selecting independent variables with small but meaningful effects. In backward regression, variables are stepwise eliminated if they do not prove to be a significant parameter for the criterion among the remaining variables.

Results

Preliminary Data Analyses

For the MCAR-test, we considered the following variables: Socio-demographic data (e.g. age, infant gender), distress during pregnancy (PESI), interaction variables and matching data (ICEP-R), cortisol data (including its potential confounders), data assessed at birth (e.g. gestation age) and breast-feeding. The test was nonsignificant (χ² = 1,056.24, df = 1,089, p = .76); the list-wise case-exclusions were valid for our sample and the sub-population is representative of the larger sample. In order to ensure comparability between the clinical and the control group and between males and females, the distribution of demographic and birth-related variables (e.g. gestational age) were compared using t-, U- and χ²-tests. As demonstrated in Table 1 no differences were found between the groups.

We only had complete cortisol data for n = 30 infants (65.2 % of study sample), but the remaining infants had at least one valid measure. We estimated the missing values for these infants (n = 16, 34.8 % of study sample) using multiple imputations [61] with all variables analyzed in this study as predictors according to standard practice [62]. Multiple imputations are a valid method of estimating missing data, if the missing-completely-at-random-condition is fulfilled, as it was in the final sample (χ² = 706.24, df = 779, p = .97). We exceeded the recommendations [62] and estimated the missing values (n = 16) in 25 data sets (fully conditional, linear, two-way-interaction between categorical variables, maximum 50 iterations). Estimated cortisol values were restricted to the limit of detection of the cortisol assay (0.1 – 15.0 ng/ml). The analyses were done on a) the original data set and b) in each of the 25 completed data sets. The results of b) were then pooled. Consequently, two results are reported: one for the original data set and one pooled result for the imputed data sets. M and SD of pooled imputed values and the pooled sample after the imputation procedure (averaged over the 25 data-sets) can be found in Table 2. A visual analysis of the iteration process revealed no systematic variations of estimated values. Variation occurred within the scope of random variations.

Table 2.

Pooled imputation result (averaged over 25 data sets) of infant salivary cortisol (in ng/ml).

	imputed values				data after imputation (n = 46)
assessment	M	SD	min	max	M	SD	min	max
C1 (n = 2 imputed values)	1.59	0.67	1.11	2.06	1.31	1.40	0.15	7.10
C2 (n = 1 imputed value)	0.97	/	/	/	1.29	1.33	0.15	6.50
C3 (n = 13 imputed values)	1.14	0.82	0.23	3.02	1.07	0.96	0.15	3.95

Primary Analyses

We used Generalized Linear Modelling with robust maximum likelihood estimations to evaluate the primary hypotheses. The dependent variable was infant cortisol reactivity (AUC_I). We included maternal anxiety disorder (dummy-coded) and latency to repair as main effects in the model. Additionally, we included an anxiety disorder x latency to repair interaction term, to evaluate if a potential effect of latency to repair differs between the groups. The analysis was adjusted for a potential effect of infant gender (dummy coded).

As demonstrated in Table 3, there was no effect of anxiety disorder or infant gender. Additionally, the anxiety disorder x latency to repair interaction term was non-significant. The only significant main effect was for latency to repair (original data: p < .01; pooled data: p < .01): The longer the latency to repair, the higher the infants’ cortisol reactivity or the slower its decline. This effect was large for the original data (w² = .34) and medium-sized for the pooled data (w² = .14).

Table 3.

Generalized linear regression model on infant cortisol reactivity (AUC_I).

	Parameter	B	SE	Lower CI bound (95%)	Upper CI bound (95%)	Wald χ2 a	p
original data (n = 30)b	anxiety disorder	11.02	14.85	−18.08	40.13	0.55	.23
	latency to repair	3.12	0.97	1.21	5.02	10.28	< .01
	anxiety disorder x latency to repair	−1.55	1.16	−3.81	0.72	1.79	.09
	female gender	−4.29	6.34	−16.71	8.14	0.46	.25
	Intercept	−29.19	12.13	−52.96	−5.41	5.79	< .01
	Scale	253.67c	71.75	145.72	441.60	/	/
pooled data (n = 46)d	anxiety disorder	7.34	13.71	−19.54	34.23	0.37	.30
	latency to repair	2.01	0.85	0.34	3.68	5.92	< .01
	Anxiety Disorder x latency to repair	−1.20	1.05	−3.25	0.85	1.46	.13
	female gender	−6.82	5.64	−17.87	4.24	1.56	.11
	Intercept	−14.13	9.34	−32.43	4.17	2.62	.07
	Scale	316.42c	83.38	152.55	480.28	/	/

Notes.

^aFor pooled analyses: averaged over original and imputed data sets;

^bLikelihood-ratio omnibus test compares the fitted model against the intercept-only model, χ² = 12.33, p = .02;

^cMaximum likelihood estimate;

^dAverage likelihood-ratio χ² = 7.76, average p = .12

The power for this analysis was approximated for linear multiple regressions. The chance of finding a large effect ( f ² = .35) of single coefficients in our sample was 1-β = .99 (1-β = .93 for original data). Medium-sized effects ( f ² = .15) could be detected with a power of 1-β = .83 (1-β = .66 for original data). Only small effects ( f ² = .02) could not be sufficiently detected in our sample (1-β = .24 for pooled data; 1-β = .19 for original data).

Additional Analyses

Pearson-correlations were carried out to evaluate the associations of additional interactive variables in different episodes of the FFSF (i.e. positive social matching states and infant self-comforting behaviors) as well as the association of distress during pregnancy with infant cortisol reactivity (AUC_I) in the original and pooled data sets (Table 4). Here, we exclusively present the pooled results, since they were more conservative. The correlation between cortisol reactivity and latency to repair was positive and significant during the play episode. The longer the latency to repair the higher infants’ subsequent cortisol secretion or the slower its decline. Additionally, the relative time durations of positive social matches and infant self-comforting during the reunion episode were significantly associated to cortisol reactivity. The longer the duration of the matches and the less infants engage in self-comforting behaviors the lower their cortisol reactivity or the stronger its decline. Furthermore, maternal distress during the first trimester of pregnancy was significantly associated to infant cortisol reactivity between the third and fourth month of infant age. The more maternal distress the higher the cortisol reactivity or the weaker its decline.

Table 4.

Pearson-correlations to infant cortisol reactivity (AUC_I)

		latency to repair (play)	positive social matches (play)	positive social matches (reunion)	self-comforting (play)	self-comforting (still-face)	self-comforting (reunion)	PESI 1st trimester	PESI 2nd trimester	PESI 3rd trimester
original data (n = 30)	r	.325	−.270	−.399	.197	.058	.332	.450	.426	.402
	p	< .05	.07	.01	.15	.38	.04	.01	.02	.02
pooled data (n = 46)	r	.309	−.214	−.298	.249	.200	.380	.315	.275	.272
	p	.03	.08	.03	.08	.11	.03	.04	.05	.05

To evaluate the unique and independent relation of latency to repair during the reunion episode with infant cortisol reactivity, we carried out a stepwise backward regression (generalized linear models) with latency to repair and all other significant associations from the correlation analyses in the pooled data as independent variables and infant cortisol reactivity (AUC_I) as the criterion. Furthermore, we were interested, if the effect of prepartum emotional distress during the first trimester of pregnancy was moderated by postpartum dyadic interaction or self-comforting behaviors. Consequently, we integrated interaction terms between the “PESI 1^st trimester and self-comforting behaviors, positive matches and latency to repair. Table 5 presents only the pooled analyses of the final regression step (step 7). Steps 1 to 6 are reported in Appendix A. The final model consisted of self-comforting behaviors during the reunion episode (w² = .25), latency to repair in the play (w² = .13) and the interaction term between maternal perceived prepartum distress during the first trimester of pregnancy and latency to repair during the reunion episode (w² = .27). The effects of self-comforting behaviors and the interaction term were large [60]: See comment on cohen's conventions above. Relative time duration of positive social matches during the reunion episode and latency to repair were significantly intercorrelated (play: p = .02; reunion: p < .01). Since all other independent variables in these analyses were not intercorrelated (all p > .28), multicollinearity can be excluded for the final model. Additionally, these analyses suggest that latency to repair during the reunion episode moderates the effect of distress during the first trimester of pregnancy.

Table 5.

Final generalized linear regression model (Step 7 of backward procedure) on infant cortisol reactivity (AUCI).

	Parameter	B	SE	Lower CI bound (95%)	Upper CI bound (95%)	Wald χ2 a	p
Step 7b	self-comforting (reunion)	29.10	15.87	−2.76	60.97	11.64	.04
	latency to repair (play)	0.65	0.28	0.11	1.20	5.98	< .01
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	11.90	< .01
	intercept	−18.03	5.85	−29.50	−6.57	9.61	< .01
	scale	206.24c	57.44	93.59	318.89	/	/

Notes.

^aAveraged over original and imputed data sets;

^bLikelihood-ratio omnibus test compares the fitted model against the intercept-only model, average χ² = 15.72, average p = .002;

^cMaximum likelihood estimate

Discussion

To our knowledge, this is the first study that supports the hypothesis that quicker reparation of dyadic mismatching states in early mother-infant interactions provides better psychobiological stress regulation in infants. Micro-analytic data reveal that the latency to positive social matches is significantly associated to infant cortisol reactivity. Due to missing values in cortisol data, especially at the second post FFSF assessment, we decided to impute missing data. The association between latency to repair and infant cortisol reactivity was significant in both the original and the pooled data. Multiple imputations are a valid method, if missing values do not depend from any other variable in the data set and if all variables used in the analyses were used as predictors. Both conditions were fulfilled. Nevertheless, for the main analyses, results out of both, the original and the pooled data were presented, leading to the same inferences. That is, the shorter the duration of mismatch, the lower the infants’ salivary cortisol output throughout the experimental paradigm. These results are in line with the predictions of Tronick’s Mutual Regulation Model, which emphasizes the critical regulatory function of reparation [15]. In addition, and as suggested in another study [39], these results support the idea that affective-behavioral regulation between caregiver and infant promotes better infant regulation at other somatic regulatory levels, such as the HPA-axis.

For the confirmatory analysis, neither maternal anxiety disorder was found to be significantly associated to infant cortisol reactivity nor was the association between latency to repair and infant cortisol reactivity different between the groups. This lack of findings contradicts recent research [33–35], but at least one other study [63] found that prepartum maternal anxiety disorder was not associated with infant physiological regulation. Though somewhat unexpected, our finding for mothers with anxiety disorders is in line with research suggesting that some women with affective disorders, such as major depression, interact with their infants in relatively sensitive ways compared to other women with similar levels of depression [64]. In addition in a critical study [65], it was observed that parenting difficulties in mothers with an anxiety disorder were only evident to a disorder-specific challenge. Of course, we might have failed to discover small effects or for the original data set medium-sized effects, since the power was low. Furthermore, our clinical group was heterogeneous, since we did not focus on specific anxiety disorders, which might have add to this null-finding. This lack of consistency requires further research in larger and homogeneous samples. Nevertheless, the results reported in this study indicate that the actual quality of the reparatory process plays a central role for infant stress reactivity and that simply using diagnostic status as a marker of problematic infant regulation might not be adequate. Infants react to what they are experiencing and not to the diagnostic status of their mothers and maternal diagnoses are hardly related in a one-to-one fashion to what the mothers actually do.

It must be noted that interactive reparations can be initiated mutually within a dyad [66]. Since we did not assess who initiated the interactive repair, we cannot infer that the mothers initiated the reparations. In another study of our group, however, it was demonstrated that latency to positive social matches is significantly associated to macro-analytical measures of maternal sensitivity [11], suggesting that it is maternal behavior that underlies reparations. Nevertheless, future studies might use time series analyses to determine the extent to which the caregiver or the infant are initiating reparations and compare those findings to more macro analytic measures of maternal sensitivity.

Furthermore, the mean durations of interactive reparation were unexpectedly extended with approximately 10 sec. for both the play and the reunion episode. Weinberg and colleagues [67] report marked lower mismatch durations of 2 to 6 sec., while Reck and colleagues [21] found comparable latencies of interactive reparation. This might be explained by cultural differences between American and German samples or be due to differences in used methods and instruments. Additionally, it may be possible, that we failed to observe some interactive matches due to the 1 sec. time interval used for the observation. If matches occurred below this time unit, we would have overestimated the mean mismatch duration. Future analyses could try to determine cultural influences on interactive reparation and use a higher time resolution in the observation of interactive matches.

We cannot draw causal conclusions between interactive reparation and infant cortisol reactivity. It is possible that high cortisol reactivity makes reparation to a positive dyadic state more difficult. Infants who respond to the FFSF with a higher release of cortisol might be less able to make appropriate adjustments to maternal signals because their arousal level narrows their perception or disrupts their behavior [68]. Thus the finding raises the question as to whether behavior is driving the cortisol response or if the cortisol response is driving the behavior. More than likely, the two processes dynamically interact, but this question is beyond the aims of this study. To clarify this issue, future studies require longitudinal data regarding behavior and cortisol as well as a causal analytic approach.

There were marked individual differences in cortisol reactivity for the infants in this study. For the whole sample there was no mean increase in cortisol following the FFSF. This might be due to the fact, that infant salivary cortisol samples were only taken prior to, immediately after and 20 minutes after the FFSF. Consequently, we may not have had a full coverage of the possible cortisol peak times, though many studies have used similar intervals as demonstrated in the review of Gunnar and colleagues [69]. This review also indicates that on average psychological stimuli (such as separation) do not provoke cortisol reactivity in young infants. Hayley and colleagues [70] were able to find an increase in cortisol-reactivity following the double-FFSF and an observation interval of 30 min. post stressor. As Haley and colleagues were able to replicate their finding [71], the double-FFSF and this sampling interval might be appropriate for future studies on infant cortisol reactivity.

Additionally, it might have been conceivable that the missing values account for the lack of cortisol increase. Nevertheless, this is unlikely since the missing values were random. However, a finding of a non-response in the mean value may be surprising given the established stressful nature of the FFSF. Nevertheless, some infants respond with an increase in cortisol to blood draws, whereas others respond to undressing, weighing and length measurement prior to the draw [72]. A low to medium rate of infant cortisol responders (30 % for our original data) and thus a decrease in cortisol means often is found in infant and child stress research [2, 69]. It must be noted that the lack of reactivity does not imply that the measurement of cortisol reactivity in response to psychological stressors is not meaningful. Rather, it has been argued that the individual differences might bring to light factors that account for the individual differences as well as potential risk factors that may adversely affect infant development [2, 69]. Accordingly, the authors of the AUC_I procedure recommend proceeding with analyses even in the case of negative values, since it may be even important to explain a decrease. Respective indices must then be interpreted as an index of decrease over time [54]. Moreover, a dampening of cortisol-responses to stressors in rodents and humans during early development [3] might play a role for our results. Although the reasons and duration of this dampening period still is unknown, there are many factors affecting stress reactivity. These include genetic influences [73], temperament differences [74], age related changes [2], individual differences in sensitivity to the nature of the stimuli and contexts [69], and a sculpting of stress reactivity by interactive history [75].

Our findings lend some support to the sculpting hypothesis [75]. In particular, that interactive reparation may play a role in maintaining low cortisol activity during this period. While we recognize that the observations of the interactions in this study were brief, it seems reasonable to assume that the quality of reparation is related to the dyad’s typical patterns of interaction as has been found for measures of sensitivity [51]. Thus, the differences in cortisol reactivity seen in this study may be chronically experienced and affect the functioning of the HPA-axis. An early and chronic dysregulation of the HPA axis might account for the risk to develop mental or affective disorders later in life [32]. Such an interpretation is not exclusive of the role of other factors, but fits with the hypothesis that quotidian stressors, especially micro-stressors such as mismatches, sculpt the regulating systems and lead to resilience or vulnerability depending on the quality of their resolution [10]. Nevertheless, the answer to this question was beyond the scope of this study. Future research might integrate repeated measures of caregiver-infant interaction over the course of development in different contexts to support this hypothesis.

The results of the additional analyses further revealed that the relative time duration of the positive social matches and infant self-comforting during the reunion episode, as well as perceived maternal distress during the first trimester of pregnancy were associated with the infants’ stress reactivity. However, in the backward regression procedure, the relative time duration of the positive social matches was no longer found to make an independent contribution in the used variable set. That is, only being in positive dyadic states for longer durations seems insufficient for affecting infant cortisol reactivity. Cortisol reactivity rather appears to need the flexible interplay between mismatching and positive matching states as indexed by latency to repair.

An effect of prepartum distress on infant cortisol reactivity is well-established [76]. These observations refer to the phenomena of fetal programming [77–79]. Prepartum influences on the fetal HPA-axis are said to occur via maternal cortisol overcoming the placental barrier [80]. Processes, such as methylation of the CpG-regions in the promotor region of the glucocorticoid-receptor-gene [78, 79], are suggested to be involved in these phenomena. Additionally, these influences might be especially marked during the first two trimesters of pregnancy, as the cortisol degrading 11-β-HSD enzyme is not expressed in the fetus that early [81, 82]. According to this assumption, it was the distress during the first trimester of pregnancy, which was associated to infant cortisol reactivity, underpinning the validity of our results. Of course, it would have been preferable to use maternal cortisol data as predictor for infant cortisol reactivity. However, according to studies that suggest epigenetic programming to be modified by postpartum caregiving [63, 79], we found that the association between distress during the first trimester of pregnancy and cortisol reactivity was moderated by postpartum interactive quality. In line with this research, the result suggests prepartum alterations of the fetal HPA-axis by maternal distress during pregnancy might be overcome by a well-adapted postpartum caregiver-infant interaction. Future research should involve longitudinal data including prepartum assessments of maternal cortisol as well as epigenetic data. Such studies could improve our understanding of the interplay between fetal programming and postpartum interactive history for infant stress reactivity.

In the final regression model, latency to repair in the play was a significant medium-sized parameter for infant cortisol reactivity besides the interaction term between maternal prepartum distress during the first trimester of pregnancy and latency to repair during the reunion episode. Surprisingly, latency to repair between play and reunion were not intercorrelated and both latency to repair during the play episode and the interaction term independently explained variance of infant cortisol reactivity. Apart from low power, this might underline the macro-temporal divergence of contexts between the play and the reunion episode of the FFSF [67]. Perhaps the function of latency to repair might be different depending on the FFSF episode. In the play episode, interactive reparation might reflect the regulation of typical micro-temporal stressors (mismatches) while in the reunion episode it might have the function of regulating the distress caused by macro-temporal disengagement of the still-face. This idea could explain why maternal prepartum distress during the first trimester of pregnancy only interacted with latency to repair during the reunion episode. It might be that solely the experience of regulatory success after a stressful event (such as the still-face) can counteract potential negative influences of maternal prepartum distress on the infants’ stress-regulatory system, the HPA axis. Future studies could examine this idea by characterizing dyads depending on their level of interactive reparation in both the play and the reunion episode as well as by varying the experimental paradigm (FFSF vs. free play).

Additionally, only self-comforting behaviors during the reunion episode after the socio-emotional stressor of maternal disengagement in the still-face episode remained as a negative parameter along with latency to repair and the interaction between prepartum distress during the first trimester of pregnancy and latency to repair during the reunion episode. Though causal conclusions cannot be drawn, this finding might underline that self-comforting behaviors are not fully effective in regulating distress [17]. Rather they may be a sign that the infants’ self-regulating strategies are overtaxed [19]. Future analyses should investigate infant self-comforting behaviors more fully regarding their change during the FFSF and their associations to markers of infant distress. It would be interesting to know, if self-comforting behaviors significantly increase during the still-face and if this increase is associated to cortisol reactivity or if cortisol values taken prior to the experiment predict the increase of self-comforting behaviors. Furthermore, the role of maternal anxiety disorder regarding infant self-comforting behaviors remains unclear and should be analyzed in future studies.

It was also surprising that latency to repair and self-comforting behaviors were not negatively intercorrelated given the reported developmental shift from self-comforting behaviors to dyadic regulation [18]. Of course, given the small sample size an interpretation of null-findings is limited. Nevertheless, it might be that self-comforting behaviors decrease later than four months of age, if dyadic regulation is well established and both regulation strategies might exist in parallel. Future studies could address this issue by repeated observations throughout the first year of life.

Summary of Limitations

The validity of null-findings (especially for small and medium-sized effects) cannot be fully evaluated given the small sample size. The clinical group was heterogeneous, since we did not concentrate on specific anxiety disorders. The data assessment for the main finding was cross-sectional. Consequently, we cannot draw causal conclusions. Additionally, we cannot draw conclusions on who initiated the reparation, since we did not apply lagged time-series analyses. Infant salivary cortisol samples were only taken prior to, immediately after and twenty minutes after the FFSF. Consequently we may not have had full coverage of possible peak cortisol reactivity times, which may in part account for the negative mean cortisol reactivity. As we had no maternal cortisol data from the prepartum period, it was not possible to validate the subjective and retrospective reports of perceived distress during pregnancy. Missing values in the cortisol assessment were imputed, which potentially can decrease the external validity for the pooled data. Our sample contains a greater proportion of controls, academic mothers and female infants, which also might decrease external validity and the discriminatory power within these variables. For the additional analyses, the findings must be interpreted cautiously because of α-error cumulation. However, for those analyses, we exclusively used the results of the imputed data sets and generalized linear modelling with robust estimators, which was more conservative. Moreover, the results might add important aspects to future research.

Conclusions

The results of this study suggest that latency to repair is associated to stress reactivity in infants as assessed with salivary cortisol. This association supports the hypothesis that the reparation of dyadic mismatching states is associated to infants’ psychophysiological regulation [10, 15]. According to the literature [32, 83, 84], individual differences in psychobiological responsiveness to stress are associated with increased susceptibility to behavioral problems and disorders in children and adolescents. Hence, preventive mother-infant interventions targeted at changing maladaptive forms of dyadic interaction to ones that more effectively reduce infant distress might be useful. As interactive reparation occurs in a clearly detectable time range (seconds), video interventions [85, 86] may be useful for increasing the flexibility in the dyadic interplay between mismatching and positive matching states. In particular, our results suggest that central elements of an intervention rather than focusing on maternal disorder or just positive matching states might better focus on reparation of mismatching states and the recognition of infant signs of distress (e.g. self-comforting behaviors).

Appendices

Appendix A.

Steps 1 – 6 of backward generalized linear regression model on infant cortisol reactivity (AUC_I).

	Parameter	B	SE	Lower CI bound (95%)	Upper CI bound (95%)	Wald χ2 a	p
Step 1b	PESI 1st trimester	0.00	0.04	−0.07	0.08	0.12	.47
	self-comforting (reunion)	35.20	20.98	−7.49	77.88	22.09	.05
	positive social matches (reunion)	26.21	82.65	−135.79	188.21	0.18	.38
	latency to repair (play)	0.60	0.79	−0.94	2.14	0.73	.22
	latency to repair (reunion)	1.11	1.19	−1.22	3.45	1.14	.18
	PESI 1st trimester x self-comforting	−0.01	0.03	−0.08	0.05	3.70	.34
	PESI 1st trimester x positive matches	0.01	0.12	−0.22	0.24	0.10	.46
	PESI 1st trimester x repair (play)	0.00	0.00	0.00	0.00	0.47	.35
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.01	0.53	.33
	intercept	−32.16	29.35	−89.69	25.36	1.40	.14
	scale	179.91c	49.81	82.24	277.59	/	/
Step 2d	self-comforting (reunion)	35.07	20.94	−7.48	77.62	20.17	.05
	positive social matches (reunion)	23.13	54.69	−84.08	130.33	0.24	.34
	latency to repair (play)	0.58	0.60	−0.59	1.75	1.10	.17
	latency to repair (reunion)	1.05	0.63	−0.19	2.29	4.36	< .05
	PESI 1st trimester x self-comforting	−0.01	0.03	−0.08	0.05	1.63	.36
	PESI 1st trimester x positive matches	0.02	0.07	−0.11	0.15	0.18	.39
	PESI 1st trimester x repair (play)	0.00	0.00	0.00	0.00	0.52	.32
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	0.82	.27
	intercept	−30.80	14.60	−59.41	−2.18	4.76	.02
	scale	180.28c	49.91	82.41	278.15	/	/
Step 3e	self-comforting (reunion)	34.43	20.80	−8.04	76.90	28.63	.05
	positive social matches (reunion)	29.56	39.30	−47.48	106.59	0.66	.23
	latency to repair (play)	0.55	0.61	−0.65	1.75	1.00	.19
	latency to repair (reunion)	1.01	0.63	−0.23	2.25	4.03	.05
	PESI 1st trimester x self-comforting	−0.01	0.03	−0.08	0.05	1.59	.36
	PESI 1st trimester x repair (play)	0.00	0.00	0.00	0.00	0.62	.29
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	0.89	.25
	intercept	−30.8	14.51	−59.24	−2.37	4.85	.02
	scale	181.31c	50.40	82.48	280.14	/	/
Step 4f	self-comforting (reunion)	31.54	15.81	−0.44	63.52	20.11	.03
	positive social matches (reunion)	34.88	34.15	−32.10	101.86	1.36	.15
	latency to repair (play)	0.62	0.57	−0.50	1.73	1.69	.14
	latency to repair (reunion)	1.09	0.58	−0.04	2.23	4.95	.03
	PESI 1st trimester x repair (play)	0.00	0.00	0.00	0.00	0.74	.33
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	0.95	.24
	intercept	−32.79	12.54	−57.38	−8.20	7.97	< .01
	scale	185.75c	51.68	84.40	287.11	/	/
Step 5g	self-comforting (reunion)	31.95	15.48	0.63	63.27	22.70	.02
	positive social matches (reunion)	37.51	31.27	−23.83	98.84	1.79	.12
	latency to repair (play)	0.77	0.34	0.11	1.44	5.61	.01
	latency to repair (reunion)	0.99	0.54	−0.08	2.05	4.07	.04
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	9.56	.01
	intercept	−33.98	10.97	−55.506	−12.46	11.11	< .01
	scale	188.17c	52.55	85.11	291.24	/	/
Step 6h	self-comforting (reunion)	29.92	15.84	−1.90	61.74	12.49	.03
	latency to repair (play)	0.65	0.31	0.05	1.26	4.70	.02
	latency to repair (reunion)	0.59	0.49	-0.37	1.55	1.75	.11
	PESI 1st trimester x repair (reunion)	0.00	0.00	0.00	0.00	6.84	.03
	intercept	−22.22	6.94	−35.81	−8.63	10.42	< .01
	scale	199.59c	55.68	90.39	308.78	/	/

Notes.

^aAveraged over original and imputed data sets;

^bLikelihood-ratio omnibus test compares the fitted model against the intercept-only model, average χ² = 19.74, average p = .03;

^cMaximum likelihood estimate;

^dAverage likelihood-ratio χ² = 19.63, average p = .02;

^eAverage likelihood-ratio χ² = 19.48, average p = .01;

^fAverage likelihood-ratio χ² = 18.78, average p = .007;

^gAverage likelihood-ratio χ² = 18.36, average p = .005;

^hAverage likelihood-ratio χ² = 16.65, average p = .004