Participants

The participants in the study were 32 mothers and their children, all from Lima, Peru. The mothers’ average age was 29.6 years (SD = 6.28, Min = 19, Max = 44). In terms of educational attainment, 43% had completed technical or college-level higher education, 53.1% were high school graduates, and 3.1% only had primary education. In terms of their domestic situation, 84.4% were married or had a stable partner, and 81% were mainly homemakers. The participating children were 20 boys (62.50%) and 12 girls (37.50%), and their ages were between 8 and 10 months 29 days (M = 8.91; SD = 0.96). In all cases, the main caregiver was the mother. Maternal reports verified that none of the children presented an evident pathology of development or health, autism, Down and related syndromes; or prematurity.

The study was approved by the financing university’s Ethics Committee. The selection of participating dyads was based on convenience: participants were selected from among those who took their children to a government-run health center. The study was explained to each mother and they were invited to participate; only one of the mothers was referred by one of the researchers’ contacts, a resident of the area where the study was conducted. Mothers who agreed to participate voluntarily signed an informed consent form describing the procedure to be followed, including benefits and possible risks; the voluntary nature of their participation was guaranteed, with the option of withdrawing at any time, as was the anonymity and confidentiality of the results.

At the end of their experience, participating mothers received an economic reward equivalent to U.S. $50.00 and a DVD recording of the observation. They were offered a psychological counselling session with two members of the research team, focused on any parenting issues they may have previously raised as concerns or on any aspects of the interaction that emerged during the observation. The research team also had a directory of professionals for the referral of possible at-risk cases, if necessary.

Instruments

Massie-Campbell Attachment During Stress Scale (ADS). The ADS is an observation guide designed for the evaluation of child-caregiver interactions in situations of moderate stress, such as a medical visit or episodes of separation (Massie & Campbell, 1983). This instrument assesses the child’s and the mother’s or caregiver’s behaviors independently. The intensity and frequency of six behaviors are evaluated for each: gaze, vocalization, support, touch, affection, and proximity seeking. Each behavior is scored on a five-level continuum, ranging from its absence to its strong expression. A score for the dyad’s interaction is assigned. Additionally, given that the original handbook does not specify a procedure for a single score, ^{Cárcamo et al. (2014}) adapted one where the presence of scores of 1 or 2 in more than half of the items is an indicator of insecure avoidant attachment; the presence of scores of 3 and 4 in more than half of the items is an indicator of secure attachment; and scores of 5 in most of the items is an indicator of ambivalent attachment.

For this study, the evaluation of mother-child interactions using the ADS was based on the observation of video recordings of structured procedures carried out as part of an earlier study. The procedure consisted of five different times. It began with 5 minutes of mother and child free play with toys placed in the space. Then there was a time when the mother left the room and the child was alone with the toys for two minutes approximately. In the third time, the mother returned to the room and interacted with the child for two more minutes, until a new person, not previously seen by the dyad, entered and removed all the toys except for one they had not played with. Finally, the dyad continued to interact for two more minutes. The procedure lasted approximately 11 minutes.

For the scoring of behaviors observed using the ADS, this study used the system proposed by ^{Cárcamo et al. (2014}) First, each of the behaviors was scored on a scale of 1, 2, 3, 4, 5 or not observed. Then, each behavior was classified as secure or insecure according to the following procedure: behaviors with scores of 3 or 4 were classified as secure and those with scores of 1, 2 or 5 were classified as insecure. Finally, dyads were classified as secure when 50% or more of their behaviors were secure; otherwise, they were classified as insecure.

The videos were rated by a group of five coders, experts in the subject of attachment, who received specialized training and rated 16 training videos. Using a prevalence-adjusted and bias-adjusted Kappa coefficient (PABAK), their interobserver reliability was moderate, .56.

To ensure the reliability of the scores in this study, two additional independent observers rated 30% of the videos. The PABAK inter-rater reliability was moderate (.45) for the global classification of the dyads, and adequate or fair (M = .36; Min = .09, Max = .64) for the classification by behavior.

Attachment Q-Set (AQS) version 3.0.^{Waters' AQS (1995}) is an observation instrument that evaluates the organization of secure base behavior with the purpose of assessing attachment security. This study used ^{Nóblega’s linguistic adaptation (Nóblega 2012}) of the Spanish version (^{Posada et al., 1999}). The AQS uses the Q Sort methodology and consists of 90 items that describe different behaviors of the child. As detailed in the introduction, this instrument has adequate psychometric characteristics for the evaluation of children 12 to 69 months old (^{Cadman et al., 2017}; Nóblega, 2012; Nóblega et al., 2016; Posada et al., 2013; ^{Posada et al., 2016}; ^{Solomon & George, 2016}; ^{Vaughn et al., 2007}). Also, it was used in the Peruvian context in a previous study with children under 12 months (Nóblega et al., 2016).

For the research from which this study is derived, free mother-child interaction was recorded during approximately an hour and two independent evaluators rated the resulting video. To improve the quality of the scores obtained for maternal behavior, the rating of the observations occurred in two stages, following the procedures of previous studies (^{Bárrig, 2004}; ^{Nóblega, 2012}; ^{Posada et al., 2004}; Posada et al., 1995; ^{Posada et al., 2002}; ^{Vaughn et al., 2007}). Average inter-rater reliability for the first stage was good, .83 (SD = 0.07, Min = .65 and Max = .92). In the second stage, items in which there was a high discrepancy between the ratings given by the observers (difference of more than 3 points) were identified for each case. The team discussed these items on the basis of behavioral referents observed during the visit, arriving at a shared description of the child's behavior. The final scores consisted of an average of the scores given by the observers to each item.

According to the procedure proposed for the instrument, the assurance of the child's attachment was calculated from the correlation between the scores given by the observer to each of the 90 items and the scores that make up a criterion agreed upon by experts regarding the scores of a hypothetical ideally secure child (^{Cadman et al., 2017}). This theoretical score is called a security criterion and has been validated in different contexts, including the Peruvian one (^{Cassibba et al., 2000}; ^{Nóblega, 2012}; ^{Posada et al., 1995}

There is no cut-off point that determines an attachment classification based on the scores obtained with the AQS (^{Van IJzendoorn et al., 2004}). However, a study by ^{Van Bakel & Riksen-Walraven (2004}) with 129 Dutch dyads found that the mean attachment security in children classified as secure was significantly higher than that in children classified as insecure on the basis of their SSP assessment (M _{secure children} = .32, SD _{secure children} = .25, M _{insecure children} = .12, SD _{insecure children} = .24; t = 4.34, p < .01). Similarly, Van IJzendoorn et al. (2004) reported that the mean of the security scores in the typical population differed significantly from the mean in the clinical population (M _{typical population} = .32, _{SD typical population} = .16; M _{clinical population} = .21, SD _{clinical population} =.08; t (297.20) = 16.06, p < .01). On their part, ^{Geenen and Conveleyn (2014}) differentiate secure and insecure attachment at a score of .32 in the AQS. Based on this evidence, those dyads with AQS scores above .32 can be considered secure.

Maternal Behavior Q-Sort (MBQS) version 2.1. The MBQS (^{Pederson & Moran, 1995}) is an observation instrument that assesses maternal sensitivity using the Q-Sort methodology to describe maternal behavior for 90 items. It has been validated in a variety of contexts (^{Posada, 2013}; ^{Posada et al., 2016}). The present study used the version adapted to the LatinAmerican context by Posada et al. (2002). Observations of maternal sensitivity were made simultaneously with observations of child behavior using the AQS. In order to improve the quality of the scores obtained for maternal behavior, the two-stage procedure described for the AQS was also carried out in this case. Two coders worked independently from each other and from their AQS counterparts. In the first stage, the scores of both observers were correlated, resulting in a good average inter-observer reliability, .82 (SD = 0.09). In 31 cases, the minimum obtained was .61 and the maximum, .92. In one case, a moderate reliability of .59 was obtained, and for this reason, a third evaluator was asked to code the recording of the interaction; after this procedure, inter-observer reliability for this video was .75. Following the pre-established procedure, a global sensitivity score was obtained for each mother based on the correlation between their score for each item and the sensitivity criteria established by expert consensus. (Pederson & Moran, 1995).

Procedure

This trans-sectional quantitative study derives from a larger project [ATI (Uruguay) & the Pontificia Universidad Javeriana of Colombia] and was carried out in a variety of contexts, exploring the relationship between attachment precursors and maternal sensitivity in children under one year. In that larger research project, two meetings were held with the participants. The first meeting took place in the mother's house, where participants signed the informed consent form and their socio-demographic data were collected. Next, the dyad’s interaction was observed and recorded for one hour, and then the mother was asked to play with her child in a fixed space for five minutes. Finally, the mother completed a narrative technique to assess attachment representations. The second meeting was held in a Gesell chamber. The mothers followed a structured procedure of separations and encounters, and these were recorded. After the procedure was completed, information was collected on their rearing practices and the use of lullabies in their child’s upbringing. Finally, the psychological counselling session described above was conducted.

This study used attachment-security scores (AQS) and maternal-sensitivity scores (MBQS) resulting from the observation of mother-child interactions in a home setting, and video recordings of the same dyads in a laboratory context, rated using the ADS.

Data Analysis

The IBM SPSS Statistics 22 data package was used to perform the statistical analyses in this study. Following recommendations by ^{Cárcamo et al. (2014}), it was decided to use only the infant scale for all the analyses.

First, the data set were analyzed for normality of distribution and homogeneity of variance. The skewnessindicators and the kurtosis, as well as a Shapiro-Wilk test and the normality graphs, led to the conclusion that the AQS’s continuous security scores (SW (32) = .949; Kurtosis = -.094; Skewness = -.675) and maternal sensitivity scores (SW (32) = .905; Kurtosis = .125; Skewness = -.880) were normal. At a specific level and based on the same indicators, it was concluded that the distribution of security scores in groups of children classified as secure and insecure was normal (SW(25) _{sec, secure children group} = .938; Kurtosis _{sec, secure children group} = 1.018; Skewness _{sec, secure children group} = -.975) vs. insecure (SW(7) _{sec, insecure children group} = .913; Kurtosis _{sec, insecure children group} = -1.189; Skewness _{sec, insecure children group} = -.269), as was the distribution of scores of maternal sensitivity for mothers of secure children, established on the basis of the ADS scores (SW(25) _{sen, secure children group} = .873; Kurtosis _{sen, secure children group} =.122; Skewness _{sen, secure children group} = -1.010) and insecure (SW(7) _{sen, insecure children group} = .939; Kurtosis _{sen, insecure children group} = 1.997; Skewness _{sen, insecure children group} = -.007). For evidence of the convergent validity of the ADS, the relationship between the ADS classification of secure and insecure attachment and AQS security scores was analyzed, using a point-biserial correlation. Similarly, a Student t-test mean comparison analysis of the AQS and ADS scores for children classified as a secure and insecure was carried out. Next, Cramer's Chi-Square and V tests were performed to examine the concordance between the secure and insecure classifications in ADS and AQS. For a classification of secure vs. insecure attachment using the AQS scores, the cut-off point was .32, which earlier studies have reported as the security mean in different non-clinical populations (^{Geenen & Conveleyn, 2014}; ^{Van Bakel & Riksen-Walraven, 2004}; ^{Van IJzendoorn et al., 2004}).

To demonstrate the scale’s discriminant validity, the relationship between the ADS and MBQS maternal sensitivity scores were analyzed, using the Student's t-test to contrast the means and compare the secure and insecure groups’ maternal sensitivity scores.

Based on the results of the discrepancy analysis, the sensitivity, specificity, positive predictive value, and negative predictive value of the ADS scale were examined, using the AQS classifications as reference criteria.

Finally, to identify a way to classify children with secure attachment that optimizes the sensitivity and specificity of the ADS, a Receiver Operating Characteristics (ROC) analysis was performed. For this to be achieved, a continuous score for the ADS was calculated by adding the scores of the 7 items converted to secure (1 point) or to insecure (0 points). Youden’s index to define an optimal cut-off point was used.

Descriptive statistics

Assessed using theAQS, the mean for attachment security was .195 (SD =.235; Min = -.358; Max = .586). Based on the ADS’s dichotomous classification, 78.1% of participating children were classified into the secure-attachment group and 21.9% into the insecure-attachment group. A media of .543 (SD =.242; Min = -.021; Max = .834) was found for maternal sensitivity using the MBQS.

Evidence of Convergent Validity: AQS and ADS Classifications

This study found a positive association between theADS’s dichotomous classification (secure attachment insecure attachment) and the AQS scores (rpb = .408, p = .021). This means that the greater the attachment security assessed with the AQS, the more likely the child will be classified as secure based on his or her ADS scores.

Based on their ADS score, children were classified in two groups, secure and insecure attachment, and a mean comparison was performed to identify any differences. The results showed that children classified as insecure had significantly lower security scores in the AQS (M _{secure children} = .25, SD secure children = .22, M _{insecure children} = .02, SD _{insecure children} = .22, t(30) = -2.446, p = .02, d = 0.89; 95% CI [ -.419; -.037]).

Similarly, a significant association between the dichotomous classifications of secure and insecure attachment provided by the ADS and the AQS [χ² (1, N = 32) = 4.69, p = .03, V = .383] was found. As shown in Table 1, in general, 56.3% of participants (18 children) have the same classification in both instruments. In addition, all participants classified as insecure in the AQS have the same classification in the ADS, and all participants classified as insecure by the ADS had scores compatible with such categorization in the AQS. However, it should be noted that 56% of the participants classified as secure by the ADS (14 children) were classified as insecure by the AQS.

Table 1 Cross Table ADS Classification and AQS Classification secure insecure (N = 32) 

Evidence of Discriminant Validity: Comparison of Maternal Sensitivity in Mothers of Secure and Insecure Children

The maternal sensitivity scores of mothers of children in the secure attachment group as determined by the ADS were not significantly higher than those of mothers of children in the insecure attachment group (M _{mothers of secure children} = .55, SD _{mothers of secure children} = .26, M _{mothers of insecure children} = .52, SD _{mothers of insecure children} = .19, t(30) = -.34, p = .74, d = -0.123; 95% CI [-.249;.179]).

Sensitivity, Specificity, and Predictive Values of ADS

The sensitivity of the ADS scale was 33.3% with a CI 95% CI = [15.48; 56.90]; this is to say, when using the secure/insecure classification based on the ADS, the probability of correctly identifying the cases that AQS classifies as insecure attachment is between 15% and 57%. At the same time, the ADS has a specificity of 99%, CI 95% = [67.85; 99.16], which means a 68% to 99% probability of correctly identifying cases with secure attachment.

In terms of the ADS scale’s positive predictive value, the probability that a child classified in the insecure attachment group by the AQS will obtain the same classification in the ADS scale is 99%, IC 95% = [56.09; 98.68]. With regards to the negative predictive value, however, the probability that a child in the secure attachment group will be similarly classified by the AQS is only 44% [95% CI = 25.02; 64.7]. Based on these results, ROC curve and Jouden’s Index analyses were performed in order to find a cut-off point for the ADS that would optimize both its sensitivity and specificity levels. The AQS scores were used as a criterion, and a continuous score for the ADS was used, adding the items’ scores.

Figure 1 shows the ROC curve; the indicator of the ADS’s discriminant capacity is the area under the curve corresponding to .28 (IC 95% = [.105; .453]; z = .09, p = .04), which indicates that the proposed criterion is not better than chance.

Figure 1 ROC curve of the ADS continuous scores, using the AQS scores as a criterion (cut-off point of .32, following earlier studies.) 

The Youden’s index analysis defines the optimal cut-off for a child to be classified in the secure attachment group based on the ADS as 4.5 secure behaviors (see Table 2.)

Table 2 Sensitivity, Specificity, and Youden’s Index for the various cut-off point obtained from ROC Curve analysis 

Since behaviors are discrete indivisible units, this datum was rounded and a new cut-off point was established: 5 or more secure behaviors. This new classification methodology resulted in a sensitivity of 47.6% (CI 95% = [26.4; 69.7]) and a specificity of 90.9 % (CI 95% = [57.1; 99.5]).

As can be seen in Table 3, using 5 or more secure behaviors as a cut-off point for the secure attachment classification in the ADS scale results in 65.6% of children classified as secure. The correlation with attachment security as measured by AQS is enhanced slightly (rbp = .43, p = .01) and the association between the two classifications continues to be significant [χ2 (1, N = 32) = 4.75, p = .03, V = .38]. In terms of the consistency between the secure and insecure attachment classifications in the two instruments, agreement results in 62.5% of cases (20).

Table 3 Cross Table ADS classification with a new cut-off point and secure - insecure AQS classification with a .32 cut-off point (n = 32)