PSYM 2132 : Les Mthodes Multivaries en Psychologie

Jan Johannes & Cedric Taverne

Analysis of attachment disorders of children in institutions

Lopez Rodriguez, David NOMA: 0371-12-00

Universit Catholique de Louvain-la-Neuve Facult de Psychologie et des Sciences de lEducation

Anne acadmique 2012-2013

1. Introduction The aim of this project is the analysis of a database focus on the field of Health psychology and psychopathology of children and adolescents. These studies evaluate the effect of the attachment versus the interactive skills of abandoned childrens from Congo, living in residential institutions. We will attempt to investigate on (1) behavior of the childrens; (2) patterns of behavior in the childrens; (3) encourage; and (4) the needs of the childrens. In this study we used a battery that is intended to cover various aspects about which we can infer answers to these questions. First, we analyze and present the scales used, validating and evaluating its internal consistency (by CPA and MCP). Then, we explore the research hypothesis. We perform and MCA analysis and a Clustering (HCPC) that allow us to obtain the relationship and distance between variables. These procedures allow us to distinguish the different existing groups. Clustering allows us to identify natural groups who provide valuable information about our sample. Finally, discussions and conclusions are presented.

2. Context Attachment is one of the most important variables for the successful development of a person during their childhood, psycho, physically, and socially. The presence of parental figures is crucial to achieve these objectives (Bowlby, 1978).. An early separation, an educational style inappropriate or growing up in a residential institution can undermine build good relationships with others (Sroufe, 2009). Our sample was performed on Kinshasa, a city that belongs to the Democratic Republic of Congo. Kinshasa has human rights to protect children, but besides that there live is difficult and many parents are forced to abandon their children, being exposed to abuse, sexual, physical and psychological (Mafu, 2006). The emergences of NGOs to work with children has provided the opportunity to give them a chance by moving to a residential institution. The relations that children maintain with the institution are the main concern of this research, e.g. affection to their caregivers, seeking comfort, attention and social problems. Some studies (Izendom, Bakerman-Kranemburg et al., 2010) showed that children living in institutions are more vulnerable to psychopathology and violence in these institutions as the main cause of disorganized attachment.

3. Sample Data were collected from 84 children from Kinshasa, capital of the Democratic Republic of Congo. We have selected two groups of participants, 42 children from residential institutions (experimental group) and 42 children living with their parents in family (control group). The age of children comprises between 4 and 7 years.

4. Hypothesis Each of the hypotheses that we will propose has been done according to the different demands of research and the types of analyzes required for verification. In the first hypothesis we aim to evaluate how the family group differs from institution group and whether a particular profile in relation to different types of attachment. We want to know if the group institution presents a kind of disorganized attachment, ambivalent or avoidant. If inhibited or uninhibited, holding low scores on intelligence and whether anthropometric measurements are below average. H1: The group institution is linked positively to different types of negative attachment (disorganized, avoidant or ambivalent, inhibited and uninhibited) and negatively related intelligence and height / weight. It is of great interest to see the profile of people who have a relationship contrary to our hypothesis. That is, children who are in the group institution with a secure attachment type, an intelligence above the average and standard anthropometric measures. In addition, children in the family group with a lack of attachment profile, low intelligence and low height / weight relationship. Furthermore, in order to give more coherence to our previous hypothesis we infer the existence of natural groups in our sample that are linked to the four categories of attachment and family groups / institution. H2: The group family and institution form distinct natural groups, in the same way that the four types of attachment. If there is a relationship between the variables described in the H1 we will find the presence of these natural groups and we will demonstrate a clear difference in the type of development that have followed these children throughout their childhood.

5. Instruments and descriptive analysis In this section we discuss briefly the descriptive contributed to each of our scales and describe how they work (Annex 1 for a data summary) 5.1 Ravens Progressive Matrix Test (Fouchey, 2007). It is an intelligence test based on the g factor, factor that dominates all cognitive skills. There are three versions but in this paper we have selected the most simple and short, the scale colorful comprises by three subscales (A, B and AB), each of which ones have 12 items, giving a total score of 36. In our sample we take as a benchmark a score of 18. This score gives a good cognitive level for these children and we will find under which an unfavorable level with respect to the mean. The percentage of children with a score below 18 on this scale is of 58.33% (19.04% belongs to the family group) while the percentage of children is above 41.67% (10.71% belongs to the institution).

5.2 Attachment Story Completion Task (Bretherton et al., 1990). This scale evaluates the representation of attachment of children through a series of seven stories. Each story is based on one of four themes: (1) parental authority, (2) the security function and reconfort of parents, (3) separation and (4) return of temporary parental figures. The aim is to activate their internal models of attachment to categorize In our sample we see a clear dominance of the group 'Secure' with 50% of the sample, of which 16.67% is in the institution group. In contrast, the group with the type of attachment 'Evitement' is only 4.76% of the sample, of which 2.38% belongs to the family. The group 'Ambivalent' represents 16.67% of the sample and the group 'Desorganis' 28.57%. 5.3 Disturbances Attachment Interview (DAI) (Smyke and Zeanah, 1999). It is a interview semistructured with 12 items that explore the signs of attachment problems regarding their caregivers. There are eight subscales, the first five pertinent to 'Inhibited Type' and the subscales V1, V6, V7 and V8 linked to 'Uninhibited Type'. Note that the subscale V1 is taken into account for two types of attachment. In our sample we found a 7.14% of children in group inhibits and 46.43% of children in the uninhibited. We found that although both types are opposites, five cases have a high score on both. The percentage of children in group inhibited and uninhibited in the institution group is respectively 2.38% and 33.33%, while in the family group is 4.76% and 13.09% respectively. While we find more cases of children with disinhibited attachment to institution, we found less cases of children inhibited in the same group. 5.4 Anthropometric measures. This study evaluated the length / weight relationship given as 75% cut. In our sample we found an average of 76.50, the 45.23% of the children are below the 75% required. We found that of these, predomnate the group institution with 36.95%.

6. Validation of tests The objective of this section is to validate our measuring instruments. We will make an analysis CPA for the questionnaire ASCT and MCA analysis for DAI. Although we have information on the scale and subscales Raven will not perform validation. 6.1 ACP of ASCT. To assess the internal consistency of this scale we performed a principal component analysis (PCA) in order to reduce the dimensionality and see if the first dimension can explain the relationship between the subcategories. So we've selected three continuous variables and an

additional factor, code.attachment. The subcategories of code.attachment are Desorganis, evitement, ambivalent and secure.

eigenvalue comp 1 comp 2 comp 3 2.4560912 0.2919343 0.2519745

percentage of variance 81.869708 9.731143 8.399149

cumulative percentage of variance 81.86971 91.60085 100.00000

Table 1 res$eig ACP of ASCT

Following the decision rules we see that the first principal component explains 81.87% of the variance and its own value is higher than 1. Therefore, it requires only one dimension to explain a high percentage of the variance on this scale (Annex 2) 6.2 MCA of DAI To validate this scale we will make an analysis of multiple components (MCA). In the validation is found the mentioned subscales in the previous section related to subtypes to which they belong, inhibited or uninhibited. It is verified graphically (Annex 3) as there is a differentiation between inhibited and uninhibited categories with respect to group nonDes and NonIn. This groups not maintaining attachment issues and they are linked to low and averages rates in the 8 subscales (V1 to V8 Low and Medium). It is verified also that there is a relationship between high scores inhibited and V1, V2 and V3 but not in V4 (note that we do not comment V5 because nobody in our sample had high scores on this subscale). The uninhibited group is linked to high scores V6, V7 and V8 showing that our scale is consistent with what it purports to measure. The dimension 1 represents 20.91% of the variance, whereas the dimension 2 represents 14.98%.

eigenvalue dim 1 dim 2 dim 3 dim 4 dim 5 dim 6 dim 7 dim 8 dim 9

percentage of variance 2.090692e+01 1.498241e+01 1.128116e+01 8.646820e+00 8.056497e+00 7.278352e+00 6.910261e+00 6.565540e+00 4.966209e+00 3.657595e+00

cumulative percentage of variance 20.90692 35.88933 47.17049 55.81731 63.87381 71.15216 78.06242 84.62796 89.59417 93.25177

3.920048e-01 2.809202e-01 2.115218e-01 1.621279e-01 1.510593e-01 1.364691e-01 1.295674e-01 1.231039e-01 9.311643e-02

dim 10 6.857990e-02

Table 2 Eigenvalue MCA of DAI (validation)

7. Results In this section, we will perform three different analysis to check our hypothesis. First, we will perform a multiple correspondence analysis to evaluate the relationship between our variables. Second, we will make a clustering analysis HPCP on MCA to discover the existence of natural groups. Third, with the aim to find the natural groups of CPA and ASCP, we will perform a clustering about the CPA on ASCP analysis. 7.1 MCA We performed a multiple correspondence analysis (Annex 4) and scales with the following variables: age, sex, ASCT, DAI, anthropometric measures and Ravens test. Dim 1 represents 29.6% of variance and clearly differentiates the control group and experimental. In the control group or family, we used the following variables (1) ages 4, 5 and 7 years, (2) secure attachment profile, (3) length / weight relationship over 75%, (4) Raven scores above 18 and (5) not disinhibited attachment. In the experimental group, we used the following variables (1) the group linked to a length / weight relationship over 75, (2) profile of disorganized, (3) ambivalent, (4) avoidance attachment and (5) Ravens low scores in uninhibited and inhibited attachment. The 15.43% of the variance (Dim2) and the 29.6% of the variance (Dim1) are shown in this figures. In the Dim2 axis, we can observe only a great differentiation with the profile of

avoidance and inhibited attachment, while the rest of variables form a cloud over the center of gravity.

eigenvalue dim 1 dim 2 dim 3 dim 4 dim 5 dim 6 dim 7 dim 8 dim 9 0.443931801 0.231430741 0.180076732 0.149497712 0.129514093 0.117442590 0.091562875 0.077784932 0.030656064

percentage of variance cumulative percentage of variance 29.5954534 15.4287160 12.0051155 9.9665141 8.6342729 7.8295060 6.1041917 5.1856621 2.0437376 1.7474244 29.59545 45.02417 57.02928 66.99580 75.63007 83.45958 89.56377 94.74943 96.79317 98.54059

dim 10 0.026211366

Table 3 Eigenvalue MCA (all variables)

In order to study the relationship with the groups, we chose the DAI (V1 to V8) subscales as supplementary. In the figure, the scores in Dim1 go from Low to Medium and High from left to right,, except V4 scale which is explained by Dim2. Dim2 explains the 15.43% of the variance and can be differentiated high scores on V1, V2, V3, V4 with respect to the rest of variables around the center of gravity. 7.2 Clustering In order to detect the number of natural groups in our dataset that allow us to accept or reject the hypothesis, we made a clustering after MCA. A priori, the number of cluster is unknown, we chose the number of groups manually with the help of the dendrogram. We used the 98.54% of the variance with the first 10 dimensions of MCA, obtaining more stable results that those with less dimensions. Note that the number of dimensions should not be larger than 15, always that the 80% of the variance is take into account. In the dendrogram, we see the hierarchical tree of our sample, suggesting an optimal distribution of three classes. An attempt to test a number of clusters higher was made, but these groups were too small (with few individuals). In the analysis of dimensions, the Dim1 is significant in the clusters 1 and 2, while cluster 3 is significant for Dim2 (annex 5). In the cluster 1, the relationships with low and medium scores of the different subscales of DAI, especially with V1, V2 and V5, V7 and V8 are found.

Again in cluster 1, a very high relation with the group nonDesinhib, secure attachment, family group and height / weight over 75%. Lastly also associated with high scores on Raven are found.

Cluster 2 shows the relationships with the average scores linked to V1, V2, V3, V4, V5, V6 and with the high of V7. Moreover, this cluster is linked with the group desinhib, desorganis and ambivalent attachment style and institution group. Also associated with lower scores in Raven. Cluster 3 shows a smaller number of subjects, quite distant from the center of gravity. At first glance, Cluster 3 is linked with the evitement attachment profile and inhibit attachment and high scores of V1, V2 and V3. All these results are linked to the cluster in an measurement larger than 45%. All relationships are shown in Annex 5. A summary of the results for each cluster are shown in Table 4. Table 4 indicates that exist three natural groups in our dataset with well-defined characteristics. The exceptions that can be found in each natural group require a different level of analysis.
Cluster 1 Famille Secure nonDesinhib height / weight >75% Low and medium rates in DAI high rates in Raven Cluster 2 Institution Desorganis and Ambivalent Desinhib height / weight <75% Medium rates in DAI Low rates in Raven Cluster 3 Evitement Inhib High rates in DAI for Inhibition.

Table 4 Information and profile of natural groups

7.2.1 Are there exceptions relevant to our hypotheses? Given the clustering (Section 7.2), the first 42 children belong to the control group and the following 42 to the experimental group. In our clustering, cluster 1 and 2 can be distinguished. Cluster 1 is formed mainly by the family and the childrens, while cluster 2 is formed by childrens living in residential institutions. A profile for each group is attended to find, as well as what happen with each subject that belong to a family, maintain a near profile to the childrens of an institution. In cluster 1, 14 cases of children who are living in institutions but maintain a profile close to the family are found. In cluster 2, 10 cases of children living in families but maintain a profile next to institution are found.
Group Famille Institution Desorganis Ambivalent Secure 5 5 0 nonDESIN 1 DESIN 9 <75pds 5 >75pds 5 R<18 8 R>18 2 14 nonIN 9 14 IN 1 0

0 0 14 14 0 0 14 0 Table 5 for the sample of 24 subjects who have a different profile than expected..

In Table 5, the descriptions of each group are shown. First, the family group in cluster 2, is related to a desorganis and ambivalent attachment profile, desinhib attachment and low rates in Raven. The institution group in cluster 1, everyone has a secure attachment profile, anthropometric good levels, good cognitive level and good attachment.

7.3 Clustering of ASCT. A clustering performed with the ACP (Section 6.1) will allow us to understand how the subscales are distributed in our sample. We conducted this clustering on three dimensions representing the 100% of the variance. The HCPC with consolidated clusters (Annex 6) suggests to select three natural groups, obtaining the best option for the desired results. Based on these three natural groups, a clear differentiation on the Dim1, which explains 81.87% of the variance, are found. Cluster 1 and 3 are significant with Dim1, while cluster 2 is not significant with Dim2. The Secure group is clearly differentiated in Cluster 3, while the other two groups are linked with the other 3 subtypes from the insecure group (ambivalence, evitation et desorganis).

. Discussion and Conclusions In this work we investigate the differences in profiles of childrens living in residential institutions and with those whom live in families. We with a sample of both profiles. Children living in families generally have a good internal model of attachment. These childrens show that there are not problems when they are temporarily separated from their parents or about when the parents will provide security and reconfort. This is essential to build a favorable attachment development. Also, these childrens maintain a level in weight, height and intelligence according to the average. From the DAI, the attachments have a normal development. The results of its subscales show that either does not exist or there is occasionally a problem of attachment but do not represent part of their normal behavior. We found reluctance to face the strange (V7-M), the lack of response to parental Reconfort (V3-M) and the absence of adult verification (V6-M). Children living in institutions are, however, more related to attachment issues. In this group profiles of ambivalent, evitance and desorganis attachment are found. We also found that 10

children have low cognitive and anthropometric index below average. On top of that, an attachment, sometimes inhibited and sometimes uninhibited must be added. Specifically, the latter is regularly found in 28 cases. These cases shown to have problems with the strange reluctance (V7-H) and to have occasional problems when having an adult of preference (V1-M), regulate their emotions (V5-M), go with a stranger (V8-M),... In Section 7.2.1, we discussed children who maintained a profile different than expected, showing that, despite living in an unfavorable situation has been developing normally. In future research is required to study under which conditions these cases occurs, in order to promote and develop these children to have a life as normal as possible. Similarly, knowing that institutionalized sample comes from the same two institutions in Kinshasa. A study about resilience would be of interest to know if they have gone ahead, because of their internal skills or the immediate environment (caregivers preferences, external support,).

11

Annex 1. Summary of Data

sexe F:38 M:46

Age 4: 4 5:14 6:24 7:42

Code.Groupe Famille :42 Institution:42

code.attach. Secure :42

Mentalisation Min. :2.580 1st Qu.:3.768 Median :4.675 Mean :4.340 3rd Qu.:4.780 Max. :5.670

Desorganis:24 Ambivalent :14 Evitement : 4

Parentalit Min. :3.060 1st Qu.:3.800 Median :4.930 Mean :4.495 3rd Qu.:4.980 Max. :5.800

collaboration Min. :2.000 1st Qu.:3.770 Median :4.770 Mean :4.285 3rd Qu.:4.820 Max. :5.320

Raven36 pdstaille V1

V2

V3

V4

V5 V5-L:48

R<18:49 <75 :31 V1-L:46 V2-L:71 R>18:35 R>75: 0 V1-M:36 V2-M:12 >75 :53 V1-H: 2 V2-H: 1 V3-H: 3

V3-L:31 V4-L:44

V3-M:50 V4-M:39 V5-M:36 V4-H: 1

V6

V7

V8

INHIBE

DESINHIBE

V6-L:16 V7-L: 9 V8-L:34

nonIN:78 nonDESIN:45

V6-M:67 V7-M:43 V8-M:47 IN : 6 DESIN :39 V6-H: 1 V7-H:32 V8-H: 3

12

Annex 2. Graph Validation of ASCT

13

Annex 3. Graph Validation of DAI (MCA)

14

Annex 4. Graph MCA results (all variables)

15

Annex 5. Clustering of MCA results (all variables)

$quanti $quanti$`1` v.test Mean in category Overall mean sd in category Overall sd p.value Dim.1 -8.455461 -0.5896034 1.825203e-17 0.2222869 0.6662821 2.779878e-17 $quanti$`2` v.test Mean in category Overall mean sd in category Overall sd p.value Dim.1 7.580691 0.6401729 1.825203e-17 0.2775565 0.6662821 3.437187e-14 Dim.3 2.219969 0.1194005 4.054916e-17 0.5012816 0.4243545 2.642088e-02 Dim.2 3.150310 -0.1920853 5.233599e-17 0.2296264 0.4810725 1.630970e-03 $quanti$`3` v.test Mean in category Overall mean sd in category Overall sd p.value Dim.2 8.155756 1.9259753 5.233599e-17 0.13065082 0.4810725 3.470017e-16 Dim.1 2.213878 0.7240821 1.825203e-17 0.22150051 0.6662821 2.683715e-02 Dim.10 -2.310107 -0.1835916 -1.352155e-17 0.04043854 0.1618992 2.088225e-02

$category $category$`1` Cla/Mod Mod/Cla Global p.value v.test DESINHIBE=nonDESIN 97.777778 100.000000 53.57143 5.889833e-23 9.865232 code.attach.=Secure 100.000000 95.454545 50.00000 1.126921e-21 9.564549 V1=V1-L 91.304348 95.454545 54.76190 1.509066e-16 8.255785 pdstaille=>75 83.018868 100.000000 63.09524 5.800325e-15 7.808216 V5=V5-L 85.416667 93.181818 57.14286 6.982782e-13 7.179772 Raven36=R>18 91.428571 72.727273 41.66667 8.134405e-10 6.142282 V7=V7-M 83.720930 81.818182 51.19048 4.215078e-09 5.875524 V8=V8-L 82.352941 63.636364 40.47619 9.546611e-06 4.427196 V2=V2-L 61.971831 100.000000 84.52381 3.845667e-05 4.116562 Code.Groupe=Famille 71.428571 68.181818 50.00000 9.320797e-04 3.310264 V4=V4-L 70.454545 70.454545 52.38095 9.981082e-04 3.291059 V6=V6-L 87.500000 31.818182 19.04762 3.108706e-03 2.956779 V3=V3-L 74.193548 52.272727 36.90476 4.110216e-03 2.869577 V7=V7-L 88.888889 18.181818 10.71429 4.236597e-02 2.029907 V3=V3-M 42.000000 47.727273 59.52381 3.589536e-02 -2.098111 V6=V6-M 44.776119 68.181818 79.76190 1.043072e-02 -2.561214 code.attach.=Ambivalent 14.285714 4.545455 16.66667 3.678841e-03 -2.904463 Code.Groupe=Institution 33.333333 31.818182 50.00000 9.320797e-04 -3.310264 V4=V4-M 30.769231 27.272727 46.42857 4.461916e-04 -3.511139 V8=V8-M 34.042553 36.363636 55.95238 2.878091e-04 -3.626033 V2=V2-M 0.000000 0.000000 14.28571 9.888858e-05 -3.893303 Raven36=R<18 24.489796 27.272727 58.33333 8.134405e-10 -6.142282 code.attach.=Desorganis 0.000000 0.000000 28.57143 1.958152e-10 -6.364588 V5=V5-M 8.333333 6.818182 42.85714 6.982782e-13 -7.179772 V1=V1-M 5.555556 4.545455 42.85714 1.019974e-14 -7.736741 pdstaille=<75 0.000000 0.000000 36.90476 5.800325e-15 -7.808216 V7=V7-H 0.000000 0.000000 38.09524 9.849608e-16 -8.028718 DESINHIBE=DESIN 0.000000 0.000000 46.42857 5.889833e-23 -9.865232 $category$`2` Cla/Mod Mod/Cla Global p.value v.test DESINHIBE=DESIN 89.743590 97.222222 46.42857 1.033986e-17 8.570096 V1=V1-M 91.666667 91.666667 42.85714 3.461241e-16 8.156062 V5=V5-M 88.888889 88.888889 42.85714 3.228652e-14 7.588807 pdstaille=<75 93.548387 80.555556 36.90476 2.017321e-13 7.347643 V7=V7-H 90.625000 80.555556 38.09524 1.877545e-12 7.043289 code.attach.=Desorganis 100.000000 66.666667 28.57143 3.899591e-12 6.940773 Raven36=R<18 67.346939 91.666667 58.33333 6.374500e-08 5.407965 Code.Groupe=Institution 61.904762 72.222222 50.00000 8.248520e-04 3.344318 V8=V8-M 59.574468 77.777778 55.95238 9.010459e-04 3.319730 code.attach.=Ambivalent 85.714286 33.333333 16.66667 9.620254e-04 3.301403 V3=V3-M 56.000000 77.777778 59.52381 5.702419e-03 2.764419 V4=V4-M 58.974359 63.888889 46.42857 1.019679e-02 2.569083 V6=V6-M 49.253731 91.666667 79.76190 3.301902e-02 2.131852 V2=V2-M 75.000000 25.000000 14.28571 3.442898e-02 2.115011 V3=V3-L 25.806452 22.222222 36.90476 2.711061e-02 -2.209922 V4=V4-L 29.545455 36.111111 52.38095 1.757816e-02 -2.374386 V6=V6-L 12.500000 5.555556 19.04762 1.081780e-02 -2.548530 Code.Groupe=Famille 23.809524 27.777778 50.00000 8.248520e-04 -3.344318 V8=V8-L 14.705882 13.888889 40.47619 2.608253e-05 -4.205223 Raven36=R>18 8.571429 8.333333 41.66667 6.374500e-08 -5.407965 V7=V7-M 13.953488 16.666667 51.19048 5.680718e-08 -5.428571 pdstaille=>75 13.207547 19.444444 63.09524 2.017321e-13 -7.347643 V5=V5-L 8.333333 11.111111 57.14286 3.228652e-14 -7.588807 V1=V1-L 6.521739 8.333333 54.76190 2.144717e-14 -7.641640 code.attach.=Secure 0.000000 0.000000 50.00000 1.461842e-17 -8.530129 DESINHIBE=nonDESIN 2.222222 2.777778 53.57143 1.033986e-17 -8.570096

16

$category$`3` Cla/Mod Mod/Cla Global p.value v.test code.attach.=Evitement 100.00000 100 4.761905 1.036537e-06 4.884572 INHIBE=IN 66.66667 100 7.142857 1.554806e-05 4.320777 V3=V3-H 100.00000 75 3.571429 8.395953e-05 3.932807 V1=V1-H 100.00000 50 2.380952 3.442341e-03 2.925201 V2=V2-M 25.00000 75 14.285714 1.693184e-02 2.388185 V2=V2-L 0.00000 0 84.523810 7.411243e-04 -3.373896 INHIBE=nonIN 0.00000 0 92.857143 1.554806e-05 -4.320777

17

Annex 6. Clustering of ASCT (CPA)

18