Children’s family drawings: association with attachment representations in story stem narratives and social and emotional difficulties

ABSTRACT

We examined representations of attachment security using story stem narratives (Manchester Attachment Story Task, MCAST) and a family drawing task in 50 Greek children between 4 and 6 years old. Additionally, we examined the association between attachment classification based on the family drawing task and maternal reports of social and emotional symptoms assessed using the Strengths and Difficulties Questionnaire. Attachment security classification based on the MCAST was not significantly associated with attachment security classification based on the family drawings. There was a small to medium strength significant association between drawing-based insecure attachment classification and higher scores on conduct problems and inattention/hyperactivity ratings. However, the associations were not statistically significant after controlling for the effects of child gender, and maternal education and age. The findings are discussed in relation to the properties of the family drawing task to measure attachment security in the preschool and early middle childhood years.

KEYWORDS:

• Attachment
• family drawings
• Manchester Child Attachment Story Task (MCAST)
• Strengths and Difficulties Questionnaire
• social and emotional difficulties
• early childhood

Child attachment refers to the affectional tie between a child and his or her caregiver (Cassidy, 2016). In the context of this tie, children are understood to be biologically predisposed to seek proximity to their caregiver in times of distress. Proximity is achieved through the activation of the child’s attachment behavioural system which reflects a set of attachment behaviours that increase the chances of staying closer to the caregiver (Cassidy, 2016). The Strange Situation (SS) is the classic assessment method of child attachment in the infant and toddler years (Solomon & George, 2016). It is based on the observation of the child’s attachment system which is activated by a series of brief episodes of child separation and reunion with the caregiver. As children grow older, developmental advances in verbal, cognitive, and motor skills, and the child’s growing appetite for autonomy bring change to the child’s attachment behaviour which is reflected in more sophisticated ways (Allen, Bendixsen, Babcock Fenerci, & Green, 2018). Therefore, classifying attachment security by observing the child's behaviour using the classic separation-reunion assessment becomes challenging (Allen et al., 2018). At the same time, the use of self-report measures with children in later early childhood or early primary years is not encouraged because of the child’s limited cognitive skills to handle their demands (Jewell et al., 2019). The limitations of these measures and children’s growing capacity to hold internal representations of the self and their relationship with their caregivers have made the use of measures based on symbolic representation of attachment very common with preschool or early primary school age children (Kerns & Brumariu, 2016).

Projective drawing is a representational measurement technique which places limited demands on the child’s verbal skills and allows the expression of emotions that would be otherwise difficult to assess (Kloft, Hawes, Moul, Sultan, & Dadds, 2017). Additionally, family drawings are quick and easy to administer so they provide a time and cost-efficient alternative to the more time consuming and expensive to administer measures of attachment, such as reunion procedures (Dallaire, Ciccone, & Wilson, 2012). Bowlby considered the potential of symbolic forms of mental representations of attachment figures (such as in photographs or letters) to activate attachment behaviour (Behrens & Kaplan, 2011). Therefore, family drawings could capture the child’s internal representation of attachment because the choice of who to draw and how may activate the child’s attachment behaviour above and beyond drawing skills (Behrens & Kaplan, 2011). This idea was originally proposed by Kaplan and Main (1986) whose work showed that certain characteristics in the drawings of children between the ages of 5 and 7 years old were associated with infant attachment classifications based on the SS procedure (Pianta & Longmaid, 1999). For instance, some of the characteristics of drawings of children classified avoidant as infants often included smiling, non-individuated family members that were distant from each other, often lacking arms, and ‘floating’; the drawings of children classified resistant in infancy often included either very large or very small figures unusually close to each other, and often accentuating vulnerable or intimate body parts (Madigan, Ladd, & Goldberg, 2003). The findings led to the development of a coding system to classify children’s attachment representations which includes a list of discrete features that reflect constructs such as the size, location, degree of movement, individuation, completeness of figures, quality of smiles, and impressions of vulnerability (Pianta & Longmaid, 1999). To make classifications the coders examine the presence or absence of discrete features and relate their patterning to the overall descriptions of the four major classifications provided in Kaplan and Main (1986): avoidant (A), secure (B), ambivalent (C), and disorganised (D). Fury, Carlson, and Sroufe (1997) introduced modifications and expanded the general scoring criteria outlined in Kaplan and Main (1986) to account for individual differences in older children. They also created aggregated scores and seven 7-point global scales to assess the drawings in an integrative manner. These scales involved the context and the patterning of the discrete features in the drawings as a whole and they reflected the following dimensions: Vitality/Creativity, Family Pride/Happiness, Vulnerability, Emotional Distance/Isolation, Tension/Anger, Role reversal, Bizarreness/Dissociation, Global Pathology. Fury et al. (1997) and subsequent studies found that aggregated and global scale scores were more strongly associated with early and concurrent attachment than individual signs (Behrens & Kaplan, 2011; Fihrer & McMahon, 2009; Madigan et al., 2003). Despite the benefits of the Fury et al. (1997) global scales, they do not classify drawings into attachment categories. Therefore, we drew on the Kaplan and Main (1986) attachment classification system to examine properties of family drawings as an independent categorical measure of attachment.

A few studies have used the original or adapted versions of the Kaplan and Main (1986) list of individual signs to classify children’s attachment relationship based on the presence or absence of the signs and their overall patterning. Fury et al. (1997) found an association between early attachment classification and attachment classification based on a modified version of the Kaplan and Main (1986) list of individual markers in a sample of children between the ages of 8 and 9 years old at risk for caregiving difficulties and with histories of stable attachment at 12 and 18 months. Later, Madigan et al. (2003) examined the association of attachment classification based on the list of individual signs modified by Fury et al. (1997) and infant attachment based on the SS in a sample of chronically ill children at the age of 7 years old. They found that the drawings of the children who were classified insecure as infants included a significantly higher proportion of individual signs reflecting insecurity than the drawings of children classified secure in infancy. However, a regression analysis showed that early attachment classification made only a very small contribution to children’s total number of signs reflecting insecurity after controlling for child and parent characteristics. The authors explained that failing to find a stronger association between early attachment and attachment classification based on drawings could be attributed to differences in attachment measurement and the composition of the sample’s characteristics including the fewer cases of disorganised attachment which may have attenuated the link with attachment. Similarly, Fihrer and McMahon (2009) found no significant association between infant attachment classification and attachment classification based on the Fury et al. (1997) modified version of the Kaplan and Main (1986) system in a sample of children between the ages of 6 and 8 years old and which included a large proportion of mothers with depression. The authors explained that the lack of association could be attributed to lack of stability in the children’s attachment relationship because of potentially experiencing unstable caregiving due to their mothers’ ongoing depression. Bar-Haim, Sutton, Fox, and Marvin (2000) suggested that the quality of attachment representations is more a reflection of the current quality of parent–child interaction than it is a reflection of earlier interaction patterns. Arguably, it is not always possible to control change in caregiving quality and its impact on children’s later attachment to examine stability of attachment with precision. The examination of the concurrent agreement between family drawing-based attachment security classification and an established measure of attachment security could contribute significantly to our understanding of the validity of family drawings as a representational measure of attachment classification. A later study by Behrens and Kaplan (2011) examined concurrently the link between attachment classification based on the Main and Cassidy (1988) reunion system and the Kaplan and Main (1986) classification system in a sample of Japanese children between the ages of 6 and 7 years old but did not find any significant association. A plausible interpretation of the lack of association could be that children at this age struggle to maintain consistent strategies in two different measures of attachment (behavioural vs. representational) (Behrens & Kaplan, 2011). Hence, the concurrent examination of family drawings with another representational measure is an important extension of the research in the validity of family drawings as an independent measure of attachment classification.

There is limited research on the association between the Kaplan and Main (1986) attachment classification system and other representational measures. The narrative story stem method is one of the most popular in the assessment of attachment representations in children in the preschool and early middle childhood years (Allen et al., 2018). The child is asked to complete several story stems that induce mild distress (e.g. an injury, a nightmare) to activate the child’s behavioural attachment system and a response to the distress which is supposed to reflect his or her attachment representations. The Manchester Child Attachment Story Task (MCAST; Green, Stanley, Smith, & Goldwyn, 2000) is one of the most common story stem methods of attachment classification and uses doll play to allow the child to act out the story stems (Solomon & George, 2016). Jin, Chung, and Hazen (2018) contrasted the Kaplan and Main (1986) classification system with the MCAST in a community and a clinical sample of Korean children between 7 and 9 years old and found that the MCAST classifications were related to the classifications derived from the family drawings in both samples. Pace, Guerriero, and Zavattini (2020) used 24 individual markers to classify family drawings in attachment categories in a clinical and small (n = 12) community sample of Italian children between the ages of 5 and 8 years old. They found a significant correspondence between disorganised vs. non-disorganised classified groups across the two measures in the community sample, but not between secure vs. insecure classified groups.

Research in the validity of family drawings as a measure of attachment should consider children’s chronological age because it is related to drawing quality (Pianta & Longmaid, 1999). The family drawing task can be administered to children as young as five because by that age they have developed important drawings skills (Sugden, Wade, & Hart, 2013). However, the findings from an earlier study which showed that aspects of family drawings of children at the age of 4 years, such as proximity to the mother, were associated with a history of infant security (Madigan et al., 2003) suggest that family drawings can be potentially used with children younger than 5 years old. Examining the potential of family drawings to classify attachment security with accuracy in children as young as 4 years old will provide support of family drawings as an ecologically valid method of attachment classification in young children. Additionally, it will contribute to the ongoing research of identifying appropriate methods of attachment in an age group of children where no dominant conceptual or methodological approach of attachment security exists (Kerns & Brumariu, 2016). Therefore, the first objective of the study was to replicate previous work by examining the agreement between attachment classification based on the family drawing task and the MCAST in children between the ages of 4 and 6 years old.

In addition to age, children’s social and emotional outcomes should be accounted in the investigation of the validity of family drawings as a measure of attachment because they are associated with attachment classification across different assessment methods of attachment (DeKlyen & Greenberg, 2016). A few studies suggest that drawing-based attachment security is sensitive to children’s social and emotional outcomes. Fury et al. (1997) demonstrated that drawing quality based on global scale ratings was independently predicted by teachers ratings of behavioural and/or emotional difficulties. Pianta and Longmaid (1999) showed that preschool children classified as securely attached based on family drawings were more likely to score well on teacher ratings of poor social adjustment independently of age, sex, socioeconomic status, intelligence, and motor skills. Furthermore, Madigan et al. (2003) found an independent link between concurrent parent reported behavioural and/or emotional difficulties and the total number of signs reflecting insecurity in 7-year-old children. As well as association with global dimensions of social and emotional difficulties a limited but emerging literature explores the association of attachment insecurity and specific types of difficulties that are related to insecure attachment such as attention deficit and hyperactivity (Storebø, Rasmussen, & Simonsen, 2016), conduct problems (Boldt, Kochanska, & Jonas, 2017; Theule, Germain, Cheung, Hurl, & Markel, 2016), and emotional difficulties (DeKlyen & Greenberg, 2016). For instance, an earlier study found that boys between 5 and 10 years old with a diagnosis of inattention and hyperactivity were more likely than those without a diagnosis to score lower on global ratings of family drawings quality independent of age (Clarke, Ungerer, Chahoud, Johnson, & Stiefel, 2002). Findings from the US-based Family Life Project, a large epidemiological study, showed that conduct problems at first grade were not associated with global ratings of family drawings (Wagner, Mills-Koonce, Willoughby, Zvara, & Cox, 2015), but they were associated with attachment classification based on the frequency of presence of individual signs in children drawings (Rehder, Mills-Koonce, Wagner, Zvara, & Willoughby, 2020). Regarding emotional difficulties, earlier research suggests an association with family drawing tasks, such as the Draw-A-Person and the Kinetic Family Drawing (Tharinger & Stark, 1990). The limited research in this area and the variability in findings suggest that further research will enhance our understanding of whether attachment security classification based on family drawings is associated to specific aspects of children’s social and emotional outcomes. Therefore, the second aim was to examine the association between attachment classification based on family drawings and early childhood symptoms of inattention and hyperactivity, conduct problems, and emotional difficulties.

Method

Sample

Data were drawn from a N = 72 families of Greek children between 3 and 6 years old who consented to take part in a study of attachment. We collected drawings from n = 50 children (M_age= 5.29 years SD = .45; range = 4.58–6.35 years; n = 21 (42%) girls). Children were recruited from a day-care centre and were free from developmental delay according to centre records. The children were assessed on the MCAST except for one who was not available on the day of assessment, so the sample size used to examine measurement agreement was n = 49 children. Most mothers (M_age = 38.36 SD = 4.88; range = 28.68–51.33 years) were married (94%), in employment (68%), and approximately 1 in 2 (58%) had a higher education degree. Ethical approval was obtained by the Ethics Research Committee of the University of Roehampton (EDU 16/119).

Children’s family drawings

Children were instructed to draw their family on a piece of paper. Teachers positioned an 8½ × 11-inch (A4) blank white page (landscape orientation) in front of the children and gave them a pencil and rubber. Then they asked them to draw their family. Teachers were instructed not to intervene with the children’s drawings on any occasion and that the whole activity should be carried out in a relaxed atmosphere so that children do not feel that they are being assessed for their drawing skills. If a child asked whether they could include friends, teachers were instructed to clarify that only family members could be included in the drawing. If a child was found to struggle to start the drawing, teachers were instructed to prompt by asking ‘Who lives in your house?’. There was no time limit. Teachers were instructed to make a note on top of the figures if their identity was not obvious.

The third author blind to participant demographics, MCAST-based attachment classification, and social and emotional ratings, and with prior experience in coding family drawings (Shiakou, 2012) used a checklist containing 53 discrete features of family drawings outlined in Kaplan and Main (1986) and Pianta and Longmaid (1999). Each drawing was examined for the presence or absence and the patterning of the 53 features and then classified in one of Kaplan and Main’s (1986) attachment classification categories: secure (B e.g. ‘firm, open-armed, embracing stance’), insecure-avoidant (A e.g. ‘figures stiff, without movement’), insecure-ambivalent (C e.g. ‘figures separated by barrier’) or insecure-disorganised (D e.g. ‘unfinished objects or figures’). The data were examined using both the 2-way (Secure/Insecure) and the 4-way classification (B, A, C, D) to examine the distribution of attachment classification and agreement with the MCAST. A senior clinical psychology student, trained by the third author and blind to sample characteristics, coded 20% of the drawings with inter-rater reliability k = . 80.

Attachment story stem narratives

The Manchester Child Attachment Story Task (MCAST; Goldwyn, Stanley, Smith, & Green, 2000; Green et al., 2000) was used to examine children’s representations of attachment. Children were presented with a doll’s house and a collection of female and male dolls and were asked to choose two dolls that represent the child and mother, respectively. Then, they were asked to use them to continue and enact four story stems (nightmare, hurt knee, feeling ill, lost in store) started by the administrator. The stories are designed to induce mild distress to allow children to represent a solution/response to it, preceded by a ‘breakfast’ story that aims to familiarise children with the testing procedure. Upon story completion, the administrator asked the child standardised follow-up questions to gain a fuller picture of the child’s and caregiver’s thoughts and feelings. The stories are video recorded to allow coding. Each story is rated on a 9-point scale for (a) task engagement and increase in arousal; (b) child attachment-related behaviour (e.g. proximity seeking); caregiver behaviour (e.g. parental responsiveness/sensitivity); (c) effectiveness of assuagement; (d) narrative coherence; and (e) attachment disorganisation based on bizarre and disoriented phenomena. A 3-point scale is used to assess mentalising of the self and mother. The examiner uses the ratings to classify each story and the child’s overall attachment style according to the four categories of attachment: Secure (B), Insecure-Avoidant (A), Insecure-Ambivalent (C), Disorganised (D). We used both the 2-way (Secure/Insecure) and the 4-way classification (B, A, C, D) to examine the distribution of attachment classification and agreement with the family drawings. We used the total disorganisation score which is based on the disorganisation score of each of the four story-stems. This score helps to capture disorganised phenomena in children’s stories that are significant but not so much as to lead to a D classification. The first author who is trained by the MCAST developers and experienced administrator and coder administered the task in the centre’s premises and coded and assigned the stories to the four attachment categories (B, A, C, D). A clinical psychologist who is trained by the developers of the MCAST and was blind to participant demographics and ratings coded the data of 20 children (out of N = 72 in the original sample with data on the MCAST). Inter-rater agreement for the 2-way and 4-way classification was κ = .90 and .77, respectively. The intraclass correlation coefficient for the total disorganisation score was r = .86, p = .00.

Social and emotional difficulties

The Strengths and Difficulties Questionnaire (SDQ) is a screening tool for social and emotional difficulties in children and young people (Goodman, 2001). It has been used widely across different populations and extensively researched for its psychometric properties (Stone, Otten, Engels, Vermulst, & Janssens, 2010). The children’s parents completed a Greek version of the Strengths and Difficulties Questionnaire (SDQ) for 4–17-year olds. It includes 25 items which comprise 5 scales of 5 items each rated on a 3-point scale (0 = not true, 1 = somewhat true, 2 = certainly true; range 0–10): (1) Emotional problems (EP); (2) Conduct problems (CP); (3) Hyperactivity/inattention (H/I); (4) Peer problems (PP); (5) Prosocial behaviour (PB). The Greek SDQ version and scoring details can be found on the SDQ website https://sdqinfo.org/. The internal structure and reliability of the parent scales of the Greek SDQ have been satisfactory with Cronbach’s α ranging between .71 and .81 (Bibou-Nakou, Markos, Padeliadu, Chatzilampou, & Ververidou, 2019).

Results

Preliminary analysis

Statistical tests were two-tailed with the alpha set at .05. We used chi-square tests to examine associations between categorical variables and t-tests and Pearson correlations to allow for the examination of between group (e.g. secure vs. insecure classification) differences and associations between dichotomous variables (e.g. secure vs. insecure classification) and mean scores on the SDQ scales, respectively. Table 1 presents the distribution of attachment secure vs. insecure classification by child and mother characteristics. Boys were generally more likely to be classified insecure by either measure. Children classified insecure on the MCAST were not more likely than those classified secure to have mothers who have spent fewer years in education. Nevertheless, there was an almost significant tendency (p = .06) for the proportion of mothers with a higher education degree to be higher in the group of children classified insecure based on the drawing task. We did not find between group differences in child age, maternal employment, and marital status for either measure. The mothers of children with drawings classified secure were significantly younger than the mothers of children with drawings classified insecure, but close inspection of the data spread revealed an outlier in the insecure group which probably inflated the difference. The disorganisation score differentiated the secure from the insecure classified groups based on the MCAST but not based on the family drawings. The group of children classified insecure based on the drawings had significantly higher scores on the parent SDQ CP and H/I scale. The t-tests did not reveal any statistically significant differences between insecure and secure groups classified based on the MCAST.

Table 1. Distribution of attachment security by child and mother characteristics across the two measures.

	Family drawings (n = 50)				MCAST (n = 49)
	Secure (n = 22)	Insecure (n = 28)			Secure (n=25)	Insecure (n=24)
	f (%)	f (%)	x^2	p	f (%)	f (%)	x^2	p
Boys	9 (41)	20 (71)	4.71	.03*	11 (44)	18 (75)	4.87	.03*
Higher education^a	16 (72)	13 (46)	3.50	.06	14 (56)	14 (58)	.03	.87
In employment^b	15 (68)	19 (68)	.00	.98	16 (64)	17 (71)	.26	.61
Married^c	21 (96)	26 (93)	.15	.70	23 (92)	23 (96)	.31	.58
	M (SD)	M (SD)	t (48)	p	M (SD)	M (SD)	t (47)	p
Childage	5.30 (.51)	5.28 (.41)	.15	.88	5.33 (.46)	5.22 (.43)	.83	.41
Motherage	36.85 (3.92)	39.55 (5.24)	−2.02	.05*	37.35 (4.35)	39.42 (5.31)	−1.49	.14
Conduct problems	1.45 (1.22)	2.25 (1.40)	−2.10	.04*	1.60 (1.24)	2.25 (1.45)	−1.68	.10
Hyperactivity/Inattention	1.95 (1.84)	3.36 (2.30)	−2.34	.02*	2.32 (1.78)	3.25 (2.52)	−1.49	.14
Emotional problems	.95 (.95)	1.50 (1.48)	−1.50	.14	1.28 (1.14)	1.21 (1.48)	.19	.85
Peer problems	.91 (1.11)	1.25 (1.18)	−1.04	.30	1.08 (1.12)	1.17 (1.20)	−.26	.78
Prosocial behaviour	8.00 (1.45)	8.54 (1.23)	−1.41	.16	8.48 (1.12)	8.08 (1.56)	1.03	.31
Disorganisation score	3.01 (2.01)	3.67 (1.63)	−1.19	.24	2.30 (1.56)	4.52 (1.41)	−5.21	.00***

Note: ^aHigher education vs. non-higher education (up to lower high school – 15 years; senior high school – 18 years; further education college).

^b In employment vs. not in employment.

^c Married vs. separated/divorced.

*p <= .05; **p <= .01; ***p <= .001.

Attachment security classification agreement between the MCAST and the family drawing task

Table 2 shows that slightly more than half (52%) of the sample was classified secure on the MCAST. On the contrary, based on the drawings less than half of the sample (43%) was classified secure. There was not a prevailing type of insecurity on the MCAST as there were almost equal proportions of children falling under each one of the three types of insecurity. The classification of drawings had a different spread as most children classified insecure met criteria for the avoidant category (47%). As shown in Table 2, the association between the two measures on the 4-way and 2-way attachment classification was statistically not significant suggesting that the agreement was low.

Table 2. Distribution of attachment security classification and agreement between the MCAST and the family drawings (FD) task.

Agreement		MCAST (n = 49)
FD (n = 49)	Secure	Avoidant	Ambivalent	Disorganised	Total
Secure	13	3	3	2	21 (43%)
Avoidant	10	6	4	3	23 (47%)
Ambivalent	2	0	1	1	4 (8%)
Disorganised	0	0	0	1	1 (2%)
Total	25 (52%)	9 (17%)	8 (17%)	7 (15%)	49 (100%)

Note: 2-way classification: x² (1) = 1.74, p = .87; 4-way classification: x² (9) = 9.35, p = .41.

Association of parent-reported social and emotional difficulties with attachment security classification based on the family drawings task and the MCAST, and with the MCAST disorganisation score

Table 3 presents the correlations between attachment security classification across the two measures, disorganisation, and children’s mean scores on the SDQ scales. There was a small to medium strength significant association between drawing-based insecure attachment classification and higher scores on CP (r = .29, p = .04) and H/I (r = .32, p = .02) symptoms. There was also a trend for a small association between drawing-based insecure classification and higher scores on EP (r = .21) and PB scales (r = .20). Additionally, a non-statistically significant small to medium correlation between the MCAST attachment classification and CP (r = .24) and a small to medium statistically significant correlation between the disorganisation score and symptoms of CP (r = .28, p = .05) was found. A non-significant small association was found between drawing-based attachment insecurity classification and disorganisation and a significant large association between MCAST insecure classification and disorganisation (r = .61, p < .001).

Table 3. Correlations between security attachment classification, disorganisation ratings and children’s parent-reported social and emotional difficulties.

	1	2	3	4	5	6	7	8
1. FD secure (=1)	–
2. MCAST secure (=1)	.19	–
3. Conduct problems	.29*	.24	–
4. Hyperactivity/Inattention	.32*	.21	.47***	–
5. Emotional problems	.21	−.03	.27	.20	–
6. Peer problems	.15	.04	.29	.09	−.11	–
7. Prosocial behaviour	.20	−.15	−.25	−.46***	−.15	−.02	–
8. Disorganisation score	.18	.61***	.28*	.01	−.03	−.01	−.09	–

*p <= .05; **p <= .01; ***p <= .001.

We run partial correlations to examine if the significant associations identified in the previous step were independent of sample characteristics that were found to be significantly associated with attachment classification on either measure (details in Table 3): child gender, and maternal education and age. Because CP and H/I were highly correlated (r = .47, p < .001) we controlled for their effects. The partial correlation between drawing-based attachment insecurity classification and symptoms of CP (r = .14, p = .36) and H/I (r = .16, p = .30) were small and not statistically significant. The partial correlations between the MCAST disorganisation score and CP were still within the small to medium range and marginally significant (r = .28, p = .06).

Discussion

The first objective was to examine the concurrent agreement between the two representational measures of attachment classification in a sample of children between the ages of 4 and 6 years old. The association between the two measures on attachment security classification was not statistically significant. These findings do not replicate a previous study which found a significant association and higher agreement between the two measures (Jin et al., 2018). One possible explanation of the difference in findings between the two studies could be attributed to developmental reasons. The children in our study were younger than the children in the study by Jin et al. 2018 (7–9 years old). Although by age five children develop important drawings skills (Sugden et al., 2013), between the ages of 6 and 7 years old more rapid development is observed (Feder & Majnemer, 2007). It is plausible that the drawing skills of the children in our sample lacked the sophistication required to express inner representations of family relationships in family drawings. Use of different versions of the Kaplan and Main (1986) coding classification system could comprise another reason. Jin et al. (2018) report that they relied on overall attachment classification and not on individual signs, whereas the coding scheme we used included part of the original individual signs (n = 53) included in the original Kaplan and Main (1986) scheme (Pianta & Longmaid, 1999). On the other hand, it is worth noting that Pace et al. (2020) also failed to find a match between secure and insecure classification on the family drawings and the MCAST in a small sample of Italian children between the ages of 5 and 8 years old. The findings suggest that further research is required to identify what is the optimal age to use the family drawing task as a valid method of attachment security assessment while ensuring consistency in attachment coding system use.

The second objective was to examine the association of attachment security classification based on the family drawing task with children’s scores on the parent scales of the SDQ. We found that children with insecure drawings were more likely to score higher on CP and H/I, but the associations were not significant once sample characteristics and associated behaviour difficulties were statistically controlled. This finding suggests that the drawing-based classification was not independently associated with child CP and H/I symptoms and it is not in agreement with studies which found associations between behaviour problems and drawing-based attachment classification (e.g. Pianta & Longmaid, 1999; Rehder et al., 2020). Differences in the distribution of child (e.g. gender) and family characteristics (e.g. risk of social disadvantage) as well as larger sample sizes could potentially explain the differences in the findings. However, a few previous studies did not find associations either. For instance, Wagner and colleagues (2015) did not find an association between CP and drawing-based global ratings of attachment in a large sample of non-urban at risk of social disadvantage families. Furthermore, Jin et al. (2018) did not find significant differences in the number of children classified insecure in the clinical vs. community sample, suggesting low association with child psychopathology. The variation in the findings of the association between child social and emotional difficulties and attachment classification based on drawings across studies suggests that further research is required to understand it.

The findings suggest that the family drawing attachment classification is sensitive to individual differences and contextual factors. The partial correlations demonstrated that gender and maternal education effects interfere with the association between drawing-based security classification and symptoms of CP and H/I. Boys are reported to score lower on tests of drawing competence compared to girls (Pianta & Longmaid, 1999). Therefore, it is possible that the association reflects poor drawing performance (e.g. body parts missing, etc.) because of gender differences. Moreover, access to resources and a rich home-learning environment has been associated with the number of years caregivers have spent in education (Jeong, Kim, & Subramanian, 2018). It is plausible that the association between maternal education, behaviour problems, and signs of insecurity in children’s drawings reflect fewer opportunities to engage with activities that advance fine motor skills at a rate that allows children to use drawings to represent the subtleties of family relationships successfully. Future studies should control for drawing and/or fine motor skills, and intellectual abilities to help understand attachment insecurity captured through family drawings in children with behaviour problems.

Attachment classification based on the MCAST was not significantly associated with the SDQ scales, however, there was a significant trend for an association between disorganisation and conduct problems even after controlling for child and mother characteristics, and H/I symptoms. This finding that the MCAST D score can capture disorganisation in attachment security in children with behaviour difficulties regardless of child characteristics and contextual factors aligns with previous research on its robustness as a measure of attachment security (Allen et al., 2018) and provide further support of its validity.

Although not directly related to the study’s main objectives an important finding was that the family drawings classified more children insecure than secure. This finding agrees with those reported in other studies that rely on the individual signs to code attachment security using the drawings. For instance, Pianta and Longmaid (1999) reported that 62% of the children in a mixed-race US-based sample of 5–7-year-olds was classified insecure. Similarly, Behrens and Kaplan (2011) reported 66% of the drawings of 6-year-old Japanese children were classified insecure. Additionally, Fihrer and McMahon (2009) found that 62% of children between 6 and 8 years old were classified insecure. These distributions are different to those reported by Jin et al. (2018) who found less than half of the drawings (38%) classified insecure in a sample of Korean boys. However, their study relied on overall classifications than individual signs. The differences in the findings emphasise the need for consistent use of validated versions of the original task to facilitate replication and ensure that findings are comparable across studies and settings. In the same vein, an important finding was that the drawing task classified a significant proportion of the children avoidant. Using the same classification system, all children in a small Greek-Cypriot sample with an insecure attachment were classified avoidant too (Shiakou, 2012). On the contrary, using the same system, most insecure children were classified disorganised in the Pianta and Longmaid (1999) study. We do not know why there is an overrepresentation of avoidant children in our sample. The findings may either reflect a measurement error or genuine differences in children’s views about family relationships. Our findings provide additional evidence of the complexities in using family drawings to assess children’s attachment classification (Gernhardt, Keller, & Rübeling, 2016; Pianta & Longmaid, 1999) and highlight the need for further research on their psychometric properties.

The findings should be interpreted in the context of certain methodological limitations. The findings may not be applicable to populations with different sociodemographic characteristics. A larger sample size could have revealed further differences and allow comparison across the four different categories of attachment (A/B/C/D), therefore, replication with larger samples is recommended. The main coder of the MCAST was not blind, but a high level of convergence with the reliability coder who was blind was achieved. We did not use additional concurrent measures of insecurity classification based on family drawings, for instance global ratings, which could have helped to understand the classification properties of the Kaplan and Main (1986) system in this age group and the agreement with the MCAST. Unfortunately, we did not collect data on children’s intellectual ability and drawing and/or fine motor skills to help understand the associations with children’s behaviour problems better. Additionally, we did not collect data on attachment with other significant adults which could have provided a fuller explanation of the findings.

Collectively, our findings suggest that further research is required to establish the validity of family drawings as a measure of attachment security in the early childhood years. The lack of a standardised version of the Kaplan and Main (1986) attachment classification system poses challenges in comparing its classification properties across studies. Future studies should contrast large samples of children in late early childhood and children in early middle childhood while using standardised versions of both aggregated and global ratings of drawing-based attachment security classification and controlling for child and parent characteristics. While family drawings should be used cautiously by practitioners to assess attachment security in young children, they are easy and quick to administer and as such can supplement the assessment process to enhance understanding of family relationships and children’s behaviour difficulties. Because the technique is highly sensitive to individual differences and contextual factors concurrent administration with measures of intellectual ability and/or drawing skill is recommended.

Acknowledgements

We would like to thank the participant families and Kostas Dimatis for reliability coding of the Manchester Child Attachment Story Task.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author [AK], upon reasonable request.

Additional information

Funding

This work was supported by Banco Santander [grant number: 2016/25].

Notes on contributors

Angeliki Kallitsoglou

Angeliki Kallitsoglou is a senior lecturer in child psychology and education at the School of Education, University of Roehampton, London, UK. Her research interests include children’s social and emotional development, parenting, parent–child relationships.

Vasiliki Repana

Vasiliki Repana is an early childhood care professional with expertise in early years settings.

Monica Shiakou

Monica Shiakou is an associate professor in psychology.