Attachment and Depression: The Relationship Between the Felt Security of Attachment and Clinical Depression Among Hospitalized Female Adolescents

Abstract

Introduction

To examine the attachment construct in terms of felt security, angry distress and perceived unavailability in hospitalized depressed adolescent females.

Method

A case comparison study of 51 adolescent females with depression and 22 adolescent females with other diagnoses was investigated.

Results

In a multivariable logistic regression model, lower selfesteem, lower peer support, greater angry distress and perceived unavailability in interaction with family type increased the likelihood of membership in the depressed group.

Conclusion

Absence of felt security increases the chances of having a diagnosis of depression.

INTRODUCTION

A long tradition of research has implicated family background factors in the etiology of depressive disorder (Whitesell & Harter, 1996, Puskar et al, 1999). Although family background factors have been among the most discriminative variables of depression in adolescents, previous studies have seldom been guided by adequate theoretical models and have often used inappropriate comparison groups. The present study was grounded in attachment theory and compared two clinical groups, one with depression and the other without.

The attachment system, as described by Bowlby in 1969, functions to promote proximity to caretakers in the service of care and security. Repeated relationship experiences with caregivers determine the adolescent’s perception of the responsive availability of the attachment figure, “felt security” (West et al, 1999). Continued lack of success in meeting the need for security leaves the adolescent prone to low self-esteem, anger and an increased likelihood of experiencing feelings of depression. Unable to view the self as competent, and, at the same time, unable to form and sustain the relationships essential for maintaining self-esteem, this adolescent is vulnerable to depression. From an attachment standpoint, therefore, the greater the lack of success in achieving felt security, the greater the probability of developing a depressive disorder.

Attachment theory provides a framework for explaining not only etiological aspects of depression. Researchers identify the value of the attachment models for both assessment and treatment (O’Connor & Zeanah, 2003; Diamond, Siqueland, & Diamond, 2003). The key to advancing this agenda lies in the ability to measure attachment in clinical settings relation to psychiatric disorders, such as depression. While it is common for studies to examine a relationship between attachment and depression in community samples (Bifulco et al. 2003; Rice, 1990), only a few studies have examined this relationship in clinical samples (Allen, Hauser, Borman-Spurrell, 1996; Armsden, McCauley, Greenberg, Burke, & Mitchell, 1990; Rosenstein & Horowitz, 1996). As a result, a number of barriers exist in developing attachment-based intervention methods in clinical populations. This rests upon an absence of established treatment guidelines or consensus regarding the mechanisms of change in relation to attachment and depressive disorders (O’Connor & Zeanah, 2003). Foremost among these barriers is the ability to economically and accurately measure attachment and depression within clinical populations.

This study was a case-comparison study of the association between felt security and depressive disorder among adolescents in psychiatric treatment. Female adolescents are more likely to receive psychiatric treatment for depressive disorder than male adolescents, this study focused on depression in females (Nolen-Hoeksema, 1990, Offord et al, 1987). We compared two clinical populations. The case group included female adolescents diagnosed with depressive disorder according to DSM-III-R criteria. We expected that female adolescents with low felt security, perceived unavailability of the attachment figure and high angry distress would be over represented in the case group of female adolescents with depressive disorder.

METHODS

Following the conjoint ethics committee approval of the study and based on informed consent, 73 female adolescent inpatients agreed to participate in the study at the time of their admissions. Inclusion criteria consisted of admission to the inpatient adolescent psychiatric unit for treatment. Exclusion criteria included thought disorders, psychoses, or developmental disorders that might have interfered with an individual’s ability to complete the questionnaires. All participants completed a computer-based diagnostic interview and measures of attachment, cognitive, and psychosocial variables.

Outcome Variable: Diagnostic Category Structured Diagnostic Interview

The computer-based Revised Version of the Diagnostic Interview Schedule for Children (Piancentini et al, 1993) was used to assign participants to the outcome diagnostic groups. The case group consisted of participants with an outcome diagnosis of dysthymia, or major depression, or both. The comparison group included participants with other psychiatric disorders. We also included a count of the total number of comorbid diagnoses within each outcome group.

Attachment Variables Adolescent Attachment Questionnaire

The Adolescent Attachment Questionnaire (AAQ) was used to assess the level of felt security that each participant perceived in her primary attachment relationship. West, et al in 1998 designed the AAQ scales around theoretically important features of the adolescent-parent attachment relationship. In this study, the AAQ consisted of two scales - Angry Distress and Unavailability. The AAQ measures perceptions regarding the availability of the primary attachment figure and the angry distress held in regard to this figure.

Cognitive Variables Beck Hopelessness Scale

The Beck Hopelessness Scale (BHS) is a 20 item, true-false response, self-report instrument that measures negative beliefs about future expectations (Beck et al, 1974 & Owen, 1989).

Rosenberg Self-Esteem Scale

Self-esteem is the belief that one is adequate to confront and master situations and that one is deserving of love (Rosenberg, 1965). This scale measures global self-esteem. In comparing the RSE and another self-esteem inventory, Crandall found a Pearson correlation (r) of 0.60, demonstrating acceptable convergent validity.

Social Support Variables Perceived Social Support from Family and Friends Scales

These two scales each consist of a twenty-item questionnaire measuring an individual’s belief that family members or friends provide support, sharing, involvement and sensitivity. Construct validation studies have indicated that higher perceived social support is associated with lower psychopathology and higher social competence (Lyons et al, 1988, Hosmer & Lemeshaw, 1989).

Statistical Analysis

Multiple logistic regression was the method employed to combine the qualifying variables in order to address the question, “Which of the independent variables (e.g., attachment) best described membership in the outcome group with depression?” Hosmer and Lemeshow in 1989 and Kleinbaum in 1998 have provided methods for the selection of variables as candidates for a multivariable model. Remaining in the multivariable model required that a particular variable made a statistically significant contribution to the model. The cut point or alpha was set apriori to testing at p< 0.10, based on the power associated with the sample size. This is lower than the arbitrary standard of p < 0.05 thereby increasing the probability of accepting a false difference in direct proportion to the probability of accepting a true difference. Indeed, when relevant , Hosmer and Lemeshow (1989 ) have argued that it matters little even to allow insignificant variables remain in the model for the sake of context and description.

RESULTS

Description of the Sample

Seventy-three females comprised the sample. All were inpatients on a psychiatric treatment unit at the time of data collection. Age was normally distributed with a mean of 15.5 years, a standard deviation of 1.4 years and a range from 13 to 18 years.

Demographic Comparison of the Diagnostic Outcome Groups

Fifty-one participants were in the case group with depression and 22 were in the comparison group with other diagnoses. Age did not distinguish between the groups. Table 1 shows the family composition of the group with depression and the group with other diagnoses. There were more participants from “Biological” and “Single Parent” families in the depressed group and slightly more participants from families with “One Step Parent” in the group with other diagnoses.

Table 1.

Family Composition of the Outcome Groups

Diagnosis
Family Composition	Other	Depression/Dysthymia	Total
Biological Parents*	10 (0, 30, [.16 .49])	23 (0.70, [.51, .84])	33
One Step-Parent*	8 (0.57, [.28, .82])	6 (0.43, [.18, .71])	14
Single Parent*	2 (0.11, [.01, .35])	16 (0.89, [.65, .99])	18
Blood Relative	0	1	1
Adoptive Parents	0	3	3
Other Guardian	2	2	4
Total	22	51	73

^*(Proportion in each outcome group, [95% CI]) are shown for these strata.)

Sixty-eight of the participants identified their biological mothers as their primary attachment figures, while three identified their adoptive mothers, and only one participant identified her father. The variable attachment figure did not distinguish between the depressed participants and those with other diagnoses.

Comorbidity

The mean number of other comorbid diagnoses in the depressed group, controlling for dysthymia and depression, was 7.28 (95% C.I. 6.36, 8.20). The mean number of comorbid diagnoses in the group without depression was 3.00 (95% C.I. 2.11, 2.89) including the primary diagnosis.

Cognitive Variables

The scores of the cognitive variables Self-Esteem and Hopelessness were higher for the group diagnosed with depression. The mean value of the cognitive variable Hopelessness was 12.59 (90% confidence interval: [11.27, 13.90]) in the group with depression. The mean value in the group with other diagnoses was 6.32 (90% confidence interval: [4.33, 8.31]). Higher scores indicated greater hopelessness. The mean values of the cognitive variable Self-Esteem were 30.29 (95% confidence interval: [28.88, 31.71]) for the group with a depressive disorder and 23.78 (90% confidence interval: [21.66, 25.88]) for the group with other disorders. Higher values indicated lower self-esteem.

Perceived Social Support Variables

The values of the Perceived Family and Friend Support variables were lower for the group diagnosed with depression. Perceived Family Support had a mean value of 6.76 (90% confidence interval: [5.61, 9.92]) in the depressed group and a mean value of 9.55 (90% confidence interval: [7.33, 11.76]) in the group with other diagnoses. Perceived Friend Support had a mean value of 11.49 (90% confidence interval: [9.95, 13.03]) in the depressed group and a mean value of 15.5 (90% confidence interval: [13.70, 17.30]) in the group with other diagnoses.

Attachment Variables

The attachment variables had higher scores for the group diagnosed with depression compared to the group with other diagnoses. Greater values in the perceived attachment variables indicated higher angry distress and greater unavailability. The mean values of the perceived attachment variable Angry Distress were 8.96 (90% confidence interval: [8.32, 9.60]) for the group with a depressive disorder and 6.68 (90% confidence interval: [5.88, 7.49]) for the group with other disorders. The results were similar for the mean values of the perceived attachment variable Unavailability; 7.90 (90% confidence interval: [7.48, 8.32]) was the value of the mean in the group with depression and 6.45 (90% confidence interval: [5.78, 7.13]) was the value of the mean in the group with other diagnoses.

Multivariable Model Predicting Diagnostic Group Membership

In the case of the variable family composition, the data structure gives rise to the type of numerical instability that results from spreading the data over too many categories when the sample size is relatively small in relation to the number of variables. To overcome this problem Hosmer and Lemeshow in 1989 suggested collapsing the categories in a meaningful way. For Family Composition (Table 1), this meant combining the categories into two groups; one group (value 1) included categories with an increased likelihood of membership in the group with depression (e.g. Biological Parents, Single Parents, and Other), and the other group (value 0) included the category with decreased likelihood of membership in the group with depression (e.g. One Step-Parent). The two category variable predicted diagnostic outcome. Compared to families with One Step-Parent, all the other constellations increased the probability of membership in the depressed group (beta coefficient 1.46, p < 0.02). This two category variable, Family Type, was included as a candidate in the selection of variables for the final model as it resolved to some extent the problem of numerical instability.

There were no significant interactions between Family Type and the other predictor variables, with the exception of Unavailability. An interaction term must be relevant (e.g., biologically meaningful) and statistically significant to be considered in the model (e.g., p < 0.10) (Hosmer & Lemeshow, 1989). In the analysis that modeled the interaction of the perceived attachment variable Unavailability and Family Type, the interaction term accounted for the increased probability of membership in the depressed group. In addition to the main effect of family type, the level of perceived Unavailability of those who had either biological parents, a single parent, or a family composition from the other groups collapsed into this variable increased the likelihood of membership in the depressed group.

Best Fit Model Considering All the Variables

While all variables but Age were eligible to enter the multivariable model, only the perceived attachment variable Angry Distress, the cognitive variable Self-esteem, the social support variable Perceived support from Friends, and the two category Family Type variable and this variable in interaction with the perceived attachment variable Unavailability were eligible to remain in the multivariable model. The final model is shown in Table 2 (model Chi square = 39.76 with six degrees of freedom; p < 0.0001).

Table 2.

Multivariable Logistic Regression Model Differentiating Outcome

n=73	Estimated Coefficient	Standard 'Error	z	p<	90%CI	Log Likelihood
Constant	−2.91	3.30	−0.88	0.377	−8.33, 251	−24.80
Family Type	−5.52	3.29	−1.68	0.093	−10.9, −0.11
Unavailability	−0.30	0.35	−0.87	0.386	−0.88, 0.27
Family Type – Unavalability	1.01	0.49	2.06	0.039	0.20, 1.81
Interaction
Angry Distress	0.25	0.14	1.77	0.076	0.02, 0.47
Self-Esteem	0.17	0.07	2.44	0.015	0.06, 0.29
Friend Support	−0.13	0.07	−1.90	0.057	−0.25, −0.02

Selection included the predictor variables having beta coefficients with p values less that 0.10, based on the power associated with the sample size. Likelihood-ratio tests were used to develop the final model shown in Table 2. The reduced model accounted for membership in the depressed group as adequately as the model with all the qualifying variables. Membership in the depressed group was more likely for those with lower selfesteem, lower peer support, and greater angry distress, together with greater unavailability in interaction with the variable representing family types other than the step-parent category.

DISCUSSION

In this case control study of the association between female adolescents’ felt security and depressive disorder in adolescents in psychiatric treatment, the expected predictions were confirmed. In particular, membership in the depressed group was more likely among those reporting higher levels of angry distress and perceived unavailability. Consistent with attachment theory, the failure to experience felt security in her primary attachment relationship increased a female adolescent’s risk for membership in the group with depression.

The study results support the model postulating a relationship between perceived attachment and depression. The concept of felt security embodies the attachment relationship quality that individuals perceive in regard to their parents. “Felt security” stems from experiences with the caregiver and reflects the conscious investment of affect held in regard to an attachment figure. Aspects of felt security, measured in this study, as angry distress and unavailability, appear to represent features of depressive disorder as important as cognitive distortions such as self-esteem and hopelessness, as well as perceived social isolation or support. In the final multivariable analysis the important main effect variables describing membership in the depressed group were the perceived attachment variable angry distress, the cognitive variable self-esteem and the social support variable perceived family support. These findings support the concept that attachment failure may be a core aspect of the depression.

With regard to Family Type, the results of the present study are most comparable and, indeed, closest to the results of the clinical study of Puig-Antich et al in 1993, who used a similar categorization of family composition. As with the Puig-Antich et al study more depressed female adolescents in the present study came from biological, single parent families, and other compositions, while similar proportions in each of the case and comparison groups were from families with stepparents. The increased risk of biological families in this sample may be explained in terms of the long term entrenchment of maladaptive family interactions that form a depressed adolescent’s perceptions and feelings about herself and her attachment figure. Single mothers have often left an untenable relationship. While single and caregivers, they must meet and attend to all the needs of their families, which make the single mothers less available for the purpose of attending to and meeting the emotional needs of their children. A large proportion of depressed female adolescents were from single parent, and like biological families, this group viewed their caregivers as less available, in attachment terms. Support for this idea comes from the final model and the observation that in this study 16 of 18 adolescent females who came from single parent families were in the depressed group. The interaction term comprised of the perceived attachment variable unavailability and the variable family type was the foremost variable in terms of statistical magnitude and significance.

This study has several limitations. We have used a selfreport measure of attachment (AAQ) rather than a structured interview (e.g., AAI) as an index of attachment. As a result, we cannot explicitly define attachment categories along the traditional lines of attachment theory. Rather, we can only indicate that there was a tendency within the diagnostic groups to perceive felt security of attachment in terms of the AAQ scale labels. Also, the study was cross-sectional, which precludes identification attachment pathology as a precursor of the clinically depressed state. Additionally, the small sample size makes interpreting the multivariable model difficult due to the levels of stratification extent within the dataset. We have left apparently insignificant variables in the model because they are of interest and would likely be significant were the sample size to increase. A larger sample would overcome many of the obstacles related to interpretation. Even with the limitations of sample size, the variables were significant in bivariate analyses and did not appear to change much in the multivariable analysis in magnitude or direction. Rather, they changed in significance alone indicating that the associations persisted though were dampened somewhat by the small sample size. Finally, while perhaps useful and of interest to clinicians, the findings cannot be generalized to a nonclinical adolescent population.