The concept of the internal working model (Bowlby, 1973, 1983) is of central importance for the theory of adult attachment. Research has shown that the development of attachment-specific internal working models of the self and others influence emotion regulation and interpersonal patterns in childhood as well as later experiences in close relationships (Mikulincer et al., 2003). A further milestone in attachment research is the typology of the secure, avoidant, and anxious-ambivalent attachment style by Ainsworth et al. (2015).

A highly discussed topic is the measurement of adult attachment (Garbarino, 1998; Roisman et al., 2007; Jewell et al., 2019), which can be divided into classifications and dimensional models (Fraley et al., 2015). Furthermore, the instruments differ in regard to the focus on the respective attachment figure (mother, father, current partner) and type of assessment (self-ratings, semi-protective methods, or expert ratings based on interviews). However, the different attachment instruments have no convergent validity or moderate at best (Shaver et al., 2000; Roisman et al., 2007; Jewell et al., 2019; Strauss et al., 2022). It is assumed that convergent validity is higher, the more the two instruments match regarding method, domain, and concept (Bartholomew and Shaver, 1998).

The Bielefeld Partnership Expectations Questionnaire (BPEQ; Höger and Buschkämper, 2002; Höger et al., 2008; Pollak et al., 2008) is a self-rating of partner-related attachment dimensions. With 30 items, the scales “fear of rejection,” “readiness for self-disclosure,” and “conscious need for care” are assessed reliably (Cronbach’s α = 0.79-0.95; Kirchmann et al., 2007; Höger et al., 2008; Petrowski et al., 2010). The items are based on the Bielefeld Client Expectations Questionnaire (BCEQ; Höger, 1999; Pollak et al., 2008) whereby the term “therapist” is often replaced by the term “partner.” The items of the BCEQ are self-descriptions and measure the patient’s expectations regarding the therapeutic relationship. However, the BPEQ is not limited to psychotherapy only, since the questionaire addresses the romantic/marital relationship.

In contrast to other attachment self-ratings, the BPEQ provides dimensional attachment measures and a classification into the attachment patterns secure, partially secure, (insecure) avoidant-withdrawing, (insecure) ambivalent-clinging, and (insecure) ambivalent-withdrawing (Höger and Buschkämper, 2002; Grau et al., 2003). Suggested by Collins and Read (1990) the classification algorithm considers the individual values of three scales (fear of rejection, readiness for self-disclosure, and conscious need for care) and maps them into a small number of attachment patterns based on cluster analyses (Höger, 1999; Pollak et al., 2008). According to Höger et al. (2008), the attachment patterns partially secure and avoidant-withdrawing can be merged to the so-called avoidant attachment pattern, and the ambivalent-clinging and ambivalent-withdrawing can be merged to the ambivalent attachment pattern.

Regarding the validity of the BPEQ scales, hypothesis-conform correlations were found with social support, relationship satisfaction (Petrowski et al., 2010), self-rated attachment anxiety, attachment avoidance (Grau et al., 2003) and attachment security ratings of the adult attachment prototype rating (Kirchmann et al., 2007), which is an interview-based expert rating of attachment (Strauss et al., 1999). The validity of the BPEQ attachment patterns is supported by the findings that the secure attachment pattern was found to be associated with positive parental rearing behavior and the insecure attachment pattern with negative experiences of parental rearing (Schumacher et al., 2004). Furthermore, as expected in a clinical sample of patients with eating disorders, the ambivalent attachment pattern was observed comparably often (Steins et al., 2002).

Recently, a short version of the BPEQ including 12 items (BPEQ-12, in German BFPE-12; Altmann et al., 2018) was developed using two different samples for exploratory factor analysis and confirmatory factor analysis. The reliability of the scales (Cronbach’s α > 0.799) and the concordance of the attachment patterns assessed with the orginal and the short version of the BPEQ were good (κ > 0.6) (Altmann et al., 2018). In a second study, factor structure and reliability were confirmed using a current representative sample from the German population (Altmann et al., 2019). However, the instrument and its factor structure have not yet been tested in patients with a psychological disorder. Furthermore, there is no validity study investigating the convergence of the BPEQ-12 and other attachment instruments.

Table 1 shows the descriptive statistics for both disorder groups. Their comparison revealed that the patient sample included a higher proportion of subjects with a high-school education and more subjects with a firm relationship than the healthy controls. Hypothesis-conform, patients with panic disorder showed more symptom load (GSI), attachment avoidance (ECR-AVO), fear of rejection, and less readiness for self-disclosure and conscious need for care (BPEQ-12). Furthermore, the clinical sample included a larger proportion of individuals who classified as ambivalent attached according to the BPEQ-12, whereas the avoidant attached attachment pattern was more often present in the sample of healthy controls.

The wording of items and item statistics are reported in Tables 2 and 3.

Next, we evaluated the factor structure of the BPEQ-12. The model fit of CFA was acceptable for the entire sample (CFI = 0.973, TLI = 0.966, RMSEA = 0.048, SRMR = 0.063; see Table 4; N = 317), the sample of patients with anxiety disorder (CFI = 0.980, TLI = 0.974, RMSEA = 0.045, SRMR = 0.072), and the sample of controls (CFI = 1, TLI = 1, RMSEA = 0, SRMR = 0.072). In Figure 1, the standardized regression coefficients and correlations of latent variables are listed for the entire sample and both groups. It should be noted that in the analysis of entire sample the standardized loading of item 22 is lower than 0.5. This holds also for item 16 in the sample of patients and item 17 in the sample of controls.

When investigating measurement invariance regarding the presence of an anxiety disorder (patients vs. healthy controls), the change of CFI and RMSEA suggest that, there is scalar, metric and strict invariance (for details see Table 5). However, subsequent analysis revealed that the loadings of item 16 and 17 differ in both groups (both p < 0.05, tests not reported in detail).

For the scales fear of rejection and readiness for self-disclosure we found an acceptable reliability in the entire sample and both sub-samples (Cronbach’s α > 0.7; see Table 6). The reliability of the scale conscious need for care reached an acceptable level only in the sub-group of controls (α = 0.703) but not in the sub-sample of patients (α = 0.619). When quantifying the reliability with McDonalds ω, we obtained very similar results (see Table 6). The AVE was only for readiness for self-disclosure larger than 0.5 (Table 6). The scale fear of rejection narrowly missed the threshold value (but all AVE ≥ 0.489, Table 2).

The evaluation of convergent validity of the BPEQ-12 scales revealed that fear of rejection is mainly associated with higher attachment anxiety (assessed with the ECR-R) and readiness for self-disclosure mainly with lower attachment avoidance. This holds for both the correlation and the regression analysis (large effect sizes, see Tables 7, 8). In the regression analysis, the scale conscious need for care was associated with higher ECR-ANX, higher GSI, and lower ECR-AVO. The corresponding effect sizes were small.

The results reported above suggested that only two scales are reliable and valid. Therefore, we evaluated in a post-hoc analysis a CFA including only fear of rejection and readiness for self-disclosure. All four fit indices were acceptable in the entire sample (CFI = 0.996, TLI = 0.994, RMSEA = 0.027, SRMR = 0.054), the patient group (CFI = 0.997, TLI = 0.996, RMSEA = 0.025, SRMR = 0.065) and control group (CFI = 1, TLI = 1, RMSEA = 0, SRMR = 0.0495). The loadings of two factor model are shown in Figure 2.

Regarding the attachment patterns of the BPEQ-12, we first compared the observed distribution of the five attachment patterns with distributions reported by Höger et al. (2008) and Altmann et al. (2018). Neither the patient sample nor the control sample had a distribution similar to these studies (all p < 0.001). However, we found that anxiety patients had insecure attachment more frequently than healthy controls (medium effect sizes). This holds for both, a categorization into five and three attachment patterns assessed with the BPEQ-12 (see Table 1). Furthermore, we found a significant but slight convergence between the attachment pattern of the BPEQ-12 and the attachment styles of the AAPR (for the three pattern secure, avoidant, and ambivalent: χ²(4) = 31.2^***, V = 0.229, κ = 0.174, N = 298; two patterns secure and insecure: χ²(1) = 12.9^***, V = 0.208, κ = 0.187, N = 298). Multi-nominal regression revealed that the ambivalent attachment pattern of the BPEQ-12 is mainly predicted by the ECR-ANX and GSI, whereas the avoidant attachment pattern is mainly associated with the ECR-AVO (see Table 9). Attachment security assessed by AAPR was neither predictive for the ambivalent pattern nor for the avoidant pattern.

Due to the low convergence between the attachment styles of the AAPR and the BPEQ-12, we conducted the following post hoc analyses. First, we tried to replicate sample partitioning into several attachment patterns proposed by Höger and Buschkämper (2002). For this we applied a latent class analysis (LCA) on the three BPEQ-12 scales. In concordance with Höger and Buschkämper (2002), the Bayes Information criteria, the entropy value, and the parametric bootstrapped likelihood ratio test suggested a solution with five latent classes (respectively, five attachment pattern). However, the concordance of these “new” attachment patterns with the “original” five attachment patterns was weak according to Cohen’s κ [χ²(16) = 291.8^***, V = 0.482, κ = 0.146, N = 314]. Then, we considered a solution of LCA with three latent classes. These three attachment patterns correlated moderately with the “original” BPEQ-12 attachment patterns [χ²(4) = 196.2^***, V = 0.559, κ = 0.488, N = 314]. In other words, our partitioning of the data confirms the three patterns by Höger and Buschkämper (2002). However, the concordance between these new BPEQ-12 attachment patterns and the three AAPR attachment styles was weak [χ²(4) = 19.5^**, V = 0.18, κ = 0.109, N = 314].

In a further post hoc analysis, we determined “new” attachment patterns of the BPEQ-12 using AAPR attachment styles as reference instead of the data-driven method above. We randomly divided the sample into two halves: a learning sample and an evaluation sample. Neither sex, age, high-school education, firm relationship nor the presence of anxiety was able to predict the assignment to the random groups. Based on the learning sample, we computed the averages of the BPEQ-12 scales depending on the AAPR attachment styles. Then, we classified each individual in the evaluation sample according to the smallest distance to these three “cluster centroids.” According to Cohen’s κ, the concordance between the resulting BPEQ-12 attachment pattern and the “original” attachment style of AAPR was significant, but slight [learning sample: χ²(6) = 15.3^**, V = 0.225, κ = 0.157; evaluation sample B: χ²(6) = 17.9^**, V = 0.243, κ = 0.162].

The present study evaluated the factor structure and convergent validity of the short version of the Bielefeld Partnership Expectations Questionnaire (BPEQ-12) in patients with anxiety disorder and healthy controls. The confirmatory factor analysis confirmed the expected structure with three factors and four items per factor in the entire sample, the sample of anxiety patients and the sample of healthy controls. Investigations of measurement invariance suggest strict measurement invariance. However, the loadings of item 16 and 17 are different in both groups.

The reliability was acceptable (Cronbach’s α > 0.7) for all three scales in our sub-sample of healthy controls. Our findings correspond to Höger and Buschkämper (2002), Kirchmann et al. (2007), Petrowski et al. (2010), and Altmann et al. (2018, 2019) (Cronbach’s α = 0.72-0.95), who examined only healthy individuals. However, in our sub-sample of anxiety patients, the scales fear of rejection and readiness for selfdisclosure had good reliability, whereas the scale conscious need for care missed the threshold for acceptable reliability (α < 0.7). Interestingly, in the studies mentioned above, the scale conscious need for care showed the lowest values for Cronbach’s α compared to the other scales. A reason might be that three items of this scale had a comparatively low (standardized) loading in the sub-sample of anxiety patients. Accordingly, in future studies, the scale conscious need for care should be used with caution, especially when anxiety patients are examined.

The convergent validity of the BPEQ-12 scales was examined with correlation and regression analyses. Our results suggest that, despite the fact that the BPEQ-12 measure three dimensions, the BPEQ-12 ought rather be assigned to two-dimensional models (e.g., Bartholomew and Horowitz, 1991; Brennan et al., 1998) than to three-dimensional models (e.g., Bäckström and Holmes, 2007). The former models include the dimensions attachment anxiety and avoidance and derive attachment security based on the constellation of both dimensions. In contrast, the model of Bäckström and Holmes (2007) measure attachment anxiety, avoidance, and security directly. Our results from the regression analysis showed that attachment security assessed with AAPR is not associated with any of the BPEQ-12 scales, whereas fear of rejection is related to attachment anxiety of ECR-R and readiness for self-disclosure to attachment avoidance of ECR-R. The lack of convergent validity of the scale conscious need for care corresponds to findings by Grau et al. (2003) who examined the original version of BPEQ. All in all, the low reliability and low convergent validity suggests the exclusion of the scale conscious need for care. However, we cannot rule out that sufficient reliability will be found in other clinical samples. Also, it is possible that convergence might be found with other attachment measures not considered in the present study. For the computation of attachment patterns according to Höger and Buschkämper (2002), all three scales must be assessed.

Furthermore, our results suggest discriminant validity. We found medium correlations (r > 0.3) between the symptom load measured with BSI and the three attachment scales of BPEQ-12. This means that the BPEQ-12 measures not the distress due to various psychological symptoms, but a related concept.

Furthermore, we evaluated the classification into the attachment patterns according to Höger and Buschkämper (2002), Grau et al. (2003), and Höger et al. (2008). Indeed, our regression analysis revealed that the avoidant attachment pattern of the BPEQ-12 is mainly predicted by the ECR-R scale attachment avoidance and that the ambivalent attachment pattern of the BPEQ-12 is associated with the ECR-R scale attachment anxiety. This corresponds with Grau et al. (2003). Moreover, the resulting patterns of an LCA with three latent classes correspond to the three patterns (secure, avoidant, ambivalent) classified according to Höger et al. (2008). However, the fit indices of our LCA indicate a solution with five patterns which correspond only slightly with the “original” five patterns of Höger et al. (2008) (see first post hoc analysis).

A critical point is the low concordance between the attachment patterns of the BPEQ-12 and the attachment styles of the AAPR. This low concordance is also present for the BPEQ-12 attachment pattern classification that was developed based on the attachment styles of the AAPR (see second post hoc analysis). In contrast, we found evidence for the convergent validity of the BPEQ-12 and ECR-R scales. These findings support the assumption by Bartholomew and Shaver (1998): the convergent validity is higher when the method of both considered attachment instruments matches (BPEQ-12 and ECR-R are self-ratings) and lower when the method does not match (the AAPR is an interview-based expert-rating).

A general point to discuss is whether there exist only the three attachment patterns secure, ambivalent, and avoidant or whether mixed patterns may also be possible (e.g., partially secure in the model of Höger et al., 2008). The attachment concepts behind the BPEQ (Höger et al., 2008) and the AAPR (Strauss et al., 1999), for example, assume that an individual can have secure and insecure “parts.” Both instruments measure attachment-relevant dimensions in the first step. In the AAPR, experts rate the degree of presence of seven prototypes on the basis of video-recorded interviews. In the BPEQ, the subjects themselves rate the degree of the three attachment dimensions. In a second step, the dimensions are mapped into a small set of attachment patterns (resp. styles) using an instrument-specific algorithm. When two attachment dimensions have a similar intensity (e.g., attachment security and attachment avoidance), the subject can be assigned to different attachment patterns depending on which attachment dimension is slightly more present. The BPEQ (Höger et al., 2008) avoids such difficult to decide classifications with additional mixed patterns (e.g., partially secure). However, the simultaneous presence of secure and insecure “parts” may be one reason why dimensional models of attachment are better suited for measuring individual differences (Fraley and Roisman, 2014; Shi et al., 2014; Fraley et al., 2015; Jewell et al., 2019). The classification in a small subset of attachment pattern (or styles) based on attachment dimensions include a loss of information. However, in other contexts, categorical models are more suited, e.g., to describing different types of patients and their attachment-related behavior in the context of medical care (Strauss and Brenk-Franz, 2016). The benefit of the BPEQ-12 is that the instrument provides attachment dimensions and categories.

One of the limitations is the focus on one specific disorder group (panic disorder and/or agoraphobia). Also, the clinical sample was not representative. Furthermore, the appropriateness of the AAPR as reference is limited by its moderate inter-rater-reliability and the untypical distribution of the attachment styles of this instrument. Moreover, it should be noted that, similar to Altmann et al. (2018), the long version of the BPEQ was assessed and that the analyses were thus based on 12 items selected from the long version. Last, the sample size of both groups was below N = 250 which leads to biased model fit indices (especially TLI and RMSEA) (Hu and Bentler, 1999).

The BPEQ-12 is suitable to measure partner-related adult attachment. However, only the reliability of the scales fear of rejection and readiness for self-disclosure was acceptable. The former showed convergent validity with higher attachment anxiety and the latter with lower attachment avoidance assessed with the ECR-R. The scale conscious need for care should be used with caution. It missed the threshold for acceptable reliability and had low convergent validity regarding attachment anxiety, attachment avoidance, and attachment security. Despite the measurement of three scales, the BPEQ-12 can be assigned to the two-dimensional attachment models including attachment-avoidance and attachment-anxiety.

The datasets presented in this article are not readily available because they are still used for further analyses. The data that support the findings of this study are available from the senior authors, BS and KP, upon reasonable request. Requests to access the datasets should be directed to BS, bernhard.strauss@med.uni-jena.de.

The studies involving human participants were reviewed and approved by Committee of Jena University Hospital, Jena, Germany (ID 3060-02/11). The patients/participants provided their written informed consent to participate in this study.

UA conducted the data analysis and did the first drafting. KB-F contributed with literature research. BS contributed with statistical advice. KP conceptualized the design of the research and provided helpful advice for drafting and statistical analysis. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.