Before the specific aims of the study can be detailed it is important to outline some of the methodological issues raised by a study such as this. These relate to the difficulties in accessing representations.
Difficulties arise when faced with the assessment of representations. First, although there is a long history of theorising about the nature of representations it is unclear how they are organised.
“We do not know whether representations .... are to be viewed as the product (some kind of entity) of various processes or as a procedure, a process itself to permit re-experiencing.” (Stem, 1995, p. 19)
Second, regardless of whether they are an entity or a process, by their nature they are internal, and as a consequence cannot be accessed directly. Therefore, representations can only be revealed through inference.
As a consequence of the lack of clarity of the stmcture of representations, in much of the recent research into the area, two methods have been adopted to infer representations: semi-structured interviews leading to some narrative account or verbal description of a person, and more stmctured rating-scales. In the language based accounts the representations are qualified by an independent observer, whereas in the stmctured rating scales the representations are qualified by the participant. The employment of both methods in many studies reflects the lack of agreement as to which method is more accurate in accessing
representations. In keeping with current research this study adopts both methods, and these are discussed below.
Inferring Representations Through Language: Free-Response Descriptions
With the proposal of Main et aL (1985) that internal Working Models of relationshçs not only direct behaviour but also “attention, memory, and cognition,” (p. 67), internal models could be inferred not only through observation of behaviour, but
through ‘"patterns of language and structures of mind”. That is, individual differences in representations of self and others will have influence on that person’s memories and thoughts, and as such will affect the way that the individual constructs narrative accounts of self, others, and relationsh%)s. \^^th the development of a semi
structured interview: the Adult Attachment Interview (AAI) (George, Kaplan and Main, 1985), which probes for accounts of childhood experiences with caregivers, inferences could be drawn about the individual’s current mental representations of attachment, through the qualitative differences in the verbal expression of these experiences. The narrative accounts are transcribed verbatim, then rated on a number of dimensions using conçrehensively described operational definitions of the
dimensions. These dimensions measure not only wkat is said (the content of the narrative), but also the way in which accounts are given (the structural dimensions) e.g. the coherence of the descriptioa Research has demonstrated both aspects to be inç>ortant when inferring the nature of internal representations. However, as we have already stated, the AAI leads to inferences about high order global representations, and does not yield explicit information regarding lower order representations of specific individuals, and particularly representations of a particular child (Zeanah and Barton 1989).
Since the development of the AAI, Benoit et a l (1997), have noted that several groups of researchers have begun to redress this issue, developing instruments designed specifically to explore mothers’ representations of their children and their relationship with them (Aber et a l 1989, Bretherton et a l 1989, and Cramer et a l
these tools. In response they developed the Working Model of the Child Interview (WMCI) (Zeanah, Benoit and Barton, 1986 and Benoit et al. 1997). The technique is similar to that enç)loyed in the AAI, using a semi-structured interview exploring aspects of the mother’s perception of the child and accounts of interactions with the child. Similarly the tra n s c r^ of the interview are rated along a series of constructs. In addition transcripts are classified into one of three categories: balanced, disengaged or distorted, depending on the quality and structure of the inferred representation.
Despite the development of robust instruments for assessing representations such as the AAI and WMCI, due to their specific nature, none of them allows for inter- generational conq)arisons of representations. Le. of child, self, and own mother. However, recently Blatt and colleagues (Blatt, Wein, Chevron and Quinlan, 1979, Quinlan, Blatt, Chevron and Wein, 1992, Bers, Blatt, Sayward, and Johnston, 1993, and Blatt, Bers and Schaffer, 1993), have devised a technique, with reasonable psychometric properties, which would allow for such conq)arisons. The method they devised conpiised a free-response procedure where particçants were singly asked to give desertions of either themselves, (Bers, Blatt, Sayward, and Johnston, 1993) or their parents (Blatt, Wein, Chevron and Quinlan, 1979, and Quinlan, Blatt, Chevron and Wein, 1992). In the original procedure the response was written down by the subject (Blatt et al. 1988), but in later studies the respondents gave verbal accounts which were transcribed verbatim (Bers et al. 1993). Once transcribed the deselections were analysed using a series of rating scales which assess both the content and the structure of the representations. Later the methodology was extended to include analysis of self descriptions (Bers et a l, 1993). Much of the
rating system was re-designed to be sensitive to the inportant aspects of such descriptions.
Such free-response methods of eliciting internal representations have been shown to be effective elsewhere (e.g. Rosenberg, 1977). Althou^ the method of analysis was different from that used by Blatt and colleagues^ Herzog et al. (1992), have shown free-response methods to be useful in eliciting representations across the generations, i.e. of Self, Child, and Own Mother. In a study of mothers identified as at risk of abusing their own children, they found that these mothers produced highly elaborate representations of their child, but these descriptions showed distinct adultamorphic features. Representations of Self in this group were either highly disjunctive with representations of parents (hypothesised as a defence against identification with the abuser) or produced ‘equivalent’ representations (identification with the abuser). ‘Normal’ adults are expected to fall somewhere between the two (Rosenberg, 1988). Those who exhibit the disjunctive pattern were felt to have unrealistically high expectations of their child; those who demonstrate the equivalence pattern tend to model their beliefs about the child as being inherently ‘bad’. However, those mothers that were able to reflect on and integrate both acceptable and unacceptable past experiences were able to be more accepting of their child’s behaviour and feel better able to cope^. Clearly, abuse is at the extreme end of negative experience, however, a similar mechanism may exist for less extreme experiences leading to mothers holding unrealistic expectations of their child.
This study adopts such a free-response method for inferring representations from verbal descriptions, as one of the measures of representations.
The analysis of the descriptions is based on that used by Blatt et al. (1993). However, in order to conpare inter-generational representations, it has been
necessary to adapt the rating scales by omitting those scales which cannot be used to rate descriptions of Own Mother, and Child, as well as Self. In addition to those scales developed by Blatt and colleagues to rate self descriptions, further scales were used to retain the sensitivity of the inportant relational characteristics of descriptions of others that where included in the earlier versions of the Blatt measure (Blatt et al.
1988). However, it was felt that the operational definitions of the concepts in the original rating system ençployed by Blatt et al. (1988), were not sufficiently well developed. It was, therefore, decided to enploy the operational definitions from the AAI (Main and Goldwyn 1993). These operational definitions are much more detailed than those described in Blatt et al. (1988), and capture, more explicitly, the individual’s perceptions of the relationshç) with the other.
Also, again due to the less well developed operational definition in the Blatt et al.
(1993), manual, the scale for rating ‘Self Reflectivity’ was replaced with the more robust definition of the ‘Reflective-Self Scale’ produced by Fonagy et al. (1991), which has been demonstrated to be an extremely in c ita n t predictive variable in language based accounts inferring representations. Finally, it was felt inportant to reflect a more global rating for each description which combined both the content and
structural features of the representation. In order to achieve this the categorisations developed from the WMCI were used, to categorise each of the three descriptions.
Inferring Representations Using Rating Scales: Repertory Grids
In addition to inferring representations from narrative or descriptive accounts, a number of studies have used some form of structured descriptive rating scale. In most cases these have taken the form of lists of dichotomous pairs of adjectives (e.g. Good Vs. Bad etc.). Typically, participants are asked to rate either themselves, their child, or their own mother along a continuum depending on which adjective best describes that person. These scales are variously referred to as “bipolar Likert scales’ (Fava Vizziello et a l 1993), “Person x Features Matrix” (Herzog, Gara and
Rosenberg 1992), or “semantic differential scales” (Blatt et a l 1979, and
Ammanitti et a l 1992). These methods bear a great similarity to methods used in Personal Construct Psychology (PCP) (Kelly 1955), for eliciting an
individual’s construct system (or more accurately specific parts of it): Repertory Grids.
Using Repertory Grids, individuals typically are asked to rate a series of individuals or objects (referred to as elements in PCP) against a series of dichotomous adjectives (referred to as constructs), similar to those methods mentioned above. These constructs and elements can be generated by the participant to produce a more idiographic view of the construct system. However, constructs and elements supphed by the researcher enable cross
participant comparisons (Fransella and Bannister, 1977). Once each element has been rated along each construct pole, it is possible to analyse the resulting elements x constructs matrix or grid, to explore the relationships between constructs, and between elements, and between constructs and elements. In so doing it is possible to inq)ly how a person represents the elements being assessed, and how these representations are organised in relation to one another; basically how similar or different they are. In order to understand this, it is important to consider what Hes behind the judgement an individual makes when rating an element along a construct pole. Fransella and Bannister (1977), explain it thus,
“Behind each single act of judgement that a person makes (consciously or unconsciously) lies his in^licit theory about the realm of events within which he is making judgements. The repertory grid technique i s a way of exploring the structure and content of such implicit theories.” (p. 2)
The strength of the Repertory Grid method is that it allows not only for
representations of individuals to be analysed, but also allows for the relationships between them to be assessed. One possibility with the grid analysis is to examine distances between currently held representations and an ‘ideal’ the participant may have by asking them to rate both ‘real’ and ‘imagined’ elements. In this way it is possible to obtain a more idiographic picture as to how a mother evaluates her current representations. In a sense it is a structural feature that provides a more subtle measure of an individual’s evaluation of a representation. Grid
analysis also allows for the intensity of the individuals construing to be examined: the extent to which elements are differentiated.