1

A model for assessingadult attachment

John Kirkland*^David Bimler^Eva Klohnen^^

^Massey University^^University of Iowa

2

Some approaches

Several approaches have been advanced for assessing Adult Attachment.

These have to migrate into distinct “two cultures”:– Developmental– Social psychology/personality(see references for details)

3

Our approach, here

We take a “social psychological/personality” approach here. This research is very much “in progress”.

Our aim is to advance assessment through systematic review of an existing data base and offer an alternative model.

A caveat: How this may correspond to facets of the

“developmental approach” is beyond the scope of our current project.

4

The frame

The starting point was a recent theoretical and measurement framework offered to accommodate the major approaches to adult attachment measurement and conceptualisation (Klohnen, 2005).

5

This frame accommodates

Major social personality approaches:- two major dimensions – anxiety &

avoidance- four “types” or “styles”

Major developmental approaches:- security-insecurity and hyper- and

deactivation (45 deg offset from anxiety and avoidance)

6

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

7

8

“Attachment” adjectives

Based upon this 2-D framework, 8 adjectives were chosen for each octet from a representative pool of 196.

It was thought these adjectives would be the best markers for these zones.

9

Examples

Secure enthusiastic outgoing playful sociable warm optimistic cooperative contented

10

Examples

Dismissing individualistic independent self-reliant tough stern indifferent unemotional harsh

11

Empirical test of 8 circumplex scales

1000 participants (Ps) rated all 196 adjectives using Likert scales.

Computed scores for all 8-item scales for each P; standardised scores (mn=0,sd=1).

Ps also completed independent attachment measures (eg obtained attachment style data).

12

Graphic outputs

The following radar-like graphs show each of the four “attachment styles” scored on the 8 adjective scales.

Thus, scores on all 8 scales can be displayed for visual inspection.

13

Typical Scales Results - Secure

-100

-80

-60

-40

-20

0

20

40

60

80Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

Secure

14

Typical Scales Results - Dismissing

-100

-80

-60

-40

-20

0

20

40

60

80Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

Dismissing

15

Typical Scales Results - Fearful

-100

-80

-60

-40

-20

0

20

40

60

80Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

Fearful

16

Typical Scales Results - Preoccupied

-100

-80

-60

-40

-20

0

20

40

60

80Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

Preoccupied

17

Typical Scales Results - All

-100

-80

-60

-40

-20

0

20

40

60

80Positive Other

Preoccupied

Negative Self

Fearful

Negative Other

Dismissing

Positive Self

Secure

Secure Dismissing Fearful Preoccupied

18

First impressions

There is definitely a trend here –

For instance, those selected items, placed within the eight groups do appear to fit into a theoretical attachment frame.

19

Our current approach

For the present study we began from a different place.

First, deleted 46 items from the 196, eliminating those which provided little contribution to the 4 attachment “types”.

Then, by applying inter-item sorting routines we “mapped” these 150 items.

This inter-item “semantic” map was subsequently adjusted (slightly) using the 1k Likert-based data set.

20

Some “maps”

We now present some maps to represent various outcomes from our analyses.

As a little warning, map viewing does require some spatial agility.

21

Map 1

“Items in space”

Shows inter-relationships amongst items spread in 3-D, as if on the surface of a hollow sphere.

22

Map 1 – Item Points

Map1.plo

23

Map 2 – Labelled Items

Map2.plo

24

Map 3

We then apply “dimensions” for slicing the sphere into manageable sections.

Dimensions can be shot through anywhere since the inter-item array itself does not change with alternative axes.

Selected dimensions may be checked for face value by interpolating any item’s location along each dimension’s axis to see if this “makes sense”.

Clearly, the “meaning” of these dimensions will be a function of their locations. But these are for illustrative purposes only and should not be taken as definitive. They are as “stepping stones” leading to a larger project.

25

Map 3 - Dimensions

Map3.plo

26

Map 4 – Items marking this selection

Map4.plo

27

Map 5 - Dimensions with all items

Map5.plo

28

Map 6

Aligning the 8-item groups within the map itself.

Any predictions?

29

Map 6 - Circumplex

Map6.plo

30

Map 7

We can then compute vectors by multiple regression.

We used published aggregated ratings from 7 “experts”.

They provided Likert-like ratings for 196 adjectives for each of the classic “types”: dismissing, secure, preoccupied and fearful.

31

Map 7 - Vectors

Map7.plo

32

Map 8

The map can be rotated to show there is a contribution of D3 (whatever that is) and this would have a bearing on interpretation. There is more work to be done on choosing optimal dimensions.

This outcome suggests a 3-D circumplex rather than a 2-D one fits these data better.

33

Map 8 – Rotated for Vectors

Map8.plo

34

Map 9

As an example of how “experts” and lay (the 1k Ps) persons differ in their word use, consider the spatial differences between experts’ vectors and the actual words.

This outcome suggests a better fit between experts and lay participants will occur when those “troublesome” words (having the largest separation) are removed. This is what we did for our revised approach, see below.

35

Map 9 – Lay and Expert

Map9.plo

36

Trilemmas

As an alternative to Likert-like scales for data collection we propose trilemmas provide rich data for less work.

Trilemmas are created by optimising equilateral triangles within the map.

Each trilemma contains items that are uncorrelated.

Within sets each item is listed once.Many parallel trilemma sets can be constructed

by changing the combinations.

37

38

Map 10

For example, one expert completed 50 trilemmas for each of the 4 classic attachment styles.

These results were plotted against the 7 expert prototype ratings (using Likert-type scales for all 196 adjectives).

The next plot shows relevant outcomes.

39

Map 10 – Trilemmas and Likert

Map10.plo

40

Where to go from here?

Since the 1k participants involved with those 196 adjectives also provided other attachment information (such as from Hazan and Shaver’s instrument) it would be possible to select persons who matched predictions from this model for further analyses.

41

Further, more work is needed about the relevance of Dimension 3’s contribution to understanding attachment dynamics.

Additionally, how can the new 3-D information contribute to revising the 2-D circumplex formulations?

42

Moreover, will this new 3-D conceptualisation allow us to extend our assessment and understanding of Adult Attachment on a broader level and therefore enable us to go beyond a purely social psychological/personality perspective?

43

Further, this research extends the current trend where research is moving away from a strictly categorical approach (here are x number of styles and shoe-horn people into these) and more towards a dimensional approach.

One future challenge is to identify the relative strengths of both of these traditions and make them more complementary, as is done in the medical arena.

44

Finally, we believe this method will help to elucidate the relative strengths and weaknesses of existing approaches to adult attachment conceptualisation and measurement, incorporate them into a new, more inclusive model that can be empirically tested and validated with a view to extending theory and, ultimately, delivering sound practical field-applicable procedures.

45

Some references

Bartholomew, K. & Horowitz, L. M. (1991). Attachment styles among young adults: A test of a four-category model. Journal of Personality and Social Development, 61, 226-244.

George, C. & West, M. (1999). Developmental vs. social personality models of adult attachment and mental ill health. British journal of Medical Psychology, 71, 285-303.

Kirkland, J., Bimler, D., Drawneek, A., McKim, M. & Scholmerich, A. (2004). An alternative approach for the analyses and interpretation of attachment sort items. Early Child Development and Care, 174, 701-719. Klohnen, E. (2005). A circumplex approach to attachment-based self-representations. Manuscript under revision.

Klohnen, E. C. & John, O. P. (1998). Working models of attachment: A theory-based prototype approach. In J. A. Simpson & W. S. Rholes (Eds.). Attachment theory and close relationships. NY: Guilford.

Shaver, P. S. & Mikulincer, M. (2002). Attachment-related psychodynamics. Attachment and Human Development, 4, 133–161.

Waters, E., Crowell, J., Elliott, M., Corcoran, D. & Treboux, D. (2002). Bowlby’s secure base theory and the social/personality psychology of attachment styles: Work(s) in progress. Attachment and Human Development, 4, 230-242.

46

Paper presented at: 14th Biennial Australasian Human Development Conference, Perth, July 2005.

Contact John Kirkland, Department of Health and Human Development, Massey University, Private Bag 11-222, Palmerston North, New Zealand.

Email: j.kirkland@massey.ac.nz