Assessing Secure Base Behavior and Attachment
Security Using the Q-Sort Method

Everett Waters

About Secure Base Behavior

The secure base concept is attachment theory's defining metaphor. Mary Ainsworth reported the first observational study of secure base behavior in her book, Infancy in Uganda (1967). Then followed her longitudinal observational study of mother infant interaction and secure base behavior in Baltimore. The goal in both studies was to provide empirical support for Bowlby's attachment theory. It was important to establish that infant attachment behavior is context sensitive and goal corrected in ways that only a control system model can explain.

The Strange Situation was developed to complement Ainsworth's observational data on the context sensitivity and goal correctedness of infant attachment behavior. The complex secure base behavior revealed in her home and laboratory observations demonstrated the need for a model as complex as Bowlby's and was critical to his success. By 1978 attachment research was almost exclusively focused on Ainsworth's system for assessing individual differences in secure base behavior. Ultimately the validity of Strange Situation classifications rests on their relation to secure base behavior at home (Ainsworth, Bell & Stayton, 1973).

About The Attachment Behavior Q-set (Version 3. 0)

The Attachment Q-Set was developed for three reasons: (1) to provide an economical methodology for further examining relations between secure base behavior at home and Strange Situation classifications, (2) to better define (via a Q-set) the behavioral referents of the secure base concept, and (3) to stimulate interest in normative secure base behavior and individual differences in attachment security beyond infancy. As a first step toward further examining relations between secure base behavior at home and Strange Situation classifications, Vaughn & Waters (1991) replicated the association reported by Ainsworth et al. (1973). This illustrated a method that can be used to test the validity of Strange Situation classifications across age, across cultures, and in clinical populations. The current version of the Attachment Q Set is Version 3.0. It was written in 1987 and consists of 90 items. The items (in English & Spanish) are available on this web site via the "Attachment Q-set Items and Info" link.

About Observing

1. The Q-set items were not written with any particular setting in mind. They can be used to describe secure base behavior of 1-5 year-olds at home or in public places, indoors or outside. Familiarity with the Q-set items (i.e., what to look for; what is "scorable" behavior) is critical. It is not necessary to memorize them, only to know them well enough to recognize "scorable behavior" when you see it. Observers who don't know the items well often become bored, lapse into play with the child, or don't have much of an idea about the child after the visit. Be alert for these signs.

2. It is also important to observe enough behavior. We try to make 2 home visits of 1.5 - 2 hours each. Additional observations can be scheduled if necessary. But fewer or shorter visits can be adequate, especially if the research design involves comparison of group means which, in reasonable sized samples, are reliable even if the scores on individual subjects are less so. Correlational analyses depend on having reliable data on each subject and thus call for longer / more observations per subject.

We prefer to have 2 observers make each visit and provide independent sorts. Sometimes you must make do with a single experienced observer. If after you arrive for a visit, or even half-way through a visit, the observation situation (or the child's behavior, mood, etc.) seems ill-suited to useful observation, the visit should be rescheduled (Ask if child has been ill, slept poorly, etc. Ask if behavior seems out of sorts. Say "Perhaps this isn't the best day for you", etc. Don't make the parent feel that you are disapproving.

Naturalistic observation does not always work as planned. Subjects can be sick; people can drop in to visit the family. Etc. If, immediately after the observation or after attempting to complete the q-sort, you conclude that you are not confident of the observations, it makes sense to discard them and schedule another visit. If the behavior was atypical, you'll get better data the next time. If it was typical, you get the same as before. The goal is to get the best estimate of the subject's typical behavior. You have to make this decision immediately and stick to it. No fair deciding that data are good or bad after seeing that they aren't favoring your hypothesis.The precent of visits that have to be discarded and rescheduled is typically small but should be noted in your research report.

3. The amount of observation time required depends on a number of factors. Infants generally require more time than toddlers. The mother's and child's activities during the visit are also a factor. You may not see much if mom is too busy to interact or prefers to stay inside while the child is outdoors. Nor will you see much if the child becomes to preoccupied playing with the visitor.

4. Bringing along materials for a few brief (5-10 min) age-appropriate mother-child activities can make a visit much more productive. You might try having mother and child make cupcakes from mix, read a pop-up book, play hide-and-seek, or play with a "curiosity box". Explain the activities when you arrange the visit. Emphasize that they are just to keep the child busy and let them enjoy the visit; they are not a "test". Don't interrupt periods of productive observation for these activities. Wait for a "slow" or "unproductive" period. Keep the planned activities informal (no rules, no time keeping). Sometimes you won't get to them at all. The point is to see representative secure base behavior, not to see these particular activities.

5. It can be hard to get good observations if siblings or other visitors are on the scene or if the child has been ill or up late. Try to work this out in advance but if necessary consider excusing yourself early and rescheduling the visit. (Note: No one likes to disappoint a guest, so don't leave abruptly; explain that this happens all the time. )

6. Observers should adopt an informal and responsive manner with the child. If the child shows an interest in playing, play along. Behave much as you would if you really were a social visitor. The mother too will be more comfortable if you seem comfortable with children.

7. Don't adopt a formal manner with the mother. If she is nervous or uncomfortable about what to do the child may well notice. Get acquainted; accept coffee or a snack. But don't let your attention wander from your task as an observer. Don't let the mother feel she is obligated to sit and talk with you. Encourage her to go about her activities; but don't end up baby sitting.

8. Taking the initiative. You can sometimes get information about a child's response to physical contact, willingness to explore away from mother, etc. by initiating the right kind of interaction. For example, if you offer or place the child on your lap to look at a book you can judge response to visitor initiated contact. If you ask the child to take you outside to play, you might learn something about willingness to play away from mother and also set up a situation in which you could better see efforts to monitor or periodically check on mother's whereabouts and activities (as long as you don't keep the child preoccupied once you are outside).

A few such "probes" can be informative. Wait for an appropriate context and don't engage in too much of this. And, as above, don't make too much out of a single instance. The idea is to get information that converges on a conclusion. (Note: Children often invite visitors to wander through the house or play in a favorite room. Inform the mother and ask her permission before going along. Be aware that parents may be uncomfortable with male visitors in the home or playing out of sight with children for more than a few minutes. Be sensible.)

9. Sometimes it makes sense to ask the mother whether something you observe is typical or "Just because I'm visiting. " During the second half of the visit you may find time to ask her about behaviors in the Q-set that you haven't or wouldn't ordinarily see (e.g., crying when left at home). There is no point in asking about very technical secure base items (e.g., sits and cries instead of approaching). These aren't the kinds of behaviors parents report very well. Parent reports are interesting. They can help you interpret specific behaviors. They are useful when they converge with indications you have from your own observations. But taken alone they are not sufficient to justify placing an item more than a point or two from the center of the sort. When asking about the child's behavior, be direct and brief. If the mother seems uncomfortable or embarrassed about her child's behavior, assure her as an observer of many children you see all kinds of things.

10. Observing is difficult. Staying alert, remembering and interpreting events across time, and anticipating critical events are exhausting. If you don't find it so you are probably overlooking things.

About Sorting

Q-set items are sorted into a predefined number of piles (usually 9) according to a predefined distribution. The distribution may be rectangular (equal numbers in each pile) or quasi-normal (more in the center piles than at the extremes). One advantage of the sorting procedure is that it demands closer attention to each item than might occur if the same items were simply rated on a 19 scale. It probably also constrains to some extent social desirability responding because (a) the attention required by the sorting procedure distracts from the desirability set, (b) the sorter does not necessarily know which constructs will be scored or how they will be scored from the data, and (c) only a few items can be "most characteristic" of a subject.

It takes practice to get used to sorting rather than rating items. The following suggestions may be helpful in using the Attachment Q-set. For a more detailed discussion of the Q-sort method see Block (1965), "The Q-sort method in personality assessment and psychiatry research". This volume is available from Consulting Psychologists Press.

1. Good sorting depends on good observing. Good observing depends on familiarity with the Q-set items. You can't make good observations or recall what you have seen if you don't know in advance what to look for. Also, it is critically important to complete your sort as soon as the home visit is over. Drive home or to your office but don't wait until the next day and don't let another visit intervene.

2. Think of yourself starting out with all the items in a single middle pile. The farther you are willing to push an item from the center, the stronger your statement about the child. If you don't have any information about a particular item (or if it is not relevant to a given age), leave it in middle.

3. Focus on how important a behavior is to a good description of the child. To what extent does a particular behavior organize or have high priority in the child's repertoire. Do not confuse importance with mere frequency. A behavior can be more frequent but less important as a descriptor. For example, even an habitual criminal performs more socially appropriate acts more often than criminal ones, but it is likely that , in the sense intended here, the criminal behaviors are the more important descriptors.

4. When you consider placing an item high in a sort, ask yourself "Is this the behavior that would let me pick this child out of a crowd?" or "Would the child alter or forego other behaviors in order to behave this way?"

5. For low placement, ask yourself "Is this the opposite of the child I have seen? "Is it the last thing this child would do" "Would the child alter or forego other behaviors in order to avoid behaving this way?" Do not place an high or low simply because you have seen one unambiguously relevant behavior.

6. Look for multiple examples, critical contexts, or a convergence of indications to support very high or very low placements. Do not place an item very high or very low in a sort solely on the basis of inference from other behaviors. Nor should you place an item very low in a sort merely because it was not seen. You would want to see opposite or incompatible behavior before placing it very low.

7. You will never see examples of every item during a single observation. Yet because of the fixed distribution of the Q-sort, you can't place lots of items in the middle pile. It is reasonable for an experienced observer to make inferences or extrapolations in placing items. If a child "brings or gives objects to mother willingly", you might place the item "stops misbehaving when mother says 'No' " in pile 6 or 7 even if you didn't observe mother having to stop the child from doing something. It would be too much, however, to infer from either of these behaviors that the child would "share willingly with a stranger". (Don't assume that all good things go together.)

8. Most Attachment Q-set items refer to a specific type of behavior and a specific context. There is a rationale for each item. Some address a specific issue related to secure base behavior, some were included to assess discriminant validity vis a vis temperament, motor maturity, or activity preferences unrelated to attachment security. "Filler" items are included because a Q-set will not sort easily if all the items are about the same topic.  

Note on Mothers As Observers

I am often asked about using mothers or other caregivers as observers in research with the Attachment Q-set.  The following is the text of a reply to a 9/98 inquiry:

"The use of mothers as observers has been discussed in an article by Teti, D. M., & McGourty (1996), Child Development, 67, 597-605. Using mothers vs. trained observers in assessing children's secure base behavior:  Theoretical and methodological considerations.

Their bottom line is that if you must use parents it is important to provide some support.  It is ideal to have them become familiar with the items by doing a quick 3 pile sort just to make sure they read the items.  You might sit with the mother and have them mention the types of behavior that makes her place an item high, middle, or low.  If she is not interpreting the item correctly, provide some feedback.

After this orientation, have the mom observe the child for 2-3 days and then do a full sort.  Again, it is best to sit with her, ask for example behaviors for items placed away from the middle, and correct misinterpretations.  All in all, I don't find this much easier than making the observations myself.  In fact, it becomes quite tedious.

In Practice
Teti & McGourty's recommendations are useful and important. Nonetheless, in my own research, .I strongly prefer to make the observations myself or to train research assiatants as observers.  This is what I had in mind in developing the Q-set and I am embarrassed to say I was surprised when most of the people who contacted me wanted to have mothers do the sorting.  My impression is that most mothers don't monitor the secure base phenomenon in detail.  They focus on safety, needs, and indirect indicators of secure base behavior such as affective state, exploration/play, and affective sharing.  The q-set, for better or worse, assumes a more focused and detailed perspective.  I now think our early success with mothers depended more than we recognized on their high motivation as participants in an ongoing longitudinal study and on the support we provided.

I might mention that in using mothers we have noticed that some mothers of (to us) very secure infants give low security scores.  When we talk to them about their scoring, they seem very aware of the infant's behavior and not at all defensive. They just notice the rough edges and see areas in which they can imagine improvement.  We also see the occasional mother of an obviously insecure child giving a very high score.  When we try to discuss their scoring they seem very defensive. If mothers are inattentive during sorting, they are likely to produce moderate to moderately low scores. It may be that such cases account for the majority of the problems with mothers' sorts.

We have used mothers' reports occasionally to identify groups of more secure vs. less secure infants or toddlers.  In group comparisons, some subjects in each group are over-scored and some are under-scored.  The group means are not affected and the loss of power due to greater within groups variance can be made up by increasing the sample size.  For correlational studies, we stick with direct observations.  Here, each subject's score needs to be as accurate as possible.  If one subject is over-scored it reduces the correlation; if another is underscored it reduces the correlation more.  There is no compensation.  So you end up with the wrong correlation (too low) and reduced power.  You can recover power by adding more subjects but the correlation remains an underestimate of the true value.  If you are interested in correlations, I would avoid mothers.  Sometimes the solution is to do several small scale studies with trained observers rather than one large on with mothers.

Keep in mind that in many designs using mothers will confound the source of the q-sort descriptions (and thus possibly the description of the child) with mother characteristics). For example, in a study of clinic vs. non-clinic children, apparent differences between groups of children might actually have arisen because the two sets of mothers see child behavior differently or use the q-set differently.  You would want to avoid using mothers as sorters in such cases.

If you decide to use mothers as observers, it makes sense to change the wording to the first person and "my child".  I would not, however, suggest simplifying the items or reducing the number of items.  Joan Stevenson-Hinde made a good attempt at this early in the q-set's history and the revised set never produced much more than trends for her or others who used it (e.g. Jay Belsky and I think also Marinus van IJzendoorn, among others).

Another suggestion.  If you feel you must use mothers as observers, you may want to have trained observers for a subset so you can report mother observer agreement and perhaps modify your procedure as soon as you detect problems. "

Data Analysis

Block ( The Q Sort Method in Personality Assessment and Psychiatric Research 1978, Consulting Psychologists Press) describes three ways to score q-sort data: item level, cluster level, and criterion sort scoring.

Item level analysis
The first is to treat each q-set item a a separate variable. This gives a large number of descriptive links (correlations or t-tests) between the q-set content and one or more continuous or dichotomous dependent variables. It isuseful to indicate which items a dependent variable IS correlated with and also to examine discriminant validity by noting the items it IS NOT correlated with.

This method is primarily descriptive. It yields too many significance tests for any one to be strongly interpreted. It can make sense however to the correlates and non-correlates. It can be useful to aggregats item level results by performing separate cluster or factor analyses of items that do (convergent) and do not (discriminant) correlate with a particular dependent variable. Although results on individual items may not be reliable, stronger conclusions can be drawn about the dmains represented by the clusters or factors.

Note that the number of significant correltions expected by chance is NOT the p-value x the number of items. This would only be the case if the items were uncorrelated which, inevitably, they are not. The only way to determine whether the number of item level correlates is greater than expected by chance would be through empirical trials correlating several thousand times the item data from a particular study with an array of random numbers representing the dependent variable and counting the number of significant results actually obtained by chance. We have an old DOS program that calculates and plots such randomization trials. It is being rewritten for Windows and will be posted on this site when it is finished.

Note that first factoring or clustering q-set items and then relating them to the dependnt variable IS NOT equivalent to sorting items into correlates and non-correlates of the dependent variable and THEN performing separate clustering or factoring. Because q-set items tend not to be very highly correlated, by design, the former is not usually very useful. Good q-sets have too many small factors. The later better fits the structure of typical q-sets and is likely to be more coherent because the structure of items correlated with the dependent variable can be quite different that the structure of the entire q-set.

Item Sub-Scale Scoring
For the reason just outlined, researchers do not ordinarily construct scales based on facotr or cluster analysis of entire q-set item pools. Instead, the focus is on scales created by rationally combining sets of theoretically related items (usually 5-10 in number). Several such content based scales might be constructed from the items of a given q-set; many of the items in the q-set not being assigned to any scale at all. The reliability and coherence of scales based on such rational item selection can be improcved and summarized using standard correlational methods to examine internal consistency and discriminant validity and add or drop a few items from the initial selection.

The easiest way to score a subject on such a scale is simply to sum the scores of the items included in the scale. Typically soem items will be worded and scored in the inverse direction of the scale (e.g., "Pets unfamiliar dogs" is an inverse indicator for an Anxiety scale). Such items should have theor scores reflected (i.e., change 9 becomes 1, 8 becomes, 2, etc) before the item scores are summed. Never compute a scale score by adding positive items and subtracting inverse items.

Scales based on subsets of items yield fewer significance tests and have a stronger rationale than item level scoring. But they are somewhat less descriptive and can be more difficult to interpret. Results should be interpreted with appropriate qualifications and replication is desirable.

Criterion Sort Scoring
Criterion sort scoring is one of the most useful methods of scoring q-sort data. It involves having experts use the q-set to define the construct of interest and then comparing the q-sort description of individual subjects to this "criterion sort". The criterion sort is constructed by having experts sort the q-set items to describe the "hypothetical most secure (or dependent, etc) subject". The scores in each expert's sort are arranged in order from Item 1 to Item n.
They are then averaged to obtain an average q-set description of the "hypothetical most secure" subject. This criterion sort is an array of n item means.

The similarity between this criterion sort (an array of n item means) and the q-sort description of a particular subject (either an array of n scores, from one observer or the average of several observers) is used as the subject's score on the construct. Similarity is usually measure by correlating the n-item array of criterion sort scores with the n-item array of scores describing the subject. This is illustrated in the following table.

 
Subject 1
Subject 2
Subject n
Criterion sort
Item 1
4
4
9
8.0
Item 2
8
4
3
1.8
Item 3
2
1
2
4.8
Item 4
7
7
5
6.2
Item 5
7
3
1
6.3
 
.
.
.
.
 
.
.
.
.
 
.
.
.
.
Item 90
2
9
4
8.3
Pearson r
r(1)
r(2)
r(n)
 

Once the q-sort description of each subject is arranged in order from Item 1 - Item n and put into a data file along with the criterion sort, it is a simple matter to compute the correlations of any number of subjects to the criterion sort. These correlation coefficients are used as the subjects' scores on the construct.

Although the Pearson correlation does not provide an equal interval scale throughout it's range, the non-linearity is not substantial at values lower than about .70 and can be ignored. If in your data subjects' scores on a criterion sort often range higher than this, use Fisher's r-to-z transformation to make them more linear. (Note: The Fisher's transformation must be computed or obtained from a table of Fisher's r-to-z values. These are NOT the values obtained from tables used to test the significance of correlation coefficients.)

Criterion Sorts

The amount of observation necessary to obtain good descriptions of a subject's typical secure base behavior depends on age, research design (correlational vs. group comparison), and observational context. Infants are typically observerved longer. Correlational designs require good estimates of each subject's typical secure base behavior. Group comparisong require only good estimates of the mean secure base behavior of each group Accordingly, correlational studies require more extensive observation of each subject. Group comparisons need enough subjects per group to obtain reliable group means.

The following criterion sorts can be used to obtain scores on Security (ability to use adult as a secure base) and Dependency ("clinginess") using the current 90-item version of the AQS. In general, scores on these constructs are between 0 and -.2.

The fact that AQS scores measure Security on a continuum is an important advantage. Classifying infants as "secure" or "insecure"overlooks information about meaningful differences within each group. Capturing these differences is an important strength of the q-sort method. In addition, a number of common research designs and statistical methods require or have greater power with continuous variables. Nonetheless, it is sometimes useful to convert continuous AQS scores to a secure/insecure dichotomy.

One approach to this is to assume that the proportion of secure and insecure infants shpuld be the same as if the Strange Situation had been used. In middle class home reared infants the split is typically 70 secure:30 insecure. In the same samples, an AQS cut score of .30 would designate about the same proportion secure and insecure.Of course this cut score could give different proportions in different populations. Attachment theory does not assume equal proportions of secure and insecure attachment in every community or culture.

 

Security Criterion Sort
Based on 8 experts' Q-sort description
of the Hypothetical Most Secure Infant

Item # 

1

2

3

4

5

6

7

8

9

0

00

8.00

1.80

4.80

6.20

6.30

2.20

4.30

3.30

6.50

2.30

10

7.50

6.00

2.70

7.80

7.70

5.20

3.50

8.50

7.70

4.20

20

8.80

6.50

2.70

4.50

2.00

3.30

6.30

7.50

4.30

2.30

30

2.50

7.20

1.30

1.20

4.30

8.80

4.80

1.20

4.70

6.50

40

8.50

8.20

4.70

7.70

5.20

5.70

7.20

6.00

6.30

3.50

50

4.70

3.80

8.50

1.50

7.00

2.70

4.00

3.20

3.80

8.50

60

1.80

5.50

2.00

7.00

1.80

7.00

4.00

5.00

2.30

8.00

70

8.80

4.50

5.20

1.50

1.20

3.20

7.70

4.50

1.00

8.50

80

1.80

4.00

6.50

5.00

7.50

6.50

5.80

1.20

6.50

8.30

 

Dependency Criterion Sort
Based on 8 experts' Q-sort description
of the Hypothetical Most Secure Infant

Item # 

1

2

3

4

5

6

7

8

9

0

00

5.20

5.80

2.00

4.80

5.80

7.20

2.40

4.60

3.00

6.00

10

7.40

2.80

7.40

6.20

4.00

5.00

4.40

5.60

5.44

3.00

20

8.80

4.80

7.00

5.40

2.80

7.60

4.00

6.40

4.00

5.00

30

8.40

4.60

5.20

5.00

1.00

3.60

4.40

7.20

5.00

4.00

40

6.80

5.00

8.60

7.40

5.00

4.60

5.00

4.00

5.20

5.40

50

2.60

5.00

6.00

4.00

5.40

5.60

2.40

3.80

1.20

3.00

60

4.60

4.00

7.80

6.00

5.00

3.60

4.40

5.00

1.20

5.60

70

3.40

5.40

5.60

6.20

8.00

2.80

5.20

2.40

5.20

4.60

80

7.40

4.80

7.00

4.60

3.40

6.20

6.60

4.40

4.80

  7.20

 

Measurement Menu