Dominance analysis of oppositional-defiant parent-child dyads in parent-child interaction therapy


Material Information

Dominance analysis of oppositional-defiant parent-child dyads in parent-child interaction therapy
Physical Description:
Wruble, Marc K., 1962-
Publication Date:

Record Information

Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
oclc - 26907988
System ID:

Table of Contents
    Title Page
        Page i
        Page ii
    Table of Contents
        Page iii
        Page iv
    List of Tables
        Page v
    List of Figures
        Page vi
        Page vii
        Page viii
    Chapter 1. Literature review
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
    Chapter 2. Method
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
    Chapter 3. Results
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
    Chapter 4. Discussion
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
        Page 80
        Page 81
        Page 82
    Appendix 1. Dominance coding system (DCS)
        Page 83
        Page 84
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
        Page 91
        Page 92
        Page 93
        Page 94
        Page 95
    Appendix 2. Dyadic interaction scale (DIS)
        Page 96
        Page 97
    Appendix 3. Time-series programs
        Page 98
        Page 99
        Page 100
        Page 101
    Appendix 4. Chi-square values of the first three LAG intervals of the children's data
        Page 102
        Page 103
        Page 104
        Page 105
        Page 106
        Page 107
        Page 108
        Page 109
        Page 110
        Page 111
    Biographical sketch
        Page 112
        Page 113
        Page 114
        Page 115
        Page 116
Full Text








I acknowledge gratefully my chairperson Sheila Eyberg

for her patience and support. She has provided not only

guidance and direction, but also has served as a fine

professional and clinical role-model. I would also like to

thank my committee members Drs. Hugh Davis, Michael Robinson,

Roger Ray, Steve Boggs, and Michael Conlon for their

assistance in the development and completion of this project.

My wife and best friend, Dee, has been instrumental in

the completion of this project. She coded data, entered it

into the computer, and provided a level of support and

patience that at times seemed almost superhuman. And I wish

to thank my friend and colleague, Dr. Roger D. Ray, for

picking up where Kantor left off and for his unconditional

technical, methodological, and personal support. Last, I

thank all the undergraduate coders for their assistance and



ACKNOWLEDGEMENTS ....................................... ii

LIST OF TABLES ........................................... v

LIST OF FIGURES ........................................... vi

ABSTRACT ................................................. vii


LITERATURE REVIEW ................................. 1
Introduction ................................. 1
Parent-Child Interaction Therapy (PCIT) ...... 2
PCIT Research into Leading and Following ..... 3
Selected Review of Relevant Social
Interaction Process Research ................. 7
Dominance in Marital Interaction ............. 12
Summary ...................................... 14
Statement of Problem ......................... 16
Defining Leading and Following
Operationally ........................... 17
Advantages of the Dominance Construct ........ 20
Statement of Objectives ...................... 22
Predictions About Dominance in PCIT .......... 24
Predictions About the Convergent and
Concurrent Validity of the Dominance Measures 27

II METHOD ............................................... 30
Subjects ..................................... 30
Measures ..................................... 31
Dominance Coding System (DCS) ........... 31
Dyadic Interaction Scale (DIS) .......... 34
Eyberg Child Behavior Inventory (ECBI).. 34
Procedures ................................... 35
Observer Training and Coding.............. 36
Training Observers and Reliability
Assessment .............................. 36
Data Analysis ................................ 37

Time Series Analysis .................... 37
Time Series Models and Summary Measures. 45

III RESULTS .............................................. 49
Reliability of the DCS ....................... 49
Groups Comparison Results .................... 53
Validity of the Dominance Measure ............ 57
Concurrent Validity ..................... 57
Convergent Validity ..................... 59

IV DISCUSSION ......................................... 62
Dominance Coding System (DCS)....:............. 62
Psychometric Properties of the Dominance
Measure ...................................... 65
Inter-observer reliability .............. 66
Validity of the DCS ..................... 70
Concurrent Validity ................ 70
Convergent Validity ................ 72
DIS .............................. 72
Discriminant Validity ................... 74
Groups Comparison Results .......... 74
Special Considerations in PCIT Interaction
Research ..................................... 77
Overview of Parent-Child Interactions ........ 78
Epilogue ..................................... 81

1 DOMINANCE CODING SYSTEM (DCS) .................. 83
2 DYADIC INTERACTION SCALE (DIS) ................. 96
3 TIME-SERIES PROGRAMS ........................... 98

REFERENCES ................................................ 105

BIOGRAPHICAL SKETCH .................................... 112




MOTHER AND CHILD ............ .................. 60



1 LEADING AND FOLLOWING MATRIX ....................... 19

ASSESSMENT INTERVAL ................................ 55

Abstract of a Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy



Marc K. Wruble

December, 1991

Chairman: Sheila Eyberg
Major Department: Clinical and Health Psychology

The process of social leading and following has been

considered important to the effectiveness of Parent-Child

Interaction Therapy (PCIT), even though this process has not

been measured directly and this assumption untested.

Subjects in this study were 12 oppositional-defiant mother-

child dyads and 12 non-referred mother-child dyads. Each

clinic dyad was observed before and after treatment, and each

non-referred dyad was observed at comparable times, in two

successive five-min clinical play situations in which the

mother was instructed to either lead or follow the child's

play. Dominance was defined as the amount of sequential

patterning between mother and child leading and following

behaviors. An observational measure of dominance, obtained

by continuously coding changes in mother and child play

activity and speech, was developed and evaluated. Results

for the children suggested that child following is an

important component of PCIT, and that in CDI at post-

treatment the clinic children's behavior was more self-

directed than in CDI at pre-treatment. The dominance measure

was sensitive to treatment changes in CDI. Inter-observer

reliability was adequate for the child data. The concurrent

validity of the dominance measure, with a parent-report

measure of conduct problem behaviors, was established.

Correlations between the dominance measure and subjective

ratings of leading and following did not yield evidence of

convergent validity. The discriminant validity of the

dominance measure was also established.




Children with conduct problem behaviors are

distinguished from other children by their behaviors during

social interaction with their parents. Children with conduct

problem behaviors display higher rates of noncompliance and

deviant behaviors, including whining and yelling, than non-

referred children (Robinson & Eyberg, 1981). Parents of

these children experience more parental and extrafamilial

distress and marital dysfunction than parents of nonreferred

children (see McMahon & Wells, 1989 for a review). These

child and family problems are especially important

considering that conduct problem behaviors are the most

common psychological disturbances of childhood (Quay, 1986).

To this point behavioral clinical interventions for

children with conduct disorders have involved parent-training

programs based on operant behavior management principles,

relationship enhancement approaches, or both.

In the late 1960s Hanf advanced a two-stage model that

taught parents to use operant principles to modify child

noncompliance (Hanf, 1969). In the first stage parents were

taught to follow the child's play using techniques of

differential social attention. The second stage involved

instructing the mother to both lead play and to respond

contingently to child noncompliance or compliance with time-

out or labeled praise, respectively. The purpose of Hanf's

two-stage model and its subsequent modifications (Barkley,

1987; Eyberg, 1979; Forehand & McMahon, 1981) has been to

improve the quality of the parent-child relationship and to

more effectively manage the child's behavior.

Parent-Child Interaction Therapv (PCIT)

Parent-Child Interaction Therapy (PCIT) (Eyberg, 1979)

extends Hanf's (1969) two-stage model. PCIT has proven

effective in treating conduct disordered dyads. PCIT,

however, is designed for a broad range of psychological

disturbances in preschool children (Eyberg & Boggs, 1988).

PCIT has two phases: Child-Directed Interaction (CDI) and

Parent-Directed Interaction (PDI).

CDI focuses mainly on changing the quality of the

parent-child relationship through enhanced communication

using techniques derived from child-centered play therapy.

In CDI parents are taught to follow the child in play using

both nondirective interaction skills and differential social

attention. Specifically, the parent follows the child by

describing, praising, or imitating appropriate child

behaviors, reflecting appropriate child verbalizations, and

consistently ignoring inappropriate child behaviors.

Problem solving skills training for the parent has been

incorporated in the second phase, Parent-Directed Interaction

(PDI). In PDI the parent is taught to lead play through the

use of both effective commands and positive or negative

consequences contingent on child compliance or noncompliance,

respectively. Specifically, the parent leads the child in

play by giving effective commands and providing positive

(i.e., labeled praise) and negative consequences (i.e., time-

out) contingent on the child's obedience or disobedience,

respectively. In practice, however, PDI is not the opposite

of CDI. This is because PDI skills are actually practiced

and conducted within the context of CDI. Hence, during PDI

the parent and the child often exchange the lead back and

forth. This would be indicated, for instance, by the child

complying to a parental command (an instance of parent lead,

child follow) and then the parent describing and praising the

child's play (an instance of child lead, parent follow).

By blending both CDI and PDI interactive skills the

parent achieves a healthy balance between being nurturant and

being firm by setting clear limits and consistently enforcing

them. In addition to increasing child compliance, clinical

observations suggest that the parent models and facilitates

reciprocal play, sharing, and social leading and following


PCIT Research into Leading and Following

As is evident from the discussion above, the ability to

lead and follow are considered important to the success of

PCIT. Parent-Child Interaction Therapy research has

attempted to measure leading and following by using frequency

counts of particular parent verbalizations and child

behaviors. For instance, a parent-child compliance

interaction, by definition, provides important information on

the outcome of a particular lead-follow event; the parent

attempts to lead by giving the child a command and the child

has an opportunity to follow the parent by complying with the


PCIT research using frequency counts of mother and child

behaviors during their dyadic interactions has demonstrated

that parents of conduct problem children have difficulty

following their children in play as indicated by using higher

rates of commands (Aragona & Eyberg, 1981; Robinson & Eyberg,

1981), questions and critical statements (Robinson & Eyberg,

1981), and by using less praise and offering fewer

descriptions (Aragona & Eyberg, 1981) than parents of

nonreferred children. Similarly, Eyberg and Robinson's

(1981) results suggest that conduct problem children have

difficulty following their parent's directions as indicated

by their high rate of non-compliance and deviant behaviors.

These studies demonstrate how frequency counts of specific

parent behaviors provide a general measure of how many times

the parent attempts to lead or follow the child's behavior

during their play interaction. One limitation of frequency

count measures of the parent-child interaction is that they

do not assess how the parent and child interact with each

other contingently on a moment-to-moment basis. Stated

another way, frequency measures do not assess the process of

how the parent and child sequence their behaviors on a

moment-to-moment basis.1 The task of determining who is

leading and who is following is complicated by the continuous

nature of the parent-child interaction in which any behavior

is both an antecedent to future behaviors and a consequent of

prior behaviors. Thus, without knowing the sequential

relationship between the parent and child behaviors it is

difficult to know who is leading and who is following, and to

what degree. For these reasons frequency counts are

considered inadequate for describing social interaction

processes in general (see for example, Gottman, 1979; Gottman

& Ringland, 1981) or the parent-child system in particular

(Sheeber, Wruble, Sorensen, & Eyberg, 1990).

More direct measures of parent-child interaction are

provided by a time-based sequential view of how the parent

and child relate to each other through time. By this account

PCIT research has provided limited study of the process of

leading and following. As a result there has been no

evaluation of the ability of a process measure of leading and

following to discriminate clinic from nonclinic dyadic

interactions, or to assess the importance of leading and

following to the PCIT program. This leaves researchers and

1 The distinction being made here is the familiar one between
content and process. PCIT research using frequency counts is
to content as PCIT research using sequence measures is to

clinicians in the position of coaching the parent in specific

leading and following skills (e.g., reflections, praise)

without empirical support for the importance of the leading

and following process. Consequently, the relationship

between specific leading and following skills and the process

of leading and following is unknown, making it difficult for

PCIT clinicians to assess adequately changes in this process

over the course of treatment.

Gottman's (1979) investigations of marital interaction

illustrate the limitations of using frequency counts to study

social interactions. He used highly detailed coding of

distressed and nondistressed couples' nonverbal behavior

while they discussed a number of decision-making issues

varying from a nonconflict to a high-conflict task. He

reported that distressed and nondistressed couples did not

differ significantly in the amount of positive or negative

affect but that distressed couples were distinguished by long

sequences of reciprocating "negative affect" (i.e., the

husband's negative affect would be followed by the wife's

negative affect, followed by the husband's negative affect,


Of course, frequency counts and process measures of the

parent-child interaction can provide convergent and

complementary information. For instance, process measures

(especially those that are time-based) require relatively

large samples of behavior which often renders them

insensitive to patterns in relatively low rate behaviors

(e.g., hitting), whereas frequency counts of these behaviors

may suggest group differences. For another instance, because

frequency measures have been used extensively they can be

employed to assess the validity of process measures.

Finally, frequency measures typically are obtained more

efficiently and are more practical for clinical use. Thus,

if a relation between sequence measures and frequency

measures is obtained then the frequency measure would provide

a clinically useful and efficient measure of the process.

Selected Review of Relevant Social
Interaction Process Research

The process of social leading and following is

considered to be the foundation upon which social competence

originates and develops (Lewis & Freedle, 1973; Stern, 1971;

Tronick, Heidelise, & Brazelton, 1977). This process

originates in the face-to-face interactions between the

parent(s) and infant (Hofer, 1981; Stern, 1971; Brazelton,

Koslowski, & Main, 1974) and continues throughout the course

of a person's social development (Davis, 1982; Maccoby,

1980). Competent parent-infant interactions occur when the

child is able to signal his status and needs effectively and

to respond effectively to parental interventions, and when

the parent is able to interpret the child's status and needs

so as to effectively respond with the right behavior at the

right time in the right intensity. A number of researchers

have attempted to quantify this and other aspects of the

parent-infant relationship in order to relate it to later

cognitive, social, emotional, and individual development

(Tronick et al., 1977; Bowlby, 1969).

A number of studies conducted in the 1970s attempted to

document interaction patterns between parent and infant

during face-to-face dyadic interactions. In 1971, Stern

studied the face-to-face interactions between a mother and

her 3-1/2-month-old twins. Citing developmental research

about the importance of gaze, Stern had observers code parent

and infant gaze behaviors (from videotapes) into dichotomous

categories of gaze toward or away for both the mother and

infant, resulting in a total of four possible states. In

addition to providing descriptions of how maternal and infant

behaviors merge to produce mutual interactive sequences

(e.g., both gazing away from each other to both gazing at

each other), he used running correlations to assert that

"Statistically, the mother is more often the initiator of the

mutual pattern; but the pattern can be initiated by either"

(Stern, 1971, p. 512). Unfortunately, it is difficult to

generalize from the results of this study given the small

sample size, and because of Stern's inappropriate use of

running correlations to analyze crosscorrelations. The

latter limitation will be addressed in more detail shortly.

Brazelton, Koslowski, and Main (1974) studied five

parent-infant dyads in which the infants were between 2 and

20 weeks old. Mother and infant behaviors were videotaped

simultaneously by two cameras and then merged on a split

screen with a time code superimposed, which allowed slow

motion or frame-by-frame analysis. A variety of maternal and

infant behaviors were coded from these videotapes. Brazelton

et al. (1974) analyzed the data visually using time-series

graphs and described a mother-infant cycle characterized by

gradual transitions from attention to inattention states.

They concluded that "The most important rule for maintaining

an interaction seemed to be that a mother develop a

sensitivity to her infant's capacity for attention and his

need for withdrawal--partial or complete--after a period of

attention to her" (Brazelton et al., 1974, p. 59).

In a more methodologically sophisticated study, Stern

(1974) examined the moment-by-moment vocal, facial, and gaze

interactions in eight mother-infant dyads. Each dyad was

composed of premature (by weight criteria) twin infants

between the ages of 3 and 4 months. He was interested in

investigating the bidirectional nature of parent-infant

interaction during feeding and spontaneous social play

situations. As before (Stern, 1971), he used a dichotomous

gaze "away-toward" category system. Stern (1974) used

transitional probability analysis and descriptive analyses to

conclude that infants used attention-inattention to control

the interaction, whereas the mother controlled the

interaction through her facial and vocal behaviors. In

particular, he reported that the mother was more likely to

lead by gazing at the infant until the infant followed by

gazing back at her. Having achieved this mutual gazing

state, the infant regulated his attentional and affective

tolerance by averting his gaze. Stern (1974) concluded that

during parent-infant interaction, gaze functions to regulate

social contact and stimulation, and that coordinated

sequences of attention-inattention represent a parent-infant

dyad that is communicating their status, needs, and

intentions effectively. In addition to the rich description

of parent-infant interaction, this study also provided a

sequential account of attention in the dyadic relationship.

Pioneering work by Tronick et al. (1977) was directed

towards quantifying how the parent-infant dyad develops, and

maintains, mutual affective and attentional states through

the course of their face-to-face interaction. These

researchers used the same videotape recording procedure used

by Brazelton et al. (1974) with three parent-infant dyads as

subjects in which the latter ranged from 11 to 13 weeks old.

Six infant and parent expressive behaviors were coded.

Behaviors were ranked on the degree to which they reflected

the parent's and child's affective and attentional

involvement. This ranking resulted in a univariate scale

ranging from maximum positive involvement to maximum negative

involvement. Each dyad's involvement score was then plotted

across time, creating a time-series graph which summarized

their interaction. Tronick et al. (1977) statistically

assessed the degree of cycling and mutuality by correlating

the rankings for each mother-infant dyad at successive 10

second intervals (i.e., each correlation was based on 10

seconds of interaction such that the first correlation was

based on seconds 1-10, the second correlation was based on

seconds 2-11, etc.). They interpreted periods of high

positive correlations as indicative of mutual cycling, and

periods of high negative correlations as indicating that the

mother and infant were almost perfectly out of phase with

each other; periods of low to medium positive correlations

indicated periods of disharmony. They concluded that

Long periods of interaction result during which the
quality of the infants' displays approaches an almost
perfect synchrony with the mothers'. During these
periods in which mutual cycling occurs, as evidenced by
high positive correlations, the infant and the mother
show a cyclic acceleration and deceleration of their
affective and attentional involvement that closely
resembles a homeostatic curve" (Tronick et al., 1977, p.

Periods of desynchrony are important because they represent

crisis periods (Riegel, 1976) when the parent and infant are

not communicating well.

Gottman and Ringland (1981) criticized the conclusions

of Tronick et al. (1977) on statistical grounds because

nonzero correlations would be produced even in the case of no

interaction They wrote that

If the mother and infant's behavior were entirely
cyclic, with unequal frequencies, the behavior of
each would be completely determined by its own past
and no knowledge would be gained by knowing the
behavior of the other. This is the case of no
interaction . and . the running
correlations would be nonzero . because . .
the mother's and baby's cycles would be
successively in phase (giving positive
correlations) and successively out of phase (giving
negative correlations) (Gottman & Ringland, 1981,
pp. 396-397).

Gottman and Ringland (1981) advanced time-series

analysis as an appropriate methodology to study these and

other types of social interaction. Gottman and Ringland's

methodology will be detailed in the method section because it

provided the basis for the time series analytic procedures

used in the present study.

In summary, the parent-infant interaction literature

reviewed is significant for its pioneering work in the area

of observational measurement and lucid descriptions of dyadic

interaction. The results of this research suggest that

social leading and following is instrumental in the

development of social competence and fundamental to the

development and maintenance of a healthy parent-infant

relationship. However, pervasive methodological problems,

including small sample sizes and inappropriate statistical

methods reliability methods, limit the generalizability of

these conclusions.

Dominance in Marital Interaction

Gottman (1979) used an intuitive univariate scaling

procedure similar to that used by Brazelton et al. (1974) and

Tronick et al. (1977) to assess patterns of affective leading

and following in his research with clinic-referred and

nonreferred married couples. Using videotaped interactions,

Gottman (1979) assigned positive or negative values to both

the speaker's and listener's affective behaviors during

discussions of low and high conflict improvisation tasks. He

summed these scores within each turn-taking episode, and then

summed across episodes, resulting in a cumulative record or

time series. Gottman defined two types of affect, mood and

expressions, which were distinguished by slower and faster

changes, respectively, in facial affect displays. He

analyzed the data using bivariate cross-spectral time-series

analysis procedures. Gottman (1977) reported an interaction

between type of affect (mood versus expression) and the level

of task conflict such that significant and predictable

patterns of leading and following in emotional expression

(rapidly changing affect component) were observed between

groups only during high conflict tasks. In particular,

during high conflict tasks the clinic husband's emotional

expression was found to predict his wife's emotional

expression, but not vice versa. Gottman (1979) interpreted

these results as suggesting that clinic-referred wives are

emotionally responsive to their husbands, but not vice versa.

Gottman's (1977) pioneering research is important for a

number of reasons. Prior to his investigations there had

been more speculation about marital interaction than

empirical research. Indeed, marital research had provided

limited study of interactions in general, and the process of

leading and following in particular. Gottman (1977)

demonstrated how observational measures of leading and

following can be used to quantify interpersonal processes

that heretofore had been addressed only metaphorically or

anecdotally. His time-series results provide an excellent

illustration of how measures of leading and following

discriminate between clinic and nonreferred dyads as a

function of situational context (i.e., level of task

conflict). Unfortunately, it is difficult to generalize from

Gottman's conclusions because of limitations in the time-

series method he selected and because the psychometric

properties (including interobserver reliability) of Gottman's

(1977) measure of leading and following are unknown. Both of

these limitations will be detailed in the methods section.

Once the psychometric properties of Gottman's measure of

leading and following is known, it has the potential to

improve clinic-nonclinic classifications, assess therapy

progress, and evaluate therapy outcome.

Gottman's (1979) marital interaction research advanced

social interaction research methods and elucidated that

dysfunctional interactions are represented by particular

patterns of leading and following, within specific contexts,

and that these interaction process measures have important

clinical implications for assessment and treatment.

The combined results of the parent-infant and marital

interaction literatures reviewed suggest the following

points. Under certain conditions dominance is indicative of

functional interactions (e.g., the parent-infant synchrony

detailed in the parent-infant literature), whereas under

other conditions dominance is indicative of dysfunctional

interactions (e.g., affective dominance as detailed in the

marital interaction literature). The fact that dominance can

indicate either functional or dysfunctional interactions

suggests that the ability to engage and disengage in dominant

interactions is necessary but not sufficient for the

development of social competence. In some ways the results

of the parent-infant and marital interaction literatures seem

contradictory. In the parent-infant interaction literature,

dominance is viewed as both the goal of the dyad and when

present, as a positive index of the dyad's developing social

competence; while in the marital interaction research

dominance is considered to be indicative of a dysfunctional

marital relationship. Thus, dominance is said to both

presuppose competent social interaction and at the same time

reflect it. This apparent confusion likely reflects the fact

that dominance is a global measure of the extent to which

each person's behavior is influenced by the behavior of the

other, that the interpretation of a particular type of

dominance depends upon the context in which the interaction

occurs, and that developmental variables affect the

interpretation of dominance. A similar argument has been put

forth by Gottman and Ringland (1981). They assert that


is a function of the interactants and of the situational
context of the interaction. The conceptual label given
to asymmetry in predictability should vary as a function
of the dependent measures, the context (e.g., the goals
and tasks of the interaction), and the nature of the
interactants (Gottman & Ringland, 1981, p.385).
In order to interpret this construct, interaction

research needs to proceed in three directions:

(a) establishing a psychometrically sound measure of

dominance; (b) manipulating different properties of the

interactants and their context to clarify its interpretation;

and (c) describing empirically the interactive behaviors that

establish dominance.

Statement of Problem

Social leading and following is considered important to

the development of social competence in general and to PCIT

in particular. PCIT research, however, has provided limited

study of the process of leading and following. Leading and

following in PCIT has been measured only minimally and at

only a relatively gross level (e.g., reflections follow,

commands lead). Consequently, the potential ability of a

measure of leading and following to improve clinic-nonclinic

classifications, assess therapy progress, and evaluate

outcome has not been evaluated. A measure of lead-follow

interaction patterns should indicate how well the parents are

able to lead and follow. This would be useful in

discriminating clinic from nonclinic dyads, assessing

suitability for treatment, and evaluating therapeutic

progress and outcome.

Clearly, a construct alleged this important must be

carefully defined and must exhibit adequate psychometric


Defining Leading and Following Operationally

The first step in evaluating the importance of leading

and following in PCIT is an operational definition of the


Gottman (1979) used the term "dominance" to describe

patterns of leading and following during marital interaction.

Gottman's notion of dominance was important for two reasons.

First, it allowed the bidirectional process of leading and

following to be incorporated into one term; the person who

was leading was dominant, and by implication the other person

was following. Second, Gottman (1979) suggested that

patterns of social leading and following (i.e., dominance)

can be quantified in probability terms. Gottman (1979) used

the notion of "asymmetry in predictability" in his studies of

marital interactions to determine which partner was leading

and which partner was following in affect and mood.

Gottman's notion of asymmetry in predictability translated

into parent and child terms simply states that if the child's

behavior is more predictable from the parent's past behavior

than from the child's own past behavior then the parent is

considered to be leading the child. Another way of saying

this is that the parent is leading the child when, over the

course of the interaction, her behavior is a more probable

antecedent to the child's behavior than the child's own past

behavior. The pattern of child leading and parent following

taught in CDI expressed in probability terms is:

p(parent behavior/child behavior) > p(parent
behavior/parent behavior)

which is read as the conditional probability of a parent

behavior given a prior child behavior is greater than the

conditional probability of a parent behavior given a prior

parent behavior. Thus, if leading and following are

important, then during CDI, for example, the mother should

lag behind the child's behavior in a predictable way. This

relationship should be represented in sequential measures of

the parent-child interaction.

Figure 1 illustrates the four possible patterns of

leading and following between the parent and child. The cell

in the upper left defines an interaction in which the child

is dominant (i.e., the child leads and the parent follows).

This is the pattern expected in CDI. The cell in the lower

right corner defines an interaction in which the mother is

dominant (i.e., the mother leads and the child follows). The

cell in the lower left defines an interaction in which the

mother and child are essentially playing independently of one

another. The cell in the upper right corner defines a

bidirectional relationship in which the mother's and child's

play alternates between leading and following. Knowing the

clinical play situation, associated affect, content, and

issue would help to determine whether a bidirectional pattern

is indicative of a "power struggle," or of the sharing, and

friendly dialogue of an egalitarian interaction.

Synchronous interactions represent a particular subset

of dominance patterns. Synchronous interactions have been


C >M C< >M


C M M >C

(adapted from Gottman & Ringland, 1981)

Figure 1


defined by either mutuality (performing the same behaviors at

the same time, as for example, during a handshake, greeting

or good-bye, postural mirroring, etc.) or by lead-follow

behaviors (i.e., performing one or more behaviors before or

after the other person in the interaction, such as taking

turns during dialogue). Synchrony, then, is indicated by

all types of dominance except independence. Mother or child

dominance are indicative of synchrony because either the

mother follows the child's lead by some period of time or the

child follows the mother's lead by some period of time.

Bidirectional interactions are indicative of synchrony

because the parent and child are acting simultaneously, or

because they are actively switching the lead back and forth.

Advantages of the Dominance Construct

Gottman's dominance construct provides a framework

within which to integrate concepts and terms that have been

used in different literatures. Parent-infant researchers

have used the term "synchrony," while PCIT researchers have

used the term "leading and following," and marital

interaction researchers have used the term "dominance" to

refer to observed patterns of asymmetrical or mutual

influence during social interaction. Gottman's dominance

construct provides conceptual clarity across these fields

because it incorporates prior definitions. At the same time

it advances these fields by providing an operational

definition and a statistical measure (i.e., a multi-parameter

equation and/or a classification) of leading and following

for a variety of interactive data. Summarizing complex and

dynamic interactions in this way is important because it

allows clinical researchers to provide measures of

interaction patterns to complement their assessment and

intervention procedures. Without such measures there is no

direct way to assess the status and change in the dyadic


The ability to detect dominance patterns in overall

interaction patterns has a number of advantages. First, it

overcomes problems of using specific behaviors or sequences

to establish dominance (Gottman 1979; Gottman & Parkhurst,

1980). For instance, the dominance statistic is superior to

frequency counts of specific behaviors because it provides a

measure of leading and following across specific types of

play interactions (compliance, sharing, giving, object

struggles, monologue, dialogue, rearranging, taking, etc).

Obviously, there are numerous ways in which a parent can

influence the child's behavior; a statistic that assesses

dominance across separate instances is more likely to provide

a more accurate measurement. Second, the ability to obtain
"macro" measures (i.e., a dominance summary equation and/or

classification) and "micro" measures (i.e., patterns in

specific codes) allows for complementary analyses at

different levels of behavioral resolution. Third, dominance

can be used as a dependent variable in studies evaluating the

influence of various factors on dyadic interactions.

Statement of Obiectives

The purpose of this study was to examine the inter-

observer reliability, and the discriminant, concurrent, and

convergent validity of an observational measure of dominance

previously developed by Wruble, Sorensen, Sheeber, & Eyberg

(1989) for assessing this aspect of parent-child interaction

during PCIT. All of these purposes have important

implications for clinical assessment and intervention.

The discriminant validity of the dominance measure was

assessed by comparing dominance in clinic dyads, before and

after treatment, with dominance in nonclinic dyads at the

same assessment intervals. Assessing dominance between

groups and across treatment addresses three important

questions: (1) is the dominance measure valid for

discriminating conduct-problem-referred from nonreferred

dyads, (2) how does treatment affect dominance, and (3) is

dominance an important component in PCIT? Dominance would be

considered important to PCIT if the clinic and nonclinic

groups differed before treatment but behaved more similarly

after treatment. A measure of dominance with good

discriminant validity can improve clinic-nonclinic

classifications by providing a sensitive measure of

functional or dysfunctional interactions. Existing frequency

count measures may not provide such a direct and sensitive

measure of the interactions (e.g., frequency patterns may be

the same but interaction patterns may be different).

Comparing a measure of dominance with an existing

treatment outcome measure allows a test of the previously

untested assumption that leading and following is an

important component of PCIT. The concurrent validity of the

dominance measure will be assessed by comparing it with the

two scores from the Eyberg Child Behavior Inventory (ECBI;

Eyberg, 1974). The ECBI has been shown to have good

psychometric properties including being able to discriminate

children with conduct problem behaviors from non-referred

children (Eyberg & Robinson, 1983), relating significantly to

observational measures of parent-child interactions (Webster-

Stratton & Eyberg, 1982), and providing a sensitive measure

of treatment outcome (Eyberg & Robinson, 1982).

Attempts to measure the dominance construct have for the

most part relied on detailed, often frame-by-frame behavioral

coding of dyadic interaction. The advantage of this type of

measure is that it allows for a precise description of the

behaviors involved in dominance between each interactant.

One disadvantage is that the procedure is tedious and

extremely time consuming which makes it impractical for

clinical use. Recently, Bernieri, Reznick, and Rosenthal

(1988) had observers rate synchrony (which, as mentioned

earlier, is a type of dominance) in parent-infant

interactions along three dimensions (simultaneous movement,

tempo/rhythm, and overall impression of interpersonal

harmony) each on a 1-9 Likert-type scale. By manipulating

the timing and partner variables in the video film clips for

each interaction they found that these scales were highly

intercorrelated and that observers could reliably assess

dominant (i.e., synchronous) interactions. The advantage of

this more subjective coding strategy is that it seems to

provide a good estimate of dominance in a period of time that

is practical clinically. One conceptual problem with this

type of scoring system is that one is unsure about what

behaviors the coders are using in evaluating dominance. This

is important for two reasons. One, researchers have

identified different types of dominance (e.g., affective,

attentional, motoric, etc.) and it would seem important to

know how these are related to each other and how they

function separately. Assessing dominance in a global way

does not allow for such specification. Two, in order to

teach dominance, therapists need to know how specific parent-

child behaviors are related sequentially over time. Global

measures of dominance do not provide this information to the

therapist. Ideally, clinical ratings of dominance would be

empirically derived from the interactional patterns that

establish dominance. This would allow clinician's to assess

these patterns efficiently and to teach these patterns. In

this study dominance measures were obtained from detailed

observational coding and from observer ratings of the parent-

child interactions.

Predictions About Dominance in PCIT

The design of the present study examined the effects of

group (clinic or nonclinic), situation (CDI or PDI), and

assessment interval (pre and post-treatment for the clinic

group) on parent and child leading and following behavior. A

basic assumption in this study is that if leading and

following are important, then the parent and child data

should take certain forms. The demand characteristics of the

two clinical play situations, CDI and PDI, facilitated

forming hypotheses. A basic rule in forming the hypotheses

was that at pre-treatment the clinic dyads will behave in a

way inconsistent with the demands of the CDI-PDI situations.

For instance, in CDI at pre-treatment clinic parents will

lead when they are supposed to follow. Hypotheses were

advanced for both parents and children. The hypotheses of

most interest were those for the participant who should be

following in the situation. This is because the statistic

used to measure dominance is based on the relation between

current behavior and past behavior. Thus, a large dominance

value for a child would indicate that the child's behavior

was highly predictable from the parent's prior behavior at a

level greater than would be expected by chance alone. By

implication, and in statistical terms, the parent would be

considered to be leading the interaction (i.e., dominant)

because the child's behavior is more predictable from the

parent's behavior than from the child's own prior behavior.

The following hypotheses were made:

(a) at CDI pre-treatment, clinic parents will follow their

children's play behaviors less than parents in the comparison

group because clinic parents often have difficulty following

their children's play behaviors.

(b) PCIT should have the effect of improving the clinic

parents' following behavior (i.e., an increase in

predictability between antecedent child behavior and

subsequent parent behavior) such that at post-treatment they

will follow their children at a level comparable with that

exhibited by the nonclinic parents.

(c) in CDI pre-treatment, children in the clinic group will

exhibit more following behavior than children in the

comparison group because clinical experience suggests that

clinic parents' often lead in CDI when they are supposed to

be following.

(d) in CDI post-treatment PCIT will have the effect of both

decreasing the parents' attempts to lead and improving the

children's self-management of their own behavior. Therefore,

clinic children will follow their parents less compared to

pre-treatment CDI, and at a level not significantly different

from that of the comparison group children.

(e) at PDI pre-treatment, parents in the clinic group will

follow their children more than parents in the comparison

group. This is because clinic parents have difficulty

leading effectively and often respond ineffectively to their

children's deviant behavior resulting in the clinic parents

following their child's behavior. In contrast, non-referred

parents are able to lead effectively, control their

children's behavior, and are better able to alternate between

leading and following.

(f) at PDI post-treatment, parents in the clinic group will

follow their children less than they did in PDI at pre-

treatment, and at a level not significantly different from

non-referred parents because PCIT teaches clinic parents

differential social attention skills and how to effectively

alternate between leading and following.

(g) non-referred parents did not receive treatment so they

should not differ significantly in the amount they follow

their children in CDI or PDI over the two assessment


(h) at PDI pre-treatment, the clinic children will follow

their parents' lead less than nonreferred children because

clinic children typically oppose their parents attempts to

lead, and clinic children control the interaction through

their deviant behavior.

(i) at PDI post-treatment, PCIT will have functioned to

improve the clinic children's following behavior such that

the clinic and nonreferred children will not differ

significantly in the amount they follow their parents.

Predictions About the Convergent and Concurrent
Validity of the Dominance Measures

The convergent validity of the dominance measure will be

indicated by a positive correlation between the measure of

dominance derived from the Dominance Coding System (DCS;

Wruble, 1990; see Appendix 1) and the measures of dominance

obtained from the Dyadic Interaction Scale (DIS; see Appendix


The concurrent validity of the dominance measure will be

indicated by the pattern of correlations between the DCS

measures of dominance and the two ECBI scores for each level

of the experimental variables. The following hypotheses were

advanced: (a) in CDI, at each assessment interval and for

each group, the ECBI will correlate positively with child

following. High ECBI scores indicate frequent conduct

problem behaviors which are experienced as problematic by

parents. High levels of child following in CDI indicates

dysfunctional interactions because the parent should be

following the child. In particular, clinic parents not only

report that their children exhibit more conduct problem

behaviors, but at pre-treatment clinic parents are also more

likely to lead in CDI when they should be following. The

ECBI will also correlate positively with child following for

clinic children at post-treatment because PCIT should have

the effect of improving the clinic parents' following

behavior resulting in levels of child following and child

deviant behaviors consistent with those exhibited by the non-

referred children at both assessment intervals. Thus, ECBI

scores and child following will be positively correlated for

the non-referred children in CDI at both assessment

intervals; (b) in PDI, at each assessment interval and for

each group, the ECBI will not be significantly correlated

with child following. Elevated ECBI scores are always

considered indicative of frequent or problematic conduct

problem behaviors. Leading and following in PDI, as detailed

previously, is expected to be reciprocal and alternating.

Thus, the interpretation of child following in PDI depends on

how much the mother attempts to lead and how successful she

is in her attempts; (c) in CDI, at each assessment interval

and for each group, the ECBI will correlate negatively with

parent following. The ECBI will correlate negatively with

parent following for the clinic parents in CDI at pre and

post-treatment, but perhaps for different reasons. At pre-

treatment CDI, clinic parents have difficulty following their

children's behavior and report high levels of child deviant

behavior. At post-treatment, PCIT will have functioned to

improve the parent's following while concomitantly decreasing

child deviant behavior. The ECBI will be negatively

correlated with parent following for non-referred parents in

CDI at the two assessment intervals, because non-referred

parents should both consistently follow their children and

report low levels of conduct problem behaviors; (d) in PDI,

at each assessment interval and for each group, the ECBI will

not be significantly correlated with parent following. The

interpretation of parent following in PDI, as discussed

previously, depends on the balance the parent strikes between

alternating between leading and following.




The videotapes of subjects used in this study were from

24 parent-child dyads who were observed as part of another

research project (Eisenstadt, 1990). One-half of these dyads

(1 female, 11 male; mean age in months = 57.2) were referred

for treatment of their child's behavior problems. The other

half (3 female, 9 male children; mean age 56.9 months) were

solicited through advertisements in local day care and

preschools in the Gainesville area. The groups did not

differ significantly by age, t(22) = .05, ns, income t(22) =

.87, ns, or child gender 2(1, N = 24) = 1.2, ns. At the

first assessment interval, the two groups' mean ECBI

Intensity, t(22) = 4.14, 2 < .0001; and Problem, t(22) =

5.32, 2 < .0001, scores were significantly different.

The mother of each dyad had completed an ECBI and had

been videotaped in two 5-min standard situations during

standard Dyadic Parent-Child Interaction Coding System

(DPICS; Eyberg & Robinson, 1983) observations. Each dyadic

interaction was conducted in the play room in the Psychology

Clinic at the University of Florida. The room was furnished

with a table, chairs, a blackboard, chalk, eraser, and a

choice of toys including an alphabet teacher, a variety of

magnetic and wooden building blocks, and Lincoln Logs. Each

dyadic interaction was videotaped from behind a one-way


Dominance Coding System (DCSI

PCIT involves, by definition, training the mothers to

use verbal (e.g., praise, descriptions) and nonverbal

behaviors (imitating, ignoring, watching) to respond to their

child's appropriate and inappropriate behaviors. Thus,

comprehensive assessment of mother-child interaction requires

assessment of both behavioral dimensions. Furthermore, the

importance of using both verbal and nonverbal behaviors in

assessing dominance in conduct problem behavior mother-child

dyads was suggested from the results of a pilot study by

Wruble et al.(1989). These clinical researchers coded a wide

range of mother and child verbal and nonverbal behavior and

reported both that mothers are more active verbally than

their children, that children are more active nonverbally

than their mothers, and that different dominance profiles

resulted when using exclusively verbal or nonverbal behaviors

(Wruble et al., 1989).

The DPICS observational system is a standardized

behavioral observation instrument which has been shown to

discriminate clinic referred from non-referred children and

to be sensitive to changes from pre- to post-treatment.

DPICS was expanded to include 48 child and mother play

behaviors, and 3 additional verbal behaviors. The play

behaviors were adopted from the Parent-Child Interaction

Coding System (PACHINCO) developed by Wruble, Sheeber,

Sorensen, & Eyberg (1987). Another addition was the

introduction of a time-base to the system allowing each

mother and child behavior to be associated with a particular

point in time. This was done to provide a more accurate

description of the sequence of behaviors between mother and


The Dominance Coding System (DCS; Wruble, 1990) was

developed for the present study based upon the results of a

pilot study (Wruble et al., 1989). The DCS coding manual is

included as Appendix 1. The DCS consisted of the expanded

DPICS verbalizations, and two general types of nonverbal play

behaviors: activity, and the referent of the activity which

were adopted from the PACHINCO coding system. With few

exceptions, only behaviors lasting for at least one second

were coded. Behaviors lasting less than one second, such as

brief glances away from the other person, or brief pauses in

activities, were not scored because they were considered less

important for the purposes of this study, and because they

would likely be difficult to code reliably. Exceptions were

made for behaviors which typically last for less than one

second, such as hitting, touching, and caressing, because

they are important interpersonal behaviors and because

clinical experience suggested that their high degree of

recognition and face validity would allow them to be coded

Each behavioral unit was defined by the natural shift

between behaviors as the mother and child interacted, such as

when the mother or child changed their behavior (e.g., from

building to watching) or made an intelligible verbalization.

Therefore, the length between behavior changes could vary

widely from a verbal acknowledgement or the length of time

necessary to speak a sentence, to the length of time

necessary to finish a complex building task and start another

task. Recording was "activated" by the onset of a new

behavior, that is, whenever the mother or child spoke,

changed their play activity, or the object with which they

were playing. Each of these changes was viewed as an

opportunity for the mother or child to lead or follow the

other's play. Observers recorded the time and valence

(positive, negative, or neutral) of each behavior.

Verbalizations were collapsed into positive (e.g., praise),

neutral (e.g., descriptions, reflections), or negative (e.g.,

criticism) statements in line with their functional

description in the Eyberg and Robinson (1983) DPICS coding

manual (Wruble, 1990). Nonverbal behaviors were assigned a

valence depending on the social impact of the behavior;

prosocial behaviors such as giving or caressing were

considered positive behaviors while deviant behaviors such as

hitting and destructive play were considered negative

behaviors (see Wruble, 1990; Appendix 1). Information about

the time of an event was provided by reference to a running

digital time code superimposed previously on each videotape.

Mother and child univariate time series, representing

natural shifts in play activity and speech, were derived from

these data.

Dyadic Interaction Scale (DISI

The Dyadic Interaction Scale (DIS) (see Appendix 2) was

created by the author for use in this study. The DIS

consists of a total of four items. Two items are directed

toward assessing how much the observer thinks the child is

leading and following, and two items are directed towards

assessing how much the observer thinks the mother is leading

and following. Each of the four items is rated on a 5 point

scale ranging from "Almost None" to "Almost Completely,"

resulting in a total of four measures. Observers rated

leading and following immediately after viewing each 5 min


Evberu Child Behavior Inventory (ECBI)

The ECBI (Eyberg, 1974) is a 36 item parent rating scale

of conduct problem behaviors for use with children between

the ages of 2 and 17. Each behavior is rated on a frequency

of occurrence scale (ranging from I = never to 7 = always

occurs) and on a yes-no problem scale. The ECBI, then,

provides two scores. The intensity score is obtained by

summing across all frequency of occurrence scores. The

problem score is obtained by summing the "yes" responses.

The ECBI provides a sensitive measure of treatment outcome in

children with conduct problem behaviors (Eyberg & Robinson,

1982; Webster-Stratton, 1984), relates significantly to

observational measures of parent-child interactions (Webster-

Stratton & Eyberg, 1982) and discriminates conduct problem

children and adolescents from other clinic-referred and non-

referred children (Eyberg & Robinson, 1983).

Behavioral observers coded from videotapes of mother-

child interactions recorded prior to this study. Both CDI

and PDI, pre- and post-treatment, interactions were coded for

the clinic group, and for the non-referred group at two

comparable assessment intervals. The coders used the

following materials: a hi-fidelity videocassette recorder

with remote control, data sheets for verbal or nonverbal

behaviors, pencils, and a copy of the coding manual.

The same videotaped interactions of the 24 dyads (12

clinic and 12 nonclinic) described previously were also coded

using the Dyadic Interaction Scale (DIS), by seven first year

graduate students in the clinical psychology program at the

University of Florida. These coders were considered more

appropriate than either naive undergraduate students or

"expert" PCIT clinicians because they were beginning their

clinical training but were not yet familiar enough with PCIT

to expect that their ratings of leading and following would

be biased by their knowledge of the program. Each coder

viewed 22, five-min interactions which were randomized along

all experimental conditions. Of these, 16 were viewed by

only one coder, while the remaining were assessed by two

coders for reliability purposes.

Observer Training and Coding

Four coders participated in coding the dyadic

interactions. Three of the coders were undergraduate

students receiving course credit for their participation.

The fourth coder was a research assistant trained to code

exclusively the verbal data. The author served as the

reliability coder.

Coders were first trained in the use of the coding

system in groups of three to four each, then in groups of

two, and finally individually. Once the coders had achieved

an acceptable level of percentage agreement (85%) with a

criterion tape, they were allowed to code the interactions of

the subjects used in this study.

For each five-min videotaped interaction, observers used

a multiple pass, slow forward-fast backward, recording

strategy to code mother and child behavior separately. Each

coder would pass through a particular segment a number of

times until he or she determined what code(s) (e.g., activity

code and activity referent code) to assign.

Training Observers and Reliability Assessment

Observers were trained according to the following

procedure. Observers were first trained with previously

coded criterion videotapes using a copy of the DCS manual and

coding sheets. A criterion tape refers to a videotaped

mother-child interaction that has been previously coded by

two or more "expert" observers. Consequently, measures of

its contents provide a standard against which other observers

can be compared. The criterion videotapes used in this study

had been used in another study (Wruble et al., 1989) and had

been coded by at least two advanced graduate students and two

undergraduate coders with demonstrated reliability.

Accuracy refers to the degree to which coders

approximate the measures of the criterion tape. Observers

were trained to achieve a minimum of 85% accuracy based on

frequency counts of agreements and disagreements, as assessed

by percent agreement, before they were allowed to score the

videotapes of dyads used in this study. Observers were blind

with respect to all experimental conditions.

Data Analysis

Time Series Analysis

A time series is defined as one or more random variables

assessed through time (Gottman, 1981). Time series

parameters relevant to the present study include trend,

autocorrelation and crosscorrelation.

Two general types of trend may exist, linear and

cyclical (Gottman, 1981). Linear trend refers to a positive

or negative slope in the time series. A positive trend, for

instance, would be represented by a cumulative record of a

consistent rate of responding (e.g., that associated with a

variable interval schedule of reinforcement). Cyclical trend

refers to a rhythmic component in the series. For example,

Glass, Wilson, & Gottman (1975) noted that time series of

retail sales exhibits a 12 month cyclical trend (i.e., prices

in December are highly correlated with prices last December,

and because of the post-holiday drop in retail sales, retail

prices in January are highly correlated with the same drop

the previous January, etc.). Trend is important because it

provides information about one of the assumptions of a time


Time-series analysis assumes that the data are

stationary (Gottman, 1981). The first assumption is that the

mean and variance of the time series remains the same

independent of time. In other words, the time series must

exhibit no linear or cyclical trend and the variance must be

constant throughout the series.

The second assumption is that autocorrelations do not

change as a function of location in the time series.

Autocorrelation refers to the relationship between present

values and past values of a series. Autocorrelation is

reportedly pervasive in behavioral data (Jones, Vaught, &

Weinrott, 1977). With regard to the present study,

accounting for autocorrelation is important for at least two

reasons. First, its presence violates the assumptions of

independent and randomly distributed error terms common to

traditional inferential statistics. Failure to correct for

autocorrelation results in biased estimates of confidence

intervals (Gottman, 1981) and of significance tests (Kazdin,

1976; Hartmann, Gottman, Jones, Gardner, Kazdin, & Vaught,

1980). For instance, the presence of positive

autocorrelation results in a smaller estimate of the standard

error term which artificially inflates the estimate of the F-

value, thereby increasing the Type I error rate. Second, the

effect of autocorrelation must be removed in order to assess


Crosscorrelation refers to the relationship between the

values in one time series with the values in another time

series. Applied to social interaction, for instance, it

refers to the degree to which an individual's behavior (e.g.,

the child) is predictable from the past behavior of the other

person (e.g., the mother) (Gottman, 1981).

Patterns of autocorrelation and crosscorrelation have

been used to determine dominance in interactions. Heidelise

et al.. (1977) used contingency table data to investigate

auto- and cross-dependency in mother-infant interactions.

Unfortunately, for reasons to be discussed shortly, their

analyses did not control for relations between

autocorrelation and crosscorrelation, which is necessary to

determine dominance.

Thomas and Martin (1976) developed a two component model

of dyadic interaction. They speculated that parent-infant

interaction reflects the operation of a self-regulatory

component and a social interactive component. They provided

a statistical model of parent-infant interactions from which

they obtained summary statistics representing both components

of the interaction. According to their model, the behavior

of a person at a point in time was considered a linear

function of the person's own past behavior and the past

behavior of the person with whom he/she was interacting.

Thomas and Martin's (1976) model, however, is unable to

provide an adequate test of dominance for two reasons.

First, they used crosscorrelation measures to estimate the

values of the interactive components, but they did not

control for the effects of autocorrelation originating from

the self-regulatory component; and autocorrelation and

crosscorrelation can be related such that crosscorrelations

can be produced from strong autocorrelations (Gottman &

Ringland, 1981). Moreover, to infer, for example, mother

dominance in parent-child interactions, the behavior of the

child must be more predictable from the past behavior of the

mother than from the past behavior of the child. In order to

infer dominance, then, two models must be compared; a model

with autocorrelation only and a model with both

autocorrelation and crosscorrelation. Second, Thomas and

Martin (1976) did not provide an algorithm to fit models with

greater than two parameters, although conceivably their model

could. Consequently, there is no way of knowing whether the

Thomas and Martin (1976) two-parameter model is the model of

best fit (i.e., explains the most variance in the


Gottman and Ringland (1981) presented a model of dyadic

interaction that accounts for the methodological and

statistical shortcomings of the Thomas and Martin (1976)

model. Gottman and Ringland's methodology will be detailed

because it provided the basis for the time series analytic

procedures used in this study.

Broadly speaking, Gottman and Ringland's main purpose

was to determine the most parsimonious model to predict the

behavior of a mother and her infant during face-to-face

interactions. Their methodology is applicable, of course, to

other types of interaction and hereafter will be applied with

reference to parent-child interaction.

The Gottman and Ringland (1981) procedure involved using

a maximum likelihood procedure to compare a reduced model, in

which only autocorrelation parameters are used to predict the

mother or child series, with the prediction from a full model

in which both auto- and crosscorrelation terms are included.

The null hypothesis of this procedure is that including

crosscorrelation terms provides no increase in prediction,

suggesting that the parent's or child's behavior is a

function only of their own past behavior.

The Gottman and Ringland (1981) algorithm involved the

following steps. First, they used an ordinary least squares

procedure to estimate the initial autocorrelation parameters.

Once estimated, slightly larger autoregressive parameters

were estimated. Different combinations of these parameters

were then combined with the crosscorrelation parameter

estimates, resulting in a full and reduced model with

slightly inflated estimates. Then, using a least squares

procedure adapted from Mann and Wald (1943), Gottman and

Ringland (1983) reduced these models in a stepwise process.

Specifically, for both mother and child time series, the

least squares procedure selected the best full model and the

best reduced model to fit each series. Then, separate

likelihood ratio tests compared the full versus the reduced

models for the null hypothesis that crosscorrelation terms do

not contribute to the prediction of the mother or child

series. The likelihood ratio statistic for this test is a

fraction in which the numerator is the liklihood of the data

fit with auto- and crosscorrelation terms, and the

denominator is the liklihood of the data fit with only

autocorrelation terms. All liklihood ratio statistics are

approximately distributed as a Chi Square with degrees of

freedom equal to the difference in the number of parameters

between the full and reduced models (Lindgren, 1968). Last,

the parameter values of each full model were varied to

examine if a better fitting model could be achieved, and

similarly, additional autoregressive terms were added to the

reduced model to examine if a better fitting model could be

achieved. Neither of these tests were expected to give

significant results, and served essentially as checks on the

internal validity of the procedure.

The advantages of the Gottman and Ringland (1981)

procedure were that it permitted the assessment of

crosscorrelation by controlling for the effects of

autocorrelation, and it used a maximum likelihood procedure

to choose a model of best fit. These models provided the

additional benefits of providing an equation with parameter

estimates, residual variance information for both the mother

and infant, and a direction of influence classification for

the dyad.

The Gottman and Ringland (1981) procedure is not without

both procedural and methodological limitations. First, the

use of step-wise ordinary least squares to estimate the

number and value of autoregressive parameters may not be the

procedure of choice (Newton, 1988). Instead, Parzen's (1977)

criterion autoregressive transfer function (CAT) has

developed as a method to obtain stable estimates of the

number of autoregressive parameters in the time series model.

The accuracy of the CAT criterion to detect white noise

(i.e., random uncorrelated observations) has been reported

elsewhere (Newton, 1988) and was assessed empirically in this

study. Parzen's CAT criterion was used on 100 white noise

processes, whose composition was randomly selected by the

Timeslab (Newton, 1988) software program. If the CAT

criterion was working properly it would be expected to

correctly diagnose each time series as uncorrelated white

noise. This process, then, served as a calibration or

accuracy check on the performance of the CAT criterion.

Results from a Monte Carlo computer simulation suggested that

approximately 85% of the white noise processes were diagnosed

as white noise by the CAT criterion.2 This means that in

2 The author is indebted to Mike Conlon for conducting the
computer simulations to empirically determine the performance
of the CAT criterion used in this study.

this study, the Type I error rate for the diagnosis of

autocorrelation actually occurred at the .15 level. This

level is more than the conventional .05 probability level,

but should not affect the results of this study because any

effect of autocorrelation on the cross-correlation parameters

is automatically removed by the time-series algorithms used

in this study. The Gottman and Ringland (1981) procedure was

also compromised by a computational error in the likelihood

ratio provided. This was corrected, making the statistical

procedure used in the study a refinement of the Gottman and

Ringland (1981) methodology.3

All analyses were conducted using TIMESLAB (Newton,

1988) statistical programs (see Appendix 3) and software.4

Inspection of a number of time series revealed that most were

non-stationary. Therefore, linear trend was automatically

removed from all time series. Detrending a time series does

not affect autocorrelation or crosscorrelation because it

involves removing a constant term from the series.

In the present study, the valence (e.g., positive,

negative, or neutral) of each verbalization or each change in

play behavior was collapsed into a binary value (I = change,

0 = no change) in order to assess the sequential patterning

3 The author is completely indebted to Mike Conlon for
detecting and ameliorating the problem discussed.

4 The author wishes to thank Mike Conlon for constructing the
time-series macros used in calculating the dominance

between mother and child. In other words, time series were

formed on the basis of assigning a value of one at the time

when the mother or child changed their activity or made a

statement, and zero values were assigned otherwise. For

instance, if a child maintained his or her play activity and

was silent for the first 9 seconds of the interaction but

made a statement at time = 10 seconds, then this was

represented in the time-series as 9 successive zeros (for the

first nine seconds) followed by a 1 for the change in

behavior at time = 10 seconds (i.e., 0000000001). Hence

separate child and mother time-series were composed of a

series of zeros and ones.

Time Series Models and Summary Measures

Two issues pervade time-based interaction research. One

is the question of how far back in time to search for

patterning. In a pilot study with conduct problem behavior

mother-child dyads, Wruble et al.. (1989) determined that as

the number of parameters (from 1 to 60) in the time-series

models increased, the amount of variance accounted for by the

models reached a point of diminishing returns. Each

parameter represented one second. Based on the results of

the Wruble et al. (1989) study, the time-series models used

in the present study were limited to quantifying the lag

between mother and child behaviors within 30 seconds.

The other question is how to summarize complex

interactions; specifically, how to turn interaction process

into interaction measures. Interaction researchers have

recommended using summary measures of interactions as

dependent variables to answer groups comparison design

questions (Bakeman & Gottman, 1985; Sackett, 1979). Six

summary measures of dominance were formed by dividing the

estimates of the 30 time-series parameters into 6 cumulative

lag intervals. Thus, Chi-square values from latter lag

intervals were based on earlier values. For instance, Chi-

square values for lag interval one were based in the amount

of crosscorrelation that occurred between the mother and

child within a five sec time span, while the Chi-square

values for interval two were based on the amount of

crosscorrelation that occurred with a ten sec time span

including the five seconds from the first lag interval, and

Chi-square values for lag interval six were based on the

amount of crosscorrelation that occurred between the mother

and child within a 30 sec time span, including the 25 sec

from the prior 5 lag intervals. Chi-square values of the

crosscorrelation parameter estimates for each of these 6 lag

intervals were calculated and served as dependent variables

for comparisons between clinic and nonclinic groups, and

between CDI and PDI within groups and across comparable

assessment intervals. Separate time-series analyses were

performed for each dyad in each experimental condition.

Chi-square values of child following in each lag

interval represented how predictable the child's behavior was

given antecedent mother behavior(s) occurring in the prior

six, 5-sec, lag intervals (i.e.,the prior 30 sec). Likewise,

Chi-square values of mother following represented how

predictable the mother's behavior was given antecedent child

behavior(s) occurring in the prior 30 seconds. Expected Chi-

Square values increased linearly as a function of lag

interval. This is because the number of parameters in the

time series-models increases as the lag increases (as the

search for pattern recedes back in time) which increases the

degrees of freedom for the Chi-Square statistic. The

expected Chi-square values for each lag interval were 11.07,

18.31, 25.00, 31.41, 37.65, and 43.77 for intervals 1 through

6 respectively. In the case of child following within a lag

interval, Chi-square values exceeding the expected value for

that interval indicated that the child was following the

mother's antecedent behavior(s) at a level greater than would

be expected by chance alone. With regard to mother

following, Chi-square values exceeding the expected value for

that interval indicated that the mother was following the

child's antecedent behavior(s) at a level greater than would

be expected by chance alone. Therefore, Chi-Square values

larger than that expected by chance alone indicate patterning

in the mother-child dyad, with increasingly greater values

indicating increasingly patterned interactions. It is

important to note that significant Chi-square values both

indicate patterns of following and imply that the other

person is leading the interaction. In statistical terms,

leading and following are mutually implied. But this does

not necessarily mean that the other person is intending to

lead the interaction. The important distinction to be made

is between what is implied in the statistical definition of

leading and following, and whether the person leading was

intending to lead. During mother-infant interaction, the

mother follows the infant's attentional and affective rhythms

in the sense that her behavior is statistically dependent on

the infant's past behavior, while the infant is considered to

be leading the interaction because his or her behavior

predicts subsequent maternal behavior. Most researchers,

however, would be hesitant to conclude that the infant was

intending to lead the interaction as this may be outside his

or her developmental level. The statistical versus

intentional interpretation of leading and following is less

than clear cut, and in fact they can complement each other.

In this study, the statistical definition and measure of

leading and following was used in conjunction with the CDI-

PDI experimental manipulation designed to control the

mother's following and leading intentions.



Reliability of the DCS

Reliability measures should reflect closely the type of

measures used in the study (Johnson & Bolstad, 1973; Gottman,

1979). This suggests that observational studies examining

time-based patterns must calculate reliability by examining

observer agreement at each point in time and by examining

congruence in these patterns (Bakeman & Gottman, 1986). The

present study examined sequential patterns through time in

mother-child dyads. Consequently, observer agreement was

assessed at a moment-by-moment level. Agreement coded in

this way requires two considerations: (a) that the observers

agree on the type of behavior (e.g., positive, negative, or

neutral in the present case) and (b) when the behavior

occurred. The first consideration is easily assessed by, for

example, comparing the number of positive behaviors coded by

the original and reliability observer. The question of

interobserver agreement on when the behaviors occurred,

however, is more difficult because of differences in

videorecorder fidelity, observer reaction time, and

individual differences in discrimination ability. In

recognition of this fact,time-based observational studies

typically allow a "window" of variation for calculating

reliability. Results from a previous study (Wruble et al.,

1989) indicate that a two-second window (i.e., interval of

variability) accounts for these problems without

substantially inflating agreement. In the present study,

then, an agreement was defined as the case when the original

and reliability observer both recorded a behavioral change

occurring within two seconds of each other. For cases in

which two events occurred within the same two second

interval, only one agreement was scored. For cases in which

two or more events occurred within the two second interval,

but only one of the events matched the other observer's

coding, then one agreement and one disagreement were coded.

Three reliability scores will be presented for the DCS

data: percentage agreement, percentage agreement controlling

for chance agreement (i.e., Kappa), and Pearson Product

Moment correlations.

Percent agreement is the most frequently used index of

observer agreement. This index is simply the number of

agreements divided by the sum of agreements plus

disagreements. Although elegant in its simplicity, this

index fails to account for the amount of agreement that could

be obtained by chance alone. Failure to account for chance

agreement can result in inflated agreement values. Kappa is

considered a superior measure (Hollenbeck, 1978) because by

controlling for chance agreement it provides a less biased

estimate of agreement (Cohen, 1960).

Reliability was formally assessed at the beginning of

the study and again just prior to the conclusion of the

study. Informal reliability spots checks were also conducted

with both specific and general feedback provided to the

coders during a group research meeting. Fully one-third of

the data was used for reliability calculations over these two

assessment intervals.

Reliability was calculated by tallying agreement and

disagreement for the natural shifts in mother and child

behavior at each point in time. This procedure was conducted

separately for the verbal and nonverbal data, for each

condition (CDI and PDI), at pre- and post-treatment, and for

the mother and child. Data were then combined across all

experimental conditions resulting in a total of over 8600

behavioral changes. Only 8% of the behaviors were coded as

either positive or negative and of these almost all were

scored as positive. Overall, reliability uncorrected for

chance agreement was moderate at 76% with an associated Kappa

of .50. Fleiss (1981) has characterized Kappas of .40 to .60

as "fair." The reliability of the more specific codes of

positive, negative, and neutral was similar at 74%, but with

an unacceptably low Kappa of .22. This low Kappa value may

be the result of low variability resulting from the fact that

only 8% of the behaviors were coded as positive (8%) or

negative (.0005%).

At the first reliability assessment, percent agreement

for behavior changes was moderate at 79% with an associated

Kappa of .55, while at the second reliability assessment,

percent agreement and Kappa were comparable at 77% and .53,

respectively. At the first reliability assessment, percent

agreement and Kappa for the more specific codes of positive,

negative, and neutral codes were .75 and .28, respectively.

But at the second assessment, percent agreement was .73, and

Kappa was .53 which was substantially higher than at the

first assessment. The subject data used for reliability

purposes was balanced across all experimental variables. The

probability of chance agreement for the first and second

assessment intervals were .65 and .53, respectively.

Furthermore, approximately the same number of behaviors were

sampled at the first interval (N=4205) as at the second

(N=4336) interval. Thus, the differences in the Kappa values

between the two assessments is likely due to greater

variability (i.e., a more equal distribution of behaviors

across codes) in the interactions upon which the second

reliability assessment was based.

Across both assessment intervals, changes in verbal

behaviors were more reliable (Agreement= 80%, Kappa=.44) than

changes in non-verbal behaviors (Agreement=70%, Kappa=-.06).

The overall reliability of the activity referent behaviors

was 64% (Kappa=.44). Percent agreement and Kappa values did

not differ substantially as a function of group or person


The inter-observer reliability of the Chi-square values

were assessed by Pearson Product Moment correlations.

Significant inter-observer reliability was obtained for only

the first three lag intervals of the child data (interval

one: L(32) =.49, 2 < .001; interval two: K(32) =.42, 2 < .02;

interval three: L(32) =.42; 2 < .02). The correlations for

intervals four through six were not significant.

Interobserver reliability of the Chi-square values was

not established for any of the lag intervals of the mother

data. Hence, the DCS observers were unable to agree on the

patterning of the mothers' behavior regardless of the time

interval. Consequently, repeated measure ANOVAs using the

mothers' data were not performed.

Dominance as discussed in this study referred to

asymmetry in predictability between the mother and child

during their dyadic interactions. Assessing dominance in

this study, then, required both mother and child data.

Unfortunately, limitations in the reliability of the mother's

data did not allow a direct test of the dominance construct.

Instead, this study evaluated the psychometric properties of

a measure of child following. In discussing the results of

this study the term dominance and child following are

sometimes used interchangeably, to refer to the

predictability between child behavior and antecedent mother

behavior. Hence, in this study the term dominance is used

more broadly.

Groups Comparison Results

The child data (see Appendix 4) were analyzed with three

separate 2 X 2 X 2 (Group X Situation X Time) ANOVAs, with

repeated measures on the last two factors, and clinic versus

nonclinic as the group variable. Dependent variables for

these analyses were the Chi-Square values from the first

three lag intervals, respectively. In lag interval three, a

main effect for Time, E(l, 22) = 6.54, 2 < .0179, and a Group

X Time interaction, E(i, 22) = 4.54, R < .04, were

significant, but they were considered less important because

a significant three-way interaction (Group X Situation X

Time) was obtained, E(l, 22) = 4.37, 2 < .0485. A

significant main effect for Time was also obtained in lag

interval one, E(l, 22) = 7.46, 2 < .0122. This effect was

not interpreted separately because Chi-Square values from

latter lag intervals are based on earlier values; hence Chi-

square values, and effects, from the third interval subsume

those from the prior two intervals. The latter is the case

because the time-series models for lag interval three may

estimate child following more comprehensively than time-

series models from the other two lag intervals, because the

models assess child following over a larger interval of time.

For purposes of clarity only effects from interval three will

be presented.

In an effort to partial out the effects of the three-way

interaction, a two-way ANOVA by Group--with Student Neuman

Keuls post-hoc comparisons--was conducted using the child

data from lag interval three. A significant Situation X Time

interaction, F(i, 22) = 4.79, a < 051, was obtained for the

clinic children. Figure 2 illustrates the Situation X Time



interaction found for the clinic children. In an effort to

better understand this two-way interaction, a one-way ANOVA

was performed. The prediction that the clinic children would

follow their mothers less in CDI at post-treatment, as

compared to pre-treatment, was confirmed, (i, 22) = 6.65, 2

<.018. The prediction that clinic children would follow

their parents more in PDI at post-treatment compared to pre-

treatment was not, however, supported, E(1, 22) = 1.30, ns.

No significant effects were observed for the non-referred
children in CDI, E(1, 22) = 3.86, ns., or in PDI, Fjl, 22) =

2.47, ns., between the two assessment intervals.

Clinic and nonclinic children were also predicted to

exhibit different levels of following behavior, as a function

of situation, within each assessment interval. The

prediction that clinic children would follow more than

nonclinic children in CDI at the first assessment interval

was supported L(22) = 2.52, a < .02. The prediction that

clinic children would follow less than nonreferred children

in PDI at the first assessment interval was not supported,

t(22) = -.66, ns. As predicted, the clinic and nonclinic

children did not differ significantly in the amount of

following at the second assessment interval in either CDI,

t(22) = -1.09, ns, or PDI, t(22) = -.76, ns.

In summary, the DCS measure of child following in CDI

was able to discriminate between clinic and nonclinic

children at the first assessment interval, and was sensitive

to treatment changes. The decrease in following behavior

from pre to post-treatment for clinic children indicates that

PCIT had the effect of making the clinic children's behavior

less dependent on antecedent maternal behavior in the CDI

situation. This effect may indicate more flexibility in the

mother-child interactive behavior patterns. The finding that

clinic children followed less at post-treatment in CDI

compared to the same situation at pre-treatment, whereas the

non-referred children's following behavior did not differ

significantly over the same assessment intervals suggests

that child following is an important component of PCIT.

Validity of the Dominance Measure

Concurrent Validity

The concurrent validity of the child following was

assessed by correlating the Chi-squares values with the ECBI

Problem and Intensity scores. The mothers' data were not

analyzed because of the reliability limitations noted


Table 1 illustrates that at pre-treatment CDI for the

clinic children, ECBI Intensity scores were significantly and

positively correlated with child following at each of the

three lag intervals (interval one, z(12) = .70, 2 < .01;

interval two, E(12) = .72, 2 < .01; and interval three, r(12)

= .63, 2 < .05), and with ECBI Problem scores for the first

lag interval, r(12) = .62, 1 < .01. Table I also illustrates

that the significant positive correlation between child

following and ECBI scores for clinic children in CDI at pre-

treatment were not evident for these children at post-





2 3
.87*** .765**













* p < .05
** p < .01
*** P < .001


2 3
.67* .63*












treatment. ECBI scores and child following were not

significantly correlated for non-referred children in CDI at

comparable assessment intervals.

Convergent Validity

The convergent validity of the dominance measure was

assessed by separately correlating the child following Chi-

square values with the DIS ratings of child leading and

following. Table 2 illustrates that the child following Chi-

square values and the DIS ratings of leading and following

for the children were not significantly correlated.

The concurrent validity of the DIS was assessed by

correlating the DIS ratings of leading and following with the

ECBI Problem and Intensity scores within each experimental

condition. The results of these analyses failed to establish

the concurrent validity of the DIS. DIS ratings of child or

mother leading and following were not significantly

correlated with ECBI Problem and Intensity scores in any

experimental condition.

The inter-rater reliability of the DIS ratings was

calculated on approximately one-half of the data. The inter-

rater reliability of the DIS ratings of child leading was

L(41) =.55, a < .0002, and the inter-rater reliability of

child following was X(41) = .38, 2 < .01. Similarly, the

inter-rater reliability of the DIS ratings of mother leading
and mother following were r(41) = .60, 2 < .0001, and r(41) =

.62, 9 < .0001, respectively.



Lag Interval
1 2

.093 .097

.022 -.015.

.011 -.013

.099 .080

Lag Interval
1 2 3

.041 -.05 -.192

.004 .082 .032



.02 .022

.001 -.124

Note: N = 96 non-independent observations.










The DIS ratings of child and mother leading and

following were intercorrelated. The correlation between

child lead and mother follow was high, positive, and

significant Z(41) = .78, 2 < .0001. Similarly, the

correlation between child follow and mother lead was high,

positive, and significant Z(41) = .82, 2 0001. The

correlation between child lead and child follow, and between

child lead and mother lead were moderate, negative, and

significant r(41) = -.53, 2 < .0001; and 1(41) = -.56, 2 <

.0001, respectively. The correlation between mother lead and

mother follow lead were moderate, negative, and significant

L(41) = -.65, 2 < .0001.



This study provided a time-based observational measure

of dominance, estimates of some of its psychometric

properties, and some empirical support for the position that

child following is an important component in PCIT.

Dominance Coding System (DCS)

A defining feature of any mother-child interaction is

that it unfolds in time. The measure of mother and child

sequential patterning used in this study captured this aspect

of the process. The present observational measure of

sequential patterning was based on coding discrete

alternations in play behaviors and speech. This coding

strategy had been used in a pilot study using a similar

subject population and provided promising results (Wruble, et

al., 1989). The DCS provided a first step toward assessing

the patterns of leading and following that occurred in the

mother-child play interactions and dialogue during two quasi-

naturalistic clinical play situations.

Dominance was defined by asymmetries in predictability

between the mother and child during interaction. The

unreliable nature of the mother data precluded an assessment

of dominance by limiting analyses to the predictability of

child behavior given prior mother behavior, but not vice


One important limitation of the DCS coding system is

that information about content (e.g., type of statement) and

affect were not retained. These areas were not assessed

mainly for pragmatic reasons. Initially, coders were trained

to code both the content of each statement and each specific

type of nonverbal behavior. Using the more detailed system,

observers needed approximately 18 min of coding time per 1

min of parent-child interaction. The highly detailed coding

system was streamlined early in the study by collapsing the

specific verbal and nonverbal behaviors into the general

categories of positive, negative, and neutral depending on

the social impact of the behavior. The DCS coding system

used in this study--composed of the three general categories

of positive, negative, and neutral--only required

approximately 6 to 12 min of coding time per min of parent-

child interaction. Affect was not coded in this study for

two reasons. First, a main emphasis of this study was to

examine the psychometric properties of the measure of

dominance developed by Wruble et al. (1989), which did not

include affect variables. Second, adding affect codes to the

DCS would have made the coding system prohibitively time


Another limitation of the DCS is that it lacked

information about the continuous change in affective

intensity and attentional involvement over the course of the

interaction. Consequently, valuable information about the

level of affective and attentional involvement between the

mothers and children was unavailable. The affective

patterning in the parent-child relationship seems especially

important given that PCIT facilitates the development of a

warm, positive, and genuine parent-child relationship, and

because PCIT teaches parents to control their child's

behavior by using differential social attention. It is

likely that a measure that captures the continuous ebb and

flow of affective and attentional involvement during mother-

child interaction would provide a more direct assessment of

the quality of the relationship in general and the quantity

of dominance in particular. This type of coding strategy is

not without historical precedent. In their study of

synchrony (i.e., dominance) in parent-infant interaction,

Tronick et al., (1977) were the first to convert specific

behaviors into values along a univariate scale of

involvement. Gottman (1979) later used the same univariate

scaling strategy in his study of dominance in marital

interaction. Converting DPICS and nonverbal behavior codes

into univariate scale values could easily be accomplished by

intuitively or empirically scaling these behaviors along a

predetermined scale (e.g., attentional involvement). The

resulting time-series could then be analyzed using the semi-

automated statistical packages developed and used in the

present study.

The DCS is not practical for everyday clinical use

because of the large investment in coding time and data

analysis necessary to obtain a measure of leading and

following. PCIT researchers are likely to profit from

correlating frequency measures of leading and following with

process measures of the same. If high correlations are

obtained then researchers and clinicians could be more

confident in their use of frequency measures as estimates of

leading and following to assess clinic versus nonclinic

status and treatment changes.

Psychometric Properties of the Dominance Measure

A major emphasis of this study was to assess the

psychometric properties of the DCS dominance measure. This

is the first social interaction study in general, and PCIT

study in particular, to assess the reliability and validity

of a time-based observational measure of social interaction

and dominance, respectively.

Parent-infant research has implicated social leading and

following as important to the development of social

competence. These studies have, however, been based on

anecdotal reports or small samples, and have used inadequate

statistical methods to assess leading and following.

Additionally, with the exception of Gottman's (1979) marital

interaction research, the dominance construct has been poorly

defined and researched. Finally, the reliability and the

multi-method tests of the validity of dominance measures have

not been previously reported in any literature.

Leading and following has also been considered important

to the effectiveness of PCIT. Likewise, prior to the present

study PCIT research had not defined, measured, or examined

the leading and following process. Consequently, prior to

the present study there was no empirical support for the

assumption that leading and following is important in the

development of social competence in general, or to the

effectiveness of PCIT in particular.

Inter-observer reliability

Inter-observer agreement corrected and uncorrected for

chance was significant for the mothers' data. The

reliability of the mothers' Chi-square values, however, was

nonsignificant. These results taken together indicate that

the observers often agreed on the overall rate or frequency

of the mothers' behaviors, but were unable to agree on the

timing and patterning of the mothers' behaviors.

The present study highlighted one limitation of using

the Kappa measure to assess sequential patterning which is

that Kappa values are based on agreement within time as

opposed to concordance in patterning across time. For

instance, consider the following contrived sequences of

mother-child interaction (where M-mother, C=Child, and the

a=positive, b=negative, and c=neutral subscripts represent

the type of behavior change):

Observer 1:

Ma Mb Ma Mb Mc

Ca Cb Ca Cb c

Time 1 5 10 15 20

Observer 2:

Ma Mb Ma Mb Mc

Ca Cb Ca Cb CC

Time 1 5 10 15 20

Both observers coded the same number (5 each for mother

and child) and types of behavior (i.e., positive, negative,

or neutral). Using a conventional two-second window they

were in perfect agreement. It is clear however, that the

observers were in complete disagreement about who was leading

the interaction. The data generated by the first observer

seems to suggest that the child is leading the interaction,

while the opposite is true for the second observer. If these

time-series were extended out and analyzed for dominance

patterns, what would be expected? With regard to the child

data it would be expected that the resulting Chi-square value

for the first lag interval from the first observer's time-

series would be low (e.g., 3.00) because the child seems to

be leading. The comparable Chi-square value from the second

observer's time-series would be high (e.g., 10.00) because

the child seems to be following the mother's lead! What this

example demonstrates is that the reliability of the

sequential patterning between the mother and child's behavior

is likely better estimated by the correlation between the

sequential patterning measures (i.e., Chi-square values of

the crosscorrelation parameters generated from the original

and reliability observer's time-series models).

This is the first study to calculate the reliability of

the sequential patterning measures themselves. The results

are quite sobering. Inter-observer agreement, Kappa, and the

reliability of the Chi-squares values of sequential

patterning were consistent only for the first three lag

intervals of the children's data. This suggests that the

coders agreed on the patterning of the children's behavior

with respect to the mothers' antecedent behavior within any

15-sec period, but not thereafter.

The inter-observer reliability of the Chi-Square values

obtained for the children's data were lower than are

typically considered acceptable using the standards developed

from questionnaire data. The DCS inter-rater reliability

measures obtained in this study were tentatively treated as

acceptable pending the establishment of reliability standards

for time-based sequential data, and because the DCS measure

was both robust enough to discriminate between clinic and

nonreferred children in CDI at pre-treatment and was

sensitive to treatment changes in the clinic children's

following behavior.

The reliability of observational measures is thought to

be influenced by a number of factors including the complexity

of the interaction and the complexity of the coding system

used to assess the interaction (Bakeman & Gottman, 1986;

Reid, Skinrud, Taplin, & Jones, 1973). Observers using time-

based observational coding systems must agree not only on the

total frequency of behaviors, but face the the additional

challenge of having to agree on the sequential patterning of

these behaviors. As indicated previously even slight

variations between observers in coding sequences can have

dramatic effects on the reliability of their observations.

The results of this study indicate that one cannot

assume that agreement in frequency means agreement in

sequential patterning. The discordance between the frequency

(i.e., percent agreement and Kappa) and pattern measures of

reliability observed for the mothers' data in this study

strongly suggests that social interaction researchers use

multiple methods to assess empirically the reliability of

their measures of patterning. In particular, the inter-

observer reliability of sequence measures should be assessed

during observer training, and each time reliability is

evaluated formally. The former would facilitate providing

specific feedback to the coders regarding the reliability of

their discriminations of the sequencing of the interaction,

while the former would allow for evaluation of changes in the

observer's sequential reliability over the course of the

study (i.e., observer drift or decay).

One explanation as to why the children's and mother's

patterning data were reliable and unreliable, respectively,

is that the child's behavior may have been easier to code.

Verbal and nonverbal inter-observer reliability was extremely

similar for the mother and child data. Comments made by the

observers, however, suggest that the mothers' behavior

changes may have been more difficult to code, perhaps because

adult behavior is more subtle. For instance, maternal

transitions in play often take the form of complex sequences

in which the mother watches the child, comments, watches

something else, or gazes at the toys with which the child is

playing. Sequences such as these were consistently difficult

to code given the camera angle and the uncertainty as to

where the mother's attention was focused. Decision rules

were developed to bring some order to these situations, but

scoring often required the informed judgement of the coder.

Children, on the other hand, would typically make distinct

transitions from playing to watching, or from playing with

one toy to playing with another toy. In contrast to the DCS,

the inter-observer reliabilities obtained for the DIS were

considered unacceptable by conventional standards.

Validity of the DCS

Concurrent Validity. ECBI Intensity scores were positively

related to the DCS measure of child following only for the

clinic children at pre-treatment CDI. At this assessment

interval, significant positive correlations between ECBI

Intensity scores and child following were found at all three

lag intervals, and a significant positive correlation was

found between ECBI Problem scores and child following, but

only in the first lag interval. No significant correlations

were observed between child following and ECBI Intensity or

Problem scores for the clinic children in CDI at post-

treatment, or for the non-referred children at comparable CDI

assessment intervals. This pattern of results indicates that

for conduct problem children, child following is related to

the frequency of conduct problem behaviors. As mentioned

previously, clinic observations and research indicate that

mothers of children with conduct problem behaviors have

difficulty following their children during clinical play

interactions which is often manifest by the mother's leading

the interaction when they should be following. Before

beginning CDI, the mothers were instructed to let the child

choose any activity, to follow the child's lead, and to play

along with the child according to the child's rules.

Mothers' reasons for not following the child in his/her game,

are likely numerous, varied, and could include not

understanding the task, not knowing how to follow the child

(i.e., a social competence deficit), or refusing to cooperate

even though the mother is socially competent, perhaps because

the mother refuses to "spoil the child," or "to let the child

have his/her way." The interpretation of the relation

between child following and ECBI Intensity scores for the

clinic dyads in CDI would be clarified by knowing the

relation between mother following and ECBI scores in CDI at

pre- and post-treatment. Unfortunately, limitations in the

reliability of the mother's following data precluded knowing

if, and how, the mother's following behavior in CDI at pre-

and post-treatment was related to ECBI scores.

Competent social following is probably related to

dimensions having to do with the quality of the mother-child

relationship. These include, but are not limited to,

effective listening and responsiveness, sensitivity, feeling
"connected," emotionally close or "in synch," as well as

perceptions about who's directing the interaction (i.e.,

who's in charge) and how much the person following trusts and

is comfortable following the other person's lead.

Convergent Validity

21a. The results of this study suggest that the DIS has poor

psychometric properties. In addition to exhibiting poor

inter-observer reliability, the concurrent validity of the

DIS with the ECBI was not established5, and the DIS was found

to be an inadequate measure for evaluating the convergent

validity of the DCS measure of leading and following.

The pattern of moderate, negative and significant

intercorrelations between child-lead and mother-follow, and

vice versa, on the DIS suggest that the observers tended to

rate leading and following within a person, and between

interactants, as being opposing tendencies. For instance, if

5 An intra-class correlation test would have been more
appropriate, but was not conducted because near zero
correlations were found using even this more liberal

the child was rated as leading then the child would also be

likely to be rated as not following the mother, and the

mother would be rated as following but not leading the child.

Taken together these results indicate that the raters tended

to pair child following with mother leading and vice versa,

and that they had some difficulty agreeing on these ratings.

This indicates a logical, simplistic, and seemingly invalid

approach to coding leading and following. Assuming that the

DSC measure quantifies, at least partly, the dominance

process between mother and child, then the DIS ratings appear

to assess the raters' biases more than mother-child leading

and following behaviors. The DIS ratings reflect the raters'

tendency to rate leading and following as opposite tendencies

as compared with behavioral sequences that establish leading

and following.

Apparently, the DCS and the DIS measures assessed

entirely different behaviors or processes. The lack of a

relationship between the DIS and the ECBI scales indicates

that it is unlikely that the ratings were based on disruptive

behaviors. At this point it is unclear what behaviors the

DIS raters were using in forming their judgements. This may

have resulted because they received insufficient instructions

(see Appendix 2). They were simply asked to rate child and

mother leading and following without being informed about

what behaviors to use in their judgements. This was done

with the intent of keeping the raters blind to the

experimental conditions, and also to avoid biasing their

ratings. The latter reason is precisely why PCIT clinicians

were not used. The raters were first year graduate students

who were familiar enough with PCIT to know the difference

between CDI and PDI but had never coded using DPICS or served

as a PCIT therapist. It was felt that if raters knew what

behaviors were thought to indicate leading and following and

how these relate to CDI and PDI, that their ratings of

leading and following might be influenced by what they

thought should be happening in the situation (e.g., the

mother should be following in CDI).

PCIT researchers who are interested in developing rating

measures of leading and following are advised to think

through these issues carefully and exert more experimental

control over the type of information given to the raters.

Preferably, future rating scales will be based on empirical

studies that have identified the interactive behaviors that

establish patterns of leading and following.

Discriminant Validity

Groups Comparison Results. Repeated measures ANOVAs of the

DCS showed that clinic children followed their mothers less

in CDI at post-treatment compared to pre-treatment. This

result was consistent with predictions for clinic children in

CDI. It was expected that clinic children would follow their

mothers less in CDI after treatment because clinic mothers in

CDI pre-treatment often lead the interaction even though they

have been instructed to follow the child in play.

Consequently, prior to treatment the clinic children are

often coerced into following as a result of their mothers'

attempts to lead. PCIT coaches mothers in effective and

satisfying leading and following interaction patterns.

Support was not provided for the prediction that clinic

children would follow more in PDI at post-treatment compared

to pre-treatment. After further reflection, perhaps this

result is not too surprising. As detailed previously, during

the PDI phase of PCIT, clinic-referred parents are coached to

alternate between leading and following. For instance,

parents are always instructed to introduce commands within

the context of a warm, genuine interaction in order to

facilitate child compliance. Moreover, clinical experience

suggests that over the course of treatment parents often seem

to feel more comfortable following their children, perhaps

because their attempts are more effective, and because it

makes them feel emotionally closer to their child. In other

words, the parent's use of non-directive techniques promotes

a secure relationship, and provides a context for change in

the parent-child relationship. This type of parent-child

relationship facilitates more effective parent management of

the child's behavior, and provides a supportive, nurturant,

and safe relationship within which the child can become more

socially competent and self-managed. At post-treatment

assessment, then, PDI may have a different meaning and demand

characteristics for clinic parents: PDI may come to mean

alternating between leading and following as opposed to

feeling that they always have to be in the lead. It is

likely that at post-treatment clinic-referred mothers focus

relatively more on the following than the leading component

of PDI in guiding their children's interactions. The

prediction that clinic children would follow more at post-

treatment PDI compared to pre-treatment assumed that the

mother would lead more at post-treatment and would be more

effective in her attempts. In fact, the clinic mothers may

have actually led less at post-treatment than at pre-

treatment for the reasons outlined above, resulting in less

dependent and more self-managed child behavior.

In any event, PCIT appears to have the effect of making

the clinic children's behavior less dependent on their

mothers' antecedent behavior, suggesting that the children's

behavior becomes more self-managed as a result of treatment.

This interpretation is buttressed by the significant between-

group differences in ECBI Intensity, _(22) = 12.08, 2 <

.0001, and Problem scores, t(22) = 12.33, < .0001, at pre-

treatment, but the groups did not differ significantly at

post-treatment on either ECBI Intensity scores, &(22) = -

1.77, ns, or ECBI Problem scores, &(22) = -.077, ns.

The DCS was able to discriminate between conduct problem

children and comparison group children in CDI at the first

assessment interval on the measure of child following, and

the DCS was sensitive to treatment changes in child following

for the clinic children in the CDI situation, despite the

high variability of the DCS and a relatively small sample

size. The fact that the two groups of children differed on

the DCS measure of following at the first assessment interval

in CDI, but not at the second CDI assessment interval,

suggests that patterns of leading and following are

indicative of the functional status of the mother-child dyad,

and that PCIT may function to restructure the dyadic patterns

of leading and following in CDI. The support for the

discriminant validity of the DCS provided by this study also

indicates that the DCS has the potential to improve clinic-

nonclinic classifications, and to provide a measure of

treatment outcome for the CDI component.
Special Considerations in PCIT Interaction Research

Conducting research, especially interaction research,

with the PCIT population requires attending to the special

needs of the population as it interacts within the quasi-

naturalistic clinical setting. Analog interaction measures

require continuous monitoring and assessment of the mother-

child interactive behaviors and processes. This presents a

problem. In an effort to facilitate naturalistic and

spontaneous interaction, the clinical play situations place

few constraints on the mother's or child's behavior during

each situation. As a result, children often wander, explore,

lie down, or even run about the playroom. Capturing this

complex behavior on videotape with preschoolers is a

formidable task, which has not been encountered by

researchers who have analyzed dominance patterns in dyads

whose movement could be restricted during relatively

naturalistic interactions. Mother-infant interaction

researchers were able to code face-to-face interaction by

strapping the infant in an infant seat fixed to a table,

similar to a normal feeding situation. Gottman (1979) was

able to restrict the movement of the married couples that he

studied by seating them across from each other during

conversation or by having them communicate through a "talk

table." These methods do not seem appropriate for studying

the interactions between conduct problem children and their

mothers mainly because restricting the child's movement

during mother-child interactions compromises the

generalizability of the results to more naturalistic

settings. Coding mother-child interactions is also difficult

from videotapes in which only one camera was used because

there are many times when it is impossible to observe the

mother and child simultaneously. Advances in video

technology such as small remote controlled cameras, video

laser discs, and advances in observational methodology

including digital images and on-line computer coding systems

may attenuate some of these problems.

Overview of Parent-Child Interactions

Competent parent-child interactions occur when the child

is able to signal his status and needs effectively and to

respond effectively to parental interventions, and when the

parent is able to interpret the child's status and needs so

as to respond effectively with the right behavior at the

right time in the right intensity. The ability to coordinate

interpersonal behaviors presupposes being able to

discriminate deviant from prosocial behaviors, being

sensitive to the status and needs of the other person, being

able to control one's own behavior effectively, and, on the

parent's part being able to differentiate between responding

to the child's immediate behavior instead of unrelated and

unresolved distal interpersonal behavior.

Interactional processes are probably best viewed as

existing on a continuum extending from uninvolved (whereby

each member's behavior is not predictable from the other's)

to involved (whereby each member's behavior is predictably

interrelated). The level of involvement corresponds to how

well the dyad negotiates the difficult task of balancing

intrapersonal and interpersonal factors. Effective social

functioning requires managing both a self-regulatory

component and an interactive component (Patterson, 1982;

Thomas & Martin, 1976). Walcher and Peters (1971) suggested

that play activities help develop intrapersonal and

interpersonal coordination through trial and error

experience. The functional status of iiany interaction is

related to the confluence and interplay of the self-

regulatory component and the interactive component. Walcher

and Peters (1976) state that:

Regulatory systems are found on all levels of the
organism's functioning ... Each self-regulatory
system presupposes the existence of several types crf
feedback resulting from the interaction of the child and
the environment and from changes in the environment.
One type of feedback results from the coordination of
the individual's general activities. The other results
from the progressive differentiation between the

individual's activities and the objects in his
environment (pp. 8-9).

According to this view, PCIT may help the parent to

manage his/her own behavior which has the effect of not only

increasing the parent's ability to manage the child, but also

has the effect of helping the child to more effectively self-

manage his/her own behavior. The importance of the ability

of the parent and child to engage in both coordinated and

independent activities cannot be overstated. Gottman (1979)

cited family systems and dyadic interaction research which

reported that functional and dysfunctional interactions could

be distinguished by the degree of predictability in the

interactions. Functional systems are distinguished by their

ability to respond in a flexible and creative manner in the

face of changing circumstances (von Bertalanffy, 1968), and

are able to alternate between coordinated and independent

activities. Dysfunctional systems, by contrast, are often

rigidly patterned, may have difficulties alternating between

coordinated and independent activities, and demonstrate

difficulties differentiating from each other which allows

disturbances in one component of the system to spill over

into other areas of the system. For instance, Wahler and

Dumas (1989) have found evidence that negative parent-child

interactions are predicted from the parent's prior negative

interactions with intimates (e.g., mother, husband) and

people in the community. The inability of the parent of a

child with conduct problem behaviors to differentiate his/her

feelings left over from prior negative interactions from

his/her feelings and perceptions about the child in the

immediate interaction suggests that self-management deficits

create or exacerbate unsatisfying or coercive parent-child


The results of this study suggest that PCIT may promote

the development of more functional parent and child self-

management patterns as well as more functional mother-child

interaction patterns. Perhaps, the most important result of

this study is that it provides empirical support for the

previously untested assumption that child following is an

important component of PCIT.

Dominance is thought to be essential to effective

communication. Dominance means being able to use the right

behavior pattern(s) at the right time in the right intensity.

In other words it is a matching phenomena. Dominance is

important to PCIT therapists given the fundamental role of

parent-child communication in establishing, maintaining, and

terminating functional parent-child relationships. The

interrelations between the dominance dimensions of type,

timing, and intensity need to be assessed systematically in

any complete account of the parent-child interaction process.

PCIT clinicians cannot be content with well-worn

metaphors such as "harmony", warmth, "being in synch", etc.,

as descriptions of parent-child interactions in general, or

parent-child dominance in particular. Moreover, PCIT

researchers need valid and reliable measures of the process

aspects of the parent-child relationship. Specifically, PCIT

clinicians and researchers need to know the behaviors and

interactional patterns involved in parent-child dominance,

how to change dysfunctional patterns of dominance, and how

dominance relates to parent-child and personal adjustment.

Other natural sciences have advanced only after the

requisite descriptive groundwork has been laid (Braitenberg,

1984; Glass, Wilson, & Gottman, 1975; Gottman, 1976; Gottman,

1979, Kantor, 1959; Ray & Delprato, 1988) emphasizing the

fact that "direct observation is the methodological

cornerstone of the natural sciences" (Hartup, 1979, p. 11).

Working from descriptive data to formal model building is the

method of the natural sciences. Gottman (1979) has suggested

that theorizing should mean explaining patterns in well-

described phenomena and that "the behavioral sciences have

suffered from the neglect of the descriptive phase of

scientific investigation" (p. 292). He issues a call for

detailed descriptive research in the study of social

interaction and suggests that investigation proceed in four

phases: description and discovery of pattern, parsimonious

model building, predictive testing of the model and

intervention studies, and understanding the model (theory

building). Descriptive research on parent-child interactions

is still in the early stages of this investigation sequence.

The present study provided some of the initial groundwork in

this process.



Marc K. Wruble

Version 3
This coding system is designed to provide information on
the patterning of mother-chid interactions in two relatively
unstructured clinical play situations. This coding system is
a descendent of a more comprehensive coding system (Wruble,
Sorensen, Sheeber, & Eyberg, 1987) designed to detail the
structural, functional, and operational characteristics of
mother-child interactions during three standard clinical play
situations varying in the amount of parental control
required. The present system attempts to capture the overall
quality (positive, negative, neutral) and patterning of
mother-child interactions.
Coders are instructed to perform three general tasks.
First, they are asked to capture the time of each verbal or
nonverbal behavior. Verbal behaviors are coded at the end of
each verbalization with the exception of 'duration relevant'
verbalizations such as crying, yelling, play talk, and
singing which are coded for the entire time they occur.
Nonverbal behaviors are coded at the beginning of the change
to a different nonverbal behavior and or the referent for a
behavior. This process is detailed in the nonverbal section.

Subject ID:
M = mother
C = Child

General Rules for coding nonverbal behaviors.

Both domains are coded at the beginning (time=000) of
each condition. However, after that initial coding, a domain
is only coded when there is a change in at least one domain.
There is a 1 second resolution; a behavior or event must last
for 1 second to be coded, except for hitting, kicking,

grabbing, kissing, head nod or head shake, take, give,
rearrange/block, and tear up. A state change is coded for
the activity domain at the time when the person touches an
object (i.e., transitions are considered part of the previous
activity). The referent of an activity is coded under two
conditions; when there is a change in the activity referent
domain itself (e.g., a change from playing with blocks to
playing with the castle figures), or when there is a change
in the activity and the referent also changes (e.g., from
playing with blocks to watching mother).

If person is engaged in two or more play activities
concurrently, then code the activity which is the most active
(e.g., if the child is holding a toy in one hand and
manipulating another toy with the other hand, then code

Only code activities that are relevant to play. For
instance, moving furniture would not be coded.

When there is a question about when a nonverbal activity
changes (e.g., when a child picks up a toy slowly and then
builds) code at the start of the activity (i.e., at the
picking up). This is consistent with the way that 'Giving'
is coded.

Fidgeting is defined as any repetitive/stereotyped
movements with the hand(s) that occurs when the person is
attending to something other than their hands or the objects
in them. Therefore, fidgeting is not coded as play activity.

Manipulating is defined as attending to and moving one
or more objects in a dynamic, i.e., non-stereotyped manner.

Non-verbal behaviors are coded as either positive, negative,
or neutral.

9 = unidentifiable If in any domain the coder cannot see the
subject to identify the proper code, use 9. However, that
domain must be coded the second that it is identifiable.

Neutral non-verbal behaviors are coded "0" and include:

HD = holding Supporting the weight of an object,
most frequently a person, using the
hand(s) or arm(s). Holding does not
include hugging, grasping, or
restraining, and it is not coded
when the activity is part of play.

BU = building

Manipulation of construction toys
(e.g., blocks, straws, legos).

HN = head nod (yes)

HS = head shake (no)

PL = pulling/pushing

WT = watching

PS = pointing/showing/

DW = drawing

Building includes taking the toys
from containers as well as the
construction of some thing.

Only code when the person is not
actively engaged in another activity
at the same time. Therefore, only
code during watching.

Only code when the person is not
actively engaged in another activity
at the same time. Therefore, only
code during watching.

Moving an object or person towards
or away from the person who is
pulling/pushing. Do not code
pulling/pushing as part of play.
Grabbing is coded when the grasping
is sudden and/or forcible.

When a person is engaged in no other
activity than looking at an object
or person. If the face cannot be
seen, code 9 as referent (except
when the person is scanning, then
code other as referent). If it is
unclear whether the person is
watching a object or a person
manipulating the object, code the
person as referent.

Enabling another person to see
gesturing or directing attention
towards an object, direction, or
place (for example pointing with a
finger or holding an object for
someone to see). Do not code
holding as part of this activity.
Also includes gesturing (using the
hands to indicate a quantity or
amount and waving) when there is no
other concurrent activity. The
referent for gesturing is always
coded 9.
Additional information:
Point/Show/Gestures can be
continuous or discrete; i.e., it
does not have to last for one

Application of a writing instrument
on a surface. Drawing includes
scribbling, writing, or drawing a

CU = clean up

BG = Board Games

picture. It also includes picking
up and moving the instruments of
writing, paper and pencils for

To take something apart by gradual
means. Includes placing materials
into containers (i.e., cleaning up).
Only code when the participants are
obviously ending play with that toy
and they are putting them away, or
when one or more of the participants
are picking up a play object from
the floor.

Any manipulation of the board game
itself, it's pieces, or it's
container within play. Board Game
must last for one-second.

Decision Rules for Board Games:
1.) If person is resting hand on board and
alternatively pushing or otherwise manipulating
objects then code as board game.
2.) If a person is contacting surface code Board Game,
whereas if not code change to either Point Show or
3.) If unsure whether person has changed activities do
not code as a change in activity.
4.) With alphabet teacher code Point Show if pointing
above game; if teaching board/manipulating code
Board Game.

FP = fantasy play

ER = erasing

RB = rearrange

Imitating human and/or animal
behaviors, typically using dolls or
figurines. This code includes
picking up and moving objects during
play and when arranging the toys.

Rubbing or crossing out something
tangible (e.g. pencil or chalk
writing). Crossing out includes
scratching out or putting an X
through something.

This includes rearranging things on
the other person's construction or
clean up, and rearranging the other
person's addition to the actor's
construction or clean up. This also
includes erasing the other person's
drawing (if whole drawing is
destroyed, code tear-up/break.)
All rearranges must occur within 2

seconds of other person making the

Decision Rules:
1.) Transitions from rearrange to building is not coded
as a change.

Negative non-verbal behaviors are coded "-" and include:

Whenever a negative nonverbal behavior occurs code
whether the other person responds to or ignores the nonverbal

Responds to negative = consists of any verbal or nonverbal
reaction by the other person following smart talk, whine,
cry, or yell. Examples include laughs, frowns, smiles,
grimaces, winces, rolls eyes, sighs, maintains or establishes
eye contact, warns other person, criticism, taking object
from other person, asks for rationale, etc (see p. 79 of
DPICS manual for additional examples).

Ignores negative = the other person remains silent, maintains
a neutral facial expression, avoids or breaks eye contact,
makes no movement in response except to turn away (see p. 81
of DPICS manual for examples).

BK = Blocking

Includes blocking the other
person from playing or
cleaning up.

Decision Rules:
1.) Transitions from blocking to playing are coded as

FG = flailing (of arms
and/or legs)

HT = hitting

Repeated gross arm/leg
a flailing manner without
making contact with another
object. This can occur in any

Applying undue force towards
an object or person using the
hand with short duration (<1
sec) without grasping. This
does not include rubbing,
patting, or kicking. It does
include slapping, punching,
pounding, and striking. Does
not require contact with the
person or object and may occur
in rapid succession.

RN = restraining

TH = throwing

GR = grabbing

KK = kicking

TU = tear up/break

TK = take

Actively restricting the
movement of any or all parts
of the other person's body
with that person actively
resisting the restraint.
Holding is not coded as part
of restraining.

Projecting or propelling an
object through the air by a
swift motion of the arm(s).
Do not code as part of play.
If object is thrown at other
person, code as hitting other

Actively taking hold
of/seizing an object suddenly
and/or forcibly.
Pulling/pushing is not coded
as part of grabbing. Includes
taking hold of an object that
is part of the other persons
construction or that is
presently in use by the other
person. Note: grabbing is
coded even if object is not
actually obtained. If a
subject knocks down or tears
up an object while grabbing,
only code grab.

Applying undue force towards
an object or person using the
feet or knees.

To separate an object into
parts by ripping, tearing, or
breaking, or to crush an
object by stepping on it or
smashing it between the hands.
Also include knocking down

To seize an object from the
other person's possession.

Decision Rule:
1.) If unsure whether to code 'Take' or not, do not
code 'take'.

Only coded when there is (1) a
nonverbal gesture with one or
more of the hands indicating


'No' or 'Stop', and (2)
verbalizations indicating
'No', 'Stop', or 'Don't'.
This gesture often consists of
waving the hand(s) in a side-
to-side motion with palm(s) of
hand facing toward the person.

Positive nonverbal behaviors are coded "+" and include:

CA = caressing

CP = clapping

KS = kissing

HG = hugging

TC = touching

GV = giving

Gentile rhythmic stroking or
patting of an object with any
part of the body. This also
includes hair ruffling and

to strike the hands together
repeatedly usually to applaud
either another person or

Touching an object with the
lips or a puckering gesture of
the lips towards an object.

One or both arms folded around
object with torsos at least
nearly in contact.

Physical contact of hand with
other person. Touching does
not include hitting, grasping,
patting, or any activity that
includes the contact of the
hand with anything other than
a person.

Placing an object in the hand
of, or in front of the other
person (when verbalizations
indicate the other person is
commanded to make use of the
object). The actor must
physically move or hold the
object (else code
point/show/gesture.) When it
is unclear whether GV or PS,
code PS.

Object of Activity (Referent)

CB = cabinet
FL = floor
CH = chair
PT = parent
TB = table
OA = other
CK = chalkboard
DO = door
MI = mirror
CI = child
AI = air when action is done towards someone
TI-T8 = toys:
Ti = Lincoln Logs
T2 = Alphabet Teacher
T3 = Magnetic Letters & Board
T4 = Baby Blocks
T5 = Magneto Blocks
T6 = Tinker Toys
T7 = Legos
T8 = Constructo Straws
9 = unknown, unclear, ambiguous object, or obstructed from

Decision Rules:
1.) If two people have possession of an object
simultaneously then code the referent of the
activity as the other person.
2.) When a person changes from one toy (e.g., castle)
to another toy (e.g., blocks) then incorporates
this 'new' toy into the 'old' toy code as follows:
(a) code as change from toy 1 to toy 2, and
(b) only code change from playing with incorporated
toy 'toy 1' to next change in referent (e.g.,
watching). That is, only code two changes.


Three rules for coding separate verbalizations. These
rules are meant to be used in combination to determine the
number of distinct statements (i.e., thought units) present
in any verbalization.

Separated by > 1 sec.?
a). If so, then likely code as two separate statements.
b). If not, then likely code as one statement.
Is statement a complete thought?
a). If so, then likely code as one statement.
b). If not, then likely code as two or more statements.
What would statement look like grammatically?
a). Punctuation typically indicates separate thoughts.

BASIC RULE FOR VERBALIZATIONS: Code the literal meaning of
the statement.

Other important Rules:
Acknowledgements occurring at the beginning or end of a
sentence are coded as part of the sentence. That is, they
are not considered to be a separate thought apart from the
sentence. For instance, '"O.K.., Now put the blue block on
top of the red one.' would be coded as one complete thought,
not two.

There are two types of verbalizations:
a). Discrete consist of distinguishable sentences,
phrases, or utterances. The times for these verbalizations
are coded at the end of the verbalization.
b). Duration Relevant consist of utterances which may or
may not have content. These statements include yell, cry,
play talk, laughing, singing, etc. The times for these types
of verbalizations are coded at the beginning and end of the

Neutral verbalizations are coded "0" and include:

DC = direct command A clearly stated order,
demand, or direction
(i.e. telling the person
what to do) in
declarative form. The
statement must be
sufficiently specific as
to indicate the behavior that
is expected from the other
person (can include a vague
command with nonverbal

IC = indirect command

IV = irrelevant

NW = No Words

AK = acknowledgement

IS = incomplete sentence

Put that block here.
Please tie your shoe.

An order, demand, or direction
(i.e. telling the person what
to do) for a behavioral
response that is nonspecific
or implies choice on the part
of the other person.

A comment or question made by
the parent (all child
verbalizations are relevant)
that pertains to an event,
verbalization, individual, or
object that is unrelated to
the ongoing activity of the
parent or child. Continued
discussion or elaboration on a
topic brought up by the other
person is not coded irrelevant

Used in circumstances where
there is not manifest content
but there are vocalizations,
e.g., laughing. This is
differentiated from '9' codes
by the fact that the "no
words" code is used for
instances in which there is no
manifest content whereas the
'9' code is used in instances
where there is content but it
is not discernible.

A brief verbal response to the
other person's verbalization
or behavior that contains no
manifest content other than a
simple yes or no response to a

There you go.

Whenever a subject makes an
incomplete statement or is
interrupted before completion.
Specifically, when a person
interrupts themselves or