PCRI manual cover_070910_Layout 1 7/9/10 3:35 PM Page 1
Parent-Child
Relat ionship Inventory
( PCRI )
Manual
Anthony B. Gerard, Ph.D.
Additional copies of this manual (W-293B) may be purchased from WPS.
Please contact us at 800-648-8857, Fax 310-478-7838 or our Web site.
W-293B
01-02_v4_ 1-2.qxd 6/28/10 2:24 PM Page 1
1
INTRODUCTION
The Parent-Child Relationship Inventory (PCRI)
assesses parents’ attitudes toward parenting and toward their
children. The PCRI yields a quantified description of the
parent-child relationship that complements other assessment
procedures used in clinical evaluations of children and
families. Rather than replacing qualitative evaluation of
parent-child interactions, the PCRI helps to put qualitative
impressions in perspective by making normative comparisons possible.
Standardized on more than 1,100 parents across the
United States, the PCRI identifies specific aspects of the
parent-child relationship that may cause problems, as well
as giving an overall picture of the quality of the relationship.
In an era when fathers are increasingly expected to take an
active role in parenting, the PCRI explicitly measures the
attitudes and behaviors of both mothers and fathers. It is
assumed that the PCRI will often be administered to couples, and there are separate norms for mothers and fathers.
Recent increases in child custody litigation and
divorce mediation have created a demand for sophisticated
assessment of the relationships between parents and children, and the PCRI may prove especially useful in child custody settings and in other institutions that specifically
address the needs of children. Used in conjunction with
interviews and other forms of clinical assessment, the PCRI
can be an important element in the making of custody
recommendations and in evaluating the possibility that a
parent is abusive.
reflect major features of parenting and the parent-child relationship. Consistent with the idea that parenting skills define a positive dimension, high scores on the PCRI scales
indicate good parenting skills and low scores indicate poor
parenting skills.
PCRI Scales and Validity Indicators
The PCRI has seven content scales and two validity
indicators. Each of the content scales explores a specific
aspect of the parent-child relationship. These scales were
developed using a combination of empirical and rational
approaches, as described in the section of chapter 4 entitled
“Item Analysis and Selection.” One of the two validity indicators gauges the client’s tendency to give socially desirable
responses. The other validity indicator, which is based on
agreement between answers on select pairs of items, measures
the tendency to give inconsistent responses. A listing of the
items in the content scales and the Social Desirability scale
appears in Appendix A.
Content scales. Of the 73 items included in the content scales, 26 are keyed positively and 47 are keyed negatively. If an item is positively keyed, a response of agree or
strongly agree increases the score for the scale on which that
item appears; conversely, if an item is negatively keyed, a
response of disagree or strongly disagree increases the scale
score. High scores indicate positive parenting characteristics.
The Parental Support scale (SUP), which has 9 items,
assesses the level of emotional and social support a parent
receives.
The Satisfaction With Parenting scale (SAT) consists
of 10 items measuring the amount of pleasure and fulfillment an individual derives from being a parent.
The 14-item Involvement scale (INV) examines the
level of a parent’s interaction with and knowledge of his or
her child.
The Communication scale (COM) consists of 9
positively keyed items that assess a parent’s perception of
how effectively he or she communicates with a child.
The Limit Setting scale (LIM) contains 12 items, all
negatively keyed, that focus on a parent’s experience disciplining a child.
The 10-item Autonomy scale (AUT) assesses the
ability of a parent to promote a child’s independence.
General Description
The PCRI is a 78-item, self-report questionnaire that
can be administered to either an individual or a group in
about 15 minutes. It has a fourth-grade reading level.
The items were selected to measure a wide range of
parenting dispositions and behaviors. Some of the items
present general attitudes toward being a parent, and others
are intended to elicit responses specific to a parent’s relationship with a particular child. All of the items have a
Likert-type, 4-point response format: strongly agree, agree,
disagree, and strongly disagree. Rather than providing a single score representing an individual’s overall ability in and
satisfaction with parenting, items are arranged in scales that
1
01-02_v4_ 1-2.qxd 6/28/10 2:24 PM Page 2
2
Parent-Child Relationship Inventory (PCRI)
The Role Orientation scale (ROL), comprising 9
items, examines parents’ attitudes about gender roles in
parenting.
Protocol validity. The PCRI has two validity indicators: Social Desirability (SOC) and Inconsistency (INC).
The Social Desirability indicator consists of five items
that are rarely endorsed in the positive direction. A low SOC
score suggests that the parent is giving distorted responses
intended to portray the parent-child relationship in an
unrealistically positive light.
The Inconsistency indicator comprises 10 pairs of
highly correlated items. The response to one item in the pair
in effect predicts the response to the other item. High scores
on this indicator suggest inattentive or random responding.
Applications
The PCRI is intended for use in a wide range of contexts, including both clinical and research settings. Because
it is multidimensional, the PCRI identifies specific areas of
difficulty between parents and children. By quantifying
aspects of parent-child interactions, the instrument makes it
possible to verify clinical hypotheses about individual and
family disturbances against a background of objective data.
Limitations
Assessment of parent-child relationships requires clinical sensitivity and a thorough knowledge of the research on
parent-child interaction. Moreover, relationships between
parents and their children do not exist in a vacuum. They are
embedded in a matrix of family, cultural, and socioeconomic factors, all of which influence parental attitudes and
behaviors. Designed for use by individuals with a background in psychological assessment, the PCRI is intended
to be one measure of the characteristics of the parent-child
relationship. Given the complexity of parent-child interactions, the inventory must never be used in isolation as the
sole basis for clinical diagnoses, treatment decisions, or
custody recommendations.
08-19_v4_ 8-19.qxd 6/28/10 2:26 PM Page 17
4
DEVELOPMENT
Interest among mental health professionals in the
assessment and treatment of dysfunctional parent-child relationships has grown dramatically in recent years. This
increased concern is not surprising. Almost daily, newspapers
and television news programs report repeated incidents of
physical or sexual child abuse. For each such highly publicized case, many more are dealt with privately or go unreported. Improved assessment methods are needed to support
research into the causes of child abuse, and to identify parents
and children who might benefit from specific interventions.
Recent increases in child custody litigation and the
greater involvement of mental health professionals in voluntary or court-mandated mediation have created a specific
demand for more sophisticated procedures to assess parentchild relationships. In their work within the legal system,
psychologists and other professionals are often called upon
to consult on decisions that can have enormous impact on
people’s lives. They may, for example, be asked to make
recommendations about custody, foster care, visitation
rights, and parental competence. In the absence of standardized tests for assessing parental attitudes and behaviors,
those charged with making these decisions must rely solely
on their clinical skills or adapt existing instruments to
purposes for which they were not intended.
At present, there are no generally agreed upon
standards for evaluating parenting skills. To begin with, the
subtle nuances of the parent-child relationship are difficult
to measure. Many factors influence the quality of the relationships within a family, and it is notoriously difficult to
identify the sources of family dysfunction. For example, it is
widely recognized that the causes of child abuse are multiple
and interactive (Spinetta & Rigler, 1972; Young, 1976).
One measure that has been used to assess parental attitudes is the Mother-Child Relationship Evaluation (MCRE).
Originally published by Roth in 1961, the MCRE is a brief,
self-report instrument designed to assess both normal and
problematic aspects of parenting. Based on a set of constructs initially presented by Symonds (1949), the MCRE
contains four subscales of 12 items each: Overprotection,
Overindulgence, Rejection, and Acceptance. Mothers are
asked to rate their strength of agreement or disagreement
with each item on a 5-point Likert scale. Clinical interpretation of the MCRE is based on individual item responses, the
clinical scale scores, the overall profile pattern, and the
integration of the test data with information from other
sources, especially clinical interviews.
Despite considerable interest in the MCRE, use of
this instrument to address the major issues outlined above
has been constrained by two major factors. First, the original standardization of the MCRE was based on a small
sample of volunteer mothers in the Midwest. The extent to
which the normative data generalizes appropriately to
other groups is unknown. Second, and perhaps more
importantly, the MCRE was originally conceptualized as a
measure of maternal attitudes toward child rearing; both
the present normative data as well as many of the individual
items do not apply to fathers. The increasing recognition of
the father’s role in child rearing almost automatically calls
into question the usefulness of a parent attitudes questionnaire designed for mothers only (Cath & Ross, 1982;
Lamb, 1981).
Furthermore, there is a need for new parenting
questionnaires that are psychometrically stronger and more
current than those of the past (Holden & Edwards, 1989).
Many of the instruments developed for circumscribed
purposes a generation ago are still in general use. In some
instances, instruments designed essentially for laboratory
use have been used as true assessment tools, with little or no
normative data and limited validation. Because attitudes
toward marriage and child-rearing have changed markedly
in the past two decades, the content of these questionnaires
is often out of date. Although the need for good tools to
assess parent-child relationships is perhaps greater than
ever, few instruments actually meet present standards.
In light of the need to redesign the MCRE and in
response to the call for new instruments of this type, a
completely new inventory for assessing parental attitudes
was developed. The development project had the following
specific goals:
1. To create an objective measure of parental
attitudes that would be useful for both clinical
and research applications
2. To develop items that are equally appropriate
for both males and females, and that are not
age dependent
3. To use items at a reading level that would
make the inventory useful with low SES
populations
4. To meet contemporary psychometric standards
5. To determine whether multiple scales would
facilitate accurate interpretation
17
08-19_v4_ 8-19.qxd 6/28/10 2:26 PM Page 18
18
Parent-Child Relationship Inventory (PCRI)
Scale Construction
The first step in scale construction was to determine
the dimensions of parenting on which to generate items.
Whereas numerous, more specific content dimensions have
been identified, factor analytic studies consistently extract
three superordinate dimensions of parenting attitudes: an
affective dimension (“good/bad parenting”); a dimension
centering around control, authority, and limit setting; and a
third dimension that varies more across studies but that
typically reflects either overprotection or punishment
(Whitman & Zachary, 1986).
Factor analysis of data collected on a preliminary,
106-item version of the PCRI (Form A) identified five major
dimensions of parenting, and the items were organized into
scales based on these factors. New factor analyses were run
on each of these scales. The results suggested the organization of the PCRI into 5 factor scales and 14 clinical scales.
A set of 24 items for each scale, balanced for strength
and the direction of the response, was written for each of
the 14 clinical scales, and a 9-item social desirability scale
was added. The result was a 345-item version of the PCRI
(Form B). Because this version contains many more items
than is desirable for most practical applications, further item
selection studies were required.
This research took three forms: (a) rating of items by
expert judges; (b) qualitative feedback from professionals
and test takers to identify objectionable items; and
(c) collection of additional empirical data and subsequent
item analysis.
Expert Rating Study
The 345-item version of the PCRI was rated by 11
judges chosen to represent a diversity of opinion. This group
included item-writing experts, clinicians and school
psychologists in active practice, a nationally known figure
on child abuse, and a minority psychologist interested
in assessment.
For the ratings, the individual items were typed onto
3 × 5 index cards and presented in packs corresponding to
scales. The actual procedure was a Q sort (Nunnally,
1978). To insure useful response variance, the judges were
required to sort the items into five piles arranged in a triangular pattern approximating a normal distribution. Items
were to be put in Pile 1 if they were extremely problematic
or bore little relation to the intended content domain;
items were to be put in Pile 5 if they were superior and
should be retained; Piles 2, 3, and 4 represented the middle
ground.
The average rating for each item across all 11 judges
was computed, as was the rank order of the item within its
particular scale. Because there were 24 items on each of the
14 content scales, the rank orderings ranged from 1 to 24
(where a ranking of 1 is highest); when there were ties, all
items having the same mean rating were given the same
rank. These rank orderings were used later as part of the
empirical criteria for item selection.
Subjective Feedback
Important information on the wording and interpretation of each item came from individuals—some professionals,
some parents, some both professionals and parents—who
carefully examined the 345-item Form B and commented on
all objectionable items. Information about the appropriateness
of items also came unsolicited from many individuals who
completed the inventory after having been told that it was an
experimental instrument not suitable for use as a basis for
important life decisions. Other individuals provided feedback by agreeing to “think aloud” as they went through each
item, so that salient comments could be recorded by another
individual. Observations about item clarity and acceptability
gathered in these ways were used only at the end of the item
selection process to refine the wording of individual items.
Item Analysis and Selection
The final selection of items was based on a combination of empirical and rational criteria, following a systematic
procedure originally presented by Jackson (1970). Much of
the data used in this step came from administering the
345-item Form B to 211 parents living in and around
St. Louis, Missouri.
Step 1. The first step was to eliminate high- and lowfrequency items—items everyone endorsed or no one
endorsed. Such items contribute little unique information,
and 91 items with extreme probabilities of endorsement or
rejection (above .90 or below .10) were discarded in this way.
Step 2. The rank order of the items based on the
ratings by the expert judges was examined for each scale,
and only items ranked 1 through 9 (out of 24) were retained.
These were all items that the expert judges perceived as
well-written and meaningfully related to the content
domains specified by the scale names. This procedure
eliminated an additional 61 items.
Step 3. An item that did not correlate significantly
(>.30) or uniquely with the total for its assigned scale was
examined. Although the majority of such items were subsequently eliminated, attention was paid to the content of
the item. If the shift of an item to a new scale was defensible,
it was usually shifted. In part for structural reasons
connected with the original development of the scales
and in part to preserve face validity, an item that was
more highly correlated with another scale than with its
own was sometimes retained on its original scale. An
item that had a low correlation with its own scale, however, and did not fit logically or statistically with another
scale was eliminated.
Systematic empirical criteria for item selection were
used, but they were not followed rigidly. The content of each
item and its relationship to the other items in the scale were
considered most important. The balancing of the best possible psychometric properties against a subjective sense of
which items logically belonged to which scale required
more than 20 iterations.
Step 4. Following Jackson (1970), a Differential
Reliability Index (DRI) was computed for each item. This
08-19_v4_ 8-19.qxd 6/28/10 2:26 PM Page 19
Chapter 4
Development
procedure compares the corrected item-scale correlation to
the magnitude of the correlation between the item and the
Social Desirability (SOC) scale. Because the SOC scale had
an alpha coefficient of .76, it was not ideal for this purpose
but did provide a systematic method for eliminating items
that were heavily biased with respect to social desirability.
Items with a DRI of less than .30 were usually eliminated.
Step 5. A major concern throughout the scale construction and item analysis process was that the final scale
be perceived as logical and useful by clinicians, school psychologists, and others who would use the scale for various
applications. Achieving this goal implied a balance between
the ideals of empirical test construction and the demands of
clinical usefulness. The final step in the initial development
procedure was to give the revised scales to a new set of five
expert judges. Rather than assigning each item a value from
1 to 5, this panel of judges made dichotomous decisions
about whether a particular item was good or bad and
whether it belonged on a particular scale. Based on this
input, items were eliminated because the judges found them
objectionable or because they could not be assigned sensibly
to a particular scale. Although a few problematic items identified at this stage were reassigned to other scales, most were
simply eliminated.
With these adjustments, the PCRI consisted of 107
items distributed over two validity scales and eight content
scales. This version of the instrument was used over a period
of approximately 4 years to conduct a variety of validity
studies. The 107-item form was also used to collect the
standardization data.
Step 6. Following the collection of the standardization
data an additional item selection process was initiated,
similar to that described in Step 3. Scale alpha levels and
item-total correlations were obtained using approximately
half of the normative sample (n = 519). Items that were not
contributing to scale reliabilities were eliminated. In a few
cases items were shifted from one scale to another, based on
a combination of empirical and rational criteria. One scale,
Moderation, was eliminated, and two of its items were
incorporated into the Autonomy scale. The results of this
procedure were cross-validated using the remaining half of
the normative sample (n = 518). The scale reliabilities
obtained using both samples are presented in Table 1. As the
values in the table show, shrinkage in the scale reliabilities was
minimal, supporting the validity of the new scale structure.
The result of these additional adjustments was a
78-item instrument with a fourth-grade Flesch-Kincaid
reading level (Thomas, Hartley, & Kincaid, 1975).
Development of Validity Indicators
The two response validity indicators for the PCRI
were generated in different ways. The Social Desirability
scale was developed at the same time as the content scales.
The Inconsistency indicator was developed using the results
of the standardization study. The cut points for both scales
were established using standardization results.
19
Table 1
Scale Alphas for Development and
Cross-Validation Samples
Scale
Development
Cross-Validation
SUP
SAT
INV
COM
LIM
AUT
ROL
.70
.83
.77
.81
.88
.80
.77
.71
.85
.77
.82
.87
.78
.75
Note. The samples for Development (n = 519) and Cross-Validation
(n = 518) were derived from the normative sample.
Social Desirability
The purpose of a social desirability scale is to identify
clients who are responding with a so-called “defensive
response set.” On a test like the PCRI, individuals may have
many reasons to “fake good.” For example, a client may be
involved in a custody dispute, which places a heavy premium
on appearing to be an excellent parent. Even when less is at
stake, many individuals tend to present themselves in a
favorable light. Signs that a client is “faking good” introduce a note of caution into the interpretation of his or her
scores, and may be sufficient reason to treat the results as
entirely invalid.
The present five-item Social Desirability scale is the
result of the same item selection procedure described in the
section of this chapter entitled “Scale Construction.” All of
the items were written to reflect the exaggeration of positive
qualities characteristic of a defensive response style (faking
good). The mean score on the SOC scale was 14.52, and the
standard deviation was 2.63. The items on this scale and the
endorsement percentage for each are presented in Table 2.
The cutting score of 9 on the SOC scale is based on
simple inspection of the responses of those in the standardization sample. Those with scores of 9 or lower constitute
5% of the sample. To use this proportion to identify those
who may be faking good appears to be a sufficiently selective screening.
Inconsistency
The purpose of the Inconsistency indicator is to
identify protocols that result from inattentive responding.
Almost by definition, such a response set will produce
patterns that resemble random responding. Although
random responding is obviously undesirable, it is also
relatively easy to detect with some certainty. The chief problem in designing the INC indicator is to arrive at a criterion
for identifying a protocol as invalid.
The INC indicator is based on interitem contingencies. Ten pairs of highly correlated items were identified.
Each pair consists of items from the same scale; intrapair
correlations range from .53 to .74, and all of the correlations
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 20
20
Parent-Child Relationship Inventory (PCRI)
Table 2
Percent Endorsement of PCRI Social Desirability Scale Items
Item
Percent Strongly Agree
18. My child is never jealous of others.
30. I never worry about my child.
37. I have never had any problems with my child.
43. I have never been embarrassed by anything my child has said or done.
47. My child never puts off doing things that should be done right away.
6.4
1.6
4.5
8.1
1.7
Note. N = 1,093.
are positive. Scores equal the number of pairs in which the
items are endorsed in opposite directions, with the limit that
there must be a 2-point difference between the ratings of the
items in a pair. The item pairs comprised by the INC indicator are as follows: 17 and 36, 24 and 27, 55 and 67, 64 and
77, 53 and 60, 63 and 72, 39 and 46, 10 and 26, 49 and 78,
and 25 and 32. The mean score was .42, and the standard
deviation was .77. Inconsistent responses to two or more of
these pairs were rare in the normative sample, occurring in
fewer than 6% of the cases. Consequently, the validity of a
protocol is questioned if it contains inconsistent responses to
more than one of these pairs of items.
The validity of using a cut point of one pair on the
INC indicator was subjected to statistical test. It is possible
to compare the probability that a given score on the INC
indicator comes from a sample of good protocols (those that
result from attentive responding) with the probability that
the same score comes from a sample of bad protocols (those
resulting from random responding). Following logic devel-
oped by Cureton (1957), Dawes (1962), and Dawes and
Meehl (1966), and applied by Cull and Gill (1988), among
others, it is assumed that the distribution of “good” protocols numerically (or spatially) overlaps the distribution of
“bad” protocols. For each possible INC score (1 through 10)
a thin statistical “slice” is taken through both distributions,
permitting an estimate of the proportions of good protocols
and bad protocols that attain that score. The deeper into the
bad distribution, the higher the proportion of bad protocols a
given score represents.
Using this procedure, it is found that the exact probability that an INC score of 1 comes from a sample of bad
(random) protocols is 29.1%, but the probability that an INC
score of 2 comes from a sample of bad protocols is 84.4%.
For scores above 3, it is a virtual certainty that the protocol
is bad. The large classification increment between 1 and 2
justifies the establishment of 1 as the INC cut point. There is
a high likelihood that protocols with INC scores greater than
1 reflect inattentive or random responding.
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 21
5
STANDARDIZATION
This chapter describes the methods employed in deriving norms for the PCRI. The first section describes the data
collection effort, including the sample selection procedures.
Subsequent sections discuss the reduction and analysis of the
normative data, as well as the structure of the PCRI norms.
filled out. In some instances, however, the packets were
distributed and completed as part of a group administration
at a PTA meeting or other parents’ gathering.
Each site received detailed data collection instructions. The individual packets also contained detailed
instructions to the parents for taking the test and filling out
the accompanying demographic form. None of the sites
reported any difficulty in following the instructions or
administering the instrument.
Standardization Sample
The normative data for the PCRI represents the
responses of more than 1,100 mothers and fathers. Data
was collected through schools and day-care centers in each
of the four major geographical regions of the United States
(Northeast, South, Midwest, and West). The first of the
following sections describes the procedures for acquiring
standardization sites and collecting data; the second
section discusses the demographic characteristics of the
normative sample.
Sample Characteristics
The main sample demographics are given in Table 3.
Where it is appropriate, Table 3 also gives U.S. Census
percentages corresponding to sample figures. In general, the
sample was geographically diverse but weighted heavily
toward the middle of the socioeconomic spectrum. Sample
characteristics are described briefly in the following sections.
Age and sex. That the PCRI is a self-report of parents
about their relationship with their children slightly complicates the collection and description of the normative data.
The possibility that parents of children at different ages
respond differently to their children meant that it was necessary to make the age of the child a factor in the normative
study. In fact, age of child was presupposed to be a factor of
greater interest than age of parent. Consequently, data was
collected through institutions expected to have students as
young as 3 years of age and at institutions expected to have
students as old as 15 years of age. The sex of the child was
also expected to exert some influence on PCRI scores. For
that reason, sex of child was also a factor in the normative
study. The distribution across sex and ages of children and
parents is given in Table 3.
Responses of mothers and fathers. Parenting instruments have typically been based on the questionnaire
responses of mothers (Holden & Edwards, 1989). As discussed in the introductory section of chapter 1, a main motivation for developing the PCRI was to present normative
information on the attitudes of both parents toward children
and child-rearing. Although there are more mothers than
fathers in the sample, the sites participating in the standardization of the PCRI were, by and large, successful in collecting responses from fathers, as the figures in Table 3 suggest.
Data Collection
The collection sites for the PCRI normative data were
schools and day-care centers.
Letters soliciting participation in the study were sent
to 2,000 individuals identified as principals of elementary
schools, middle schools, and junior high schools or as directors of preschools and day-care centers. Those interested in
participating were to return a postage-paid card requesting
basic demographic information about the parents in their
institutions, and 88 such cards were returned. The light
return rate (4.4%) was anticipated, and the number of
returns was in excess of that needed to fulfill the requirements of the normative study.
Based on geographic distribution and on the demographics supplied on the return cards, 18 data collection
sites were selected. Each site received a specified number of
data collection packets, and each packet contained enough
materials so that both parents of a given child could participate in the study. In most cases it was the joint responsibility
of the parents’ organization at the school and a designated
member of the school administration to distribute the packets, see that they were properly completed, collect them, and
return them for analysis. Also in most cases, the packets
were sent home with students and returned after they were
21
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 22
22
Parent-Child Relationship Inventory (PCRI)
Table 3
Sample Demographics
Child’s Age
n
Sample %
<3
54
4.5
Child’s Sex
n
Sample %
Censusa
Male
595
50.2
51.2
3
56
4.7
Ethnic Backgroundc
n
Sample %
Censusd
5
181
15.2
6
145
12.2
7
121
10.2
8
89
7.5
9
129
10.8
10
105
8.8
11
68
5.7
12
43
3.6
13
42
3.5
>13
37
3.2
Female
588
49.7
48.8
Relationship to Child
n
Sample %
Parents’ Age
n
Sample %
Censusb
4
122
10.2
Mother
669
55.2
18–24
27
2.3
13.8
Father
474
39.1
25–34
389
32.8
22.9
Asian
22
1.9
2.9
Stepparent
33
2.7
35–44
657
55.4
19.1
Black
81
6.9
12.1
45–54
95
8.0
13.0
Hispanic
18
1.5
9.0
Native
American
12
1.0
0.8
Other
37
3.1
>55
17
1.4
27.5
White
1,018
85.7
80.3
Parents’
Education
Level
n
Sample %
Censuse
Less Than
High School
Graduation
52
4.5
13.5
High School
Graduation
381
32.6
40.0
Some
College
322
27.6
21.1
Four Years
of College
or More
413
35.4
25.4
Region
n
Sample %
Censusd
Northeast
338
28.7
20.4
South
491
41.7
34.4
Midwest
201
17.1
24.0
West
147
12.5
21.2
Other
35
3.0
—
Note. N = 1,192; sample size varies slightly across subtables.
a
Based on U.S. Census figures for children (U.S. Bureau of the Census, 1991).
b
Based on U.S. Census figures for those over 16 (U.S. Bureau of the Census, 1991).
c
Based on respondents’ reports of children’s ethnicity.
d
Based on U.S. Census figures (U.S. Bureau of the Census, 1991).
e
Based on U.S. Census figures for adults 25–44 (U.S. Bureau of the Census, 1991).
Socioeconomic status (SES) and ethnicity. The
figures in Table 3 also show that the PCRI standardization
sample is somewhat better educated and less diverse than
the U.S. population as a whole. For example, the median
number of years of education for the sample was 14, whereas
the median for U.S. residents between 25 and 44 (a span that
encompasses virtually all of the participants in the normative study) is approximately 13 (U.S. Bureau of the Census,
1991). The figures for parents’ occupational status, which
were also collected, closely parallel those for education;
approximately 54% of the sample participants were
employed in technical, managerial, or professional positions.
Because the bulk of the data comes from both parents’
ratings of a single child, the respondents’ identification of
their children’s ethnicity was used as the primary indicator
of ethnicity. The actual difference between respondents’
reports of their own ethnicity and their children’s was small.
As shown in Table 3, whites are overrepresented in the
normative sample.
Region. Of the 18 schools that agreed to participate,
13 returned data—4 in the South and 3 each in the
Northeast, Midwest, and West. In each region, one of the
institutions returning data was a preschool or day-care
center. Given that the South has a substantially larger population than do the other major geographic regions (U.S.
Bureau of the Census, 1991), the number of participants in
each region, presented in Table 3, is roughly proportional to
the distribution of population across regions.
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 23
Chapter 5
Standardization
Effects of Sample Characteristics
Because divergence of standardization sample characteristics from population characteristics could produce
anomalies in the PCRI norms, sample characteristics were
examined for their influence on PCRI scores. The effects of
sample characteristics were evaluated for their statistical
significance through the use of multivariate analysis of variance (MANOVA), and the statistical results were evaluated
for their practical importance through the use of standard
techniques for assessing the meaning of individual test
results. Simplifying assumptions were made in order to
reduce the complexity of the analysis, and criteria were
developed for assessing the relationship of statistically
significant group effects to the interpretation of individual
PCRI scores. The effects of purely demographic factors
(ethnicity, SES, and region) are discussed separately from
the effects of factors more directly related to the content of
the PCRI (age and sex of parent and child).
Evaluating Effects
Such an examination of the data set raises two
methodological issues. First, the number of variables makes
the number of potential comparisons so large that it would
be virtually impossible to determine the actual significance
of statistically rare results. Second, statistically significant
group differences may or may not reflect differences of
practical significance in evaluating individual PCRI results.
Efforts were made to address both of these problems and to
render results comparable across subscales.
Standard metric. To assist in evaluating differences
among groups on the seven content scales of the PCRI, all of
the raw data was transformed to T-scores based on the mean
and standard deviation (SD) of each scale for the entire standardization sample. The overall mean of the sample is set at
50 and the standard deviation is set at 10. Therefore, the
group mean for a scale is below 50 when the group has
endorsed fewer than the average number of items in the
scale-positive direction, and it is greater than 50 when the
group has endorsed greater than the average number of
items. This linear T-score transformation is not the T-score
transformation ultimately used in constructing the PCRI
norms. Instead, it is a preliminary centering that represents
all of the scale results in terms of a common, familiar metric.
The T-score transformation does not affect the presence and
size of effects in the data set.
Effect significance. Two features of the analysis
reduced its complexity and protected against mistakenly
identifying effects as significant (Type I error). First,
although the initial steps of the analysis included the assessment of interaction effects, no interaction terms were anticipated, permitting the analysis to focus on simple main
effects; the MANOVA models employed were designed to
establish the need to pursue more detailed analysis of interaction effects. Second, the real purpose of the analysis was
to uncover and explain any statistically significant effects
that might distort the norms. Consequently, the acceptance
23
of spuriously significant findings posed little danger, and it
was actually conservative not to control for the acceptance
of spurious effects through the use of a technique for adjusting the size of the statistical rejection region.
Effect size. It is necessary to place those statistically
significant results that do arise in their proper perspective.
Accordingly, the standard error of measurement (SEM) was
chosen as the criterion for assessing the practical importance
of statistically significant effects. The SEM, which corresponds to the 68% confidence interval, is frequently used in
test interpretation. It therefore provides a familiar benchmark for assessing the meaning of effects.
On each scale, differences among group means and
differences of group means from the overall mean have been
interpreted through the application of the SEM to individual
scores. Differences between means of greater than one SEM
are assumed to be of substantive importance to the clinical
interpretation of individual scores. No clinical importance is
attached to differences of less than one SEM .
The formula for the SEM is:
SEM = SD * 1 – r
√
In this instance, the SD is always 10, because the scores in
question are T-scores. In general, the r on the right side of
this equation is an estimate of the reliability of the scale in
question; in this instance, the internal consistency reliability
of each scale (coefficient alpha) is used as the value of r.
(The scale alphas for the PCRI are discussed in chapter 6 in
the section entitled “Internal Consistency.”) The interpretation criteria arrived at in this way for the scales of the PCRI
are presented in Table 4. The reported values are based on
the preliminary T-scores described earlier, and they are
rounded for simplicity.
Demographic Variables
Several analyses were done in order to evaluate the
effects of purely demographic factors, namely ethnicity,
SES, and region. In some cases these analyses were run on
stratified subsamples drawn from the standardization
Table 4
Standard Errors of Measurement (SEM s)
and Criteria for Interpretation of Substantive
Differences in the PCRI Normative Sample
PCRI Scale
SEMa
Interpretation
Criterion
SUP
SAT
INV
COM
LIM
AUT
ROL
5.42
3.84
4.76
4.24
3.49
4.53
5.03
5
4
5
4
3
5
5
a
Computed using Cronbach’s alpha as the estimate of scale reliability and
based on preliminary T-scores for each scale.
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 24
24
Parent-Child Relationship Inventory (PCRI)
sample. Each subsample was designed to approximate U.S.
population proportions of the variables in question.
MANOVAs on these samples address the possibility that
interactions among demographic variables could produce
substantial distortions in the data. The analyses also reveal
the effects of demographic variables on PCRI results.
Effects of ethnicity and education. A sample of 240
individuals stratified by education level and ethnicity to
approximate U.S. population proportions was drawn from
the larger standardization sample. Because representation of
other ethnic groups in the sample was too small for a meaningful analysis, only blacks and whites were included. A
MANOVA was run with ethnicity of child and respondent
education treated as independent variables and scores on the
seven content scales of the PCRI treated as dependent variables. The interaction term was not significant, but there
were significant main effects of both independent variables.
For each of the independent variables, there were significant
differences between groups on two of the seven PCRI
scales, as noted in Table 5. The relevant cell means are
presented in the table where there are significant group
differences on a given scale.
There were significant differences between blacks and
whites in their level of parental satisfaction and in their level
of autonomy, as measured with the PCRI. In both instances
these differences appear to be substantive: The mean
Satisfaction With Parenting score of blacks was nearly 6
T-score points lower than that of whites, clearly exceeding
the criterion for interpretation of a substantive difference;
the mean Autonomy score for blacks was approximately 4.5
T-score points below that of whites, almost precisely the
magnitude of the SEM for the autonomy scale.
There were significant differences related to level of
education on the support and autonomy scales. As might be
expected, parents with four or more years of college had a
higher mean score on the support scale than did parents at
the other three levels of education, a difference that exceeds
the criterion for interpretation of a substantive difference.
Furthermore, the difference between the mean for parents
with at least a college education and the overall mean
approximately equalled the criterion for interpretation of a
substantive difference. Scores on the autonomy scale for
parents with a high school education or less were lower than
the overall mean and lower than the scores of parents with at
least some college by amounts that exceed the criterion for
interpretation of a substantive difference.
Regional differences. A one-way MANOVA was run
with region as the independent variable and scores on the
seven content scales of the PCRI as dependent variables.
(The relatively representative distribution of the sample
across regions precluded the need for the drawing of a
stratified subsample.) There were significant regional
differences on four of the seven scales, as shown in Table 6,
which includes the relevant weighted cell means for those
four scales. In no instance do differences between regions
or differences of regions from the mean exceed the criterion
for interpretation of a substantive difference.
Table 5
Analysis of Variance Results and Associated Group Means
for Ethnicity and Education
F
Ethnicity
SUP
SAT
INV
COM
LIM
AUT
ROL
Education
SUP
SAT
INV
COM
LIM
AUT
ROL
*p < .05; **p < .01.
Group Means
.02
7.72**
1.53
2.08
.03
5.15*
.21
2.88*
2.53
1.40
2.60
2.56
4.75**
1.52
Less Than
High School
Graduation
48.29
—
—
—
—
42.39
—
Black
—
44.90
—
—
—
44.69
—
White
—
50.62
—
—
—
49.21
—
High School
Graduation
48.45
—
—
—
—
44.82
—
Some
College
48.34
—
—
—
—
51.66
—
Four Years
of College
or More
54.80
—
—
—
—
52.25
—
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 25
Chapter 5
Standardization
Content Variables
On a priori grounds, four variables are assumed to
have the potential of interacting in important ways with
factors the PCRI measures directly. They are the age of the
parent, the relationship of the parent to the child (the sex of
the parent), the sex of the child, and the age of the child.
Strong, interpretable relationships between these presumably content-related variables and scores on the PCRI
content scales could have direct implications for the structure of the PCRI norms.
Parent’s age. A one-way MANOVA was run with
parent’s age (four age levels) as the independent variable
and scores on the seven PCRI content scales as dependent
variables. The small number of parents in the sample over
the age of 54 was excluded from the analysis, as was the
small number of stepparents. There was a main effect for
age. The results are presented in Table 7. As before,
relevant cell means are given where there are significant
differences.
Direct examination of the means for parents of
different ages suggests that there is a marked difference in
both range and magnitude between the mean PCRI scale
scores of parents 24 years old and younger and those of
parents from 25 to 54. In part, this difference may be a
result of the smaller sample of younger parents. When the
youngest parents are excluded from the analysis, there is
no significant multivariate main effect of parent’s age nor
are there significant main effects for any of the PCRI
content scales.
Parent’s relationship to child. A one-way
MANOVA was run with parent’s relationship to the child
(mother or father) as the independent variable and scores on
the seven PCRI content scales as the dependent variables.
As shown in Table 8, differences between mothers and
fathers are among the most pervasive of the effects examined here. Although the difference on no subscale exceeds
the criterion for interpretation of a substantive effect, the
overall pattern of results strongly suggests a systematic
difference between mothers and fathers in their responses on
the PCRI. Consequently, there will be separate sets of PCRI
norms for mothers and fathers, as described in a later section
of this chapter (“Construction of the PCRI Norms”).
Child’s sex. A one-way MANOVA was run with sex
of child as the independent variable and scores on the seven
content scales of the PCRI as dependent variables. As
shown in Table 9, there was a significant difference favoring
females only for the Satisfaction with Parenting scale, a
difference not exceeding the criterion for interpretation.
Therefore, sex of child was not a factor in the construction
of the PCRI norms.
Child’s age. A one-way MANOVA was run with age
of child as the independent variable and scores on the seven
content scales of the PCRI as dependent variables.
(Respondent’s children were classified into three age levels:
5 years and younger; 6 through 10 years; 11 years and
older.) There was a main effect for age of child; there were
significant differences between age groups on six of the
25
seven scales, as shown in Table 10. As discussed in the
following section, the observed age differences were not
used in the construction of norms for the PCRI. The group
differences shown in Table 10 may be used as a guide by
those interested in adjusting the interpretation of individual
scores based on the age of the child.
Effect Size and Norms Construction
With one exception, the differences reported here
were not used as the basis for the construction of groupspecific norms. In general, these differences were neither
large enough nor pervasive enough to warrant the loss of
statistical power and convenience associated with the
construction of a series of separate norms tables.
The pattern of results for ethnicity does not justify
separate norms. The reliable difference on the Satisfaction
With Parenting scale does not, by itself, constitute evidence
of systematic differences in parenting between blacks and
whites. Nevertheless, the information presented in Table 5
does permit the making of slight adjustments in the interpretation of the PCRI profiles of black parents.
From a wider perspective, it seems appropriate to
perceive the lower scores of low-SES individuals as
possibly reflecting systematic disadvantages. The
observed differences may reflect a real need for some type
of intervention. Norms that adjusted for differences based
on SES would tend to reduce the proportion of clinically
significant elevations identified among low-SES individuals. Furthermore, the observed difference on the
Parental Support scale transparently reflects the more
privileged high-SES circumstances, leaving the Autonomy
scale alone as a possible reflection of structural differences in parenting related to SES. Therefore, the norms for
the PCRI were constructed without specific adjustments
for SES.
Parent’s age was also not used as a basis for norms
construction. The sample of young parents was small. As
with SES, the use of norms based solely on this segment of
the sample might under-identify the presence of serious
problems. The presentation of separate norms for these
groups of special concern might obscure problems where
they do exist.
Of the variables closely related to the content of the
PCRI, only the parent’s relationship to the child is a factor in
the norms. Although the effects of the child’s age were even
more pervasive than were those of relationship, age was not
used as the basis of norms construction for two reasons.
First, the effects of age were pervasive, but they were based
on slightly smaller differences overall than were those for
relationship. Second, the interests of statistical consistency
compete with those of economy of presentation. The small
but reliable differences accounted for by age of child do not
appear to justify complicating the presentation of the norms
literally by a factor of three. In contrast, the claim of
relationship (mother or father) as a basis for a distinction in
the norms is strong.
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 26
26
Parent-Child Relationship Inventory (PCRI)
Table 6
Analysis of Variance Results and Associated Group Means
for Region
F
Region
SUP
SAT
INV
COM
LIM
AUT
ROL
5.91***
4.73**
2.13
.38
2.68*
1.14
5.60***
Group Means
Northeast
50.92
51.20
—
—
50.07
—
49.50
South
49.92
49.76
—
—
50.61
—
50.95
Midwest
47.76
49.04
—
—
48.72
—
51.02
West
51.67
50.61
—
—
49.87
—
46.83
*p < .05; **p < .01; ***p < .001.
Table 7
Analysis of Variance Results and Associated Group Means
for Parent’s Age
F
Parent’s Age
SUP
SAT
INV
COM
LIM
AUT
ROL
2.39
10.46***
4.94**
2.09
1.20
3.32
.45
Group Means
18–24
—
38.76
42.94
—
—
43.61
—
25–34
—
50.68
51.24
—
—
50.29
—
35–44
—
50.84
50.47
—
—
50.46
—
**p < .01; ***p < .001.
Table 8
Analysis of Variance Results and Associated Group Means
for Relationship to Child
F
Relationship
SUP
SAT
INV
COM
LIM
AUT
ROL
*p < .05; **p < .01; ***p < .001.
9.80**
.28
14.09***
4.95*
8.74**
.10
4.57*
Group Means
Mother
48.83
—
52.24
51.76
49.51
—
50.81
Father
51.90
—
48.01
48.25
51.19
—
49.92
45–54
—
49.73
49.59
—
—
50.13
—
20-28_v4_20-27.qxd 6/28/10 2:28 PM Page 27
Chapter 5
Standardization
27
Table 9
Analysis of Variance Results and Associated Group Means
for Sex of Child
F
Sex
SUP
SAT
INV
COM
LIM
AUT
ROL
Group Means
Male
—
49.61
—
—
—
—
—
.00
4.40*
.02
1.03
3.26
2.06
.01
Female
—
50.83
—
—
—
—
—
*p < .05.
Table 10
Analysis of Variance Results and Associated Group Means
for Child’s Age
F
Child’s Age
SUP
SAT
INV
COM
LIM
AUT
ROL
5.07**
13.31***
7.08***
3.10*
4.64**
2.66
3.18*
Group Means
5 and under
49.68
49.74
49.98
50.17
49.12
—
49.12
6–10
51.18
51.84
51.31
50.83
51.04
—
50.80
11 and over
48.86
48.08
48.34
48.73
50.00
—
50.12
*p < .05; **p < .01; ***p < .001.
Construction of the PCRI Norms
The final normative sample included 668 mothers and
471 fathers, 1,139 parents in all. As discussed in the immediately preceding sections, respondents over 54 years of age
were excluded from the normative sample. (Protocols with
more than 10% of the responses missing had already been
excluded.)
Separate norms tables for mothers and fathers are
presented in Appendix B. Scale scores are presented as
T-scores with a mean of 50 and a standard deviation of 10. A
T-score conversion transforms raw scores so that the number of items on a scale does not greatly influence profile
elevations. Because the distribution of scores on some of the
scales was skewed, the T-score distributions for all of the
scales have been normalized. The normalization function
transforms the raw score distribution to fit a normal curve.
Therefore, the T-scores are based on the cumulative probabilities in the raw score distribution rather than on the mean
and standard deviation of the scores directly (Anastasi,
1988; Guilford & Fruchter, 1978).
As an aid in interpreting scores, percentile equivalents
of the T-scores are given for the 99th, 95th, 90th, 75th, 50th,
25th, 10th, 5th, and 1st percentiles.
The use of the norms in scoring is discussed in chapter
2 of this manual in the section entitled “Scaling Scores and
Displaying Results.”