University of Nebraska at OmahaDigitalCommons@UNO
Student Work
8-1981
Attitude-behavior relationships: A comparison ofthe Fishbein-Ajzen and Bentler-Speckart modelsArlene J. FredericksUniversity of Nebraska at Omaha
Follow this and additional works at: https://digitalcommons.unomaha.edu/studentwork
Part of the Psychology Commons
This Thesis is brought to you for free and open access byDigitalCommons@UNO. It has been accepted for inclusion in StudentWork by an authorized administrator of DigitalCommons@UNO. Formore information, please contact [email protected].
Recommended CitationFredericks, Arlene J., "Attitude-behavior relationships: A comparison of the Fishbein-Ajzen and Bentler-Speckart models" (1981).Student Work. 92.https://digitalcommons.unomaha.edu/studentwork/92
ATTITUDE-BEHAVIOR RELATIONSHIPS: A COMPARISON OF THE FISHBEIN-AJZEN
AND BENTLER-SPECKART MODELS
A Thesis
Presented to the
Department of Psychology
and the
Faculty of the Graduate College
University of Nebraska
In Partial Fulfillment
of the Requirements for the Degree
Master of Arts
University of Nebraska at Omaha
by
Arlene J. Fredricks
August 1981
UMI Number: EP72742
All rights reserved
INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
Published by ProQuest LLC (2015). Copyright in the Dissertation held by the Author.
UMI EP72742
Microform Edition © ProQuest LLC.All rights reserved. This work is protected against
unauthorized copying under Title 17, United States Code
ProQuest LLC.789 East Eisenhower Parkway
P.O. Box 1346 Ann Arbor, Ml 48106-1346
THESIS ACCEPTANCE
Accepted for the faculty of the Graduate College, University of
Nebraska, in partial fulfillment of the requirements for the degree
Master of Arts, University of Nebraska at Omaha.
Thesis CommitteeDepartmentame
Chairman
Date
ii
ACKNOWLEDGEMENTS
The author is indebted to D. Kenneth Spenner of the Boys Town
Center for Youth Development for not only making the LISREL computer
program available, but for his invaluable instruction and guidance that
made use of it possible; Dr. Dennis Dossett for innumerable suggestions,
guidance, editorial assistance, and encouragement; the Committee--
Dr. J. Brad Chapman, Dr. Carl Greenberg, Dr. Gaylon Oswalt, for reading
and helpful comments; Mr. Patrick Jordan for coding data; Instructors
who collected daily attendance data and permitted the use of class time
for questionnaire administration--Mr. David Arnold, Dr. Carl Greenberg,
Mr. William Grisham, Mr. Gary Kinstlinger, Dr. Barbara Manning, Dr. C.
Raymond Millimet, and Mr. Edward Ward; Instructors who permitted use of
class time for pilot sample data collection--Dr. Kenneth Deffenbacher,
Dr. Joseph LaVoie, Mr. Jack Leon, Mr. Russell Montgomery, Dr. Gaylon
Oswalt; and Mr. Jim Smith of the UNO Computer Network for much appre
ciated technical assistance.
TABLE OF CONTENTS
Page
T.i st. of T a b l e s . ........................................................... vi
List of F i g u r e s ..............................................................vii
Abstract.................................................................... viii
The Attitude-Behavior Relationship....................................... 1
The Fishbein-Aj zen Model.......................................... 3
Conceptual Framework of the Fishbein-Aj zen Model.............. 5
The Fishbein-Aj zen Model of Attitude Formation................ 7
Fishbein and Ajzen’s Specifications for Measurement ......... 9
The Bentler and Speckart Model........................................... 12
Method....................................................................... 17
Subjects..................... 17
P r o c e d u r e ........................................................... 18
Measures............................................................. 20
Analysis............................................................. 22
R e s u l t s ..................................................................... 24
Pilot and Primary Study Sample Comparisons..................... 24
Measurement M o d e l .................................................. 24
Structural Model.................................................... 25
Model C o m p a r i s o n s .................................................. 25
Chi-Square Goodness of Fit Tests................ 34
Chi-Square Difference Tests ...................................... 37
Discussion.................................................................. 40
iv
Page
Reference Note................ 46
References.................................................................. 47
Appendix A--Questionnaire for Pilot Study ............................. 50
Appendix B--Questionnaire for Primary Study ........................... 55
Appendix C--Tables of Results............................................ 69
Appendix D--Discussion of Path Analysis .............................. 82
v
LIST OF TABLES
Table Page
1* Structural Model Specifications.................................... 29
2. Path Comparisons between Models, t-Values, . . . ............. 31
3. Chi-Square Goodness-of-Fit Tests ..................... 35
4. Chi-Square Difference Tests for Model Comparisons................ 38
vi
LIST OF FIGURES
Figure Page
1. Schematic Representation of Fishbein-AjzenModel of Attitude-Behavior Relationship....................... 6
2. Schematic Representation of Fishbein-Ajzen Modeland of the Bentler-Speckart Modification . . . ...............14
3. Fishbein-Ajzen Model (Prior Behavior Excluded)with Paths Labeled for LISREL Analysis ....................... 26
4. Bentler-Speckart Model (Prior Behavior Included)with Paths Labeled for LISREL Analysis ....................... 27
vii
ABSTRACT
This study compared the Fishbein-Ajzen (1975) model of attitude-behavior
relationships with Bentler-Speckart's (1979) modifications of the model.
Subjects were 236 undergraduate college students and the measures of
behavior were repeated self-reports of class attendance. An analysis of
linear structural relationships, using multiple indicators for each under
lying construct, supported the Bentler-Speckart addition to the Fishbein-
Aj zen model of prior behavior as a direct causal influence on both sub
sequent behavior and behavioral intentions. However, consistent with the
original Fishbein-Ajzen model, a direct causal path from attitude to sub
sequent behavioral intentions was not found. Directions for future studies
and respecification of the model were discussed.
viii
1
The Attitude-Behavior Relationship
That attitude can sometimes predict behavior has been documented
and reported in recent reviews of the attitude-behavior relationship
literature (Ajzen § Fishbein, 1977; Calder § Ross, 1977; Eagly £
Himmelfarb, 1978; Kelman, 1974). However, discovery of the conditions
and processes that permit predictions of behavior remains a challenge
for behavioral scientists. This research question must necessarily be
partitioned into a conceptualization of attitudes on the one hand and
behavior on the other. Since behavior is more easily observed and
measured, it is the attitude construct that has attracted the greatest
amount of methodological attention.
Numerous definitions of attitude have been promulgated. However,
the present discussion will be limited to only relatively recent
approaches. For example, Rokeach defined attitude as "an organization
of interrelated beliefs around a common object, with certain aspects of
the object being at the focus of attention" (1968, p. 116).
Triandis (1971) presented a definition which he felt included many
previously developed central ideas as follows: "An attitude is an idea
charged with emotion which predisposes a class of actions to a particu
lar class of social situations" (p. 2). This definition references
three components of attitude: (a) the cognitive or "idea" component;
(b) the affective or emotional component; and (c) the behavioral or
predisposition to action component. In this context predisposition to
action does not necessarily imply actual behavior. Triandis (1971),
in discussing the attitude-behavior relationship, notes that attitudes
involve:
2
What people think about, feel about, and how they would like
to behave toward an attitude object. Behavior is not only
determined by what people would like to do but also by what
they think they should do, that is, social norms, by what they
have usually done, that is, habits, and by the expected
consequences of the behavior. (p. 14)
Implicit in this definition is the concept of the determination of
attitudes by the cognitive component, beliefs.
Triandis, in a more recent discussion (1979), points out a basic
source of controversy in social psychology; that operationalization and
measurement of a construct is dependent on how the construct is defined.
If attitude is linked to behavior by definition; then, the behavioral
scientists concern is to explore the conditions under which either a
strong or weak relationship between verbal attitudes and behavior are
likely to be observed.
Calder and Ross (1976), in considering the psychological founda
tions of attitudes, conceive of attitudes as evaluative summaries of
underlying beliefs. According to this view, in order to understand
attitudes, it is first necessary to understand the information structures
or belief systems which underlie attitudes. Secondly, it is necessary
to understand how these beliefs are processed or integrated to produce
the evaluative summary called an attitude.
Some of the major contributions relevant to these issues include
the cognitive consistency approach of Heider's (1946) balance theory
which stresses the function of the perception of consistency in attitude
formation and change. Along the same line of reasoning was Festinger's
3
(1957) theory of cognitive dissonance which examined perceptual incon
sistencies and modes of reducing the resulting dissonance. Bern (1967,
1972), in an alternate approach, proposed a theory of self-perception
which suggested that behavior might well be an antecedent rather than
a result of attitude in that individuals infer what their attitudes must
be from observation of what their behavior is. Thus, there is a vast
complexity of attitude constructs and attitude-behavior relationships
and a multiplicity of approaches taken in attempting to gain an under
standing of attitudinal components and their processes.
The theoretical approach focused upon in this study is that
developed by Fishbein (1967) and elaborated by Fishbein and Ajzen (1975).
This theoretical framework has served to integrate much of the currently
accepted attitude-behavior knowledge into a theory that is explicit,
testable, and widely generalizable. It is the application of this
theory in the area of attitudes toward attendance/absenteeism that is
the subject of this study.
The Fishbein-Ajzen Model
Rather than providing a simple definition of attitude, Fishbein
and Ajzen (1975) proposed a conceptual framework systematically integrat
ing theoretical attitude components and their underlying processes.
They note that various interpretations or definitions of "attitude"
have in the past implied different measurement procedures which conse
quently produced different results in attitude studies. The subsequent
confusion as to exactly what "attitude" is has been the logical result.
Fishbein and Ajzen (1975) prefer to define attitude by its generally
agreed upon most essential component which is the major characteristic
4
that distinguishes attitude from other constructs, that is, its evalu
ative or affective nature.
According to this definition, a distinction must be made between
belief and attitude. This distinction implies the testable proposition
that beliefs and attitudes have different determinants and that changes
in them can lead to different consequences. Accordingly, Fishbein and
Ajzen (1975) use the term, "attitude,” to refer to affect, i.e., feelings
toward or evaluation of an attitude object, and the term "belief,” to
represent cognition or knowledge about the object, specifically the
linkage of an object to some attribute. This definition, while basically
the same as Rokeach’s (1968) definition of attitude, emphasizes the
separation of the concepts of attitude and belief.
The third generally recognized component of attitude, the behavior
al component, is divided in the Fishbein-Ajzen model to refer to both
behavioral intentions and actions with respect to or in the presence of
the attitude object. This distinction between behavioral intentions
and actual behavior is made since most theorists agree that attitudes
are concerned with predispositions to behave rather than with the
behavior itself.
The Fishbein-Ajzen (1975) model, then, is a descriptive framework
of the relationship between four broad categories: affect (feelings,
evaluation), cognition (opinions, beliefs), conation (behavioral inten
tions), and behavior (observed overt acts). The term "attitude" is
reserved for only one of these categories, affect.
5
Conceptual Framework .of the Fishbein-Ajzen Model
The conceptual framework of the Fishbein-Ajzen model for the rela
tionship of attitudes to behavior can perhaps best be presented
schematically (see Figure 1}. According to this framework, the perform
ance or nonperformance of a specific behavior is determined by the
intention to perform that behavior. Consequently, the prediction of
behavior toward an object from knowledge of attitude toward that same
object is accurate only insofar as that attitude influences the intention
to perform the behavior. This behavioral intention is a function of
beliefs, not about the object of the behavior, but instead beliefs con
cerned with the behavior itself.
A person’s attitude toward performing a given behavior is represented
as a function of two types of beliefs. One of these is that performing
the behavior will lead to certain consequences along with his/her evalu
ation of these consequences. The other relevant beliefs, labeled sub
jective norms since they are normative in nature, are beliefs that
certain relevant others think that the person should or should not
perform the behavior in question. Subjective norms are combined multi-
plicatively with the subject’s motivation to comply with these norms.
According to this conceptual structure, beliefs are the fundamental
building blocks, an informational base that is the ultimate determinant
of attitudes, intentions, and behavior. The formation of attitudes,
then, is viewed in terms of an information processing approach wherein
a person’s salient set of beliefs about the object determines his/her
attitude toward that object. Applied to behavior, it is the set of
beliefs as a whole, including behavioral intentions, which are viewed
Info
rmat
iona
l
6
pco3X*33cdcd
tin 33P3£OP
CD P33 O3 •Hp >•H 3P r 3P CD< CQ
to<DCD3I CD
3 3<P o'O CDP > to3 to w 3O CD O
X i O X CD3 3CD p <P
3 o O<p o ' •HCD > 3•H to 3 Oip 3 X •HCD O CD P03 U 0Q 3
PO■H>£3 3P XO CD2 CQ
CD toO> 3•H •HP 3O PCD CD•i—>OrC3 33 Oco U/
Xto
tp i t— \CD toO to•H P •H Pp O CO CD OCD -H 3S PCQ > <P P3 o O 3CD r3 O> CD to P H
•H CQ <p 3 PP CD 3 3 X3 P ■H O > pS 3 iP •H H pHP O CD <P p gO JQ 03 •H O O2 3 '— ' 3 S U
PS0 3
pO"dCDcd
tin1
CDP3bO•rHPU
J
Sche
mati
c Re
pres
enta
tion
of
Fish
bein
-Ajz
en
Model
of At
titu
de-B
ehav
ior
Rela
tion
ship
7
as a special case of beliefs, that is the determinant of attitudes in
the attitude-behavior relationship.
The Fishbein-Ajzen Model of Attitude Formation
Fishbein and Ajzen (1975) suggest an expectancy-value model of
attitude formation. The prediction of behavioral intentions is a
function of the weighted sum of two variables, the attitude toward
performing the behavior and the subjective norm as follows:
B 'v I = (A ) + (SN)B''w1 v 'w2
where B is the behavior, I is the intention to perform the behavior,
A is the attitude toward performing the behavior B, SN is the subjective
norm, and w^ and w^ are empirically determined weights. The attitude
toward performing a specific behavior is proposed to be a function of
the perceived consequences of performing that behavior and of the
person's evaluations of those consequences:
nA = L b.e. B . . 1 1 i=l
where b is the belief that performing behavior B leads to consequence
or outcome, i_, e_ is the person's evaluation of outcome i , and n is the
number of beliefs the persons hold about performing behavior B.
The normative component, SN, deals with the influence of the social
environment on behavior. The subjective norm is the person's perception
that people who are important to him/her think he/she should or should
not perform the behavior in question. According to Fishbein and Ajzen
8
(1975), the general subjective norm (SN) is determined by the perceived
expectations of specific referent individuals or groups, and by the
person's motivation to comply with those expectations:
nSN = I b . m .
1-1 1 1
where b^ is the normative belief, m^ is the motivation to comply with
referent jl, and n. is the number of relevant referents.
These two major determinants of behavioral intentions are given
empirical weights in the prediction equation proportional to their
relative importance. Since adequate estimates of these weights for
each individual are not generally available, the accepted practice has
been to use multiple regression techniques and standardized regression
coefficients as estimates of the weights for the theory components.
The present version of the theory, in the form of a multiple regression
equation i s :
i = (3b)ab ♦ ceSN)SN
where I is the behavioral intention.
The component of motivation to comply can be expressed as the
person's intention to comply with the referent in question. The
determinants of this intention are the same as those of any other
behavioral intention and can be expressed in equation form,
m ^ I « (A„) i (SN)— C v CJ w^ v 'w2
9
where m is the motivation to comply with the referent, 1 , is the
intention to comply with the referent , A is the attitude toward comply-Vj
ing with the referent, SN is the subjective norm concerning compliance
with the referent, and w^ and are weights.
The Fishbein-Ajzen (1975) theory accounts for the influence on
intentions of additional variables external to the model only through
their indirect influence on either of the two components (attitude and
subjective norms) or on the relative weights of these components.
Accordingly, the attitude toward the target object or person will be
unrelated to the behavioral intention itself if it is not related to
either the attitudinal or normative component of behavioral intention.
Only if the component variable in question carries a significant weight
in the regression equation predicting behavioral intention will attitude
toward the object be related to or predictive of intentions. It is the
behavioral intention which is considered to be the determinant of overt
volitional behavior. A number of studies are cited by Ajzen and Fishbein
(1977) and Fishbein and Ajzen (1975) in support of this theoretical
formulation.
Fishbein and Ajzen’s Specifications for Measurement
A major factor in the prediction of overt behavior from behavioral
intentions is the necessity for correspondence with respect to the
level of specificity between intentions and behavior and also between
intentions and the components of intentions. Fishbein and Ajzen (1975),
in their development of a framework for linking attitudes and behaviors,
are highly specific as to the measurement methods that they consider
appropriate. Consequently, any study designed to investigate this model
10
risks the possibility of testing constructs other than those designated
if other methods of measurement are employed. The procedure recommended
by Fishbein and Ajzen is to measure attitude "by a procedure which
locates the subject on a bipolar affective or evaluative dimension
vis-a-vis a given object" (Fishbein § Ajzen, 1975, p. 11). This procedure
is an exact definition of the semantic-differential scale (Osgood, Suci,
§ Tannenbaum, 1957).
In contrast to the evaluative nature of attitudes, beliefs represent
information. Differences among individuals in this respect are defined
in terms of belief strength or the perceived likelihood that an object
has or is associated with a particular attribute. The recommended
procedure for the measurement of belief strength, then, places the sub
ject along a dimension of subjective probability involving an object and
some related attribute. For example, the more money a person is per
ceived to possess, the higher should be the subjective probability that
the person is wealthy.
Since behavioral intention is conceptualized as a special case of
beliefs, the strength of a behavioral intention is appropriately
measured by a procedure which places the subject along a subjective
probability dimension involving a relation between himself/herself and
some action. For example, the strength of an intention to attend church
on Sunday would be measured by the subject’s probability rating of the
concept, "I will attend church Sunday,” on scales anchored by "probable-
improbable” or "agree-disagree."
According to Fishbein and Ajzen (1975) only a limited number of
salient beliefs can serve as determinants of attitude at any given time.
11
Therefore, measures of beliefs should also include assessment of their
saliency in the subject's belief hierarchy. This can be accomplished
by considering as salient only the first few responses elicited in a
free-response format when subjects are asked for a listing of charac
teristics, qualities, and attributes possessed by an object or for the
consequences of performing a behavior.
This procedure is similar to content analysis. When a modal set
of beliefs for a population is to be identified, the responses obtained
from a representative sample are first grouped according to similarity
and the frequency of each similar belief is counted. It is necessarily
a matter of judgment as to whether or not similar-appearing beliefs are
to be considered identical. The set of beliefs to be included in the
modal set is then arbitrarily set as the number of beliefs sufficient
to account for a stated percentage, such as 75%, of all beliefs emitted
by the sample. This procedure is described by Ajzen and Fishbein (1980).
By definition, overt behaviors are observable acts. Fishbein and
Ajzen (1975) view behavior as consisting of four elements: behavior,
target, situation, and time. They make a distinction between different
types of behavioral criteria in terms of the variance of the criteria
with respect to one or more of these elements.
A single-act criterion, i.e., the single observation of a single
act, is always specific with respect to the four elements of behavior
as it involves a directly observable response to a specific target, in
a given situation, at a given point in time. A repeated-observation
criterion is an index of behavior derived from repeated observations of
the same behavior, such as observations across several trials in an
12
experiment. Such criteria can represent generalizations across targets,
across situations, or across time. A multiple-act criterion represents
a behavioral index computed from observations of different behaviors
with respect to a given target, in a given situation, at approximately
the same point in time. For example, withdrawal behavior in a social
situation can be measured by the degree of conversational participation,
eye contact, physical distance, and the amount of time spent with a
group. A combination of repeated observations of more than one behavior
would be considered a multiple-act, repeated-observation criterion.
Since behavioral observations are data, such observations can be
subject to the same problems of unreliability and invalidity as any other
form of data. For this reason, Fishbein and Ajzen (1975) note that
rigorous analyses of behavioral data are essential for an understanding
of the relation between attitude and behavior and that inconsistent
research findings from attitude-behavior studies may be due to the use
of inappropriate behavioral measures. They conclude that, in regard to
single-act criteria:
Not every behavior with respect to some object is related to
the attitude toward that object [and that] multiple-act and
repeated-observation criteria . . . when properly constructed
on the basis of standard scaling procedure . . . can serve as
indicants of attitude. (Fishbein § Ajzen, 1975, pp. 356-357)
The Bentler and Speckart Model
Bcntlcr and Speckart (1979) proposed and tested a modification and
extension of the Fishbein and Ajzen model. In the Bentler-Speckart
model, which they consider to be "the most theoretically adequate
13
causal-predictive system relevant to a variety of behavioral domains"
(1979, p. 455), affect (attitude) has a direct effect on behavior in
addition to its indirect influence on behavior by means of its influ
ence on intention. A second modification is the addition of previous
behavior to the model. This previous behavior is postulated to have an
effect on both current intentions and on future behavior that cannot be
accounted for by the original Fishbein and Ajzen model. A schematic
of these two approaches is presented in Figure 2.
Bentler and Speckart (1979) reason that, since behavioral intention
is conscious and thereby cognitive in nature, the Fishbein-Ajzen theory
which proposes that affect impacts behavior only by means of the regu
lation of intention or premeditation (conation-cognition) is counter
intuitive in most domains of behavior, less accurate and generalizable,
and has less predictive power than their conceptualization.
They also propose that the role of previous behavior in accounting
for future behavior is inadequately modeled in the Fishbein-Ajzen
approach of indirect influence through attitudes and subjective norms.
Bentler and Speckart state that previous behavior may "circumnavigate
these factors in its causation of subsequent behavior in the same way
that attitudes circumnavigate intentions" (1979, p. 454). This the
oretical formulation, whereby behavior has an independent role in the
prediction of future behavior, is consistent with other theoretical
formulations such as Bern's (1967, 1972) self-perception theory in which
attitudes may be generated from self-perceptions of behavior. Other
research findings indicating a relationship between past and subsequent
behavior would appear to be consistent with both the direct and indirect
Schematic Representation of Fishbein-Ajzen Model14
Attitudetoward
Behavior
SubjectiveNorms
BehavioralIntentions
TargetBehavior
Schematic Representation of Bentler-Speckart Modification
PreviousBehavior
BehaviorIntentions
TargetBehavior
Attitudetoward
Behavior
Subjective Norms
Figure 2
15
previous-behavior influence models.
In the Bentler-Speckart (1979) study, three models were tested with
the same data: the two schematically represented in Figure 2 and an
intermediate modification of the Fishbein-Ajzen model. Bentler and
Speckart (1979) collected their data following Fishbein and Ajzen1s
recommended approach of using semantic-differential scales. A sample of
228 college students were asked three questions on each of the constructs
of attitudes, subjective norms, and intentions at one point in time.
Behavior was measured twice within a two-week period. To obtain replica
tions of the tests for each model, these five variables were assessed
for each of three attitudinal domains: alcohol, marijuana, and hard
drug use. The behavior measures were not observations of behavior but
rather questionnaire responses of self-report of behavior for the two-
week period prior to the time of data collection. The first behavior
measure was taken at the time of the complete questionnaire administra
tion; the second two weeks later.
The first analysis was a comparison of the Fishbein-Ajzen model as
shown in the upper half of Figure 2 with a modification which added
only a direct path from attitude to behavior to the Fishbein-Ajzen
model. Thus, this modified model did not include prior behavior.
Attitudes and subjective norms were the exogenous or independent vari
ables; intentions and subsequent behavior were dependent or endogenous
variables. Endogenous variables are defined as variables whose causes
are completely determined by variables included in the causal model;
exogenous variables are determined by causes lying outside the model.
For the additional comparison of models with the fully-expanded
16
Bentler-Speckart model, as presented in the lower half of Figure 2; the
measures of previous behavior were included. Thus, prior behavior is
conceived as a latent variable which was included in the causal model.
A factor analysis of the data supported the conclusion that the
theoretical constructs hypothesized as latent factors were adequately
assessed and were reasonably indicated by the observed variables.
Parameters for each causal model (Fishbein-Ajzen model, Bentler-Speckart
first modified model, and Bentler-Speckart fully-expanded model) were
estimated by a computer program, LISREL IV (Joreskog § Sorbom, 1978).
Hierarchical models were generated by adding parameters (causal paths)
to or removing them from the models being tested. Chi-square difference
tests of the null hypothesis that each parameter in question is not
present in the proposed causal structure in the population were used to
compare competing models. Results of the statistical analysis supported
Bentler and Speckart's (1979) hypothesis that the addition of three
structural parameters, i.e., direct paths from attitude to subsequent
behavior and from previous behavior to both intention and subsequent
behavior, is necessary for the causal model to successfully reproduce
the data.
The Bentler-Speckart (1979) study was carefully conceived and
executed; however, some comments on it appear to be in order. The five
latent variables assessed in their study (attitude, subjective norms,
prior behavior, intention, and target behavior) were each assessed by
an overall measure rather than by separate measures of their components.
Attitude was measured as the evaluative component only and did not
include beliefs and belief strength. Subjective norms were measured by
17
belief strength but did not encompass the motivation to comply.
Behavior was measured by self-report rather than by observations of
overt behavior. The element of response bias inherent in self-reports
of socially-censured or potentially illegal acts should be considered
since the target behaviors of the study were the use of alcohol,
marijuana, and hard drugs. Thus, the measurement procedure did not
correspond to Fishbein and Ajzen's recommendations and may have unduly
biased the results against the Fishbein and Ajzen model.
The study reported here is a comparison between the Fishbein-Ajzen
model and the Bentler-Speckart fully-expanded model, testing for dif
ferences in predictive power between them. If the expanded model were
to show no significant increase in prediction over the basic Fishbein-
Aj zen model and if the additional causal paths hypothesized in the
Bentler-Speckart model do not demonstrate significant structural path
coefficients, then support for the Fishbein-Ajzen model would seem war
ranted on the basis of the most parsimonious explanation of the data
and the relationships represented.
Method
Subjects
Subjects were 259 college students, of both sexes, who were
enrolled in the 1980 summer session at the University of Nebraska at
Omaha in psychology classes which met five days per week. Extra credit
toward the student's course grade was given for completion of the
questionnaire. For the final data analysis, 23 subjects, comprising
one class, were eliminated due to insufficient behavioral attendance
data. This left a final sample of 236 for the primary study sample.
18
Procedure
Demographic data were collected on each subject in a pilot sample
and in the primary study sample. These data included information on
sex, age, marital status, student status (number of hours in which
currently enrolled and expected to be enrolled in the coming fall
semester), expected date of graduation, employment status (number of
hours worked per week currently and expected in the fall), family
income, U.S. citizenship, and race. The purpose of this data collection
was to assess the representativeness of the pilot sample relative to
the population from which the primary study sample was also drawn.
Data on expected fall semester school/work status and on the expected
date of graduation were collected because of the possibility that these
variables might serve a moderating function.
A pilot sample of 123 summer school students in psychology classes
was used to identify modal beliefs of the consequences of the target
behaviors (class attendance/absenteeism) and the significant others in
relation to these behaviors. The procedure outlined by Ajzen and
Fishbein (1980) and previously described was followed. Subjects were
asked in a free-response format to list the beliefs that came to mind
as possible consequences of the behavior of attending or of being absent
from class. They were also asked to identify individuals or groups
whose opinions with respect to these behaviors are important to them.
See Appendix A for the pilot sample questions.
Tabulation of results from the 123 returned questionnaires yielded
1,093 responses to the beliefs (behavioral consequences) questions.
Of these, 90 responses were directed at the consequences of attending
19
summer school per se and an additional 97 responses were "none" or
blanks. Elimination of these null or nonrelevant responses left 906
attendance/absence behavioral consequences which were then grouped and
tabulated as categories. Eleven response categories represented 698
responses or 77% of total relevant responses elicited. An arbitrary
designation of 75% of total responses had been established as the
criteria for salient modal responses to be retained for the primary
study questionnaire. These eleven response categories were then used
in formulating corresponding questions to be rated by subjects in the
primary study (see Appendix B for the primary study questionnaire).
Question 6 of the pilot study questionnaire was designed to elicit
modal referents for the population. The total number of responses to
this question was 260. Of these, 67 (25.77%) were "myself" or "myself
alone" and 4 were "no one." Of the remaining 189 responses, five
categories included 157 (83%) of the total. These were the response
categories used in the primary study as modal referents.
The primary study used students in nine, five-week summer school
classes. Complete data were obtained from 236 subjects out of 295
students originally registered for these classes. Of these 295 students,
25 either provided no attendance data or provided none after the first
week of attendance data collection. If the assumption is made that
these students dropped out of the classes, the volunteer rate was 87.4
for participation in the study.
The target behavior or behavior of interest was class attendance/
absence which was assessed by passing attendance sheets for students
to sign. Students were uniformly told at the beginning of the session
20
that class attendance was not a factor in grading, but that attendance
sheets would be passed for "administrative purposes." In order to avoid
confounding regular class attendance with the special case of attending
class on a test day, no attendance was collected on test days. In each
class there were some days, unsystematically distributed, on which
attendance was not collected. Because of the unequal number of data
collection days between classes, the measures of behavior were the ratios
of days attending to the number of days in that time period.
Prior behavior was the attendance/absence data collected for two
weeks before the collection of attitude, subjective norms, and intention
data. The only explanation given for collection of these attendance
data was that they were "for administrative purposes." At the midpoint
of the summer school session, (i.e., between weeks 3 and 4), a semantic-
differential questionnaire was administered to all subjects to assess
attitudes toward the target behavior, subjective norms, and behavioral
intentions. Target behavior was the attendance/absence data obtained
after collection of the questionnaire data and until the end of the
summer school class session, the final two weeks of the session. The
first week of the five-week session was omitted from the study as that
is typically a week of instability in which many students disenroll
from their courses.
Measures
Multiple indicators (measures) were used to assess the reliability
of measurement for each latent variable (construct) and to remove
measurement error from the relationships among the latent variables.
Thus, behavioral measures were combined into several pre- and
21
post-intention periods. The eleven days of class prior to and including
the day of questionnaire administration were designated as prior behavior
measures and divided into three periods. Likewise, the post-questionnaire
period of eight days was designated as target behavior and divided into
two periods of four days each. As stated earlier, these measures were
the ratio of days attended to days attendance was taken.
Intention was assessed by two questions, "I intend to attend this
class every session . . .," and "I intend to be absent from this class
some times . . ." Scales were anchored with "likely" and "unlikely" at
their respective endpoints with reversed scoring for the intention to
be absent question.
Two measures of the attitude toward the behavior were used. One
was a scale score derived from the summation of seven items (item 3
in the questionnaire in Appendix B) evaluating the behavior of class
attendance on seven-point scales with endpoints anchored by "important-
unimportant," "worthless-valuable," "good-bad," "rewarding-punishing,"
and so forth. The other measure of attitude was derived from the rating
of each of eleven consequences of the behavior (items 5 through 15 of
the questionnaire in Appendix B ) . These consequences were obtained from
the pilot sample; the rating given to each consequence on the seven-;
point scale was multiplied by a specific evaluation of it (items 1
through 11 in the evaluation section of the questionnaire). These eleven
products of belief strength and evaluation were summed to form a score.
Subjective norms were also assessed by two measures. The first
was a general subjective norm measure, "Most people who are important
to me think I should attend this class every day during the summer
22
session.” The rating given to this question was multiplied by a rating
of motivation to comply with this perceived belief which was obtained
by asking, "Generally speaking, how much do you want to do what others
who are important to you think you should do?” The other subjective
norm measure was obtained from ratings given to perceived beliefs of
specific significant others, such as parents, friends, instructors, and
so forth obtained from the pilot sample (items 3 through 7 in the ques
tionnaire section on '-how you think other people would like you to
behave”). These ratings were then multiplied by the subject's motivation
to comply with these perceived beliefs (items 16 through 20 on the last
page of the questionnaire) and the products summed.
Analysis
The method of data analysis used in this study was an extension of
path analysis developed by Joreskog and Sorbom (1978) in an effort to
combine the efficacy of path analysis in explicating underlying rela
tionships with a confirmatory factor analytic approach used to identify
the factors (latent variables) in such relationships. While a path
analytic approach is appropriate for theory testing and clarification
(Billings § Wroten, 1978; Cook $ Campbell, 1979; Kerlinger § Pedhazur,
1973; Li, 1975; Namboodiri, Carter, § Blalock, 1975), path analysis has
certain limitations and shortcomings (see Appendix D for a general dis
cussion of path analysis). The technique developed by Joreskog and
Sorbom (1978) eliminates or avoids many of these problems and has been
used in recent studies exploring causal models (e.g., Maruyama §
McGarvey, 1980; Pedhazur, Note 1). Also, this method was deemed
especially appropriate since it was used by Bentler and Speckart (1979,
23
1981) in their tests of modifications of the Fishbein-Ajzen model.
This method is an analysis of linear structural relationships by the
method of maximum likelihood operationalized in the computer program
LISREL IV. Multiple observed indicators of unobserved latent constructs
are used to infer relationships between the latent, unmeasured variables.
This analysis provides a measurement model and a causal model. The
inclusion of measurement error or unique variance as explicit parameters
in the model permits causal regression parameters to be estimated with
out the influence of measurement error.
The Fishbein-Ajzen and Bentler-Speckart models were compared by
estimating the various structural parameters of a saturated model, i.e.,
a model in which all paths possible were estimated, and then computing
estimates for parameters of nested modifications of this saturated model.
This was accomplished by setting various parameters equal to zero, that
is, removing paths from the model.
Two comparative indices were used. First, the critical ratio of a
specific parameter gives a significance test for that parameter. The
critical ratio of each causal parameter reported as a t-value is the
ratio of the unstandardized LISREL parameter estimate divided by its
standard error. These ratios, due to the large sample size, are inter
preted as standard normal deviates and represent levels of significance
for the parameter. At an alpha level of .05, a critical ratio of less
than 1.96 would indicate the non-significance or expendability of the
specific parameter.
In addition, a chi-square difference test between the hierarchical
models tested the null hypothesis that the specific parameter omitted
24
in the nested model is not present in tbe causal structure in the popula
tion. The rationale for the chi-square difference test (Bentler § Bonett,
1980; Bentler § Speckart, 1979, 1981) is based on goodness-of-fit chi-
square tests and the associated degrees of freedom for each model compared.
In the case of parameter nesting, the models to be compared differ only in
that the parameter vector of the more restricted model is a special case
of the parameter vector of the less restricted model, with certain para
meters constrained to equality or known constants. The null hypothesis is
of model equivalence and the difference between chi-squares with degrees
of freedom equal to the difference in parameters estimated provides a
statistical test of that null hypothesis and of the statistical necessity
of the parameters that differentiate the models.
Results
Pilot and Primary Study Sample Comparisons
Analyses of demographic data obtained from subjects in the pilot
and primary study samples supported the hypothesis that the two samples
were drawn from the same population. Only two of the fifteen variables
measured differed at the .05 level of significance; report of family
income and expected year of graduation. The primary study sample
reported an earlier date of graduation and a greater proportion reported
lower family income. All of the comparisons are reported in Table 1 in
Appendix C.
Measurement Model
The measurement model presents the standardized factor loadings of
the measured variables on the latent factors. These loadings, or
lambdas (A), may be interpreted as validity coefficients reflecting the
25
degree to which the observed variables adequately measure the specified
underlying construct. These parameters range from .609 to 1.0, a range
reported by Bentler and Speckart (1979) as adequate. Unique variance
represents the proportion of the variables' variance that is not
accounted for by the factors and includes measurement error. This is
given by epsilons (e) and deltas (6 ) in the schematic representations
(Figures 3 and 4). The top half of Tables C2 and Cll in Appendix C
presents the measurement model for each causal structure tested.
Structural Model
The structural or causal model estimates parameters of the rela
tionships between latent, unmeasured variables with error of measurement
removed. Gammas (y) are interpreted as path coefficients from exogenous
to endogenous variables; betas (3 ) are interpreted as path coefficients
between endogenous variables. The relationships between latent exogenous
variables are given by the phis ($). Residuals of latent endogenous
variables are represented by zis (£). The bottom half of Tables C2
through Cll presents the structural model for each causal structure
tested. The following summary of results is based on the data presented
in these tables in Appendix C.
Model Comparisons
Figure 3 presents the structural models of relationships between
attitudes, subjective norms, intentions, and target behavior. Prior
behavior is not included. Circles represent latent, unmeasured vari
ables and rectangles represent the observed, measured variables.
Double-headed arrows represent covariance and single-headed arrows
represent hypothesized causal paths. Figure 4 presents the structural
26
ta•H(AXr—IP5-4Wc£co
p DOo P4h •HPdPd bfl P0 p •HT-( •H m0 PdrQ p rPP •H -P-4 PPH,— -PPd ■P P0 cd PPd Ph PP or-1 •P •rHo P CHX cd •Hw o P•H bOP cp •HO •H (A•H P P> fcuOoP •H p,P tA0 p04 cd 0P 0 pO P p•H cd oP o •HOh •H pd'—' Pd PP •rHr-H •H0pd0 sp0tsi•r-><1p
•H0rOX tA •HH,
toCDuPbfl
•HP-,
<A U) 0 <D £ P -H•H i—I r“H PPd <D •H ^i—I o O U CO CQ
COXi—tctSG<04OiSco
27
h3 000 Gi-H •H0j=> 00 Gcj c •H
,-q •Hrt3.
4-1
/—\ G JG•H P
0 4-1 pi04
03r—1 +J Po p3 Gg 04 Gi—i o
+J •Hp G 4-1o pj •H
•H o G> •H bOPj 4H •HA •rH to0 c GCQ bO O
•H cCO
O 0•H 0 pG +-> GCu p3 O'—1 o •H
•H T ii—t T3 c0 G •H
h3 •H
tOS -
to CO CD CD C C -H
• H i—ii—I
gh3 0 •H ^ i-i O O f-t CO CQ
>- o
28
models tested with prior behavior included.
Exogenous variables, defined as variables whose causes lie outside
of the hypothesized model, are attitude (A), subjective norms (SN), and
prior behavior (PB) which is added in Figure 4. Endogenous variables,
defined as those whose causes are completely determined within the
hypothesized model, are intention (INT) and target behavior (TB) .
The models corresponding to the Fishbein-Ajzen (1975) formulation,
i.e., those which did not include prior behavior data are labeled FA-1,
FA-2, and FA-3 (see Table 1). The parameters are schematically
presented in Figure 3. These models are differentiated as follows:
Model FA-1, saturated model (all possible paths estimated);
Model FA-2, deletion of path from SN to TB (^22 *
Model FA-3, deletion of path from SN to TB (y22) an( P ath
from A to TB (y2i) •
In the analyses of these models the path from attitude to intention
(y-1) was consistently significant. Also consistent with expectations, * 1 1the path from intention to target behavior {$21) was beyond the .0005
probability level in its critical value. When a path from attitude to
target behavior (y2 i) was tested in Models FA-1 and FA-2, this parameter
was not significant. This finding supports the Fishbein-Ajzen model in
contrast to Bentler and Speckart’s modification.
Contrary to the predictions of both theoretical approaches, the
path from subjective norms to intention (y ^ ) did not achieve a sig
nificant level. However, the relationship (covariation) between sub
jective norms and attitude (^j) was significant (t-values from 5.199
to 5.278, all ps < .00006).
29
Table 1
Structural Model Specifications
Fishbein-Ajzen Models (Figure 3): k, SN^ INT^ TB
FA-1:®21} Y i r Y 1 2 * Y 2 1 9 Y 22
FA-2:^21’ Y n > y 12’ Y 21
FA-3: 3 21/ Y n > Y 12
Bentler-Speckart Models (Figure 4): A,, SN, PB, INT, TB
BS-1: 321j Yll^ y 12’ Y13* Y23BS-2: 321' Yi r y12’ Y13j Y23BS-3: 321* Yi r y 12’ Y13j Y23BS-4: 321> y 11# y12’ Y13BS-5: 321’ yn j Y12BS-6: 321’ Yn * yl2} Y13* Y21BS-7: 321' Yi r Y12j Y21
30
Prior behavior data (PB) were included in Models BS-1 through
BS-7. The parameters of these models are presented schematically in
Figure 4 and are differentiated (see Table 1) as follows:
Model BS-1, saturated model (all possible paths estimated);
Model BS-2, deletion of path from SN to TB
Model BS-3, deletion of paths from SN and A to TB (Y22 and Y 21^ ’
Model BS-4, deletion of paths from A, SN, and PB to TB
^21* Y 22* and Y 23^;Model BS-5, deletion of paths from A, SN, and PB to TB;
and of path from PB to INT (y2 i> y 22’ y 23* Y 13^ ’
Model BS-6 , deletion of paths from SN and PB to TB (y22 and Y 23^'
Model BS-7, deletion of paths from SN and PB to TB, and of
path from PB to INT (y 2 2 > Y 23> and Y 13) •
Tables C5 through Cll in Appendix C present the measurement and
causal model parameter estimates for these BS-series models. In order
to facilitate comparisons of the significance levels of the parameter
estimates between models, t-values of estimated paths and of the
estimated residuals of intention and target behavior for all models
tested are presented in Table 2.
Of the models in which a path from attitude to target behavior
(y 2 i) was tested, a significant parameter estimate was attained only
once and that was in a model which excluded all paths from prior
behavior (Model BS-7). This finding supports the Fishbein-Ajzen model
as does the finding that the path from attitude to intention (Y-q) was
significant (maximum £ - .0250) for all models tested.
Where paths from prior behavior were estimated, parameter estimates
31
cd
CDi-HXcdH
C/30i-HnS>ii
0Og00p0X ■C/3GOC/3•H
I-OUXPccjCl,
TJ0 0•3c •3c •3c •3c i-H i-Hr- •3c •3c •3c He ■3c •Hi r—1 03 *d" r- O LO cd cdCO O 1 H 1 1 r- CM CM P PCQ • • 1 • 1 1 • • • t ivD CM to rH -d- to o o1p p^^ — y
* * ■3c ■3cVO •3c He He He •3c \D 001 'Cj' rH r- LO 00 LO d 03 LOCO to CM o vD 3 1 LO CM • «■CQ • • • • 3 1 • • • rH CMCM rH LO 'd- •d to A A
0 0•3c ■3c •3c •3c ■3c o o•3c •3c •3c •3c •3c C GLO O 00 00 O LO cd cd1 to 03 ■<d- o CM o aCO • • 1 I 1 1 • • • •H *HCQ vO CM 1 3 1 1 vD to to <4H <4-41 •H *Hcn & &i-H •rH ~*H0 •3c He •3c He tf) tf)T3 -d" •3c He •3c He HeO 1 \ D \0 ■'d- OO to \ D t+H t+HS CO rH CM 1 1 1 r- ■'d- rH o oCQ • • • 1 1 1 • • •to CM rH LO vD d to rH rHCQ 0 0
> >0 0•3c •3c He •3C rH rHto •3c •3c •3c •3c1 rH VO 00 O ■<d- CM to LO rHco oo 00 CM 1 1 O 03 O LO O OCQ • • • 1 1 • • • • • •CM 'd- \ D LO rH ■3c -3c■3c
He •3c •3c HeCM •3c •3c ■3c ■3c1 CM LO O LO \ D 03 toCO 00 oo to CM 1 03 vD 03 LOCQ » « • « 1 • • • •CM ■<d- LO d rH
tf)•3c •3c •3c 0 *r—1 •3c •3c •3c •3c G tf>1 r- LO LO LO o LO 00 r- rH 0CO CM 03 CM LO CM vD LO 03 CM cd PCQ • * • - • • • • • > cdCM ■<d- rH LO d rH •Hp >00 T3He ■3c •3c •3c tnCO •3c * -3c He He <D i”H1 *d" rH CM LO oo rG cd
< O to 3 1 1 1 LO rH to p SPL, • ■ 1 1 1 1 • • • pLO 1 LO LO to - ovD CC/3 totf) 0 CM X)i-H 3 p0 He •3c rH •3c •3c Cp cdCM He •3c cd •3c He O T?o 1 LO to ■'d- 'd- > LO d G< o> CM 1 03 1 1 i rH to 0 cdPP • » i • 1 1 • P • • N p< 'Cj' 3 to LO to •H tf)PL, tf)
tf)C/3 0 cd0 »-H•3c He O •3c He P h-GtH •3c •3c G •3c •3c g 01 r- CM to CM cd r- 03 cd P< oo rH l to rH 1 v£> •H rH to tf) 0PL. • • i • • 1 Eh • • Eh'Cj' 1 rH rH to cd LO to 0 P1 > X PP 0i—i Pcd o GE- E-h CQ 3 P -HE- 2 2 CQ CQ E- TJX 2 t-H I-H CQ H H + •rH 0 0p T—1 t— 1 CM f to + rH H CM f to f rH E-1 c/3 E- G Pccj rH -f rH Z rH CQ CM + CM 2 CM CQ CM Z 0 2 CQ Q cdCL, ?*- < CO CL, >- < > - CO >- Cl, CO. I-H OS I-H E- cd
32
were consistently significant both to intention (jd < .00006) and to
target behavior (£ < .00006). These findings support one of Bentler
and Speckart’s hypothesized modifications.
No path was hypothesized from subjective norms to target behavior,
therefore the finding of nonsignificance for this parameter estimate
^22^ saturatec* m odels was expected. However, contrary to expecta
tions hypothesized by both the Bentler-Speckart and Fishbein-Ajzen
models, the path from subjective norms to intentions (Y-j achieved
significance (£ < .05) only in Models BS-5 and BS-7. These were models
which excluded all paths from prior behavior, a model specification
consistent with Fishbein and Ajzen's formulation. Interestingly, this
path (Y-j^p was n°t significant in Models FA-1, FA-2, or FA-3 which did
not include prior behavior data.
Model BS-5 was a formulation of the Fishbein-Ajzen model which
included prior behavior data in the measurement model only. That is,
the paths estimated were from attitude and subjective norms to intention
only, and just from intention to target behavior. In Model BS-5 all
causal paths estimated were significant; but in looking at the relative
strengths of the paths, it is apparent that the path from subjective
norms to intention (Y-j^* t-value = -2.984, £ = .0028) did not achieve
the level of significance reached by the path from attitude to intention
(Yu * t-value = 6.302, p < .00006) or by the path from intention to
target behavior t-value = 6.480, £ < .00006) (see Table C8 ) .
Model BS-7 differed from Model BS-5 only in the addition of a
path from attitude to target behavior (Y2 j) anc was a test of Bentler
and Speckart's (1979) first model modification. In this model also,
33
comparison of t-values and their associated probabilities for the
estimated path parameters (presented in Tables CIO and Cll) shows that
the path from subjective norms to intention ^ _va^ue = "2.086,
£ = .0366) did not achieve a level of significance as high as that
reached by the path from attitude to intentions (Y-^* t> value = 6.409,
£ < .00006) or the path from attitude to target behavior il2 1 * Ji”value =
3.145, £ = .0016). However, a strong relationship between subjective
norms and attitude ( ^i^ was demonstrated in all models (t-values >
4.956, £ < .00006). Also, in Model BS-7, when attitude was allowed a
direct path to target behavior (Y2i)> t le P at^ from intention to target
behavior (f^P failed to reach the .05 level of significance (t>value =
1.770, £ = .0768).
The relationship between attitude and prior behavior ($3^) and
between subjective norms and attitude (^2 1 was uniformly high in all
models Qp < .00006). The relationship between prior behavior and
subjective norms (^3 2 was nonsignificant (£ > .05).
Estimates of the path from intentions to target behavior (^2 1
were extremely variable. In those models which did not include prior
behavior data (Models FA-1, FA-2, and FA-3), this path was consistently
significant (£ < .002). However, when alternate paths to target
behavior were tested, this parameter decreased in value. In the models
which included prior behavior data (BS series models), the parameter
estimate for this path (f^ ) reached significance (t-values > 4.00,
p < .00006) only in models which deleted all other paths to target
behavior (Models BS-4 and BS-5) and in Model BS-6 which included a path
from attitude to target behavior (Y2 P * Apparently, the value of this
34
parameter is inversely related to the availability of alternate paths
to target behavior and the inclusion of prior behavior. In other words,
when other variables are included in the model, intention is a less
influential determinant of target behavior than the other variables.
In summary, when the t-values of estimated path parameters are
compared, the only paths demonstrating consistent significance are
those from prior behavior to intention Cy-^* £ < .00006), from prior
behavior to target behavior (Y2 3 > £ < *00006), and from attitude to
intention (y^ j £ < .023). Also consistent was the finding that the
path from subjective norms to target behavior (Y2 2) was nonsignificant
(£ > .05) whenever it was tested. All other paths varied in levels of
significance depending on the inclusion or exclusion of other variables
in the model. This instability suggests a need for model re-specification
to attain consistency of results.
Inspection of the t-values for the residuals of target behavior
reveals that when prior behavior is not permitted a direct path' to target
behavior (Models BS-4, BS-5, BS-6 , and BS-7), this residual increases
from a nonsignificant 1.27 or 1.53 to a significant level (£s < .003).
The addition of a path from attitude to target behavior without a path
from prior behavior to target behavior (Models BS-6 and BS-7) did not
result in an insignificant ^t-value for this residual. The t-values for
the residual of intention were consistently significant (all £s < .003).
Chi-Square Goodness-of-Fit Tests
The chi-square Goodness-of-Fit test is a test of the model's fit
to the observed data, that is, the variance unaccounted for by the model.
Table 3 presents the chi-squares, degrees of freedom, associated
Chi-
Squa
re
Good
ness
-of-
Fit
Test
s
35
oo IN- r -o rH CN
CM oo■Tf to• • •
rH rH rH
o O orH rH
V V V
04 04 o4V V Vo o oCM CM CN
■Tf- LO nOrH rH rH
CN \ D o>T“l LO tocm r— CN CN^NrH to rH CQ# . . E-O rH CNCN CN CN *\E-<2CQ I-H
H *\CQE-* C l,
2HH •s2•s CO
2co •s<
< C
/—\ ■Tf
t— \ CNto CN <D
> - Pa» 0u *\ oo3 rH rH •H00 CN CN tL*•rH > > - v— 'U h'— ' COCN CN CN rHCO r“1 r-1 rH CDI— 1 > - > - >- -d<D oT3 SO
S>- u
o cd<D *\ •stsl r“1 r“1 o
•i—> CN CN CN CD< CO. ca ca O'C/3
1o•H O<D rH CN to CDJO 1 1 i t-HX < < C pCO P-, P-, tu e•H CDCL, CQ
rH rH nO LO o 03d" oo LO rH O \D o03, oo d" 00 LO LOLO LO LO OO to 03 CN• • • • • • •rH rH rH CN to CN to
LO LO LO rH i-H r—Io o o o o o oo o o o © o ©• • • • • • •
04V V V V V V Vpj eg Ph| 04 P-.I Ph| Ph|
*4-1 Id" LO nO 00 nO f'-T3 [to to to to to to to
nO nO oo d" CN 03 00,o to rH d" d"° oo d" rH to rH oo|CN LO NO NO 03 d" CN• * • • • • •d" LO LO NO NO NO oLO LO LO o CN O CN
rH rH rH rH
CNCN>-
CNI CNI
>- >-•sto to to rHCN CN CN CN
>- >- >- >-•s •S •s
to to to to to r HrH rH . rH r H CN
>- >- >- >- >- >-•s n •sCN CN CM CM CN CN (NrH T— 1 T“ l T“ l rH rH
>- >- >- >- >- >- >-*\ •S *\
>- >- >- >- >- >- >-•s •s *\ •srH rH rH rH rH rH rHCN CN CN CN CN CN CNCO. CO. CO. CO. CO. CO. CO.
rH CN to d"• i
LO NOco C/3 C/3 co
1CO co
1co
CQ CQ CQ CQ CQ CQ CQ
36
probabilities, and ratios of chi-square divided by the degrees of
freedom for each model.
The x 2 statistic provides a test of the proposed model
against the general alternative that the MVs (measured
variables) are simply correlated to an arbitrary extent.
If the x 2 is large compared to degrees of freedom, one
concludes that the model does not appropriately mirror
the causal process that generated the data. (Bentler,
1980, p. 428)
A nonsignificant chi-square value, then, supports the hypothesis that
the model provides a plausible representation of the causal process,
that is, the chi-square indicates whether or not the factors specified
in the confirmatory factor analysis extract sufficient variance so that
the residuals are nonsignificant. The ratio of chi-squared divided by
its degrees of freedom is also an index of goodness-of-fit, with a
better fit being indicated by a smaller ratio. There is no associated
significance test for this ratio, however.
A problem that arises from the exclusive use of the chi-square
Goodness-of-Fit test for the evaluation of a model's fit to the data is
that the chi-square variate is a direct function of sample size and the
number of parameters estimated. Consequently, with large samples (this
analysis is not appropriate for small samples) and a large number of
parameters to be estimated, this statistical test would result in the
rejection of virtually all models. That is to say that the appropriate
statistical conclusion would be that the residual matrix contains sig
nificant additional information that could be explained by a better model.
37
While the value of x 2 depends on sample size, the associated degrees
of freedom are only determined by the number of variables and
hypothesized factors in the model.
The results reported in Table 3 show that the chi-squares for"'all
FA series models attained a similar nonsignificant level of probability
(.20 < £ < .10). The chi-squares for all BS series models, which
included more variables along with their associated parameters and
consequently additional measurement error, were at less than the .005
level of significance while the sample size was the same for both series.
According to Bentler and Bonett (1980), one method of addressing
this problem is to inspect the absolute values of residuals which
provide an estimate of the amount of statistical information extracted
from the data. (This information on residuals is presented in Table 2.)
However, a "key ingredient" in appropriate statistical methodology for
comparisons of causal models in their view is the use of hierarchical
(nested) models to provide a chi-square difference test between models.
The primary use of the information presented in Table 3 is for the
computation of these chi-square difference tests between hypothesized
models. The results of these difference tests are presented in Table 4.
Chi-Square Difference Tests
The chi-square difference test is based upon the rationale that the
difference between chi-squares is also distributed as chi-square with
degrees of freedom equal to the difference between the number of para
meters estimated by each model. This statistic is used to test the
importance of the parameters that differentiate the models and to assess
the relative adequacy of the models in explaining the observed data.
38
Table 4
Chi-Square Difference Tests for Model Comparisons
Fishbein-Ajzen Models
FA-1/FA-2
FA-1/FA-3 1
FA-2/FA-3 1
Bentler-Speckart Models
BS-l/BS-2 1
BS-2/BS-3
BS-l/BS-3 1
BS-3/BS-4 50
BS-4/BS-5 20
BS-4/BS-6
BS-6/BS-7 13
BS-3/BS-5 71
BS-3/BS-7 64
xi df £
.7983 1 ' .50 < £ < .30
.9527 2 .50 < £ < .30
.1544 1 .30 < £ < .20
xi df £
.3830 1 .30 < £ < .20
.0582 1 .80 < £ < .70
.4412 2 .70 < £ < .50
.9726 1 £ < .001
.3198 1 £ < .001
.1995 1 .70 < £ < .50
.8699 1 . £ < .001
.2924 2 £ < .001
.6430 1 £ < .001
39
The null hypothesis appropriate for these tests is one of model
equivalence. The chi-square differences, degrees of freedom, and
associated probabilities for the model comparisons are reported in
Table 4.
For the models which did not include prior behavior data (FA
series, see Table 4), the null hypothesis of model equivalence cannot
be rejected. Essentially, this means that the model with a path to
target behavior from intention only (Model FA-3) is equivalent to
models with additional paths to target behavior (Models FA-1 and FA-2).
Thus, with prior behavior data excluded, the Fishbein-Ajzen model is
supported as the most parsimonious and adequate explanation of the
observed data.
When prior behavior data are included (BS series models, see
Table 4), the null hypothesis of model equivalence cannot be rejected
for Models BS-1 (the saturated model), BS-2 (path from SN to TB, y
deleted), and BS-3 (path from A to TB, y a l s o deleted). These results
support the model with paths to target behavior from prior behavior and
intention, but not from attitude and subjective norms to target behavior
as the best fit to the observed data. Comparison of Model BS-3 with
Model BS-2 which includes the attitude to target behavior path (Bentler-
Speckart’ s expanded model) does not support the Bentler-Speckart (1979)
model in that the attitude to target behavior path does not fit the
observed data significantly better than the model with this path omitted./
Comparison of Model BS-3 with Model BS-4 (paths to target behavior
deleted from all variables with the exception of intention, results
in rejection of the null hypothesis of model equivalence. This hypothesis
40
must also be rejected for comparisons of Model BS-3 with subsequent
BS series models as well, indicating the necessity of paths from prior
behavior to both intention and to target behavior (y-j^ and Y 2 3) • The
lower ratio for this model of y2 to degrees of freedom (1.5456), as
compared to the value of that ratio for the other models, supports the
conclusion that this model provides the best fit to the observed data
of the models tested. These findings support the Bentler-Speckart
(1979) modification calling for paths from prior behavior to both
intention and target behavior.
Because of an equal number of degrees of freedom, Models BS-3 and
BS-6 (no path from PB to TB, Y 2 3 > ^ut inclusion of a path from A to TB,
Y 2 1 ) cannot be directly compared by a y2 difference test. However,
comparison of the x2/df ratios of these models supports the better fit
of Model BS-3 to the data.
Discussion
Before further discussion of the results of this study, one caveat
is in order. It must be borne in mind that this study concerned only
a single behavioral content domain, class attendance, and a single
population, university summer school students. Replication, utilizing
other behavioral domains and different populations of subjects, is a
necessary prerequisite to hypothesizing changes in structural equation
models. This present discussion of results is directed toward future
directions for additional study in this area. In considering behavioral
domains, it should be noted that Bentler and Speckart, in their most
recently published study in this area (1981), found different results
for different behavioral domains among the same subjects.
41
To address the initial research question posed by this study,
comparison of the Fishbein-Ajzen (1975) model with Bentler-Speckart’s
(1979) modifications, the results of this study lend support to the
Bentler-Speckart hypothesis of direct paths from prior behavior to both
intention and target behavior. However, the results do not support
their model modification of a direct path from attitude to target
behavior.
Another area for additional research suggested by these findings
is a test of a respecification of the causal model with subjective norms
and prior behavior acting upon (causal to) attitudes along with
hypothesized direct effects of prior behavior on the other endogenous
variables of intention and target behavior. The finding in this study
that the path from subjective norms to intention (y ^ ) achieved the .05
level of significance in only two of the models tested was contrary to
both the Fishbein-Ajzen (1975) and Bentler-Speckart (1979) model
predictions. The Fishbein and Ajzen formulation does allow the weights
of attitude and subjective norms to vary with the type of behavior,
,with the context or situation in which the behavior is to be performed,
with the target, and with individual differences between actors. But,
while the hypothesized path (causal relationship, was nonsignificant,
subjective norms and attitude exhibited a strong relationship (cor-4
relation) in all models (minimum t-value of cj^ of 4.956). This
covariation suggests the possibility of a respecification of the model
based upon the notion that social influences are indirect determinants
of intention through their effect on attitudes. Thus, a reasonable
respecification of the model would be a test of subjective norms prior
42
to attitude in the causal model specifications.
The most recently published Bentler-Speckart (1981) study found
support for a model which placed intention as an equal determinant of
target behavior along with attitude, subjective norms, and prior
behavior rather than as a mediating variable. While this modification
of the original model is different from the hypothesis suggested above,
it too supports the need for additional research on variants of the
Fishbein-Ajzen model.
Likewise, the relationship between prior behavior and attitude
^ 3 1 ^ was extremely strong in all models which included prior behavior.
A reasonable respecification would be the placement of prior behavior
causal to attitudes as well as to intention and target behavior in a
model respecification. As mentioned earlier, the instability of some
path parameter estimates, notably the path from intention to target
behavior seems to indicate the lack of satisfactory model
specifications. In model BS-3 which demonstrated the best overall
goodness-of-fit (y2/df), the t>value for the estimate of this path
^21^ WaS nons Snificaivt •Another possible explanation of the nonsignificant subjective norm
path might lie in the behavioral domain of the target behavior, class
attendance during a five-week summer session. The development of
group identification and group cohesiveness with classmates for such a
brief period is obviously minimal. This leaves as primary sources of
social influence significant others not directly involved in the target
behavior itself, such as parents, spouse, and friends. It is con
ceivable that under such conditions, these social influences may have
43
generalized to a broad spectrum of attitudes toward academic behavior
in general. Ajzen and Fishbein (1980) presented consistent findings
derived from a variety of target behaviors which included voter
behavior, consumer behavior, and family planning behavior. However,
Bentler and Speckart (1981) found that in the same sample, three
behavioral domains, e.g., exercise, studying, and dating, led to dif
ferent results with respect to path significance. Again, additional
studies across behavioral domains are indicated.
A third possible explanation might reside in the samples used in
the pilot and primary studies. The "significant others" of the study
questionnaire were determined by means of a pilot study on a sample of
summer school students at the same university who attended the session
immediately preceding the session attended by the subjects of the
primary study. These two samples did not differ significantly on 13
out of 15 demographic variables (see Table Cl in Appendix C ) . The two
variables in which significant differences were observed were year of
graduation and economic status as measured by parent's income. The
primary study sample was composed of students who reported an earlier
date of graduation and a greater proportion of whose parents had lower
income levels. Possibly, students who are closer to graduation per
ceive the instrumentality of class attendance in attaining the goal of
graduation differently from students for whom the expectancy of gradu
ation is not so immediate. For them, the social influence might be
overridden by the perceived instrumentality reflected in attitude which
was measured by the consequences of the behavior multiplied by an
evaluation of these consequences. Also, the difference in economic
44
status between samples might have led to the omission of relevant
"significant others" for consideration in the questionnaire used in
the primary study.
The question of generalizability requires attention at this point,
for the value of theory lies in its generalizability. If separate
theoretical formulations were required for each behavioral domain
investigated, the usefulness of such theory would be questionable.
It is the ultimate goal of theory-testing not only to provide empirical
support or disproof, but to define the limits of generalizability.
The behavior of interest in this study was classroom attendance,
a behavior that in its own right is of proper concern to the educational
community. Attendance behavior is likewise of prime interest to
organizations in the industrial, governmental, or service communities,
where successful and efficient operation depends heavily upon the
presence of organization members or employees.
The practical significance of identification of those variables
which can ultimately affect this behavior is obvious. For example, if
group norms were significant predictors of this behavior, then organi
zational interventions targeted at the development of group cohesiveness
and desirable norms might be very appropriate. If prior behavior were
found to be most significant, as this study indicates, interventions
such as new-employee supervision, indoctrination, and prompt attention
to the first indications of attendance problems might forestall the
establishment of a pattern of poor attendance behavior.
A population of students with class attendance as a target
behavior is not the same as a population of workers with work attendance
45
as target behavior. However, the basic interrelationships between vari
ables hypothesized by Fishbein and Ajzen have been found to possess
wide generalizability. So it seems that an initial study applying this
model and variants of it to the target behavior of attendance could
serve as a reasonable preliminary step toward the more general applica
tion of the model to this behavioral domain.
46
1. Pedhazur, E.
Reference Note
J. Personal communication* December 8, 1980.
47
References
Ajzen, I., § Fishbein, M. Attitude-behavior relations: A theoretical
analysis and review of empirical research. Psychological Bulletin,
1977, 84, 888-918.
Ajzen, I., § Fishbein, M. Understanding attitudes and predicting
social behavior. Englewood Cliffs, N.J.: Prentice-Hall, 1980.
Bern, D. J. Self-perception: An alternative interpretation of cognitive
dissonance phenomena. Psychological Review, 1967, _74, 183-200.
Bern, D. J. Self-perception theory. In L. Berkowitz (Ed.), Advances in
experimental social psychology. New York: Academic Press, 1972.
Bentler, P. M. Multivariate analysis with latent variables: Causal
modeling. Annual Review of Psychology, 1980, _31, 419-456.*
Bentler, P. M . , § Bonett, D. G. Significance tests and goodness-of-fit
in the analysis of covariance structures. Psychological Builetin,
1980, 88, 588-606.
Bentler, P. M., § Speckart G. Models of attitude-behavior relations.
Psychological Review, 1979, 86 , 452-464.
Bentler, P. M., § Speckart, G. Attitudes "cause" behaviors: A
structural equation analysis. Journal of Personality and Social
Psychology, 1981, 40 , 226-238.
Billings, R. S., § Wroten, S. P. Use of path analysis in industrial/
organizational psychology: Criticisms and suggestions. Journal of
Applied Psychology, 1978, ^3, 677-688.
Calder, B. J., § Ross, M. Attitudes and behavior. Morristown, N.J.:
General Learning Press, 1973.
48
Calder, B. J., § Ross, M. Attitudes: Theories and issues. In J. W.
Thibaut, J. T. Spence, § R. C. Carson (Eds.), Contemporary topics
in social psychology. Morristown, N.J.: General Learning Press,
1976.
Cook, T. D., § Campbell, D. T. Quasi-experimentation design 5 analysis
issues for field settings. Chicago: Rand McNally College
Publishing Co., 1979.
Eagly, A. H., fi Himmelfarb, S. Attitudes and opinion's. Annual Review
of Psychology, 1978, 29, 517-554.
Festinger, L. A theory of cognitive dissonance. Evanston, 111.: Row,
Peterson, 1957.
Fishbein, M. Attitude and the prediction of behavior. In M. Fishbein
(Ed.), Readings in attitude theory and measurement. New York:
Wiley, 1967.
Fishbein, M . , £ Ajzen, I. Belief, attitude, intention and behavior:
An introduction to theory and research. Reading, Mass.: Addison-
Wesley, 1975.
Heider, F. Attitudes and cognitive organization. Journal of
Psychology, 1946, 21_, 107-112.
Joreskog, K. G., £ Sorbom, D. LISREL IV: Estimation of linear
structural equation systems by maximum likelihood methods. Chicago:
National Educational Resources, Inc., 1978.
Kelman, H. C. Attitudes are alive and well and gainfully employed in
the sphere of action. American Psychologist, 1974, 29, 310-324.
Kerlinger, F. N., § Pedhazur, E. J. Multiple regression in behavioral
research. New York: Holt, Rinehart and Winston, 1973.
49
Li, C. C. Path analysis: A primer. Pacific Grove, Calif.: Boswood
Press, 1975.
Maruyama, G., § McGarvey, B. Evaluating causal models: An application
of maximum-likelihood analysis of structural equations. Psychological
Bulletin, 1980, 87_, 502-512.
Namboodiri, N. K., Carter, L. F., $ Blalock, H. M., Jr. Applied
multivariate analysis and experimental designs. New York: McGraw-
Hill, 1975.
Osgood, C. E., Suci, G. J., § Tannenbaum, P. H. The measurement of
meaning. Urbana, 111.: University of Illinois Press, 1957.
Rokeach, M. Beliefs, attitudes, and values. San Francisco: Jossey-
Bass, 1968.
Triandis, H. C. Attitude and attitude change. New York: John Wiley
§ Sons, 1971.
Triandis, H. C. Values, attitudes, and interpersonal behavior. In
H. E. Howe, Jr., § M. M. Page (Eds.), Nebraska Symposium on Motivation
(Vol. 27). Lincoln: University of Nebraska Press, 1980.
Appendix A
Questionnaire for Pilot Study
51
Informed Consent for Participation in a Research Project
Your participation in a research study of student attitudes being
conducted by Arlene Fredricks of the UNO psychology department is
requested. Participation involves filling out a questionnaire during a
class meeting on the subject of some of your attitudes and providing
some personal information about yourself.
All information will be confidential and the anonymity of youri
responses will be guaranteed. Your responses will not be identified to
your instructor/professor.
Your decision on whether or not to participate in this study or to
withdraw from the study at any time will in no way prejudice your
relationship with the instructor or the university.
Your signature on this consent form indicates your willingness to
participate in this study and authorizes the use of the information
collected along with classroom data for research purposes only. There
are no hidden conditions or manipulations involved and you are free to
withdraw from the study at any time. I will be happy to answer any
questions you might have on this project and appreciate your cooperation.
Arlene Fredricks, 554-2704 or334-1177
Signature__________________________________________________________________________
Date
Investigator
Social Security Number __________________
Sex _________________ Age
Marital Status: Single ___Married ___Separated ___Divorced ___Widowed
Student Status:Number of credit hours you are currently taking. ____
Number of credit hours you expect to take in the fall.
Expected date of graduation (if you are in a degree program).
Check here if you are not in a degree program. ______
Employment Status:Average number of hours that you work per week currently. __
Average number of hours that you expect to work per week this fall.
Race: CaucasianBlack SpanishAmerican Indian Asian
Parents’ Approximate Annual Income (please check your best estimate)Don’t know________________________Less than $10,000 ___Between $10,000 and $14,999 ___Between $15,000 and $25,000 ___Over $25,000 ___
Expected grade in this class:
A > 0 , F
GPA: 4.003.90-3.993.75-3.893.50-3.743.00-3.492.50-2.992.00-2.491.50-1.99 Less than 1.50
Citizenship: U.S.
Other
53
Briefly list the beliefs that come to mind when you are asked the following questions. If you run out of space for your answers, use the back of the sheet and indicate the question number.
1. What do you believe are the advantages of attending your summer school class(es) every day for the summer session?
2. What do you believe are the disadvantages of your attending your summer school class(es) every day for the summer session?
3. What do you believe are the advantages of your being absent fromyour summer school class(es) once? Several times? One day a week? Twice a week? More than twice a week?
54
4. What do you believe are the disadvantages of your being absent from your summer school class(es) once? Several times? One day a week? Twice a week? More than twice a week?
5. What else do you associate with your attendance and/or absence from this summer school class?
6. Who are the individuals or groups of people whose opinions orinfluence is important to you with respect to your attendance and/or absence from your summer school class(es)?
Appendix B
56
Informed Consent for Participation in a Research Project
Your participation in a research study of student attitudes being conducted by Arlene Fredricks of the UNO Psychology Department is requested. Participation involves filling out a questionnaire either during a class meeting or at home on the subject of some of your attitudes, beliefs, and intentions about attending or not attending classes and also in providing some personal information about yourself.
All information will be confidential and the anonymity of your responses will be guaranteed. Your responses will NOT be identified to your instructor/professor. They will be used for research purposes only.
Your decision on whether or not to participate in this study or to withdraw from the study at any time will in no way prejudice your relationship with the instructor or the university.
Your signature on this consent form indicates your willingness to participate in this study and authorizes the use of the information collected along with classroom data for research purposes only. You are free to withdraw from the study at any time. You are also free to omit any questions that you do not desire to answer, but it would be appreciated if you would answer all questions on the information sheet and on the attitude questionnaire.
After collection of all the data, the entire study and how the data that you have provided is to be interpreted and used will be explained to you.
I will be happy to answer any questions that you might have on this project and appreciate your cooperation.
Arlene Fredricks, 554-2704 or(home) 334-1177
Psychology Dept., 554-2592
Signature
Date
Investigator
57
INSTRUCTIONS FOR QUESTIONNAIRE ADMINISTRATION
You are being asked to participate in a research study which involves student attitudes and opinions about attending or not attending class.As the Informed Consent Form indicates, participation involves providing some information about yourself, filling out a questionnaire about your beliefs and intentions concerning attending or not attending class, and authorizing the use of classroom data about yourself. If there are any questions that you do not wish to answer, you may leave them blank. However, it is important to the study to collect as much of this requested information as possible and it would be very much appreciated if you did answer all the questions.
It is vital to the study that your responses be identified by your Social Security Number. However, your responses will NOT be identified to your instructor and will be used for the research purposes of this study ONLY.
After all the data have been collected, you will be informed of the entire scope of the study and the methods used and of how the data that you have supplied will be interpreted and used.
I will be happy to answer any questions that you might have on this study. My phone number is on the Informed Consent Form (334-1177) or you may contact me through the UNO Psychology Department at 554-2592.
Thank you very much.
Arlene Fredricks
PLEASE NOTE: IF ANY OF THE QUESTIONS ABOUT OTHER PEOPLE'S OPINIONS ORYOUR DESIRE TO COMPLY WITH THE OPINIONS OF THESE OTHERS (SUCH AS HUSBAND, WIFE, OR FIANCE) DO NOT APPLY TO YOU, PLEASE LEAVE THE RATING SCALE BLANK FOR THAT QUESTION AND WRITE N/A (NOT APPLICABLE) AFTER THE QUESTION.
The
first
page
of the
Qu
esti
onna
ire
asks
for
pers
onal
in
form
atio
n.
The
question
on pa
rent
s' annu;
income
asks
for
your
best
estimate
or,
if you
have
no idea,
there
is a
place
for
"Don't
know."
If yo
uhave
been
out
of school
for
some
time
and
are
now
retu
rnin
g and
consider
that
question
irre
leva
nt for
you,
you
may
change
it to
"My
annual
income"
and
answer
acco
rdin
gly.
58
i-i so3 OP i—ip •H i—i 0 ndo3 • • o > 0e PCD co5 <p •HP Picopi X O CO o3 o3o P I— 1 o3 30 o3 rH t-H 0PS CD o Pi o3 U XO 5 <p Po3 >0 02: i-H0etf CD to C PiPi 03 P 3 CO •H0 E-* nd Td P T3 0 O t-HPi o3 Pi o3 3 o3 p X o Co3 Pi P Pi O Pi p i-H 3B CDP X 03
B X X X o 4h• H t-HoQ o3 t—l t—H Cl) t-H CO UhP P 0S P3 0B
o03 0B
o3 •s0 CO0 O 0 O 0 i—t 0 i-H i-H o3P P p X p Ph p i-H Pho3od pX 0 pX Td pX 05 i Po33 CD 0 o 0 t-H 0 X 0O Pi •« o3 .. 3 •. CO 0 Ph • •X coo3 # # i-HPH O o3u §•• </) 0 0 0 co o10 CD p od p 3 p P Ph Td0 P o •H t-H •H O •rH p t-H
O CD i-H 3 3 3 X 3 •H 3o3 £ i-H cr O cr cr o • Oi—t o 5 c Ph Xp , 3 <p 0 nd 03O • • 3 . • p .. C S P • •e<u X COo3 X Ox X p X 0 0 •H> m i—t t—H •N i-H X »N0 • H od p p p od P i-H 0CO •N 0p rPtoO 0p: bO ' o3 Pi
pbO 0P i CO i-Ho3p j CD 0 •P p •H •rH •H 0 Op t-H P i—1 i—1 p i—1 t-H p : CO•H P H O h CO •N CO o CO e p5 P0 • • odo • • p • • 3i p o3 • •CO X P p o p od p X •H p :CD CD P 0 bO 0 o 0 i-H oi— 1 •H p: p : o p 0 od 3o3 P p X p too X p Pi 0 COCJ O 0 •H t-H •H •rH •H Pco tin Pi 0P 0
B0e p0 0c i-H 0P Potoo *rd 0 p p : PhP . t-H • • p • • p • • p P • ••H P £ p X •H •H 0 •P O O X X 0 is S P Xo3 • H rC r—H 0 t-H 3 i-H P f“HP P•HO h
COco
Pp:bO CO• H4->pbO CO•H
PpbO0t-Ho3
•H0
3•“i Xo o 0 •H •H •H O Pi P t—Ho r—H o3 t-H o3 i-H CO •H 3CD p o3 CO p: CO p co O *-iCO 0 i—t •« o3 .. o3 . . too P P • •3 o PH <§ g e O CX o •H 0 p : •rHCD P 0 0 0 P PPi co 0 P C p c P o3 03 CO Po3 CD > •H •H •H •H •H P •H oB 0 3 3 3 CO p :• H co • • cr P • • cr p . • cr o3 0 03CO P CO 0 CO 0 CO i-H p : COe O 0 • H p •H p •rH bO o3 o3 •rHo CO p : . • p .. p •. C O g • •• H CD p o3 X o3 03 X o3 03 X •H CO o o3P od .O t-H 0 p : f—i 0 p H CO p :co n o3 0 £ o3 0 £ o3 0 3 0 P o3CD p 0 B B g B g B co •H gCO i—t O 0 0 <5 X 0 0 <5 0 0 0 Qcr CD o3 P p: P p P Pi p : PPi O e P p p P p p P 0 PbO co • H X •H X •H X O p : • HP p 0 Pi 0 pi 0 co p• H o3 o3 p e P e p i—H • o3 Pp p : 0 Td • H 0 od • H 0 Td nj CO 0 0 X• H p p : p : o p : p : o p p o p 2: rC rHo3 o p o p p o p p o t—H •H P CD€ o 0 o3 toO o3 bO o3 bO t-H o3 0 o3CD o CO 0 3 0 3 0 •H Ph
0 f0 • HP o3t-H P S OX £ OX S 2: S t—H
CD Ph o 0 0 0 3 > 0PSH 0p :p
p•H Omah
a" S <pHH S <pl— l £ OX •HPo0•r— kTd03 to ra
te p :H
extremely
quite
slightly
neither
slightly
quite
extr
emel
y
If you
think
it is
extrem
ely
likely
that
the
weather
in Omaha
is hot
in
July,
you
would
mark
the
scale
as fo
llow
s:The
weather
in Omaha
is hot
in
July
.likely
X :
: :
: :
:
unli
kely
extr
emely
quite
slightly
neither
slightly
quite
extr
emel
y
59
Social Security Number
Sex _________________ Age
Marital Status: Single ___Married ___Separated ___Divorced ___Widowed
Student Status:Number of credit hours you are currently taking.
Number of credit hours you expect to take in the fall. ______
Expected date of graduation (if you are in a degree program).
Check here if you are not in a degree program. ______
Employment Status:Average number of hours that you work per week currently.
Average number of hours that you expect to work per week this fall.
Race: CaucasianBlack SpanishAmerican Indian Asian
Parents1 Approximate Annual Income (please check your best estimate)Don’t know ___Less than $10,000________________Between $10,000 and $14,999 ___Between $15,000 and $25,000 ___Over $25,000 ___
Expected grade in this class:
A ____, B , C , D , E , F
GPA: 4.003.90-3.993.75-3.893.50-3.743.00-3.492.50-2.992.00-2.491.50-1.99 Less than 1.50
Citizenship: U.S.
Other
61
unpl
easa
nt
My at
tend
ing
this
class
every
day
will
enable
me to
learn
more
about
the
subj
ect.
%62
0ps
xrH0e0pPX0
0P•H3cr
p32w>•Hf pto
p32W>•Hto
0PS
3•HP0p3e0p3po0
T3333O•HP03epom3•H3ojPPo&•HtoC/3
OP
0£C/30t/33oj0
Cl 3 PS•i-tr—l33
0e0ppx0
3cr
pw>•HC/3
P032p•H03
W>•H
pH00
4-1opo
op0eC/30t/333OXerf
cl)PS•HipS
0e0ppx0
0p•H3cr
pxbo•Ht-H
in
p032p•H03
bO•H
bO3•HXT33pC/33•HT33•HJ0X
Oj4-1OP
0eC/30C/33ojO
Cl)PS
0eoppx0
0p•H3cr
32bO
PXbO•HrHin
Xp
0T303PbO
P0XW>•H32oj
p0bO
T33oj
P0PP0XoT3OP
0e0rHxoj30
Sf
0PS•HrP33
0eoppx0
0p•H3cr
p32tuO•Ht-H
xHp32bO•HipC/3
3 O ■ *H P 3 0 P O 0 PC/33Xo3C/3C/30•HP•H>•HPO3P032
3OP3o3 X 3 P
•rH *H6 E 3
O *4-1 P T3 0 3 £ 3C/3 30 T3t/333OX3
30■HP4H
C/3 TJ 3 *3 TJ E3 30 3 0 X 0 3 0 X 0 X 32p rH P p P H P P P P P•H o •H 0 •H o •H 0 •H 0 -H3 3 > 3 3 > 3 > £cr C/3 cr 0 cr t/3 cr 0 cr 0•H •H 032 t/3 32 t/3 3 32P • • to • • P • • t/3 • • 3X X 3 X X 3 X 3 OrH £ rH rH rH £ rH rH rH -p P0 o 0 o 0 O 0 o 0 0s p e E P E E 00 4-1 0 t/3 0 4-1 0 3 0 3 £p p •H P P •H P •H #Hp P p 32 p P p 32 p 32 PX 3 X P X 3 X P X P0 0 0 0 0 0 0 *\C/3 tuO 3 W> b/3 pi33 X 3 X 33 X 3 X 3 P3 rH •H rH 3 rH •H rH •H O0 T3 0 0 T3 0 33 £bO PS 3 PS bO PS 3 PS 33 •H 0 •H 3 •H 0 •H 0 O•H ip P ip •H ip p ip P P0 P 0 p p33 3 3) 3 3 0C2X X X X x - S
s s s s S P
# #vD 00 03 O
fl)PS•HH§
0e0ppx0
0p•H3cr
xw>
p032P•H03
P32W>
eoTJ0po3)X3OPo3O
•H330P
032
I£4-10•HrH0pT30T300332P•H■£0EC/30T3•H>OPPh
0PS•HrH33
>?0e0ppx0
•H3cr
32bO•H
P32bO•H
t-H
in• • 0
t/3 3t/3 •H0 3 P 0p rH 3 P
•H o O •H3 P 3r t/3•H32 XIpcr
• * X P •H3 • • XrH E 33 rH0 o 0E p 0 E0 4-1 33 0P P PP P pX 3 4-1 X0 0
t/3O 0
X 32 3 XrH 3 O ^H0 •H 0PS bO C/3 PS•H 3 3 •HrH •H0
32
£
0P
T333
ip
12.
My being
absent
from
this
class
is a
waste
of the
money
paid
for
tuit
ion.
63
P.X•HiH§
P£pppXCD
CDP•rH3a"
pXbO•Pi-HCO
"dpccj
pp•rHr—IP h
•rH0 to
•rH
1pI— tpto
X£top
•rHX
ccjXXTJ3
M•rH i—tp3
p£pPPXP
Pp•rHPcr
pC5bO•rH
PXpco•3ppp>oap3
ccjrP5POPo•rH
oPHp•rH
P■X•rH i—t
PP•rHPcr
xbO•rHr-Hto
popoppptop
•rH
PX
TJpccj
toPPP•PPPto
ppXPo
(I.)rX•rHi-HPP
PCCjPi
PP•rHPcr
.pbO•rH
bOP•rH£ooPhpX
pp•rHPcr
pxbO•rHi-Hto
to•rH
totoCCji-Hoto
•rH
Xpp
•rHP • < P . 1 p • • p too P o ccj rH
P P bO P ccj p tp i-H P ajP P P P p o a P •HX TJ X O X P tP X pp O p P p P p to to P p•rH o •rH •rH P •rH P •H •rH pP bO P P P •rH P o X P ccjP
toPh
P £
to
Ptop
P •rHPccj
Pt+n
P £
pX
OlH X piH rH X "d•rH X 3
HO X p
drH■MXb0 •rH
P>P
TJ
t-HPXbO•rH
•8
§
CtJ•rHPPP
rH■M, PbO•rH
>opPh
rH+->XbO•rH
aj>pu
PPPP3
i-HP, PbO•rH
puprH
••
i-Hto X
CtJTJ
X
••
i-Hto sf
"d
X
ccj£
Pccj
••
i-Hto X
ccj"d
X
••
i-Hto 3O
X
p
£
pop
i-Hto rd
Pccj
pP p p P tP P P P o •r-> P OP p p P P P P P < p X P •rH•rH > •rH > •rH > •rH 3 •rH PP p 3 P Ph. 3 P 3 to to 3 ccjcr
toto
crtoto
oto
crtoto
cr tXtoccj
PX
cr £po•. ccj •. ccj P • 1 ccj • • P Mh
X i-H X i-H o X r-H X to X Pi-H a r-H a •rH i-H a i-H p P rH •HP P P P P o 3 P£ to £ to ccj £ to £ •rH o £ PP •rH P •rH P P •rH P P rQ P PP X . P X ccj P X P to P P ccjp P £ p p i-H p p p P P pX P X P h X X 3 p X pP b£>
PPP
P bOP
XP P b£> . P
p cr pO
P oPh
X •rH to X •rH X •rH X m £ "d £i-H *3 P i-H "d i-H "d H o O •rHP p i P P p p P p bO oX p P Pi P ccj tX P rX p P bO bO•H p i-H •rH P •rH P •H p •rH Pi-H p
ccj
1 ?
Pto
X£
i-H PCtJ
XS
toptopp
i-H P3Xs
i-H to
pXPl-H
PPccjPrJ
•rHCOto•p
*3CCjPi
T3OObO
to LO PXE- CN
extr
emely
quite
slightly
neither
slightly
quite
extr
emel
y
64
P43
P43
PJO
mPPO•HPpxe43+->•H
Pcti43Pcti4=1
Pcti43Pcti43>
X X X 5 X X Xi-H r-H f-H r-H • • r-H i-Ho o o o o o oE E E 43 E •rH E Ea> a> o o o op p P O p o P pp p +-> P p p p pX X X X •rH X Xo o O o o P o os • • P • • • ••H o tn+-> p •rHo o • • O o o o+-> +-> m +-> P p X p p s•H •rH •H •rH P •rH i-H •rH •H op P P P •rH P p ocr cr O cr cr cr cr pm rd mp ttf) o• • P • • P • • O iX X oo X •rH X 434_> X X pr-Hi-H i"H f-H i"H i"H <DP +-> m +-> r-io p q_l p p to43 4P • rH 43 43 o 43 43bO bO r- bO bO bO bO•Hr-H •Hi-H P •rHr-H Po •rHr-H Po •rHr-H •rHi-H ptn m P in uh w •rH m in UJ
• • • • •rH • * • • in • • • • r\ • •o P op P <D P E P o P P po O P O •H o p o o •rH43 43 cti 43 P 43 43 43 r-H+-> +-> P +-> P nd P P &•H •rH bO •rH •rH P •rH •rH •rH<D • • a> O P o rt a> a> oP m P 43 P O p P P m•rH bO •H r\ •rH• • •rH • ' P • • E • • P • *X w
m X 43 X ctiO X o"■p X X 1Pi-HP ctii-Hr-HP ctJ r-H+-> P(j r-Hp o i-HP i-HP r-H<D43 (J 43 bfi 43 <D 43 o 43 43 mbO bO £ bO ttf) _Q bO bO•rH t/) •H •rH •rH •rH H—* •rH •rHi-H •H i-H 4-> r-H W r-H r\ i-H i-H min 43 m +-> to cti w X w in p• • +-> O • • • • p • • m •H • •bo 43 O •rH rOP O Po •rH o P o P o O a> P o 43p +-> P +-> m p P p o p
• • •H bo •rH cti •H •H O •rH p •H Xw P P P P in P E P tn P P•H cr •H , cr in cr o cr cr cr PX m •rH r\ S PPh P P go P r-H ♦ • •H • • • • in \ • •O X +-> X o X > X •rH X •rH X rXr-H r-H w i"H r-H •rH i"H P r-H i"H Pm o o 01 a> P o o a> P 4) oe P E •rH E o E p E EP o •rH CD 43 o ct* o a> 43 oO p P +-> P p a> p P P p+-> P +-> P P •• p 43 p p o+-> X P X P X o w X p X X X oO o •rH o •H o •rH O o o o bor-H 43 P E pP o p r-H H3 P X P o p o p bOcti O 43 O i-H O O 1-H o p o E o p oO O o O •H o P o O •rH ottf) ttf) bO bO *: bO bO B bo ttf) P bo & bOP p p c3 P o o•rH •H bO •H p a> •rH f-H
Ol i-H P tn ♦ rH P <Dtn rH •rH in P i-H •H >•H cti o •h c o P os Ph a od Q. • • . . #
to '3' LO \D r-- oo CO
+->43bO■H01
+->43bO
O+->•H35cr
10.
Getting
information
on what will be
covered
on the
tests
and explanations of
that material is:
65
ndX
X7— <CDEa)ppxCD
•rH
a"
pxbO
Pa)Xp•HCDPXr—IPxbO
tn
CDP•HPcr
xr-Ha)Ea)f-tpxa)TJoobO
CDx:
T3Pccj
PCDndP
f-tCDXPO,p
po•HPOccjf-tP
'SX
a)ECDf-tPXa)
CD
Pcr
xi— HpdbO
f-tPxp•HCDid
Xi— Hp.ebO
C/3
pp•rHPcr
xi—ia)eCDPPXa)
T3 o obO
opoCDP htnPH
CD •> P rt o d X P
X O pop a) f-tp PoX c/3
pCD O rX -H •H Gt-H *iH
P h TJ Oi-Hp a) o tn 2 P XZ
CD Pi-H
P h P O P CD cd p p f-lf-t O <0 P5 - 3O £rX OG if•HrP -tJP PPPo tn X P P
I-Ho o sd
tntn *h p sd p pCDo S P o O P O «Ptn p P P O <D •H tn Ptn p CDp b£)cr p•rH <P CDo Xi
CDXpbJOp
c/3
p p X -HCD T3 P P
CDCD P
X P H P
CD X P X tn X XX i-H T3 i-H p r—1 i-Hp CD p tn P pbO PC tn M -X ,X•rH p •rH P •rH •HP i-H •§ i-H p i-H i-H•rH c p g P eP p P g P • pP p PT3 Xi-H Eo Xf-l tn XfH o•H Xi-HX CD tn p p P tn PP E E X E tn ETd CDP tn
tnPP p PP p
tnPPX P P P bO p pp X i-H X P X P XCD CD o P •rH p p P> • • • • P • • g • •CD C/3•rH Pnd §
tn CD x P p tn Ptn P p P X p PP •rH •rH cd •H P •rHr—1 P E P Td P sd Po cr oP cr X cr p crtn tp p bO•rH • • p • • p • *sdp X pG X >p X •rHp Xi-H P
tnXiP
i-H i—t pTdi-H
ndGpsd prP tn
tnpx
pXCD+-> bC•rH bO•rH PT—H
bO•H XPbO•H
P i-H p i—t o i—t "G T—♦P , , tn ,P .. tn
tntn
X(/)
Td nd •rH PI—1 P i-H P Xi P P Pp CD P p p P > Po sd o rP sd P Xsd p rP P nd p Pc/3 •rHCD tn •rHP PP •rHP tntn
•rHPt-H P t-H P PP P Pi—t PPC • • rX • • P . • O • •P•rH X P•rH X Td X tn XX i-H xz I--I rH i-H •rH r-Hp +->sd p p
X Po +->X Xp
+->X
CD tuO p bO ,G bO bOE •rHi-H E •rHi—t tn •rHi-H T3 •rHi-Ho C/3 o tn t-H tn V-tP tnp •4 p • •
tn• * PP • •
p p .X cdp P p P P P Pp P p p •rH p ’id pp •rH p •rH sd •rH i-H •rHp P p P p P P Po cr o cr cr o cr& p sdE E o tn•rH • • X •rH • • X tn
tn* * X i—t • • XCD r—1 P r—1 p rH i—tp CD P P tp P ,X pp EP P EP oP EP P•rH EPo P o • P Ph P X P
X P xi P P P p P£ X 5 o X P X XP •rH P O P r—s PCD p tn P tni-H X i-h tn X O X \___/ XPh i-H Ot P r—t P i-H P i-HO • <D O C/3 p P P P PCD p P rX P rX P rXPh o Ot P •H tn •rH P •rH•H rH P i-H P i-H cd i-HP tn P E •H Oht/3 in c/3 gO CD o p X Xs tn S W s s
to
My hu
sban
d/wi
fe/f
ianc
e thinks
I should
attend
this
class
every
day
during
the
summer
sess
ion.
66
xrHM•rHrH§
xi-HCD
P i•H
X
CO
CD42P
(“ 1CD GOs S3CD •HP PP 3X XCD
XCtJ
XCDP X•rH p3 CDc r >
X
CD
COtoccjr-H »—H+->
43 oGO
•H CO•Hi-H
CO 43P
PXS3
CD CD42 PP P•rH ccSCDS3 X
Xi-Hp
i-H3o
43CO
43 HHGO
•Hi-HCO
P iS3
•H
CD
43P
S3P CD•rH P3 Xc r i-H
X
•rH42O
\i-H COCD CD£ >CD •rHP PP ccjX i-HCD CD
PNXi—H•rH
CDP i•rHi-H
CDeCDPPXCD
CDP•H3cr
4->43GOH—ICO
P43GO
•Hi-HCO
CD4->•rH3cr
CD£CDPPXCD
CDP i•rH
xs
CD43P
GOc
•HP3XXccjXXpCD>CD
COCOccSi-HoCO
•H42 P
XGCDPPCCS
Xr-H3O
43CO
G•H43pCOCDPCCSbCOCOccSi—loCOpCDCD
COXGCD
•rHP
m
xs
<uPi•HrHc3
pCD
X §i—H toCD
P ! CD•rH 43rH PG
G 3 GOO GX *H X *HCO ■—• P
CD CO CD 3B CD B XCD CO CDP P COP P P CDX CD X BCD B CD *H
S * ' PCD
CO BCD CD OP CD ' P CO•H 43 •H3 P 3 tocr cr co
GO CCJG i-H• i-l . • o
> . § X to
■§> M
.iH43 r S GO ^
3 * *H e ^ o* X * p
P <pCD
P > p pCD CD CD G
42 43 CDP CO P CO•rH CO •H 43CD CCS CD CCjG -H G
o CD43
PXGO
•Hi-HCO
CDP•rH3cr
CDBoppXCD
CDPi•rH
•rH43p
x3CDPPCCS
Xi-H3O
43CO
pCtJ42 pPiG
•H43 P
Pi-H
CDCOXe
p43GO
CDP•H3cr
CDsCDppXCD
CDPi
3O
43co
coPiS3•rH
43P
POCOCOCD
POPCP
Xpopo3ppCOS3
Xs
CDPi•Hi-H
§CDeCDPPXCD
CDP•rH3cr
p43GO•rHi-HCO
P CD
43 P • H CD S3
P43GO
•Hi-HCO
CDP•H3cr
CDfiCDPPXCD
CDP i•rH
C/>•H43P
gP
P
PS3CDCO•8CD43
3O
43CO
PiS3
•H43P
CO \—/ P S3 CD P ccj SPX
s
CD Pi • H rH§
CDSCDPPXCD
CDP•H3cr
p42GO
•Hi-HCO
PCD
42P•rHCDS3
P42GO
•Hi-HCO
CDP•rH3cr
CDbCDPPXCD
Xi-HCDPi
•rH
CDPi•rH
Xi-HCD£CDPPXCD
•H3cr
p43GO
PCD
43P•HCDS3
P43GO
•Hi-HCO
CDP•H3cr
CDbCDPPXCD
CDPi•rH
LO vO CO
12.
My fa
mily
/rel
ativ
es/c
hild
ren
think
I should
be absent
from
this
class
some
time
s during
the
summer
sess
ion.
C-*Od
67
CD2O
diHdod X XX X (/)X CD X X ur—1 X r—H rH d d d0 X 0) (D ■ H fi QX x • X X X•H bO •rH d •H X X
r-H d rH o rH X md •H d •H d d 0 Xd fH d mC/) 3 oX > d•rHX d X CD X Xr-H rH tn r—H o r-H XCD m CD CD X 0e 0 E fH E E fHCD E CD CD CD X 0 ofH •rH X g X d fH (/)+-> +-> +-> H X d X (/)X CD X 3 X X X 0
•• CD EoC/) •• CD in
CD •• CD Xop •• 0 4hOfH
CD X ECD d +-> CD •rH 0 \+-> C/) +-> X X fH•H d •rH bO •H 0 •H od r—H d d 2 fH d XCT om•rH
cr •rH:3d
cr dOX
cr odfHX
X +-> X tf) X 5 X C/)Gr—H+-> |
rH+-> CD r—H+-> V) rHX •HX X •H X 0X+->
X XdobO•H £
bO•rH +->CDbO*H bO•H
• •rHC/) +->d0 ••
rHV) §C/) ••rHm
oXd
rHm
XXd
fH C/) fH C/) fH X fH XCD X CD C/) 0 0d X d X X+-> +-> r-H X o X o•rH 0 •rH o •rH d -rH X!CD CD 0 0d d1—Hd C/)•rHX
d oX d oXX o X +-> X XG X XG
r-H +-> rGX!C/)
r—H+->Xeof_,
rHXXd rH
XXd
bO•H l—H bO•H m bO•H dobO•H do
••rHc/) Xd•rHX+->
••rHm +->dCDdX
••rHm X
od ••rHm X
odCD CD cj 0 d 0 X+-> C/) +-> X o X o•H CD •H CD •H d •H dd +-> d X 2 E d Ecr dEC/)C/)
cr dr-H3
cr soXcr SoXx d X O X X
r—H r—H rH X r—H r* rHCD o CD C/) 0 bO 0 bo6 \ 6 E d E dCD C/) CD i— i 0 •H 0 •HfH fH fH fH X fH X+-> CD +-> X X d X dX CD X d X 0 X 0CD p. CD •rH 0 & 0 PX C/) C/)X m X X X e * rH
r-H d r-H r—H X O d XCD d • CD <4H CD <—< d rHX CD d X r—H X rH rH•H •H O •H <D •rH d d X drH fH *H «+H I/) C/)X CDS w
rH C/)XEh—1
r—H fH rH0 d d o0 X CJ tn
dXo2
X0d0CJ
X01■xX0>
d
LO vO
X Xo o od "d dfs. E cO T3XJ X r—H XfH d XT3 0 o 0
rH > X >dXrH £
Xr—H d X
r—H0 C/) 0 o 0E E X E0 d 0 0fH o fH C/) XX X X X XX X d X0 X 0 •H 0d • • X•H XX0 X 0 0 0X X o X• H ,—N •H d •Hd tn d d dcr '__t cr •H crX IXd \
0 • • 0 • •
X fHd X 4h•H XrHX P r—H+-> £ i-H
XX X 'd XbO•H dobO
•rH dd bO•HrH X rH
in XC/) rHin
X • • d • •d XfH X fH X0 £ 0 X 0X X d XX o X o X
•rH •rH X •H0 0 0d O d X dX d• • X • •
X XG X £ XrH d r—H o rHX 2s X d XX X XbO•H do
bO•H oXbO•H
rH s. rH rHC/) O C/) X inO • • dd£0 X 0 0X o X d X•rH d •H o •Hd E d X dcr cr crs oo dX • •X X X Xr-H r—H o r—H0 bO 0 d 0E d E E E0 •H 0 0fH X fH XX d X o XX 0 X X X0 P 0 0C/) •N
rH r—H rH,X rH d i—1rH d X <T,rH 0d X d XfH d 0 d0 bOd X X0 o d oCJ 2 IH 2
X ooi-H rH
19.
In ge
nera
l, how
much
do you
want
to do
what
your
fami
ly/r
elat
ives
/chi
ldre
n think
you
should
do?
68
Appendix C
70
cj
PE -i
tO0i-H&PcoXnd34->COX!3.§Pa.ndpP
a.p0<D4->0CQtOpoto• H&&ou
0 o LOCM r—1 • oP-i 0 ••H > • • • Va 0 to • to CO •
P<|V
cmr—1 • • to • • to • • • •
VI p • p p * H V) V) tO Vio LO • p p •s • • • •o P ft. •V ft oo ft. P P P p0 ft rH ft. 'vf 00 ft o o0 *\ nt CM to vO LO CM LO •v ft, ft, •S
p ft r H CM 00 0> to o r H oo r H CM 00 ap too o vD oo r H r H v£> CM to P" ■P" CM CM0 •H oo cn r- a t-v. » ft•H to • * to • • • r- CM 1 1 • 1CM (M to • CM r H r H r H•H II II II IIP II II ii II II II II IItoO /— \ /— , r— \ r— \•H t— \ /— \ f— \ /— , \ /— , ,— \ f— \ CM 1 CM CM 1 CMCO CM 1 CM 1 CM 1 CM I cm I CM I CM 1 CM 1 nd nd nd ndCM 'P 1 nd 1 nd 1 nd nd nd j nd nd | O o o o
r H nt rH vO CM r H r. LO oo oo oo 00O '_/ '— ' '— ' '__/ '— > '— ' '— ' '— ' to to to toCM CM CM CM CM CM CM CM '— ' ,__ / '__ / ,__ /
M X X X X X X X X M j M j +->l +->to0E-h
0
S Ch P LO CO CMX IIndP 2MCO
0r-H
&<S to p cmCO i— i
r H \D LOto CM to to 00 CMr H nt LO •nf \D v Ont oo to to to LO 00 v O• • • ft • • • •
P" LO p- c- LO CM r H LO oo lncr> \ o CM r H r H r H
'vf COto oo
top4->
CD Pr-H Mftp . to cop 0•H to 1—1 1—1p 0 p pp .M S M> P 0 -H
X 2 tin P0 Pco 2
o 00 cr> rHLO P" LO 00 o o CM LO\0 rH tO LO tO rH oo toO ret CM tO nt X to to
• • • • • • ft ftp- C" LO CM rH LO 00 LOCM rH rH i— l
4->o2po
p00opCl,00P000a
to4-> O 0
•r—iX>Pco
4->P0ndP4-> PCO 00 P4-> E-*P PP 0 0nd ftX ,P XP P r HP E- O 4->CJ M PN X 0M r-H nd PP 4-> 0 P0 P •M Pnd P 0 0 CJP o P 0P, +-> -H X P P h nd•H CO 4-> i— 1 P X 0ftp P -H CJ UJ ftXto 0 p B pp +-> nd P to to o0 p P CL, p p 2
tM P p p p•H nd Cd cm o o to-M P O 2 2 P•H p Cm PCJ toO O 0 P M P M P oP r £ P -H P -H P , 20 0 P O 0 Q 1 nd 0 Q nd 0 a0 co nd P 0 0 2 co 0 2 co 0 2 co •P P 0 P too P P oCX 2 z> X tH < CJ CJ 2
PP0 Q 2 co
71
0)OP0)PiCD
PH r-HPh CD•H >O CD
i—1 • •P=H lO • VI
O LO too P • P
CD • PO •\ *\P • •sd- r 00to w> tO CNJ tMO *M * i—1 *
•rH (/) 1 • 1Ph I•rH II IIP IIW) /—\ /— \
•rH Ph 1 ,— , Ph 1C/D T3 Ph t 3
O T J | ©PH oo o ooO to oo to
v__> to '__'
+-> ■Pi ’ p 1to P iCD
H
CDi-HP h cn o oo£ ctj r-H 00 LO O r—1 OTO LO o \o lo o f"- \ D
C/D CM LO O O i O lO• • • • • •
X II C7) LO CM r-H r-H”0 r-H r-H3 24->C/D
CDrHp . r-H r-H 00£ to t o cn cm cn cn lo
TO CM ^}" r-H r-H CD tOC/D r-H OD H OS O vO vO
• m • • f •H-> II 00 LO CM r-H r-HO r-H r-H
'—i 2•rHCl,
TO3CD
PSuO
3=
CDrQ
of—\ +->"TOCDP CDp +-»
*H O4-> CD CDp P ho X TO
o CD u HV._/ ,-H CD
rO COr-H TO M TJcd •H P CD
fH O P P +-» P0) TO 2 TO TO O TO
i—1 > CD Q CD Q CD CD QrQ • 2 w < 2 w P h S c/ dTO o CL, x
E - 2 CD PJ
72
Table C2
Model FA-1(Prior behavior data omitted ATT § SN to INT § TB„
all paths estimated saturated model)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (A )
Unique Variance (6 or £) 1-A2 (stand.)
AttitudeXIX2
.708
.754.499.431
Subjective Norms X3 X4
.850
.849.278.278
IntentionY1Y2
.826
.798.317.363
Target Behavior Y3 Y4
.654
.661.572.563
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ATT-SN .542 5.278
Y 11 .613 4.886
Y 21 .211 1.334
Y 12 -.012 -.116
Y 22 -.124 -1.120
3 21 .501 3.670
Residual Variances Intention Target Behavior
.632
.6285.1683.392
73
Table C3
Model FA-2(Prior behavior data omitted AT § SN to INT § AT
to target behavior)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (A)
Unique Variance (6 o r e ) 1-A2 (stand.)
AttitudeXIX2
.711
.753.494.434
Subjective Norms X3 X4
.859
.840.261.295
IntentionY1Y2
.824
.801.322.359
Target Behavior Y3 Y4
.671
.644.549.585
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ATT-SN .539 5.199
Y 11 .620 4.948
Y 21 .121 .942
Y 12 -.023 -.234
Y 22 DELETED
e21 .516 3.735
Residual Variances Intention Target Behavior
.631
.6435.1513,431
74
Table C4
Model FA-3(Prior behavior data omitted paths ATT § SN to INT not to TB)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (A)
Unique Variance (6 or c) 1-A2 (stand.)
AttitudeXIX2
. 713
.749.492.439
Subjective Norms X3 X4
.859
.840.263.294
IntentionY1Y2
.819
.801.329.358
Target Behavior Y3 Y4
.655
.660.571.565
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ATT-SN .543 5.214
Y 11 .633 5.039
Y 21 DELETED
Y 12 -.031 -.310
Y 22 DELETED
321 .603 5.516
Residual Variances Intention Target Behavior
.619
.6365.1523.384
75
Table C5
Critical Ratios of Parameters--Model BS-1 All Paths Estimated-Saturated Model
Measurement ModelMeasureFactors £ Variables
Standardized Factor Loadings (A)
Unique Variance (6 or £ ) 1-A.2 (stand.)
AttitudeXIX2
.716
.746.487.444
Subjective Norms X3 X4
.879
.821.228.325
Prior Behavior X5 X6 X7
.682
.662
.674
.535
.561
.546Intention
Y1Y2
.777
.849.396.279
Target Behavior Y3 Y4
.616
.702.621.507
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^A-S .331 5.152
^A-PB .223 4.225
^S-PB .015 .323
Y 11 .390 2.268
Y 21 -.077 -.552
y 12 .096 .947
y 22 .115 1.205
y 13 .556 4.252
76
Table C5 (Continued)
Standardized Parameters Factor Correlations
Causal Model ParametersStandardWeights
Critical Ratio (t-values)
Y 23 .934 5.644
321 .059 .550
Residual VariancesIntention .509 4.984Target Behavior .161 1.273
77
Table C6
Model BS-2 (Path SN to TB deleted)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (X)
Unique Variance fo or e ) 1-A2 (stand.)
AttitudeXIX2
.712
.749.493.438
Subjective Norms X3 X4
.862
.837.257.300
Prior Behavior X5 X6 X7
.683
.664
.679
.534
.559
.539Intention
Y1Y2
.778
.847.394.282
Target Behavior Y3 Y4
.611
.708.627.499
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^A* S .544 5.243
^A-PB .437 4.196
^SN-PB .050 .611
Yll .355 2.824
Y21 .028 .251
Y12 .082 .851
Y 22 DELETED
^13 .433 4.304
78
Table C6 (Continued)
Causal Model ParametersStandardized Parameters Standard Critical Ratio
Factor Correlations Weights Cl-values)
y 23 .833 5.960
21Residual Variances
.086 .674
Intention .510 4.994Target Behavior .190 1.528
79
Table C7
Model BS-3 (Paths SN to TB, ATT to TB deleted)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (A)
Unique Variance (6 or e ) 1-A2 (stand.)
AttitudeXIX2
.713
.749.492,438
Subjective Norms X3 X4
.861
.838.258.298
Prior Behavior X5 X6 X7
.682
. 664
.679
.535
.559
.539Intention
Y1Y2
.779
.846.393.284
Target Behavior Y3 Y4
.609
.710.629.497
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^AT-SN .543 5.245
^AT-PB .441 4.312
^SN-PB .054 .655
Yll .355 2.812
*21 DELETED
*12 .082 .857
*22 DELETED
*13 .431 4.278
80
Table C7 (Continued)
Causal Model ParametersStandardized Parameters Standard
Factor Corx'elations WeightsCritical Ratio
(t-values)
Y 23
21Residual Variances
Intention Target Behavior
.835
.103
.512
.190
6.003
.943
5.0171.527
81
Table C8
Model BS-4(All paths to TB from exogenous variables deleted)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings CO
Unique Variance fc or e ) !-X2 (stand.)
AttitudeXIX2
.715,747
.448
.443Subjective Norms
X3 X4
.875
.825.234.320
Prior Behavior X5 X6 X7
.690
.639
.687
.524
.591
.528Intention
Y1Y2
.745
.825.445.320
Target Behavior Y3 Y4
.643
.673.587.548
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ AT • SN .542 5.174
4>AT*PB .444 4.223
SN * PB .029 .340
Y 11 ..304 2.463
Y 21 DELETED
Y 12 .110 1.164
Y 22 DELETED
Y13 .556 5.245
82
Table C8 (Continued)
Causal Model ParametersStandardized Parameters Standard Critical Ratio
Factor Correlations Weights (t-values)
y 23 DELETED
B21 .714 6.783
Residual VariancesIntention .397 4.432Target Behavior .490 3.155
83
Table C9
Model BS-5(F-A Models only paths from ATT § SN to INT § from INT to
target behavior--prior behavior not included except in measurement model)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings (X)
Unique Variance (6 or e ) 1-A2 (stand.)
AttitudeXIX2
.615
.637.621.594
Subjective Norms X3 X4
.873,822
.238
.324Prior Behavior
X5 X6 X7
.683
.644
.690
.533
.586
.524Intention
Y1Y2
.762
.824.419.321
Target Behavior Y3 Y4
.646
.669.582.553 /
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^AT-SN .646 5.481
^AT-PB .607 4.934
^SN-PB .022 .261
■Yll 1.035 6.302
^21 DELETED
^12 -.395 -2.984
y 22 DELETED
y 13DELETED
84
Table C9 (Continued)
Standardized Parameters Factor Correlations
Causal Model ParametersStandardWeights
Critical Rati< (t-values)
y 23 DELETED
321 .687 6.480
Residual VariancesIntention .309 2.995Target Behavior .527 3.254
85
Table CIO
Model BS-6(Prior behavior to INT only ATT to INT £ TB, SN to INT, PB to INT)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings a )
Unique Variance (6 or e ) 1 - x 2 (stand.)
AttitudeXIX2
.713
.747.491.442
Subjective Norms X3 X4
.876
.823.282.322
Prior Behavior X5 X6 X7
.689
.639
.688
.525
.592
.527Intention
Y1Y2
.749
.835.439.303
Target Behavior Y3 Y4
.650
.665.577.558
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ATT-SN .540 5.159
^ATT-PB .457 4.321
^SN-PB .029 .339
*11 .283 2.239
*21 .080 .654
*12 .116 1.209
*22 DELETED
*13 .548 5.068
86
Table CIO (Continued)
Causal Model ParametersStandardized Parameters Standard Critical Ratio
Factor Correlations Weights (t-values)
y 23 DELETED
321 .649 4.777
Residual VariancesIntention .426 4.554Target Behavior .510 3.238
87
Table Cll
Model BS-7(No paths from prior behavior path from ATT to TB^
Bentler-Speckart Modification 1)
Measurement ModelMeasureFactors § Variables
Standardized Factor Loadings M
Unique Variance (6 or e ) 1-A2 (stand.)
AttitudeXIX2
.610
.634.628.597
Subjective Norms X3 X4
.892
.808.205.347
Prior Behavior X5 X6 X7
.678
.649
.689
.540
.578
.525Intention
Y1Y2
.770
.853.408.273
Target Behavior Y3 Y4
.680
.636.537.596
Causal Model ParametersStandardized Parameters
Factor CorrelationsStandardWeights
Critical Ratio (t-values)
^ATT-SN .575 4.956
^ATT-PB .695 5.416
^SN-PB .025 .296
Y 11 .844 6.409
Y 21 .500 3.145
Y 12 -.217 -2.086
y 22 DELETED
y 13 DELETED
88
Table Cll (Continued)
Standardized Parameters Factor Correlations
Causal Model ParametersStandard 'Weights
Critical Ratio (t-values)
y 23 DELETED
321 * 260 1.770
Residual VariancesIntention .451 4.199Target Behavior .496 3. 249
Appendix
90
Discussion of Path Analysis
Path analysis is the logical precursor of structural analysis.
This discussion is based upon the following references: Billings §
Wroten, 1978; Cook § Campbell, 1979; Kerlinger § Pedhazur, 1973; Li,
1975; and Namboodiri, Carter, § Blalock, 1975.f
In order to use path analysis the theoretical framework being
studied must first be made explicit. The basic technique uses ordinary
least squares regression to derive path coefficients which may be defined
as standardized regression coefficients and which may be interpreted as a
ratio of two standard deviations. These path coefficients between vari
ables are then used to test proposed causal relationships among a set of
variables. In this manner, path analysis can be used to test an a priori
causal hypothesis against a set of observed correlations. According to
Li (1975) the usefulness of path analysis over simple multiple regression
techniques lies in its ability to extend the single multiple regression
equation treatment to a network of variables involving more than one
equation. The use of path analysis for decomposing a correlation into its
components of direct and indirect effects within a given causal model is
an additional important function of this analytic approach.
Kerlinger and Pedhazur (1973) explain the use of path analysis as
a tool for theory testing as follows:
Path analysis is an important analytic tool for theory
testing. Through its application one can determine whether
or not a pattern of correlations for a set of observations
is consistent with a specific theoretical formulation
. . . . a correlation between two variables can'be expressed
as a composite of the direct and indirect effects of one
variable on the other. Using path coefficients it is there
fore possible to reproduce the correlation matrix (R) for all
the variables in the system. . . . however, as long as all
variables are connected by paths and all the path coefficients
are employed, the R matrix can be reproduced regardless of
the causal model formulated by the researcher. Consequently,
the reproduction of the R. matrix when all the path coefficients
are used is of no help in testing a specific theoretical model.
What if one were to delete certain paths from the causal model?
This, in effect, will amount to setting certain path coeffi
cients euqal to zero. The implication is that the researcher
conceives of the correlation between the two variables whose
connecting path is deleted as being due to indirect effects
only. By deleting certain paths the researcher is offering a
more parsimonious causal model. If after the deletion of some
paths, it is possible to reproduce the original R matrix, or
closely approximate it, the conclusion is that the pattern of
correlations in the data is consistent with the more parsimoni
ous model. . . .
If after the deletion of some paths there are large dis
crepancies between the original R matrix and the reproduced
one, the conclusion is that, in the light of the relations
among the variables,, the more parsimonious theory is not
tenable. (p. 317)
92
Path analysis requires a distinction between variables labeled
exogenous and endogenous. Exogenous variables are one or more variables
whose causes lie outside the causal system. Variables so labeled may
be correlated with no causal direction specified. Endogenous variables
arc those whose cau3e3 lie within the system* These variables musl be
specifically ordered with respect to hypothesized cause-effect relation
ships. The variance of an endogenous variable is considered to be
accounted for by the effects of other endogenous and exogenous variables
prior to it in the ordering and by a residual or error term.
For each endogenous variable it is necessary to create a causal
model which is a weighted function of variables prior to (causal to)
that variable and an error term, Actual weights are determined by means
of multiple regression in which each endogenous variable is treated as
the criterion and the variables hypothesized to affect it are treated
as predictors. The resulting standardized beta weights are path coef
ficients representing the direct effects of the causal variables upon the
criterion variable. An indirect path is computed by multiplying together
all the direct path coefficients which comprise the indirect path.
A basic theorem of path analysis is that "the correlation between
two variables is the sum of all connecting paths between the two vari
ables" (Lij 1975^ p. 149). Therefore^ indirect effects are the differences
between total effects (the correlation) and direct effects (path coef
ficients) . In this way^ path analysis may be used to decompose a
correlation into its components of direct and indirect effects within
a causal model.
93
In testing a theory which predicts the absence of one or more
direct paths, those variables hypothesized to have only indirect effects
are deleted from the regression equation for that particular dependent
variable. If the hypothesized paths are statistically or meaningfully
significant (beta weights usually greater than .05) and the model is
able to reproduce the observed correlation matrix, the theory as modeled
is supported.
The major assumptions that must be met in the use of path analysis
are: (a) a priori specification of the causal sequence of the variables;
(b) the specified causal sequence must be a one-way flow (recursive);
(c) the residual of each endogenous variable affects only that specific
variable and is uncorrelated with other system variables or with their
residuals; and (d) the data are linear, additive (no interaction effects),
and on a ratio or interval scale of measurement.
There are, however, some basic shortcomings in the use of path
analysis. The degree of reliability with which latent variables or
constructs are measured by the observed variables is not assessed when
multiple regression equations using one observed indicator for each latent
variable are used to obtain the beta weights (path coefficients). Because
of this, any error in measurement is confounded with estimation of causal
parameters pertaining to the latent variables. Also, the method requires
the assumption of a recursive or unidirectional causal model which may
not always represent reality.