UvA-DARE (Digital Academic Repository)
High- and low-dose alcohol-related expectancies and the differential associations with
drinking in male and female adolescents and young adults
Wiers, R.W.H.J.; Hoogeveen, K.J.; Sergeant, J.A.; Gunning, W.B.
Published in:
Addiction
DOI:
10.1111/j.1360-0443.1997.tb02956.x
Link to publication
Citation for published version (APA):
Wiers, R. W. H. J., Hoogeveen, K. J., Sergeant, J. A., & Gunning, W. B. (1997). High- and low-dose alcoholrelated expectancies and the differential associations with drinking in male and female adolescents and young
adults. Addiction, 92(7), 871-888. DOI: 10.1111/j.1360-0443.1997.tb02956.x
General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),
other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations
If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating
your reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Ask
the Library: http://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,
The Netherlands. You will be contacted as soon as possible.
UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)
Download date: 17 Jun 2017
Addiction (1997) 92(7), 871± 888
RE SE AR C H RE PO R T
H igh- and low-dose alcohol-related
expectancies and the differential associations
with drinking in m ale and fem ale adolescents
and young adults
1,2
2
REINOU T W . W IERS, KEES-JAN H OOGEVEEN,
2
1
JO SEPH A. SERGEANT & W . BOUDEWIJN G UNNING
1
Academic M edical Centre Amsterdam, the N etherlands, Department of C hild and A dolescent
2
Psychiatry and Amsterdam Institute for Addiction Research (AIAR) & Department of
Clinical Psychology, University of Amsterdam, The N etherlands
Abstract
A Dutch question naire was developed consisting of positive and negative expectancies relating to low and high
doses of alcohol. The associations of these four types of expectancies with current alcohol consu mption were
investigated in three sam ples: seconda ry school pupils of 11± 15 years old, seconda ry school pupils of 16 and
older, and university underg raduate student s (total n 5 554). Using restrictive factor analyses, a com m on
factor-m odel of the expectancies was show n to ® t adequately across subgroups. W hich expectancies were
associated with current alcohol consum ption varied substantially across the subgroups. As expected from
previous research, inclusio n of high dose expectancies did not substantially improve the prediction of drinking
in university students . H ow ever, positive and negative high-dose expectancies were found to be powerful
predictors of current alcohol consu m ption in secondary school boys of 16 and older, the subgroup with the
highest average alcohol consu m ption on each occasion. Possible implications are discussed for future research
and interventions.
Introduction
In the past 15 years, alcohol-related expectancies
have emerged as powerful predictors of alcohol
consumption. Expectancies have consistently
been found to be associated with current alcohol
consumption in college students (e.g. Leigh &
Stacy, 1993), com munity samples (e.g. Brown et
al., 1980) and adolescents (e.g. Christiansen,
Goldman & Inn, 1982). Expectancies have also
been shown to predict future drinking in adolescents after 1 year (Christiansen et al ., 1989), 2
years (Smith et al., 1995) and 9 years (Stacy,
Newcom b & Bentler, 1991). Furtherm ore, expectancies have proven to be useful in predicting
treatment-outcome in alcoholics (e.g. Jones &
M cMahon, 1994). Although expectancies are
now widely believed to be important in alcohol
research, several issues concerning their
Correspondence: R. W . W iers, Department of Child and Adolescent Psychiatry, Academ ic M edical Centre, PO
Box 22700 , 1100 DE Am sterdam, T he Netherlands. Tel.: 1 31 20 5256818 /5662230 ; fax: 1 31 20
6391369- 6971937 ; e-mail:kp wiers@m acm ail.psy.uva.nl
Subm itted 18th June 1996; initial review com pleted 15th November 1996; ® nal version accepted 6th January 1997 .
0965± 2140/97/070871 ± 18 $9.50
Carfax Publishing Company
Ó
Society for the Study of Addiction to Alcohol and Other Drugs
872
Reinou t W . W iers et al.
measurem ent remain (e.g. Leigh, 1989a; Goldman et al., 1991; Leigh & Stacy, 1991, 1993).
It is now well established that people hold
both positive and negative alcohol-related expectancies (e.g. Fromm e, Stroot & Kaplan, 1993;
Leigh & Stacy, 1993; C hen et al ., 1994; M cM ahon, Jones & O’ Donnell, 1994). In addition, it
has been shown that expectancies vary with the
dose of alcohol concerned (Southwick et al .,
1981; Connors et al ., 1987; Collins et al ., 1990;
Earleywine & M artin, 1993). W ith respect to
dose and valence, four types of expectancies can
be distinguished: positive expectancies for a low
dose, positive expectancies for a high dose, negative expectancies for a low dose, and negative
expectancies for a high dose.
Previous research ® rst focused on positive expectancies. The ® rst and the most widely used
instrument in alcohol-expectancy research
(AEQ , Brown et al ., 1980) was aimed at: ª the
dom ain of hum ans’ expectancies about the reinforcing effects of moderate alcohol consumptionº (Brown et al ., 1980, p. 424). The authors
followed Rotter’ s de® nition of expectancies as
ª the probability held by the individual that a
particular reinforcement will occur as a function
of a speci® c behaviourº (ibid ., p. 419). The
instrument contained item s concerning a low
dose of alcohol (ª a drink or twoº ) as well as
items about an unspeci® ed dose of alcohol (e.g.
ª Drinking relieves boredomº ), but no items
speci® cally targeted high doses of alcohol. In the
typology de® ned above, all six factors of the
AEQ fall into the category of positive expectancies for a low (to moderate) dose of alcohol.
Negative expectancies were introduced on a
small scale in two modi® ed versions of the AEQ :
the AEQ-A (AEQ modi® ed for adolescents,
Christiansen et al ., 1982) contained one negative
scale, one m ixed scale and ® ve positive scales. In
the modi® ed AEQ by Rohsenow (1983) two
negative scales were added to the six abridged
positive expectancy-scales of the AEQ . In subsequent years, several instrum ents were developed with an approximately equal number of
positive and negative scales (e.g. Young &
Knight, 1989; Fromme et al., 1993; Leigh &
Stacy, 1993). It was shown with various instruments that negative expectancies signi® cantly
improve the prediction of current drinking
(Fromm e et al., 1993; Leigh & Stacy, 1993;
Chen et al., 1994; M cMahon et al., 1994). The
issue of whether positive or negative expectancies
are better predictors of current drinking is an
unresolved issue: the ® ndings of Stacy, Widaman
& Marlatt (1990) and of Leigh & Stacy (1993)
favour positive expectancies, but those by
Grube, Ames & Delaney (1994) and McM ahon
et al. (1994) favour negative expectancies. The
comparison between positive and negative expectancies is confounded by the fact that the
expected positive effects are more proximal than
the expected negative effects. Negative effects
are expected to occur later in a drinking session;
students expect more positive effects of alcohol
for the rising lim b of the blood± alcohol curve
and expect more negative effects for the descending limb of the blood± alcohol curve, which
is in agreem ent with the actual effects of alcohol
(Earleywine & M artin, 1993). Recently, it has
been found that still more distal negative expectancies (e.g. feeling sick the day after, or losing
one’ s job with continued drinking), are also
strongly associated with current drinking
(McM ahon et al., 1994).
Even though it has been recognized in the
literature that expectancies vary with dose of
alcohol (Southwick et al., 1981; C onnors et al.,
1987; Collins et al ., 1990; Earleywine & Martin,
1993), the assessm ent of dose-related expectancies has received relatively little attention. One
reason could be that the AES (Southwick et al.,
1981), an early scale which assessed dose-related
expectancies, predicted current alcohol consumption less accurately than two questionnaires
which did not assess high and low dose effects
independently (Leigh, 1989b). The AES was
also criticized in an in¯ uential review of the
expectancy literature for the use of a bipolar
response format (Leigh, 1989a). This could well
be the main reason for the poor association with
current drinking. In a m ore recent expectancy
questionnaire (Fromm e et al., 1993), the relation
between dose and expectancy was investigated
by asking the subjects to indicate the num ber of
drinks they would need to consume in order to
experience each of the expected alcohol effects.
Positive effects were expected to occur after a
signi® cantly lower dose than negative effects. In
this way, the two most plausible types of expectancies are m easured which correspond to the
biphasic response to alcohol (low dose positive
and high dose negative expectancies). However,
the other two types of expectancies (negative low
dose and positive high dose) are not measured by
this response format. This might be an adequate
H igh- and low-dose expectancies
strategy in students but not necessarily in other
populations.
What exactly constitutes a ª positiveº or a
ª negativeº expectancy? One person’ s positive
expectancy might be another person’ s negative
expectancy. Aggression, for exam ple, has been
reported both as a positive expectancy (Brown et
al., 1980) and as a negative expectancy (e.g.
Leigh & Stacy, 1993). Some researchers have
argued that the valence (or ª valueº or
ª evaluationº ) of the expected effect should be
measured on an individual basis (e.g. Fromme et
al., 1993; G rube et al., 1995). However, as observed by Grube (1995), the usually applied
strategy in expectancy research is not to include
valence on an individual basis. In this study
value was not m easured on an individual basis,
but this approach holds prom ise for future
expectancy research and clinical applications
(e.g. Mooney & Corcoran, 1989).
The expectancy questionnaire used in this
study contains items stated in the general format
concerning the expected effects of alcohol on ª an
average personº . This strategy has several advantages: one questionnaire may be used irrespective
of the alcohol experience of the (young) subject,
which m akes a direct com parison of the expectancies of drinking and non-drinking subjects in
the youngest subject-group possible. A second
advantage is that it will allow for future comparisons with a different subject-group that our research group is studying (children of alcoholics,
see W iers, Sergeant & Gunning, 1994). The
third advantage is that the results can more easily
be compared with the most widely used expectancy questionnaire for adolescents which
also contains items in the general format (the
AEQ-A; e.g. Christiansen et al., 1982, 1989;
Smith et al., 1995). It has been argued, however,
that the general format is not optim al with
drinkers (Leigh, 1989a; Young & Knight, 1989).
In direct comparisons it has been found that
subjects expect stronger effects for others than
for themselves (Rohsenow, 1983; Leigh, 1987).
It is not clear, however, which format results in
a better prediction of current drinking. In the
only reported study known to the authors in
which both personal and general expectancies
were included, the correlations of personal expectancies and general expectancies with current
drinking were judged to be so similar that the
latter were left out of the report (Wood, Nagoshi
& Dennis, 1992). Hence, this paper reports a
873
study which measured expectancies about the
effects of a high or a low dose of alcohol on an
average person and the associations of these
expectancies with current drinking patterns in
adolescents and adults.
M ethod
M easures
Expectancies. A Dutch expectancy questionnaire was developed to m easure the four types of
expectancies described in the introduction. It
consisted of a prim ary section of 42 items concerning the response of an average person to a
low dose of alcohol, followed by a smaller secondary (exploratory) section of 14 items regarding the response of an average person to a high
dose of alcohol. An unipolar Likert scale was
used as the response format as in recent expectancy questionnaires, with a ® ve-point scale from
ª don’ t agreeº to ª strongly agreeº . Several
speci® c scales from the AEQ and AEQ-A served
as models for our a priori positive scales. The
general positive scale of these measures was not
included. General positive expectancies are accounted for using the m ethod of ® tting a second
order factor m odel, as in Leigh & Stacy (1993).
M ost negative items were constructed by changing positive items into the opposite meaning. For
exam ple, the item ª after a few drinks people say
clever thingsº (cognitive/motor enhancement)
becam e ª after a few drinks people say stupid
thingsº (cognitive/motor impairment). Som e
item s were translated from the AEQ and the
AEQ-A, others from a questionnaire developed
by Sher et al. (1991). The remaining items were
generated by the authors with the help of several
regular bar visitors.
Alcohol consu m ption. A self-report measure of
alcohol-consumption was used. Self-report measures have generally been found valid when used
in research settings with sober subjects who are
given assurances of con® dentiality (Sobell &
Sobell, 1978, 1990; Sobell et al., 1988; Knight &
Godfrey, 1993). These requirements were
ful® lled here.
Subjects were asked to indicate on a checklist
of alcoholic beverages how many they would
typically drink of each beverage on an average
weekday (Monday± Thursday). All alcoholic
consumptions were transformed into ª standard drinksº (a standard drink contains
874
Reinou t W . W iers et al.
( n 5 216; 104 boys and 112 girls, mean
age 5 13.3 years) and secondary school pupils of
16 and older ( n 5 163; 101 boys and 62 girls,
m ean age 5 17.0 years). The third group
(ª university studentsº ) consisted of introductory
psychology students ( n 5 175; 73 men and 102
women, mean age 5 21.7 years). The second and
third subject-groups showed some overlap in
age: university students were 18 years or older,
and the age of some of the 16 1
group of secondary school pupils extended into the range of
the university students. The overlap is not large:
when 18 years is used as an overlapping border,
the age of 8% (13/163) of the subjects of group
2 had an age in the range of group 3 (nine
subjects were 19, three subjects were 20, and one
subject was 21 years old). The reason why the
subject-groups were differentiated with respect
to current education level rather than age was
that the subcultures of the subject-groups is
rather different: university students generally live
on their own, in student houses or in private
rooms, whereas the secondary school pupils generally live with their parents. In addition, the
university students were selected for school
ability: they must have ® nished the highest level
of education in secondary school. The secondary
school pupils, in contrast, are of different
levels of education: only about 25% of them
would eventually qualify for university. Further
details concerning the subgroups can be found in
Table 1.
approximately 15 ml of pure ethanol). In addition to the amount consum ed on an average
weekday, subjects were asked to estim ate the
number of weekdays in a month that they would
drink this quantity (range: 0± 16). The same response format was used to ascertain the number
of standard drinks consum ed during weekenddays (Friday± Sunday; range: 0± 12). Subjects
were also asked to indicate on how m any days
they drank ® ve or m ore standard drinks during
the past month. At the end of the questionnaire,
the subject was asked to give an estim ate of the
number of standard drinks he or she consum ed
during an average week (ª one guessº ). From
these data, an estimate of the subject’ s weekly
alcohol consumption was calculated (abridged
as ª QFº , indicating quantity 3 frequency 1 ), as
well as the number of days a subject had consumed ® ve or m ore alcoholic consum ptions
(ª Dº , indicating number of days intoxicated).
Subjects
Subjects were 554 secondary school pupils and
university students. The secondary school pupils
were drawn from four secondary education levels: low level professional training, medium general education, m edium level professional
education and high-level general education. The
secondary school pupils were divided into two
groups with respect to age: 11± 15-year-olds
Table 1. Characteristics of the six subgroups de® ned by cohort and gender
Subgroup
G ender
15 or younger
boys
104
girls
112
boys
101
girls
62
m en
73
Adolescents
16 or older
Students
wom en
1
n
102
M ean age
( 6 SD)
%
abstainers
13.4
1.0)
13.3
( 6 1.0)
46
17.1
1.1)
16.8
( 6 0.8)
6
(6
(6
22.2
5.0)
21.4
( 6 3.8)
(6
M ean standard
drinks/week
1
QF ( 6 SD)
(6
60
(6
(6
15
3
10
T he form ulae of the drinking variables are given in Note 1.
(6
2.0
4.0)
1.5
3.6)
15.3
13.8)
6.5
( 6 7.0)
15.9
11.7)
9.5
( 6 9.6)
M ean standard
drinks/occasion
1
Q ( 6 SD)
(6
(6
(6
(6
(6
(6
2.1
2.8)
1.3
1.9)
6.5
4.0)
3.8
2.7)
5.3
3.5)
3.5
2.4)
Number of
intoxications
past m onth
1
D ( 6 SD)
(6
(6
(6
(6
(6
(6
0.8
2.6)
0.5
1.6)
4.7
4.5)
2.3
3.4)
5.8
6.7)
3.2
4.5)
H igh- and low-dose expectancies
875
After a few drinks ... (low dose expectancies)
0.6
0.7
1.0
1.4
1.0
0.8
It is arousing to dance
People talk about sex more easily
People kiss more readily
0.5
0.7
0.9
0.8
People
People
People
People
1.1
1.0
1.2
People find a dinner party festive
People enjoy watching tv together
One is more readily accepted by a group
1.0
1.6
0.8
People have difficulties expressing themselves
People become bad at snooker
People cannot think clearly
0.6
0.5
0.6
0.6
0.8
People
People
People
People
A party
0.6
0.5
0.5
0.5
get good ideas
can write poems more easily
can ride their bike fast
are good at pinball
do not feel like making love
feel unattractive
feel insecure
become gloomy about the future
becomes annoying
0.7
0.6
0.7
0.7
0.6
0.9
0.4
0.5
0.5
0.5
0.5
SEX+
low-dose sexual
enhancement
CM+ low-dose
cognitive and motor
enhancement
CEL+ low-dose
celebration
group-acceptance
CM ± low-dose
cognitive and motor
impairment
INH ± low-dose
inhibition/
negative mood
After many drinks ... (high dose expectancies)
1.0
0.8
0.6
1.0
It is exciting to drink with friends
People enjoy a party
People get good ideas
People become romantic
0.9
0.7
1.7
1.4
People
People
People
People
want to fight
say stupid things
dance badly
become gloomy
0.6
0.7
0.4
0.5
0.8
0.7
0.6
0.8
HD+ high-dose
positive
HD ± high-dose
negative
Figure 1. R estrictive factor analysis model with ® rst-order factors as ® t to all subgroups. The rectangles contain the items
of the latent factors. Small ellipses contain the error components, the numbers above the arrows are the common factor scores
obtained in the multi-group analyses (see Appendix).
Procedure
In a pilot study (Hoogeveen, 1994), two versions
of the expectancy questionnaire were compared
in a university student sample ( n 5 370). One
version contained positive and negative expectancies, as described above. The other version
consisted of only positive a priori scales for a low
and for a high dose of alcohol. The version
which included negative expectancies resulted in
a better prediction of current drinking. It was
then decided to continue the research with adolescents with the version that included negative
items. The general instruction and the wording
of some item s were changed to make them more
suitable for secondary school pupils. The items
in the ® nal common factor model presented
in Fig. 1 were identical across all subjects. All
questionnaires were completed in a surveyed
assem bly setting (the secondary school pupils in
a regular classroom, and the university students
as part of a questionnaire battery for freshmen).
C on® dentiality was assured verbally and in
writing.
Statistical analyses
The m ain aim of the statistical analyses was to
compare the endorsement of the different expectancies and the prediction of current drinking
across the three subject-groups and gender.
876
Reinou t W . W iers et al.
Before these com parisons could be done in
meaningful way, it had to be shown ® rst that one
measurem ent model could be used across the
subject-groups. Because the present study used a
new expectancy instrument, the analyses were
started with two exploratory factor analyses
(PCAs), to get a rough idea of the factor structure in the secondary school pupils and in the
university students. Then a series of m ulti-group
restrictive factor analyses were conducted to test
whether one comm on factor model held up
across the subgroups. This procedure is not
strictly con® rmatory (the exploratory analyses
were performed on the sam e sam ple, see
JoÈ reskog, 1993), but not trivial because it is
explicitly tested that one com mon model can be
used to compare the subject groups. This test is
more stringent than the standard Box test, because the covariances of two groups can be of the
sam e order of magnitude while the sam e factor
structure does not hold across groups.
After it was shown that one measurement
model could indeed be used across the subgroups, regression analyses were performed for
each subgroup in which it was investigated which
expectancies predicted current drinking. Finally,
the endorsement of the different expectancies
across subgroups was analysed using m ultivariate
analyses of variance (M ANO VA).
R esults
Drinking variables
The two m easures of the average number of
standard drinks consumed in a week, QF (QF is
1
quantity 3 frequency ) were highly correlated
(0.85). In all three subject groups the ª one
guessº measure was signi® cantly lower than the
estimate based upon an average weekday and an
average day at the weekend (paired t -tests: group
1 t (215 df) 5 2.9, p , 0.01, group 2 t (162
df) 5 8.4, p , 0.01, group 3 t (174 df) 5 6.7,
p , 0.01). In subject-group 3 (university students), QF could be compared with a detailed
measure of drinking during the past week. QF
correlated 0.77 with the num ber of standard
drinks consumed during the past week, but was
again signi® cantly lower (12.3 vs. 14.6; t (183
df) 5 2.70, p , 0.01).
The second drinking variable which was assessed in the entire subject-group was the number of days in the past month on which ® ve or
more alcoholic standard drinks were consum ed
(D). This drinking-variable correlated 0.75 with
QF. For each of the subject groups further subdivided with respect to gender, means and standard deviations of the variables concerning
self-reported drinking can be found in Table 1.
A M ANO VA of the drinking variables showed
a signi® cant m ain effect for gender F (2,547
df) 5 22.5, p , 0.001) and subject group
( F 5 22.5; 4,1094 df, p , 0.001), and a
signi® cant interaction between subject group and
gender F (4,1094 df) 5 5.8, p , 0.001). Followup univariate tests showed that the two m ain
effects and the interaction were signi® cant for
both drinking variables (both p , 0.01). Pairwise
comparisons of the harmonic m eans (adjusted
for unequal group sizes, Stevens, 1992, p. 204)
with the Student± Newman± Keuls m ethod
( a 5 0.05) showed that males in groups 2 and 3
drink signi® cantly more standard drinks each
week than girls of the same groups. The
youngest boys and girls consum ed signi® cantly
less than the older subjects. On the number of
drinks on each occasion (Q), the pattern of
results showed one difference: boys of group 2
drink signi® cantly more on each occasion than
m ale university students (and all other subgroups).
E xpectancies: analyses of the measurem ent m odel
Preliminary expectancy scales were constructed
separately for the university students (group 3)
and the secondary school pupils (groups 1 and 2)
by using exploratory factor analysis with oblim in
rotation. Items measuring low and high dose
expectancies were analysed separately. The factor structures in the students and secondary
school pupils appeared to be quite similar. To
enable a test of a common factor model for the
different subject groups (multi-group restrictive
analysis), a second exploratory factor analysis
was done excluding those item s which had been
slightly changed in wording for the adolescents.
The ® rst six exploratory factors on this reduced
item set explained 54% of the variance. These
factors were used to construct six preliminary
scales concerning a low dose of alcohol. Three
scales concerned positive effects of a low dose of
alcohol and three concerned negative effects of a
low dose of alcohol. Following the sam e procedure, two prelim inary comm on scales were
constructed concerning a high dose of alcohol:
high-dose positive and high-dose negative.
H igh- and low-dose expectancies
Unfortunately, the small number of the highdose item s did not allow for speci® c subscales.
Restrictive factor analyses were used to investigate whether the prelim inary common factor
structure ® tted the expectancy data for each of
the subject groups and across gender (with
LISREL VIII, JoÈ reskog & SoÈ rbom , 1993). In
all structural equation models that follow,
maximum likelihood estimation was used.
The covariance matrix of the latent factors of
the prelim inary model (F, see Appendix) was not
positive de® nite in every subgroup. As a solution
to this problem, the model was reduced to seven
factors. The aggression scale was removed because a preliminary m ultiple regression analysis
showed that it did not correlate signi® cantly with
alcohol consumption in any of the subgroups.
The ® nal model consisted of three low-dose
positive expectancy scales (cognitive± m otor enhancement, sexual enhancement, celebration±
group acceptance), two low-dose negative
expectancy scales (sexual± social inhibition and
cognitive± m otor impairm ent), a scale of highdose positive expectancies and a scale of highdose negative expectancies. The ® nal m odel with
all rem aining item s is shown in Fig. 1. The ® t of
this model was adequate in the adolescent sample: F 2 (278 df) 5 531 ( n 5 379), with ® t indices:
RMSEA 5 0.049, AGFI 5 0.88, NN FI 5 0.82
(for an explanation of the ® t indices see note
[2]). The ® t for the student sample was reasonable: F 2 (278 df) 5 421 ( n 5 175), RM SEA 5 0.054, AGFI 5 0.81, NNFI 5 0.80.
Does one com m on factor m odel hold across subgroups? The question addressed in this section is
whether the same factor structure for the expectancies holds in the six subgroups (details concerning the m ulti-group model may be found in
the Appendix). Fitting the com mon model to all
six subgroups in one restrictive m ulti-group
analysis was not possible, because this would
have required a larger number of subjects in each
subgroup. The ® rst m ulti-group analysis concerned the question whether the same factor
structure held for students and adolescents. The
® t was reasonable: F 2 (575 df) 5 991 (n 1 5 175,
n 2 5 379), with ® t indices: RM SEA 5 0.051,
GFI 5 0.84, NN FI 5 0.81. Within the adolescent group, the comparison between subject
group 1 (m axim um age 15) and subject group 2
(secondary school pupils 16 1 ) also resulted in
a moderate ® t, F 2 (575 df) 5 901, (n 1 5 163,
877
n 2 5 216),
RM SEA 5 0.055,
GFI 5 0.83,
NNFI 5 0.77. This is interesting, because in the
youngest age-group the majority did not drink
alcohol (see Table 1). When comparing all m ale
and female subjects the ® t was also found to
be reasonable: F 2 (575 df) 5 999 ( n 1 5 278,
n 2 5 276),
RM SEA 5 0.052,
GFI 5 0.88,
NNFI 5 0.81. Note that for all the multi-group
results, the inform al ® t indices indicated a moderate to adequate ® t (RM SEA between 0.050
and 0.055 2 ), suggesting that a comm on factor
m odel was acceptable for further comparisons
across groups.
Second-order factors. Restrictive factor analysis
was perform ed on a model consisting of the four
types of expectancies as second order factors (see
Fig. 2). This resulted in a reasonable ® t in adolescents, F 2 (288 df) 5 587 ( n 5 379), with ® t
indices RMSEA 5 0.052, AGFI 5 0.87, NNFI 5
0.80 and again a modest ® t for students: F 2 (288
df) 5 472 (n 5 175) RM SEA 5 0.061, AGFI 5
0.79, NN FI 5 0.75. The ® t indices of the model
with second-order factors indicated a somewhat
poorer ® t to the data than the model with only
® rst-order factors. Nevertheless, the model containing second-order factors was used to predict
drinking in addition to the m odel containing
only the ® rst-order factors. The reason is that
under certain circum stances, second-order factors m ay predict drinking better than ® rst-order
factors. The reason for this difference is that the
® rst-order factors contain a residual error com ponent that the second-order factors do not contain (see Fig. 2). The presence of the residual
variance of the ® rst-order factors has the sam e
effect as error in the predictors: it results in a
(downward) bias of the estim ates of the regression coef® cients and a reduction of explained
variance. This is not the case when the residual
error is ® xed to zero, in the case of second-order
factors with only one ® rst-order indicator (here
the two high-dose expectancies and age). A similar model (with two higher-order factors: lowdose positive and low-dose negative) was used in
students in a study by Leigh & Stacy (1993). An
advantage of this procedure is that no general
scale is introduced at the level of the ® rst-order
factors (which correlate highly with other predictors) while the general effect of a type of expectancy can still be investigated (Leigh, 1989; Leigh
& Stacy, 1993).
878
Reinou t W . W iers et al.
Low-dose sexual
enhancement
Low-dose cognitive
and motor
enhancement
Low-dose
positive
expectancies
Low-dose
celebration
group-acceptanc e
+
Low-dose cognitive
and motor
impairment
Low-dose
negative
expectancies
Low-dose
inhibition/
negative mood
Alcohol
use
±
+
High-dose positive
High-dose negative
Age (within
subject-group)
1
High-dose
positive
expectancies
1
High-dose
negative
expectancies
1
Age (within
subject-group)
±
Figure 2. R egression model with second-order latent factors as predictors. The small ellipses in the left column contain the
error components. The bigger ellipses in the second column depict the ® rst-order expectanc y factors (as in Fig. 1) and age within
group. The five ellipses in the third column are the second-order factors used to predict drinking. Note that the high-dose factors
and age are equivalent at the ® rst- and second-order levels. The result is that prediction of drinking is not improved for these
predictors due to the introduction of the higher-order factors. The prediction of drinking can be improved by the introduction
of the second-order factors of the low-dose expectancie s (see Appendix).
Predictio n of alcohol consu mption
The prediction of current alcohol consumption
with the latent expectancy variables was also
performed within the fram ework of structural
equation modelling (LISREL VIII, JoÈ reskog &
SoÈ rbom , 1993). The regression with ® rst-order
factors only is equivalent to a conventional regression analysis. In this procedure nonnorm ality of the dependent variables is undesir-
able. Therefore, both indices of alcohol consumption used (QF and D) were log
transformed. 1 Prediction of the number of standard drinks consumed in an average week (QF)
by the ® rst-order expectancy scales and withingroup age were conducted for each of the six
subgroups separately. In all analyses the sam e
procedure was followed. All regression analyses
were carried out using the scores on the scales
H igh- and low-dose expectancies
derived from the common factor m odel. First, all
expectancies and age were entered as predictors
and QF as dependent variable. The estimates of
the regression coef® cients were judged against
their standard errors to determ ine their
signi® cance. On the basis of these results,
signi® cant predictors were identi® ed and retained in a second regression analysis. The effect
of rem oving the non-signi® cant predictors was
evaluated by inspecting the c 2 goodness-of-® t
indices. Using ® rst-order factors as predictors,
rem oval of the non-signi® cant predictors did not
result in a signi® cant decrease in goodness-of-® t
in any instance. None of the analyses were substantially in¯ uenced by outliers; regression without the outliers resulted in virtually identical
results. Regression analyses for all six subgroups
with ® rst-order latent factors are displayed in
Fig. 3a± f.
Boys 11± 15 years (Fig. 3a). Signi® cant predictors were two low-dose positive expectancies
(sexual enhancement and cognitive± motor enhancement), the low-dose negative expectancy of
sexual and social inhibition, the high-dose negative scale and age. Together these predictors
accounted for 27% of the variance in current
drinking. Expectancies alone (without age) accounted for 23% of the variance. Because of the
large percentage of non-drinkers in the youngest
subject group, drinking boys were analysed separately (n 5 56). Interestingly, apart from age,
only positive expectancies signi® cantly explained
drinking for this group: cognitive± motor enhancement, celebration± tension reduction and
the high-dose positive scale. Including age as
predictor 41% of the variance was accounted for,
and without age 29%.
G irls 11± 15 years (Fig. 3b). Signi® cant predictors were one low-dose negative expectancy scale
(sexual and social inhibition) and age, accounting for 26% of the variance in current drinking.
Expectancy alone (without age) accounted for
14% of the variance. The analysis of the subgroup of drinking girls (n 5 45) showed that,
besides age, one expectancy scale signi® cantly
predicted drinking: cognitive± motor enhancement. It is noteworthy that the regression
coef® cient of this positive expectancy was negative in this subgroup.
Adolescent boys 16 1
(Fig. 3c). High-dose
879
expectancies in this group appeared to be of
prim ary importance in the prediction of current
drinking; all other predictors were not signi® cant
(including age). Positive and negative high-dose
expectancies together accounted for 24% of the
variance in current drinking.
Adolescent girls 16 1 (Fig. 3d). Signi® cant predictors were the two negative expectancies for a
low dose of alcohol (sexual and social inhibition
and cognitive± m otor impairment), one positive
scale (sexual enhancement) and age. Together
these variables accounted for 29% of the variance in current drinking. Expectancies alone
(without age) accounted for 23% of the variance.
M ale student s (Fig. 3e). Signi® cant predictors
were cognitive± m otor enhancement, cognitive±
m otor impairment and the high-dose negative
expectancies, together accounting for 25% of the
variance in current drinking.
Fem ale student s (Fig. 3f). Signi® cant predictors
were sexual enhancement, cognitive± motor impairm ent and the high-dose negative scale, together accounting for 22% of the variance in
current drinking.
Prediction of the second drinking variable, the
number of times subjects drank more than ® ve
alcoholic beverages on one occasion during the
past month (D), was analysed in the same way as
QF. The results were very similar in most subgroups. Only those subgroups where different
predictors were found in com parison with the
prediction of QF are reported. D was explained
by other expectancies for boys in the youngest
subject group. Signi® cant predictors were highdose positive expectancies and high-dose negative expectancies. In the subgroup of drinking
young boys, high-dose positive expectancies and
the low-dose expectancy of celebration± tension
reduction signi® cantly predicted drinking. In
girls in the youngest subject group, D was
signi® cantly predicted by the expectancy of sexual enhancem ent (like QF in the other fem ale
subgroups). For the boys in the second subject
group, D was not only signi® cantly explained by
the two high-dose expectancies (like QF), but
also by the low-dose expectancy of cognitive±
m otor enhancement. In the other subgroups the
signi® cant predictors of D were the same as
those for QF.
Finally, the second-order factors were used to
880
Reinou t W . W iers et al.
(a) Boys 11-15 years
(b) Girls 11-15 years
SEX+
SEX+
CM+
0.
17
0. 2
CEL+
CM+
7
CEL+
CM ±
INH ±
HD+
HD ±
± 0.16
QF
INH ±
. 23
± 0
19
0. 27% variance QF
explained
without age: 23%
AGE
CM ±
(c) Boys 16+ years
HD ±
CM+
CEL+
CEL+
CM ±
QF
24% variance QF
explained
(e) Male students
0.
29
± 0.35
± 0.23
QF
0.
23
AGE
29% variance QF
explained
without age: 23%
(f) Female students
SEX+
SEX+
0. 2
CM+
3
CEL+
0.
37
CEL+
CM ±
± 0.24
.3
± 0
QF
1
HD ±
AGE
26% variance QF
explained
without age: 14%
HD+
HD ±
AGE
HD+
CM ±
INH ±
3
0.4
. 26
± 0
HD ±
INH ±
34
(d) Girls 16+ years
CM+
CM+
0.
AGE
SEX+
HD+
QF
HD+
SEX+
INH ±
± 0.34
CM ±
± 0.20
QF
INH ±
HD+
.3 0
± 0
HD ±
25% variance QF
explained
AGE
22% variance QF
explained
Figure 3a± f. Regression models for each of the six subgroups as de® ned by gender and subject group (see Table 1). In each
of the six ® gures, the small ellipses in the left column denote the residuals. The bigger ellipses in the second column depict the
® rst-order expectanc y factors and age (within the subgroup). Legend for the expectancie s (as in Fig. 1): sex 1 5 sexual
enhance ment for a low dose of alcohol; cm 1 5 cognitive± motor enhance ment for a low dose of alcohol, cel 1 5 celebration±
group acceptanc e for a low dose of alcohol, cm 2 5 cognitive± motor impairment for a low dose of alcohol, inh 2 5 sexual and
social inhibition for a low dose of alcohol, hd 1 5 positive expectanci es for a high dose of alcohol, hd 2 5 negative expectancie s
for a high dose of alcohol. The dependen t variable is QF: the number of standard alcoholic consumptions consumed per week
1
(log-transformed ), with the small ellipse depicting the residual. O nly signi® cant predictors of QF are depicted with the arrows
from the expectanci es and age. Details of the analyses are given in the text and in the Appendix.
H igh- and low-dose expectancies
predict drinking (QF) in all subgroups. In the
regression of the second-order factors upon QF,
the same scale scores were used as above, but the
second-order factors were de® ned as predictors
in the LISREL model (see Fig. 2). Note that
there is a 1 ;1 relationship between three ® rst-order factors and their second-order equivalents:
the two high-dose expectancies and age. As indicated above, the explained variance will not be
increased for these variables by using the secondorder factors. The results were dif® cult to compare across the different subgroups for two
reasons. First, in the smallest subgroup (girls of
subject group 2), the m odel did not converge.
Secondly, the ® t indices indicated an inadequate
® t of this model to the data in some subgroups.
As a solution, group-speci® c alterations to the
model had to be introduced. These alterations
will be explained.
In the youngest male subject group, 48% of
the variance in current drinking was accounted
for by two signi® cant second-order predictors:
low-dose positive and low-dose negative expectancies; m odel ® t: c 2 (18 df) 5 20, p 5 NS. Note
that a signi® cant p denotes a failure of the model
to ® t the data (see JoÈ reskog, 1993). In girls of the
youngest subject group 29% of the variance in
current drinking was accounted for by the second-order low-dose negative expectancies and
age, model ® t: c 2 (19 df) 5 30.6, p 5 0.045. The
LISREL VIII output indicated (ª M odi® cation
Indicesº , JoÈ reskog & SoÈ rbom, 1993) that the
speci® c alteration to the model required in this
subgroup was a direct path from the residual of
sexual± social inhibition to QF (a sim ilar alteration as described in Leigh & Stacy, 1993). For
boys in the second subject group, prediction of
current drinking was increased to 30%. Two
predictors were signi® cant: high-dose positive
expectancies and the second-order low-dose
negative expectancies (adequate ® t: c 2 [19 df] 5
29.9, p 5 NS). In m ale students the ® t of the
second-order model was inadequate ( c 2 [19
df] 5 41.1, p , 0.01). M odi® cation indices indicated that the residual errors of two ® rst-order
factors (sexual enhancement and sexual± social
inhibition) should be allowed to correlate. W hen
the m odel was modi® ed in this way, the ® t was
adequate ( c 2 [18 df] 5 25.4, p 5 NS) and the
variance in current drinking accounted for was
32% with low-dose positive and low-dose negative expectancies as signi® cant predictors. In female students the original second-order model
881
also indicated an inadequate ® t (c 2 [16 df] 5
32.6, p , 0.01). M odi® cation indices indicated
that in this subgroup the residual error of age
was correlated with the residual errors of sexual
enhancement, sexual± social inhibition and cognitive± motor impairment. When the model was
m odi® ed in this way, the ® t was good ( c 2 [15
df] 5 15.7, p 5 NS). Low-dose positive, low-dose
negative and high-dose negative expectancies
were the signi® cant second-order predictors that
together accounted for 42% of the variance in
current drinking.
D ifferences in the endorsem ent of the four types of
expectancies between subject groups
Given that a comm on factor model ® tted adequately across subgroups and that different types
of expectancies associated with drinking in these
groups, the next issue concerned differences between subject groups in the endorsement of the
expectancy scales. M ultivariate analyses of variance were perform ed on the four types of expectancies (the second-order factors) for the three
groups by gender (3 3 2 MANOVA). Both m ain
effects were signi® cant: the expectancies differed
between males and females F (4, 545 df) 5 6.8,
p , 0.001,
and
between
the
groups
F (8, 1090) 5 20.2, p , 0.001, with a signi® cant
interaction F (8, 1090) 5 3.6, p , 0.001. Pairwise
comparisons of the harm onic means were conducted
with
the
Student± Newman± Keuls
m ethod (c 5 0.05).
Low-dose positive expectancies. No signi® cant
gender differences were found on low-dose positive expectancies in the youngest group, nor in
university students. In contrast, boys of group 2
m ore often endorsed low-dose positive expectancies than girls of the sam e group. These expectancies were mostly endorsed by university
students and boys of group 2, followed by girls of
the second group. The youngest group showed
the lowest endorsement of positive expectancies
for low dose of alcohol.
Low-dose negative expectancies. The opposite
pattern was found for low-dose negative expectancies: they were most endorsed by boys and
girls of group 1 (no gender difference), followed
by girls of group 2. The lowest level of endorsem ent was found in boys of group 2 and university students (m ale and fem ale).
882
Reinou t W . W iers et al.
H igh-dose positive expectancies. Boys of group 2
endorsed
high-dose
positive
expectancies
signi® cantly more than females of group 2 and 3.
No further signi® cant differences were found.
H igh-dose negative expectancies. Boys and girls
of group 2 endorsed high-dose negative expectancies signi® cantly less in com parison with all
other
subgroups.
No
further
signi® cant
differences were found.
H igh-dose expectancies of boys of group 2. The
pattern of results suggested that boys of the
second subject group showed an extreme pattern
of expectancy endorsem ent, with the negative
expectancies endorsed least and positive expectancies m ost in com parison with other groups.
This suggestion was tested with a contrast comparing boys of group 2 with the other three
subgroups with somewhat overlapping ageranges: girls of the same group, male and female
university students. It was found that boys of
group 2 showed the highest level of endorsement
of positive expectancies for a high-dose and the
lowest level of endorsement of negative expectancies for a high dose (p , 0.001). This is illustrated in Fig. 4. The same contrast was also
signi® cant for low-dose positive expectancies
(p 5 0.020), but not for low-dose negative expectancies. The extreme endorsement of the highdose expectancies in boys of group 2 paralleled
their high average number of drinks on each
drinking occasion (Q in Table 1), which was also
signi® cantly higher for boys of the second subject
group in com parison with these other three
subgroups (same contrast, p , 0.001).
Differences with alcohol consu m ption as a
covariate. The general pattern of results was that
a group with a higher alcohol consumption more
endorsed positive expectancies and less endorsed
negative expectancies in comparison with groups
with a lower alcohol consumption. Further
analyses showed, however, that not all differences in expectancies between the subgroups
could be explained by current drinking pattern.
When current weekly alcohol consumption was
introduced as a covariate, the main effect for
subject group was still signi® cant F (8,1088 df) 5
14.6, p , 0.001, the main effect for gender was
no longer signi® cant F (4,544) 5 2.2, p 5 NS, and
the interaction between gender and subject
group was still signi® cant F (1088) 5 2.3,
p 5 0.019. The pattern of results remained the
same when average alcohol consumption on each
occasion (Q) was introduced as a covariate.
Differences in the endorsem ent of expectancies of
youn g drinkers and non-drink ers. Because approxim ately half the subjects in the youngest subject
group reported experience with alcohol ( n 5 115)
and the other half did not ( n 5 101), the differences in the endorsement of the four types of
expectancies between drinkers and non-drinkers
in the youngest group was tested. Because the
non-drinkers were found to be signi® cantly
younger, t (214 df) 5 3.4, p , 0.01, age was
introduced as a covariate in the analysis (2 3 2
M ANC OVA). The main effect for gender was
not signi® cant F (4,208 df) 5 0.14, p 5 NS, nor
was the interaction between gender and drinking
status F (4,208 df) 5 1.4, p 5 NS. The main effect for drinking status was signi® cant, F (4,208
df) 5 6.8, p , 0.001. Univariate follow-up tests
indicated that the non-drinkers signi® cantly
m ore endorsed negative expectancies for a low
dose of alcohol, F (1,211 df) 5 21.1, p , 0.001,
and negative expectancies for a high dose of
alcohol F (1,211 df) 5 5.2, p 5 0.023.
D iscussion
The ® rst important ® nding of this study was that
the factor structure of the alcohol-related expectancies was similar across different subject
groups and gender. With multi-group restrictive
factor analyses, it was shown that one model
with seven ® rst-order factors ® tted reasonably
well across subgroups. A second model with four
second-order expectancy factors also ® tted
across subgroups. These second-order factors
were the four types of expectancies, distinguished by valence (positive or negative) and
dose of alcohol (a few drinks or many drinks).
This allowed for further comparisons of the prediction of current alcohol consum ption and of
the differential endorsement of the four types of
expectancies across subgroups.
The m ain ® nding of this study was that different expectancies predicted current drinking
across the subgroups. Positive expectancies for a
high dose of alcohol, for example, did not
signi® cantly predict current drinking in students,
which is in agreement with several studies (e.g.
Fromm e et al ., 1993; Leigh & Stacy, 1993).
H igh- and low-dose expectancies
883
Positive expectancies for a high dose
(a)
4.5
4.25
4.0
3.75
3.5
3.25
1
2
Group
3
1
2
Group
3
Negative expectancies for a high dose
(b)
11.0
10.5
10.0
9.5
9.0
8.5
8.0
Figure 4. Differences in (a) high-dose positive and (b) high-dose negative expectancie s between subject-groups and gender
( e males; h females).
However, it was found that these ª least plausibleº expectancies were the strongest predictor
of drinking in secondary school boys of 16
and older. In addition, these expectancies
signi® cantly predicted drinking in secondary
school boys of the youngest subject-group who
had initiated drinking. A comparison between
the older secondary school pupils and students
884
Reinou t W . W iers et al.
suggested that male and female university students and secondary school girls of 16 and older
maintained a pattern of social drinking due to
their relatively strong negative expectancies for a
high dose and their relatively weak positive expectancies for a high-dose of alcohol. The boys
in the second subject group, on the other hand,
showed the reverse pattern of high-dose expectancy endorsement (highest on positive expectancies for a high dose and lowest on negative
expectancies for a high dose of alcohol). This
paralleled their current drinking pattern: they
reported the highest average alcohol consumption on each occasion (6.5 standard drinks).
The comparison of the expectancies of
drinkers and non-drinkers in the youngest subject group showed that non-drinkers scored
signi® cantly higher on negative expectancies (for
a low and a high dose), but not on positive
expectancies. Interestingly, in those boys of the
youngest group who had initiated drinking, positive expectancies predicted the amount of alcohol
they consumed. This pattern of results suggested
that the relevant positive expectancies were acquired before alcohol consumption actually commenced, which is in agreement with the ® ndings
of Christiansen et al ., (1989) and M iller, Smith
and Goldm an (1990). Negative expectancies appeared to be moderated as a result of drinking
experience. A similar pattern was found for the
comparison between the younger boys in secondary education and the older boys in secondary education: the younger boys held stronger
negative expectancies than the older boys, both
for a high and for a low dose of alcohol. In
contrast, no signi® cant difference was found between these two subgroups on the positive high
dose expectancies that most strongly predicted
drinking in boys of group 2. Again, the pattern of
results suggested that the most relevant positive
expectancies did not alter much as a result of
increased drinking, but the negative expectancies
did.
There are several lim itations in the present
study. First, high-dose expectancies were introduced on an exploratory basis. Therefore, it was
not possible to construct speci® c high dose scales
beyond ª positiveº and ª negativeº high-dose expectancies. The introduction of more speci® c
scales is important for two reasons: to arrive at a
more precise understanding of these expectancies in those groups where they are important
predictors of current risky drinking patterns. In
addition, the introduction of speci® c high-dose
scales would make the second-order factor structure more powerful and would enable a more
accurate comparison of the predictive power of
high- and low-dose expectancies in different
samples.
A second lim itation of the present study concerns the form at of the expectancy questionnaire, which m easures the effects of alcohol on
an average person. It has been argued that this
m ay not be the most appropriate way to measure
expectancies in subgroups primarily consisting of
experienced drinkers (here the older secondary
school pupils and the university students). However, in a direct com parison between the two
form ats in college students no difference was
found (W ood et al., 1992). The advantage of the
general form at was that (young) non-drinkers
could be included in the study and that, therefore, the results could be compared across subgroups, as well as with results obtained with the
widely used AEQ-A (Christiansen et al., 1982).
Nevertheless, a replication study with a questionnaire measuring the four types of expectancies in
a personal format would be desirable. This especially concerns the ® ndings on the importance
of high-dose expectancies for predicting
ª weekend-bingeingº in secondary school boys of
16 and older. Another possible extension of the
present questionnaire is to measure separately
expected effects and the desirability of the
effects. This m ight be an interesting way to
proceed. However, in the present set-up subjects
would have to ® ll in some items four times: for
a low dose and for a high dose of alcohol,
concerning expectancy and desirability. This
m ight be a rather taxing procedure for younger
subjects.
The results of this study have implications for
the prevention of problem drinking in adolescents. C hallenging positive expectancies would
seem a better prevention strategy than further
strengthening negative expectancies which appear to be already unrealistically high before
drinking has started. Darkes & Goldm an (1993)
showed that an experim ental challenge of (lowdose) positive expectancies in m ale problem
drinkers (students) was effective. Present results
suggest that the prevention of problem drinking
in adolescent boys could be improved by challenging their positive high-dose expectancies. The
m ore established strategy of enhancing negative
(high-dose) expectancies could also be of
H igh- and low-dose expectancies
relevance in adolescents at risk. These expectancies appear to moderate drinking successfully in
students and are relatively weak in adolescent
boys with a high alcohol consumption on each
drinking occasion. Learning to distinguish for
dose effects could be a helpful strategy in prevention of problem drinking. An application of the
ª extended balanced placebo designº in a challenge could be helpful in this respect (Lapp et al .,
1994).
The results for the older adolescent boys obtained in this study can be related to models of
risks for alcoholism (Zucker, 1987). The development of an individual’ s high dose expectancies
could be an important factor in determ ining the
outcome of a period of high alcohol consumption in late adolescence: returning to m oderate
levels of consum ption without professional help
(a ª developmentally limitedº alcohol problem ,
Zucker, 1987) or developing a m ore serious alcohol problem . The interaction of expectancies
with other risk factors seems to be of great
importance in understanding vulnerability for alcoholism (Sher, 1991), and constitutes a main
goal in our ongoing study with young children of
alcoholics (Wiers et al ., 1994). In this line of
research the recent ® ndings of Deckel, Hesselbrock & Bauer (1995) are particularly interesting: neuropsychological measures of frontal
cortex functioning predicted positive expectancies in young-adult children of alcoholics. Impaired functioning of the frontal cortex has been
related to the genetic vulnerability for alcoholism
(e.g. Pihl, Petersen & Finn, 1990), and might
in¯ uence the development of high-dose expectancies. Another important risk factor that could
be related to the development of (high-dose)
expectancies concerns the subjective response to
an alcohol challenge in early adulthood which
was found to be a strong predictor of future
alcoholism (Schuckit & Smith 1996). Longitudinal evidence for the reciprocal in¯ uence of alcohol experience and expectancies has been
found (Smith et al ., 1995).
The main conclusion of this study is that the
measurem ent of dose-related alcohol expectancies is a prom ising addition in future research
with alcohol-related expectancies, especially
when other populations than university students
are studied. Including the ª implausibleº positive
expectancies for a high dose of alcohol could be
important in relation to the prevention of alcohol
problems. Further longitudinal studies are re-
885
quired including the four types of expectancies
and other individual risk factors.
A cknow ledgem ents
Special thanks to C. Dolan for help with structural equation modelling, and to T. Neve for
data collection. We further gratefully acknowledge the helpful comments of K. Sher, J.
Elshout, S. Huismans, H. Vorst, W. van den
Brink, and those of two anonymous reviewers.
W e would also like to thank M . Goldman for
promptly sending us requested m aterials and
SGO for funding the research.
N otes
[1]
The following variables are de® ned:
5
NW D
DRINKSW D 5
5
NW ND
DRINW ND 5
5
W EEKG SS
NDDrunk2w 5
the number of weekdays per
m onth (M onday± Thursday) on
which alcohol is consum ed (0±
16).
the number of standard drinks
on an average weekday on which
alcohol isconsum ed (as sum m ed
from the beverage checklist).
the num ber of w eekend-days per
m onth
(Friday± Sunday)
on
which alcohol is consum ed (0±
12).
the number of standard drinks
on an average weekend-day on
which alcohol is consum ed (as
sum med from the beverage
checklist).
the number of standard drinks a
subject reports to drink in an average week.
the num ber of days in the past 2
weeks on which the subject
drank ® ve or m ore standard
alcoholic drinks on one occasion
The estim ate of weekly alcohol consum ption
(Q F, or quantity 3 frequency) was calculated as
follows:
(1) QF 5 [{(NW D*DRINKSW D)
1 (N W ND*DRINW ND)/4}
1 W EEKG SS]/2.
The estim ate of the average number of drinks per
drinking occasion w as calculated as follows:
(2) Q 5 QF/F
where F denotes the frequency of drinking:
(3) F 5 {(NW D 1 NW ND)/4}/2.
For non-drinkers (F 5 0), Q was set to zero.
The number of days intoxicated in the past
m onth was calculated as follows:
(4) D 5 2* NDDrunk2w.
All alcohol consumption variables w ere log transform ed in the following way:
886
[2]
Reinou t W . W iers et al.
(5) lnDrinkVar 5 ln(DrinkVar 1 1),
where one is added to m ake the inclusion of
non-drinkers possible.
A problematic aspect of a transformation is
that it m ight obscure the interpretation of the
outcom es (Stevens, 1992) . H ere, however, the
interpretation is fairly straightforward: ln(drinks 1 1) equals zero indicates non-drinkers,
ln(drinks 1 1) equals one indicates about three
drinks a week, ln(drinks 1 1) equals about seven
drinks a w eek, ln(drinks 1 1) equals about 19
drinks a week, and ln(drinks 1 1) equals about 50
drinks a week, or seven drinks a day.
To assess the m odel-® t, several indices can be
used. H owever, it is known that ª statistical goodness-of-® t tests are often m ore a re¯ ection on the
size of the sample than on the adequacy of the
modelº (Browne & Cudeck, 1993, p. 137). It is
therefore recom m ended to concentrate on errors
of approxim ation such as the root mean square
error of approximation, or RM SEA (Browne &
Cudeck, 1993; JoÈ reskog, 1993). A RM SEA of
below 0.05 indicates a close ® t of the model in
relation to the degrees of freedom , which is ª no
less subjective than the choice of 5% as a
signi® cance levelº (Browne & Cudeck, 1993, p.
144 ± 147). To facilitate comparison with other
structural equation m odels of expectancy, the
adjusted goodness of ® t index (AG FI, as in
Fromme et al., 1993) is given and the NNFI
(non-normed ® t index, as in Leigh & Stacy,
1993) . In m ulti-group analyses the GFI is given
instead of the AG FI. In the m ulti-group analyses
the corrected RM SEA is given, which is equal
to the RM SEA given by LISREL 8.12 tim es
the square root of the number of groups (S.
Gregorich,
personal
comm unication
on
S M N E T , S E M N E T @ U A 1 V M . U A .E D U . ,
February 13, 1997) .
R eferences
B ROW N , S. A., G O LDM AN , M . S., I NN , A. & A NDERSO N ,
L. (1980 ) Expectations of reinforcem ent from alcohol: their domain and relation to drinking patterns, Journal of Consulting and Clinical Psychology,
48, 419± 426.
B ROW NE , M . W . & C U DECK R. (1993 ) Alternative w ays
of assessing m odel ® t, in: B OLLEN , K. A. & S CO TT
L O NG , J. (Eds) Testing Structural Equation Models
(N ew bury Park, Sage).
C HEN , M. J., G RU BE , J. W . & M ADDEN , P .A. (1994 )
Alcohol expectancies and adolescent drinking: differential prediction of frequency, quantity, and intoxication, Addictive Behaviors, 19, 521± 529.
C HRISTIANSE N , B. A., G OLDM AN , M. S. & INN , A.
(1982 ) Developm ent of alcohol-related expectancies
in adolescents: separating pharm acological from social-learning in¯ uences, Journal of Consulting and
Clinical Psychology, 50, 336± 344 .
C HRISTIANSE N , B. A., S M ITH , G . T., R OEHLIN G , P. V. &
G OLDM AN , M. S. (1989 ) U sing alcohol expectancies
to predict adolescent drinking behavior after one
year, Journal of Consulting and Clinical Psychology, 57,
93± 99.
C OLLINS , R. L., L APP , W . M ., E M M ONS , K. M. & I SAAC ,
L. M. (1990 ) Endorsement and strength of alcohol
expectancies, Journal of Studies on Alcohol, 51, 336±
342.
C O NNORS , G . J., O’ F ARRELL , T. J., C UTTER , H. S. G. &
L OG AN- T HO M PSO N , D. (1987 ) Dose-related effects
of alcohol among m ale alcoholics, problem drinkers,
and nonproblem drinkers, Journal of Studies on
Alcohol, 48, 461 ± 466.
D ARKES , J. & G OLDM AN , M . S. (1993 ) Expectancy
challenge and drinking reduction: experim ental evidence for a m ediational process, Journal of Consulting
and Clinical Psychology, 61, 344± 353 .
D EC KEL , A. W ., H ESSELB ROCK , V. & B AUER , L. (1995)
Relationship between alcohol-related expectancies
and anterior brain functioning in young m en at risk
for developing alcoholism, Alcoholism: Clinical and
Experimental Research, 19, 476 ± 481.
E ARLEYW INE , M . & M ARTIN , C. S. (1993 ) Anticipated
stim ulant and sedative effects of alcohol vary w ith
dosage and lim b of the blood alcohol curve, Alcoholism, Clinical and Experimental R esearch, 17, 135± 139 .
F ROM M E , K., S TROOT , E. & K APLAN , D. (1993 ) Comprehensive effects of alcohol: development and
psychom etric assessm ent of a new expectancy questionnaire, Psychological A ssessment, 5, 19± 26.
G OLDM AN , M . S., B RO WN , S. A., C HRISTI ANSE N , B. A.
& S M ITH , G . T. (1991 ) Alcoholism and m emory:
broadening the scope of alcohol-expectancy research, Psychological Bulletin, 110, 137 ± 146.
G RU BE , J. W ., A M ES , G. M . & D ELANEY , W . (1994)
Alcohol expectancies and w orkplace drinking,
Journal of A pplied Social Psychology, 24, 646± 660.
G RU BE , J. W ., C HEN , M . J., M ADDEN , P. & M ORGAN ,
M. (1995 ) Predicting adolescent drinking from alcohol expectancy values: a com parison of additive,
interactive, and nonlinear m odels, Journal of Applied
Social Psychology, 25, 839± 857.
H OOG EVEEN , K. J. (1994 ) Alcohol verw achtingen en
alcohol gebruik bij jeugdigen in Nederland; ontwikkeling van een Nederlandse Alcohol Verwachtingen Vragenlijst [Alcohol expectancies and
alcohol use in youngsters in the Netherlands;
developm ent of a Dutch Alcohol Expectancy
Questionnaire],
Master’ s
thesis,
Am sterdam :
University of Am sterdam , Faculty of Psychology.
J ONES , B. T. & M C M AHON , J. (1994 ) Negative and
positive alcohol expectancies as predictors of abstinence after discharge from a residential treatm ent
programm e: a one- and three-m onth follow-up study
in males, Journal of Studies on Alcohol, 55, 543± 548 .
J OÈ RESKOG , K. (1993) Testing Structural Equation
Models, in: B OLLEN , K. A. & S CO TT L ONG , J. (Eds)
Testing Structural Equation Models (N ewbury Park,
Sage).
J OÈ RESKOG , K. G . & S OÈ RBO M , D. (1993) LISREL VIII
User’ s Reference Guide (C hicago, Scienti® c Software
International Inc.).
K NIG HT , R. G. & G ODFREY , H. P. D. (1993 ) The role
of alcohol-related expectancies in the prediction of
drinking behaviour in a sim ulated social interaction,
Addiction, 88, 1111± 1118.
L APP , W . M., C O LLINS , R. L., Z YW IAK , W . H. & I ZZO ,
C. V. (1994 ) Psychopharm acological effects of alcohol on tim e perception: the extended balanced
H igh- and low-dose expectancies
placebo design, Journal of Studies on Alcohol, 55,
96± 112.
L EIG H , B. C. (1987 ) Beliefs about the effects of alcohol
on self and others, Journal of Studies on Alcohol, 48,
467± 475.
L EIG H , B. C. (1989a ) In search of the seven dwarves:
issues of m easurem ent and m eaning in alcohol expectancy research, Psychological Bulletin, 105, 361±
373.
L EIG H , B. C. (1989b ) Attitudes and expectancies as
predictors of drinking habits: a com parison of three
scales, Journal of Studies on Alcohol, 50, 432± 440 .
L EIG H , B. C. & S TAC Y , A. W . (1991 ) On the scope of
alcohol expectancy research: rem aining issues of
m easurem ent and m eaning, Psychological Bulletin,
110, 147 ± 154.
L EIG H , B. C. & S TACY , A. W . (1993 ) Alcohol outcom e
expectancies: scale construction and predictive utility in higher order con® rmatory factor m odels,
Psychological Assessment, 5, 216 ± 229.
M C M AHO N , J., JO NES , B. T . & O’ D ONNELL , P. (1994)
Comparing positive and negative alcohol expectancies in m ale and fem ale social drinkers, Addiction
Research , 1, 349± 365 .
M ILLER , P. M ., S M ITH , G . T. & G OLDM AN , M. S.
(1990 ) Emergence of alcohol expectancies in childhood: a possible critical period, Journal of Studies on
Alcohol, 51, 343± 349 .
M O ONEY , D. K., & C ORC ORAN , K. J. (1991 ) The relationship between assertiveness, alcohol-related expectations for social assertion, and drinking patterns
among college students, Addictiv e Behaviors, 14,
301± 305.
M O ONEY , D. K., F ROM M E , K., K IVLAHA N , D. R. &
M ARLATT , G. A. (1987 ) Correlates of alcohol consum ption: sex, age, and expectancies related differentially to quantity and frequency, Addictive
Behaviors, 12, 235± 240 .
P IHL , R., P ETERSO N , J. & F INN , P. (1990 ) Inherited
predisposition to alcoholism : characteristics of sons
of male alcoholics, Journal of Abnormal Psychology,
99, 291± 301
R OHSENO W , D. J. (1983 ) Drinking habits and expectancies about alcohol’ s effect for self versus others,
Journal of Clinical and Consulting Psychology, 51, 752±
756.
S C HU CKIT , M. A. & S M ITH , T. L. (1996 ) An 8-year
follow -up of 450 sons of alcoholic and control
subjects, A rchives of G enera l Psychiatry, 53, 202± 210.
S H ER , K. J. (1991 ) Children of Alcoholics, a critical
appraisal of theory and research (Chicago, U niversity
of Chicago Press).
S H ER , K. J., W ALITZE R , K. S., W OO D , P. K. & B RENT ,
E. E. (1991 ) Characteristics of children of alcoholics:
putative risk factors, substance use and abuse and
psychopathology, Journal of Abnormal Psychology,
100, 427 ± 448.
S M ITH , G. T., G O LDM AN , M . S., G REENBA U M , P. E. &
C HRISTIANSE N , B. A. (1995 ) Expectancy for social
facilitation from drinking: the divergent paths of
high-expectancy and low-expectancy adolescents,
Journal of Abnormal Psychology, 104, 32± 40.
S O BELL , L. C. & S OBELL , M . B. (1978 ) Validity of
self-reports in three populations of alcoholics,
887
Journal of Consulting and Clinical Psychology, 46, 901±
907.
S OBELL , L. C. & S OBELL , M. B. (1990 ) Self-report
issues in alcohol-abuse: state of the art and future
directions, Behavioral A ssessment, 12, 77± 90.
S OBELL , L. C. & S O BELL , M. B., L EO , G . I. & C ANCILLA , A. (1988 ) Reliability of a tim eline method:
assessing normal drinkers’ reports of recent drinking
and a com parative evaluation across several populations., British Journal of Addiction, 83, 393± 402 .
S OU THW IC K , L., S TEELE , C., M ARLATT , A. & L INDELL ,
M. (1981 ) Alcohol-related expectancies: de® ned by
phase of intoxication and drinking experience,
Journal of Consulting and Clinical Psychology, 49, 713±
721.
S TACY , A. W ., W IDAM AN , K. F. & M ARLATT , G . A.
(1990 ) Expectancy m odels of alcohol use, Journal of
Personality and Social Psychology, 58, 918± 928 .
S TACY , A. W ., N EW COM B , M . D. & B ENTLER , P. M.
(1991 ) Cognitive motivation and drug use: a 9-year
longitudinal study, Journal of Abnormal Psychology,
100 , 502± 515 .
S TEVENS , J. (1992) A pplied M ultivariate Statistics for the
Social Sciences, 2nd edn (Hillsdale, NJ, Lawrence
Erlbaum).
W IERS , R. W ., S ERG EANT , J. A. & G UNNING , W . B.
(1994 ) Psychological mechanism s of enhanced
risk of addiction: a dual pathway? Acta Pediatrica
Supplement, 404, 9± 13.
W OOD , M . D., N AG OSHI , C. T . & D ENNIS , D. A. (1992 )
Alcohol norm s and expectations as predictors of
alcohol use and problem s in a college student sam ple, Am erican Journal of Drug and A lcohol Abuse, 18,
461± 476.
Y OU NG , R. M C D. & K NIG HT , R. G. (1989 ) T he drinking expectancy questionnaire: a revised measure of
alcohol-related beliefs, Journal of Psychopathology and
Behavioral Assessment, 11, 99± 112.
Z U CKER , R. A. (1987 ) The four alcoholism s: a developmental account of the etiological process, in: R IVERS ,
P. C. (E d.) Nebraska Symposium on Motivation, 1986:
alcohol and addictive behavior, pp. 27± 83 (L incoln,
University of Nebraska Press).
A ppendix
S
A . Multi-group restrictive ® rst order factor analyses in
LISREL
In the multi-group restrictive ® rst order factor analysis,
the following m odel w as speci® ed in each of the
subject-groups used in that analysis:
i5
S
iS
t
1
H
i
(1)
H ere S is a m atrix of factor loadings, S is the covariance m atrix of the (zero mean) factors, and H i is the
diagonal covariance m atrix of the (zero m ean) errors,
w ith the subscript i indicating the subject-group. The
structure of S was speci® ed a priori, because each item
w as assigned to a predetermined factor (in contrast to
exploratory factor analysis) Note that S has no groupindicator because it was constrained to be invariant
across groups. For reasons of identi® cation, one indi-
888
Reinou t W . W iers et al.
cator of each factor was ® xed to 1 in all subgroups.
The errors in H i were allow ed to vary across subjectgroups, as w as the covariance m atrix Y i. All free
parameters in Y, Y i and H i were estim ated using the
LISREL program .
Y
B. Second-order (multi-group) factor analysis in LISREL
The object of second order factor analysis is to explain
the covariance of the ® rst order factors by postulating
a smaller num ber of second order factors. The m odel
used to this end is:
i5
Y
(Y
iY
t
1
Y
i)
Y
t
1 H
i
(2)
Here Y contains factor loadings on ® rst order factors,
Y contains factor loadings of ® rst order factors on
second order factors, Y i is the covariance matrix of the
second order factors, Y i is the diagonal covariance
m atrix containing the residual variances, i.e. variances
of the ® rst order factors that are not explained by the
second order factors, and H i is again the diagonal
covariance m atrix of the errors. The covariance m atrix
of the ® rst order factors is now m odelled as
t
(Y
1 Y i). Note that in this analysis both the ® rst
iY
and second order factor loadings (Y
and Y) w ere
estim ated under the restriction that these parameters
w ere invariant across the subject-groups. The param eters in Y i , Y i and H i were allowed to vary over the
subject-groups. For reasons of identi® cation, again one
indicator of each ® rst order factor was ® xed to 1 in all
subgroups, and Y i and Y i were standardized in one
subject-group.