The Validity of Self-Reports of Smoking: Analyses by Race/Ethnicity in a School Sample of Urban Adolescents

Abstract

This research compared the validity of self-reports of cigarette smoking for African-American, Hispanic, and White respondents. Previous research has raised a question about the validity of self-report for African Americans.
A self-report of cigarette smoking was obtained together with a measure of carbon monoxide from expired air. Convergence between self-reported smoking and the biochemical measure was analyzed separately for three ethnic groups at 7th grade, 8th grade, 9th grade, and 10th grade.
Analyses indicated that the validity of self-reports of smoking was generally comparable across ethnic groups. Sensitivity and specificity were comparable with data reported in recent meta-analyses. Though sensitivity was slightly lower for minority adolescents than for White adolescents, prevalence rates corrected for group differences in sensitivity showed significantly lower smoking rates for African-American and Hispanic adolescents than for White adolescents.
The lower smoking rates reported for African-American adolescents are real and are not substantially a consequence of reporting artifacts.

Full text

The
Validity
of
Self-Reports
of
Smoking:
Analyses
by
Race/Ethnicity
in
a
School
Sample
of
Urban
Adolescents
Thomas
A.
Wills,
PhD,
and
Sean
D.
Cleary,
PhD,
MPH
Introduction
.. 'f
.: .
:- . ..
..
. . ..
: .
*: .:
- : ..
. . .
. . :
.
.:
*:.
::
:
.....
.. ; .......
... .
... .
.. . .
.
...
, j0: .i
.: .. '
. . . . .
. .. ...
. : .
*' ,
- ; ...
.:
...
.. .
. . i
,"^ .X
. ,ffl,
.... ..
.. ...
.
..
,X0<EDo.ff4
: ......
f
.fE '
,0"'
'i
X
..
..
....
..
.:
:.
*:
:.:
::
::;:
:.
.:
:.
:.
.:
.;:
...
..
::
.. . ::
:.:: ::
. .. ..
.. : ...
.. .
..
.:
.... i.;.
;. :.e.::
....... ..
... .. .
. . . ...
.... ::
.. : . .s
{: ..
f: : . :.
.. : :.:
:.::
. ... :
:::
::
...
::::
: : :
. :.
:.: .
. ...
. ::
. :
::.:
.....
.
.
. .
.:
*: :
a.; ..
....
: -
. .
:
:.
*: : . ... t: r :
*;
:::
:t::
..
Y '
ShG
hf
J
S
; .....
..
..
. ...
t .
Epidemiologic
research
on
adoles-
cent
substance
use
has
investigated
the
predictors
of
cigarette
smoking
and
other
substance
use
in
large
community
samples.
1"2
Research
based
on
self-reports
of
cigarette
smoking,
alcohol
use,
and
marijuana
use
has
consistently
shown
lower
rates
of
use
among
African-
American
adolescents
than
among
Whites.
This
research
is
based
on
studies
using
self-report
questionnaires
or
interview
methods,
in
which
subjects
are
assured
of
the
confidentiality
of
their
responses.
Similar
results
have
been
found
both
in
regional
samples
and
in
national
samples.3-8
A
recent
paper
by
Bauman
and
Ennett
raised
a
question
about
the
intemal
validity
of
this
literature.9
Using
data
from
a
sample
of
younger
adolescents
from
the
southeastem
United
States,
these
investi-
gators
compared
self-reports
of
cigarette
smoking
with
measurements
of
carbon
monoxide
(CO)
from
expired
breath.
In
this
type
of
research
with
adolescents,
where
rates
of
regular
smoking
are
low,
higher
rates
are
expected
for
self-report
because
the
half-life
of
CO
is
about
8
hours;
hence
this
biochemical
measure
will
be
negative
for
subjects
who
smoke
only
a
few
times
a
week.
However,
Bauman
and
Ennett
noted
that
the
differ-
ential
between
self-reported
and
CO-
indicated
smoking
may
be
informative;
a
substantial
difference
in
the
White:Black
ratios
for
the
two
indices
would
suggest
that
Blacks
were
systematically
underre-
porting
cigarette
smoking.9
These
investi-
gators
found
that
the
differential
was
greater
for
self-report
than
for
CO.
From
this
they
concluded
that
there
was
system-
atic
underreporting
among
African-Ameri-
can
youth,
and
they
suggested
accord-
ingly
that
findings
on
ethnic
differences
in
smoking
may
be
substantially
artifactual,
a
consequence
of
reporting
biases
rather
than
of
true
differences
in
prevalence.
Because
previous
methodological
re-
search
has
supported
the
validity
of
self-report
methods
used
to
index
ciga-
rette
smoking,
both
for
majority
popula-
tions
and
for
samples
for
minority
youth,'0-2
the
conclusions
advanced
by
Bauman
and
Ennett
need
careful
examina-
tion.
Several
aspects
of
the
procedures
used
by
Bauman
and
Ennett
may
have
a
bearing
on
the
interpretation
of
their
results.9
First,
their
research
was
con-
ducted
in
a
younger
age
group,
and
the
sampling
frame
was
smaller
towns
in
the
southeastem
United
States,
an
area
with
typically
lower
rates
of
substance
use
than
other
areas
of
the
United
States.'3
Given
the
generally
low
prevalence
rates
found
within
this
age
group
and
sampling
frame,
there
are
questions
about
whether
the
effect
they
suggested
is
a
general
one.
For
example,
there
were
only
two
self-
reported
Black
smokers
in
the
sample
studied
by
Bauman
and
Ennett.
These
low
rates
make
ratio
results
potentially
un-
stable,
so
that
replication
with
other
ages
and
samples
is
desirable.
School-based
research
provides
data
from
a
different
research
design,
in
which
ethnic
differences
in
validity
may
be
examined.
In
this
research
method
the
data
are
obtained
in
group
administration,
subjects
are
assured
of
the
confidentiality
of
their
responses,
and
participants
are
The
authors
are
with
the
Ferkauf
Graduate
School
of
Psychology
and
the
Department
of
Epidemiology
and
Social
Medicine,
Albert
Ein-
stein
College
of
Medicine,
Bronx,
NY.
Requests
for
reprints
should
be
sent
to
Thomas
A.
Wills,
PhD,
Health
Psychology
Training
Program,
Ferkauf
Graduate
School
of
Psychology,
Albert
Einstein
College
of
Medicine,
1300
Morris
Park
Ave,
Bronx,
NY
1046
1.
This
paper
was
accepted
April
19,
1996.
January
1997,
Vol.
87,
No.
I

Self-Reported
Smoking
informed
that
a
biochemical
measure
of
smoking
status
will
be
obtained.
This
method
has
been
shown
to
enhance
the
validity
of
self-reports
of
smoking."1
Though
school-based
samples
have
been
criticized
on
grounds
of
bias
through
dropout,9
this
issue
is
open
to
empirical
examination
as
to
whether
it
is
relevant
for
interpreting
school-based
data
on
adolescent
smoking.
To
provide
a
careful
examination
of
the
conclusions
suggested
by
Bauman
and
Ennett,9
we
analyzed
data
from
a
represen-
tative
sample
of
adolescents
in
a
northeast-
em
United
States
metropolitan
area.
The
sample
included
substantial
proportions
of
both
African-American
and
Hispanic
adolescents.
This
school-based
research
obtained
data
on
cigarette
smoking
both
through
self-report
and
through
measure-
ment
of
CO
from
expired
breath,
and
data
were
obtained
on
four
occasions
when
the
subjects
were
between
12
years
of
age
and
16
years
of
age
so
that
the
generality
of
effects
could
be
examined
over
a
range
of
ages
and
rates
of
smoking.
We
used
the
same
analytic
procedures
employed
by
Bauman
and
Ennett9
so
that
the
replicabil-
ity
of
their
findings
could
be
tested.
Methods
Participants
included
students
in
7th
through
10th
grades
in
six
public
schools
in
lower
Westchester
County,
New
York.
The
communities
from
which
the
schools
draw
are
representative
of
the
New
York
State
population
on
overall
socioeco-
nomic
census
statistics.'4
In
the
recruitment
procedure,
parents
were
notified
by
mail
about
the
nature
and
methods
of
the
research;
parents
could
have
their
children
excluded
from
the
data
collection
if
they
wished.
Students
were
also
informed
about
the
methods
of
the
research
at
the
time
of
questionnaire
administration;
they
were
informed
that
participation
was
voluntary
and
that
they
could
refuse
or
discontinue
participation.
The
sampling
frame
for
the
study
was
all
7th-grade
students
in
the
six
schools.
For
the
initial
data
collection
in
the
fall
of
1990
there
were
1702
partici-
pants;
the
completion
rate
(number
of
participating
students/number
in
eligible
population
from
school
lists)
was
92%.
Reasons
for
nonparticipation
were
parent
exclusion
(<
1%),
student
refusal
(<
1%),
and
student
absenteeism
(6%).
The
data
collection
was
continued
with
the
same
group
over
the
next
3
years
(fall
1991
through
fall
1993);
students
newly
en-
rolled
in
the
school
were
also
surveyed
at
each
grade
level.
The
sample
size
was
n
=
1775
in
8th
grade
(completion
rate
88%),
n
=
1895
in
9th
grade
(completion
rate
84%),
and
n
=
1699
in
10th
grade
(completion
rate
80%).
Completion
rate
decreased
over
the
study
period
primarily
because
of
an
increase
in
student
absentee-
ism.
In
the
7th-grade
sample
the
partici-
pants
were
47%
female
and
53%
male,
and
the
mean
age
was
12.4
years.
By
self-reported
ethnic
identification,
29%
were
African-American,
23%
were
His-
panic,
37%
were
White,
and
11%
were
of
other
or
mixed
ethnicity.
Data
on
family
structure
indicated
that
53%
of
the
partici-
pants
were
living
with
two
biological
parents,
13%
were
in
a
blended
family
(one
biological
parent
and
one
steppar-
ent),
and
34%
were
living
with
a
single
parent.
Demographic
characteristics
of
the
sample
were
generally
comparable
over
the
study
period.
(A
complete
table
of
demographic
characteristics
is
avail-
able
from
the
authors.)
In
the
research
procedure,
a
self-
report
questionnaire
was
administered
to
students
in
school
classrooms
by
project
staff,
who
followed
a
standardized
proto-
col
in
giving
instructions
to
students
and
answering
questions
about
individual
items.
The
data
collection
staff
were
recruited
from
local
colleges
and
included
African-American,
Hispanic,
and
White
staff
members.
Students
were
informed
that
the
responses
were
confidential
and
would
not
be
reported
to
their
parents
or
teachers.
Students
were
instructed
not
to
write
their
names
on
the
survey;
they
were
assured
that
all
their
answers
were
confi-
dential,
and
that
the
research
data
were
protected
by
a
certificate
of
confidentiality
from
the
US
Public
Health
Service.
The
questionnaire
included
items
on
cigarette
smoking
and
other
substance
use.
The
lead-in
instruction
stated:
"Here
are
some
simple
questions
about
sub-
stances.
Please
give
the
best
answer
you
can
for
each
one.
Remember
to
check
only
one
answer
for
each
question."
The
item
on
cigarette
smoking
asked,
"How
often
do
you
smoke
cigarettes?"
and
had
the
following
response
options:
"I
have
never
smoked
in
my
life,"
"I
have
tried
one
or
two
cigarettes,"
"I
have
tried
cigarettes
four
or
five
times,"
"I
usually
smoke
a
few
cigarettes
a
month,"
"I
usually
smoke
a
few
cigarettes
each
week,"
and
"I
usually
smoke
cigarettes
every
day."
At
the
time
of
questionnaire
adminis-
tration,
students
were
informed
that
breath
samples
would
be
used
to
determine
how
many
students
had
smoked
cigarettes.
The
breath
samples
were
obtained
in
a
private
location
outside
the
classroom
while
other
students
were
completing
the
self-report
questionnaire.
Participants
were
instructed
and
tested
individually,
following
a
stan-
dardized
protocol.
The
participant
pro-
vided
a
sample
of
expired
air,
which
was
analyzed
in
situ
for
CO
by
means
of
the
Breath
CO
Analyzer
(Vitalograph
Corpo-
ration,
Lenexa,
Kan).
The
participant
was
instructed
to
take
a
deep
breath,
hold
it
for
15
seconds
and
"blow
slowly
into
the
machine
and
push
all
the
air
out
of
your
lungs."
The
CO
level
was
then
read
from
the
instrument
and
recorded
on
a
separate
sheet
with only
a
numerical
identification
code,
to
maintain
confidentiality.
The
CO
instruments
were
calibrated
regularly
dur-
ing
the
field
period,
using
the
manufactur-
er's
instructions
and
calibration
samples.
Results
For
the
following
analyses,
a
partici-
pant
was
considered
a
smoker
if
he
or
she
reported
weekly
or
daily
cigarette
use.
By
this
definition,
the
prevalence
of
smoking
based
on
self-report
was
2.4%,
5.9%,
13.3%,
and
16.0%
for
the
7th
through
10th
grades,
respectively,
consistent
with
the
typical
age
effect
noted
in
studies
of
adolescent
substance
use.13
For
compari-
son
with
Bauman
and
Ennett's
data
(which
were
collected
between
end
of
7th
grade
and
beginning
of
8th
grade),
we
computed
the
mean
prevalence
for
7th
and
8th
graders
combined.
This
preva-
lence
was
4.2%,
somewhat
higher
than
the
prevalence
of
3.4%
found
by
Bauman
and
Ennett,9
but
this
difference
would
be
expected
from
the
differential
for
the
northeastem
region
and
larger
Standard
Metropolitan
Statistical
Area
noted
in
national
studies.13
The
lifetime
prevalence
rates
for
smoking
in
the
present
study
were
comparable
to
lifetime
prevalence
rates
for
cigarette
smoking
from
other
studies
of
adolescents.'5
Thus
the
ob-
served
rates
in
the
present
sample
are
consistent
with
other
sources
of
data
on
prevalence
of
cigarette
smoking
in
adoles-
cents.
A
one-way
analysis
of
variance
conducted
with
Proc
GLM
tested
for
variation
in
CO
level
across
self-reported
level
of
smoking.'6
These
analyses,
pre-
sented
in
Table
1,
indicated
a
significant
overall
effect
at
7th
grade
(F[5,
1667]
=
17.30;
P
<
.0001),
8th
grade
(F[5,
1539]
=
9.65;
P
<
.0001),
9th
grade
(F[5,
1777]
=
117.53;
P
<
.0001),
and
10th
grade
(F[5,
1552]
=
176.08;
P
<
.0001).
Pairwise
comparisons
indicated
that
mean
American
Journal
of
Public
Health
57
January
1997,
Vol.
87,
No.
1

Wills
and
Cleary
CO
levels
of
subjects
reporting
weekly
or
daily
smoking
were
significantly
elevated
above
those
of
subjects
reporting
any
lower
frequency
(nonsmoking
through
monthly
smoking).
(There
was
an
overall
decrease
in
mean
levels
of
CO
from
7th
grade
to
8th
grade
[e.g.,
from
3.7
ppm
to
2.9
ppm
among
nonusers].
Because
the
instruments
and
measurement
protocol
were
identical
throughout
the
study,
we
believe
the
decrease
may
reflect
seasonal
ambient
levels.
In
7th
grade,
some
stu-
dents
were
surveyed
during
November-
December
and
some
were
surveyed
in
February-March,
whereas
in
8th
and
9th
grades
all
subjects
were
surveyed
between
October
and
December.
The
decrease
may
reflect
a
small
change
in
ambient
level
of
CO
during
the
heating
season.)
Pairwise
comparisons
contrasting
daily
smokers
with
other
groups
were
significant
(at
P
'
.05)
at
all
time
points.
Pairwise
comparisons
contrasting
subjects
report-
ing
weekly
smoking
with
lower-fre-
quency
groups
were
significant
at
P
<
.05
for
the
8th,
9th,
and
10th
grades;
there
was
one
nonsignificant
comparison
(weekly
vs
monthly
smokers;
P
=
.13,
NS,
at
7th
grade).
These
results
for
the
total
sample
are
consistent
with
previous
research
on
the
validity
of
self-reports
of
smoking,
demonstrating
that
self-reports
of
regular
(weekly
or
daily)
smoking
are
corrobo-
rated
by
significantly
higher
levels
of
the
biochemical
indicator.10
Variations
in
self-reported
rates
of
smoking
across
ethnic
groups
are
pre-
sented
in
Table
2.
These
analyses
indi-
cated
that
Blacks
had
lower
rates
of
smoking
across
the
study
period.
The
overall
chi-square
was
marginal
at
7th
grade
(X2[2,
n
=
1503]
=
4.83;
P
=
.09)
and
significant
at
8th
grade
(X2[2,
n
=
1548]
=
8.20;
P
<
.05),
9th
grade
(X2[2
n
=
1651]
=
28.72;
P
<
.0001),
and
10th
grade
(X2[2,
n
=
1456]
=
69.47;
P
<
.0001).
Cell
chi-squares
for
most
study
points
indicated
that
Blacks
were
significantly
underrepresented
in
preva-
lence
of
weekly
or
daily
smoking,
whereas
Whites
were
significantly
overrepre-
sented,
consistent
with
previous
re-
search.-8
Hispanic
adolescents
tended
to
be
intermediate
between
African
Ameri-
cans
and
Whites;
these
findings
also
are
consistent
with
previous
research
on
Hispanic
adolescents.'
157
To
compare
the
validity
of
self-
reports
of
smoking
for
different
ethnic
groups,
analyses
were
performed
for
smoking
indexed
both
through
self-report
and
through
CO
levels.
Subjects
were
classified
as
smokers
on
the
basis
of
self-report
if
they
indicated
weekly
or
daily
smoking,
and
they
were
classified
as
smokers
on
the
basis
of
CO
measurement
if
they
showed
a
level
of
9
ppm
or
higher.
The
results
are
presented
in
Table
2.
The
prevalence
of
smoking
as
indicated
by
CO
measures
was
lower
than
the
prevalence
for
self-report,
consistent
with
Bauman
and
Ennett9
and
with
other
58
American
Journal
of
Public
Health
TABLE
1-Mean
Breath
Level
of
Carbon
Monoxide
(ppm)
in
a
Sample
of
African-American,
Hispanic,
and
White
Urban
Adolescents,
by
Self-Reported
Smoking
Level
and
Grade
7th
Grade
8th
Grade
9th
Grade
10th
Grade
Self-Reported
Smoking
Level
No.
Mean
(SE)
No.
Mean
(SE)
No.
Mean
(SE)
No.
Mean
(SE)
Never
1094
3.71
(.03)a
822
2.86
(.04)a
777
2.71
(.09)a
619
2.69
(.12)a
1
or2
times
389
3.87
(.06)bc
403
2.78
(.06)a
442
2.59
(.11)a
386
2.67
(.15)a
4or5times
111
3.95
(.11)bcW
168
2.80
(.10)a
225
2.62
(.16)a
217
2.77
(.21)a
Monthly
39
4.03
(.18)abcd
63
2.81
(.16)a
101
2.53
(.24)a
89
2.79
(.32)a
Weekly
18
4.50
(.27)cd
31
3.39
(.22)b
57
4.25
(.32)b
52
5.79
(.42)b
Daily
22
5.77
(.24)e
58
3.91
(.16)b
181
7.14
(.18)c
195
9.49
(.22)c
Note.
Within
column,
values
with
common
subscripts
do
not
differ
significantly
at
P
<
.05.
TABLE
2-Prevalence
of
Smoking
by
Self-Report
Index
and
by
Carbon
Monoxide
(CO)
Index,
by
RacelEthnicity
and
Grade
African
Americans
Hispanics
Whites
Smoking
-X
Index
No.
%
No.
%
No.
%
(2
do
7th
grade
Self-report
6
1.2
10
2.6
20
3.2
4.79
CO
2
0.4
3
0.8
3
0.5
0.60
n
494
389
622
8th
grade
Self-report
19
3.7
27
6.1
46
7.7
8.39*
CO
2
0.4
2
0.5
4
0.7
0.50
n
521
442
595
9th
grade
Self-report
43
7.9
70
12.8
107
18.9
28.95**
CO
8
1.5
25
4.6
35
6.2
15.97**
n
541
546
566
10th
grade
Self-report
37
7.5
64
13.7
132
26.6
69.85**
CO
18
3.6
23
4.9
58
11.7
29.01**
n
495
466
497
Note.
A
subject
was
defined
as
a
smoker
by
self-report
if
he
or
she
reported
smoking
weekly
or
daily.
A
subject
was
defined
as
a
smoker
by
the
CO
index
if
the breath
analyzer
showed
a
CO
level
of
9
ppm
or
higher
for
that
person.
Chi-square
value
is
for
3
(ethnicity)
x
2
(smoking
status,
smoker
vs
nonsmoker)
analysis
for
each
smoking
index.
*P
<
.05;
**P
<
.0001.
January
1997,
Vol.
87,
No.
I

Self-Reported
Smoking
studies
of
this
type.18'19
The
ratio
of
ethnic-group
differences
for
self-report
and
CO,
respectively,
was
computed
as
in
Bauman
and
Ennett.9
For
example,
the
ratio
of
self-reported
smoking
among
Whites
to
self-reported
smoking
among
Blacks
was
computed,
and
this
was
compared
with
the
ratio
of
CO-indicated
smoking
among
Whites
to
CO-indicated
smoking
among
Blacks.
The
results
are
presented
in
Table
3.
The
White:Black
ratio
was
similar
for
the
self-report
index
and
the
CO
index
in
three
of
the
four
measurements,
contrary
to
the
findings
of
Bauman
and
Ennett,
who
reported
a
large
difference
between
the
White:Black
ratio
for
self-report
(11.5)
and
that
for
CO
(3.3).
Only
in
the
7th
grade,
where
rates
of
smoking
were
very
low
in
absolute
terms,
was
the
ratio
substantially
greater
for
self-report
(2.67)
than
for
CO
(1.20).
Comparable
findings
were
noted
for
Hispanics,
where
the
ratio
of
White:Hispanic
reports
was
similar
for
self-report
and
CO-indexed
smoking.
One
marked
difference
occurred
for
9th-grade
data,
where
the
White:Black
ratio
for
CO
(4.13)
was
actually
higher
than
the
ratio
for
self-report
(2.39);
this
is
the
reverse
of
the
pattem
noted
by
Bauman
and
Ennett9
and
is
attributable
to
a
proportionately
greater
rate
of
false
positives
(which
may
indicate
overreporting)
among
Blacks.
Thus
the
tests
comparing
self-report
and
CO
measures
are
not
entirely
consistent
with
the
findings
of
Bauman
and
Ennett.
Computations
of
sensitivity
and
specificity
for
the
four
grade
levels
are
presented
in
Table
4.
Specificity
[true
negatives/(false
positives
+
true
nega-
tives)]
was
generally
high,
consistent
with
previous
research,10
and
did
not
vary
appreciably
by
ethnicity.
Sensitivity
[true-
positives/(false
negatives
+
true
posi-
tives)]
tended
to
increase
as
age,
and
hence
smoking
prevalence,
increased.
In
7th
and
8th
grades
the
sensitivity
analyses
were
based
on
small
numbers
for
all
ethnic
groups
because
of
the
generally
low
prevalence
of
smoking
at
this
age;
in
9th
and
10th
grades
the
prevalence
of
smoking
was
greater.
The
data
indicate
that
the
sensitivity
of
self-report
among
Whites
was
gener-
ally
high.
Sensitivity
tended
to
be
lower
among
minority
respondents,
but
the
magnitude
of
the
effect
was
relatively
small.
For
example,
data
for
9th
grade
indicated
2
false
negatives
for
Blacks
out
of
a
total
of
45
indicated
smokers
(including
6
jointly
verified
and
37
self-reported).
Findings
for
comparisons
of
ethnic
groups
were
variable;
for
ex-
ample,
in
10th
grade
there
were
relatively
more
false
negatives
among
Blacks
(8
out
of
18
verified
smokers)
than
among
Hispanics
(3
out
of
23
verified
smokers),
but
for
upper
grades
the
sensitivity
for
minorities
was
mostly
in
the
range
from
.75
to
.87,
a
figure
that
is
not
greatly
discrepant
from
the
range
for
sensitivity
observed
in
other
studies.'0
There
remains
a
question
as
to
whether
the
relatively
lower
sensitivity
among
minority
respondents
could
ac-
count
for
the
ethnic
differential
observed
in
previous
studies
of
adolescent
smoking.
To
test
this
possibility,
for
data
from
each
grade
level
we
applied
the
sensitivity
from
White
respondents
to
data
for
other
respondents
and
computed
adjusted
preva-
lence
rates
for
African
Americans
and
for
Hispanics.
The
adjusted
prevalence
of
smoking
for
minority
respondents
was
then
compared
with
the
observed
preva-
lence
for
White
respondents
to
determine
whether
a
significant
difference
remained.
Adjusted
prevalences
for
Black
respon-
dents
were
1.4%,
3.8%,
8.3%,
and
9.1%,
and
those
for
Hispanic
respondents
were
3.1%,
6.3%,
13.7%,
and
14.4%,
for
the
7th
to
10th
grades,
respectively.
Each
adjusted
prevalence
rate
for
Blacks
was
significantly
lower
than
the
prevalence
rate
for
Whites
(P
<
.05
by
chi-square
test).
For
Hispanic
respondents,
adjusted
prevalence
rates
in
7th
and
8th
grades
were
not
significantly
different
from
the
prevalence
for
Whites,
but
in
9th
and
10th
grades
the
prevalence
for
Hispanics
was
significantly
lower
(P
<
.01
by
chi-square
test)
than
that
for
Whites.
The
slightly
lower
sensitivity
for
self-reports
of
smok-
ing
among
minority
groups
does
not
account
for
previously
observed
differen-
tials
in
smoking
rates.
From
this
result
we
conclude
that
ethnic
differentials
in
smok-
ing
are
real
and
cannot
be
attributed
to
reporting
artifacts.
In
interpreting
data
from
school-
based
research,
the
issue
of
dropout
rates
needs
to
be
considered.
It
is
typically
found
that
school
dropouts
tend
to
smoke
more
than
nondropouts.20'2'
If
the
pattem
of
smoking
among
dropouts
were
differ-
ent
across
ethnic
groups,
this
could
serve
as
a
biasing
factor
for
school-based
studies.
To
examine
this
issue,
we
per-
formed
analyses
by
ethnicity,
using
Proc
CATMOD
for
log-linear
analysis'6
to
compare
smoking
rates
for
persons
not
represented
in
the
data
set
from
one
wave
to
another
(here
termed
study
dropouts)
and
persons
continuing
over
waves
(here
termed
continuing
participants).
The
drop-
out
rate
was
13%
between
the
7th
and
8th
grades,
18%
between
the
8th
and
9th
grades,
and
25%
between
the
9th
and
10th
grades.
Chi-square
analyses
indicated
that
the
dropout
rate
was
higher
among
minority
students
in
all
three
waves
of
data
(P
<
.0001).
Log-linear
analyses
showed
significantly
higher
rates
of
smok-
American
Journal
of
Public
Health
59
TABLE
3-Ratio
of
Indexed
Smoking
between
Racial/Ethnic
Groups,
for
Two
Indices
of
Smoking
Whites
vs
Whites
vs
Hispanics
vs
Smoking
Index
African
Americans
Hispanics
African
Americans
7th
grade
Self-report
2.67
1.23
2.17
CO
1.20
0.62
1.75
8th
grade
Self-report
2.12
1.27
1.67
CO
1.76
1.49
1.18
9th
grade
Self-report
2.39
1.48
1.62
CO
4.13
1.35
3.09
10th
grade
Self-report
3.55
1.94
1.83
CO
3.25
2.36
1.36
Note.
A
subject
was
defined
as
a
smoker
by
self
report
if
he
or
she
reported
smoking
weekly
or
daily.
A
subject
was
defined
as
a
smoker
by
the
carbon
monoxide
(CO)
index
if
the
breath
analyzer
showed
a
CO
level
of
9
ppm
or
higher
for
that
person.
January
1997,
Vol.
87,
No.
I

Wills
and
Cleary
ing
among
dropouts
than
among
continu-
ing
participants
in
7th
and
8th
grade,
8th
and
9th
grade,
and
9th
and
10th
grade
(x2=6.88,
P<.01;
X2=2l.3l1
P<
.0001;
x2
=
40.62,
P
<
.0001,
respec-
tively)
and
higher
rates
of
smoking
among
Whites
than
among
minority
group
mem-
bers
(X2
=
5.15,
NS;
x2
=
8.85,
P
<
.05;
X2
=
31.19,
P
<
.0001,
respectively).
These
findings
are
consistent
with
previ-
ous
research.2'
However,
the
three-way
interaction
was
nonsignificant
in
all
three
waves
of
data,
indicating
that
the
pattern
of
smoking
among
the
study
dropouts
was
not
significantly
different
from
the
pattern
among
the
continuing
participants.
Thus,
including
the
dropouts
in
the
analytic
sample
would
not
change
conclusions
about
ethnic
differentials
in
smoking.
Discussion
This
research
examined
racialethnic-
group
differentials
in
the
prevalence
of
cigarette
smoking,
using
data
from
adoles-
cents
who
were
surveyed
at
four
age
levels.
The
results
are
consistent
with
previous
research
in
showing
a
substantial
increase
in
smoking
over
the
period
from
early
through
middle
adolescence
and
in
showing
a
lower
prevalence
of
smoking
among
African-American
adolescents
than
among
White
adolescents.
The
data
are
also
consistent
with
available
findings
on
Hispanic
adolescents,
indicating
a
lower
prevalence
of
smoking
for
Hispanic
ado-
lescents
than
for
White
adolescents.
157'22
Indices
of
sensitivity
and
specificity
for
the
total
sample
were
consistent
with
other
research
on
the
validity
of
self-
reports
of
smoking.'0
Thus
the
general
properties
of
the
present
data
are
consis-
tent
with
previous
findings
obtained
with
a
variety
of
methods
and
populations.
Findings
on
ethnic
differentials
in
the
validity
of
self-reported
smoking
support
three
conclusions.
First,
the
valid-
ity
of
self-reports
of
smoking
among
White
respondents
was
high
in
absolute
terms.
Second,
the
results
of
Bauman
and
Ennett
were
not
confirmed
for
older
age
groups.9
The
White:Black
ratio
for
self-
reported
smoking
was
similar
to
the
ratio
for
CO-indexed
smoking;
hence,
using
the
methods
of
Bauman
and
Ennett,9
the
present
results
are
not
entirely
consistent
with
the
suggestion
of
the
invalidity
of
reports
by
African-American
adolescents.
Third,
while
the
observed
sensitivity
for
self-reports
of
smoking
was
slightly
lower
among
both
African-American
and
His-
panic
adolescents,
adjusted
rates
indicated
that
ethnic
differentials
in
smoking
rates
could
not
be
accounted
for
by
reporting
factors.
In
the
interpretation
of
research
on
the
sensitivity
of
self-reports
of
smoking
in
adolescent
samples,
the
base
rate
of
smoking
must
be
considered.
Research
with
younger
samples
will
necessarily
find
low
rates
of
regular
smoking,
and
accordingly
there
will
be
few
subjects
indicated
as
smokers
either
on
the
basis
of
CO
measurements
or
on
the
basis
of
self-report.
This
was
evident
in
the
present
data
for
7th
grade,
where
only
two
or
three
subjects
in
any
ethnic
group
were
indi-
cated
as
regular
smokers;
this
was
also
the
case
in
the
data
reported
by
Bauman
and
Ennett.9
Therefore,
conclusions
about
validity
based
on
samples
of
younger
respondents
may
have
limited
generaliz-
ability;
because
of
the
small
cell
sizes,
a
shift
in
one
or
two
cases
will
change
sensitivity
results
appreciably.
In
the
present
article
we
report
data
on
smoking
at
older
ages,
where
rates
of
regular
smoking
were
higher
for
each
ethnic
group.
We
think
greater
confidence
should
be
attached
to
these
data
because
the
base
rates
of
smoking
are
higher
in
older
samples.
It
should
be
noted
that
the
CO
measure
is
not
a
perfect
indicator
of
smoking
status
in
younger
populations.
CO
levels
for
some
individuals
may
be
affected
by
passive
exposure
to
smoking
in
the
home.
Most
important,
adolescents
are
infrequent
smokers
rather
than
regular
smokers,
and
the
times
at
which
adoles-
cents
are
able
to
smoke
(usually
after
school)
are
limited.
Thus
false-positive
rates
will
tend
to
be
elevated
because
an
individual
might
report
correctly
that
he
or
she
smoked
on
a
weekly
basis,
but
this
individual
would
not
necessarily
test
positive
on
the
CO
measure
if
he
or
she
had
not
smoked
during
the
previous
8
to
12
hours.
However,
although
CO
measure-
ments
will
tend
to
underestimate
the
absolute
level
of
smoking
prevalence
60
American
Journal
of
Public
Health
TABLE
4-Designation
of
Adolescents'
Smoking
Status
by
Self-Report
and
Carbon
Monoxide
(CO)
Measures,
by
Race/Ethnicity
and
Grade
African
Americans:
CO
Hispanics:
CO
Whites:
CO
Yes
No
Yes
No
Yes
No
7th
grade
Self-report
Yes
1
5
1
9
3
17
No
1
487
2
377
0
602
Sensitivity
.50
.33
1.00
Specificity
.99
.98
.97
8th
grade
Self-report
Yes
1
18
1
26
3
43
No
1
501
1
414
1
548
Sensitivity
.50
.50
.75
Specificity
.97
.94
.93
9th
grade
Self-report
Yes
6
37
19
51
33
74
No
2
496
6
470
2
457
Sensitivity
.75
.76
.94
Specificity
.93
.90
.86
10th
grade
Self-report
Yes
10
27
20
44
57
75
No
8
450
3
399
1
364
Sensitivity
.56
.87
.98
Specificity
.98
.90
.83
Note.
A
subject
was
defined
as
a
smoker
by
self
report
if
he
or
she
reported
smoking
weekly
or
daily.
A
subject
was
defined
as a
smoker
by
the
carbon
monoxide
(CO)
index
if
the
breath
analyzer
showed
a
CO
level
of
9
ppm
or
higher
for
that
person.
January
1997,
Vol.
87,
No.
I

Self-Reported
Smoking
among
adolescents,
they
are
useful
for
assessing
the
validity
of
group
differences
in
smoking
rates.
With
regard
to
the
sensitivity
of
self-reported
smoking
measures
among
adolescents,
the
present
results
provide
a
somewhat
mixed
message.
While
the
sensitivity
for
White
adolescents
was
almost
perfect,
there
were
more
occur-
rences
of
false
negatives
among
African-
American
and
Hispanic
respondents,
and
this
suggests
that
the
rate
of
underreport-
ing
is
relatively
greater
among
minority
adolescents.
However,
even
when
we
adjusted
for
sensitivity
differences
there
was
a
significantly
lower
prevalence
of
smoking
for
both
African
Americans
and
Hispanics
than
for
Whites,
thereby
provid-
ing
evidence
that
smoking
differentials
among
minorities
are
not
attributable
to
reporting
effects.
The
data
do
suggest
that
researchers
working
with
minority
adoles-
cents
should
give
attention
to
issues
such
as
assuring
confidentiality
and
working
to
build
rapport
with
respondents.
An
additional
issue
for
interpreting
data
on
ethnic
differences
in
smoking
is
the
false-positive
rate.
There
were
propor-
tionately
more
false
positives
among
both
Blacks
and
Hispanics
than
among
Whites.
Although
false
positives
affect
the
preva-
lence
rate,
their
effect in
the
present
data
would
be
to
inflate
the
prevalence
rates
for
minority
groups
and
thereby
reduce
the
apparent
magnitude
of
the
White:minority
differential.
If
there
are
White
vs
minority
differences
in
temporal
patterns
of
smok-
ing
(e.g.,
on
weekends
vs
on
weekdays),
then
the
apparent
overreporting
of
one
group
could
be
attributable
to
different
patterns
of
smoking.
This
question
re-
mains
to
be
clarified
through
further
research
to
provide
detailed
data
on
temporal
pattems
of
adolescent
smoking.
The
present
data
indicate
that
the
White:
minority
differential
in
smoking
is
real
and
may
be
underestimated.
Some
possible
limitations
of
the
present
study
should
be
noted.
We
used
one
method
of
biochemical
verifica-
tion-CO
level-and
were
not
able
to
verify
smoking
by
other
biochemical
methods.
The
school-based
research
de-
sign
is
not
an
exact
random
sample
of
the
community,
but
we
tested
for
an
effect
through
dropout
from
the
sample
and
found
that
dropout
did
not
bias
conclu-
sions
about
ethnic
differentials
in
smok-
ing.
The
sample
was
obtained
from
a
metropolitan
area
in
the
northeastemn
United
States,
and
some
of
the
findings
might
not
be
the
same
in
other
regions
of
the
country.
However,
we
think
the
present
study
yields
valid
conclusions
about
ethnic
differentials
in
smoking.
In
summary,
results
based
on
a
large
sample
of
adolescents
indicated
that
self-
reports
of
cigarette
smoking
were
gener-
ally
valid
and
that
differences
in
false-
negative
and
false-positive
rates
across
ethnic
groups
did
not
seriously
qualify
the
results.
We
conclude
that
the
ethnic
differentials
in
cigarette
smoking
indi-
cated
by
previous
epidemiologic
research
are
real
and
are
not
a
consequence
of
reporting
artifacts.
The
results
raise
inter-
esting
questions
for
further
research
to
examine
pattems
of
smoking
among
adolescents
and
to
determine
the
reason
for
the
lower
rate
of
cigarette
smoking
among
African-American
adolescents.
C]
Acknowledgments
This
research
was
supported
by
grant
RO
I
-DA-
05950
and
grant
K02-DA-00252
from
the
National
Institute
on
Drug
Abuse.
We
thank
the
superintendents
of
the
school
districts
for
their
support
and
the
principals
and
students
of
the
participating
schools
for
their
cooperation.
References
1.
Hawkins
JD,
Catalano
RF,
Miller
JY.
Risk
and
protective
factors
for
alcohol
and
other
drug
problems
in
adolescence
and
early
adulthood:
implications
for
substance
abuse
prevention.
Psychol
Bull.
1992;
112:64-
105.
2.
Wills
TA.
Stress
and
coping
factors
in
the
epidemiology
of
substance
use.
In:
Ko-
zlowski
LT
et
al.,
eds.
Research
Advances
in
Alcohol
and
Drug
Problems.
New
York,
NY:
Plenum
Press;
1990;10:215-250.
3.
Barnes
GM,
Welte
JW.
Adolescent
alcohol
abuse:
subgroup
differences
and
relation-
ships
to
other
problem
behaviors.
JAdolesc
Res.
1986;
1:79-94.
4.
Kandel
D,
Single
E,
Kessler
RC.
The
epidemiology
of
drug
use
among
New
York
State
high
school
students:
distribu-
tion,
trends,
and
change
in
rates
of
use.
Am
J
Public
Health.
1976;66:43-53.
5.
Newcomb
MD,
Maddahian
E,
Skager
R,
Bentler
PM.
Substance
abuse
and
psycho-
social
risk
factors
among
teenagers:
associa-
tions
with
sex,
age,
ethnicity,
and
type
of
school.
Am
JDrugAlcoholAbuse.
1987;13:
413-433.
6.
Welte
JW,
Barnes
GM.
Alcohol
use
among
adolescent
minority
groups.
J
Stud
Alco-
hol.
1987;48:329-336.
7.
Bachman
JG,
Wallace
JM,
O'Malley
PM,
Johnston
LD,
Kurth
CL,
Neighbors
HW.
Racial/ethnic
differences
in
smoking,
drink-
ing,
and
illicit
drug
use
among
American
high
school
seniors,
1976-89.
Am
J
Public
Health.
1991
;8Q1:372-377.
8.
Windle
M.
A
longitudinal
study
of
antiso-
cial
behaviors
in
early
adolescence
as
predictors
of
late
adolescent
substance
use:
gender
and
ethnic
group
differences.
J
Abnorm
Psychol.
1990;99:86-91.
9.
Bauman
KE,
Ennett
SE.
Tobacco
use
by
Black
and
White
adolescents:
the
validity
of
self-reports.
Am
J
Public
Health.
1994;
84:394-398.
10.
Patrick
DL,
Cheadle
A,
Thompson
DC,
Diehr
P,
Koepsell
T,
Kinne
S.
The
validity
of
self-reported
smoking:
a
review
and
meta-analysis.
Am
J
Public
Health.
1994;
84:1086-1093.
I.
Murray
DM,
Perry
CL.
The
measurement
of
substance
use
among
adolescents:
when
is
the
'bogus
pipeline'
method
needed?
Addict
Behav.
1987;
12:225-233.
12.
Wallace
JM
Jr,
Bachman
JG.
Validity
of
self-reports
in
student-based
studies
on
minority
populations:
issues
and
concerns.
In:
De
La
Rosa
MR,
Adrados
JR,
eds.
Drug
Abuse
among
Minority
Youth:
Advanices
in
Research
and
Methodology.
Rockville,
Md:
National
Institute
on
Drug
Abuse;
1993:167-200.
US
Dept
of
Health,
Educa-
tion,
and
Welfare
publication
NIH
93-
3479.
13.
Johnston
LD,
O'Malley
PM,
Bachman
JG.
Drug
Use
among
American
High
School
Seniors,
College
Students
and
Young
Adults,
1975-1990.
Vol.
1,
High
School
Seniors.
Rockville,
Md:
National
Institute
on
Drug
Abuse;
1991.
DHHS
publication
91-1813.
14.
U.S.
Department
of
Commerce.
1990
Census
of
the
Population:
Social
and
Economic
Characteristics
(New
York).
Washington,
DC:
US
Government
Printing
Office;
1993.
15.
Oetting
ER,
Beauvais
F.
Adolescent
drug
use:
findings
of
national
and
local
surveys.
J
Consult
Clin
Psychol.
1990;58:385-394.
16.
SAS/STAT
User's
Guide.
Release
6.03
ed.
Cary,
NC:
SAS
Institute
Inc;
1988.
17.
Substance
Abuse
and
Mental
Health
Ser-
vices
Administration.
National
Household
Survey
on
Drug
Abuse:
Population
Esti-
mates
1992.
Rockville,
Md:
US
Dept
of
Health
and
Human
Services;
1993.
DHHS
publication
93-2053.
18.
Bauman
KE,
Koch
GG.
Validity
of
self-
reports
and
descriptive
and
analytical
conclusions:
the
case
of
cigarette
smoking
by
adolescents
and
their
mothers.
Am
J
Epidemiol.
1983;1
18:90-98.
19.
Murray
DM,
O'Connell
CM,
Schmid
LA,
Perry
CL.
The
validity
of
smoking
self-
reports
by
adolescents:
a
reexamination
of
the
bogus
pipeline
procedure.
Addict
Be-
hav.
1987;12:7-15.
20.
Kandel
D,
Kessler
RC,
Margulies
RZ.
Antecedents
of
adolescent
initiation
into
stages
of
drug
use.
In:
Kandel
DB,
ed.
Longitudinal
Research
on
Drug
Use.
New
York,
NY:
John
Wiley
&
Sons
Inc;
1978:73-100.
21.
Pirie
PL,
Murray
DM,
Luepker
RV.
Smok-
ing
prevalence
in
a
cohort
of
adolescents,
including
absentees,
dropouts
and
trans-
fers.
Am
J
Public
Health.
1988;78:176-
178.
22.
Chavez
EL,
Swaim
RC.
An
epidemiologi-
cal
comparison
of
Mexican-American
and
White
non-Hispanic
8th-
and
12th-grade
students'
substance
use.
Am
J
Public
Health.
1992;82:445-447.
January
1997,
Vol.
87,
No.
1
American
Journal
of
Public
Health
61