You are on page 1of 7



Using Social Cognitive Theory to Explain

Discretionary, Leisure-time Physical Exercise
Among High School Students



Purpose: To examine whether knowledge of high utilized in this study. Society for Adolescent Medicine,
school students actions of self-regulation, and percep- 2003
tions of self-efficacy to overcome exercise barriers, social
situation, and outcome expectation will predict non-
school related moderate and vigorous physical exercise. KEY WORDS:
Methods: High school students enrolled in introduc- Adolescents
tory Physical Education courses completed question- High school students
naires that targeted selected Social Cognitive Theory Leisure time
variables. They also self-reported their typical leisure- Physical exercise
time exercise participation using a standardized ques- Social Cognitive Theory
tionnaire. Bivariate correlation statistic and hierarchical
regression were conducted on reports of moderate and
vigorous exercise frequency. During recent years the importance of promoting
Results: Each predictor variable was significantly asso-
physical activity during adolescence has become
ciated with measures of moderate and vigorous exercise
frequency. All predictor variables were significant in the
increasingly recognized. This shift toward promot-
final regression model used to explain vigorous exercise. ing adolescent physical activity has been supported
After controlling for the effects of gender, the psychoso- by the consistent association between cardiovascular
cial variables explained 29% of variance in vigorous fitness and decreased risk of mortality among adult
exercise frequency. Three of four predictor variables subjects [13]. Another factor has been the realization
were significant in the final regression equation used to that maintaining even a moderately active life will
explain moderate exercise. The final regression equation reduce the risk of premature death among adults
accounted for 11% of variance in moderate exercise [4,5]. Unfortunately physical activity rates decline
frequency. consistently during the adolescent years [6,7]. It is
Conclusions: Professionals who attempt to increase the clear that one important path to extending life and
prevalence of physical exercise through educational
preventing disease begins by increasing physical
methods should focus on the psychosocial variables
activity rates during adolescence. The Office of the
United States Surgeon General echoed this logic [8].
Effective interventions should be based on re-
From the Department of Physical Education, Denison University, search that identifies determinants of physical activ-
Granville, Ohio (E.R.W.); Health and Physical Activity Behavior, Ohio
State University (R.L.P.), Columbus, Ohio; and Denison University, ity among adolescents [9]. Ideally these empirically
Granville, Ohio (T.E.C.). verified determinants would then be tested for mod-
Address correspondence to: Eric Ross Winters, Ph.D., ATC Assis- ifiability in the context of school based programs.
tant Professor, Department of Physical Education, Denison University,
Granville, OH 43023. E-mail: Unfortunately there are only a few published studies
Manuscript accepted November 12, 2002. identifying valuable constructs according to a review
1054-139X/03/$see front matter Society for Adolescent Medicine, 2003
doi:10.1016/S1054-139X(03)00046-6 Published by Elsevier Inc., 360 Park Avenue South, New York, NY 10010
June 2003 WINTERS ET AL 437

of forty-five studies of environmental and psycho- participating on an organized athletic team during
social determinants of adolescent physical activity the previous year. The percentage of female athletes
[10]. Social factors were found to be associated with in the sample is approximately 2% higher than
adolescents physical exercise. The authors con- population estimates. The percentage of male ath-
cluded that many potential constructs were of inde- letes in the sample is approximately 7% greater than
terminate magnitude owing in part to weaknesses in population estimates.
study methods, including: lack of theoretical basis
for study, errors in measurement, poor sampling,
and inappropriate data analysis strategies. Instruments
The purpose of this study was to test selected Leisure-time physical exercise was measured using
Social Cognitive Theory (SCT) constructs in explain- the Godin/Shephard exercise self-report instrument
ing discretionary, leisure-time physical exercise which asks subjects to provide the number of times
among high school aged adolescents. Specifically the during a typical week they participated in strenuous
SCT constructs of social situation, self-efficacy, out- exercise and moderate exercise [11]. For this study,
come expectations values, and self-regulation were subjects were instructed to report only leisure-time
used to predict discretionary moderate and vigorous strenuous and or moderate exercise (i.e. bouts that
physical exercise. In an effort to address measure- lasted a minimum of 15 minutes and occurred out-
ment problems cited previously, each of the instru- side of physical education class and which was not
ments used in this study had established validity and associated with organized athletics). Strenuous exer-
reliability. The results of this study will identify cise was operationally defined as exercise that re-
potentially modifiable constructs useful for the de- sulted in sweating, labored breathing, and rapid
sign of effective interventions to promote physical heart rate. Moderate exercise was defined as exercise
exercise. that could be conducted while holding a conversa-
tion and did not result in labored breathing or rapid
heart rate. The Godin Leisure-time Exercise Ques-
Methods tionnaire previously has been reported to have retest
reliability at one month for vigorous, .84, as well as
Three high schools in central Ohio were recruited for moderate, .36. This instrument also has been re-
the study. Data were collected in intact physical ported to have criterion-related validity, .36 with
education classrooms during two consecutive days. accelerometry. [11,12].
One class period was dedicated to collecting the SCT Social Situation was measured using a self-report
questionnaires, the Godan Leisure-time Physical Ac- questionnaire containing eight items. This instru-
tivity questionnaire, resting pulse, resting blood ment was originally developed by Reynolds et al
pressure, height, and weight. The second class pe- [13]. Later it was expanded upon by Trost et al [14].
riod was dedicated to obtaining student times in a This instrument measures instrumental social sup-
one and one half-mile walk/run. Only data provided port, social encouragement, and social expectations
by students who returned a satisfactory parental that are provided by friends and family members for
consent form and completed a minimum of 80% of physical exercise. This instrument has previously
the study activities were entered into the research been demonstrated to have construct validity [15].
database. The Institutional Review Board of Denison This instrument has also been reported to have
University approved all research procedures. internal reliability, .75, and retest reliability, .78 [14].
In order to increase the internal reliability of the
instrument by increasing response variability, the
Sample original reporting scale was extended from dichoto-
A total of 248 students returned an acceptable paren- mous responses to a five-point Likert-type scale.
tal informed consent and completed a minimum of Self-efficacy for exercise is defined as the per-
80% of the activities associated with this study. The ceived ability to overcome barriers to exercise. A
sample represents a convenience sample. Partici- previously developed instrument was selected for
pates in this study were 150 females, and 98 males. this study [15]. This Strength of self-efficacy instru-
Average age among participants was 15 years. All ment has been demonstrated to have construct va-
but 5 of the subjects reported being in 9th or 10th lidity [15] and predictive validity for both 5th-grade
grade. Seventy-eight, or 52%, of females in the sam- boys and girls, .23 and .27 [14]. Retest reliability of
ple and 67, or 68% of males in the sample reported this scale has been reported to be .82 [14]. For the

Table 1. Frequency Distribution of Self-Reported The instrument upon which outcome expectation
Participation in Moderate and Vigorous Physical Exercise values were measured was based on a conceptual
(N 236)
model that has been previously published [17]. This
Moderate Vigorous model has been used by other authors to develop
Physical Exercise Physical Exercise and validate similar instruments [18 20]. The instru-
Days n (cum%) n (cum%)
ment used in this study assesses outcomes expecta-
0 18 (8) 27 (11) tions and their associated expectancies for physical
1 25 (18) 28 (23)
exercise by requesting information on eight dimen-
2 39 (35) 27 (35)
3 43 (53) 45 (54) sions. These include: relaxation, fitness, competition,
4 27 (64) 22 (63) social growth, social continuation, thrills, expressive
5 34 (79) 48 (84) movement, and beautiful movement. Each of the
6 15 (85) 15 (90) eight dimensions is assessed by five items. In a
7 35 (100) 24 (100)
previous study, an earlier version of this instrument
was examined within high school students for con-
struct validity and internal reliability, and ranges of
purposes of this study three additional items were internal reliability were from .86 to .94 [20].
included in order to broaden the topics of self- Self-regulation is the personal regulation of goal-
efficacy sampled, thereby increasing the content va- directed behavior. Although self-regulation is a ma-
lidity of the instrument. By adding these items, the jor SCT construct, few studies of adolescent health
construct validity of the instrument may be compro- behavior have examined this factor. Self-regulation
mised, however, owing to the limited number of has been found to be a significant predictor of
subjects within this study, confirmatory factor ana- moderate physical exercise [20]. The Self-regulation
lytic procedures was not possible. Results from this for Physical Activity instrument measures self-regu-
study will be used to examine the predictive validity lation in five dimensions: goal-setting, self-monitor-
for this newly altered self-efficacy instrument. In ing, gaining and maintaining social support, plan-
order to increase the internal reliability of the instru- ning to overcome barriers, and securing
ment by increasing response variability, the response reinforcements [21]. This 38-item instrument has
scale was shifted from dichotomous to a five-point construct validity and internal reliability for the five
Likert-type scale. dimensions range from .78 to .94 [20].
Outcome expectation value is a construct created
by multiplying an outcome expectation with
matched outcome expectancies [16]. Outcome expec-
tations are the anticipated consequences of a behav- Results
ior. For example, the belief, regular exercise will Table 1 provides the frequency distributions for both
make me more attractive, is a belief about the effects discretionary leisure-time moderate and vigorous
of regular exercise on the body seen through the lens physical exercise. 65% of subjects reported engaging
of values regarding attractiveness. Outcome expect- in three or more days of vigorous exercise in a typical
ancies are the relative values that an individual week, and 36% of subjects reported engaging in
places on each outcome expectation. Referring to the moderate exercise on five or more days each week.
previous example, people would vary widely on the Table 2 presents the mean scores and reliability for
personal importance of physical attractiveness. If an each predictor variable. Cronbach alpha was used to
individual believes a behavior is linked to a desirable test the reliability of each instrument. Range of
outcome, the behavior is more likely to be practiced reliability for these predictor variables were .88 to
regularly. .93. The observed mean scores for the predictor

Table 2. Descriptive Means, Instrument Reliability, and the Number of Items Used to Measure Each of the Selected
Social Cognitive Theory Variables
Variable N Mean (SD) Alpha Possible Range
Outcome expectation values 190 55.23 (19.64) .93 17100
Self-regulation 232 55.07 (16.62) .90 17100
Social situation 243 55.63 (20.63) .88 20 100
Strength self-efficacy 247 68.66 (18.36) .89 20 100
June 2003 WINTERS ET AL 439

Table 3. Intercorrelations Between Selected Social Cognitive Theory Variables and Measures of Frequency of
Adolescent Leisure-time Vigorous Physical Exercise and Moderate Physical Exercise
Subscale 1 2 3 4 5 6 7
1. Gender .25* (n 236) .05 (n 236) .04 (n 190) .10 (n 232) .13 (n 243) .23* (n 247)
2. Vigorous .38* (n 235) .49* (n 182) .44* (n 221) .26* (n 232) .34* (n 235)
3. Moderate .27* (n 181) .26* (n 221) .15* (n 232) .15* (n 235)
4. Outcome exp. values .59* (n 184) .42* (n 186) .54* (n 190)
5. Self-regulation .63* (n 228) .60* (n 231)
6. Social situation .52* (n 242)
7. Strength of self-efficacy
* p .05.

variables were relatively low as none of the predictor 10%. Self-regulation, entered in the fourth stage,
variables achieved a mean score that exceeded 75% accounted for 8% of unique variance.
of their respective maximum score. Table 5 presents the hierarchical multiple-regres-
The intercorrelation matrix (Table 3) reveals a sion analysis on self-reported discretionary leisure-
significant association between each of the predictor time moderate exercise. The regression model,
variables and leisure-time vigorous and moderate which included each of the four SCT variables,
exercise bouts. Males and females expressed similar accounted for a significant quantity of variance in
perceptions regarding their self-regulation, social discretionary leisure-time moderate exercise, 11%.
situation, and outcome expectancy values, as well as The gender of the subjects was not associated with
statistically similar participation in moderate physi- the dependent variable in this case and was therefore
cal exercise. Males reported more frequent bouts of not entered into this regression equation. The SCT
vigorous exercise and stronger perceptions of self- variable, social situation, did not capture significant
efficacy than their female counterparts. The SCT unique variance in discretionary leisure-time mod-
variables were demonstrated to have high intercor- erate exercise, R2 .01. Self-regulation captured 6%
relations. The strongest association among SCT vari- of the variance in moderate exercise, which accounts
ables was between self-regulation and social situa- for 55% of the variance explained by the full model.
tion (r .63). Weakest intercorrelation among SCT
variables was the association between outcome ex-
pectancy value and social situation.
Hierarchical multiple-regression was used to test Discussion
the ability of the SCT variables to predict bouts of The percentage of subjects in this investigation who
self-reported discretionary leisure-time vigorous reported engaging in vigorous exercise on three or
exercise. Table 4 reveals that each of the predictor more days a week approximates the population
variables were retained in the final model and col- estimate, 65% in this study against 64% for high
lectively accounted for 29% of the variance in bouts school students nationally [22]. As students progress
of self-reported discretionary leisure-time vigor- through high school they tend to decrease their
ous exercise after effect of gender was controlled. participation in vigorous exercise [22]. Population
Outcome expectancy value, although entered in the estimate for the percentage of 9th and 10th grade
fifth stage, accounted for the most unique variance, students who report engaging in vigorous exercise

Table 4. Summary of Hierarchical Regression Analysis for Variables Predicting Frequency of Adolescent Leisure-time
Vigorous Physical Exercise (N 174)
Stage Variable -R2 R2 hSS df F-value P
1 Gender .06 64.83 1 14.57 .05
2 Social situation .12 .06 57.56 1 12.85 .05
3 Strength of self-efficacy .17 .05 51.86 1 11.47 .05
4. Self-regulation .25 .08 72.24 1 16.42 .05
5. Outcome expectation value .35 .10 47.09 1 10.35 .05
Full model .35 293.58 5 18.40 .000
Residual 536.01 168

Table 5. Summary of Hierarchical Regression Analysis for Variables Predicting Frequency of Adolescent Leisure-time
Moderate Physical Exercise (N 172)
Stage Variable -R2 R2 hSS df F-value p
1 Strength of self-efficacy .02 22.71 1 4.65 .05
2 Social situation .03 .01 11.87 1 2.40 .05
3. Self-regulation .07 .06 34.40 1 7.15 .05
4. Outcome expectation value .11 .04 21.06 1 4.89 .05
Full model .11 90.04 4 4.93 .001
Residual 767.19 168

on three or more days each week is 71.9% and 67%, that knowledge of student outcome expectation
respectively [22]. value is a significant explanatory factor of frequency
Percentage of subjects in this investigation who of leisure-time moderate and vigorous physical ex-
reported engaging in moderate exercise on five or ercise.
more days a week was 10% higher than the popula- Based upon the existing literature it is not possible
tion estimate. Thirty-six percent of subjects within to determine if self-efficacy is associated with phys-
this study reported exercising moderately on five or ical exercise during adolescence [10]. Explanation for
more days a week whereas the population of high this indeterminate relationship may be attributed to
school students is estimated to be 26% [22]. errors in measurement. The literature upon which
It appears that subjects within this study engaged Sallis et al [10] made their conclusions clearly con-
in vigorous exercise at a similar rate as the typical tained data from a heterogeneous set of definitions of
grade nine and ten student, and engaged in substan- self-efficacy including: self-efficacy to exercise, self-
tially more moderate exercise than the typical high efficacy to gain social support, self-efficacy to exer-
school student. The finding that the participants in cise even though other activities are available, and
this investigation were more active than the typical self-efficacy to overcome barriers to exercise. The
high school student indicates that results from this current investigation explored the relationship be-
study are most applicable to more active early high tween self-efficacy to overcome barriers to physical
school students. exercise and the frequency of moderate and vigorous
This investigation explored the possibility that leisure-time physical exercise. Results of this analysis
measurement error and the presence of a mediating revealed a significant association with leisure-time
variable may explain the previously reported inde- moderate and vigorous physical exercise bouts.
terminate relationship between outcome expectation Therefore while self-efficacy variously defined may
and adolescent physical exercise [10]. Correction for not be associated with adolescents exercise partici-
measurement error required the development of a pation it is reasonable to believe that self-efficacy to
new instrument based on the model for physical overcome barriers to exercise is associated with mod-
exercise developed previously by Kenyon [17] and erate and vigorous physical exercise.
later extended by Simon et al [18] and Schultz et al Social Situation has previously been shown to be
[19]. It was also postulated within this investigation associated with adolescent physical exercise [10].
that the value ascribed to expected outcomes of Results of this study confirm this conclusion in that
regular exercise are mediating variables that when Social Situation was significantly associated with
unaccounted for diminish the explanatory power of both frequency of vigorous and moderate leisure-
anticipatory thought. Bandura [16] suggested that time physical exercise. However this study indicates
this mediation effect should be taken into account. In that Social Situation is closely associated with the
order to account for the values ascribed to the other SCT variables used in this study and therefore
potential outcomes of physical exercise, an outcome fails to capture significant unique variance in lei-
expectancy instrument was developed. Within this sure-time moderate exercise.
investigation, the values (expectancies) were multi- Although many previous interventions to increase
plied by the students conviction that potential prod- moderate to vigorous physical exercise among
ucts (expectations) will be gained through regular younger students have made use of the self-regula-
exercise. This treatment of the relationship between tion construct, it does not appear that any empirical
outcome expectancies and expectations is supported evaluation of the relationship between self-regula-
by Bandura [16]. Results from this analysis indicated tion and physical exercise participation has been
June 2003 WINTERS ET AL 441

undertaken [10,2325]. Results of the present study Although results from the two regression models
indicate that self-regulation is associated with both accounted for a statistically significant amount of
the frequency of leisure-time vigorous and moderate variance in self-reported leisure-time moderate
physical exercise. These results signify the impor- and vigorous exercise participation, the importance
tance of continuing to address the self-regulating of this finding is undetermined. Currently there are
capacities of early high school students when at- few published evaluations of exercise promotion
tempting to increase the frequency of their physical educational interventions among high school stu-
exercise participation. dents [20,23,26]. While these studies demonstrated
Results of the hierarchical regression analysis for effectiveness at improving exercise participation or
frequency of leisure-time vigorous physical exer- physiological adaptations to exercise, the quantity of
cise revealed that the linear combination of all four exercise variance actually manipulated is but a small
selected SCT variables explained a significant quan- fraction of the exercise variance that can be ex-
tity of variance. After controlling for the demo- plained. The practical importance from this investi-
graphic variable gender, the four SCT variables were gation and others like it will likely remain unfulfilled
able to account for 29% of the variance in frequency until more powerful educational experiences can be
of vigorous exercise. developed. Finally, this article has provided what
Results of the regression analysis for frequency of these authors believe to be the first evidence that
leisure-time moderate physical exercise revealed predictors of adolescent exercise differ based on the
that a linear combination of outcome expectation intensity of the exercise. This result indicates that
values, self-regulation, social situation, and strength future studies in this area should include a separa-
of self-efficacy explained a significant quantity of tion of moderate to vigorous physical exercise into
variance, 11%. Although only a small fraction of the both moderate and vigorous physical exercise.
total variance was accounted for by the selected SCT
variables, these results are believed to be important.
It appears that this investigation is the first published Limitations
report in which significant quantity of variance in Results from this investigation should be viewed in
adolescent self-reported moderate exercise was ex- light of certain limitations. The use of the conve-
plained. This investigation also provides empirical nience sampling technique employed here does not
evidence to support the belief that within the adoles- provide necessarily accurate estimates of the under-
cent population, self-reported moderate exercise lying population parameters. The frame from which
may be behaviorally different than self-reported vig- these subjects were selected is limited to an immedi-
orous exercise. Future authors should explore the ate geographical area and therefore results may not
possibility that adolescents moderate exercise may hold for other adolescents in other locations. It
be determined by a larger and somewhat different appears that subjects within this investigation were
set of variables than adolescents vigorous exercise. engaged in more frequent moderate physical exer-
Potentially, a subset of the outcome expectancy val- cise than the average American high school student,
ues may be identified that accounts for vigorous and therefore results may hold most closely for a
exercise, while a different subset may account for more active population of adolescents. Failure to
moderate exercise. An additional possibility is that obtain information on subjects ethnicity precluded
psychological response to physiological excitation an analysis that would determine the degree to
during exercise may differentiate those who exercise which this samples ethnic distribution compares to
vigorously from those who exercise moderately. An the ethnic distribution among American high school
equally important factor in the inability to explain students.
similar amounts of moderate and vigorous exercise
may be measurement error. Adolescent subjects may
have been uncertain whether certain lifestyle activi-
ties qualified as moderate exercise. If this is true, an References
alternative method for measuring moderate exercise 1. Blair S, Kampert J, Kohl H, et al. Influences of cardiorespira-
is a pressing problem. This investigation does pro- tory fitness and other precursors of cardiovascular disease and
vide justification for future authors to initiate and all-cause mortality in men and women. JAMA 1996;276:205
report methods by which moderate exercise can be
2. Blair S, Kohl H, Barlow C, et al. Changes in physical fitness
measured as well as additional predictive modeling and all-cause mortality: A prospective study of healthy and
of moderate exercise. unhealthy men. JAMA 1995;273:10938.

3. Blair S, Kohl H, Paffenbarger R, et al. Physical fitness and 14. Trost SG, Pate RR, Saunders R, et al. A prospective study of
all-cause mortality: A prospective study of healthy men and the determinants of physical activity in rural fifth-grade
women. JAMA 1989;262:2395401. children. Prev Med 1997;26:25763.
4. Kampert J, Blair S, Barlow C, Kohl H. Physical activity, 15. Saunders RP, Pate RR, Felton G, et al. Development of
physical fitness, and all-cause and cancer mortality: A pro- questionnaires to measure psychosocial influences on chil-
spective study of men and women. Ann Epidemiol 1996;6: drens physical activity. Prev Med 1997;26:2417.
4527. 16. Bandura A. Social Foundations of Thought and Action: A
Social Cognitive Theory. Englewood Cliffs, NJ: Prentice-Hall,
5. Lee I, Paffenbarger R, Hennekens C. Physical activity, physical
fitness and longevity. Aging 1997;9:211.
17. Kenyon GS. A conceptual model for characterizing physical
6. Trost SG, Pate RR, Sallis JF, et al. Age and gender differences activity. Res Q 1968;39:96 105.
in objectively measured physical activity in youth. Med Sci
18. Simon JA, Smoll FL. An instrument for assessing childrens
Sports Exerc 2002;34:350 5.
attitudes toward physical activity. Res Q 1974;45:40715.
7. Kann L, Kinchen S, Williams B, et al. Youth Risk Behavior 19. Schultz R, Smoll FL, Wood T. A psychometric analysis of an
Surveillance: United States, 1999. MMWR CDC Surveill Summ inventory for assessing childrens attitudes toward physical
2000;49:196. activity. Res Q 1981;45:40715.
8. US Department of Health and Human Services. Physical 20. Winters ER. Test of a Social Cognitive Theory-based Educa-
Activity and Health: A Report of the Surgeon General. At- tional Treatment to Increase the Frequency of Voluntary
lanta, Georgia: US Department of Health and Human Services, Moderate and Vigorous Physical Exercise Among Adolescents
Public Health Service, CDC, National Center for Chronic School Students. Columbus, OH: Ohio State University, 2001.
Disease Prevention and Health Promotion, 1996. 21. Petosa PS. Use of Social Cognitive Theory to Explain Exercise
9. Baranowski T, Anderson C, Carmack C. Mediating variable Behavior Among Adults. Columbus, OH: Ohio State Univer-
framework in physical activity interventions. How are we sity, 1993.
doing? How might we do better? Am J Prev Med 1998;15:266 22. Grunbaum J, Kann L, Kinchen S, et al. Youth Risk Behavior
97. Surveillance: United States, 2001. MMWR CDC Surveill Summ
10. Sallis JF, Prochaska JJ, Taylor WC. A review of correlates of
physical activity of children and adolescents. Med Sci Sports 23. Killen JD, Robinson TN, Telch MJ, et al. The Stanford adoles-
Exerc 2000;32:96375. cent heart health program. Health Educ Q 1989;16:26383.
24. Parcel GS, Simons-Morton B, OHara NM, et al. School pro-
11. Godin G, Shephard RJ. A simple method to assess exercise
motion of healthful diet and physical activity: Impact on
behavior in the community. Can J Appl Sport Sci 1985;10:
learning outcomes and self-reported behavior. Health Educ Q
12. Sallis JF, Saelens B. Assessment of physical activity by self- 25. Resnicow K, Cohn L, Reinhardt J, et al. A three-year evalua-
report: Status, limitations, and future directions. Res Q Exerc tion of the know your body program in inner-city schoolchil-
Sport 2000;71:S114. dren. Health Educ Q 1992;19:46380.
13. Reynolds KD, Killen JD, Bryson SW, et al. Psychosocial 26. Robert M, Harrell J, Bangdiwala S, et al. A school-based
predictors of physical activity in adolescents. Prev Med 1990; intervention can reduce body fat and blood pressure in young
19:54151. adolescents. J Adolesc Health 2002;31:12532.