Full text of DOI:10.1207/S15324796ABM2302_3

A Computerized Social Cognitive Intervention for Nutrition Behavior:
Direct and Mediated Effects on Fat, Fiber, Fruits, and Vegetables,
Self-Efficacy, and Outcome Expectations Among Food Shoppers
Eileen S. Anderson, Ed.D., Richard A. Winett, Ph.D., and Janet R. Wojcik, Ph.D.
Center for Research in Health Behavior
Virginia Tech
Sheila G. Winett, M.S. and Todd Bowden, B.S.
PC Resources, Inc.
Blacksburg, Virginia
ABSTRACT
structural equation analysis revealed self-efficacy and physical
outcome expectations mediated treatment effects on nutrition. In
addition, physical outcome expectations mediated the effect of
self-efficacy on nutrition outcomes. Implications for future com-
puter-based health promotion interventions are discussed.
This study examined the direct and mediated impact of a
self-administered, computer-based intervention on nutrition be-
havior, self-efficacy, and outcome expectations among super-
market food shoppers. The intervention, housed in kiosks in su-
permarkets and based on social cognitive theory, used tailored
information and self-regulation strategies delivered in 15 brief
weekly segments. The study sample (N = 277), stratified and
randomly assigned to treatment or control, was 96% female,
was 92% White, had a median annual income of about $35,000,
and had a mean education of 14.78 ±2.11 years. About 12% of
the sample reported incomes of $20,000 or less, and about 20%
reported 12 years or fewer of education. Analysis of covariance
immediately after intervention and at a 4- to 6-month follow-up
found that treatment led to improved levels of fat, fiber, and
fruits and vegetables. Treatment also led to higher levels of nu-
trition-related self-efficacy, physical outcome expectations, and
social outcome expectations. Logistic regression analysis deter-
mined that the treatment group was more likely than the control
group to attain goals for fat, fiber, and fruits and vegetables at
posttest and to attain goals for fat at follow-up. Latent variable
(Ann Behav Med
2001, 23(2):88–100)
INTRODUCTION
Diets high in fat and low in fiber and fruits and vegetables
have been associated with increased risk for cancers, diabetes,
cardiovascular diseases, other morbidities, and all-cause mortal-
ity (1,2). Although major initiatives during the last decade have
resulted in some favorable dietary changes (2), onlya minority of
Americans reach all recommended dietary goals of 30% or fewer
kilocalories(kcals)fromfat, 25gfiberperday, and5servingsper
day of fruits and vegetables (3). In addition, some purported di-
etary changes such as a reduction in calories from fat may be in
part illusionary. For example, for the last 2 decades, the mean
dailyconsumptionoffatgramshasremainedatabout80g,butca-
loricconsumption, apparentlyfromlowerfatbuthighercarbohy-
drate, foodshasincreasedbyabout150to200kcalsperday(4,5).
Thus,reportsofreductionsinpercentagekcalsfromfatfrom37%
to 34% (3) may need some qualification. In addition, despite
ready access by most population segments to a wide variety of
healthier foods, overweight and obesity may have increased be-
cause of high-fat, high-calorie “fast,” prepared, and snack foods
and by foods with an emphasis on quantity, such as “super-sized”
portions (4–8). Thus, a range of intervention strategies is needed
to alter the nutritional and activity patterns of population seg-
ments to reduce overweight, obesity, and risk for cancer and car-
diovascular and other diseases.
Self-administered computer-based interventions have the
potential to influence dietary changes in large, albeit self-se-
lected, groups of people. Computerized programs designed to
assess users’ health-related behaviors and provide personalized
behavior change recommendations (9–14) have shown promise,
but these “assessment and prescription” programs are generally
designed to complement care from physicians and other health
care professionals (15). Computer-based programs designed to
be self-administered and that go beyond assessing behavior and
prescribing remedies are beginning to emerge in the areas of
AIDS risk-reduction (16), eating disorders (17), adult fitness
This study was funded by The National Cancer Institute Grant
2RO1CA45926–06A1. The Institutional Review Board of the Virginia
Polytechnic Institute and State University approved the study. An ear-
lier version of this article was presented at the 2000 Annual Conference
of the Society of Behavioral Medicine, April 5–8, 2000.
We acknowledge the contributions of Janet Walberg-Rankin, Ph.D., in
developing NLS content and measurement strategies; Patricia G. Bick-
ley, Ph.D., in developing NLS content and procedures; Jessica Whitely
for background research; and Archie Fralin from the Kroger Co., Inc. in
assisting the conduct of the project in five stores. We also acknowledge
the work of Jennifer Fretz, Jessica Psujek, Felicity Mitton, Brad Harris,
and David Wang in database management. These staff members, as
well as Susan Allaie, Ann Jones, Oliver Chen, Anup Sharma, Ruby
Cockran, Kristi Graves, and Rosell Jeffries, were essential to partici-
pant recruitment, data entry, and related tasks. We are also grateful to
the store managers and personnel of Kroger supermarkets for their help
and support.
Reprint Address: E. S. Anderson, Ed.D., CRHB (0274), Department of
Psychology, Virginia Tech, Blacksburg, VA 24061. E-mail:
eileen@vt.edu
© 2001 by The Society of Behavioral Medicine.
88

Volume 23, Number 2, 2001
Computerized Nutrition Intervention
89
(18), youth nutrition and fitness (19), general health promotion
(20), and adult nutrition (21,22). The development of com-
puter-based interventions, however, must rest on a firm theoreti-
cal foundation to realize the potential for behavior change and
risk reduction (23,24). The Nutrition for a Lifetime System©
(NLS) is a self-administered, computer-based intervention
based on social cognitive theory (24) developed through a series
of small efficacy studies centered in supermarkets (22,25).
Though of minimal dose, the NLS provides users with personal-
ized information, behavioral strategies, and incentives for
change as well as a vehicle for planning and receiving feedback
on personal behavior change goals.
pectations—processes that would also directly improve nutri-
tion behavior. The direct and mediated effects of treatment
should be evaluated with measurement systems designed to as-
sess the nature and extent of the influence of constructs used to
develop the intervention (24). Unfortunately, such measurement
systems for less intensive individual, community-based, and
large-group interventions are virtually nonexistent. Our study
used measures of self-efficacy and outcome expectations shown
to explain nutrition behavior among food shoppers (30) to ex-
plore the extent to which treatment effects were mediated by
these social cognitive variables.
Although there is a growing consensus that intensive, indi-
vidual-based interventions can be effectively evaluated by mul-
tiple, diet-recall interviews (26), such assessments often are not
feasible or warranted for evaluating less intensive individual-,
group-, or community-based interventions such as the NLS
(26,27). Food frequency surveys have become more prevalent in
assessing community- and population-based dietary behavior
(26), and when collected concurrently with diet records, they
have been shown to be sensitive to change associated with clini-
cal and self-help nutrition interventions (27). However, self-re-
port measures of health behavior and risk factors, especially
when used alone, have been found to seriously underestimate
risk behaviors in the general population (28). One effective way
to increase the accuracy of self-reported food frequency surveys
is to collect them in conjunction with perceived or real verifiable
objective data (28). Previous research with the NLS found that
annotated food shopping receipts, collected over a period of
time, can provide an objective, sensitive measure of nutrition be-
havior of individual food shoppers (21,22,25). In conjunction
with a database of supermarket foods developed specifically to
evaluate nutritional content of food purchases (21), supermarket
receipts can provide rich data regarding the nutritional status of
individuals and families whose diets consist largely of foods
purchased at food stores. For individuals who purchase a
smaller proportion of foods at food stores, food shopping re-
ceipts are less sensitive to intervention-related dietary change
(21,22,29). In light of the cost and response burden of annotat-
ing, collecting, and analyzing comprehensive samples of receipt
data, food shopping receipts may be most useful as an objective
measure used to corroborate self-report food frequency or
diet-recall data. Our study used a system that included food
shopping receipts and food frequency questionnaires. These
measures—individually, together in a composite, and together
in a latent-variable model—provide data to evaluate the effects
of the NLS on nutrition behavior and nutritional goal attainment
among a diverse group of food shoppers.
METHOD
Participants
Of 795 shoppers expressing interest in the study (i.e., re-
turned at least some part of the enrollment packet; see the fol-
lowing Procedure section), 464 (58%) were recruited for the
study (i.e., returned complete packets and one initial food shop-
ping receipt). Of the recruited sample, 363 (79%) returned at
least 4 weeks of food shopping receipts and were enrolled in the
study. Of the enrolled sample, 67 (18%) dropped out prior to in-
tervention. Of the 296 who participated in the intervention phase
of the study, 148 were assigned to the control condition, and 148
were assigned to use the NLS kiosk. All 148 control participants
completed the intervention phase and provided posttest data.
Nineteen NLS participants dropped out, however, during the in-
tervention phase (13% of NLS participants), resulting in 129
NLS participants providing posttest data (posttest N = 277). At
follow-up, 91 control participants and 72 NLS participants pro-
vided data (59% of intervention participants, 45% of enrolled
participants).
Procedure
Participants were recruited in five supermarkets using pro-
cedures consistent with store operations involving brief
face-to-face contact followed by a mail back of enrollment ma-
terials (these procedures are described in detail elsewhere) (25).
To enroll in the study, participants were required to complete a
detailed demographic survey and to mail back at least 4 weeks
of annotated food shopping receipts. After enrolling, partici-
pants continued to send baseline food shopping receipts and
completed the Block95 Food Frequency Questionnaire (FFQ)
(31) and the NLS Food Beliefs Survey. After baseline, enrolled
participants were stratified by race, education, and family size
and randomly assigned from stratification groups into the NLS
treatment or the no-treatment control condition. Participants
submitted food shopping receipt data throughout the interven-
tion phase and completed FFQs and NLS Food Belief Surveys
immediately after intervention. Four to 6 months after interven-
tion, participants completed final FFQs and submitted 6 weeks
of food shopping receipts.
The NLS intervention provides information, behavioral
strategies, and incentives to guide users to set and evaluate a se-
ries of goals regarding specific nutrition-related behaviors. So-
cial cognitive theory suggests nutrition goal setting and self-reg-
ulation would directly improve participants’ nutrition-related
behavior. In addition to the direct effect of these intervention
strategies, social cognitive theory suggests such interventions
would increase nutrition-related self-efficacy and outcome ex-
Participants received $10 for completing enrollment forms,
$10 for baseline, $15 for posttest FFQs and NLS Food Beliefs
Surveys, and $20 for follow-up FFQs. Participants also received
$5 per week for returning annotated food shopping receipts dur-

90
Anderson et al.
Annals of Behavioral Medicine
ing baseline, $7 per week during the intervention phase, and $10
per week during the follow-up phase. Participants using the NLS
program had access to coupons (see following) and redeemed a
meanof$34incouponsduringintervention. Controlparticipants
received an equivalent payment at the end of intervention.
The NLS–SFD contains information on grams of fat, grams of
fiber, kcals, and servings of fruits and vegetable for each ounce
of the purchased food. The GRRP automatically updated the
NFS–SFD with new food items and receipt label links; the
GRRP also automatically initiated needed updating of the
NFS–SFD nutrition information. Food purchases were aggre-
gated for study phase generating baseline, posttest, and fol-
low-up values for total grams of fat, total grams of fiber, total
kcals, and total servings of fruits and vegetables for each partici-
pant. These total phase values served as the basis for percentage
of kcals from fat, grams of fiber per 1,000 kcals, and servings of
fruits and vegetables per 1,000 kcals.
FFQs were collected from participants 4 weeks into base-
line, at the end of posttest, and at the beginning of follow-up.
Participants received instruction to report food consumption for
the respective phase of the intervention. Block Dietary Data
Systems (BDDS) (31) scored the FFQs and provided values for
percentage of kcals from fat. Based on the scores provided by
BDDS, we generated values for fiber grams per 1,000 kcals
(daily fiber grams/[daily kcals/1,000]) and servings of fruits and
vegetables per 1,000 kcals (daily servings of fruits + daily serv-
ings of vegetables/[daily kcals/1,000]).
For participants submitting both food shopping receipt data
and FFQ data, we calculated composite (i.e., mean) nutrient val-
ues for percentage of kcals from fat, fiber grams per 1,000 kcals,
and fruit and vegetable servings per 1,000 kcals at each assess-
ment point (baseline, posttest, and follow-up). In addition to
providing an outcome measure incorporating both nutrition in-
struments, these composites served as the basis for gauging the
magnitude of nutritional change associated with the interven-
tion and for determining nutritional goal attainment.
NLS Intervention
Stand-alone kiosks located in the five recruitment super-
markets housed the NLS computers. NLS users accessed the
system with assigned passwords. Overall, the NLS program
guided participants to decrease fat and to increase fiber, fruits,
and vegetables in their food purchases and consumption. Spe-
cifically, the NLS segments focused on increasing purchases
and consumption of cruciferous vegetables, fruits rich in vita-
mins C and A, moderate- to high-fiber bread and cereals, lowfat
or nonfat dairy, and other lean protein sources. In addition, the
NLS focused on decreasing purchases and consumption of dis-
cretionary fat (especially butter and margarine), high-fat ground
beef, and high-fat snack foods. The NLS consisted of 15 weekly
segments: 10 content segments followed immediately by 5 seg-
ments devoted to maintenance. Using pictures, graphics, and an
audio track, each segment provided prescriptive information,
suggested strategies for monitoring and planning food pur-
chases and meal preparation, and provided opportunities for
personalized goal setting and feedback for each targeted food
group or behavior change strategy. In addition, the NLS offered
targeted food coupons with the type, order, and value of coupons
dependent on the program segment’s content, a product’s cost,
and a weekly coupon allotment. The total value of the coupons
available each week ranged from about $8 to $12. Coupons were
printed at the kiosk, and to promote immediate use, redemption
was limited to the NLS user at the kiosk store within 1 week of
printing. Completion of all parts of each segment required 5 to
10 min, with longer segments providing behavior change plan-
ning and feedback. More details about the NLS program can be
found elsewhere (21,25).
Social cognitive variables. The NLS Food Beliefs Survey
(30) measured self-efficacy and outcome expectations 4 weeks
into baseline and at the end of the intervention phase. The sur-
vey’s self-efficacy questions emerged from formative work with
past users of the NLS (21,22) and assessed self-efficacy for buy-
ing, preparing, eating, and serving lower fat and higher fiber
foods and more fruits and vegetables. The scale prompted par-
ticipants to rate their certainty in performing the behaviors “all
or most of the time … for a long time … in a lot of different situ-
ations,” with the scale ranging from 1 (very sure I cannot) to 10
(very sure I can). Examples of self-efficacy items included “I
can bring a slice of bread with fiber to work or school for a
snack” and “I can get at least 4 servings from every pound of
ground beef I buy.” Factor analysis of the items (30) resulted in
the three self-efficacy factor-based scales (32) used in our analy-
ses: (a) Self-Efficacy for Increasing Fiber and Fruit and Vegeta-
bles (internal consistency, α = .90), (b) Self-Efficacy for De-
creasing Fat in Snacks (α = .88), and (c) Self-Efficacy for
Decreasing Fat in Meals (α = .83).
Measures
Nutrition outcomes. Nutrition outcomes were measured
with food shopping receipts, FFQs, and a composite of the two
measures. Values for percentage kcals from fat, fiber grams per
1,000 kcals purchased, and servings of fruits and vegetables per
1,000 kcals were gleaned from each measure.
This study utilized food shopping receipt data collected
during the last 6 weeks of each study phase. Receipts spanned an
average of 34 days for each phase of the study, 135 food items
per participant at baseline, 126 items at posttest, and 140 items
at follow-up. Food purchases were analyzed for fat, fiber, and
fruit and vegetable content utilizing the NLS Supermarket
Foods Database (SFD) (21). In general, receipted food items
along with amount purchased, price paid, date and store of pur-
chase, and participant identifiers were entered into the NLS
Grocery Receipt Recording Program (GRRP). The GRRP, com-
puter software culminating 10 years of NLS research (21,22),
linked each food item through its receipt label to the NLS–SFD.
TheNLSFoodBeliefsSurveyalsoassessedthethreecompo-
nents of outcome expectations defined by Bandura (24)—physi-
cal, social, and self-evaluative. Questions on the scale asked par-
ticipants to “tell us what would happen if you ate and served your
family a healthy diet every day.” Participants used a 5-point

Volume 23, Number 2, 2001
Computerized Nutrition Intervention
91
Likert-type scale to rate how much they agreed that certain out-
comes would result. Items included physical outcomes (e.g., “I
will be hungrier,” “The food I eat won’t taste good,”) and
self-evaluative outcomes (e.g., “I will be doing what I know I
should.”). Social outcomes, specifically family reactions to spe-
cific nutrition-related behaviors, were measured with a
Likert-type scale ranging from 1 (My family would strongly dis-
like or be angry) to 5 (My family would strongly like or be glad).
Social outcome items included “If serve bread with fiber to my
family …” and “If I serve more vegetables at meals …”. Factor
analysis (30) yielded four factor-based (32) outcome-expecta-
tions scales: (a) Expected Family Reactions (α = .88), (b) Ex-
pected Health Outcomes (α = .91), (c) Expected Budgetary Out-
comes(α =.84),and(d)ExpectedAppetiteSatisfaction(α =.76).
Demographic characteristics. Participants provided demo-
graphic information during study recruitment and enrollment. In
addition to participant age, sex, and race or ethnic background,
participants reported the highest level of education achieved by
all adults in the home, the occupation of all adults in the home,
the household income, and the number of children living in the
home. Although age and education variables were measured in
years, income was measured in categories ranging from 1 (less
than $10,000 per year) to 10 (greater than $90,000 per year).
Primary occupations of adults in the home reported by partici-
pants were coded based on Duncan’s system (33) ranging from 0
(no paid job or profession indicated) to 13 (school teacher).
Standardized values for the highest level of education of adults
in the home, highest level of occupation of adults in the home,
and family income contributed equally to a composite variable
of household socioeconomic status (SES).
FIGURE 1 Social cognitive model of treatment effects on nutrition
behavior.
allowed us to evaluate to what extent the total effects of treat-
ment were direct or independent of social cognitive variables
and to what extent they were indirect or mediated by the social
cognitive variables. SEM also allowed us to evaluate how well
the models fit the NLS study data: (a) goodness-of-fit index of
.90 or greater, (b) adjusted goodness-of-fit index of .80 or
greater, (c) normed fit index of .90 or greater, (d) root mean
square error of the approximation equal to or less than .05 (α =
.05), and (e) chi-square. The squared multiple correlation (R2)
associated with the latent fat, fiber, and fruit and vegetable vari-
ables allowed us to evaluate the effectiveness of the models in
explaining the variance observed in the sample’s nutrition be-
havior at post and follow-up.
Statistical Analysis
Prior to conducting the SEM analysis, variables were exam-
ined for normality. Although several variables were slightly
skewed, servings of fruits and vegetables and grams of fiber
from food shopping receipts had severe kurtosis (kurtosis statis-
tics ranging from 5.99 to 50.06). These variables, at posttest and
follow-up, were transformed using the Log10 transformation
operation in SPSS, Version 10.0 (35), resulting in normally dis-
tributed variables. LISREL software, Version 8.3 (36), then gen-
erated a covariance matrix of measures of treatment, social cog-
nitive, and outcome variables from raw data imported (including
transformed variables) from SPSS (35). The structural models
used measures of nutrient values from food shopping receipts
and FFQs. Self-efficacy, physical outcome expectations, and so-
cial and self-evaluative expectations variables in the model re-
flect those measured variables influenced by the NLS treatment
(i.e., ANCOVA). Error terms for latent variables in the models
with single indicators were estimated as the indicator’s variance
times one minus the indicator’s reliability (37).
Analysis of covariance (ANCOVA) determined treatment ef-
fects on FFQ, receipt, and composite measures of percentage of
kcals from fat, fiber grams per 1,000 kcals, and fruit and vegetable
servings per 1,000 kcals at post and follow-up. ANCOVA was also
used to determine treatment effects on social cognitive variables
(self-efficacy and outcome expectations) at posttest. Participant
demographic variables (SES, age, and number of children in the
home) and baseline values for the corresponding dependent vari-
ables served as covariates in these analyses.
Latent variable structural equation modeling (SEM) exam-
ined the extent to which any observed treatment effects on nutri-
tion outcomes were mediated by treatment effects (if any) on so-
cial cognitive variables. Consistent with social cognitive theory,
SEM also examined the extent to which the effect of self-effi-
cacy on the nutritional outcomes was mediated by outcome ex-
pectations. A technique most commonly used to analyze
nonexperimental data, latent-variable SEM is a preferred
method for examining mediating effects within experimental
studies (34). SEM controls for measurement error by incorpo-
rating multiple measures or measurement reliabilities, and it al-
lows simultaneous testing of relevant relations (i.e., paths). Fig-
ure 1 illustrates the six models tested—models at posttest and
follow-up for fat, fiber, and fruit and vegetable outcomes. SEM
Logistic regression analysis determined the effects of treat-
ment and social cognitive variables on nutritional goal attain-
ment at post and follow-up. A participant was considered to
have attained the goal for fat if the composite measure of fat in-
dicated 30% or fewer kcals from fat. The fiber goal was attained

92
Anderson et al.
Annals of Behavioral Medicine
if the composite measure of fiber indicated at least 25 g fiber per
day (based on a 2,000-calorie-per-day diet). Last, the goal for
fruits and vegetables was attained if the composite measure of
fruit and vegetable servings indicated at least five servings of
fruits and vegetables per day (based on a 2,000-calorie-per-day
diet). Goal attainment analysis was conducted in two steps. The
first step determined the effect of treatment on posttest and fol-
low-up goal attainment controlling for demographic variables
and baseline goal attainment rates. The second examined to
what extent the social cognitive variables added to the predic-
tion of goal attainment outcomes. The multivariate odds-ratios
(MORs; α < .10) resulting from the logistic regression analyses
represent the relative frequency of goal attainment when the pre-
dictor variable changes by one unit: no treatment compared to
treatment or 1 SD change in other noncategorical variables.
= $67.92, SD = $28.28). Of the coupons selected, about 45%
were redeemed (M = $33.92, SD = $23.86). Small, positive cor-
relations were observed between the composite measure of
posttest fiber and coupon redemption (r = .22, p < .05) and be-
tween composite posttest servings of fruits and vegetables and
kiosk revisits (r = .23, p < .05), but such correlations did not hold
at follow-up. Posttest and follow-up composite measures were
independent of the number of segments viewed by NLS partici-
pants (Pearson r, α = .05).
Nutritional and Social Cognitive Outcomes
Control group and NLS user group means and standard de-
viations at baseline, posttest, and follow-up for the three nutri-
tional measures are displayed in Table 1. ANCOVA at baseline
indicated no between-group differences on any measure of fat,
fiber, and fruit and vegetable servings controlling for partici-
pants’household SES, number of children, and age (α = .10). At
posttest, controlling for demographic characteristics and base-
line fat levels, NLS users had lower levels of fat in the food
shopping receipts, the FFQ, and the composite fat measure. Al-
though the composite percentage of kcals from fat among con-
trol group participants increased slightly from baseline to
posttest, among NLS users composite fat decreased by 9%.
These differences were maintained at follow-up in the FFQ and
composite measures but not in the food shopping receipt mea-
sure (see Table 1).
Among the covariates, only household SES influenced par-
ticipants’ posttest percentage of kcals from fat as measured by
the food shopping receipts, F(1, 255) = 5.986, p < .05, and the
composite fat measure, F(1, 189) = 4.115, p < .05. Household
SES influenced fat levels neither from these measures at fol-
low-up nor from fat as measured by the FFQ at either assess-
ment point. Although household SES negatively correlated with
fat levels from food shopping receipts (r = –.27, p < .001) and
from the composite fat measure (r = –.22, p < .01), post hoc
analyses of interactions between SES and treatment indicated
the effects were independent (α = .10).
In addition to lower levels of fat at posttest, NLS users had
higher levels of fiber grams per 1,000 kcals (controlling for de-
mographic and baseline fiber variables) on the food shopping re-
ceipts, on FFQs, and in the posttest composite fiber measure.
Although the composite fiber grams per 1,000 kcals among con-
trol participants decreased slightly from baseline to posttest,
among NLS users composite fiber increased by almost 19%.
These differences were maintained at follow-up in the FFQ and
composite measures but not in the food shopping receipt mea-
sure. None of the demographic covariates influenced any of the
posttest or follow-up measures of fiber.
At posttest, NLS users also had higher levels of fruits and
vegetable servings per 1,000 kcals in the FFQ and the composite
fruit and vegetable measure but not in the food shopping re-
ceipts. Although the composite servings per 1,000 kcals among
control participants remained unchanged from baseline to
posttest, among NLS users composite servings increased by
over 20%. These differences were maintained at follow-up.
None of the demographic covariates influenced any of the
posttest or follow-up measures of fruits and vegetable servings.
RESULTS
The sample participating in the NLS intervention (n = 277)
was 96% female, was 92% White, had a median annual income
of about $35,000, and had a mean of 14.78 ±2.11 years of educa-
tion. About 12% of the sample reported incomes of $20,000 or
less, and about 20% reported 12 years or fewer of education.
Compared to participants completing intervention, participants
who were recruited but did not enroll in the study (n = 433) were
younger, had less education, had lower levels of occupational
status, and had more children. Participants who did not enroll (n
= 433) had a mean age of 42.37 (SD = 11.86, p < .001) and mean
years of education of 14.37 (SD = 1.92, p < .05). Participants not
enrolling in the study compared to those who completed inter-
vention had mean Duncan-based (33) occupational levels of
8.28 (SD = 3.40) versus 9.25 (SD = 3.34, p < .001) and 1.13 chil-
dren (SD = 1.18) versus 0.70 children (SD = 1.00, p < .001). The
33 participants who enrolled but did not participate in the inter-
vention (i.e., baseline dropouts) had fewer children (M = 0.76,
SD = 0.74, p < .05) than those who did not enroll. In addition,
baseline dropouts had lower incomes than participants who did
not enroll and than participants who completed intervention
(mean of about $25,000 vs. mean of about $35,000, p < .01).
Finally, baseline dropouts had lower levels of occupational sta-
tus (M = 6.44, SD = 3.85) than either of the other groups (see
earlier, p < .05).
Use of the NLS Intervention
Of the 148 participants assigned to NLS intervention condi-
tion, 129 viewed at least 3 segments of the 15-segment interven-
tion with a mean of 10.36 segments viewed (SD = 3.96). Within
the week of each segment’s first viewing, NLS users were al-
lowed to visit the kiosk to review content as many times as they
wished. Most participants did not revisit segments (M = 0.86,
SD = 1.41), but about 20% of the participants made 2 to 10 total
revisits to the NLS kiosk to review 1 or more of the 15 segments.
Coupons were offered to participants only once for each seg-
ment (i.e., not when revisiting program segments). Of the ap-
proximately $140 in coupons available to NLS users if they
viewed every segment and selected all available coupons, NLS
users selected a range of $10.15 to $127.81 in total coupons (M

TABLE 1
Baseline, Posttest, and Follow-up Means and Standard Deviations for Control and NLS Users: Three Measures of Nutrient Values
Food Shopping Receipts
Food Frequency Questionnaires
Composite
NLS
Variable
Control
NLS
F (df)
Control
NLS
F (df)
Control
F (df)
Fat (% calories)
Baseline
M
SD
n
31.67
8.80
148
32.78
9.60
129
0.860
(1, 267)
33.64
7.86
137
33.49
7.51
125
0.035
(1, 254)
32.74
6.85
137
33.24
7.28
124
0.240
(1, 253)
Posttest
M
SD
n
33.55
9.40
145
31.13
10.15
121
6.676***
(1, 255)
33.25
8.33
121
30.37
7.51
87
11.167***
(1, 200)
33.36
7.02
118
30.10
6.92
79
15.228****
(1, 189)
Follow-up
M
SD
n
31.93
8.42
100
30.60
8.44
72
1.564
(1, 167)
34.29
7.96
93
31.39
6.96
72
5.47**
(1, 158)
33.19
6.93
90
31.00
6.42
72
4.535**
(1, 155)
Fiber (g/1,000kcals)
Baseline
M
SD
9.29
3.78
9.00
3.22
0.498
(1, 267)
8.72
3.45
8.86
2.76
0.194
(1, 254)
9.00
3.32
8.97
2.57
0.007
(1, 253)
Posttest
M
SD
n
8.95
4.14
145
9.96
4.72
121
4.438**
(1, 255)
8.97
3.70
121
10.87
3.08
87
23.487****
(1, 200)
8.87
3.21
118
10.63
3.29
78
19.808****
(1, 189)
Follow-up
M
SD
n
9.73
3.75
100
10.48
4.96
72
1.302
(1, 167)
8.65
3.46
93
10.74
3.26
72
20.993****
(1, 158)
9.21
3.26
90
10.61
3.37
72
8.203***
(1, 155)
Fruit and vegetables
(servings/1,000kcals)
Baseline
M
SD
n
2.96
1.86
147
2.76
1.35
129
0.867
(1, 266)
2.77
1.40
137
2.79
1.14
125
0.039
(1, 254)
2.85
1.34
136
2.78
1.06
124
0.164
(1, 252)
Posttest
M
SD
n
2.75
1.95
145
3.08
2.61
121
1.571
(1, 254)
2.89
1.46
121
3.52
1.31
87
14.436****
(1, 200)
2.77
1.27
118
3.37
1.79
79
6.871***
(1, 189)
Follow-up
M
SD
n
3.05
1.37
100
3.35
2.39
72
1.106
(1, 167)
2.80
1.40
93
3.34
1.46
72
5.249**
(1, 158)
2.5
1.18
90
3.35
1.56
72
3.286*
(1, 155)
Note. Analysis of covariance (ANCOVA) analyses of between-group differences controlled for household socioeconomic status, participant age, and number of
children in the home. ANCOVA analyses at posttest and follow-up also controlled for baseline outcome values. NLS = Nutrition for a Lifetime System©; kcals =
kilocalories.
*p < .10. **p < .05. ***p < .01. ****p < .001.
Control group and NLS user group means and standard de-
viations for the measures of self-efficacy and outcome expecta-
tions at baseline and posttest are displayed in Table 2. ANCOVA
analyses with demographic covariates indicated the groups were
equivalent on all social cognitive measures at baseline. At
posttest, NLS users had higher levels of self-efficacy for de-
creasing fat in meals than control participants. NLS users also
had higher expectations at posttest that their families would ac-
cept healthier foods (social outcome expectations) and higher
expectations that healthier foods would be satisfying (physical
outcome expectations). There were no differences between
study groups on other measures of posttest self-efficacy and out-
come expectations. Finally, none of the demographic covariates
influenced any of the posttest social cognitive variables.
Mediation of Treatment Effects and Effects of
Self-Efficacy on Nutrition Outcomes
Figure 1 illustrates the hypothesized model of mediating ef-
fects of social cognitive variables influenced by treatment. In the
93

94
Anderson et al.
Annals of Behavioral Medicine
TABLE 2
Baseline, Post, and Follow-Up Means and Standard Deviations for Control and NLS Users: Self-Efficacy and Outcome Expectations
Self-Efficacy
Low-Fat Meals
NLS
Low-Fat Snacks
Fruits, Vegetables, & Fiber
Control
F (df)
Control
NLS
F (df)
Control
NLS
F (df)
Baseline
M
SD
n
6.70
1.79
139
6.89
1.82
125
1.004
(1, 256)
7.24
1.67
139
7.29
1.86
125
0.025
(1, 256)
7.29
1.78
139
7.50
1.58
125
1.038
(1, 256)
Posttest
M
SD
n
6.68
1.73
132
7.01*
1.67
98
3.999**
(1, 224)
7.18
1.61
132
7.38
1.80
98
0.809
(1, 224)
7.38
1.67
132
7.58
1.73
98
0.058
(1, 224)
Outcome Expectations
Budgetary Outcomes Health Outcomes
Appetite Satisfaction
Family Reactions
NLS
Control
NLS
F (df)
Control
NLS
F (df)
Control
NLS
F (df)
Control
F (df)
Baseline
M
SD
n
3.92
0.90
139
3.97
0.93
125
0.024
(1, 256)
3.39
1.09
139
3.40
1.10
125
0.101
(1, 256)
4.29
0.66
139
4.37
0.59
125
1.462
(1, 256)
3.15
0.79
138
3.17
0.75
122
0.010
(1, 257)
Posttest
M
SD
n
3.94
0.91
132
4.13*
0.88
98
2.799*
(1, 224)
3.40
1.07
132
3.39
1.14
98
0.292
(1, 224)
4.32
0.63
132
4.40
0.58
98
0.225
(1, 224)
3.16
0.76
131
3.31** 5.729**
0.72
96
(1, 224)
Note. Analysis of covariance analyses of between-group differences at posttest controlled for household socioeconomic status, participant age, number of
children in the home, and baseline values. NLS = Nutrition for a Lifetime System.
*p < .10. **p < .05.
model self-efficacy, physical outcome expectations and social
outcome expectations mediate treatment effects on nutrition be-
havior. In addition, physical outcome expectations and social out-
come expectations mediate the effects of self-efficacy on nutri-
tion behavior. In the tested models, the measures of self-efficacy
and outcome expectations were those found to have been influ-
enced by the NLS treatment (i.e., ANCOVA analyses: self-effi-
cacy for decreasing fat in meals, expectations for family reac-
tions, and expectations for appetite satisfaction). Latent-variable
SEM analysis (LISREL software, Version 8.3) (36) tested six so-
cial cognitive models of treatment effects on nutrition behavior:
posttest and follow-up models each for fat, fiber, and fruit and
vegetables. Table 3 contains the intervariable correlations, means,
and standard deviations associated with the measured variables in
the latent-variable models. Table 3 also contains the completely
standardized factor loadings of latent variables on the measured
variables within the models. Table 4 contains the standardized co-
efficients associated with the total, direct, and mediated effects of
treatment and social cognitive variables on post and follow-up fat,
fiber, and fruit and vegetables along with the fit statistics associ-
ated with each tested model.
ble levels at post and follow-up than control participants. In addi-
tion, SEM analysis confirmed the NLS’s effects increasing
posttest self-efficacy, physical outcome expectations, and out-
come expectations. Overall, the social cognitive models of NLS
treatment effects (Figure 1) provided excellent fit to the data from
food shoppers, explaining from 33% to 59% of the variance ob-
served in the latent fat, fiber, and fruit and vegetable variables at
posttest and follow-up (see Table 4).
In addition to confirming treatment effects, SEM analysis
revealed that the total effects of treatment on nutritional out-
comes were partly direct or independent of the social cognitive
variables (see Table 4). The total effects of treatment were also
partly mediated, or were exerted through treatment’s effects on
self-efficacy and outcome-expectations variables, which then
influenced nutrition behavior. Self-efficacy and physical out-
come expectations mediated the total effect of treatment on fiber
and fruits and vegetables at both posttest and follow-up and on
fat at follow-up. Self-efficacy, alone, mediated treatment effects
on fat at posttest. Social outcome expectations did not mediate
the effects of NLS treatment among food shoppers. Self-effi-
cacy exerted the strongest total effects on nutritional outcomes
observed in the SEM analyses. Physical outcome expectations
mediated the effects of self-efficacy on fiber at posttest and fol-
low-up and on fat and fruit and vegetable servings at follow-up.
SEM analysis confirmed the NLS treatment’s total effects on
posttest and follow-up fat, fiber, and fruit and vegetable servings.
NLS users had lower fat, higher fiber, and higher fruit and vegeta-

95

96
Anderson et al.
Annals of Behavioral Medicine
TABLE 4
Direct, Mediated, and Total Effects of NLS Treatment and Social Cognitive Variables at Posttest and Follow-Up:
Standardized Path Coefficients From Structural Equation Modeling Analyses
Dependent Variables (Posttest Coefficient/Follow-Up Coefficient)
Fruits &
Vegetables
(Servings/1,000
kcals)
Independent
Variables
Physical Outcome Social Outcome
Fat
(% kcals)
Fiber
(g/1,000 kcals)
Self-Efficacy
Expectations
Expectations
Total Effects
Treatment
.15**/.14**
.17**/.18*
.47****/.48****
.14**/.14**
–.31***/–.23**
.34****/.28***
.30***/.14*
.48***/.65***
.13/.21**
Self-efficacy
Physical outcome
expectations
Social outcome
expectations
Direct Effects
Treatment
Self-efficacy
Physical outcome
expectations
Social outcome
expectations
Mediated Effects
Treatment
.68****/.66**** –.69***/–.51**** .57****/.70****
–.08/–.17**
.19**/.22**
.03/.–03
.02/.04
.10/.11
.15**/.14*
.10/.11
.47****/.48****
.04/.04
–.20**/–.12*
.23**/.16***
.47***/.47****
.19**/.22**
.21***/.02
.35**/.47***
.13/.21**
.68****/.66**** –.57****/–.55***
–.08/–.17*
.03/.–03
.02/.04
.10/.11
.07**/.07*
.10**/.09*
–.11**/–.11**
–.02/–.10
.11***/.13**
.10/.20*
.09**/.12**
.13/.17*
Self-efficacy
Fit Statistics
(Posttest/Follow-Up)
GFI
AGFI
NFI
.99/.99
.96/.95
.98/.97
1.00/.98
.98/.88
.99/.95
1.00/.99
.98/.94
.99/.97
RMSEA
.03 (p = .54)/
.03(p = .52)
4.74 (p = .32)/
4.56 (p = .54)
.57/.49
.00 (p = .84)/
.10 (p = .11)
2.23 (p = .69)/
10.16 (p = .04)
.46/.59
.00 (p = .91)/
.05 (p = .41)
1.54 (p = .82)/
5.67 (p = .23)
.33/.47
χ2(4)
R2
Note. NLS = Nutrition for a Lifetime System; kcals = kilocalories; GFI = goodness-of-fit index; AGFI = adjusted goodness-of-fit index; NFI = normed fit
index; RMSEA = root mean square error of the approximation.
*p < .10. **p < .05. ***p < .01. ****p < .001.
Nutritional Goal Attainment
Logistic regression also determined the extent to which the
social cognitive variables affected by NLS treatment contrib-
uted to posttest and follow-up goal attainment. In these analyses
self-efficacy for preparing and eating low-fat meals, expecta-
tions for appetite satisfaction, and expectations for family ac-
ceptance of healthier foods along with treatment and the base-
line covariates predicted goal attainment; Table 5 displays the
resulting MOR statistics with their confidence intervals. Higher,
posttest self-efficacy for low-fat meals contributed to higher
rates of posttest and follow-up attainment of fat goals. Higher,
posttest expectations for family acceptance of healthier foods
contributed to higher rates of posttest and follow-up attainment
of fruit and vegetable goals and posttest attainment of fiber
goals. Posttest expectations of appetite satisfaction did not con-
tribute to nutritional goal attainment.
Logistic regression analyses determined the effects of NLS
treatment on participants’ nutritional goal attainment at posttest
and follow-up. Table 5 contains MOR statistics associated with
the treatment along with their 90% confidence intervals; each
analysis included participant age, number of children in the
home, household SES, and baseline goal attainment as
covariates. At posttest, NLS users attained 30% or fewer kcals
from fat at almost four times the rate of the control group (p <
.001). NLS users reached 25 grams of fiber per day and five
servings of fruits and vegetables a day (based on a 2,000
kcals/day diet) at about two times the rate of control participants
at posttest. At follow-up, NLS users attained fat goals at four
times the rate of the control group (p < .05) but were not more
likely to reach fiber and fruit and vegetable goals.

Volume 23, Number 2, 2001
Computerized Nutrition Intervention
97
TABLE 5
Multivariate Effects of NLS Treatment and Social Cognitive Variablesa on Nutritional Goal Attainment Rates
at Posttest and Follow-Up Among Food Shoppers
Fat
(≤ 30% kcal)
Fiber
(25+ g/Day)
Fruits & Vegetables
(5+ Servings/Day)
Variable
MORb
90% CI
MOR
90% CI
MOR
90% CI
Posttest
NLS treatment versus control
3.75****
1.97–7.12
2.07*
1.01–4.25
1.81*
1.01–3.25
NLS treatment versus control
Self-efficacy for low-fat meals
Expect appetite satisfaction
Expect family acceptance
Follow-up
3.38**
1.33*
0.96
1.73–6.60
1.03–1.73
0.65–1.43
0.96–2.99
2.10
1.05
1.10
1.90*
0.98–4.50
0.78–1.41
0.68–1.76
1.08–3.33
1.62
1.12
1.24
1.81**
0.86–3.05
0.90–1.41
0.87–1.76
1.11–2.96
1.70
NLS treatment versus control
2.26**
1.18–4.35
1.49
0.68–3.27
1.25
0.66–2.35
NLS treatment versus control
Self-efficacy for low-fat meals
Expect appetite satisfaction
Expect family acceptance
2.09*
1.67**
1.38
1.02–4.24
1.24–2.26
0.88–2.15
0.64–2.08
1.40
1.23
1.85
1.09
0.61–3.21
0.88–1.70
1.01–3.39
0.59–2.03
0.95
1.27
1.34
1.98**
0.46–1.93
0.99–1.63
0.91–1.96
1.12–3.50
1.15
Note. NLS = Nutrition for a Lifetime System; kcals = kilocalories; MOR = multivariate odds ratios; CI = confidence interval.
DISCUSSION
preparing and eating low-fat meals, expected appetite satisfac-
tion, and expected family reactions. Further, changes in
self-efficacy and outcome expectations led to improved nutri-
tion behavior, explaining part of the total effect of the NLS treat-
ment on nutrition behavior among food shoppers. Indeed, the
social cognitive model of the NLS treatment provided a good fit
to this study’s outcome data and explained a substantial portion
of the variance observed in fat, fiber, and fruit and vegetable (la-
tent) variables.
This test of a computerized social cognitive intervention to
improve nutrition behavior found that although use of the
self-administered system involved self-selected participants, di-
verse users of the minimal-dose program made and maintained
meaningful nutritional changes when compared to no-treatment
control participants. Users of the NLS decreased fat and in-
creased fiber and fruits and vegetables and generally maintained
these changes at the 4- to 6-month follow-up. The nutrition be-
havior of participants in the no-treatment control group, on the
other hand, generally worsened over the course of the study. In
addition, NLS users were more likely than control participants
to meet nutritional goals immediately after treatment and were
able to maintain higher fat goal attainment rates at follow-up.
These changes compare well to interventions involving consid-
erable interpersonal contact and supports (38), suggesting that
the NLS represents a type of intervention that can be used to ef-
fectively alter the fat, fiber, and fruit and vegetable content of
food purchases and intake.
The NLS content, based on social cognitive theory (24),
provided information, behavioral strategies, and incentives to
guide users to set and evaluate a series of goals regarding spe-
cific nutrition-related behaviors. Social cognitive theory sug-
gests such planning and evaluation of goals can change behavior
directly, but also by increasing participants’ self-efficacy and
subsequent outcome expectations—changes that would also
lead to behavior change. By modeling and testing the relations
among the intervention, the nutritional outcomes, and the theo-
retical constructs on which the intervention was based, we dem-
onstrated not only that the NLS treatment was successful, but
also how the treatment was successful. The NLS increased spe-
cific areas of participants’ self-efficacy, physical outcome ex-
pectations, and social outcome expectations: self-efficacy for
These results add to a growing body of evidence suggesting
that changes in self-efficacy mediate the effects of a wide range
of health-related interventions (see chapter 7 of Bandura’s Self
Efficacy: The Exercise of Control [24] for a comprehensive re-
view). In addition, these results provide evidence as to how out-
come expectations operate within a health-related intervention.
Recent studies have demonstrated that outcome expectations
can contribute to health behaviors such as exercise and activity
(39–42), smoking cessation (43,44), nutrition behavior
(30,45,46), and condom use (47). Few studies, however, have in-
vestigated how outcome expectations operate within appropri-
ately ordered, social cognitive causal models (i.e., models in
which self-efficacy precedes other social cognitive variables)
(24). By modeling outcome expectations as a mediator of the ef-
fects of self-efficacy and demographic variables on nutrition,
Anderson, Winett, and Wojcik (30) demonstrated how and to
what extent outcome expectations can contribute to nutrition be-
havior among food shoppers. Physical outcome expectations,
and possibly social and self-evaluative outcome expectations,
influenced fat, fiber, and fruit and vegetable levels directly, but
they also appeared to mediate the effects of household SES,
family configuration, and self-efficacy on nutrition behavior
(30). Although studies examining predictors of and attempting
to change health-related outcome expectations (48–50) are

98
Anderson et al.
Annals of Behavioral Medicine
emerging in the literature, our study goes one step further by
demonstrating how changes in outcome expectations can con-
tribute to improvements in nutrition.
onstrated how researchers might control for the measurement
error associated with dietary data.
The challenge for future applications of the NLS and simi-
lar stand-alone, computer-based programs will be to expand ac-
cess to programs to larger groups of individuals looking for help
in improving their nutrition-related behaviors. In addition, fu-
ture applications of computer-based self-administered programs
will want to explore how and if programs situated in other set-
tings and social contexts can be more successful in reaching
people ready and less ready for change. Self-efficacy for change
is increased by supportive social environments as well as by
comfortable and familiar settings (24). Taking the NLS out of
the supermarket and moving it to familiar, socially supportive
settings may alone be enough to attract users with lower levels
of self-efficacy. As larger proportions of people from diverse
ethnic and SES backgrounds have Internet access at home (51),
an Internet version of the NLS would immediately expand ac-
cess to larger groups of users. An Internet version could also be
tailored to be incorporated into more comprehensive dietary or
other health behavior change programs delivered in a wide vari-
ety of supportive settings (52).
These results suggest nutrition interventions targeting effi-
cacy beliefs and expected outcomes (yielded from an investiga-
tion and understanding of the domains and situations relevant to
specific nutritional outcomes and intervention populations) are
more likely to succeed. Previous research on this sample (30),
for example, suggested changes in self-efficacy for decreasing
fat in meals and self-efficacy for decreasing fat in snacks would
improve fat, fiber, and fruit and vegetable levels among food
shoppers. In addition, although the NLS treatment did not influ-
ence self-efficacy for decreasing fat in snacks, the effects of
NLS treatment on self-efficacy for decreasing fat in meals
clearly led to improved nutrition. Similarly, initial findings with
the sample suggested that changes in expectations that healthier
foods would be satisfying would improve nutrition overall and
that changes in expectations that families would accept healthier
foods would lead to higher levels of fruit and vegetable servings
(30). The NLS treatment did influence expectations for appetite
satisfaction, and these changes appeared to be especially impor-
tant in maintaining dietary changes.
Although unintentional, the NLS recruitment strategies did
not result in substantial numbers of participants with low nutri-
tion-related self-efficacy or outcome expectations. Such partici-
pants arguably might have had “worse” initial nutrition-related be-
havior than the study group but also arguably might have needed
more detailed rationales and information and preliminary guided
mastery experiences prior to goal setting and evaluation activities
than provided by the very brief (5–6 min) NLS content (24).
By using multiple, self-administered dietary instruments,
this study demonstrated a feasible, appropriate approach for
evaluating less intensive individual-, group-, or commu-
nity-based interventions (26,27). Although there is a growing
consensus in dietary intervention research that such multiple
measures of nutrition-related behavior may be more effective in
collecting accurate dietary data (26,27,28), there are few recom-
mendations regarding how to retain the multiple-measure nature
of the data during analysis. Some authors have recommended
averaging nutrient values from multiple measures (26); by using
a self-report measure and an objective measure of nutrition to
build such a composite measure we demonstrated how this ap-
proach can be used to detect change associated with a nutrition
intervention (28). Other authors have suggested analyzing mul-
tiple measures separately to determine to what extent each is
sensitive to change (27)—an approach demonstrated in our FFQ
and food receipt ANCOVA tests of treatment effects. Still other
authors have recommended allowing more costly or cumber-
some objective measures to provide the aura of verification to
participants’ self-reported behavior in the hope of improving
self-report accuracy (28). Latent variable structural analysis al-
lows researchers to utilize data from multiple measures, in effect
defining nutrition-related variables as the underlying factors
that explain the pattern of correlations among multiple mea-
sures. By using SEM to explore the effects of treatment (and the
other variables) on the underlying nutrition variables, we dem-
REFERENCES
(1) Willet WC, Trichopoulos D: Nutrition and cancer: A summary
of evidence. Cancer Causes and Control. 1996, 7:178–180.
(2) U.S. Department of Health and Human Services: Healthy Peo-
ple 2000: National Health Promotion and Disease Prevention
Objectives, DHHS Publication No. (PHS) 91–50212. Washing-
ton, DC: U.S. Government Printing Office, 1990.
(3) U.S. Department of Health and Human Services: Healthy Peo-
ple 2010, Conference Edition, Volumes I and II. Washington,
DC: U.S. Department of Health and Human Services, 2000. Re-
trieved July 24, 2000 from the World Wide Web:
http://web.health.gov/healthypeople/
(4) Hill JO, Peters JC: Environmental contributions to the obesity
epidemic. Science. 1998, 280:1371–1374.
(5) Must A, Spandano J, Coakley EH, et al.: The disease burden as-
sociated with overweight and obesity. Journal of the American
Medical Association. 1999, 282:1523–1529.
(6) Serdula MK, Mokdad AH, Williamson DF, et al.: Prevalence of
attempting weight loss and strategies for controlling weight.
Journal of the American Medical Association. 1999,
282:1353–1358.
(7) Jones DA, Ainsworth BE, Croft JB, et al.: Moderate lei-
sure-time activity: Who is meeting the public health recommen-
dations? Archives of Family Medicine. 1998, 7:285–289.
(8) Koplan JP, Dietz WH: Caloric imbalance and public health pol-
icy. Journal of the American Medical Association. 1999,
282:1579–1581.
(9) Brug J, Glanz K, Van Assema P, Kok G, Van Breuklen GJP: The
impact of computer-tailored feedback and iterative feedback on
fat, fruit and vegetable intake. Health Education and Behavior.
1998, 25:517–531.
(10) Campbell MK, DeVillis BM, Strecher VJ, et al.: Improving di-
etary behavior: The effectiveness of tailored messages in pri-
mary care settings. American Journal of Public Health. 1994,
84: 783–787.
(11) Clark NM, Ghandour G, Houston Miller N, et al.: Development
and evaluation of a computer-based system for dietary manage-

Volume 23, Number 2, 2001
Computerized Nutrition Intervention
99
ment of hyperlipidemia. Journal of the American Dietetic Asso-
ciation. 1997, 97:146–150.
lation: A critical review. American Journal of Preventive
Medicine. 1999, 17:211–229.
(12) Glasgow RE, LaChange PA, Toobert DJ, et al.: Long term effect
and costs of brief behavioral dietary intervention for patients
with diabetes delivered from the medical office. Patient Educa-
tion and Counseling. 1997, 32:175–184.
(29) Walberg-Rankin J, Winett RA, Anderson ES, et al.: Food pur-
chase patterns at the grocery store and their relationship to fam-
ily characteristics. Journal of Nutrition Education. 1998,
30:81–88.
(13) Lowensteyn I, Joseph L, Levinton C, et al.: Can computerized
risk profiles help patients improve their coronary risk? The re-
sults of the Coronary Health Assessment Study (CHAS). Pre-
ventive Medicine. 1998, 27:730–737.
(14) Marcus BH, Bock BC, Pinto BM, et al.: Efficacy of an individu-
alized, motivationally-tailored physical activity intervention.
Annals of Behavioral Medicine. 1998, 20:174–180.
(15) Skinner CS, Siegfried JC, Kegler MC: The potential of comput-
ers in patient education. Patient Education and Counseling.
1993, 22:27–34.
(16) Thomas R, Cahill J, Santilli L: Using an interactive computer
game to increase skill and self-efficacy regarding safer sex ne-
gotiation: Field test results. Health Education and Behavior.
1997, 24:71–86.
(30) Anderson ES, Winett RA, Wojcik JR: Social cognitive determi-
nants of nutrition behavior: An exploratory analysis using
SEM. Health Psychology. 2000, 19:479–486.
(31) Block Dietary Data Systems: Block 95 Food Frequency Ques-
tionnaire. Berkley, CA: Block Dietary Data Systems, 1995.
(32) Pedhazur EJ, Schmelkin LP: Measurement, Design, and Analy-
sis: An Integrated Approach. Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc., 1991.
(33) Reiss AJ: Occupations and social status / Albert J. Reiss, Jr.,
with Otis Dudley Duncan, Paul K. Hatt, Cecil C. North. New
York: Free Press of Glencoe, 1961.
(34) Byron RM, Kenny DA: The moderator mediator variable dis-
tinction in social psychological research: Conceptual, strategic,
and statistical considerations. Journal of Personality and Social
Psychology. 1986, 51:1173–1182.
(17) Winzelberg AJ: The analysis of electronic support group for in-
dividuals with eating disorders. Computers in Human Behavior.
1997, 13:393–407.
(35) SPSS, Inc.: SPSS (Version 10.0) [Computer software]. Chi-
cago: SPSS, Inc., 1999.
(18) Annesi JJ: Effects of computer feedback on adherence to exer-
cise. Perceptual and Motor Skills. 1998, 87:723–730.
(19) Winett RA, Roodman AA, Winett SG, et al.: The effect of the
Eat4Life Internet-based health behavior program on high
school girls’nutrition and exercise practices. Journal of Gender,
Culture, and Health. 1999, 4:239–254.
(20) Robinson TN: Community health behavior change through
computer network health promotion: Preliminary findings from
Stanford Health-Net. Computer Methods and Programs in
Biomedicine. 1989, 30:137–144.
(21) Winett RA, Anderson ES, Bickley PG, et al.: Nutrition for a
Lifetime System: A multimedia system for altering supermar-
ket shoppers’ purchases to meet nutritional guidelines. Com-
puters in Human Behavior. 1997, 13:371–382.
(22) Winett RA, Moore JF, Wagner JA, et al.: Altering shoppers’su-
permarket purchases to meet nutrition guidelines: Interactive
information systems. Journal of Applied Behavior Analysis.
1991, 24:95–105.
(36) Jöreskog KG, Sörbom D: LISREL (Version 8.30) [Computer
software]. Chicago: Scientific Software International, Inc., 1999.
(37) Schumacker RE, Lomax RG: A Beginner’s Guide to Structural
Equation Modeling. Mahwah, NJ: Lawrence Erlbaum Associ-
ates, Inc., 1996.
(38) Campbell MK, Denmark-Wahnefried W, Symons M, et al.:
Fruit and vegetable consumption and prevention of cancer: The
Black churches unified for better health project. American
Journal of Public Health. 1999, 89:1390–1396.
(39) Resnick B, Palmer MH, Jenkins JS, Spellbring AM: Path analy-
sis of efficacy expectations and exercise behaviour in older
adults. Journal of Advanced Nursing. 2000, 31:1309–1315.
(40) Clark DO, Nothwehr F: Exercise self-efficacy and its correlates
among socioeconomically disadvantaged older adults. Health
Education and Behavior. 1999, 26:533–546.
(41) Hayes LM, Clark DO: Correlates of physical activity in a sam-
ple of older adults with type 2 diabetes. Diabetes Care. 1999,
22:706–712.
(23) Robinson TN, Patrick K, Eng TR, Gustafson D: An evi-
dence-based approach to interactive health communication. A
challenge to medicine in the information age. The Journal of the
American Medical Association. 1998, 280:1264–1269.
(24) Bandura A: Self Efficacy: The Exercise of Control. New York:
Freeman, 1997.
(25) Anderson ES, Winett RA, Bickley PG, et al.: The effects of a
multimedia system in supermarkets to alter shoppers’food pur-
chases: Nutritional outcomes and caveats. Journal of Health
Psychology. 1997, 2:221–232.
(42) Sorenson M: Maintenance of exercise behavior for individuals
at risk for cardiovascular disease. Perceptual and Motor Skills.
1997, 85:867–880.
(43) Pavone RM, Burnett KF, LaPerriere A, Perna FM: Social cogni-
tive and physical health determinants of exercise adherence for
HIV–1 seropositive, early symptomatic men and women. Inter-
national Journal of Behavioral Medicine. 1999, 5:245–258.
(44) Kuppens R, Eriksen MP, Adriaanse HP, et al.: Determinants of
fat and fiber consumption in American rural energy workers.
Preventive Medicine. 1996, 25:212–217.
(26) U.S. Department of Health and Human Services: Consensus
Workshop on Dietary Assessment: Nutrition Monitoring and
Tracking the Year 2000 Objectives. Hyattesville, MD: DHHS,
Public Health Service, Centers for Disease Control and Preven-
tion, National Centers for Health Statistics, 1994.
(27) Kristal AR, Beresford AA, Lazovich DA: Assessing change in
the diet-intervention research. American Journal of Clinical
Nutrition. 1994, 59(Suppl.):185S–189S.
(45) Resnicow K, Davis-Hearn M, Smith M, et al.: Social-cognitive
predictors of fruit and vegetable intake in children. Health Psy-
chology. 1997, 16:272–276.
(46) Sieving R, Resnick MD, Bearinger L, et al.: Cognitive and be-
havioral predictors of sexually transmitted disease risk behav-
ior among sexually active adolescents. Archives of Pediatric
Adolescent Medicine. 1997, 15:243–251.
(47) Thompson NJ, Potter JS, Sanderson CA, Maibach EW: The re-
lationship of sexual abuse and HIV risk behaviors among het-
erosexual adult female STD patients. Child Abuse and Neglect.
1997, 21:149–156.
(28) Newell SA, Girgis A, Sanson-Fisher RW, Savolainen NJ: The
accuracy of self-reported health behaviors and risk factors re-
lated to cancer and cardiovascular disease in the general popu-

100
Anderson et al.
Annals of Behavioral Medicine
(48) Domel SB, Baranowski T, Davis HC, et al.: A measure of stages
of change in fruits and vegetable consumption among fourth-
and fifth-grade school children: Reliability and validity. Jour-
nal of American College of Nutrition. 1996, 15:56–64.
(49) Clark DO, Patrick DL, Grembowski D, Durham ML:
Socioeconomic status and exercise self-efficacy in late life.
Journal of Behavioral Medicine. 1995, 18:355–376.
Dutch field experiment. Health Education Research. 1998,
13:197–206.
(51) Wojcik JR, Winett RA, Harris BE: Computer and Internet ac-
cess in two rural majority churches and one small-city Afri-
can-American church in Southwest Virginia. Society of Behav-
ioral Medicine Annual Meeting. Nashville, TN: 2000.
(52) Winett RA, Anderson ES, Whiteley JA, et al.: Church-based
health behavior programs: Using social cognitive theory to for-
mulate interventions for at-risk populations. Applied and Pre-
ventive Psychology. 1999, 8:141–157.
(50) Dijkstra A, DeVries H, Roijackers J: Computerized tailored
feedback to change cognitive determinants of smoking: A