Many employers offer worksite wellness programs to their employees. These programs can take
different forms and focus on different health outcomes and health-related behaviors, depending
upon the needs of the workforce at a particular employer. A large telecommunications company
located in the upper Midwest has a worksite wellness program with different modules: a smoking
session module, a stress management module, and so on. Employees are free to enroll in any or
all of the modules they choose. Some do not enroll in any, others enroll in just one, and some
enroll in two or more. One of the modules focuses on weight management. This module
consists of an integrated web- and mobile device-based interface that offers tailored goal-setting
and self-monitoring facilities. Through the website, participants enter information about
themselves and work through a structured goal selection process. Once they have identified
goals, they use the mobile app to monitor their behaviors (physical activity and diet) as well as
their weight. The app also provides tailored feedback and motivation-boosting messages, also
tailored to the individual employee.
Up to the present, the weight management module of this worksite wellness program has never
been formally evaluated. No one knows for certain, therefore, whether or not that module
actually helps people manage their weight successfully. Now, however, for various reasons, the
company finds out: Does participating in the weight management module of the worksite
wellness program actually help employees either lose weight or, at least, avoid putting on more
weight? The company hires an outside evaluator, who argues that the strongest evaluation
design would involve randomizing some employees to receive (or be offered) the intervention,
and others not to receive (or be offered) it. Because the program has been up and running for
some time, however, this turns out not to be feasible. The company and outside evaluator will
have to use an observational study, rather than a randomized trial, to try to measure the effect of
participation in the weight management module on change in weight. That is, they will need to
compare the change in weight of employees who enroll in the weight management module to the
change in weight of employees who did not enroll.
The evaluator realizes, however, that employees who enroll be systematically different from
employees who do not enroll in ways that may render the simple comparison of the groups
problematic. Perhaps heavier people are more likely to enroll than lighter people. Perhaps
people who are more satisfied with their own bodies are less likely to enroll. Perhaps people
with higher self-efficacy in the domains of physical activity and diet are more likely to enroll.
Perhaps age, gender, or type of job at the company influence whether or not employees enroll in
the weight management module. If so, and if these variables also exert an independent influence
on weight change, they might function as confounders in the relationship between enrollment
and weight change. The evaluator is able to convince the company that, if they cannot do a
randomized trial, they can at least measure and control for potential confounders.
And that is what they do. The evaluator, with assistance from the employers, conducts a survey
of a random sample of employees. She collects information on each employees weight, age,
gender, and job category. The questionnaire also includes multi-item measures of body
satisfaction (four items on a 1-to-5 scale), self-efficacy for physical activity (five items on a 1-to-
4 scale), and self-efficacy for dietary behavior (five items on a 1-to-4 scale). Six months after
the original data collection, she contacts the same employees again and obtains updated
information on their weights. She computes each participants’ weight change but subtracting her
or his baseline weight from her or his weight at the six-month follow-up assessment. Finally,
from information collected automatically through the web-based component of the company’s
weight management module, the evaluator is able to determine whether each employee was or
was not enrolled in the weight management module at any point between the baseline and sixmonth follow-up data collections.
Data from this evaluation are available in the dataset WorkWell.sav. That dataset includes
demographic variables, baseline weight, the scale scores, enrollment status, and weight change.
The meaning of all variables should be clear from the codebook, WorkWellCodebook.doc. Your
task now is to analyze that dataset. Ultimately your goal is to determine whether or not
enrollment in the weight management module has any influence on weight change over the sixmonth period. To do this, however, you must take into account the possibility of confounding by
baseline demographic and/or psychosocial factors. Additionally, the employer wishes to know if
the effect (if any) of the weight management module is different for male versus female
employees, or differs as a function of baseline body satisfaction. They hypothesize that the
program may be more effective for female employees than for male employees, and for
employees with low levels of baseline body satisfaction than for those who were more satisfied
with their bodies at baseline. Please use SPSS, therefore, to carry out the following analyses.
1. Create a table of frequencies and/or descriptive statistics for all variables in the dataset (other
than ID).
2. The first condition for a variable to be a confounder is that it must be associated with the
independent variable of interest. Therefore, to determine whether WGTPRE, AGE, BODYSAT,
SEPHYSACT, SEDIET, MALE, and/or JOBCAT are associated with ENROLL, run seven
logistic regression models, each predicting ENROLL, and each containing one of the seven
potential confounders.
3. As a further test of whether WGTPRE, AGE, BODYSAT, SEPHYSACT, SEDIET, MALE,
and/or JOBCAT are associated with ENROLL, run a single multivariable logistic regression
model using all seven of these potential confounders simultaneously to predict ENROLL.
4. The second condition for a variable to be a confounder is that it must be associated with the
dependent variable. To determine whether WGTPRE, AGE, BODYSAT, SEPHYSACT,
SEDIET, MALE, and/or JOBCAT are associated with WGTCHANGE, run seven linear
regression models, each predicting WGTCHANGE, and each containing one of the seven
potential confounders.
5. As a further test of whether WGTPRE, AGE, BODYSAT, SEPHYSACT, SEDIET, MALE,
and/or JOBCAT are associated with WGTCHANGE, run a single multivariable linear regression
model using all seven of these potential confounders simultaneously to predict WGTCHANGE.
6. Next, as a preliminary but naïve test of whether the weight management module has any
effect, compare the mean weight change for employees who enrolled in that module to the mean
weight change for employees who did not enroll in it. Do not control for any other variables
7. As a more rigorous test of whether the weight management module has any effect, compare
the mean weight change for employees who enrolled in that module to the mean weight change
for employees who did not enroll in it while controlling for baseline differences in any or all of
the seven potential confounding variables: WGTPRE, AGE, BODYSAT, SEPHYSACT,
SEDIET, MALE, and JOBCAT.
8. Next, run two linear regression models to see if the effect of enrollment in the weight
management module appears to vary by gender. Do not include any control variables (other than
gender) in the first model. Include all of the control variables in the second. Note that each
model should include a gender-by-enrollment interaction variable.
9. Similarly, run two linear regression models to see if they effect of enrollment in the weight
management model appears to vary according to baseline body satisfaction. Do not include any
control variables (other than body satisfaction) in the first model. Include all of the control
variables in the second. Note that these models should include a body-satisfaction-by-enrollment
interaction variable, and remember the importance of mean-centering continuous variable before
forming interaction terms.
Prepare a written report summarizing the results of all of these analyses. That report should
integrate narrative text with results in tabular form. It should include six tables: one containing
results from Part 1; the second containing results from Parts 2 and 3; the third containing results
for Parts 4 and 5; the fourth containing results from parts 6 and 7; the fifth containing results
from Part 8; and the sixth containing results from Part 9. Make sure to relate the results to the
following research questions:
• Which if any of the seven potential confounders meet the first criterion for confounding,
i.e., being associated with the focal independent variable?
• Which if any of the seven potential confounders meet the second criterion for
confounding, i.e., being associated with the dependent variable?
• What is the estimated effect of enrollment in the weight management program on weight
change over the six month follow up period?
• How, if at all, does the effect of enrollment on weight change vary by gender?
• How, if at all, does the effect of enrollment on weight change vary by baseline body
satisfaction?
There are two deliverables for this assignment:
(1) An SPSS syntax file that accomplishes Tasks 1 through 9. The file should contain all of the
commands necessary for accomplishing these tasks, in the specified order, and should not
contain any extraneous commands.
(2) Your written report summarizing the results. Please do not copy-and-paste SPSS output into
the report; rather, please create your own summary tables.
