Research conducted by: Alia J. Crum and Ellen J. Langer

Case study prepared by: Robert F. Houser and Alyssa Koomas

Overview
The "placebo effect" is an effect that cannot be attributed to a drug or remedy, but rather to a change in a person's mind-set or perception. The placebo effect is widely accepted in clinical trials and its effects may shock you. For instance, one study found that subjects developed real rashes after being exposed to fake poison ivy (Blakeslee, 1998)! This study examined the placebo effect with relation to physical activity and health. Could becoming aware of how much you exercise result in weight loss even if you didn't make any changes to your diet or exercise routine?

The subjects were 84 female maids of ages 19 to 65 years at seven hotels. They were told the purpose of the study was to improve the health and happiness of hotel maids. According to the authors, "[e]ach of seven hotels was randomly assigned to one of two conditions: informed or control" (page 166). "Four hotels were assigned to the informed condition, and three were assigned to the control condition" (pages 166-167). Each subject filled out a questionnaire asking about her perceived amount of exercise during and outside of work. Physiological measurements were taken for weight, body mass index, body-fat percentage, waist-to-hip ratio, and blood pressure. The maids in the informed condition were then given an oral presentation and handouts explaining how their work as hotel maids is good exercise, so good in fact that it meets or exceeds the Surgeon General's recommendations for physical activity. The maids in the control condition were not given this information. After four weeks, the researchers readministered the questionnaire and took follow-up physiological measurements.

Questions to Answer Does the placebo effect play a role in the health benefits of exercise? If we alter a person's perception of the exercise she performs, does it result in weight loss?

Design Issues
Instead of assigning individual maids randomly to either the informed or control condition, all of the maids in the same hotel were assigned to the same condition. This was done in an effort to prevent information contamination. This type of study design is known as a "cluster randomized trial," and calls for advanced statistical practices that we will not worry about in this case study.

Simple random sampling with a sufficient number of subjects randomly assigned to intervention and control groups ideally leads to intervention and control groups that are similar with respect to many demographic characteristics. Simple random sampling of individuals and random assignment of individuals to conditions were not used in this study. The authors of this study pointed out that "[s]ubjects in the informed group were significantly younger than subjects in the control group." Consequently, they attempted to control for age differences in their statistical analysis.

The questionnaire asked about self-reported levels of exercise and dietary intake. Future research should use more rigorous methods to assess physical activity and diet.

Descriptions of Variables

VARIABLE

DESCRIPTION

cond

Condition: Either Informed or Control

age

Age in years

ex1

Perceived amount of exercise at Time 1 (On a scale from 0 to 10 with 0 = “none” and 10 = “a great deal”)

ex2

Perceived amount of exercise at Time 2 (On a scale from 0 to 10 with 0 = “none” and 10 = “a great deal”)

wt1

Weight in pounds at Time 1

wt2

Weight in pounds at Time 2

aex

Change score for exercise equal to the perceived amount of exercise at Time 2 minus the perceived amount of exercise at Time 1

awt

Weight change equal to the weight at Time 2 minus the weight at Time 1

References

Crum, A. J., Langer, E. J. (2007). Mind-set matters: Exercise and the placebo effect. Psychological Science, 18, 165-171.

Please read the original publication first, and then perform the exercises using the data set. The data set is a modified version of the original data, which yields the same results as those in the article. For all analyses, statistical significance is based on p<0.05.

Could the researchers have used a simple random sample of participants for this study? What problems might have occurred had they not used cluster sampling?

The researchers found that the informed and control groups were statistically different with regard to age. It was therefore important to control for age in the analysis, since people of different ages might be affected differently by the intervention. Perform an independent samples t test comparing the mean age of subjects by condition (informed versus control).

What are the mean age and standard deviation of the informed group?

What are the mean age and standard deviation of the control group?

Which group was older?

Is the difference statistically significant at the 0.05 level?

Is the difference statistically significant at the 0.01 level?

Which transformation of the age variable would be useful for this analysis?

Apply the transformation and perform an independent samples t test. Do you obtain statistical significance?

Check your answers with footnote 4 on page 167 of the article. (Note that p_{rep} was reported instead of the p value. For information on p_{rep}, see Wikipedia: http://en.wikipedia.org/wiki/P-rep)

Create a histogram for perceived amount of exercise at Time 1 (ex1). Do the same for perceived amount of exercise at Time 2 (ex2).

Looking at the graphs, are these variables normally distributed, positively skewed, negatively skewed, or bimodal?

Perform a frequency distribution for each of the variables ex1 and ex2. What percentage of subjects reported not doing any exercise at all at Time 1? What percentage of subjects reported not doing any exercise at all at Time 2?

How do the informed and control groups compare in terms of average perceived amount of exercise at Time 1 (ex1)? Perform an independent samples t test by condition for the variable ex1.

Are the two group means significantly different?

Would a transformation be useful for this analysis? Why or why not?

Perform a Wilcoxon rank sum test by condition for the variable ex1. Does it give the same conclusion as in part a? Explain why this nonparametric test is appropriate for this analysis.

Check your means and standard deviations with the values in Table 1.

Perform an independent samples t test comparing the average perceived amount of exercise at Time 2 (ex2) by condition.

Are the two group means significantly different?

Perform a Wilcoxon rank sum test by condition for the variable ex2. Does it give the same conclusion as in part a?

Check your means and standard deviations with the values in Table 1.

Consider the change score variable for exercise (aex). A positive value indicates that a subject's perceived amount of exercise increased from Time 1 to Time 2.

Create a histogram for the variable aex. How does its shape compare with the distributions of variables ex1 and ex2 in Exercise #3?

Perform an independent samples t test comparing the variable aex by condition.

What are the mean change scores and standard deviations for the informed and control groups?

Are the mean change scores for the two groups significantly different?

Which group had the larger increase in perceived amount of exercise?

Compare the informed and control groups in terms of weight by performing the t tests below.

Perform an independent samples t test by condition for weight at Time 1 (wt1). Did the two groups differ significantly in mean weight at Time 1?

Perform an independent samples t test by condition for weight at Time 2 (wt2). Is there a statistically significant difference in mean weight at Time 2?

Check your means and standard deviations with the values in Table 1.

Perform an independent samples t test by condition for the variable awt, which is the change in weight between Time 1 and Time 2. Is there a statistically significant difference in mean weight change between the two groups?

Which group had the greater weight loss?

The results of this study suggest that a change in mind-set in the informed group (i.e., the perception of how much exercise one was getting) led to a decrease in weight without any actual increases in physical activity at or outside of work or changes in diet. What other factors may have contributed to the change in weight? What measures could future researchers take to more rigorously investigate this phenomenon? (Hint: After you come up with your own recommendations, check the article for the authors' suggestions.)