Chapter 9 – Producing Data: Experiments
Observation Vs Experiment
• Observational study – observes individuals and measure variables of interest but does not
attempt to influence the responses. Purpose is to describe some group/situation.
• Experiment – imposes some treatment on individual in order to observe responses. Study
whether treatment causes a change in the response.
• “when our goal is to understand cause and effect, only experiments are the only source of fully
• Example 9.1) Drink a little but not a lot. People who drink moderate amount of alcohol have
less heart disease than people who don't drink or drink heavily. But there is a lurking variable.
Diet and lifestyle can affect whether or not they would be prone to heart disease or not.
◦ Drinking habits (explanatory variable), but is this the only reason what causes heart disease
(response variable). Diet and lifestyle (lurking variable) may also be at work.And it's hard
to distinguish effects of drinking habits from effects of overall diet/lifestyle
◦ two variables (explanatory or lurking variables) are confounded when their effects on a
response variable cannot be distinguished
▪ “observational studies of effect of one variable on another often fail b/c explanatory
variable is confounded w/ lurking variables
Subjects, Factors, Treatments
• factors – are explanatory variables in an experiments
• Individuals – studied in experiments are often called subjects, especially people
• Treatment – any specific experimental condition applied to subjects. If experiment has more
than one factor, treatment is a combination of specific values of each factor
◦ assigning subjects to treatments, we can avoid confounding
◦ we can study combined effects of several factors simultaneously
Example 9.3 Effects of TV advertising
Effects of repeated exposure to an advertising message? Might depend on both length of ad and how
often it is repeated. Undergraduate students are the subjects. Some saw 30s of it, 90 s of it shown either
1,3,5 times during program.
2 factors: length of commercial w/ two values and repetitions with three values. 6 Combinations of 1
value of each factor form 6 treatments.Attitude toward the camera, recall of ad and intention to
purchase were the response variables.
How to Experiment Badly
• In controlled environment (like a laboratory), simple designs often work well. Outside the lab,
uncontrolled experiments often yield worthless results because of confounding w/ lurking
Randomized Comparative Experiments
• Remedy for confounding is do a comparative experiment and then do random selection so
subjects can't choose
◦ control group • Randomized comparative experiment – uses both comparison of two or more treatments and
random assignment of subjects
◦ designed to give good evidence th