1. Observational Studies
Simply observing what happens
A sample survey is an observational study.
There are other observational studies that are not surveys.
For observational studies we simply observe what is happening or has happened. There is no intervention, simply observation. A sample survey is a type of observational study.

2. Tanning and Skin Cancer
The observational study involved 1,500 people.
Checked medical records to see if a person had experienced skin cancer or had never had skin cancer.
Asked all participants whether they had used tanning beds.
In this study we are observing people. We check their medical records to see if they have or have not had skin cancer. We also question the people to see if they had used tanning beds. Who? – People. What? – Have you had skin cancer? Yes/No. Have you used tanning beds? Yes/No.

3. Sodium and Blood Pressure
Enroll 100 individuals in the study.
Give them diet diaries where they record everything they eat each day for a month. From this the amount of sodium in the diet is found.
Measure their blood pressure.
In another observational study we enroll 100 individuals and give them diet diaries that they fill out for a month. At the end of the month, the participants blood pressure is taken (observed). We are simply observing what the people eat over the month and their blood pressure. Who? – People. What? Amount of sodium in the diet (mg). Blood pressure (mmHg).

4. Differences Retrospective – look at past records and historical data.
Tanning and Skin Cancer
Prospective – identify subjects and collect data as events unfold.
Sodium and Blood Pressure
The two observational studies we have just mentioned are different in that one looks back at past records and historical data (retrospective) while the other follows people and observes them into the future (prospective).

5. Experiments Explanatory variable – Factor. Treatments.
Subjects – Participants – Experimental Units.
Response variable.
Another way to collect data is through an experiment. In an experiment we will encounter special terms. In experiments we will have an explanatory variable and a response variable. Because in an experiment, someone actively changes the explanatory variable, we often refer to this explanatory variable as a factor of interest that is manipulated to take on several of the factor levels. The factor levels are often referred to as treatments. What the treatments are applied to are the experimental material: units, subjects, participants.

6. Experiments
The experimenter must actively and deliberately manipulate the factor(s) to establish the method of treatment. Experimental units are assigned at random to the treatments.

7. Sodium and Blood Pressure
20 subjects.
Factor – amount of dietary sodium.
Treatments: low sodium diet and high sodium diet.
10 subjects randomly assigned to each treatment.
Response – systolic blood pressure.

8. Experimental Principles
Control
Randomize
Replicate
within an experiment.
repeating the entire experiment.
Block

9. Control Control outside variables that may affect the response.
Have subjects of the same age, gender, general health, ethnic group.
By controlling outside variables you prevent those variables from causing variation in the response.

10. Randomization
Randomization tends to spread the effects of uncontrolled outside variables evenly across the treatment groups.
Randomization reduces the chance that an uncontrolled outside variable will bias the results.
The best way to randomize is to draw “names out of a hat.” We put the 20 participants names in the hat. Mix, draw one without replacement, put that person in treatment 1. Continue to mix, draw without replacement, and place in treatment 1 until you have 10 people in treatment 1. The remaining names in the hat go to treatment 2.
Alternatively, put 20 chips in a bag. Ten of the chips have treatment 1 on it, ten have treatment 2 on it. Have the participants come up one at a time and draw a chip from the bag without replacement. The person goes to the treatment group indicated on the chip.

11. Replication Within an experiment.
must have several experimental units in each treatment group.
can assess the natural variation in the response for units treated the same.
If asked what replication is the simple answer is having several experimental units for each treatment group.

12. “Replication” Repeating the entire experiment.
This is especially important if the subjects in an experiment are not a random sample from a population.
Are the results of the entire experiment repeatable?
The more common notion of “replication” is repeating the entire experiment with different people (experimental units).

13. Diagram Group 1 several subjects Treatment 1 random allocation Compare
Response
Subjects
Group 2
several
subjects
Treatment 2

14. Block
There may be attributes of the experimental units that can’t be controlled but may affect the response.
Group similar experimental units into blocks and then randomize the assignment of treatments within each block.
There may be attributes of experimental units that can’t be controlled. For example, gender. We could do an experiment with only males but then we could only make inferences about the effect of the treatment on males. If we include females, we could us randomization to spread the effect of gender evenly across the treatment groups but this may increase the natural variation in the experiment. Instead, use males and females as blocks and assign treatments at random within each block.

15. Blocking Math ability very high, high, average, low, and very low.
assign, at random, students from each math ability group to each treatment.
Suppose we want to see if using a calculator or not using a calculator affects the score on a statistics exam. People taking the statistics exam may have different levels of ability in math which could affect the score on the statistics exam. We could form blocks according to math ability; very high, high, average, low and very low. Within each math ability group, randomly assign some students to use a calculator while others will not.

16. More Ideas Control “treatment”. Blinding. Placebos.
Single blind and double blind.
Placebos.
A control “treatment” or group is sometimes used in an experiment. If we are looking at the effectiveness of a drug we might have a high dose and a low dose treatment. Having a no dose treatment would be a control “treatment”. Without the control “treatment” we can compare to see if the high dose is better than the low dose. With the control “treatment” we can also assess if either or both dose levels are effective at all (i.e. better than nothing).
An experiment can be blind to the participant (i.e. the participant does not know which treatment group he/she is in) or blind to the researcher (i.e. the researcher does not know which treatment each participant is getting) or blind to both (a double blind experiment). Blinding is important if knowledge of the treatment might affect the response.
A placebo is a fake or false treatment that should not have any affect on the response (this is often a sugar pill or saline injection that contains no active ingredient). Placebos are often given to control “treatment” participants so that they experience a similar protocol (taking a pill or getting an injection) as the other participants. This is again to prevent knowledge of which treatment group a participant is in.

17. Multiple Factors Factors Treatments can use calculator (yes, no)
can use a formula sheet (yes, no).
Treatments
calculator and formulas, calculator but no formulas, formulas but no calculator, no calculator and no formulas.
Often times we can manipulate more than one factor in an experiment and get a more efficient use of our participants or experimental material. The idea is to make treatments out of combinations of levels of 2 or more factors.

18. Confounding Sodium and blood pressure.
All subjects on the low sodium diet had their blood pressure measured by a registered nurse using a standard manual cuff and stethoscope.
All subjects on the high sodium diet had their blood pressure measured using an automated cuff and digital readout.
Confounding is something that we wish to avoid in an experiment. Controlling outside variables in this example the method of taking the blood pressure usually does the trick. If we cannot control then randomization can often lower the chance that a confounding variable biases the results of the experiment.