Observational Studies

• Researchers only observe and measure variables

• No treatments are imposed

• Provide evidence of association but not causation

Experiments

• Treatments are assigned to subjects

• Evidence of causation can only be given if there is proof that the only explanation for results is random choice

Natural Experiment

Smallest collection of units that treatment is applied to.

Subjects

When units are human beings they can be referred to as subjects.

Explanatory Variables

• What the experimenter is purposefully changing (factors)

• Each factor has 2 or more levels

• Independent variable (what is changing)

Response Variables

• Outcomes of experiment after treatment is administered (measured outcome)

• Dependent variable (what is being measured/observed)

Control Variable

• The element that is not changed

• Serves as a standard for comparison

• Constant (what should not be allowed to change)

Control Group

• Collection of experimental units not given or inactive treatment

• Without it, researchers wouldn’t know if the difference in treatment is what caused the change in the response variables

Factor

• Controlled independent variable

• General type of category of treatments

Levels

Different values of a factor

Treatment

The specific condition applied to the individual

Completely Randomised Design

• Treatments are assigned to people at random

• Helps to balance out confounding variables so differences can be attributed to treatments

Blocks

• Create blocks (groups) of people and assign each member to treatments

• Blocks are organized like strata (similar within and different between)

• Creates homogeneous units which reduces variance

Randomized Complete Block Design: first separated into blocks, then assign treatments

Matched Pairs Design: two treatments, grouped into pairs based on some blocking variable

Confounding Variable

• Related to the independent variable

• Might influence response variable (creates false perception between variables)

• Randomization helps (“evens out”)

Blind

All subjects in both groups cannot know which group they are in

Double Blind: Neither the subjects nor the researchers knows what treatment the subject is receiving (ensures equal treatment)

Single Blind: One party knows, other doesn’t

Statistical Significance

• When observed effect is unlikely the result of chance or random allocation

• The differences need to be real

Causation Requirements

• Outcome is proceeded on time

• Strong statistical relationship between predictions and outcome

• All causal factors have been accounted for

Inferences

Population: subjects must be randomly selected

Cause and Effect: treatment must be randomly assigned

Lack of realism will prevent viable data.

Good Experiments

1. Comparison

Only one variable needs to be tested at a time

2. Controls

Other potential variables should stay the same for everyone

3. Random Assignment

Subjects, treatments, and control groups

4. Replication

Enough subjects are used and there is room for replication