CAOT 31 Lecture Notes - Lecture 2: Random Assignment, Analysis Of Variance, Test Statistic
Design and Analysis of Experiments and Observational
Studies
Observational studies
= simply observing behaviour from given data
Retrospective studies = observational study that studies outcome in
the present by looking into historical records
Prospective studies = indentifying subjects in advance and collecting
data as events unfold
Experiments
- only way to show cause-effect relationship
- experimenter manipulates attributes of what is being studied
and observes the consequences
- attributes(=factors) are manipulated by being set to a
particular level and then allocated to individuals
- difference to observational study : manipulating factors,
random assignment (very important! )
factors = attributes of what is being studied ( e.g. interest rate,)
response = outcome that will be measured, often quantitative
treatment = combination of factor levels assigned to subject
subjects/participants = human individuals who participate in
experiment
experimental units = other individuals that participate in experiment
control group = receives control treatment („normal treatment“ )
replicates = repeated observations at each treatment
balanced = if number of replication at each treatment is the same, the
experiment is balanced
Interaction = when the combination of two factors has a different
effect than you would expect
The 4 principles of experimental design
1) control. Controlling sources of variation other than factors by
making conditions as similar as possible
2) randomize. Subjects have to be assigned to treatments at
random. That allows us to equalize the effects of unknown
variation
Document Summary
Retrospective studies = observational study that studies outcome in the present by looking into historical records. Prospective studies = indentifying subjects in advance and collecting data as events unfold. Experimenter manipulates attributes of what is being studied and observes the consequences. Attributes(=factors) are manipulated by being set to a particular level and then allocated to individuals. Interaction = when the combination of two factors has a different effect than you would expect. The 4 principles of experimental design: control. Controlling sources of variation other than factors by making conditions as similar as possible: randomize. Subjects have to be assigned to treatments at random. That allows us to equalize the effects of unknown variation: replicate. We need to estimate the variability of our measurements so we need to perform more than one experiment. Completely randomized designs: when each of the treatments is assigned to at least one subject at random. Factorial designs: an experiment with more than one manipulated factor.