Observation Studies
There are two types of Data Collection
- Observation Studies and Statistical Experiments
In ‘‘Observation Studies’’ we observe existing characteristics of a subset of individuals in a population
- typically done via surveys, by following smb, etc
- this method doesn’t directly interfere with how the data appear (in contrast to Statistical Experiments)
the goal is to
- draw conclusions about the population or
- find differences between 2 or more groups or
- find out about the relationships between variables
Types
- Prospective Study
- collect the data as an event unfolds
- Retrospective Study
- use the data of some event that already took place
Finding Relationships
Types of variables:
- outcome - the variables of our interest
- explanatory - the variables that are used to analyze and explain the outcome
Types of Relationships
The relationships between the explanatory variable and the outcome
- '’independent’’: there is no association between the variables
- '’association’’: the variables are dependent, but it’s not clear what kind of relationship there is
- '’causes’’: changes in the explanatory variables case the outcome to change
- '’reverse causation’’: changes in outcome cause the explanatory variable to change
- '’coincidence’’: just pure chance
- '’common cause’’: some other variable causes both the explanatory variables and the outcome to change (see also Confounding Variables)
Correlation and Causation
- with this type of studies it is possible to find association relationship between the variables
- but it’s not possible to show the causation here - need to run a controlled Statistical Experiment for that
- beware of Confounding Variables
Example
- Suppose we run a sunscreen study and collected some data
- We saw that the more sunscreen is used, the more chances to have skin cancer
- does sunscreen causes the cancer?
- cannot say it here because the study is observational - we didn’t run a controlled Statistical Experiment to make sure there are no other variables that might have caused it
- e.g. in this case we don’t see the exposure to sun - it’s correlated with both sunscreen and cancer variables
- this is a Confounding Variable that is likely to have caused the effect