SDA: Survey Analysis Flashcards
What are Causal Models?
Models highlighting bi-variate or multi-variate relationships
Table must show more than correlation because correlation does not imply causation!
What is a bi-variate relationship/model?
‘x’ causes or has an effect on ‘y’ (two variables)
What is a multi-variate relationship/model?
Independent causes of a response e.g. age, gender, social class and smoking all influence illness rates (more than one independent variable)
What 3 conditions must be met for causality models?
- Covariance
- Temporal Precedence
- Production
What is Covariance?
If ‘x’ causes ‘y’, variations in ‘x’ should lead to variations in ‘y’
What is Temporal Precedence?
If ‘x’ causes ‘y’, changes in ‘x’ should occur before corresponding changes in ‘y’
What is Production?
Changes in ‘x’ should really produce changes in ‘y’
V. difficult to prove e.g. wet periods follow dry periods, but dry periods are not PRODUCED by wet periods
Why do we undertake ‘control by analysis’/sub-group analysis?
It is very difficult to prove the third condition of causality models of Production in non-experimental research, therefore sub-group analysis allows looking to see if relationships hold for different subgroups of a population
What do the following show in a Contingency Table:
- Columns
- Rows
- Summing across the rows
- Summing down the columns
- Dependent variables (y)
- Independent variables (x)
- Row marginal
- Column marginal
What is the calculation for turning counts into row %s to allow for comparisons?
Row % = (cell count/row marginal) * 100
What is a good graphical representation used for contingency tables?
Clustered bar charts
What is a perfect positive relationship?
E.g. ALL people WITH bronchitis live in high pollution areas
What is a perfect negative relationship?
E.g. ALL people WITHOUT bronchitis live in high pollution areas
What does the Chi-squared test show?
Shows whether there is a relationship between recorded nominal data, but nothing else
What does Cramer’s V test show?
The size of the relationship
For the Chi-squared and Cramer’s V test, both +ve and -ve relationships give positive values, so how must the sign of the relationship be found?
Through inspection of contingency tables
If the value of the chi-squared and cramer’s v test is 0, what does this say about the relationship?
There is no relationship
What is the issue with the chi-squared test?
The value depends on the sample size, i.e. a big value means there was a big sample size, not that there is a very strong relationship
What is confounding?
A type of bias arising when the relationship between two things is influenced by a third thing (a confounder)
Can be caused by selection bias when the individuals or groups under study are different
Leads to SPURIOUS relationships
Example: carrying a lighter and smoking 20 day are both related to lung cancer prevalence rates. But carrying a lighter is for the cigarettes, not a cause of lung cancer
APPARENT relationships between two variables that are in fact spurious, is caused by what?
A ‘hidden’ extraneous confounding variable
Why is checking for confounders important?
Because any variables that are deemed to be confounders should not be included in multi-variate analysis