Question

Many statistical methods and techniques we are looking at are intrinsically related and/or provide different views at the same underlying phenomena. This week you read about contingency tables. As we discussed, this is a way to examine possible association between two (categorical) variables. An example we have considered was association between patient's sex and remission status. We know other methods for testing for associations (not necessarily perfectly fit for the specific problem).

Can you think about setting up a linear model for our example?

Any other method?

How would you do it and what would you look for in the output?

You do not have to program, just describe what you would do.

Answer 1

To find the association between patient's sex and remission status:

(i) Linear model:Logistic Regession Model, since the dependent variable is dichotomous (binary)

(ii) Any other method: test of association, Fisher's Exact test, McNemar's test, Odds Ratio, Yate's Correction

(iii) How would you do it and what would you look for in the output:

Here:

One Dependent Variable: Remission Status: Status of whether the disease is controlled so that the patient is not as ill as he was.

Independent Variable: Sex (Male/ Female). It is dichotomous in nature.

In the Logistic Regression Analysis, the log odds of an event is estimated.

The model itself simply models probability of output in terms of input and performs statistical classification.

Logistic regression is used to predict the risk of developing a given disease based on observed characteristics of the patient.