Session 6 - Comparison of Categorical Variables within two groups?

Question 1

What is the null hypothesis for Person’s Chi-squared test?

Answer 1

The categorical variables are uncorrelated

Question 2

What is the alternate hypothesis?

Answer 2

At least two of the categorical variables are correlated

Question 3

What are some key assumptions of the Pearson’s chi-squared test?

Answer 3

Independent subjects
Un-ordered categories

For contingency tables:
-Assumes that the marginal totals are fixed
-Distribution of test statistic is approximately chi-squared
-Test statistic is the difference between observed and expected cell frequencies

Question 4

What are the problems with chi-squared test?

Answer 4

Small cell counts: the chi-squared approximation breaks down with small numbers. How small?

Less than 5 in a cell (Fisher)
Less than 10 in a cell (Cochran)
n less than 40 - total across all cells (Cochran)

-Fixed margin total

Question 5

Is the Yates correction universally accepted?

Why is it done?

What’s the message to do it with a chi-square test?

Answer 5

No it isn’t

It is done as correction for low (expected, not observed) cell counts.

Page 31 - session 6 lecture

R will do it by default with any column sample less than 5

To preven this, correct=FALSE

Question 6

What do we do when we are considering “as or more extreme”?

Answer 6

Fisher’s exact or Fisher-Irwin test

Question 7

What test do we use if a priori we expect there to be agreement between the binary outcomes?

Give an example

Answer 7

McNemar’s test

For example, comparing the outcomes of two supposedly inter-changeable tests

Question 8

What tests whether the proportion follows a trend?

Answer 8

Chi-square test for trend