Categorical Data Flashcards
What are the assumptions for calculating the CIs for proportions?
Independent random sample
The sample proportions are Normally distributed
r1, r2, (n1 - r1), (n2 - r2) are all >5, where
r = number of variable of interest
What are the assumptions for a z-test?
Independent random sample
The sample proportions are Normally distributed
r1, r2, (n1 - r1), (n2 - r2) are all >5, where
r = number of variable of interest
Tests of association can estimate the extent of the association. T/F?
False
What statistical test can be used to compare 2 or more proportions across 2 groups?
Chi-squared tests
How does a chi-squared test work?
The actual observed values are compared to the values we would expect to see if the Null hypothesis (that there is no relationship between the variables) is true. The further apart the observed and expected values are, the more evidence there is against the Null hypothesis.
The chi-squared test tells us about the form of the association. T/F?
False - in this case the form of the association can be revealed by inspection of the table of observed and expected values to see where the excesses and deficits are
The chi-squared test is only valid if…?
The observations are from independent subjects / experimental units
The expected values are not too small - no more than 20% of the expected values should be <5 and none of them should be <1
The chi squared test for trend is a more powerful test to use when one of the variables is ordered and the other has exactly 2 categories. T/F?
True
When is it not appropriate to use a chi square test for trend?
If the trend across the groups is not changing in a linear manner then this test. is inappropriate
When should continuity correction be used when doing a chi squared test?
When there is a 2 x 2 table of categorical data
When should Fisher’s exact test be used?
When the smallest expected values are very small (<2)
Outline how the Fisher’s exact test works?
It assumes that the row and column totals in the test are fixed and evaluates the probability of every possible 2x2 table with these totals.
It then looks at how extreme the observed table is relative to all the others.
What test can be used to compare proportions from paired samples?
McNemar’s test
Why can a chi squared test not be used ti compare proportions from paired samples?
Because the samples are not independent
If the expected values in McNemar’s test are <20, then what should be used?
Continuity correction or sign test
