The Analysis of Categorical Data- Fisher's Exact Test Flashcards
How do you calculate the P-value in a Fisher’s Exact Test compared to Chi squared test
Calculate the P-value DIRECTLY FROM THE DATA instead of from marginal totals equal to observed values , although not necessarily to do this when using the method
What can you deduce from the probabilities are the same from the Fisher’s Exact Test
The Null hypothesis is true
How can the P-value be found?
By adding up the possibilities that are LESS THAN OR EQUAL TO THE PROBABILITY OF THE OBSERVED TABLES
What is ‘Exact’ about the the exact test?
- Gives a P-value that is calculated on the basis of the correct probability distribution for the different tables (no need for asymptotic approximation) assuming that the null hypothesis is true
- If performed the experiment many times with the null hypothesis true, and the MARGINS ARE FIXED AT THE OBSERVED VALUES, the P<0.05 might occur in FEWER THAN 5% OF THE EXPERIMENTS. This is NOTICED IN SMALL DATA SETS where the effects of discrepencies are inescapable. This is not a problem if the EXACT P-VALUE is quoted.
Why Not Use the Exact Test all the time?
- For larger counts it would be long to process and present an algorithmic challenge
- CONFIDENCE LEVELS are WIDE and often NOT GOOD
When should you use an exact test rather than x2 (chi squared) test?
- If any EXPECTED VALUE in a 2x2 table is LESS THAN 5
For larger tables
- OVER 20% of the cells have EXPECTED VALUES BELOW 5
- ANY CELLS have EXPECTED VALUE BELOW 1 then the x2 method may be unreliable
To do a Fisher’s Test for larger values what do you need?
- AMALGAMATE CATEGORIES APPROPRIATELY so that the EXPECTED VALUES IN THE NEW TABLES ARE LARGER THAN THESE THRESHOLD
- STATISTICAL PACKAGES the exact test for rxc tables
(Advances have made exact tables on quite large tables entirely feasible but specialist advice is needed)