Selecting the Most Appropriate Statistical Test - Nominal Data

Question 1

What statistical test would you use if the data was nominal and you were comparing frequencies or proportions of 2 groups that were independent or had independent assessments, and no cell had an expected frequency of less than 5? Would the test change if the cell had an expected frequency of less than 5?

Answer 1

• (Pearson’s) Chi-Square Test
• Yes, would use Fisher’s Exact Test instead

Question 2

What statistical test would you use if the data was nominal, there was 2 groups with paired or related data, and you wished to compare frequencies or proportions?

Answer 2

McNemar Test

Question 3

What statistical test would you use if the data was nominal and there was 3 or more groups that had independent data, you wished to compare frequencies or proportions, and there was no cell with expected count of <5? What about if the cells had an expected count of <5?

Answer 3

• Chi-Square Test of Independence
• Fisher’s Exact Test if EXPECTED cell count of <5

Question 4

If a Chi-Square or Fisher’s Exact test was statistically significant for nominal data that had 3 or more independent groups, what post-hoc test would be ran and why?

Answer 4

• Bonferroni Test of Inequality (Bonferroni correction)
  
  • to determine specificaly which groups are different.
  • Adjusts the p-value for # of comparisons being made
  • Very Conservative

** NOTE: multiple chi-square test to find the differences between groups is NEVER acceptable because it increases the risk of Type 1 Error with each additional test. (almost guaranteed after 4-5 tests) **

Question 5

Which statistical test would you use if the data was nominal, there were 3 or more groups that had paired or related data, and you wished to compare frequencies or proportions? Which test would you run if this test showed statistical significance? What if it did not?

Answer 5

• Cochran Test
  
  • same principle and assumptions as chi-square test yet mathematically factors in concept of paired, or related, data.
• Showed Statistical Significance
  
  • Bonferroni Test of Inequality
• No Statistical Significance
  
  • Report Results

Question 6

Which statistical test would you choose if the data was nominal, you did not wish to compare frequencies or proportions, but you wanted to assess for associations or predict group membership?

Answer 6

Logistic Regression

Question 7

Which statistical test would you choose if the data was nominal, you did not wish to compare frequencies or proportions, did not want to assess for associations or predict group membership, but you did want to check for correlation?

Answer 7

Contingency Coefficient

Question 8

Which statistical test would you choose if the data was nominal, you did not wish to compare frequencies or proportions, did not want to assess for associations or predict group membership, you did not want to check for correlation, but you did wish to assess for proportion to event (survival)?

Answer 8

Log Rank