Statistics:Non parametric statistics Flashcards
What form of statistics does nonparametric statistics methods work with?
The ranking of the subjects on the response variable, they do not assume a particular form of distribution
What two cases are non parametric statistics particularly useful in?
- When the data are ranks for the subjects
- When it’s inappropriate to assume normality, and when the ordinary statistical method is not robust to violations of normality assumption.
What are the null and alternative hypotheses templates for the Wilcoxon test
H0: Identical population distributions for the two groups (equal expected values.)
Ha: Different expected values for the sample mean ranks (two sided)
Ha: Higher expected value for mean sample rank for a specific rank (one sided)
What are the assumptions of the Wilcoxon test?
Independent random samples from two groups
How do you calculate the test statistic in the Wilcoxon test?
Calculate the difference of the sample mean ranks for the two groups.
How do you get the P value in the Wilcoxon test? does this differ for a two sided statistic?
For one sided Ha: P(difference between sample mean ranks at least as large as observed.) Double this for the two sided statistic.
What alternative test statistic can the Wilcoxon test use?
The sum of the ranks
If two subjects tie in their rank what happens?
They share that rank and are given the mean of the two ranks either side. i.e if they’re ranked second the mean of rank 2 and 3 are found and they are both assigned rank 2.5.
What is done when the response variable is quantitative rather than a rank?
The subjects are assigned ranks with the smallest value being ranked 1.
How do you get the proportion for better responses for a group
See all possible pairs (if two variables in one group and three in another then 2x3 = 6 pairs) between the two groups and count how many times each group comes out better. (see pg 731)
What proportion corresponds with no effect?
1/2
What variant of the Wilcoxon test is used for multiple groups?
Kruskal-wallis test
When do we use the sign test?
When the sample is dependent (uses matched pairs.)
What are the null and alternative hypothesis for the sign test?
H0: Population proportion p=0.50 who have a better response for a particular group
Ha: p=/= 0.5 Two sided
Ha: p<0.5 (two sided)
Name some disadvantages of nonparametric methods
Confidence intervals not as well developed, methods not well developed for multivariate procedures such as multiple regression also most parametric methods are robust.
