Non parametric tests

Question 1

What are the conditions for a parametric test?
What level of data?
Distribution? test? p-value?
Variance? test? p-value?

Answer 1

Interval or ratio-level data
Normal distribution - Komogrov Simonov (done by spss) test (p-value has to be lower than .05 for normal distribution)
Homogeneity of variance (The variance between conditions should roughly be equal) - Levines test p-value less than 0.05 = violated

Question 2

What are the assumptions of a non-parametric est?
sample
distribution?

Answer 2

Does not need a large sample size
Does not assume a normal distribution
Overall non-parametric tests do not make any assumptions

Question 3

What is the non-parametric equivalent of:
Independent samples t-test?
Repeated samples t-test?
Chi-squared?
Pearson’s correlation?

Answer 3

Independent = Mann Whitney U Test
Repeated = Wilcoson signed ranks test
Chi sqaure has is a non parametric test non paratric test of independence and does not have a parametric eualivan
Pearsons correlation = Spearmans Rho

Question 4

How to check for normal distribution
Two criteria (sample/distribution)
The most appropriate measure of central tendency?

Answer 4

By inspecting histograms, it can show if the data has a normal distribution
For data that does not have a normal distribution and small sample size the most appropriate measure of central tendency is the median rather than the mean

Question 5

What is the associated probability?
What does it show? Which hypothesis?

Answer 5

Associated probability refers to the p-value it shows the probability of gaining the statistic value if the null hypothesis was true.

Question 6

How to decide if there is a significant result

Answer 6

Look at the associated probability value
The p-value has to be less than 0.05 for a significant result

Question 7

Cohen
Classifications for effect sizes

Answer 7

Non = 0
weak = 0.1 - 0.2
Moderate = 0.3 - 0.4
Strong = 0.5 - 0.9
Perfect = 1

Question 8

Mann Whitney and Wilcoxon
Effect size
What does Z refer to?
Does Mann Whitney and wilcoxon calculate effect size?
What value can be used to approximate effect size?
Calculation?
What does N stand for in:
Mann Whitney
Wilcoxon?
Use this to determine effect size

Answer 8

Z = measure of effect size
The test for Mnn Whitney does not calculate effect size but the z-value can be used to approximate the value of r (calculation has to be done by hand)
Calculation: r = Z / square root of N
Mann Whitney, N = number of participants
Wilcoxon. N = Number of observations

Question 9

Reporting results in research reports
What values are italicized?
What is the p-value labelled as on spss? When is the result significant?
Structure for reporting Mann Whitney?
What does Md stand for?
What is the U-value?

Answer 9

P and R should be italicized
The p-value is labelled sig of spss and has to be less than 0.05 for a significant result
Structure = (Md = …) and … (Md =…), U, Z =, p =, r =
Md = median
U = Mann Whitney

Question 10

How to check for normal distribution on Spss

Answer 10

Analyse, descriptive statistics, Explore
Move the DV into the Dependent list test
Move the IV into the factor list test box
Click on plots to open Explore: Plots dialogue box
Selct Normality plots with tests and histogram

Question 11

How to do a Levines test on Spss

Answer 11

Analyse, descriptive statistics, explore
DV into the dependent list test box and Iv into the Factor list test box
Select Plots to open the Explore: Plots dialogue box
Select Power estimation
The result is on the first row

Question 12

Why do the distribution of scores in the non-parametric tests Mann Whitney and Wilcoxom tests not matter?
What are scores converted to?
A statistical significant difference between?

Answer 12

The distribution of scores don’t matter as the scores are converted into ranks
Tests whether there is a statistically significant difference between the mean ranks of two conditions

Question 13

Wilcoxon Signed ranks test
Reporting Spss output
Which table does not need to e reported?
Which hypothesis is reported in the table? how to calculate one-tailed hypothesis?

Answer 13

Don’t report anything from the ranks table
The descriptive statistics table shows the median values
The test statistics table shows the measure of effect size and the significant value for a 2 tailed hypothesis (for a one-tailed hypothesis divide by 2)

Question 14

Wilcoxon Signed ranks test
Converts scores into?
and compares them at?

Answer 14

Converts scores to ranks and compares them at time 1 and time 2

Question 15

Mann Whitney U-test
converts scores into? across how many conditions?

Answer 15

Converts scores to ranks across two conditons

Question 16

Non parametric tests

Answer 16

For statistically significant results, it is important to describe the direction of the findings- which group is higher. This can be seen in the mean rank in the ranks table
But as normal distribution is not assumed it is more appropriate to report the median rather than the mean
How to find the median:
Analyse, compare means, means which opens the means dialogue box
DV into the dependent list box
IV into the independent test box
Options, Means: options dialogue box
Statistic box transfer median to cell statistic box
From the cell, statistic box put the mean and standard deviation in the statistics box

Question 17

Wilcoxon
When are non-parametric tests used (sample size), however, what could the data still be? Therefore what kind of test is not suitable?
How to create a histogram

Answer 17

The kolmogrov-simnov test indicates if scores are normally distributed. It can also be checked by creating histograms
Non-parametric tests are used when the sample sizes are unequal however even though in some cases sample sizes are equal, as they are small there is a high chance that the data is skewed so a t-test would not be suitable.

Create histograms for each time condition
- graphs, histogram, one at a time move each variable to the variable box once the histograms are created click on them to open chart editor and select show distribution curve: normal distribution

If one graph is skewed use a non-‘
parametric test