STATS 16- Non-parametric tests Flashcards

1
Q

Why so many different tests

A
  • Different data types
  • Different experiments have different numbers of conditions
    • One way, Two ways
  • Different subjects design require different tests, because different assumptions can be made
    • Independent vs repeated measures
    • Normal vs non-parametric
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Chi-squared test

A
  • One variable chi-squared
  • Chi-squared tests of association
  • Both calculate the difference between
  • OBSERVED frequencies (your data) AND
  • EXPECTED frequencies (arising by chance)
  • If the difference (O-E) very large: a real effect
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

It’s all greek

A
  • X2= Chi-squared
  • E= sigma (sum of)
  • O= observed value
  • E= Expected value
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Discrimination in the workplace

A
  • 6 male managers
  • 4 female managers
  • Is this discrimination against women?
  • Would expect 5 of each sex but the reality is only one person out- most likely just chance
  • Due to small numbers can’t say it is due to discrimination
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Bigger firm: Contingency table

A
  • Frequency of male-female managers
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What might we expect

A
  • Chi-squared divides each difference squared by the expected frequency and sums them all- this is the value of chi
  • If the differences are larger, chi will be larger= more likely to be a trend as oppose to chance
  • X2= 100/50 + 100/50 = 4
  • The value of 4 (chi) is our CALCULATED STATISTIC in this case (p<0.05)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What a computer would do for us

A
  • 60 men
  • 40 women
    *
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

One variable chi square test

A
  • Throwing a die on the following numbers with the following frequencies
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Interpreting the output

A
  • Because p value= <0.05 (0.013) we can say there is a significant difference between what we expect and what actually happened with a Chi2 value of 14.4
  • NB- for Chi2 to be valid, the observed frequency in each cell of a contingency table must be at least 5
    *
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Is it significant?…see more later

A
  • Compare your CALCULATED STATISTIC
    • t…From t-test
    • f…From ANOVA
    • X2…chi-squared test
  • Compare with
  • A CRITICAL VALUE
    • In a stats table
    • A computer does it for you
    • Usually for different df, and different levels of confidence (5%, 10%)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

The sign test

A
  • The simplest but least powerful test for a difference between 2 conditions
  • Subject design: repeated measures
  • Description: Counts the number of times one conditions is larger than the other and compares this number to what would be expected by chance
  • NB: need at least 6 pairs of scores
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Calculation of the sign test

A
  1. Discard people who scored the same in both conditions (discard pairs)
  2. Count how many scored more in 1st condition and how many scored less in 1st condition
  3. Take the lower number from 2
  • If the result is equal or less than the critical value for the number of people in the experiment, THERE IS A SIGNIFICANT DIFFERENCE
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Example

A
  • Hypothesis: there will be a difference between IQ scores between the first and the second time to take the test
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Use of sign test

A
  • Generally only when the direction but not the size of the difference is important
    • Does not give information about size of difference
  • Can handle Dichotomous data (nominal data)
    • (Only 2 possible values) e.g. yes/no
    • Not many tests can do this
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Wilcoxon Test

A
  • Subject design: repeated measures
  • Description: The value of the differences between the 2 conditions (e.g. Before vs After) for each person is ranked
    • Small differences have low (small numbers) ranks
  • Sum ranks of those scoring MORE in 2nd condition, and sum those that score LESS
  • The smaller of the 2 summed values is “T”
    • Compared against a critical value
    • T should be equal or smaller to be significant
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Calculation of the Wilcoxon Statistic “T”

A
  • Discard SUBJECTS who scored the same in both conditions
  • Take the smallest score from the largest score for each SUBJECT
  • Rank these differences (smallest=1, largest= n)
    • Add up all the ranks for SUBJECTS who did best in Condition A and Condition B separately (Sum rank of those who did worse 2nd time)
  • Take the smaller of these values= “T”
    • T must be equal to or lower than the critical value for the conditions to be different
17
Q

Example

A
18
Q

Critical Values- Wilcoxon

A
  • 3.5 is greater than 2 therefore there is no significant difference
  • T must be equal to or less than the critical value in the wilcoxon table for there to be a significant difference
19
Q

Wilcoxon- advantage/disadvantages

A
  • No good with dichotomous data because there are too many rank ties
    • Differences always the same
  • Takes into account the size and direction of the difference
    • As you sum the ranks of each direction separately
20
Q

Sign vs Wilcoxon

A
  • Wilcoxon takes into account the size of the differences between the conditions as well as the direction of the difference so is better than sign test
  • Both require the calculated statistic to be smaller than the critical value
    • A smaller calculated statistic suggests more consistency in the data
    • (i.e. a pattern different from a random error
  • Both have the advantage of being able to analyse repeated measures (within subjects) design