Upload
mazhar-perdana
View
214
Download
0
Embed Size (px)
Citation preview
8/10/2019 Vonk Cochrans Q 2011 June 7
1/38
Jet Vonk
Cochrans Q test
Methodology and Statistics for Linguistic Research(LTR002M10, 2010/2011)
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
2/38
Jet Vonk
Content
Types of data Cochrans Q test
Who?
What?
When?
How?
Application?
Cochrans Q test: Example Limitations of the Cochrans Q test
Discussion
References
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
3/38
Jet Vonk
Types of data
10-6-2011
nominal
8/10/2019 Vonk Cochrans Q 2011 June 7
4/38
Jet Vonk
Types of data
10-6-2011
nominal
Nominal data: no specific order, nonumerical meaning, differentiatedby a naming system
Examples:
Men/women
Set of countries
8/10/2019 Vonk Cochrans Q 2011 June 7
5/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
8/10/2019 Vonk Cochrans Q 2011 June 7
6/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
Ordinal data: items have a
specific order on the scale
Examples:
Positions within a company Winner, second and third in a
race
8/10/2019 Vonk Cochrans Q 2011 June 7
7/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
8/10/2019 Vonk Cochrans Q 2011 June 7
8/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
Interval data: ordered,constant scale, but no naturalzero
The intervals between valuesare equal
Cannot be multiplied or
divided
Examples:
Temperature
Dates
8/10/2019 Vonk Cochrans Q 2011 June 7
9/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
ratio
8/10/2019 Vonk Cochrans Q 2011 June 7
10/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
ratio
Ratio data: ordered, constantscale, natural zero
Can be multiplied or divided
Examples:
Age
Height
Weight
Length
Time
8/10/2019 Vonk Cochrans Q 2011 June 7
11/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
ratio
Non-parametric
8/10/2019 Vonk Cochrans Q 2011 June 7
12/38
Jet Vonk
Types of data
10-6-2011
nominal
ordinal
interval
ratio
Non-parametric
Parametric
8/10/2019 Vonk Cochrans Q 2011 June 7
13/38
Jet Vonk
Types of data
10-6-2011
nominal
Non-parametric
8/10/2019 Vonk Cochrans Q 2011 June 7
14/38
Jet Vonk
Types of data
10-6-2011
nominal
2 samples
k samples
8/10/2019 Vonk Cochrans Q 2011 June 7
15/38
Jet Vonk
Types of data
10-6-2011
nominal
dependent
independent
k samples
2 samples
dependent
independent
8/10/2019 Vonk Cochrans Q 2011 June 7
16/38
Jet Vonk
Cochrans Q test
10-6-2011
nominal
dependent
independent
k samples
2 samples
dependent
independent
8/10/2019 Vonk Cochrans Q 2011 June 7
17/38
Jet Vonk
Cochrans Q test: who?
William Gemmell Cochran
(1909 1980)
Cochran, W.G. (1950). The Comparison ofPercentages in Matched Samples.Biometrika, 37, 256-66.
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
18/38
Jet Vonk
Cochrans Q test: what?
McNemar (1947) considered the problem of testing thesignificance of the difference between two correlatedsample proportions
Cochran suggested a generalization of the problem in
which there are k(>2) matched samples
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
19/38
Jet Vonk
Cochrans Q test: what?
To statistically analyze success rate data
Tests the hypothesis that several related dichotomous
variables have the same mean The variables are measured on the same individual or
on matched individuals
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
20/38
Jet Vonk
Cochrans Q test: when?
Nominal data More than 2 samples
Dependent
Binary response: succes (1) versus failure (0)
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
21/38
Jet Vonk
Cochrans Q test: how?
Each of k treatments is independently applied to bblocks (or subjects) and each outcome is measured asa success (1) or as a failure (0)
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
22/38
Jet Vonk
Cochrans Q test: how?
Hypotheses: H0: treatments are similarly effective
H1: treatments differ in effectiveness
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
23/38
Jet Vonk
Cochrans Q test: how?
Test statistic:
k is the number of treatments
X jis the column total for thejth treatment
b is the number of blocks
Xi is the row total for the ithblock
Nis the grand total
For significance level , the critical region is
where 21 ,k 1 is the (1 )-quantile of the chi-square distribution
with k 1 degrees of freedom
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
24/38
8/10/2019 Vonk Cochrans Q 2011 June 7
25/38
Jet Vonk
Cochrans Q test: application?
Cochrans Q is often used for meta-analyses, e.g: Is there a difference in treatments (to test)?
Is there a difference in tasks (to test)?
Is there a difference in materials (to test)?
But also: are used in the same rate or is there a difference?
Methods Materials
Devices
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
26/38
Jet Vonk
Cochrans Q test: example
Background: A teacher has been observed on 20 different moments in
his lessons (when he was about to explain something)
Observed on 3 different teaching strategies:
Point out the subject
Correspond with questions of students
Use of a stimulating beginning
Question: does the teacher use these strategies in thesame rate or is there a difference?
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
27/38
Jet Vonk
Cochrans Q test: example
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
28/38
Jet Vonk
Cochrans Q test: example
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
29/38
Jet Vonk
Cochrans Q test: example
10-6-2011
There is a significant difference in the use of teachingstrategies used by this teacher (X2(2) = 13.37, p = .001)
8/10/2019 Vonk Cochrans Q 2011 June 7
30/38
Jet Vonk
Cochrans Q test: example
When you find any significant effect, you need to do apost-hoc test (as you do for ANOVA)
For Cochran's Q test: run multiple McNemar's testsand adjust the p values with Bonferroni correction(a method used to address the problem of multiple
comparisons, overcorrects for Type I error)
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
31/38
Jet Vonk
Cochrans Q test: example
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
32/38
Jet Vonk
Cochrans Q test: example
10-6-2011
Bonferroni correction: = 0.05
3 comparisons
0.05/3= 0,0166666666666667
Only the difference between the use of point out thesubject and stimulating beginning is significant
8/10/2019 Vonk Cochrans Q 2011 June 7
33/38
Jet Vonk
Limitations of the Cochrans Q test
Only determines the occurrence of a change, but does not evaluatesthe extent of change
Possible to do multiple McNemars tests, but no interactioneffect can be measured
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
34/38
Jet Vonk
Limitations of the Cochrans Q test
Only determines the occurrence of a change, but does not evaluatesthe extent of change
Possible to do multiple McNemars tests, but no interactioneffect can be measured
The test is known to be poor at detecting true heterogeneity amongstudies as significant
Meta-analyses often include small numbers of studies, and thepower of the test in such circumstances is low
Because the test is poor at detecting true heterogeneity, a non-significant result cannot be taken as evidence of homogeneity
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
35/38
Jet Vonk
Limitations of the Cochrans Q test
Only determines the occurrence of a change, but does not evaluatesthe extent of change
Possible to do multiple McNemars tests, but no interactioneffect can be measured
The test is known to be poor at detecting true heterogeneity amongstudies as significant
Meta-analyses often include small numbers of studies, and thepower of the test in such circumstances is low
Because the test is poor at detecting true heterogeneity, a non-significant result cannot be taken as evidence of homogeneity
The test does not accommodate a control group, because it is a test
for use with dependent observations
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
36/38
Jet Vonk
Discussion
To begin with:
Questions?
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
37/38
Jet Vonk
Discussion
Discussion points?
for example:
Is this analysis usefull in Linguistic analyses? Is it applicable in your research (e.g. pilot study)?
10-6-2011
8/10/2019 Vonk Cochrans Q 2011 June 7
38/38
Jet Vonk
References
Cochran, W.G. (1950). The Comparison of Percentages in Matched Samples.
Biometrika, 37, 256-66.
Higgins, J.P.T., Thompson, S.G., Deeks, J.J., Altman, D.G. (2003). Measuringinconsistency in meta-analyses.BMJ, 327, 557-560.
Pett, M.A. (1997).Nonparametric Statistics for Health Care Research: Statistics for
Small Samples and Unusual Distributions. Thousand Oaks, CA: SAGE Publications.
10-6-2011