|
Size: 372
Comment:
|
← Revision 24 as of 2013-03-08 10:17:35 ⇥
Size: 2582
Comment: converted to 1.6 markup
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 1: | Line 1: |
| = A quick guide to choice of sample sizes for Cohen's effect sizes = | |
| Line 2: | Line 3: |
| = A quick guide to sample sizes = | Dunlap and Myers (1997) suggest rules of thumb for sample sizes using Cohen's effect size rules of thumb and equations that they have derived in some simple cases. These are given in the table. The sample sizes guarantee between 80% and 90% power to detect the given effect sizes. |
| Line 4: | Line 5: |
| Dunlap and Myers (1997) suggest rules of thumb for sample sizes using Cohen's effect size rules of thumb and equations that they have derived in some simple cases. | ||||||||<style="TEXT-ALIGN: center"> '''Effect Size''' || '''Formula''' || '''Small''' || '''Medium''' || '''Large''' || ||||||||<style="TEXT-ALIGN: center"> Cohen's d unpaired t, || $$\frac{\mbox{16}}{\mbox{d}^text{2}}$$ + 2 || 0.2 || 0.5 || 0.8 || ||||||||<style="TEXT-ALIGN: center"> equal group sizes || || '''402''' || '''66''' || '''28''' || ||||||||<style="TEXT-ALIGN: center"> Correlation || 8/$$r^text{2}$$ || 0.1 || 0.3 || 0.5 || ||||||||<style="TEXT-ALIGN: center"> Total sample size || || '''800''' || '''88''' || '''32''' || ||||||||<style="TEXT-ALIGN: center"> $$\phi$$, 2x2 table || 8/$$\phi^text{2}$$ || 0.1 || 0.3 || 0.5 || ||||||||<style="TEXT-ALIGN: center"> Total sample size || || '''800''' || '''88''' || '''32''' || |
| Line 7: | Line 14: |
| Dunlap and Myers (1997) show in their appendix that for a 2x2 table of proportions of form ||<tablewidth="52%"17%> ||<38%> Column 1||<45%> Column 2|| ||<17%>Row 1||<38%> $$p_text{11}$$||<45%> $$p_text{12}$$|| ||<17%>Row 2||<38%> $$p_text{21}$$||<45%> $$p_text{22}$$|| with A = $$\frac{p_text{11}}{p_text{11}+p_text{12}}$$, B = $$\frac{p_text{21}}{p_text{21}+p_text{22}}$$ C = $$\frac{p_text{11}}{p_text{11}+p_text{21}}$$, D = $$\frac{p_text{12}}{p_text{12}+p_text{22}}$$ then we can define $$\phi = \sqrt{\mbox{abs(A-B)} \mbox{abs(C-D)}}$$ In addition Maxwell (2000) mentions rules of thumb for power to detect the $$R^text{2}$$ of a single predictor with outcome in a multiple regression with a total of p predictors and a type I error of 0.05. These are give below: ||||||||||<style="TEXT-ALIGN: center"> Regression on p predictors || Small:$$R^text{2}$$=0.02 || Medium:$$R^text{2}$$=0.13 || Large:$$R^text{2}$$=0.26 || ||||||||<style="TEXT-ALIGN: center"> Total sample size || 80% Power || '''392+p''' || '''52+p''' || '''22+p''' || ||||||||<style="TEXT-ALIGN: center"> Total sample size || 90% Power || '''526+p''' || '''70+p''' || '''30+p''' || |
|
| Line 11: | Line 42: |
Maxwell SE (2000) Sample size and multiple regression analysis, ''Psychological methods'' '''5(4)''' 434-458. |
A quick guide to choice of sample sizes for Cohen's effect sizes
Dunlap and Myers (1997) suggest rules of thumb for sample sizes using Cohen's effect size rules of thumb and equations that they have derived in some simple cases. These are given in the table. The sample sizes guarantee between 80% and 90% power to detect the given effect sizes.
Effect Size |
Formula |
Small |
Medium |
Large |
|||
Cohen's d unpaired t, |
$$\frac{\mbox{16}}{\mbox{d}^text{2}}$$ + 2 |
0.2 |
0.5 |
0.8 |
|||
equal group sizes |
|
402 |
66 |
28 |
|||
Correlation |
8/$$r^text{2}$$ |
0.1 |
0.3 |
0.5 |
|||
Total sample size |
|
800 |
88 |
32 |
|||
$$\phi$$, 2x2 table |
8/$$\phi^text{2}$$ |
0.1 |
0.3 |
0.5 |
|||
Total sample size |
|
800 |
88 |
32 |
|||
Dunlap and Myers (1997) show in their appendix that for a 2x2 table of proportions of form
|
Column 1 |
Column 2 |
Row 1 |
$$p_text{11}$$ |
$$p_text{12}$$ |
Row 2 |
$$p_text{21}$$ |
$$p_text{22}$$ |
with
A = $$\frac{p_text{11}}{p_text{11}+p_text{12}}$$, B = $$\frac{p_text{21}}{p_text{21}+p_text{22}}$$
C = $$\frac{p_text{11}}{p_text{11}+p_text{21}}$$, D = $$\frac{p_text{12}}{p_text{12}+p_text{22}}$$
then we can define $$\phi = \sqrt{\mbox{abs(A-B)} \mbox{abs(C-D)}}$$
In addition Maxwell (2000) mentions rules of thumb for power to detect the $$R^text{2}$$ of a single predictor with outcome in a multiple regression with a total of p predictors and a type I error of 0.05. These are give below:
Regression on p predictors |
Small:$$R^text{2}$$=0.02 |
Medium:$$R^text{2}$$=0.13 |
Large:$$R^text{2}$$=0.26 |
||||
Total sample size |
80% Power |
392+p |
52+p |
22+p |
|||
Total sample size |
90% Power |
526+p |
70+p |
30+p |
|||
Reference
Dunlap WP, Myers L. (1997) Approximating power for significance tests with one degree of freedom. Psychological methods 2(2) 186-191.
Maxwell SE (2000) Sample size and multiple regression analysis, Psychological methods 5(4) 434-458.