Timeline for sample size calculated with power.prop.test cs. sample size for significance in chisq.test?
Current License: CC BY-SA 3.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 20, 2017 at 7:36 | vote | accept | Lena743 | ||
| Apr 19, 2017 at 12:36 | comment | added | Lena743 | I guess I deleted my comment earlier. For those who try to follow the conversation, this was my first comment: Thanks for your fast answer and nice function! But I'm not sure if I got it.. The power.prop.test() states how many observations I need in each group to decide for the testgroup in 80% of the tests where the testgroup is better and only in 5% of the tests where the control group is better. Right? So this is my needed sample size, isn't it? Does this mean that my chisq.test() with only half of the sample size just has a higher probability to make the wrong decision? | |
| Apr 19, 2017 at 12:33 | comment | added | Lena743 | yes, right, 50% is what I got there.. Never thought about all this, this way! ok so using the power.prop.test() for calculating the needed sample size is the right approach, right? :) | |
| Apr 19, 2017 at 12:13 | comment | added | Scortchi♦ |
Close: (1) "80% of the tests where the test group is better by 0.34 vs 0.33", to be more precise (more than 80% of course when the test group is better still), & (2) remember you've used a two-tailed test, so it's a 5% chance of deciding for the wrong group when their performance is the same. If you calculate the power for the smaller sample size calc.power(0.34,0.33, 17485, 0.05, 1000) you should find it falls to about 50%, i.e. a 50% chance that you fail to reject the null hypothesis (of equal probabilities) when the true probabilities are 0.34 for test & 0.33 for control.
|
|
| Apr 19, 2017 at 9:04 | history | answered | Scortchi♦ | CC BY-SA 3.0 |