Timeline for Reagent batch change, sample size, results significancy
Current License: CC BY-SA 4.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 4, 2020 at 12:33 | history | edited | Bernhard | CC BY-SA 4.0 |
added 1625 characters in body
|
| Sep 4, 2020 at 11:32 | comment | added | Olivier D. |
Thanx @Bernhard for this extra explanation. The test is not cheap, and using patient blood is a critical resource so we need to optimize the sample size. We have used the pwr.t.test function already 2 months ago and found 3,5 or 10 samples depending on expectations. This guided our experiments so far. We are wondering 1/ how can we interpret the t.test result (you answered with overlapping CI - still wondering if 0 needs to be contained or not) and 2/ how can we know this result is significant and we can trust it (we thought t.test p-value would answer but it's not obvious for us)
|
|
| Sep 3, 2020 at 14:30 | comment | added | Bernhard |
(ctd) The question of what sample size you need for a one sample t test to discover an deviation of exactly 5.4 (you will need to compute the effect size from that) with a power of at least 95% can be computed via the function pwr.t.test after installing the package pwr : rdocumentation.org/packages/pwr/versions/1.3-0/topics/…
|
|
| Sep 3, 2020 at 14:30 | comment | added | Bernhard | @OlivierD. Great. Now we need some critical thinking on what defines an optimal size. If tests are cheap do as many as you can. That will make the confidence intervall small and thus the chance of erroneously throwing away reagents. If they are not, you will probably want to define a minimal accepted power. (to be continued) | |
| Sep 3, 2020 at 14:03 | comment | added | Olivier D. | @Bernhard I appreciate your addendum. It helps a lot on question 2. | |
| Sep 3, 2020 at 13:28 | history | edited | Bernhard | CC BY-SA 4.0 |
added 75 characters in body
|
| Sep 3, 2020 at 13:25 | comment | added | Bernhard | @StatsStudent Thank you. Looking forward to read yours. | |
| Sep 3, 2020 at 13:24 | comment | added | Bernhard | Well it helped as much as it made you specify an acceptable deviation. Once StatsStudent posts his answer he is likely to refer to that. Meanwhile, I made an addendum to my answer, hopefully expanding usefully with the directions you gave in the first comment. | |
| Sep 3, 2020 at 13:21 | history | edited | Bernhard | CC BY-SA 4.0 |
added 706 characters in body
|
| Sep 3, 2020 at 11:59 | comment | added | Olivier D. | OK @StatsStudent ; to be honest this answer didn't help much | |
| Sep 2, 2020 at 20:18 | comment | added | StatsStudent | Excellent answer, IMHO. But I'll add another when I wrap up some work here. +1 @OlivierD. | |
| Sep 2, 2020 at 13:26 | comment | added | Olivier D. | Our acceptable deviation is 1*sigma to 3*sigma (sigma=1.8) ; so say 3*sigma=5.4 We are ok with 90 or 95% confidence. As you can read in the question, we have already done the t.test with paired=T. We need help in interpreting the results of this t.test, and help us understand if it is significant or not. Thanks, Olivier. | |
| Sep 2, 2020 at 12:19 | history | answered | Bernhard | CC BY-SA 4.0 |