Timeline for Pre-specified comparisons procedure: How comparisons can be made without looking at the data
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| May 24, 2023 at 19:06 | comment | added | Ute | If you have more than one treatment group, you can expand my example by a third group C, or many more groups, as needed. Then the tests you would do is really a one-way test for each variable separately. But 100 tests are too many, so you have to select some variables that look interesting. Or use some "variable selection" methods that do that based on the data, which is a bit snooping (and snooping should be corrected for) | |
| May 24, 2023 at 19:02 | comment | added | Ute | I get really confused by the "one-way model with 100 variables". Do you have a treatment and a control group in your data, and there has been measured height, weight, bmi etc? In that case, you can say A=treatment group, and B = control group (A, B from my explanation). A and B stand for the individuals where you measured all these variables. So you would test 100 hypotheses: (I use the variable names for the means now): bmi_A = bmi_B, height_A=height_B, ... and so on. This gives 100 tests. Prespecified: you only do some selected of these 100 tests. | |
| May 24, 2023 at 5:05 | comment | added | Tran Khanh | The $H_{0}$ for the one-way model is all treatment (or population) means are equal, $H_{a}$ at least two means are not equal. And actually in the your example, what confounds me is how to construct hypothesis: There're 100 variables (height, weight, bmi,...) and the two locations (A, B), so I guess it's two-way models. But I'm currently stuck on the Bonferroni comparison for the one-way model. | |
| May 23, 2023 at 11:23 | comment | added | Ute | I need a bit more information to map your case of multiple treatment comparison to my answer in a helpful manner: What is the test you are looking for? What is the null hypothesis? (Sorry, I don't have time to find a book online and read it :-)) | |
| May 23, 2023 at 1:02 | comment | added | Tran Khanh | It's just the Center mean + Random error (CE model) for multiple treatments: $Y = \mu_{i} + \epsilon_{ij}, i=0, 1, ..., k$ (number of treatments) and $j=0, 1,..., n_{i}$ where $n_{i}$ is the number of observations in treatment $i$, $\epsilon_{ij} \sim N(0, \sigma^2)$. Also, the book is available online. | |
| May 22, 2023 at 17:10 | comment | added | Ute | I don't know your one-way model. Can you give more details, perhaps in your question? | |
| May 22, 2023 at 12:50 | comment | added | Tran Khanh | thanks for your answer! However, the example is beyond my grasp. It would be nice if you explain it using the one-way model which I'm studying, or some resources that I can elaborate on are more than enough. | |
| May 22, 2023 at 12:19 | vote | accept | Tran Khanh | ||
| May 22, 2023 at 11:25 | history | answered | Ute | CC BY-SA 4.0 |