Log in
Register|
Ravi ★ India, 2008-11-20 13:50 (6431 d 01:13 ago) Posting: # 2748 Views: 4,096 |
|
|
Dear All, I read in ICH guidlines for stability that a level of significace of 0.25 is to used. But as far as my knowedge is concern significance level is usually taken as 0.05. So Please can any one explane me the statistical rational of choosing signifinace level of 0.25 in testing equality of slope and intercepts in stability analysis. — Thanks & Regards Ravi Pandey |
|
d_labes ★★★ Berlin, Germany, 2008-11-20 16:11 (6430 d 22:52 ago) @ Ravi Posting: # 2751 Views: 3,030 |
|
|
Dear Raci, Ravi? ❝ [...] So Please can any one explane me the statistical rational of choosing signifinace level of 0.25 in testing equality of slope and intercepts in stability analysis. First I hope you are talking about ICH Q1E. The significance level of 0.25 is mentioned here, section B.2.2, for a preliminary test of the equality of slope and intercept of regression lines before pooling. The test mentioned actually is not a test of equality, but rather a test for differences. Means the hypotheses tested are not the right one of interest. Additionally the sample size of such studies leads to low power of the test, i.e. to detect real deviations. Due to this it is an old fashioned approach (used in many statistical text books, look for instance for tests of variance homogeneity in the ANOVA contexts) to rise the significance level above the usual used 0.05. You will find values of 0.1 up to 0.4 for such tests. BTW, check your signature (Regarrrrrrrrrds). Or is your keyboard defect  .— Regards, Detlew |
|
Helmut ★★★   Vienna, Austria, 2008-11-20 16:37 (6430 d 22:25 ago) @ d_labes Posting: # 2752 Views: 3,073 |
|
|
Dear DLabes! ❝ Dear Raci, Ravi? ❝ BTW, check your signature (Regarrrrrrrrrds). Or is your keyboard defect I guess Ravi (according to the profile). The forum’s software does not allow to change the User’s name (only the password). So once you are registered (with a typo), you’re put. Sometimes users ask me to change their name – which I can do only by editing entries in the MySQL-database directly. @Ravi: If you want to change your user’s name, please let me know. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! ![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png) Helmut Schütz ![[image]](https://static.bebac.at/img/CC by.png) The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
Ravi ★ India, 2008-11-21 06:16 (6430 d 08:47 ago) @ Helmut Posting: # 2756 Views: 2,947 |
|
|
Dear HS, ❝ ❝ @Ravi: If you want to change your user’s name, please let me know. Yes you are right it was a typo error So please correct my name i.e. Ravi Pandey — Thanks & Regards Ravi Pandey |
|
Ravi ★ India, 2008-11-21 07:01 (6430 d 08:02 ago) @ d_labes Posting: # 2758 Views: 2,995 |
|
|
Dear d_labes, First of all thanks for ur reply. ❝ ❝ First I hope you are talking about ICH Q1E. Yes u are right I am taking about ICH Q1E. ❝ ❝ it is an old fashioned approach (used in many statistical text books, look for instance for tests of variance homogeneity in the ANOVA contexts) to rise the significance level above the usual used 0.05. ❝ ❝ You will find values of 0.1 up to 0.4 for such tests. From ur reply I still haven't got my answer. You have said that it is an old fashioned approach to rise the significance level above the usual used 0.05. But my question was what is the statistical rational for which this approach was used in old days. Hope you don't mind in answering my question again. — Thanks & Regards Ravi Pandey |
|
d_labes ★★★ Berlin, Germany, 2008-11-21 10:34 (6430 d 04:29 ago) @ Ravi Posting: # 2762 Views: 2,997 |
|
|
Dear Ravi, ❝ Hope you don't mind in answering my question again. Ok. The answer was contained in my previous post, eventually not as explicit as necessary to be understandable. Lets try the other way. Because the test is a test for difference, but we are interested in testing for equality (here poolability of regression lines) we are more interested in controlling the power of the test. As said in the Guideline, power is usually low given the sample size of such analyses. Now it comes to the old fashioned "trick". Raising the significance level of a test increases power. Why this is so may be illustrated by an rather artificial "test": We always accept the alternative hypothesis. Ok this is not really a test to chose between hypotheses but rather a prejudice  .With this "test" the significance level (error to reject H0, the Null if H0 is true) is 1, raised to his maximum possible value. But the type II error to not reject H0 if HA, the Alternative, is true, the beta error is zero. Because power=1-beta then power is maximum power=1. Hope this helps. — Regards, Detlew |
|
martin ★★ Austria, 2008-11-21 11:33 (6430 d 03:30 ago) @ Ravi Posting: # 2764 Views: 3,218 |
|
|
dear all ! you may find this thesis (Kiermeir, 2003) of interest which gives a nice overview regarding stability testing. hope this helps martin |
Ing. Helmut Schütz