From: A.J. Rossini (rossini@blindglobe.net)
Date: Fri Jul 11 2003 - 17:03:50 EST
Based on a few private comments, a bit of clarification to my statements in my prior email, quoted below, which seem to have been taken in a different manner than intended, regarding the quality of the statistically procedure given. 1. There are good (theoretically supported) approaches. This isn't one of them. 2. There are ad-hoc approaches, which rely on the reader/reviewer to approve/disapprove. This category is different than bad approaches. Feel free to hate them, the onus is on the presenter to justify. You can make poor or incoherent decisions with these, but not always. The SAM procedure for differential expression in microarrays falls into this category, as does the proposed approach. I personally think the SAM is a reasonable approach, among many, but its characteristics are poorly understood. This last issue holds for the proposed approach (it has poorly or impossible to understand characteristics!). 3. There are bad approaches (either of the above, where the data generating process violates assumptions required to apply the procedure). One can claim that all ad-hoc approaches are bad, but that is being fairly conservative, requiring the data to speak for itself without the benefits of scientific theory or knowledge. Of course, the FDA falls into this category. best, -tony rossini@blindglobe.net (A.J. Rossini) writes: > "Bryant, Jenny (NSW)" <JBryant@arcbs.redcross.org.au> writes: > >>> I have the following advice to determine whether or not to average all >>> three readings in a set of 3 measurements:- >>> >>> Average the three Mean Channel values. To determine whether a value is >>> too low or too high to be averaged with the others use the formula: >>> Y = (X' - X")/(Xh - Xl) >>> Where (X' - X") is the difference between the 2 closest values, and (Xh - >>> Xl) is the difference between the highest and lowest values. >>> If Y is less than or equal to 0.25, average the closest values. >>> If Y is greater than 0.25, average all 3 values. >>> >>> We have been looking at a number of statistical methods and there appears >>> to be some truth in the formula however if 2 readings are the same the >>> third will always be excluded no matter how close it is. Also, we can't >>> figure out the derivation of the value 0.25. >>> >> Does anyone have any suggestions? > > > I've never seen that ad-hoc approach before, but here's my > interpretation with respect to 3 observations: > > one of the closest two MUST be an endpoint (max or min, i.e. one > of X' and X" must be one of Xh or Xl). > > therefore, if the third (not X' or X"; must be one of Xl or Xh) is > greater than 3 times the distance of the closest two (X'-X") from > the middle one, it is excluded. > > 3 times is maybe a reasonable cut off. I don't think that there is > any good justification with just 3 points (though the three points in > this case sound like summary measures, so there is some > accuracy/information behind them), but it doesn't sound like a bad > approach, as long as it is documented. > > best, > -tony > > -- > A.J. Rossini / rossini@u.washington.edu / rossini@scharp.org > http://software.biostat.washington.edu/ UNTIL IT MOVES IN JULY. > Biomedical and Health Informatics, University of Washington > Biostatistics, HVTN/SCHARP, Fred Hutchinson Cancer Research Center. > FHCRC: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email > > CONFIDENTIALITY NOTICE: This e-mail message and any attachments may be > confidential and privileged. If you received this message in error, > please destroy it and notify the sender. Thank you. > -- A.J. Rossini / rossini@u.washington.edu / rossini@scharp.org http://software.biostat.washington.edu/ UNTIL IT MOVES IN JULY. Biomedical and Health Informatics, University of Washington Biostatistics, HVTN/SCHARP, Fred Hutchinson Cancer Research Center. FHCRC: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email CONFIDENTIALITY NOTICE: This e-mail message and any attachments may be confidential and privileged. If you received this message in error, please destroy it and notify the sender. Thank you.
This archive was generated by hypermail 2.1.6 : Thu Jan 01 2004 - 17:43:47 EST
