Winsorizing in meta-analysis
Locked
 
|
I am not near my books right now, so I would like to
get feedback from the list.
A colleague was saying that she took a course a few weeks ago, and the prof said that winsorizing usually should be done in meta-analysis. I found this surprising but she was adamant. In general, over analyses in general I have found that a subjective 80% of anomalous values are data entry errors. I have grudgingly trimmed or winsorized only when the value of a variable is substantively implausible? What is your experience? Is routinely winsorizing in meta-analysis what you teach? If you do winsorize in meta-analysis, how often do you do it? How do you decide when to do it? Is this the current practice? Art Kendall Social Research Consultants ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD
Art Kendall
Social Research Consultants |
Re: Winsorizing in meta-analysis
Administrator
|
Hi Art. Several years ago, I worked for a group that did a lot of meta-analysis, and Winsorizing was never part of their procedure. But as I say, that was several years ago. However, I've just searched the Cochrane Handbook (http://www.cochrane-handbook.org/) for "Winsorize" (and "Winsorise"), and found no mention of it there either.
Just to be clear, is your colleague suggesting Winsorization of the point estimates from the various studies before computation of the pooled estimate? Are they computing a fixed or random effects estimate? Do they have a priori hypotheses about possible heterogeneity of the point estimates? HTH. Cheers, Bruce
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
I'm in a situation similar to that of Bruce. I focused on meta-analysis
methods about a decade ago and was surprized to hear about the use of Winsorization. A quick Google search (also of scholar.google.com) in fact turns up a number of references which refer to the use of trimmed means and winsorized variances/standard deviation. One article that might be of relevance is the following: Keselman, H. J.; Algina, James; Lix, Lisa M.; Wilcox, Rand R.; Deering, Kathleen N. (2008). A generally robust approach for testing hypotheses and setting confidence intervals for effect sizes. Psychological Methods, Vol 13(2), Jun 2008, 110-129. doi: 10.1037/1082-989X.13.2.110 Keselman & Co argue for the use of robust effect size measures and propose on based on trimmed means and winsorized variances/SD. Their Table 4 on page 117 provides a listing of the different effect size measures that have suggested as well as their own for independent groups and correlated groups designs. Keselman & Co have several articles on this topic searching scholar.google.com and/or PsycInfo will turn them up as well as articles that cite them. -Mike Palij New York University [hidden email] ----- Original Message ----- From: "Bruce Weaver" <[hidden email]> To: <[hidden email]> Sent: Saturday, April 30, 2011 8:37 PM Subject: Re: Winsorizing in meta-analysis > Hi Art. Several years ago, I worked for a group that did a lot of > meta-analysis, and Winsorizing was never part of their procedure. But as I > say, that was several years ago. However, I've just searched the Cochrane > Handbook (http://www.cochrane-handbook.org/) for "Winsorize" (and > "Winsorise"), and found no mention of it there either. > > Just to be clear, is your colleague suggesting Winsorization of the point > estimates from the various studies before computation of the pooled > estimate? Are they computing a fixed or random effects estimate? Do they > have a priori hypotheses about possible heterogeneity of the point > estimates? > > HTH. > > Cheers, > Bruce > > > > Art Kendall wrote: >> >> <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 >> Transitional//EN"> >> >> >> I am not near my books right now, so I would like to >> get feedback from the list. >> >> A colleague was saying that she took a course a few weeks ago, and >> the prof said that winsorizing usually should be done in >> meta-analysis. I found this surprising but she was adamant. >> >> In general, over analyses in general I have found that a >> subjective 80% of anomalous values are data entry errors. I have >> grudgingly trimmed or winsorized only when the value of a >> variable is substantively implausible? >> >> What is your experience? >> >> Is routinely winsorizing in meta-analysis what you teach? >> >> If you do winsorize in meta-analysis, how often do you do it? How >> do you decide when to do it? >> >> Is this the current practice? >> >> >> Art Kendall >> Social Research Consultants >> >> >> >> >> >> >> >> >> >> ===================== >> To manage your subscription to SPSSX-L, send a message to >> [hidden email] (not to SPSSX-L), with no body text except the >> command. To leave the list, send the command >> SIGNOFF SPSSX-L >> For a list of commands to manage subscriptions, send the command >> INFO REFCARD >> > > ------- > Bruce Weaver > [hidden email] > http://sites.google.com/a/lakeheadu.ca/bweaver/ > > "When all else fails, RTFM." > > NOTE: My Hotmail account is not monitored regularly. > To send me an e-mail, please use the address shown above. > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Winsorizing-in-meta-analysis-tp4361802p4362030.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: Winsorizing in meta-analysis
|
Administrator
|
Hi Mike. Just a couple comments.
1. When I searched the text of that Keselman article for "meta-analysis", it came up only in the titles of some of the cited articles. 2. Someone doing a meta-analysis would only be able to use effect size measures based on Winsorized means and variances if they were reported in the original article, or if they were able to get their hands on the original raw data. 3. Interesting that the authors chose an example where the two distributions are skewed in opposite directions (Fig 1) to illustrate the short-comings of classical methods. I suspect many ardent defenders of classical methods would agree that they don't work very well in that particular case. 4. Very often, meta-analysis in the world of medical research uses measures associated with 2x2 tables (e.g., log-OR, log-RR) or survival time (log-HR) rather than mean differences. This probably accounts for my failure to find any mention of Winsorizing on the Cochrane website. Cheers, Bruce
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |