Dear R- and Omega-list-members, I am trying to make statistical inference about the following design: A dependent variable y has been measured multiple times, i.e. 4 times (y1,y2, y3, y4), unfortunately suffering from some successive dropouts (i.e. the sample sizes varies for y1, y2, y3, and y4). For every y, two other variables (covariates) were also measured: x & z, and both do presumably exert an effect on y. At some cutoffs, x can also be trichotomized into 3 ordinal levels constituting a presumed factor (fx) influencing the level of y at all of the different measurings (1-4). x and fx are rather stable arcoss the 4 measurements whereas z is not. H0 is that x and fx are not influencing the level of y (i.e. explaining any variance of y) irrespective of (i.e. controlled for) z at any of the measurements. (1) What would be a good way of testing for the hypothesis in the context of a GLM or a canonical regression analysis? I can see that in a parametric testing and for a single measurement of y a simple multiple linear regression (MLR) or an ANCOVA (or a RANCOVA in a nonparametric approach) would do the trick. However, I am not sure how to tackle the issue facing repeated and at least biologically somehow interdependent measurements and the goal to include as much measurements as possible even though the sample size differs. I thought about running an MLR for x and an ANCOVA for fx seperately for y1, y2, y3 and y4 but I am not sure if this would require a correction for multiple testings and if this is in fact the best approach at all. Pooling all 4 measurements, on the other hand, would mean to pretend that they were all derived from different subjects which is clearly not the case. So, basically I don?t really know how to treat the testing. (2) Is there an implementation in R or Omega to perform such testing? How would that run? Hopefully, the question isn?t too trivial to the list - I am not a statisticician and just fed up with the comercial stats software... Thank you very much in advance- Andreas Bartsch, MD -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
A lot of basic information is missing. For example, is y a continuous measurement? What do you mean by GLM? (Warning, SAS and R mean different things.) .... If y is indeed continuous, I think lme (in R's recommended package nlme) will do what you want, with a random effect for subject. It does need some learning. The Pinheiro & Bates book will help a lot, provided you have the background knowledge they presume. AFAIK there is no suitable software in Omegahat. On Sat, 11 May 2002, Andreas Bartsch wrote:> Dear R- and Omega-list-members, > I am trying to make statistical inference about the following design: > A dependent variable y has been measured multiple times, i.e. 4 times > (y1,y2, y3, y4), unfortunately suffering from some successive dropouts (i.e. > the sample sizes varies for y1, y2, y3, and y4). For every y, two other > variables (covariates) were also measured: x & z, and both do presumably > exert an effect on y. At some cutoffs, x can also be trichotomized into 3 > ordinal levels constituting a presumed factor (fx) influencing the level of > y at all of the different measurings (1-4). x and fx are rather stable > arcoss the 4 measurements whereas z is not. H0 is that x and fx are not > influencing the level of y (i.e. explaining any variance of y) irrespective > of (i.e. controlled for) z at any of the measurements. > (1) What would be a good way of testing for the hypothesis in the context of > a GLM or a canonical regression analysis? I can see that in a parametric > testing and for a single measurement of y a simple multiple linear > regression (MLR) or an ANCOVA (or a RANCOVA in a nonparametric approach) > would do the trick. However, I am not sure how to tackle the issue facing > repeated and at least biologically somehow interdependent measurements and > the goal to include as much measurements as possible even though the sample > size differs. I thought about running an MLR for x and an ANCOVA for fx > seperately for y1, y2, y3 and y4 but I am not sure if this would require a > correction for multiple testings and if this is in fact the best approach at > all. Pooling all 4 measurements, on the other hand, would mean to pretend > that they were all derived from different subjects which is clearly not the > case. So, basically I don´t really know how to treat the testing. > (2) Is there an implementation in R or Omega to perform such testing? How > would that run? > Hopefully, the question isn´t too trivial to the list - I am not a > statisticician and just fed up with the comercial stats software... Thank > you very much in advance- > Andreas Bartsch, MD > > -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- > r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html > Send "info", "help", or "[un]subscribe" > (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch > _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._ >-- Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272860 (secr) Oxford OX1 3TG, UK Fax: +44 1865 272595 -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Dear Professor Ripley, dear list members- thanks a lot for your comments. I am sorry for the missing basics - I wrote the email after an on-call service this morning. y is indeed continuous. x and z are as well, fx is ordinal (x trichotomized at cutoffs). And yes, I think I am after a linear mixed effects model (by GLM I meant a general linear modelling) but again: I am not sure and don?t know how to treat the data. I will read through the nlme examples but maybe someone may point me to crucial aspects my description was missing and a very brief outline how to proceed using nlme. TIA for any comments- Andreas ----- Original Message ----- From: <ripley at stats.ox.ac.uk> To: "Andreas Bartsch" <bartsch at neuroradiologie.uni-wuerzburg.de> Cc: <r-help at stat.math.ethz.ch> Sent: Saturday, May 11, 2002 6:31 PM Subject: Re: [R] modelling a particular design A lot of basic information is missing. For example, is y a continuous measurement? What do you mean by GLM? (Warning, SAS and R mean different things.) .... If y is indeed continuous, I think lme (in R's recommended package nlme) will do what you want, with a random effect for subject. It does need some learning. The Pinheiro & Bates book will help a lot, provided you have the background knowledge they presume. AFAIK there is no suitable software in Omegahat. On Sat, 11 May 2002, Andreas Bartsch wrote:> Dear R- and Omega-list-members, > I am trying to make statistical inference about the following design: > A dependent variable y has been measured multiple times, i.e. 4 times > (y1,y2, y3, y4), unfortunately suffering from some successive dropouts(i.e.> the sample sizes varies for y1, y2, y3, and y4). For every y, two other > variables (covariates) were also measured: x & z, and both do presumably > exert an effect on y. At some cutoffs, x can also be trichotomized into 3 > ordinal levels constituting a presumed factor (fx) influencing the levelof> y at all of the different measurings (1-4). x and fx are rather stable > arcoss the 4 measurements whereas z is not. H0 is that x and fx are not > influencing the level of y (i.e. explaining any variance of y)irrespective> of (i.e. controlled for) z at any of the measurements. > (1) What would be a good way of testing for the hypothesis in the contextof> a GLM or a canonical regression analysis? I can see that in a parametric > testing and for a single measurement of y a simple multiple linear > regression (MLR) or an ANCOVA (or a RANCOVA in a nonparametric approach) > would do the trick. However, I am not sure how to tackle the issue facing > repeated and at least biologically somehow interdependent measurements and > the goal to include as much measurements as possible even though thesample> size differs. I thought about running an MLR for x and an ANCOVA for fx > seperately for y1, y2, y3 and y4 but I am not sure if this would require a > correction for multiple testings and if this is in fact the best approachat> all. Pooling all 4 measurements, on the other hand, would mean to pretend > that they were all derived from different subjects which is clearly notthe> case. So, basically I don?t really know how to treat the testing. > (2) Is there an implementation in R or Omega to perform such testing? How > would that run? > Hopefully, the question isn?t too trivial to the list - I am not a > statisticician and just fed up with the comercial stats software... Thank > you very much in advance- > Andreas Bartsch, MD > > -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-> r-help mailing list -- Readhttp://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html> Send "info", "help", or "[un]subscribe" > (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch >_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._. _._>-- Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272860 (secr) Oxford OX1 3TG, UK Fax: +44 1865 272595 -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._