Dear Kersten, Thanks for reply. I have new question. When I set model use total_i_vol_stats = lme_fit_FS(X,[1 2],Y(:,2),ni) and total_i_vol_stats = lme_fit_FS(X,[1],Y(:,2),ni). But Bhat of two models' year have a lot different, for example, with 2 is 1.2422 without 2 is -0.5737. The direction has already changed. I feel confuse about this. Looking for your reply.
Thanks, Lanbo On Wed, Mar 28, 2018 at 2:05 PM, Diers, Kersten /DZNE <kersten.di...@dzne.de > wrote: > Hi > > sorry for the delayed response, I was not able to reply during the last > week. > > You are right that the F-test provided in the LME toolbox initially > does not provide information about the direction of the effects. > > You could do the following: > > First, it is always useful to plot the data to get a first impression. > However, if the statistical model also contains additional covariates, > this plot does not completely reflect the effects as estimated by the > model. So this may not give a complete picture. > > A second option is to take a look at the sign of the estimated beta > values, which also reflect the direction of the effects. These beta > values are stored in the output of the 'lme_fit_FS' script, e.g. > 'total_hipp_vol_stats.Bhat' in the tutorial example. For proper > interpretation, you'd need to relate the beta values and their > corresponding columns of the design matrix, and also take your > contrasts of interest into account. So this is somewhat complicated. > > As an alternative, the LME toolbox also provides so-called "signed p- > values". Note that these are not p-values in the conventional sense. A > signed p-value is the sign (-1 or +1) of the scalar product of a row of > the contrast matrix and the vector of the estimated betas, and it can > give information about the direction of a specific effect. > > These signed p-values can be accessed from the output of the 'lme_F' > script, see e.g. the 'F_C.sgn' variable for the tutorial example. > > There will be as many signed p-values as there are rows in the contrast > matrix, so each signed p-value corresponds to one row of the contrast > matrix. > > The interpretation of this signed p-value depends on how the contrast > was formulated: > > In a hypothetical example (different from and simpler than the tutorial > example), suppose that one row of a contrast matrix starts like > > 0 1 -1 ... > > and the entries correspond to group1, group2, group3, ..., then a > positive signed p-value will in this particular case indicate that the > difference 'group2 minus group3' is greater than zero, which means that > group2 has greater values than group3. On the other hand, a negative > signed p-value will in this particular case indicate that the > difference 'group2 minus group3' is less than zero, and hence group3 > must have greater values than group2. > > However, also consider the equally valid alternative that a row of a > contrast matrix starts like > > 0 -1 1 ... > > and the entries again correspond to group1, group2, group3, ..., then a > positive signed p-value will in this particular case indicate that the > difference 'group3 minus group2' is greater than zero, which means that > group3 has greater values than group2. On the other hand, a negative > signed p-value will in this particular case indicate that the > difference 'group3 minus group2' is less than zero, and hence group2 > must have greater values than group3. > > So the second interpretation is just the opposite as the first one. > This illustrates that careful attention needs to be paid to the > formulation of the contrasts. Also, it's best if the interpretation > gained from the signed p-values agrees with the interpretation of a > simple plot of the data. > > To summarize, if we are testing for simple group differences, the F- > test only provides a single p-value, which indicates if there is at > least one significant difference among the several groups. To get an > idea which groups actually differ, and in which direction, one needs to > take a closer look, for example at those effects that are reflected in > the rows of the contrast matrix; for this, one option is to use the > signed p-values as described above. > > Hope this helps, > > Kersten > > On Di, 2018-03-27 at 15:31 +0200, lanbo Wang wrote: > > Dear Kersten, > > > > I have a question about LME model. After I acquired p value, could I > > know which group is bigger? > > > > Thanks, > > Lanbo > > > > On Fri, Mar 16, 2018 at 12:13 AM, lanbo Wang <drram...@gmail.com<mail > > to:drram...@gmail.com>> wrote: > > Dear Kersten, > > > > Thanks a lot, it's really help. I have another question, after I got > > results that two group have significant, then how could I get > > direction? > > > > Thanks, > > Lanbo > > > > On Tue, Mar 13, 2018 at 5:40 PM, Diers, Kersten /DZNE <Kersten.Diers@ > > dzne.de<mailto:kersten.di...@dzne.de>> wrote: > > Hello, > > > > On Di, 2018-03-13 at 21:48 +0100, lanbo Wang wrote: > > > > > > Dear Kersten, > > > > > > Thanks, I find it. And I have other questions: > > > 1. The intercepts all set as one, so in this model it doesn't > > > separate different subjects, or can say no individual subject > > > change > > > rate? > > If I understood correctly, the question is whether or not we can get > > estimates for individual slopes across time? > > > > If so, then yes, that's possible - but I'd have to run an analysis > > myself and look up how to do it exactly - I'll get back on this. > > > > > > > > 2. Should we set age according to different timepoint, or just use > > > baseline age? > > It's better to use age at baseline. > > > > One of the nice things of the LME is that it can separate the cross- > > sectional effect of age (at baseline) and the longitudinal effect of > > aging (=effect of time). So the aging effect is already incorporated > > within the 'time since baseline' variable, which of course should > > also > > be present in the model. > > > > Since it is difficult to estimate and interpret effects that are very > > redundant (such as time vs age at each timepoint), it's better to > > just > > use age at baseline for the other regressor. > > > > Best regards, > > > > Kersten > > > > > > > > Thanks, > > > Lanbo > > > > > > On Tue, Mar 13, 2018 at 3:34 PM, Diers, Kersten /DZNE > > > <Kersten.Diers@ > > > dzne.de<http://dzne.de><mailto:kersten.di...@dzne.de<mailto:Kersten > > > .di...@dzne.de>>> wrote: > > > Hello Lanbo, > > > > > > the univariate example data can actually be downloaded from the LME > > > tutorial website: > > > > > > Search for: "An optional sample dataset which can be used to become > > > familiar with the LME Matlab tools can be found here". The linked > > > tar.gz archive contains two folders, one for the univariate and one > > > for > > > the mass-univariate example data. > > > > > > Best regards, > > > > > > Kersten > > > > > > On Di, 2018-03-13 at 13:38 +0100, lanbo Wang wrote: > > > > > > > > > > > > Dear Experts, > > > > > > > > Hi, > > > > There is no example detail on website of LME tutorial. I have > > > > some > > > > question about it. > > > > Could you send me the table of ADNI univariate example data? > > > > > > > > > > > > Thanks, > > > > Lanbo > > > _______________________________________________ > > > Freesurfer mailing list > > > Freesurfer@nmr.mgh.harvard.edu<mailto:freesur...@nmr.mgh.harvard.ed > > > u><mailto:Freesurfer@nmr.mgh.harvard.edu<mailto:freesur...@nmr.mgh. > > > harvard.edu>> > > > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer > > > > > > > > > The information in this e-mail is intended only for the person to > > > whom it is > > > addressed. If you believe this e-mail was sent to you in error and > > > the e-mail > > > contains patient information, please contact the Partners > > > Compliance > > > HelpLine at > > > http://www.partners.org/complianceline . If the e-mail was sent to > > > you in error > > > but does not contain patient information, please contact the sender > > > and properly > > > dispose of the e-mail. > > > > > > > > _______________________________________________ > > Freesurfer mailing list > > Freesurfer@nmr.mgh.harvard.edu<mailto:Freesurfer@nmr.mgh.harvard.edu> > > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer > > > > > > _______________________________________________ > Freesurfer mailing list > Freesurfer@nmr.mgh.harvard.edu > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer >
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.