Anicca-vijja

Archiv für die Kategorie „Methodology“

At first, if you are not familiar with R, you need to install R from CRAN and (under Win32) Tinn-R. Tinn-R enables to control R via script. If you use Linux, just install R via your distribution of choice and then there is an add-on for Emacs to control R (ESS). The next is to start R via Tinn-R or Emacs. Don’t forget to choose you ‘hotkeys‘ in Tinn-R (or any other editor of your choice to control R).

For multiple imputation, I choose the R-package ‘mice‘ from van Buuren et al. You have to install it manually. There are also other packages that deal with it, see:

library.search("imputation")

If you are not familiar with the R-style to formulate linear models, start with

?lm
example(lm)

or read the usual intros and manuals that are linked via CRAN (or the contributed documentations) – otherwise search on the internet with your search machine of choice with the add-on ‘cran’ like ‘mulitple imputation cran’ or ‘missing data cran’, etc.

The folllowing are excerpts from the man-pages of ‘mice’-package commandos.

?read.table       # import of data - see tutorials and intros to R
                  # of 'how to import data'
library(help=mice)# what is inside package 'mice'?
library(mice)     # load library for MI
data(nhanes)      # use data from 'mice'-package
str(nhanes)
nhanes            # show data
?mice             # produce Multivariate Imputation by Chained Equations
imp <- mice(nhanes)
imp
str(imp)
?lm.mids          # Performs repeated linear regression
                  # on multiply imputed data set
lm.mids           # R-source code of 'lm.mids'
fit <- lm.mids(bmi~hyp+chl,data=imp)
fit
summary(fit)
str(fit)
?pool             #
pool(fit)         # pool results
summary(pool(fit))# better output
pool              # R-source code of 'pool'

That’s all. Multiple imputation (ordinary ANOVA) is quite easy to peform in R.

Slashdot reported on Sunday Sept 20 the outcome of a new study on the apropriateness of fMRI research which can be read in much more detail on Wired. This study can be seen as a serious warning to ‘blindly’ believe results of fMRI research without doing proper quality management. Craig M. Bennett and colleagues realized an experiment which is quite close to a gag of ‘Monthy Python’s Flying Circus.’ However, the experiment is serious. They scanned a mature, but dead atlantic salmon. The experimental task is worth to be mentioned in its original language:

“The salmon was shown a series of photographs depicting human individuals in social situations with a specified emotional valence. The salmon was asked to determine what emotion the individual in the photo must have been experiencing.”

Several photos of human beings were shown to the salmon and reactions were measured. Reviewing the results – surprise, surprise – they showed clear activity in the dead salmon’s brain:

“Several active voxels were discovered in a cluster located within the salmon’s brain cavity.”

The dead salmon and its emotional reactions (Photo courtesy Craig Bennett)

The dead salmon’s ‘emotional reactions’ look quite impressive at it can be seen in the figure right next to this text. It seems the (dead) salmon reacted to photos of human beings. Unfortunately, the study was turned down by several publications. However, a poster is available. Thinking about the mass of research in the field of fMRI, it seems a little bit confusing what we can believe and what not. What is needed is a good control of random but significant voxels. Additionally, we should not take brain research too serious. But false positives should be taken very seriously. At last, I want to point also to an older posting on this blog about the works of Ed Vul and his colleagues at the MIT. Further infos dedicated to fMRI and the dead salmon are available on Craig Bennett’s personal blog.

References:

Bennett CM, Baird AA, Miller MB, and Wolford GL. (submitted) Neural Correlates of Interspecies Perspective Taking in the Post-Mortem Atlantic Salmon: An Argument For Multiple Comparisons Correction.

Much has been said about induction. I like the following very old citation:

“Yet, in fact, as I shall show here with very good reasons, the properties of the numbers known today have been mostly discovered by observation, and discovered long before their truth has been confirmed by rigid demonstrations. There are even many properties of the numbers with which we are well acquainted, but which we are not yet able to prove; only observations have led us to their knowledge.”

[Euler, Opera Omnia, ser. 1, vol. 2, pp.459, Specimen de usu observationum in mathesi pura]

There were and there are still many arguments about induction and whether induction is possible or not. However, I think this is the wrong question, because it is quite clear that there is nothing which can explain everything. We live in a relative world and if we think we work with mental or mental-somatic models and not with reality itself. Furthermore, anything is part of a context and this leads to the important question of:

• What from our present context (which elements) has influence on our topic of interest and what has not?
• How can we describe, understand, explain, forecast, and change these influences?

One of the best short outlines of induction can be found in Jaynes (2003). There, the author also performs a dedicated criticism of Popper’s argument against induction. In short, Jaynes argument that one should work on realistic problems and not just in abstract theory like Popper did. This reveals that one should not compare one theory with every possible imaginable theory, because then no solution at all is possible. This becomes clear at once. Moreover, what is needed is a critical comparison of all available and existent theories. Then, induction makes sense. This process of drawing inferences can be labeled as ‘plausible reasoning’ which was elaborated by George Polya (1954).

Technically, Bayesian statistics gives much respect to plausible reasoning and consistent argumentation. The Bayes Theorem allows to update the formula in accordance to all known information. In case of new information the formula is updated. By application of the Bayes Theorem it is possible to compare different theoretical approaches and explanations. Then, induction makes sense, because the best of all available theories or hypotheses can be chosen. But ‘best’ is always ‘relatively best’ and not an absolute term.

The same can be learned from the teaching of the Buddha: There is relative reality and there is absolute reality (nibbana). However, as long as we live in the realm of relative reality and in the field of sensual experiences of mind and body, it does not make sense to search for anything absolute with is not a product of cause and effect. So we can learn much from Buddhist teaching for scientific purposes.  But we have to remember that the teaching of the Buddha is a practical path of direct experience. It is not an academic and intellectual discussion. It is meant to be applied directly in life. Besides that, we can also mention that the Buddhist teaching involves a very complex system of logic. Some of these ideas actually found their way into psychotherapy and systemic structural sculpturing as it is shown very successfully by Varga von Kibed and Insa Sparrer (2009).

References:

Jaynes, E.T. (2003). Probability Theory. The logic of science (Edited by G.L. Bretthorst). Cambridge University Press.

Polya, G. (1990). Mathematics and plausible reasoning. Volume I: Induction and analogy in mathematics. (First print 1954). Princeton University Press.

Polyga, G. (1990). Mathematics and plausible reasoning. Volume II: Patterns of plausible inference. (First print 1954). Princeton University Press.

Sparrer, I. & Kibed, Varga von (2009). Ganz im Gegenteil: Tetralemmaarbeit und andere Grundformen Systemischer Strukturaufstellungen – für Querdenker und solche, die es werden wollen (6th edition). Heidelberg: Carl-Auer-Systeme.

This is the ‘first’ blog and I knew at once what it should be about: The new findings that heavily criticize the general findings of neuroimaging studies. A summary in German can be found in one of the lastes issues of ‘Gehirn und Geist’ (04-09, p.69) by Prof. Fritz Strack. Let’s look what is problematic besides the fact that neuroimaging studies do permanently perform a false logical conclusion, because they attribute characteristics of ‘whole’ human beings solely to their brains and neglect the rest of the body (and the mind)??

Telepolis published a short summary of the research of Edward Vul, a PhD student at the MIT in cambridge (USA) on ‘voodoo correlations’ and ‘non-independence error.’  In a first article he and colleagues write about artifical exaggerated correlations between voxels and external variables. These correlations are sometimes higher than the reliabilities which is far from being realistic! This was found not only in one, but in many studies that were re-analyzed … and not to mention all those articles that were published in high rated journals (but not re-analyzed). Additionally, in another article Edward Vul and colleagues concentrate on the selection of the analyzed brain areas (voxels). These were not independent from the behavior measurements that were done at the very same time. This means that from thousands of voxels those were selected for further analyses that showed a maximum correlation with the external behaviour measurements. Of course all further statistical relationships were high – but are they real or a methodological artefact? – that’s another story to be told.

Another article from Sirotin & Das (Nature 457) questions one of the most basic assumptions of neuroimaging studies: the covariation of local brain activity and blood flow. In an experimental study with animals the authors compared the neuronal firing of cells (direct measurement of cell firing) with the intensity of the blood flow. The results showed that both parameters did not correlate continuously with each other.

So what now? At first we should congratulate the cited researchers for their courage and the journals also for their courage to publish these important findings. What is needed in neurobiology and neuroimaging studies is a methodological discussion. Results should not or even must not be discussed without giving importance to methodological questions. Other disciplines like psychology or sociology have regular discussions of this kind (although sometimes the discipline does not considers change, e.g. ask a German psychologist why nobody does  Bayesian statistics?). I don’t like the last sentence of so many articles like ‘further research is needed,’ but this time I think – yes – these findings have to be understood properly and the experiments that led to their results have to be repeated.

References:

Sirotin, B. Y. & Das, A. (2009). Anticipatory Haemodynamic Signals in Sensory Cortex not Predicted by Local Neuronal Activity. Nature 457, 475–479.

Vul, E., Harris, C., Winkielman, P. & Pashler, H. (in press/ 2009). Voodoo correlations in social neuroscience. Perspectives on Psychological Science.

Vul, E. & Kanwisher, N. (in press/ 2009). Begging the question: The non-independence error in fMRI data analysis. To appear in Hanson, S. & Bunzl, M. (Eds.). Foundations and Philosophy for Neuroimaging.