%PDF-1.7
1 0 obj
<< /Type /Catalog
/Outlines 2 0 R
/Pages 3 0 R >>
endobj
2 0 obj
<< /Type /Outlines /Count 0 >>
endobj
3 0 obj
<< /Type /Pages
/Kids [6 0 R
16 0 R
18 0 R
20 0 R
22 0 R
30 0 R
]
/Count 6
/Resources <<
/ProcSet 4 0 R
/Font <<
/F1 8 0 R
/F2 9 0 R
>>
>>
/MediaBox [0.000 0.000 595.280 841.890]
>>
endobj
4 0 obj
[/PDF /Text ]
endobj
5 0 obj
<<
/Producer (ţ˙dompdf 1.0.2 + CPDF)
/CreationDate (D:20220807183317+00'00')
/ModDate (D:20220807183317+00'00')
>>
endobj
6 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Annots [ 10 0 R 12 0 R 14 0 R ]
/Contents 7 0 R
>>
endobj
7 0 obj
<<
/Length 5894 >>
stream
0.000 0.000 0.000 rg
BT 35.000 736.932 Td /F1 21.0 Tf [(\(20\) Neuromythology – What Happens When You Violate)] TJ ET
BT 35.000 711.984 Td /F1 21.0 Tf [(Statistical Premises)] TJ ET
BT 35.000 672.110 Td /F2 14.0 Tf [(Description)] TJ ET
BT 35.000 631.330 Td /F1 12.0 Tf [(A moment ago \(this was first published in 2016\), probably the biggest publicity bomb I have seen in a long time)] TJ ET
BT 35.000 617.074 Td /F1 12.0 Tf [(exploded: A group of Swedish authors, together with an English statistician, have published a )] TJ ET
0.000 0.000 0.800 rg
BT 487.628 617.074 Td /F1 12.0 Tf [(huge simulation )] TJ ET
0.000 0.000 0.800 RG
0.6 w 0 J [ ] 0 d
487.628 614.494 m 567.632 614.494 l S
BT 35.000 602.818 Td /F1 12.0 Tf [(study)] TJ ET
0.6 w 0 J [ ] 0 d
35.000 600.238 m 61.004 600.238 l S
0.000 0.000 0.000 rg
BT 61.004 602.818 Td /F1 12.0 Tf [(. It shows that possibly up to 70% or more of the total of more than 40,000 published neuroscience studies)] TJ ET
BT 35.000 588.562 Td /F1 12.0 Tf [(that have used functional magnetic resonance spectroscopy \(fMRI\) have produced useless results and therefore)] TJ ET
BT 35.000 574.306 Td /F1 12.0 Tf [(actually belong to be cleaned out or replicated [1].)] TJ ET
BT 35.000 548.050 Td /F1 12.0 Tf [(This seems to me to be one of the biggest scientific collective scandals of recent times. And one can learn a lot)] TJ ET
BT 35.000 533.794 Td /F1 12.0 Tf [(about statistics from it. But in order.)] TJ ET
BT 35.000 507.538 Td /F1 12.0 Tf [(Before we turn to this study: This is not to say that MRI methodology is wrong and that so-called structural)] TJ ET
BT 35.000 493.282 Td /F1 12.0 Tf [(imaging methods are useless. It is solely about statements about spatial spread of activity in functional magnetic)] TJ ET
BT 35.000 479.026 Td /F1 12.0 Tf [(imaging. But even that is a huge chunk. Follow me.)] TJ ET
BT 35.000 452.770 Td /F2 12.0 Tf [(What happened? )] TJ ET
0.000 0.000 0.800 rg
BT 35.000 426.514 Td /F2 12.0 Tf [(Functional magnetic resonance spectroscopy)] TJ ET
0.6 w 0 J [ ] 0 d
35.000 423.934 m 263.312 423.934 l S
0.000 0.000 0.000 rg
BT 263.312 426.514 Td /F1 12.0 Tf [( or imaging \(fMRI\) is very popular as a research method. The)] TJ ET
BT 35.000 412.258 Td /F1 12.0 Tf [(technique is based on the fact that hydrogen atoms – which are found everywhere – can be aligned via strong)] TJ ET
BT 35.000 398.002 Td /F1 12.0 Tf [(external magnetic fields. By simultaneously applying and scanning electromagnetic high-frequency waves, the)] TJ ET
BT 35.000 383.746 Td /F1 12.0 Tf [(atoms can be localized. Depending on which frequency is chosen, one can also make different types of structures)] TJ ET
BT 35.000 369.490 Td /F1 12.0 Tf [(or molecules visible. This can be used, for example, to determine the difference between blood whose red blood)] TJ ET
BT 35.000 355.234 Td /F1 12.0 Tf [(cells are saturated with oxygen and that which has given up its oxygen.)] TJ ET
BT 35.000 328.978 Td /F1 12.0 Tf [(This so-called BOLD signal, short for „blood oxygenation level dependent signal“, can be used to deduce how)] TJ ET
BT 35.000 314.722 Td /F1 12.0 Tf [(high the metabolic activity is in a certain area of the body, e.g. in an area of the brain. An increase indicates)] TJ ET
BT 35.000 300.466 Td /F1 12.0 Tf [(increased oxygen consumption, increased blood supply, increased metabolism and thus increased activity in an)] TJ ET
BT 35.000 286.210 Td /F1 12.0 Tf [(area of the brain. A decrease indicates the opposite.)] TJ ET
BT 35.000 259.954 Td /F1 12.0 Tf [(Now, in order to see anything at all in a functional magnetic resonance imaging \(i.e., imaging\) study, one must of)] TJ ET
BT 35.000 245.698 Td /F1 12.0 Tf [(course create differences between experimental and control conditions. This is usually done by having people in)] TJ ET
BT 35.000 231.442 Td /F1 12.0 Tf [(the MRI tube do different tasks in a specific sequence, called blocks. For example, they have to read a text on a)] TJ ET
BT 35.000 217.186 Td /F1 12.0 Tf [(screen, or think of something specific, or recite a memorized poem in their mind; in another block lie down and)] TJ ET
BT 35.000 202.930 Td /F1 12.0 Tf [(relax instead. This happens in fixed sequences. Thus, one can compare the sequences in which something defined)] TJ ET
BT 35.000 188.674 Td /F1 12.0 Tf [(happens in the mind with those in which calmness reigns.)] TJ ET
BT 35.000 162.418 Td /F1 12.0 Tf [(The difference in the signals is then used to calculate the difference in the activation levels of the two conditions)] TJ ET
BT 35.000 148.162 Td /F1 12.0 Tf [(in specific areas of the brain and to make deductions about which areas of the brain are responsible for which)] TJ ET
BT 35.000 133.906 Td /F1 12.0 Tf [(functions. In addition, such conditions are often compared with situations in which control subjects are only)] TJ ET
BT 35.000 119.650 Td /F1 12.0 Tf [(measured \(„scanned“ is the neuro jargon\) without anything happening.)] TJ ET
BT 35.000 93.394 Td /F1 12.0 Tf [(To be clear, let me add: One can also use the method to visualize anatomical structures or to record the)] TJ ET
BT 35.000 79.138 Td /F1 12.0 Tf [(functionality of connections within the brain. These two applications are not covered by the study discussed here,)] TJ ET
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 1)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
8 0 obj
<< /Type /Font
/Subtype /Type1
/Name /F1
/BaseFont /Times-Roman
/Encoding /WinAnsiEncoding
>>
endobj
9 0 obj
<< /Type /Font
/Subtype /Type1
/Name /F2
/BaseFont /Times-Bold
/Encoding /WinAnsiEncoding
>>
endobj
10 0 obj
<< /Type /Annot
/Subtype /Link
/A 11 0 R
/Border [0 0 0]
/H /I
/Rect [ 487.6280 615.9937 567.6320 627.8737 ]
>>
endobj
11 0 obj
<< /Type /Action
/S /URI
/URI (http://www.pnas.org/content/113/28/7900)
>>
endobj
12 0 obj
<< /Type /Annot
/Subtype /Link
/A 13 0 R
/Border [0 0 0]
/H /I
/Rect [ 35.0000 601.7377 61.0040 613.6177 ]
>>
endobj
13 0 obj
<< /Type /Action
/S /URI
/URI (http://www.pnas.org/content/113/28/7900)
>>
endobj
14 0 obj
<< /Type /Annot
/Subtype /Link
/A 15 0 R
/Border [0 0 0]
/H /I
/Rect [ 35.0000 425.4337 263.3120 437.3137 ]
>>
endobj
15 0 obj
<< /Type /Action
/S /URI
/URI (https://en.wikipedia.org/wiki/Magnetic_resonance_imaging)
>>
endobj
16 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Contents 17 0 R
>>
endobj
17 0 obj
<<
/Length 6915 >>
stream
0.6 w 0 J [ ] 0 d
0.000 0.000 0.000 rg
BT 35.000 760.485 Td /F1 12.0 Tf [(but only the activation of brain areas as a result of activity change due to experimental instruction.)] TJ ET
BT 35.000 734.229 Td /F2 12.0 Tf [(Now the signals that arise from the measurement)] TJ ET
BT 285.956 734.229 Td /F1 12.0 Tf [(, it is easy to imagine even as a layman, have to run through a)] TJ ET
BT 35.000 719.973 Td /F1 12.0 Tf [(series of complex mathematical and statistical procedures before the pretty colourful pictures we admire in the)] TJ ET
BT 35.000 705.717 Td /F1 12.0 Tf [(publications and glossy brochures emerge at the end. In which experts then explain that the brain „lights up“)] TJ ET
BT 35.000 691.461 Td /F1 12.0 Tf [(when a person does this or that. This „lighting up“ refers to the false colour representation of the increase or)] TJ ET
BT 35.000 677.205 Td /F1 12.0 Tf [(decrease of the BOLD signal in certain areas, which has been statistically isolated as a significant effect from the)] TJ ET
BT 35.000 662.949 Td /F1 12.0 Tf [(background noise. It is this statistical filtering procedure that then leads to the colouring – which is, after all,)] TJ ET
BT 35.000 648.693 Td /F1 12.0 Tf [(nothing more than the pictorial implementation of statistically significant signal detection – that was examined in)] TJ ET
BT 35.000 634.437 Td /F1 12.0 Tf [(this publication and found to be unreliable in the vast majority of cases. Why?)] TJ ET
BT 35.000 608.181 Td /F2 12.0 Tf [(This statistical filtering procedure is unreliable – why? )] TJ ET
BT 35.000 581.925 Td /F1 12.0 Tf [(Signal detection in an fMRI study is essentially a two-step process. The first step is to pick up the raw signals)] TJ ET
BT 35.000 567.669 Td /F1 12.0 Tf [(from the pulsed application of the magnetic fields and their deactivation, and to sample them with a high-)] TJ ET
BT 35.000 553.413 Td /F1 12.0 Tf [(frequency electromagnetic field. This provides the raw data about changes in the activity of the blood supply in)] TJ ET
BT 35.000 539.157 Td /F1 12.0 Tf [(the brain, i.e. about the oxygen saturation of the blood and the change in the distribution of the blood in the brain.)] TJ ET
BT 35.000 524.901 Td /F1 12.0 Tf [(Of course, as you can see immediately, this results in millions of data points that are determined in rapid)] TJ ET
BT 35.000 510.645 Td /F1 12.0 Tf [(succession and which, as such, are not usable in raw form.)] TJ ET
BT 35.000 484.389 Td /F2 12.0 Tf [(The second and crucial step)] TJ ET
BT 177.008 484.389 Td /F1 12.0 Tf [( is now the statistical discovery and summarization procedure. This is done by)] TJ ET
BT 35.000 470.133 Td /F1 12.0 Tf [(analysing the raw data with special programmes. The study discussed here examined the three most popular)] TJ ET
BT 35.000 455.877 Td /F1 12.0 Tf [(programmes. In order to understand how complex the whole thing is, one has to imagine that the fMRI signals)] TJ ET
BT 35.000 441.621 Td /F1 12.0 Tf [(are initially picked up at different points on the surface of the head and also originate from differently deep areas)] TJ ET
BT 35.000 427.365 Td /F1 12.0 Tf [(of the brain. We are therefore dealing with three-dimensional data points, which, analogously to the two-)] TJ ET
BT 35.000 413.109 Td /F1 12.0 Tf [(dimensional data points of a screen, where they are now known to all as „pixels“, are called voxels. Voxels are)] TJ ET
BT 35.000 398.853 Td /F1 12.0 Tf [(therefore three-dimensional pixels that originate from a defined location and vary in intensity. Since voxels cover)] TJ ET
BT 35.000 384.597 Td /F1 12.0 Tf [(just 1 cubic millimetre, the image that would emerge would be extremely confusing if one had to analyse them all)] TJ ET
BT 35.000 370.341 Td /F1 12.0 Tf [(individually.)] TJ ET
BT 35.000 344.085 Td /F1 12.0 Tf [(For this reason, one usually groups the voxels into larger areas. This is done by making assumptions about how)] TJ ET
BT 35.000 329.829 Td /F1 12.0 Tf [(the activity of neighbouring points relate to each other when a larger functional brain area, say the language)] TJ ET
BT 35.000 315.573 Td /F1 12.0 Tf [(centre in generating mental monologue, is activated. This happens via so-called autocorrelation functions of a)] TJ ET
BT 35.000 301.317 Td /F1 12.0 Tf [(spatial nature. We are all familiar with autocorrelation functions of a temporal nature: If the weather is very nice)] TJ ET
BT 35.000 287.061 Td /F1 12.0 Tf [(today, the probability that it will also be very nice tomorrow is higher than if it has already been nice for two)] TJ ET
BT 35.000 272.805 Td /F1 12.0 Tf [(weeks. Because then the probability that tomorrow will be worse is gradually higher, and vice versa.)] TJ ET
BT 35.000 246.549 Td /F1 12.0 Tf [(Analogous to such a temporal autocorrelation, one can also imagine a spatial one: Depending on how high the)] TJ ET
BT 35.000 232.293 Td /F1 12.0 Tf [(activity is at a point in the voxel universe, the probability that a neighbouring voxel belongs to a functional unit)] TJ ET
BT 35.000 218.037 Td /F1 12.0 Tf [(will be higher or lower. In the early days of programme development for the analysis of such data, relatively little)] TJ ET
BT 35.000 203.781 Td /F1 12.0 Tf [(information was available. So a reasonable, but as it now turns out wrong, assumption was made: namely, that the)] TJ ET
BT 35.000 189.525 Td /F1 12.0 Tf [(spatial autocorrelation function behaves as a spatially propagating Gaussian curve or normal distribution.)] TJ ET
BT 35.000 163.269 Td /F2 12.0 Tf [(Control data)] TJ ET
BT 35.000 137.013 Td /F1 12.0 Tf [(Now there are thousands of data sets of people measured by MRI scanners for control purposes, so to speak,)] TJ ET
BT 35.000 122.757 Td /F1 12.0 Tf [(without any tasks, and thanks to the possibility of open platforms, these data are made openly available to)] TJ ET
BT 35.000 108.501 Td /F1 12.0 Tf [(scientists. Anyone can download it and make analyses with it. Taking advantage of this opportunity, the scientists)] TJ ET
BT 35.000 94.245 Td /F1 12.0 Tf [(have recalculated data from nearly 500 healthy people from different regions of the world, measured in a scanner)] TJ ET
BT 35.000 79.989 Td /F1 12.0 Tf [(without any task, using simulated analysis methods by applying the three most popular analysis software)] TJ ET
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 2)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
18 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Contents 19 0 R
>>
endobj
19 0 obj
<<
/Length 6257 >>
stream
0.6 w 0 J [ ] 0 d
0.000 0.000 0.000 rg
BT 35.000 760.485 Td /F1 12.0 Tf [(packages to them.)] TJ ET
BT 35.000 734.229 Td /F1 12.0 Tf [(In total, they tested 192 combinations of possible settings in more than 3 million simulation calculations. So,)] TJ ET
BT 35.000 719.973 Td /F1 12.0 Tf [(somewhat simplistically, the scientists have pretended that the data from these 500 people came from real fMRI)] TJ ET
BT 35.000 705.717 Td /F1 12.0 Tf [(experiments with on and off blocks of specific tasks or questions. But it is clear that this was not the case because)] TJ ET
BT 35.000 691.461 Td /F1 12.0 Tf [(the data was control data.)] TJ ET
BT 35.000 665.205 Td /F2 12.0 Tf [(One would expect in such a procedure)] TJ ET
BT 230.336 665.205 Td /F1 12.0 Tf [( that a certain number of false positives would always be found, i.e.)] TJ ET
BT 35.000 650.949 Td /F1 12.0 Tf [(results where the statistics say: „Hurrah, we have found a significant effect“, but where in fact there is no effect.)] TJ ET
BT 35.000 636.693 Td /F1 12.0 Tf [(This so-called error of the first kind or alpha error is controlled by the nominal significance level, which can be)] TJ ET
BT 35.000 622.437 Td /F1 12.0 Tf [(set by convention and which is often 5% \(p = 0.05\), but in the case of fMRI studies is often set lower from the)] TJ ET
BT 35.000 608.181 Td /F1 12.0 Tf [(outset, namely at 1% \(p = 0.01\) or 0.1% \(p = 0.001\). This is because this alpha error indicates how often we make)] TJ ET
BT 35.000 593.925 Td /F1 12.0 Tf [(a mistake when we claim an effect, although there is none. At 5% level of alpha error, we make such an error 5)] TJ ET
BT 35.000 579.669 Td /F1 12.0 Tf [(times out of 100. At a 1% level of alpha error in one out of 100 cases. And at a 1 per thousand level in one out of)] TJ ET
BT 35.000 565.413 Td /F1 12.0 Tf [(1,000 cases.)] TJ ET
BT 35.000 539.157 Td /F1 12.0 Tf [(Now, of course, if we apply many statistical tests in parallel to the same set of data, this error multiplies because)] TJ ET
BT 35.000 524.901 Td /F1 12.0 Tf [(we get the same probability of making a mistake again in each test if we make a factual claim that is not in fact)] TJ ET
BT 35.000 510.645 Td /F1 12.0 Tf [(true. The nominal probability of error of p = 0.05, i.e. 5%, then becomes the probability of error of p = 0.1 or)] TJ ET
BT 35.000 496.389 Td /F1 12.0 Tf [(10% for two simultaneous tests We therefore make twice as many errors. So, in order to comply with the nominal)] TJ ET
BT 35.000 482.133 Td /F1 12.0 Tf [(probability of 5%, the individual probabilities must be set to p = 0.025 for two simultaneous tests on the same)] TJ ET
BT 35.000 467.877 Td /F1 12.0 Tf [(data set, so that the joint error probability p = 0.05 remains. This is called „correction for multiple testing“.)] TJ ET
BT 35.000 441.621 Td /F1 12.0 Tf [(Because a large number of tests are made at once in the fMRI evaluation packages, one sets the detection)] TJ ET
BT 35.000 427.365 Td /F1 12.0 Tf [(threshold there for what one is prepared to accept as a significant signal right from the start at p = 0.01 \(i.e. an)] TJ ET
BT 35.000 413.109 Td /F1 12.0 Tf [(error probability adjusted for 5 simultaneous tests\) or even at p = 0.001. This is an error probability adjusted for)] TJ ET
BT 35.000 398.853 Td /F1 12.0 Tf [(50 simultaneous tests and thus adheres to the nominal error level of 5% for 50 tests. This correction is already)] TJ ET
BT 35.000 384.597 Td /F1 12.0 Tf [(built into the software packages studied; thus, the problem found is not related to it.)] TJ ET
BT 35.000 358.341 Td /F1 12.0 Tf [(All these parameter settings were used in the study conducted here. At the same time, scenarios were run that are)] TJ ET
BT 35.000 344.085 Td /F1 12.0 Tf [(common practice in the real world of fMRI research, i.e. that one takes, for example, 8 mm clusters and merges)] TJ ET
BT 35.000 329.829 Td /F1 12.0 Tf [(the neighbouring voxels with a detection threshold of p = 0.001, which seems to be quite reasonable.)] TJ ET
BT 35.000 303.573 Td /F1 12.0 Tf [(Then, in complex simulation calculations, all sorts of putative experimental comparisons were superimposed on)] TJ ET
BT 35.000 289.317 Td /F1 12.0 Tf [(this control data, and it was documented how often the various software packages make significant „discoveries“,)] TJ ET
BT 35.000 275.061 Td /F1 12.0 Tf [(even though it is known that there are no signals hidden in the data at all.)] TJ ET
BT 35.000 248.805 Td /F1 12.0 Tf [(When clusters are formed, i.e. voxels are combined into larger areas, false positives, i.e. signals where there are)] TJ ET
BT 35.000 234.549 Td /F1 12.0 Tf [(none, are found in up to 50% of analyses. Or to put it another way: some software packages detect signals with a)] TJ ET
BT 35.000 220.293 Td /F1 12.0 Tf [(50% probability of error where there are no signals at all. Put another way, in 50 out of 100 studies, the analysis)] TJ ET
BT 35.000 206.037 Td /F1 12.0 Tf [(says „there is a significant effect here“ where there is no effect at all.)] TJ ET
BT 35.000 179.781 Td /F1 12.0 Tf [(When the cluster size is smaller and the threshold for combining voxels into clusters is higher, the probability of)] TJ ET
BT 35.000 165.525 Td /F1 12.0 Tf [(error approaches the 5% nominal significance threshold. For voxel-based analysis, i.e. when one makes no)] TJ ET
BT 35.000 151.269 Td /F1 12.0 Tf [(assumptions about the correlation of voxel activities and accepts that one has to interpret a chaotic image of many)] TJ ET
BT 35.000 137.013 Td /F1 12.0 Tf [(voxels, the analysis remains close to the error probability of 5% for almost all software packages.)] TJ ET
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 3)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
20 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Contents 21 0 R
>>
endobj
21 0 obj
<<
/Length 6908 >>
stream
0.6 w 0 J [ ] 0 d
0.000 0.000 0.000 rg
BT 35.000 748.485 Td /F1 12.0 Tf [(And for the so-called non-parametric method, i.e. a statistical analysis based on a simulation calculation in which)] TJ ET
BT 35.000 734.229 Td /F1 12.0 Tf [(the probability is not derived from an underlying and assumed distribution, but from an actual simulation)] TJ ET
BT 35.000 719.973 Td /F1 12.0 Tf [(calculation based on the available data, the nominal significance values are always preserved.)] TJ ET
BT 35.000 693.717 Td /F2 12.0 Tf [(The problem is, however: )] TJ ET
BT 171.656 693.717 Td /F2 12.0 Tf [(The software packages are used because)] TJ ET
BT 377.300 693.717 Td /F1 12.0 Tf [( one does not want to do a laborious)] TJ ET
BT 35.000 679.461 Td /F1 12.0 Tf [(interpretation of a voxel-based evaluation oneself, but delegate it to the computer, and because one does not want)] TJ ET
BT 35.000 665.205 Td /F1 12.0 Tf [(to carry out weeks of simulation calculations to determine the true probability. In addition, signal noise or)] TJ ET
BT 35.000 650.949 Td /F1 12.0 Tf [(artefacts, such as those caused by movements, would be too much of a factor in a voxel-based evaluation. So one)] TJ ET
BT 35.000 636.693 Td /F1 12.0 Tf [(tries to find supposedly more robust quantities, precisely those clusters, which one then tests.)] TJ ET
BT 35.000 610.437 Td /F1 12.0 Tf [(For a very common scenario, the 8 mm clusters described above with an apparently conservative detection)] TJ ET
BT 35.000 596.181 Td /F1 12.0 Tf [(threshold of p = 0.001 from voxel to voxel before one is inclined to consider a cluster „significantly activated“ or)] TJ ET
BT 35.000 581.925 Td /F1 12.0 Tf [(„significantly inactivated“, the values look grim: the error frequency rises up to 90% depending on the program,)] TJ ET
BT 35.000 567.669 Td /F1 12.0 Tf [(and )] TJ ET
BT 55.328 567.669 Td /F2 12.0 Tf [(a 70% error probability across the literature is a robust estimate)] TJ ET
BT 386.612 567.669 Td /F1 12.0 Tf [(.)] TJ ET
BT 35.000 541.413 Td /F1 12.0 Tf [(Only a non-parametric simulation statistic would not make excessive errors here, either. However, this one is)] TJ ET
BT 35.000 527.157 Td /F1 12.0 Tf [(almost non-existent. Incidentally, the same problem was also found for active data from real studies. Here, too, a)] TJ ET
BT 35.000 512.901 Td /F1 12.0 Tf [(so-called inflation of the alpha error or a far too frequent detection of effects where there are none at all has been)] TJ ET
BT 35.000 498.645 Td /F1 12.0 Tf [(demonstrated.)] TJ ET
BT 35.000 472.389 Td /F2 12.0 Tf [(Where does the problem come from? )] TJ ET
BT 35.000 446.133 Td /F1 12.0 Tf [(You can use this example to study the importance of preconditions for the validity of statistics. First, the software)] TJ ET
BT 35.000 431.877 Td /F1 12.0 Tf [(packages and the users make assumptions about the interrelation of the voxels via spatial autocorrelation)] TJ ET
BT 35.000 417.621 Td /F1 12.0 Tf [(functions, as I described above. Users also choose the size of the areas to be studied, and the smoothing used in)] TJ ET
BT 35.000 403.365 Td /F1 12.0 Tf [(the process. These original assumptions were reasonable to begin with, but they were made at a time when there)] TJ ET
BT 35.000 389.109 Td /F1 12.0 Tf [(was relatively little data. No one checked them. Until now. And lo and behold, precisely this central assumption)] TJ ET
BT 35.000 374.853 Td /F1 12.0 Tf [(describing the mathematical relationship of neighbouring voxels was wrong. So: back to the books; modify)] TJ ET
BT 35.000 360.597 Td /F1 12.0 Tf [(software programs, implement new autocorrelation functions closer to empirical reality. And recalculate.)] TJ ET
BT 35.000 334.341 Td /F1 12.0 Tf [(Other assumptions have to do with assuming statistical distributions for the data. This is something that is done)] TJ ET
BT 35.000 320.085 Td /F1 12.0 Tf [(often. So the inference procedures involved are called „parametric statistics“ because you assume a known)] TJ ET
BT 35.000 305.829 Td /F1 12.0 Tf [(distribution for the data. You can normalize the known distribution. One then interprets the area under the curve)] TJ ET
BT 35.000 291.573 Td /F1 12.0 Tf [(as „1“. If you then plot a value somewhere on the axis and calculate the area behind it, you can interpret this area)] TJ ET
BT 35.000 277.317 Td /F1 12.0 Tf [(fraction of 1 as a probability.)] TJ ET
BT 35.000 251.061 Td /F1 12.0 Tf [(So, for example, more than 95% or less than 5% of the area lies behind the axis value „2“ \(or „-2“\) of the)] TJ ET
BT 35.000 236.805 Td /F1 12.0 Tf [(standard normal distribution. Because the area is normalized to „1“, this can then be interpreted as a probability.)] TJ ET
BT 35.000 222.549 Td /F1 12.0 Tf [(So you can calculate error probabilities from a known distribution. A common distribution assumption is that)] TJ ET
BT 35.000 208.293 Td /F1 12.0 Tf [(based on normal distribution, but there are plenty of other statistical distribution curves where you can then)] TJ ET
BT 35.000 194.037 Td /F1 12.0 Tf [(calculate the area fraction of a standardized curve in the same way and thus determine the probability.)] TJ ET
BT 35.000 167.781 Td /F2 12.0 Tf [(On the other hand, we rarely know whether these assumptions are correct)] TJ ET
BT 414.284 167.781 Td /F1 12.0 Tf [(. Therefore, as this analysis shows,)] TJ ET
BT 35.000 153.525 Td /F1 12.0 Tf [(a non-parametric procedure, i.e. one that makes no distributional assumption about the data, is actually wiser. The)] TJ ET
BT 35.000 139.269 Td /F1 12.0 Tf [(discussion about this is already very old and well-known, as are the procedures [2]. We have used them on)] TJ ET
BT 35.000 125.013 Td /F1 12.0 Tf [(various occasions, especially in critical situations [3,4]. If you use such simulation or non-parametric statistics)] TJ ET
BT 35.000 110.757 Td /F1 12.0 Tf [(properly, you actually have to use the empirically found data. You let the computer generate new data sets, say)] TJ ET
BT 35.000 96.501 Td /F1 12.0 Tf [(10,000, that have similar characteristics, e.g. the same number of points with certain features, and then count how)] TJ ET
BT 35.000 82.245 Td /F1 12.0 Tf [(often the features that appear in the empirically found data also appear in the simulated data. If you then divide)] TJ ET
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 4)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
22 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Annots [ 24 0 R 26 0 R 28 0 R ]
/Contents 23 0 R
>>
endobj
23 0 obj
<<
/Length 6547 >>
stream
0.6 w 0 J [ ] 0 d
0.000 0.000 0.000 rg
BT 35.000 760.485 Td /F1 12.0 Tf [(the number of features that occurred empirically by the number of features found by chance, you have the true)] TJ ET
BT 35.000 746.229 Td /F1 12.0 Tf [(probability that the empirical finding could have occurred by chance.)] TJ ET
BT 35.000 719.973 Td /F1 12.0 Tf [(Of course, such simulations, often called Monte-Carlo analyses \(because Monte-Carlo is the big casino\) – or non-)] TJ ET
BT 35.000 705.717 Td /F1 12.0 Tf [(parametric analyses – are very costly. Even modern, fast computers often need weeks to perform complex)] TJ ET
BT 35.000 691.461 Td /F1 12.0 Tf [(analyses.)] TJ ET
BT 35.000 665.205 Td /F1 12.0 Tf [(Anyway, you can see from this example what happens when you violate statistical assumptions: You can no)] TJ ET
BT 35.000 650.949 Td /F1 12.0 Tf [(longer interpret probability values based on assumptions and feed the results to the rabbits. In this specific case, a)] TJ ET
BT 35.000 636.693 Td /F1 12.0 Tf [(huge literature of neuromythology has emerged. More than half, perhaps as many as 70%, of the approximately)] TJ ET
BT 35.000 622.437 Td /F1 12.0 Tf [(40,000 studies on fMRI methodology would actually have to be repeated or at least re-evaluated, the authors)] TJ ET
BT 35.000 608.181 Td /F1 12.0 Tf [(complain. If the data were publicly available, this would be feasible. Unfortunately, in most cases they are not.)] TJ ET
BT 35.000 593.925 Td /F1 12.0 Tf [(This is where the complaint of the neuroscientific community meets the call just made by psychologists for)] TJ ET
BT 35.000 579.669 Td /F1 12.0 Tf [(everything, but really everything, to be made publicly accessible, protocols, results, data [5]. The authors are)] TJ ET
BT 35.000 565.413 Td /F1 12.0 Tf [(calling for a moratorium: first do your homework, first work through the old problems, then do new studies. This)] TJ ET
BT 35.000 551.157 Td /F1 12.0 Tf [(will not work everywhere. Because in many cases study data was deleted after 5 years due to applicable laws.)] TJ ET
BT 35.000 524.901 Td /F2 12.0 Tf [(Fabled)] TJ ET
BT 35.000 498.645 Td /F1 12.0 Tf [(Now that’s beautifully silly, I think. One has to consider: Most major clinical units in hospitals and most major)] TJ ET
BT 35.000 484.389 Td /F1 12.0 Tf [(universities in Germany and the world maintain MRI scanners; the English Wikipedia estimates 25,000 scanners)] TJ ET
BT 35.000 470.133 Td /F1 12.0 Tf [(are in use worldwide. The problem with these devices is that once they have been put into operation, they are)] TJ ET
BT 35.000 455.877 Td /F1 12.0 Tf [(always connected to the power grid and thus generate high operating costs. You can’t simply switch them off like)] TJ ET
BT 35.000 441.621 Td /F1 12.0 Tf [(a computer, because that could damage the device, or switching them off and starting them up is itself a very)] TJ ET
BT 35.000 427.365 Td /F1 12.0 Tf [(complex and time-consuming process. That’s why these devices have to be kept in continuous use, so that their)] TJ ET
BT 35.000 413.109 Td /F1 12.0 Tf [(purchase, now worth several million euros, is worthwhile. That is why so many studies are done with them.)] TJ ET
BT 35.000 398.853 Td /F1 12.0 Tf [(Because whoever does studies pays for scanner time. No sooner does someone come up with an idea that seems)] TJ ET
BT 35.000 384.597 Td /F1 12.0 Tf [(reasonably clever – „let’s see which areas of the brain are active when you play music to people or show them)] TJ ET
BT 35.000 370.341 Td /F1 12.0 Tf [(pictures they don’t like“ – than he finds the money to get such a study funded, even in today’s climate.)] TJ ET
BT 65.000 332.085 Td /F1 12.0 Tf [(That brain research has a number of other problems has been noticed by others, )] TJ ET
0.000 0.000 0.800 rg
BT 449.240 332.085 Td /F1 12.0 Tf [(as brain researcher )] TJ ET
0.000 0.000 0.800 RG
0.6 w 0 J [ ] 0 d
449.240 329.505 m 542.192 329.505 l S
BT 65.000 317.829 Td /F1 12.0 Tf [(Hasler points out in an easy-to-read article)] TJ ET
0.6 w 0 J [ ] 0 d
65.000 315.249 m 268.952 315.249 l S
0.000 0.000 0.000 rg
BT 268.952 317.829 Td /F1 12.0 Tf [(.)] TJ ET
BT 35.000 291.573 Td /F1 12.0 Tf [(And so it comes to pass that we have a huge stock, by now we must say, of storybooks about what can happen in)] TJ ET
BT 35.000 277.317 Td /F1 12.0 Tf [(the brain when Aunt Emma knits and little Jimmy memorizes nursery rhymes. Beautiful pictures, pretty)] TJ ET
BT 35.000 263.061 Td /F1 12.0 Tf [(narratives, all suggesting to us that the most important thing in the world of science at present is knowledge about)] TJ ET
BT 35.000 248.805 Td /F1 12.0 Tf [(what makes the brain tick. Except that, in the majority of cases, all these stories have little more value than the)] TJ ET
BT 35.000 234.549 Td /F1 12.0 Tf [(sagas of classical antiquity. The sagas of classical antiquity sometimes contain a kernel of truth and are at least)] TJ ET
BT 35.000 220.293 Td /F1 12.0 Tf [(exciting. Whether the kernel of truth of the published fMRI studies is greater than that of the sagas? Indeed: the)] TJ ET
BT 35.000 206.037 Td /F1 12.0 Tf [(colourful images of the fMRI studies are the baroque churches of postmodernity: beautiful, pictorial narratives of)] TJ ET
BT 35.000 191.781 Td /F1 12.0 Tf [(a questionable theology.)] TJ ET
BT 35.000 137.383 Td /F2 18.0 Tf [(Sources and literature)] TJ ET
BT 49.800 106.766 Td /F1 12.0 Tf [(1.)] TJ ET
BT 65.000 106.761 Td /F1 12.0 Tf [(Eklund, A., Nichols, T. e., & Knutsson, H. \(2016\). Cluster failure: Why fMRI inferences for spatial extent)] TJ ET
BT 65.000 92.505 Td /F1 12.0 Tf [(have inflated false-positive rates. Proceedings of the National Academy of Science, early edition. Doi: )] TJ ET
0.000 0.000 0.800 rg
BT 65.000 78.249 Td /F1 12.0 Tf [(https://doi.org/10.1073/pnas.1602413113)] TJ ET
0.6 w 0 J [ ] 0 d
65.000 75.669 m 263.348 75.669 l S
0.000 0.000 0.000 rg
BT 263.348 78.249 Td /F1 12.0 Tf [(.)] TJ ET
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 5)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
24 0 obj
<< /Type /Annot
/Subtype /Link
/A 25 0 R
/Border [0 0 0]
/H /I
/Rect [ 449.2400 331.0052 542.1920 342.8852 ]
>>
endobj
25 0 obj
<< /Type /Action
/S /URI
/URI (http://www.beobachter.ch/gesellschaft/artikel/forschung_es-geht-auch-fast-ohne-hirn/)
>>
endobj
26 0 obj
<< /Type /Annot
/Subtype /Link
/A 27 0 R
/Border [0 0 0]
/H /I
/Rect [ 65.0000 316.7492 268.9520 328.6292 ]
>>
endobj
27 0 obj
<< /Type /Action
/S /URI
/URI (http://www.beobachter.ch/gesellschaft/artikel/forschung_es-geht-auch-fast-ohne-hirn/)
>>
endobj
28 0 obj
<< /Type /Annot
/Subtype /Link
/A 29 0 R
/Border [0 0 0]
/H /I
/Rect [ 65.0000 77.1692 263.3480 89.0492 ]
>>
endobj
29 0 obj
<< /Type /Action
/S /URI
/URI (https://doi.org/10.1073/pnas.1602413113)
>>
endobj
30 0 obj
<< /Type /Page
/MediaBox [0.000 0.000 595.280 841.890]
/Parent 3 0 R
/Contents 31 0 R
>>
endobj
31 0 obj
<<
/Length 1942 >>
stream
0.6 w 0 J [ ] 0 d
0.000 0.000 0.000 rg
BT 49.800 760.490 Td /F1 12.0 Tf [(2.)] TJ ET
BT 65.000 760.485 Td /F1 12.0 Tf [(Edgington, E. S. \(1995, orig. 1987\). Randomization Tests. 3rd Edition. New York: Dekker.)] TJ ET
BT 49.800 746.234 Td /F1 12.0 Tf [(3.)] TJ ET
BT 65.000 746.229 Td /F1 12.0 Tf [(Wackermann, J., Seiter, C., Keibel, H., & Walach, H. \(2003\). Correlations between brain electrical)] TJ ET
BT 65.000 731.973 Td /F1 12.0 Tf [(activities of two spatially separated human subjects. Neuroscience Letters, 336, 60-64.)] TJ ET
BT 49.800 717.722 Td /F1 12.0 Tf [(4.)] TJ ET
BT 65.000 717.717 Td /F1 12.0 Tf [(Schulte, D., & Walach, H. \(2006\). F.M. Alexander technique in the treatment of stuttering – A randomized)] TJ ET
BT 65.000 703.461 Td /F1 12.0 Tf [(single-case intervention study with ambulatory monitoring. Psychotherapy and Psychosomatics, 75, 190-)] TJ ET
BT 65.000 689.205 Td /F1 12.0 Tf [(191.)] TJ ET
BT 49.800 674.954 Td /F1 12.0 Tf [(5.)] TJ ET
BT 65.000 674.949 Td /F1 12.0 Tf [(Open Science Collaboration. \(2015\). Estimating the reproducibility of psychological science. Science,)] TJ ET
BT 65.000 660.693 Td /F1 12.0 Tf [(349\(6251\), aac4716.)] TJ ET
BT 35.000 611.187 Td /F2 12.0 Tf [(Date Created)] TJ ET
BT 35.000 596.931 Td /F1 12.0 Tf [(Mai 2022)] TJ ET
q
35.000 594.080 0.000 0.000 re W n
Q
0.502 0.502 0.502 rg
0.502 0.502 0.502 RG
35.000 781.760 m 585.280 781.760 l 585.280 783.260 l 35.000 783.260 l f
0.000 0.000 0.000 rg
BT 416.240 816.290 Td /F2 10.0 Tf [(PROF. DR. DR. HARALD WALACH)] TJ ET
BT 377.370 804.386 Td /F1 10.0 Tf [(https://harald-walach.de https://harald-walach.info)] TJ ET
0.502 0.502 0.502 rg
35.000 52.000 m 585.280 52.000 l 585.280 50.500 l 35.000 50.500 l f
0.000 0.000 0.000 rg
BT 44.000 31.996 Td /F1 10.0 Tf [(Page 6)] TJ ET
q
44.000 17.740 532.280 11.880 re W n
0.000 0.000 0.000 rg
BT 263.855 20.116 Td /F1 10.0 Tf [(© Prof. Harald Walach)] TJ ET
Q
endstream
endobj
xref
0 32
0000000000 65535 f
0000000009 00000 n
0000000074 00000 n
0000000120 00000 n
0000000319 00000 n
0000000348 00000 n
0000000497 00000 n
0000000633 00000 n
0000006579 00000 n
0000006688 00000 n
0000006796 00000 n
0000006924 00000 n
0000007015 00000 n
0000007141 00000 n
0000007232 00000 n
0000007359 00000 n
0000007467 00000 n
0000007572 00000 n
0000014540 00000 n
0000014645 00000 n
0000020955 00000 n
0000021060 00000 n
0000028021 00000 n
0000028159 00000 n
0000034759 00000 n
0000034887 00000 n
0000035023 00000 n
0000035150 00000 n
0000035286 00000 n
0000035411 00000 n
0000035502 00000 n
0000035607 00000 n
trailer
<<
/Size 32
/Root 1 0 R
/Info 5 0 R
/ID[]
>>
startxref
37602
%%EOF