1 The Purpose of This Book, Who Should Read It, and How to Use It

All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

1.1 What Is Cognitive Electrophysiology?

Cognitive electrophysiology is the study of how cognitive functions (including perception,

memory, language, emotions, behavior monitoring/control, and social cognition) are sup-
ported or implemented by the electrical activity produced by populations of neurons.
Cognitive electrophysiology is a broad field that contains a spectrum of researchers. At
one end of the spectrum are researchers who are mainly interested in cognitive processes.
For these scientists electrophysiology is useful because it is more sensitive than behavioral
measures such as reaction time or introspective self-report and therefore is better able to dis-
sociate cognitive processes and their subcomponents. At this end of the spectrum task design
and theory development are crucial, and sophisticated data analyses with precise neurophys-
iological interpretations are less important. Understanding neural mechanisms is relevant,
but ultimately the goal of the research is to dissect and understand the cognitive components
of behavior rather than the physiological properties of the brain.
At the other end of the spectrum are researchers who are mainly interested in discover-
ing the functional properties of neural networks and who use cognitive paradigms as tools
to elicit specific patterns of neural activity. For these scientists electrophysiology is useful
because it is a direct measure of neural population-level activity, can link research in humans
to computational models of neurobiological and neurophysiological processes, and offers an
opportunity for cross-species comparisons. At this end of the spectrum sophisticated data
Copyright @ 2014. The MIT Press.

analyses and neurophysiological interpretations of results are most important, and cognitive
tasks (if used at all) are generally as simple as possible, containing few conditions and many
trials. Cognitive theories are useful as interpretational frameworks, but ultimately the goal
of the research is to understand how the brain works rather than to dissect components of
behavior.

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 4 Chapter 1

Cognitive Electrophysiology
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

(Psychology) (Neuroscience)

Figure 1.1
Cognitive electrophysiology is a field defined by a spectrum from cognitive to electrophysiology. As a
cognitive electrophysiologist, where on this spectrum you consider yourself will help guide your experi-
ments, hypotheses, data analyses, target journals and conferences, and career choices.

Probably you find yourself somewhere between these two extremes (figure 1.1). It is use-
ful to think about where you place yourself in this spectrum because it will help guide how
much time and energy you devote to reading cognitive psychology papers versus neurosci-
ence papers, the kinds of analyses you should perform on your data, and the types of jour-
nals and conferences that will serve as outlets for your findings and ideas. You should not
criticize other scientists, or allow yourself to be criticized, for being on different positions in
this spectrum.

1.2 What Is the Purpose of This Book?

The purpose of this book is to teach you the conceptual, mathematical, and implementa-
tional (via Matlab programming) bases of time-, time-frequency-, and synchronization-based
analyses of magnetoencephalography (MEG), electroencephalography (EEG), and local field
potential (LFP) recordings from humans and nonhuman animals. If you go through this
book chapter by chapter and implement the examples in Matlab using the provided sample
data or your own data, you will develop an understanding of why and how analyses are per-
formed, how to interpret results, what are the methodological and practical issues of these
analyses, and how to perform single-subject-level and group-level statistical analyses.
Noninvasive human electrophysiology has been around for a long time, at least since the
1960s and of course dating back to Hans Berger’s recordings in the late 1920s. However, in
the past decade there has been a surge of publications using advanced analysis techniques
for EEG and MEG data (figure 1.2). These analysis techniques include spectral analysis, time-
frequency decomposition, and synchronization analyses. This is in large part thanks to
Copyright @ 2014. The MIT Press.

increases in computing power, software programs such as Matlab with reasonably easy script-
ing languages, and the development of open-source toolboxes (such as eeglab, fieldtrip, spm,
nutmeg, cartool, and BrainStorm) and commercial software (such as Brain Vision Analyzer,
Curry, and BESA) that can perform those techniques. These developments have allowed
researchers to explore their data and to link electrophysiological dynamics to behavior and
cognition in ways not previously possible.

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 The Purpose of This Book 5

200
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

180
Number of publications

160

140

120

100

80

60

40

20

0
1985 1990 1995 2000 2005 2010
Year of publication

Figure 1.2
This histogram shows that the number of publications utilizing time-frequency analyses has increased
considerably over the past years. The data were obtained from a pubmed search of EEG or MEG or LFP
and time-frequency, and were accessed in December 2012.

However, despite the increase in availability of automated programs to perform advanced

analyses, an understanding of the mathematical and methodological details and the knowl-
edge of the pitfalls associated with these analyses are lacking in many researchers who use
the automated programs. That is, the analysis methods available are becoming increasingly
sophisticated, but many of the researchers who apply those methods do not fully understand
what happens “under the hood” when they click buttons in a software program or run some
lines of Matlab script that they were given.
People often ask whether there are tutorials or books from which to learn what those
analyses mean and how they work. Unfortunately, the answer has always been: No, you have
to be a good Matlab programmer, be willing to learn math on your own, and spend lots of
time struggling in Matlab until these concepts become internalized and you can apply them
flexibly. Widespread use of advanced data analysis techniques in cognitive electrophysiology
is limited by a lack of resources from which to learn those techniques. This is unfortunate
Copyright @ 2014. The MIT Press.

because cognitive electrophysiology will benefit from researchers who, for example, have a
strong background in psychological theories but less formal mathematical training.
There are good resources for learning more generally about EEG and ERP research
(Handy 2004; Luck 2005; Regan 1989), and there are online tutorials (e.g., on the eeglab
and fieldtrip websites) that provide general introductions about using software tools and
interpreting results from time-frequency-based analyses. But these resources are less useful

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 6 Chapter 1

for understanding the math or the potential pitfalls of such analyses, and they provide insuf-
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

ficient instructions for people who want to implement the methods themselves or further
adapt existing methods. On the other side, there are resources that assume a considerable
amount of background knowledge in mathematics or physics. These resources often leave
variables in equations undefined, have little useful information about practical implementa-
tion matters such as how to perform statistics or how to deal with limited data, and they are
largely inaccessible to those without a graduate-level mathematical or physics education.
Thus, what is missing is a resource designed for cognitive scientists, neuroscientists,
cognitive neuroscientists, psychologists, and the like, who are intelligent and motivated to
understand and implement data analyses but who lack the formal training in mathemat-
ics and computer science to utilize the mathematically dense analysis resources or to read
the raw computer code behind software packages. This book is designed to provide such a
resource.

1.3 Why Shouldn’t You Use <Insert Name of M/EEG Software Analysis Package>?

You can be confident that the algorithms in commercial software packages and Matlab
toolboxes such as Brain Vision Analyzer, eeglab, fieldtrip, spm8, nutmeg, ELAN, cartool,
BrainStorm, and others are correct and valid. If you want to perform the analyses that these
toolboxes are specifically designed to perform (often, these are the analyses that the research
groups developing the toolboxes find most relevant for their research), then there is noth-
ing wrong with using point-and-click kinds of analysis programs. But if you want to under-
stand the details of the methods, or if you want to have the flexibility to perform analyses
not included in these packages, you will need an intimate understanding of the math and
logic behind time-frequency decomposition as well as an understanding of the practical
implementation-level details, effects of parameter selection, and potential methodological
issues of the analyses. The more you understand how the analyses work and how they are
implemented in Matlab, the more flexibility and freedom you will have to analyze your data
in a manner best tailored to your specific needs, research questions, and experiment design.
Ultimately, this book is about maximizing your freedom, at least when it comes to analyzing
cognitive electrophysiology data.
Copyright @ 2014. The MIT Press.

You should not feel discouraged from using existing analysis packages, and you should
not feel pressured to write all of your own code for your analyses. However, it is important
to understand what happens to your data from raw form to publication-quality figures. Writ-
ing your own code will ensure that you understand each step, but even if you never write
your own code, and even if you never open Matlab, you should have at least a basic working

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 The Purpose of This Book 7

understanding of what happens to your data when you click on the “analyze now” button
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

in your analysis software.

It does not matter if you use point-and-click graphical user-interface programs or run some
lines of Matlab code; neither guarantees that you will know what you are doing. Graphical
user interfaces and Matlab scripts can be equally well used, abused, or misunderstood. The
hope is that, by going through this book, you will understand what happens to your data
when you click that button or run a few lines of code that someone gave you.

1.4 Why Program Analyses, and Why in Matlab?

Perhaps you do not need to spend time programming analyses and instead can understand
how time-frequency-based analyses work simply by looking at equations and illustrative fig-
ures. If this strategy works for you, then that is great; you can consider yourself in the upper
echelon of gifted mathematically minded people. For the rest of us, mathematical equations
may seem to make sense while we are looking at them, but this level of understanding is
fairly superficial and rigid. In my experience teaching this material, most students look at
an equation and slowly utter “okay,” as if they hope that by declaring the equation sensible,
they will somehow understand what it means. But then they have a bewildered expression
when it is time to turn to Matlab and implement the equation. Once they have worked
through the implementation and can see visually what the function does to input data,
their eyelids raise, their backs straighten, and they smile, because now they finally realize
what the equation means. Experiences like these have convinced me that most people learn
time-frequency-based analyses by implementing them, not by looking at equations. This is
the reason that this book comes with over 10,000 lines of Matlab code, and this is the reason
why you will learn best from the book if you go through the book in front of a computer with
Matlab open and work through the equations in computer code as you read the book (more
on this point in section 1.6).
Then there is the question of why one specific programming language—Matlab—is high-
lighted in this book over other programming languages. Matlab is a high-level program-
ming environment that is relatively easy to learn and use. Several of the most widely used
M/EEG analysis packages are Matlab toolboxes. Because it is so widely used, Matlab code can
Copyright @ 2014. The MIT Press.

easily be shared with people in other labs and in other countries. Matlab has a command
interface that can store and make accessible large amounts of data. This is advantageous
because you can easily inspect data during each stage of processing and analysis, and you
can easily inspect results from different subjects. Matlab also makes very nice-looking plots
that are customizable and that can be exported as pixel-based image files (jpg, bmp, png,

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 8 Chapter 1

tiff), vector files (eps), or movies, which can be used to make presentations and publication-
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

quality figures.
Despite the focus on Matlab, the analyses and algorithms presented in this book are plat-
form independent, and you could translate the code that accompanies this book to any
other programming language, such as Python, C++, and maybe even BASIC. If you do not
have access to Matlab, Octave is an alternative program you can use. Octave is a free software
that runs on Windows, Mac, and Unix systems and can interpret most Matlab code. Octave
is not quite as fast as Matlab and does not have many bells and whistles, but it works. You
should be able to use almost all of the sample code in Octave with little or no modification.
Translating the code to other languages such as Python will take more time and effort but
should be possible. If anyone is courageous enough (and interested enough) to translate the
online Matlab code into another language, I would be happy to post it online for anyone
else to download.
For the remainder of the book I will simply write “Matlab” without implying that Matlab
is the only software package or programming language that could be used. When specific
Matlab commands, functions, or syntax are referred to, Courier typeface is used. Most of the
analysis code presented in this book does not rely on any extra Matlab toolboxes. In cases
when functions from toolboxes are used, alternatives will be suggested when available. The
three most helpful Matlab toolboxes for EEG data analyses are, in order of likely use, the
signal-processing toolbox, the statistics toolbox, and the image-processing toolbox.

1.5 How Best to Learn from and Use This Book

This book is organized in what I hope is a logical fashion, meaning you can start on page
1 and continue reading until the end. The best way to go through this book is in front of
a computer with Matlab installed and the code and sample data accompanying this book
downloaded, so you can run the code and generate the figures that appear in the book. How-
ever, concepts are also explained in plain English and with pictures, so you should be able to
develop an intuitive, semantic understanding of the analyses, their mathematical bases, and
their interpretations even without going through the Matlab code.
The book is organized in a somewhat monotonic fashion such that the ordering of book
Copyright @ 2014. The MIT Press.

chapters corresponds roughly to the order in which you perform analyses, progressing from
simple to advanced single-subject analyses and finally group-level statistics. Because the
material is cumulative, the information in each chapter is built on the information discussed
in previous chapters. Thus, you will maximize your learning if you read the chapters in the
order in which they are presented.

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 The Purpose of This Book 9

1.6 Sample Data and Online Code

All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

There are sample data and Matlab code provided with the book, downloadable from www
.mikexcohen.com/book. Almost every figure in this book was made using the online code
and sample data. Running the code as you follow the book will not only reproduce the
figures but will also help you to understand each step of each analysis and will facilitate
further exploration and development of the methods. Note that in many cases the code
for one figure relies on variables that were created when previous figures from that chapter
were made. Thus, if you open the Matlab file for chapter 12 and try to generate figure 12.6,
Matlab might crash because of undefined variables. If this happens, start from the beginning
of that chapter and run the code to create figure 12.1, then figure 12.2, and so on. I checked
all the code for each chapter using Matlab versions 2011a and 2012 on 64-bit Windows
and on 64-bit Mac. If you have a different Matlab version or a different operating system,
it is possible that you might need to make minor adjustments to the code. Some features of
the code will crash on older versions of Matlab (such as using a tilde to suppress output of a
function); these situations and their solutions are presented on first use in the code.
I thought carefully about which EEG dataset to include as sample data in this book.
I decided not to use the cleanest and most ideal dataset I could find. Rather, I decided to
use a dataset that I think represents the signal-to-noise characteristics of a typical subject
in a scalp EEG experiment. Thus, the single-subject results shown in this book reflect the
quality of results you can expect for single-subject data analyses. Labels and locations of
electrodes are shown in figure 1.3. The dataset is one condition from one subject taken
from Cohen and Ridderinkhof (2013). The data have been cleaned by manual trial rejec-
tion and independent components analysis using the eeglab software package (Delorme and
Makeig 2004). Some intracranial EEG data are also provided, specifically for cross-frequency
coupling analyses.
You can use your own data instead of the sample data, but keep in mind that you might
need to change some code depending on the format and nature of your data compared to
the sample data.
I encourage you to use and adapt the Matlab code for your data. However, keep in mind
that the scripts were written specifically for the analyses presented in the book and for the
Copyright @ 2014. The MIT Press.

sample dataset. If you modify the code it is your responsibility to do so correctly and appro-
priately. Using the code without reading this book may result in inappropriate or incor-
rect analyses or misinterpretations of results. Again, it is your responsibility to use the code
appropriately. If you do not feel comfortable adapting the code to your own data, then it is
best to use the online code only for the provided sample data when learning from this book

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 10 Chapter 1

Fp1 Fpz Fp2

All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

AF7 AF8
AF3 AFz AF4
F7 F5 F8
F3 F1 Fz F2 F4 F6
FT7 FC5 FC3 FT8
FC1 FCz FC2 FC4 FC6

T7 C5 C3 C1 Cz C2 C4 C6 T8

CP1 CPz CP2 CP4 CP6

TP7 CP5 CP3 TP8
P3 P1 Pz P2 P4
P7 P5 P6 P8
P9 PO3 POz PO4 P10
PO7 PO8
O1 Oz O2

Iz

Figure 1.3
Electrode names and two-dimensional topographical locations of EEG electrodes in the sample data
provided with the book.

and to use the analysis software package that you are comfortable with when analyzing your
own data.

1.7 Terminology Used in This Book

Most of the material in this book can be used for EEG, MEG, electrocorticogram (ECoG,
also called intracranial EEG), LFP, electromyography (EMG), functional magnetic resonance
imaging (fMRI), near-infrared spectroscopy, or any other discretely sampled time-varying
signal. Some of the analyses and statistics in this book are geared toward human scalp EEG
and MEG measurements; using the material provided here for other purposes may require
some adaptation.
For simplicity and consistency throughout this book, I write “EEG,” but this does not
imply that the analyses are inappropriate for MEG or LFP or any other time-varying signal. In
some cases there are differences between the treatment of MEG and EEG data (for example,
differences arise concerning volume conduction and some spatial filters); these situations are
Copyright @ 2014. The MIT Press.

specified. For the same reason, the term “electrode” is used for convenience, but this is not
meant to exclude those analyses from being applied to MEG sensors.
Data analyses have terms. Analysis terms are useful short-hand references for a set of
assumptions and mathematical equations that are applied to data. For example, the terms
“correlation,” “factor analysis,” and “general linear model” refer to specific data analysis

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 The Purpose of This Book 11

procedures. In the cognitive electrophysiology literature, the same or very similar mathemat-
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

ical analyses have been given different terms by different groups. This is unfortunate because
it leads to confusion about what analysis was actually applied to the data. In many cases
terms are ambiguous and could refer to very different analyses, with different interpretations
and different neurophysiological or cognitive significances. The position taken in this book
is that analysis terms should be concise descriptions of the data analysis rather than inter-
pretations of the results or hypothesized neural mechanisms. This argument is further devel-
oped in chapters 21 and 37. When many terms for the same analysis exist, I have tried to list
all of them when introducing the method and then justify the term preferred in this book.

1.8 Exercises

Many chapters contain exercises at the end. These exercises will reproduce and further
develop the material covered in that chapter. These could be used as homework assignments
in a class or simply as a personal challenge.
Most of the exercises ask you to take Matlab code presented in that chapter and further
develop the code to perform additional analyses that were introduced. Thus, the results you
produce from the exercises will, in many cases, look similar to the figures in that chapter.
For this reason it will be helpful to use the provided sample data for the exercises rather than
using your own data.
Some exercises have right and wrong answers; others do not. Some exercises, for example,
ask you to perform an analysis using a variety of parameters. In these cases the point is for
you to appreciate the impact (or perhaps lack thereof) of the parameters of different tech-
niques on the results and to consider what range of parameter settings is appropriate and
what range of parameter settings is inappropriate and should not be applied to your data.
Answers to the exercises are not provided, but pictures showing what the correct solution
might look like are available online along with the code and data. In some cases your solu-
tion might look slightly different from mine if you used different parameters or because the
exercises involved using randomly generated numbers. Nonetheless, in most cases, if your
solution produces a plot that looks like what you can see in the exercises solution online, it
is likely that you solved the exercises correctly.
Copyright @ 2014. The MIT Press.

1.9 Is Everything There Is to Know about EEG Analyses in This Book?

Certainly not. There are limitless possibilities for analyzing data, and publications detailing
new ways to analyze data appear almost monthly in peer-reviewed journals. Included in

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 12 Chapter 1

this book are what I consider the most useful, promising, and accepted approaches for link-
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

ing EEG dynamics to cognitive processes. Others may disagree. There are simply too many
possible methods to detail in one book. To those scientists whose published methods are
not included in this book but who think that their methods are relevant for understanding
neurocognitive function, please accept my apologies and do not take the omission of your
method as a negative scientific evaluation of your work.
That said, by going through the methods in this book, you will be well prepared to learn
and develop other analysis approaches that are not presented here or are not discussed in
depth. Many of the basic methods detailed in this book (including convolution, the Fourier
transform, Euler’s formula, wavelets, circular variance, and permutation testing) are funda-
mental and form the groundwork for other advanced data analysis methods.
None of the methods presented in this book is my invention, although in some cases,
I suggest minor adaptations or improvements to existing methods (the only equations to
which I can claim partial credit are 14.1, 19.3, and 26.10). Rather, this book explains in plain
English, math, and Matlab how analyses that are commonly used in the literature can be
performed, how the results can be interpreted, and what pitfalls and potential methodologi-
cal concerns arise with the use of certain methods. Not all of the methods are appropriate
for all datasets and all experiments. Whether an analysis method is appropriate for your data
and your experiment depends on many factors. Although I include some suggestions for the
circumstances or research questions for which particular analyses are well suited, it is ulti-
mately your responsibility to determine whether an analysis method is appropriate for your
data and for your experiment.

1.10 Who Should Read This Book?

This book is written specifically for individuals who have little formal mathematical training
but who are motivated to learn the conceptual, mathematical, and implementational bases
of EEG data analyses. The book is appropriate for advanced undergraduates up to full profes-
sors but is probably of most practical use for graduate students and postdoctoral researchers
who are conducting or would like to conduct EEG research. Some experience with EEG and
Matlab (or other programming language) is useful but not necessary.
Copyright @ 2014. The MIT Press.

Scientists with formal and extensive training in mathematics, engineering, or physics will
find some of the material in this book below their level (e.g., the mechanisms of convolution
and the Fourier transform). Other material, however, particularly regarding practical matters
of dealing with statistics, trial count, data interpretation, and so on, should be useful to even
the most mathematically savvy readers. There are other books that are more dense with math

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 The Purpose of This Book 13

and Matlab for advanced users (Chatfield 2004; van Drongelen 2006), although resources
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

such as these books tend to have less information about practical matters and implementa-
tional details and, in some cases, are not specifically about EEG analyses, and thus, the reader
must still have some additional expertise to adapt the material to EEG data analyses. Refer-
ences are cited throughout the book for further reading.
This book may also be useful to individuals who do not plan on analyzing EEG data but
who would like to learn more about how the analyses work and how to evaluate results of
time-frequency analyses. If you are completely new to EEG, the general introduction to EEG
and time-frequency results in chapters 2, 3, and 5 should be a useful starting point.
The book is written specifically with a view toward scalp-recorded EEG and MEG recorded
in humans. Although some of the discussions on preprocessing (chapters 6–8) and spatial
filtering (chapters 22 and 24) apply primarily to scalp EEG and MEG data, the sections on
time-frequency decomposition methods, connectivity, and statistics can be applied to any
time series data of sufficient sampling rate (generally, over 200 Hz), regardless of the spatial
scale or species from which the data were obtained.
Finally, this book is not meant to repudiate any existing EEG analysis package, nor do
I suggest abandoning EEG analysis packages in favor of writing 100% of your own analy-
sis code. Understanding the concepts and math behind EEG analyses is important whether
you write your own code, modify someone else’s code, supplement existing packages with
custom-tailored code, or use user-friendly point-and-click software programs. Understanding
the mathematical bases of advanced data analyses will help you correctly perform those anal-
yses, appropriately interpret results, get more out of your data, avoid errors or suboptimal
analysis choices, and critically evaluate results you see in publications, talks, and conference
posters. The sample data are given in eeglab format for convenience, but this is not meant
as an explicit endorsement of eeglab, nor is it meant as an implicit repudiation of any other
Matlab toolbox for analyzing EEG data. There are advantages and disadvantages to each
analysis package, and I have tried to make this book as package-independent as possible.
There are format-conversion utilities from eeglab to other major packages, and it should be
fairly straightforward to modify the code in this book to use with the data storage formats of
other Matlab toolboxes.
Copyright @ 2014. The MIT Press.

1.11 Is This Book Difficult?

Probably. It starts out easy but becomes progressively more challenging. How difficult it is for
you depends on your background level of math, programming, and EEG analysis. Unfortu-
nately, most advanced analyses involve concepts, math, and programming implementations

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost
 14 Chapter 1

that are likely to be unfamiliar to many psychology and neuroscience students. But the learn-
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

ing curve is steep, and almost anyone with sufficient motivation should be able to learn the
material covered in this book. As mentioned earlier, the book was designed for you to prog-
ress through chapters in monotonic order, and cumulative knowledge is expected. Therefore,
the book may be more difficult if you skip over some chapters. The chapters you should
certainly not skip over are chapters 10–13; these chapters explain the basics of the Fourier
transform, convolution, and Euler’s formula, which play central roles in nearly all other data
analysis techniques.

1.12 Questions?

If you do not understand some of the material in the book, the first thing you should do is
go back to the last section you understood, and then work forward again. Make sure you are
doing all the exercises in Matlab. Work with other people, and ask colleagues who might
understand the material better than you.
Science is and should be a challenge that you relish and look forward to each morning.
Probably you find the experience of learning to be rewarding, which is why you are in sci-
ence. I hope this book gives you double pleasure: you get to learn something new (how to
analyze electrophysiology data), and you get to use that knowledge to learn other new things
about how the brain works. Good luck and have fun.
Copyright @ 2014. The MIT Press.

EBSCO : eBook Collection (EBSCOhost) - printed on 10/8/2018 5:19 AM via UNIVERSITY OF AUCKLAND
AN: 689432 ; Cohen, Mike X..; Analyzing Neural Time Series Data : Theory and Practice
Account: s2802549.main.ehost