The Statistics Hell has expanded: An interview with Prof. Andy Field

FieldDoes the mention of the word “statistics” strike fear into your heart and send shivers down your spine? The results section of your thesis seeming like that dark place one should avoid at all cost? Heteroscedasticity gives you nightmares? You dread having to explain to someone what degrees of freedom are? What is the point of using ANOVA if we can do a series of t-tests? If any of these remind you of the pain of understanding statistics, or the dread of how much more lies ahead during your studies, when all you really want is someone to explain it in a humanly understandable way—look no further. Quite a few fellow students might tell you “You should go and look at Andy Field’s books. Now, at least, I understand stats”. The “Discovering statistics using …” is a gentle, student friendly introduction to statistics. Principles are introduced at a slow pace, with plenty of workable examples so that anyone with basic maths skills will be able to digest it. Now add a lens of humor and sarcasm that will have you giggling about statistics in no time!

There is a new book!

As JEPS has been excited about introducing Bayesian statistics into the lives of more psychology students (see here, here, and here for introductions, and here for software to play around with the Bayesian approach), the idea of a new book by Andy Field—whose work many of us love and wholeheartedly recommend—which incorporates this amazing approach was thrilling news.

We used this occasion to talk to Andy Field—who is he, what motivates him, and what are his thoughts on the future of psychology?

With your new book, you expand the Statistics hell with Bayesian statistics. Why is this good news for students?


There has, for a long time, been an awareness that the traditional method of testing hypotheses (null hypothesis significance testing, NHST) has its limitations. Some of these limitations are fundamental, whereas others are more about how people apply the method rather too blindly. Bayesian approaches offer an alternative, and arguably, more logical way to look at estimation and hypothesis testing. It is not without its own critics though, and it has its own set of different issues to consider. However, it is clear that there is a groundswell of support for Bayesian approaches, and that people are going to see these methods applied more and more in scientific papers. The problem is that Bayesian methods can be quite technical, and a lot of books and papers are fairly impenetrable. It can be quite hard to make the switch (or even understand what switch you would be making).

My new book essentially tries to lay some very basic foundations. It’s not a book about Bayesian statistics, it’s a book about analysing data and fitting models and I explain both the widely used classical methods and also some basic Bayesian alternatives (primarily Bayes factors). The world is not going to go Baysian overnight, so what I’m trying to do is to provide a book that covers the material that lecturers and undergraduates want covered, but also encourages them to think about the limitations of those approaches and the alternatives available to them. Hopefully, readers will have their interest piqued enough to develop their understanding by reading more specifically Bayesian books. To answer the question then, there are two reasons why introducing Bayesian approaches is a good thing for students: (1) it will help them to understand more what options are available to them when they analyse data; and (2) published research will increasingly use Bayesian methods so it will help them to make sense of what other scientists are doing with their data.

Your books are the savior for many not-so-technical psychology students. How did you first come up with writing your classic ‘Discovering Statistics with ….’ book?

Like many PhD students I was teaching statistics and SPSS to fund my PhD. I used to enjoy the challenge of trying to come up with engaging examples, and generally being a bit silly/off the wall. The student feedback was always good, and at the time I had a lot of freedom to produce my own teaching materials. At around that time, a friend-of-a-friend Dan Wright (a cognitive psychologist who was at the time doing a postdoc at City Univerity in London) was good friends with Ziyad Marar, who now heads the SAGE publications London office but at the time was a commissioning editor. Dan had just published a stats book with SAGE and Ziyad had commissioned him to help SAGE to find new authors. I was chatting to Dan during a visit to City University, and got onto the subject of me teaching SPSS and my teaching materials and whatever and he said ‘Have you ever thought of turning those into a book?’ Of course I hadn’t because books seemed like things that ‘proper’ academics did, not me. Subsequently Dan introduced me to Ziyad, who wanted to sign me up to do the book, I was in such a state of disbelief that anyone would want to publish a book written by me that I blindly agreed. The rest is history!

As an aside, I started writing it before completing my PhD although most of it was done afterwards, and I went so over the word limit that SAGE requested that I do the typesetting myself because (1) they didn’t think it would sell much (a reasonable assumption given I was a first-time author); and (2) this would save a lot of production costs. Essentially they were trying to cut their losses (and on the flip side, this also allowed me to keep the book as it was and not have to edit it to half the size!). It is a constant source of amusement to us all how much we thought the book would be a massive failure! I guess the summary is, it happened through a lot of serendipitous events. There was no master plan. I just wrote from the heart and hoped for the best, which is pretty much what I’ve done ever since.

Questionable research practices and specifically misuse of statistical methods has been a hot topic in the last years. In your opinion, what are the critical measures that have to be taken in order to improve the situation?

Three things spring immediately to mind: (1) taking the analysis away from the researcher; (2) changing the incentive structures; (3) a shift towards estimation. I’ll elaborate on these in turn.

Psychology is a very peculiar science. It’s hard to think of many other disciplines where you are expected to be an expert theoretician in a research area and also a high-level data analyst with a detailed understanding of complex statistical models. It’s bizarre really. The average medic, for example, when doing a piece of research will get expert advice from a trials unit on planning, measurement, randomization and once the data are in they’ll be sent to the biostats unit to fit the models. In other words, they are not expected to be an expert in everything: expertise is pooled. One thing, then, that I think would help is if psychologists didn’t analyse their own data but instead they were sent to a stats expert with no vested interest in the results. That way data processing and analysis could be entirely objective.

The other thing I would immediately change in academia is the incentive structures. They are completely ****** up. The whole ‘publish or perish’ mentality does nothing but harm science and waste public money. The first thing it does it create massive incentives to publish anything regardless of how interesting it is but it also incentivises ‘significance’ because journals are far more likely to publish significant results. It also encourages (especially in junior scientists) quantity over quality, and it fosters individual rather than collective motivations. For example, promotions are all about the individual demonstrating excellence rather than them demonstrating a contribution to a collective excellence. To give an example, in my research area of child anxiety I frequently have the experience that I disappear for a while to write a stats book and ignore completely child anxiety research for, say, 6 months. When I come back and try to catch up on the state of the art, hundreds, possible thousands of new papers have come out, mostly small variations on a theme, often spread across multiple publications. The signal to noise ratio is absolutely suffocating. My feeling on whether anything profound has changed in my 6 months out of the loop is ‘absolutely not’ despite several hundred new papers. Think of the collective waste of time, money and effort to achieve ‘absolutely not’. It’s good science done by extremely clever people, but everything is so piecemeal that you can’t see the word for the trees. The meaningful contributions are lost. Of course I understand that science progresses in small steps, but it has become ridiculous, and I believe that the incentive structures mean that many researchers prioritise personal gain over science. Researchers are, of course, doing what their universities expect them to do, but I can’t help but feel that psychological science would benefit from people doing fewer studies in bigger teams to address larger questions. Even at a very basic level this would mean that sample sizes would increase dramatically in psychology (which would be a wholly good thing). For this to happen, the incentive structures need to change. Value should be maximised for working in large teams, on big problems, and for saving up results to publish in more substantial papers; contribution to grants and papers should also become more balanced regardless of whether you’re first author, last author or part of a team of 30 authors.

From a statistical point of view we have to shift away from ‘all or nothing thinking’ towards estimation. From the point of view of publishing science a reviewer should ask three questions (1) is the research answering an interesting question that genuinely advances our knowledge: (2) was it well conducted to address the question being asked – i.e. does it meet the necessary methodological standards?; and (3) what do the estimates of the effects in the model tell us about the question being asked. If we strive to answer bigger questions in larger samples then p-values really become completely irrelevant (I actually think their almost irrelevant anyway but …). Pre-registration of studies helps a lot because it forces journals to address the first two questions when deciding whether to publish, but it also helps with question 3 because by making the significance of the estimates irrelevant to the decision to publish it frees the authors to focus on estimation rather than p-values. There are differing views of course on how to estimate (Classical vs Bayes, confidence intervals vs. credibility intervals etc.) but at heart, I think a shift from p-values to estimation can only be a good thing.

At JEPS we are offering students experience in scientific publishing at an early stage of their career. What could be done at universities to make students acquainted with the scientific community already during their bachelor- or master studies?

I think that psychology, as a discipline, embeds training in academic publishing within degree and PhD programs through research dissertations and the like (although note my earlier comments about the proliferation of research papers!). Nowadays though scientists are expected to engage with many different audiences through blogs, the media and so on, we could probably do more to prepare students for that by incorporating assignments into degrees that are based on public engagement. (In fact, at Sussex – and I’m sure elsewhere –  we do have these sorts of assignments).

Statistics is the predominant modeling language in almost any science and therefore sufficient knowledge about it is the prerequisite of doing any empirical work. Despite this fact, why do you think do many psychology students are reluctant to learn statistics? What could be done in education to change this attitude? How to keep it entertaining while still getting stuff done?

This really goes back to my earlier question of whether we should expect researchers to be data analysis experts. Perhaps we shouldn’t, although if we went down the route of outsourcing data analysis then a basic understanding of processing data and the types of models that can be fit would help statisticians to communicate what they have done and why.

There are lots of barriers to learning statistics. Of course anxiety is a big one, but it’s also just a very different thing to psychology. It’s a bit like putting a geography module in an English literature degree and then asking ‘why aren’t the students interested in geography?’. The answer is simple: it’s not English literature, it’s not what they want to study. It’s the same deal. People doing a psychology degree are interested in psychology, if they were interested in data they’d have chosen a maths or stats degree. The challenge is trying to help students to realize that statistical knowledge gives you power to answer interesting questions. It’s a tool, not just in research, but in making sense in an increasingly data-driven world. Numeracy and statistics, in particular, has never been more important than it is now because of the ease with which data can be collected and, therefore, the proliferation of contexts in which data is used to communicate a message to the public.

In terms of breaking down those barriers I feel strongly that teaching should be about making your own mark. What I do is not ‘correct’ (and some students hate my teaching) it’s just what works for me and my personality. In my previous books I’ve tried to use memorable examples, use humour, and I tend to have a naturally chatty writing style. In the new book I have embedded all of the academic content into a fictional story. I’m hoping that the story will be good enough to hook people in and they’ll learn statistics almost as a by-product of reading the story. Essentially they share a journey with the main character in which he keeps having to learn about statistics. I’m hoping that if the reader invests emotionally in that character then it will help them to stay invested in his journey and invested in learning. The whole enterprise is a massive gamble, I have no idea whether it will work, but as I said before I write from my heart and hope for the best!

Incidentally if you want to know more about the book and the process of creating it, see

What was your inspiration for the examples in the book? How did you come up with Satan’s little SPSS helper and other characters? How did you become the gatekeeper of the statistics hell?


The statistics hell thing comes from the fact that I listen to a lot of heavy metal music and many bands have satanic imagery. Of course, in most cases it’s just shock tactics rather than reflecting a real philosophical position, but I guess I have become a bit habituated to it. Anyway, when I designed my website (which desperately needs an overhaul incidentally) I just thought it would be amusing to poke fun at the common notion that ‘statistics is hell’. It’s supposed to be tongue-in-cheek.

As for characters in the SPSS/R/SAS book, they come from random places really. Mostly the reasons are silly and not very interesting. A few examples: the cat is simply there to look like my own cat (who is 20 now!); the Satan’s slave was because I wanted to have something with the acronym SPSS (Satan’s Personal Statistics Slave); and Oliver Twisted flags additional content so I wanted to use the phrase ‘Please sir! Can I have some more …’ like the character Oliver Twist in the Dicken’s novel. Once I knew that, it was just a matter of making him an unhinged.

The new book, of course, is much more complicated because it is a fictional story with numerous characters with different appearances and personalities. I have basically written a novel and a statistics textbook and merged the two. Therefore, each character is a lot deeper than the faces in the SPSS book – they have personalities, histories, emotions. Consequently, they have very different influences. Then, as well as the characters the storyline and the fictional world in which the story is set were influenced by all sorts of things. I’d could write you a thesis on it! In fact, I have a file on my hard drive of ‘bits of trivia’ about the new book where I kept notes on why I did certain things, where names or personalities came from, who influence the appearance of characters or objects and so on. If the book becomes a hit then come back to me and ask what influenced specific things in the book and I can probably tell you! I also think it’s nice to have some mystery and not give away too much about why the book turned out the way it did!

If you could answer any research question, what would it be?

I’d like to discover some way to make humans more tolerant of each other and of different points of view, but possibly even more than that I’d like to discover a way that people could remain at a certain age until they felt it was time to die. Mortality is the cloud over everyone’s head, but I think immortality would probably be a curse because I think you get worn down by the changing world around you. I like to think that there’s a point where you feel that you’ve done what you wanted to do and you’re ready to go. I’d invent something that allows you to do that – just stay physically at an age you liked being, and go on until you’ve had enough. There is nothing more tragic than a life ended early, so I’d stop that.

Thank you for taking the time for this interview and sharing your insights with us. We have one last question: On a 7-point Likert scale, how much do you like 7-point Likert scale?

It depends which way around the extremes are labelled …. ;-)


For more information on ‘An adventure in statistics: the reality enigma’ see:




Lea Jakob

Lea Jakob

Lea Jakob is currently finishing her psychology Master’s degree at University of Zagreb, Centre for Croatian Studies. Her research interests include clinical psychology within which she is writing her masters thesis on the topic of cognitive impairment in pulmonary patients as well as music perception and cognition. Apart from her passion for research, she has a serious case of wanderlust paired with polyglotism.

More Posts