Please Santa, no more ugly slideshows!

The objections of the venerable Edward R. Tufte notwithstanding, slideshow presentations have become an essential part of science. I sit through a great number science talks over the course of any given year. And probably 90% of them are awful. Not in terms of content, necessarily, or delivery; but in terms of ugly, overladen, whatever-the-visual-equivalent-of-tone-deaf-is slides. It doesn’t have to be that way. And frankly, I’ve had enough. So here are some golden rules for slideshow presentations, purely in terms of graphical design (i.e. how to illustrate a good talk, rather than how to give a good talk). Be warned. This is an opinionated rant that has been a long time gestating. But it is also 100% true. First, a disclaimer: If you are a graphic designer, please ignore the following and get on with designing your own beautiful work. But you’re not, are you? You’re a busy scientist trying to put together a talk in a hurry. Me too. Fortunately the nice people who design templates for your slideware package of choice actually do have some understanding of design principles. Yes, even in PowerPoint I daresay (although I switched to Keynote a few years ago, and have recently gone about 18 months without being able to open PowerPoint on my computer without it crashing, so I am perhaps not best placed to say). So, just as you wouldn’t want a designer tinkering with your data – probably best not to insist on putting your own distinctive stamp on their designs. My first rule, then, is:

1. Use design templates. And take them seriously. If a template has room for one image, insert one image. If it has room for three bullet points, that’s how many you should use. The template designers have decided on the optimal size for figures, fonts, etc. that will be appropriate for viewers. Don’t second guess them. My record spot, from a reasonably eminent speaker at a Royal Society discussion meeting, is 24 graphs on one slide. Not good. I have a much lower opinion of his science as a result of that one slide. Why would anyone who has ever seen a scientific talk even think of doing that? It’s 20 years since slides were physical objects that you might wish to use economically. So if you need to use a load more slides to fit in everything you want, then do so. In fact, that’s rule 2:

2. Use as many slides as you need. Given that digital slides are free, I am constantly amazed at how much information people try to cram on each one. There is no penalty for including sufficient slides to allow you to show, say, one figure per slide. But, I hear you say, the figure you want to include is part of a four panel figure from the paper you are promoting. So surely all four parts will have to go on the slide? No. See rule 3…

3. Redraw your figures. The figures that you used in a publication are almost certainly not going to be appropriate for a presentation. Labels will be too small, lines too faint, points invisible (and probably too numerous), and overall the figures will be too complex. So, you should have versions of relevant figures produced specifically for presentations. If, like me, you write little functions to produce all your figures, that’s quite easy – just modify arguments for line thickness, colour, axis labels, etc. If not, redrawing from scratch may seem like a pain, but the likelihood is that you’ll be using these figures multiple times, so the effort will certainly be repaid. If a figure for which you don’t have the raw data is not good enough, consider if you really need to use it. If so – can you digitise it, and so produce your own, prettier version? If so, do. (GraphClick is useful for this, or there’s a digitize packages in R too.)

4. Keep text to a minimum, but no less. I don’t want to read a full set of your notes, in bullet point form, on every slide. Your slides are a communication tool, not an aide memoire. So just write important stuff. But, write it in proper sentences, keeping abbreviations and acronyms to an absolute minimum. (And get your bloody indentations formatted neatly too.)

5. Avoid clutter. Never leave yourself in the position where you have to say, ‘You won’t be able to read this / make this out / discern what I want you to see, but…’ There is one exception, a Joker which can be played at most once per talk, and that is if you wish to illustrate the concept of complexity itself. For instance, you could include a slide of tedious but straightforward algebraic manipulations, to illustrate the tedious (but straightforward) nature of the maths required to reach your result. Or you may wish to stick in a horrendously complex food web network diagram to be accompanied by a wry comment along the lines of ‘as you can see, this system is rather complex…’ (An aside: I included such a diagram in a talk, given to a general audience a few years ago, on communicating complexity, as shorthand for complexity in ecological systems. A few months after, at an ecological conference, I saw a rather high profile marine biologist use the exact same diagram – in fact, 10 of them on a single slide – and proceed to talk as if the audience was able to process them and to discern the salient features. As you can imagine, I was delighted to note this down!)

6. Consistency is more important than specifics. Taste in design is personal. I have my preferred Keynote design templates, others I think are nasty. A lot of PowerPoint templates are pretty ugly, in my opinion. I prefer certain fonts and colour schemes over others. You make your choice (although do familiarise yourself with the basics of colour theory – 5 minutes work, max). Then, whatever your choice, stick to it. All your slides should follow the same scheme. If you’re using a few slides provided by other people, ensure they conform too. (This will probably involve re-drawing them. See rule 3.)

OK, I could go on but that’s probably enough (oh, but, of course – 7. ClipArt. Don’t). Just please, if you want me to concentrate on what you have to say over 5 to 60 minutes, have the courtesy to present a well-designed talk. I won’t respect your science if you don’t. A talk is a story, not a comprehensive CV: give some thought as to what needs to go in, and (most importantly) what can be left out – because the flip side is, if you give a good talk, I will probably go and read your papers, or at least collar you afterwards, and can catch up on details then.

One final point. Stick to all of the rules above, at all times. However, if you should ever see me break any of them, I am doing so knowingly, and that’s fine. Have a beautiful 2012!

Science and criticism, good and bad

It was always going to be tough for whoever took over weekly science column duties in the Guardian from Ben Goldacre. His Bad Science was entertaining about very important stuff – stats, data, the importance of evidence, and so on. So it’s perfectly possible that I’m judging Philip Ball’s first effort, which appeared last Saturday, too harshly. But, it put my back up from the very first sentence. What’s especially frustrating is that I think his aim, to be “like an arts critic, but for science”, is laudable. I’ve argued before for a more prominent place for science in public intellectual debate, and this kind of informed criticism would be an essential part of that. Unfortunately, over the course of this first column Ball has shown few signs that he has any of the attributes required to be such a critic. First, he seems not to know much about how scientists work. Forgive me if I quote in full the first paragraph to illustrate this point:

Scientists don’t like being criticised. Well, who does? But I don’t mean that they don’t like it when people say they are wrong, biased, self-serving or insular. I don’t like it when people say those things either, because in my experience scientists tend to be right, fair, generous and – well, OK, they could do with getting out more. But scientists don’t like being criticised in the proper sense of the word: in the way that books and plays and music are judged, for better or worse, by critics.

OK, so what’s wrong here? Lets ignore the tired stereotype about scientists needing to ‘get out more’ (unlike, say, anyone else who has a demanding and satisfying job with no set working hours? And especially poorly judged given the continuing revelations in the Leveson Inquiry about the less than social behaviour of some in Ball’s profession). No, let’s think instead about the nature of criticism in science.

I spend a good proportion of my working life either criticising the work of others, or responding to others’ criticisms of me. And I mean criticism in the sense – more or less – that Ball favours. That is, assessing the worth of science and placing it in its proper context. It is an essential part of the process of scientific publication, and – with the exception (on paper anyway) of a few journals like PLoS ONE – is always about much more than right and wrong. Constantly, we make judgements about the novelty, significance and interpretation of a set of results. I can’t think of many scientists who would seriously argue that “science is a question of fact – either it’s right or it isn’t”, as Ball charicatures us. Put simply, methods and results sections of papers would, we hope, fall into that kind of binary category; but Introductions and Discussions are pure interpretation, frequently subjective and designed to put a particular spin on the results to convince a journal editor that they fit within the journal’s stated aims.

We all know that, don’t we? And do we (i.e., all scientists) really have, as Ball claims, “a kneejerk aversion to any claim that science is shaped by culture”? Most of the scientists I know are a bit smarter than that. One example: Greg McInerney from Microsoft Research, who last week gave a talk in the Ecology & Environment Seminar Series that I run in Sheffield, spent some time discussing the sociology of the methodological choices that people make when modelling species distributions. To my knowledge, no-one in the audience thought this to be an especially unusual or outrageous departure. More generally, the science blogosphere is full of scientists grappling with such ideas.

One can’t help but suspect that Ball is not a great participant in the thriving culture of science online, and is building his straw men from (if anything) conversations with professors emeritus rather than with those actually doing and thinking about science. This comes across too in his claim that science journalism does have a pedagogical role in expalining science to the public “in language that scientists have forsaken” – again, discounting countless scientist-run outreach projects and terrific science communication (random example of excellence from Deep Sea News) in favour of a lazy stereotype about scientists using difficult language (which we do in our technical papers, of course, because well, they’re technical).

So, I am absolutely all in favour of a broader public debate about the social context of science. I think that would be a really positive step. But Ball’s column has failed to convince me that science journalists like himself are the people to steer this debate. Rather, science online has left him and his kind somewhat out of the loop, and scientists are already having these discussions between ourselves, and communicating them directly to the public.

Interactions and main effects in simple linear models

Bet that title’s got you itching to read on! Feel free to skip this post if you think stats are boring. You’d be wrong, but I won’t judge you… Anyway, getting straight down to the nitty gritty, and assuming that if you’ve hung in until now then you’re not afraid of words like ‘linear’ and ‘model’, here’s the thing: when I learnt statistics, one thing that was drummed in to us was that, if you’re fitting a linear model which includes interactions, you can’t sensibly interpret the main effects. I’ve been telling people the same ever since. But, in almost every paper that I review or edit, and which uses such models, people do just that. My purpose here then is to explain why I think that’s wrong. And hopefully, to find out if I am wrong to think that way.

Let’s start with a contrived example. Suppose we’ve measured activity levels at different times of the day across 100 individual birds, 50 of which are larks and 50 are owls, and we get the following:

Of course, in analysing these data the obvious thing to do would be to fit a linear model with activity modelled as a function of hour, species, and their interaction. The fitted lines on the figure above illustrate this model. And the explanation is straightforward: activity increases towards dawn in larks, and decreases towards dawn in owls.

What doesn’t make sense is to make any statements about general differences between larks and owls, or about any general trend in activity with time from dawn, because these ‘main effects’ are completely entangled within their interaction.

Unfortunately, most statistical software packages will give an output which includes significance levels for both main effects and interactions, for instance:

Or in Anova form (R guys – I know this is wrong, but its particular flavour of wrongness is not important for this point!):

Both of these outputs make it look like the main effects are ‘significant’, and the coefficients even seem to tell you the direction of these effects. So the kind of interpretation I read again and again would look something like: “There is a significant interaction between activity and species (p < 0.0001). In addition, activity increases with hour from dawn (p < 0.0001), and is higher in owls than in larks (p < 0.0001).” Which, as we’ve just demonstrated, is nonsense. And just because this example has been contrived to emphasise this point, doesn’t mean it doesn’t apply equally whenever there is an interaction; sensible plotting of your data will usually reveal this.

What should you do instead? Report the significant interaction, and then describe it, using the table of coefficients (and your knowledge of how your stats package of choice uses aliasing) to calculate the slope (and intercept if you like) for each level in the interaction. Here for example, the slope for the relationship for larks is 1.04, and for owls is 1.00 (1.045 – 2.048) – and you can get confidence intervals for both easily enough. And if the interaction doesn’t seem to be important, take it out, and interpret main effects to your heart’s content.

That’s my take on it anyway. The very concept of the ‘significance’ of a main effects is meaningless in the presence of an interaction between them. And so I keep telling people. So if I’m missing something, please let me know!