Writing a good dissertation
It is now dissertation writing season for British Masters students at many universities, including at my illustrious, benevolent, and wise employer (for the next week), the University of Bath. I supervise a lot of dissertations and find myself giving the same advice over and over again. I don’t mind doing this (it’s part of my job – and I think people respond better to individualized feedback even if it’s actually the same for everyone). But I thought that some of my students, and perhaps some others, would find some of my insights helpful.
Caveat #1: This advice is mainly intended for dissertations, although a lot of it also applies to term papers or journal articles. Unless you’re my student, I’m not the one marking (or “grading,” for the Yanks) your dissertation. So, caveat emptor.
Caveat #2: Your goal is not necessarily the same as mine. What you want is to write a dissertation that gets a good score. What I want is for you to write a dissertation that is good, full stop. These are correlated but not always the same thing – unless I’m the one marking.
Caveat #3: I’m an experimental cognitive and behavioral scientist. I have pretty good interdisciplinary chops – I double-majored in undergrad in Philosophy and Cognitive Science; wrote an interdisciplinary honors thesis (not to brag, but it won the Philosophy department’s dissertation award), did my PhD at Yale in Cognitive Psychology (not to brag even more, but that dissertation also won prizes); my post-doc was in Behavioral Economics (with a Sociologist as my boss); and I’m currently a Lecturer (Assistant Professor) in Marketing. I publish in journals across the cognitive and behavioral sciences. But what I definitely am NOT is an engineer or a mathematician or a biologist or (most emphatically) any kind of humanities scholar. I say all this not to impress you but to help you evaluate where my advice might or might not be reliable.
Caveat #4: Immanuel Kant wrote a famously convoluted book called The Critique of Pure Reason. Convoluted but brilliant, and many philosophers have spent their careers trying to understand it. One of the reasons it is hard to understand is because it hardly has any examples. It’s filled with concepts like the “synthetic a priori” which is supposed to be an entire category of existence but of which only a couple examples are given in a 700+ page book. Why so few examples? Kant says the book would have been too long if it had examples. (Sorry bud, that ship sailed around page 400. I say lean into it.) Unfortunately, unlike Kant, I am spending a couple afternoons, not a decade, writing this post, so I don’t have that many examples here. I’ll try to update this later on with more examples from real dissertations, but for now I just wanted to get this out so you can use these tips. If you are already writing a dissertation, it will usually be easy to supply your own examples from your own work.
Okay, caveats are emptoring and buyers are bewaring. On to the advice.
BEFORE YOU START – YOUR AUDIENCE
1. Figure out who will be reading your dissertation. This is crucial. Your dissertation isn’t a blog or a diary or a haiku. It is a scientific document used for communication. It may even be put into an online repository so that other scientists can read it in the future. It is a serious thing. In the short term, your dissertation will be read and evaluated by one or more people who have specific expertise and specific expectations. It could be just one person who marks it, but more often in the UK, dissertations are double-marked. Try to find out at least what field the double-marker will be in if you can and whether they are likely to be an expert on the topic. This is important both for writing a good dissertation (because “goodness” is relative to the audience) and also for getting a good grade (because in this case your audience is judging you).
In addition to knowing your readers’ area of expertise, it’s important to understand your relationship with your advisor because this determines a lot about how your day-to-day work itself will be structured, and it will also influence aspects of the optimal writing strategy. These fall into three basic categories and there are pros and cons to each. The key thing is for you to get ahead of this and understand who you are working with, ideally – if your program works this way – while you have some freedom to select who you want to work with.
– Coaching. If you have a coaching (or apprenticeship) relationship, your advisor thinks of you as contributing to her own research agenda. Your advisor probably helped you come up with the idea, design the studies, maybe gave you a lot of feedback on your statistical analyses or literature suggestions. Your advisor might have come up with an idea with you together, but probably gave you a lot of feedback (likely much of it negative) on your early ideas, before settling on something you both found interesting. Because your advisor thinks of your dissertation as part of her research, she doesn’t necessarily think of working with you as “teaching time” (which professors usually try to limit to their contract) but instead may think of it as her “research time,” and that working with you is an investment in her own research. Thus, she may gladly spend quite a bit more time with you than she is contractually obliged to. Your advisor might have strong opinions about every little thing in your dissertation and you can/should ask. However, there is a good chance she doesn't care much about the formal rules – and her opinion about this counts the most, especially if she is the only person grading you.
– Cheerleading. If you have a cheerleading relationship, your advisor isn’t thinking of your dissertation as part of her research, but is eager for you to do the best you can and to contribute expertise in whatever ways she can. Typically your advisor won’t be an expert on the particular subject you are studying and will encourage (or likely require) you come up with your own idea, perhaps with some feedback. By the time you’re done, you should know more about the topic than your advisor does; if you don’t, it is unlikely you wrote a good dissertation. This kind of advisor may well be generous with her time – not because it’s her research, but because she is nice and wants to help. But the amount of help she is able to give may be modest, compared to the coaching model. You may have to figure a lot out by yourself. It’s possible that your advisor will care about formal rules quite a bit, since that is a concrete aspect of your work that she can advise you on.
– Refereeing. If you have a refereeing relationship, your advisor thinks of herself as helping to keep score and keeping you within the rules of the game. Many universities and departments allocate specific numbers of workload hours to each faculty member for each dissertation; for example, a total of 4 hours of scheduled meeting time is common, plus some time allocated to written feedback and grading. Referees will want to keep to these hours, in part because she thinks of the advising relationship as more transactional and professional and in part because she may think it is unfair to exceed this allotment. You are going to be largely on your own, although referees will do their best to answer your specific questions within the allotted time constraints. They may or may not have specific expertise in the topic area of your dissertation, but when they are grading your work they are likely to pay a lot of attention to whether the formal requirements are satisfied.
Why do you need to figure this out? Because the dynamic will be very different depending on which kind of advisor you have. If you have any choice about which one to take, meet with a few options and try to figure out which kind of relationship they tend to have with their dissertation students. It’s usually not hard to tell how much someone knows about a topic, how excited they are to help you, and how they view their obligations toward students. Ask them how they typically work with students, how often they want to meet, who typically comes up with the ideas, whether they like to publish with their students, and so on. There’s no right or wrong choice necessarily, because some students would rather come up with their own ideas (and maybe have more trouble executing them) and others would rather have an expert help them a lot with the idea part (but then maybe learn more technical skills one-on-one).
2. Figure out how diligently you must follow “the rules.” In addition to shaping the dynamics of your project, what kind of advisor you have will very much shape the dynamics of your writing. Some people care more about the “rules” than others. Personally, I am reasonable and fairly flexible. I want your dissertation to be good and when I mark it, there will be a correlation of +1.0 between how good it is and how good of a mark it gets (though do remember that most dissertations are double-marked).
Many advisors are not as flexible as I am. They will care about things like whether you follow the formatting, whether your dissertation is the right length, whether you have enough references, and so on. Now, these things do matter up to a point. But they are important not in themselves but as signs of deeper problems. If you are not formatting your dissertation consistently, that’s sloppy. Sloppy writing is usually a signal of sloppy thinking and sloppy science. If you are writing something way too short, you’re not including enough ideas; and if you’re writing something too long, you’re not writing concisely. If you don’t have enough references, you’re not connecting your work to the current state of knowledge. These are all objectively bad things.
But notice that these are the root problems, not your formatting, length, and references. I will look for real problems rather than relying on these superficial cues to evaluate your work. (Please do try to make your work presentable though.) Instead of reading your work carefully, some graders will evaluate it largely on whether it follows the rules and “ticks all the boxes.”
Sometimes the rules are there for a reason. There is a reason why all your references are supposed to be formatted the same way – it makes it easier to know what article you are talking about. There’s a reason why dissertations are not supposed to be too long or too short. There’s a reason why we have uniform guidelines for reporting the results of statistical tests (up to a point; see below). But sometimes so-called “rules” don’t make much sense at all in many situations. When a rule manifestly does not make sense in your case, you should at least consider breaking it – if you can get away with it.
3. Read articles in the field you are studying. You would think this is obvious advice, but I’m afraid it’s not. Many students really don’t know how scientific articles are structured, which is very suspicious because they are citing dozens of them! You don’t have to read every word of every paper you talk about, but (i) you do need to know the central argument of each paper you cite, and (ii) you do need to carefully read every word of some of the papers you cite.
The reason for (i) is self-evident, but the reason for (ii) is less obvious. You need to read some papers in detail because you need to know how people write articles in your field. Your dissertation is basically structured like a journal article, maybe with a longer-than-usual literature review. So for example, if people in your field spend a lot of time talking about the demographics of your sample, you’ll want to do that. If they don’t spend much time talking about that, you won’t want to do it either. Why not? Because your audience will not care unless you can make a special argument about why it particularly matters in your case.
Which articles should you read? Focus on articles that are important to your theory development, are important precedents, or whose methods you are borrowing. As a tie breaker, focus on articles in higher-quality journals (the journal rankings are not perfect but they’re absolutely not random). How many articles to read in detail? I suggest reading at least 5 to 10 articles in detail (and skim many, many more). Is that a lot to expect? Not at all – you have months to write your dissertation, and even reading an article end-to-end only takes a few hours at most. If it takes you longer than that to get through the technical parts, there’s a pretty good chance you need to practice reading and understanding statistics.
4. Pick a topic you care about. All of this advice will be easier to follow if you actually care about what you are studying. If you don’t care, no one will. The neuroscientist Donald Hebb said, “If it isn’t worth doing, it isn’t worth doing well.” 99% of possible research topics don’t matter – pick one of the trillion things to study in the 1% that do.
5. Don’t assume your audience knows more than it does. Often when I go to conferences, they are attended by people in very different fields. My favorite conference is CogSci, which attracts psychologists, computer scientists, philosophers, linguistics, etc. A lot of the work is really good and I love to pieces many of the people who go there. But by god, some of the talks are impossible. It’s like they are trying to make the work incomprehensible to anyone but an expert on their very narrow topic. You’ll have computational modeling people getting really excited that they found some task where the beta parameter is double its normal magnitude and the root-mean square error is lower than any existing model (including the quadruple-helix-attractor model!) and there’s no correlation at all between alpha and theta when the tasks are in one order, but a super-high correlation when they are in the other order. BUT WHAT ARE THESE THINGS AND WHY DO THEY MATTER?????
You don’t want the audience of your dissertation to feel like you are talking down to them. It sort-of works at conferences because the audience will feel like they “should” know what the beta parameter is and they might vaguely remember that the quadruple-helix-attractor model is supposed to be pretty good. They won’t actually UNDERSTAND what you are talking about – so, pointless and not-great! – but they won’t necessarily judge you – so, maybe not actively harmful to you. They might very well give you the benefit of the doubt and assume you’re saying something smart. I guarantee that the person marking your dissertation will not feel this way. They are, in fact, being paid to judge you. If they don’t understand, they will not assume it is because they are dumb. They will assume – correctly – that it is because you are being unclear. Why is that assumption correct? Because, by definition, it is your job to make your audience understand what you are saying. If your dissertation cannot be understood, it is bad – sorry, but that’s tough love.
Examples of things that DO need to be explained: models or theories (drift–diffusion model; dual process theory; elaboration likelihood model; stereotype content model, etc.); how statistical tests work if they are not commonly used (e.g., one sentence explaining multi-level modeling and one sentence explaining why it’s appropriate here is a good idea if you are using this technique); almost all scientific jargon (or better yet, just use ordinary words when it’s not too cumbersome); concepts borrowed from other fields (e.g., if you talk about marketing implications of a psychology finding, explain terms like “segmentation” or just avoid them altogether and use ordinary words).
6. Don’t assume your audience knows less than it does. This is a much less common problem, but this comes up occasionally. It NEVER comes up in scientific conferences or journals, where scientists are crawling all over each other trying to sound as smart as possible. But sometimes dissertation students don’t know what can be safely taken as background. When in doubt, over-explain. But try not to if you can avoid it.
The hallmark of something that does NOT need to be explained is that it is a concept that comes up in many or most articles in your field but is never explained in those articles. This is one of the reasons for suggestion #3: you won’t know what is taken as background unless you actually read some articles carefully.
Some examples of things that DO NOT need to be explained if you are writing a dissertation in the cognitive or behavioral sciences (in other fields these will be different): the definitions of basic statistical concepts like p-values or interaction effects; how basic statistical tests work (such as correlation, t-tests, ANOVA, multiple regression, mediation, chi-squared tests – there’s a grey area for things like logistic regression or Mann-Whitney tests that are common but not universal, and I suggest explaining these briefly along with justifying their appropriateness); the general advantages of experimental vs. correlational vs. qualitative methods (although you should talk about specific reasons why it is useful for you to do the specific experiment you do in adding to the existing literature).
GENERAL WRITING TIPS
Some of this advice is closely aligned with that given by psycholinguist Herb Clark in his classic “Everyone Can Write Better (and You Are No Exception)” which is circulating on the Internet. I’m not covering every point made there, but it’s all excellent advice and I endorse every word. Read Clark’s advice too as a complement to this blog post – and he has lots of examples from student papers. I also recommend my favorite blogger Scott Alexander’s “Nonfiction Writing Advice”. Nominally this is aimed at bloggers, but most of it is good general advice that applies equally to scientific writing.
7. Prize clarity. I’m not sure this is necessary to point out, but the absolute most important characteristic in technical writing is not poetry or humor or nuance or scientific-sounding-ness, but clarity. If the reader does not understand what you are saying, no communication has happened – of cleverness or poetry or nuance or anything else, much less scientific discourse. Basically all of my writing tips are variants on the dictum to be clear.
8. Make an outline and have your advisor look at it. Dissertation writing can be bad in multiple ways. It can be bad because your sentences are confusing. It can be bad because your paragraphs are discombobulated. But the worst way it can be bad is if it does not talk about the right things. A careful reader will do their best to understand your sentences and paragraphs, and will usually not hold grammatical mistakes against you very much (check your grammar anyway!). But anyone will be alarmed to see sections that don’t flow into each other, an argument that doesn’t make sense, a huge and critical topic omitted, or a bizarrely long digression on something irrelevant. Those are killer problems.
For dissertations in the cognitive and behavioral sciences, the typical meta-structure is something like the following (but this can usually be more flexible than you are led to believe). What you need to do is fill in the different pieces of these sections and outline, paragraph by paragraph, what you will talk about in each section. I say what goes in each section here, with detailed tips to follow later on.
– INTRODUCTION. Gets the reader invested in the topic (usually with either an entertaining or vivid example or with a compelling argument); briefly explains what gap in knowledge is being filled; explains what the contributions of the dissertation are; briefly summarizes how your argument will proceed (the sections in the rest of the paper). These don’t necessarily have to go in this order or even in separate paragraphs – but this all needs to be crystal clear, and in no more than 2 pages, ideally less.
– LITERATURE REVIEW. Not every dissertation has this and most published articles don’t. Instead, published articles usually will review only the aspects of the literature that are directly relevant to explaining the article’s contribution to the literature or motivating the theoretical framework. The reason why dissertations usually have a separate literature review section is that your advisor wants to see that you have read the literature carefully and thought about how your work relates to it. A good literature review is selective (covering the most relevant articles), reflective (doesn’t just summarize the articles but talks about the advantages and disadvantages of each study or theory), and generative (it will give you the building blocks to motivate your study later on). The length of this section can vary greatly. I personally favor shorter and more selective literature reviews, but (alas) some people are less reasonable and want you to cover most of the work in your area even if only tangentially relevant. You’ll want to find this out from your advisor. Although you should outline this section early, you should write it near the end when you are trying to hit your word target, since it is easy to make it longer or shorter by talking about more/fewer articles or talking about them in more/less depth.
– THEORETICAL FRAMEWORK. Not every dissertation has this either and it’s not always required, but most good ones have it (or its equivalent interspersed throughout the literature review). You can go one of two ways here.
* Developing a theory. Classic science mode. What are the variables you are investigating and how do you think they fit together? A good theory can sometimes just be something that a scientist dreams up and later tests – I think Einstein’s special relativity might have been like that. But most of us are not Einstein, and we base our ideas on other scientists’ theories and findings. In your theoretical framework section, you can appeal to common sense (including anecdotes), other theories, and existing empirical findings. You shouldn’t appeal to your own results that you will talk about later – that is circular reasoning. However, from your theory you should develop a set of hypotheses that you will then test in your study. (You can even write things out like “Hypothesis 1” to make this super-clear. This is more common in some fields than others.) Some people think you should only hypothesize things that turn out to be true (called hypothesizing after the results are known or HARKing) and this is very common in published papers. However, for a dissertation I advise against this. Aim for logical consistency in your theory section; let the results show what they show and deal with it later.
* Motivating a research question. Instead of developing an a priori theory, the idea is to ask a well-defined research question that is bite-sized enough that you can answer it in your dissertation. Often this will take the form of a “broad” question (too big for you to tackle) and a “narrow” question (a specific way of operationalizing the broad question that you can handle in the space of your dissertation). The central idea of this approach is to explain why *multiple* possible answers to these questions are plausible. Don’t just argue for the answer you think most plausible or the one that turns out to be supported by your data. Why not? Because this makes it unclear why you needed to do the study. If all the arguments push one direction, it’s barely an empirical question anymore. You need to do your best to motivate both what you actually find and what you don’t find, using your three motivational tools – intuitions and anecdotes, prior theory, and prior evidence. The beauty of this approach is that it can make your results meaningful even if you don’t find what you expect.
Either way, the point of this section is to give the readers the tools needed to understand what your results mean. What would finding result X mean versus finding result Y? What would it say about the world more broadly if your theory were supported? How would it challenge existing theories? These are the kinds of questions that are the meat of science and which are the most exciting to think about. It’s fun to learn how things work!
– OVERVIEW. If you have multiple experiments, each one gets its own methods, results, and discussion sections, as well as its own special introduction. (Just read any published paper for examples of how this works.) If you only are reporting one study, I suggest a short section after your theory development section that gives an overview of what your study is doing, how it tests your hypotheses, and either what you predict (if you are developing a theory) or what each result would mean (if you are motivating a research question). Since this is just a very short section (not a whole chapter), I don’t talk about this below when I give specific advice on each section, but please don’t forget to include this somewhere.
– METHODS. Explain what you did in enough detail that someone could reproduce your study. Give examples. Walk us through your study from the participant’s point of view. Talk somewhere about the design of the study in the abstract (e.g., 2x2 between-subjects) but focus mainly on what it is actually like to be a participant. This is much easier to follow.
– RESULTS. Explain what your study showed. One excellent piece of advice I received from one of my grad school mentors was to start the results section with a short paragraph that gives an overview of the findings. This is especially important for multi-study journal articles where people are trying to piece together what all of your results mean together. But even for a one-study dissertation, it’s useful to give an overview to help the reader understand how the results relate to one another. You could churn out pages and pages of results – but don’t do this. Focus on the results that are necessary to test your hypotheses or adjudicate your research questions. It’s fine to do some exploratory data analysis (but please say that this is what you are doing) and report interesting results for other people to follow up on someday. But don’t get distracted by these, even if they are your only significant results. Focus on the bread and butter of science, which is asking and answering questions.
– DISCUSSION. Like the literature review, the length of this section can vary greatly, but unlike the literature review, this section is fun to write. Why? Because this is the part of your dissertation that most resembles a manifesto – and manifestos are awesome (hence this blog post). The “rules” are usually looser for the discussion section (but talk to your advisor).
9. Diverge strategically from your outline. You’re the one who wrote your outline, so if you now know better than you did when you wrote it, it’s okay to change your mind. If it turns out that aspects of your outline don’t make sense – e.g., because it makes more sense to combine three paragraphs into one or divide one paragraph into three – then change it and don’t worry about it. But be strategic about this. Oops, your outline doesn’t have a section for that giant digression you feel like going on? Skip it. Oops, your outline doesn’t include the part where you prattle on for pages about unnecessary statistical tests that you want to include to reach the word limit. Resist! If you need to “pad” your dissertation to get up to a word limit, pad strategically where you can actually add new and interesting content, such as in the literature review and discussion sections. If you add a bunch of stuff to your methods or results, it just makes your dissertation hard to understand.
10. One paragraph = one idea. The basic unit of scientific writing is the paragraph, not the word or sentence. Why? Because science is about ideas and one paragraph expresses one idea. It’s “boring” advice, but it’s soooo true: With very few exceptions, every paragraph must have a topic sentence. You can be subtle. Don’t write: “This paragraph is going to be about topic sentences.” Instead, write “The basic unit of scientific writing is the paragraph, not the word or sentence.” The former is clunky and long, whereas the latter introduces the topic while also explaining why we’re talking about paragraphs now. In fact, your paper should make sense to a reader who only reads the first sentence of every paragraph. Obviously they won’t get the details – but they should get the central message. Try it with this blog post and you’ll find you can get the gist of what I’m talking about by reading just the bolded headings, more wisdom by reading the first one or two sentences of every paragraph, and maximal wisdom by reading every sentence.
What the heck are the other sentences for then?! They are for argumentation. Typically, the topic sentence is an assertion, but this needs to be backed up by evidence. The other sentences give evidence, for instance by presenting a logical argument, by summarizing other research, by giving examples, by relaying anecdotes, by appealing to common sense, or by citing statistics from your own study. The form of argumentation differs across different sections of your paper, but in all sections (with the partial exception of the methods section, which as we will see below is a narrative rather than an argument) your paragraphs will take this form.
11. Use transitions. If the biggest mistake in technical writing is not having a clear outline (where the paragraphs fit together into an overall story or argument), the second biggest mistake is writing paragraphs where the sentences do not cohere. The reader should never be wondering why you are talking about something. Good writing has strong momentum from the start, where every sentence seems to logically entail the next. In really strong writing, this often seems to happen like magic – although in fact this is the result of careful thinking and planning. You don’t need a bunch of extra “therefore”- and “however”-type words if the logical relations are sufficiently clear. But some arguments are inherently intricate; “however” and “therefore” absolutely have a place and inexperienced writers can benefit a lot from using these words to signpost their arguments. You can also use punctuation strategically to achieve the same thing: A colon suggests that the part of the sentence after the colon is a consequence or elaboration of the first part of the sentence. (Re-read the last sentence for an example.)
12. Use parallel structure. Sentences are easier to read when they use parallel structure. A first clause uses a given structure; a second clause relies on that same structure. (See what I did there?) This is because the human mind relies on structure and analogy to understand information: We readily pick out the similarities and differences between things and line them up next to each other in our minds to do this. The easier it is to line them up, the easier it is to understand. Novice writers are often afraid of parallel structure because they think it makes them sound repetitive. This is wrong – it makes them sound clear.
13. Don’t over-qualify. Academics have a lot of bad writing habits; resist being sucked into the vortex. The worst is over-qualifying their sentences. Results don’t show; they “seem to demonstrate.” The procedure wasn’t “based on XX,” but was “a slight variant on an experiment by YYY.” The conclusions don’t follow, but “appear to be consistent with the hypothesis that…” This is bad writing. There is a sound principle at work here – scientists are reluctant to over-claim and they don’t want to offend readers, much less reviewers who might disagree with them. But guess what? If you claim X and the reviewer thinks Y, they’re not going to fall in love with your paper anyway. It’s called confirmation bias and adding a bunch of “somewhats” and “seems” and “appears” and other so-called nuance is not going to help. This is fake nuance. If your claims need to be qualified, be specific. What exactly is preventing you from making a strong statement? If you aren’t saying something specific, you’re just being long-winded and needlessly hard to follow, not being nuanced. (Note: This advice is controversial. It might even get you into trouble. It’s still correct.)
14. Be specific and use examples. Excessive abstraction is another very bad academic habit. The humanities are by far the worst about this, to the extent that their papers and books are often written at such a level of abstraction that they say nothing identifiable about the real world. But I have to admit that psychologists are right up there. Especially when talking about other people’s work, we often write at such a high level of abstraction that the reader doesn’t actually understand what the other person’s study actually found. Abstraction is necessary sometimes because we need to be able to generalize. But we don’t only generalize. Cognitive science research shows that people learn best when they are taught abstract principles alongside concrete instantiations of those principles. If you are teaching calculus, you might give a proof of a theorem but also give a worked example of how that theorem applies in a real-world problem. (As another example, check out the previous two sentences.) The reasoning can be abstract, but you need to support it by talking about how it applies to real things in the actual, flesh-and-blood world.
And while we’re on examples, choose your examples carefully. Examples are usually not a perfect mirror of the point they are illustrating. The example doesn’t have to fit perfectly but it should be a pretty close fit or readers will get confused rather than enlightened. (Note that I didn’t use an example in this paragraph because I couldn’t easily think of one and didn’t want to use a bad one that might be confusing!)
15. Read over your draft several times and viciously cut needless words. Wordiness is the enemy of clarity. Often there is a jewel of an argument buried in a long, meandering paragraph or sentence. Cut the vines; find the jewel. Be vicious and be mean to yourself: If you can cut a word from a sentence without changing the meaning, you should almost always do it. It is theoretically possible to shorten sentences to the point that they are hard to follow – but students almost always have the opposite problem. This shouldn’t be an afterthought. You should be pruning as much as you can while you are writing, and you should make shortening and clarifying your draft the over-arching goal of your revising. Just by shortening you will often go a long way down the road to clarity.
16. Being chatty, funny, silly, and poetic. I’ll lay my cards on the table – I’m a kinda chatty writer. I love first-person and even second-person language (“I” and “you”), I despise unnecessary qualifiers and pretentiousness, and I use informal language whenever I can get away with it. None of this violates the dictates above and indeed is all in the service of writing clearly and vividly. You, of course, do not need to be chatty – frankly most students are not strong enough writers to get away with it. But it’s okay to be informal as long as you are being clear and conceptually rigorous (which is not at all the same as using formal language).
I have a silly streak. Every time I read or write a little joke or whimsical comment, I get a little adrenaline rush. I like being funny and I like being entertained, and this is true even if it doesn’t serve the immediate goal of the argument. However, this is a trade-off and you can’t do it too much – remember, it’s a scientific report, not the Laugh Factory. Still, some levity makes reading your dissertation much more enjoyable and (sadly?) these are often the most memorable parts of even published papers. If you choose to be whimsical, try to make sure you are actually being funny rather than just confusing. Again, not everyone is experienced enough to get away with this (possibly including me) – but if you want to give it a try, you have my blessing.
If you’re going to write poetically, the first and last paragraphs are the place to do this. Did I mention that when I was an undergrad I won a university-wide poetry contest? I did – again, not to brag, but to say that if anyone is authorized to write poetically in their science writing it is probably me. Alas, when co-authors read my first drafts they almost always take all the poetic stuff out. To my chagrin, I have learned that they are usually right to do so. Poetry is not clear. It is not an argument. And it is usually redundant – in the sense that you’ll have to explain clearly whatever you meant somewhere else. If you are a strong writer, I say go for it anyway – I personally enjoy the novelty and appreciate the effort. But you don’t need to worry about sounding like the second coming of Wordsworth (who, by the way, visited our home of Bath many times). I’m probably the only person who ever wanted his cognitive science articles to sound poetic anyway.
Does this advice conflict with the above advice about writing concisely? There is no conflict between writing informally and writing concisely. Chattiness does not mean writing long sentences or writing “the way you talk.” Do NOT write the way you talk. What it means instead is writing that, if read aloud, would sound reasonably natural but preternaturally clear. Writing this way does not come naturally at all to most people.
There is more of a conflict with my advice about whimsical and poetic writing, but it is in service of a broader principle. Your dissertation does not require jokes or poetry. At the same time, jokes and poetry can have a function, and if they do, they are appropriate. The function is not communicating information, but instead keeping the reader’s attention. People like talking to funny, clever, and poetic people. It’s fun. Do you want to be fun or boring? As long as you are clear on what you are doing and your whimsy and poetic turns of phrase are enhancing your reader’s experience rather than confusing your readers, I say go ahead with it if you can handle it.
17. Invest the time in reading a book about writing – but only a good one. Writing is not like talking; it does not come naturally. Most people never become good at writing, and those who do typically don’t do so until they are well out of university. Writing skills are developed over a lifetime of practice and benefit a LOT from feedback from good mentors. But there are ways to jumpstart this process.
The easiest is to read a good book about writing. But it needs to be a good one. Some writing books are just lists of rules without explanation, and some of these rules are actually masquerading as scientific claims about the English language that linguists have long known to be incorrect. However, here are two excellent ones.
– The Elements of Style by Strunk & White. Absolutely classic. This book is super-short and a PDF is even online for free (no idea if this is legal but it is sure easy to find on Google). This book includes possibly the single best paragraph on writing ever composed:
“Vigorous writing is concise. A sentence should contain no unnecessary words, a paragraph no unnecessary sentences, for the same reason that a drawing should have no unnecessary lines and a machine no unnecessary parts. This requires not that the writer make all sentences short, or avoid all detail and treat subjects only in outline, but that every word tell.”
The whole book is excellent, but the crucial part is just section 2, which is fewer than 20 pages and loaded with examples.
– The Sense of Style by Steven Pinker. This is my new favorite writing guide. It’s written by the perfect person for the job – Pinker is a cognitive scientist at Harvard who studies language and is himself a best-selling author. In fact, I think he is the best living science writer by far, for reasons that will take us on too much of a detour. The reason this book is so good is because Pinker really understands how language works and gives both the linguistic theory and lots and lots and lots of practical advice and examples. There’s a big section of the book that’s digging into great detail into specific sentences and paragraphs, analyzing in minute detail about how they can be optimally organized to maximize communicative efficiency. Honestly that part is a long slog, and you don’t have to read it all – but if you are not a confident writer, you will benefit from reading it all. Even so, the first few chapters are a breeze and filled with smart advice and anyone can benefit from reading these.
ADVICE ON SPECIFIC SECTIONS
No one would seriously disagree with most of the above advice, which is largely pretty conventional advice about technical writing. But I also have various opinions about how some of this advice should be implemented in writing for the cognitive and behavioral sciences, and some of these opinions are more controversial. You should probably find out if you’re going to get into trouble for doing any of these things – unless I am your advisor, in which case you will get into trouble for not doing them! A lot of this advice is specific to behavioral and cognitive research, especially experimental work, so you can safely ignore advice that doesn’t make any sense in your specific situation (such as advice about reporting statistics if you are conducting a qualitative study).
18. Writing a good introduction. So if your introduction isn’t a poem, a joke, or a tête-à-tête chat with your reader, what is it? The introduction serves four main functions and you need to hit them all.
– Getting the reader excited. Honestly, students almost never succeed at this, so don’t feel bad if you find this very challenging. Science is work; what’s exciting about that? Yes, but: It’s also a journey of discovery, a flight into the unknown, a plunge into the abyss of ignorance only to swim back to the surface, puffing for the cold sweet air, to unclench a mammoth pearl you picked from the bottom of the sea. Point is: Make your work sound interesting, like a beautiful and dangerous dive, not a trip to the curb to dump the trash or vomiting into the toilet to purge yourself of your hangover.
What’s interesting? Novelty is interesting. The real world is interesting. Stories are interesting. Puzzles and paradoxes are interesting. Real problems that real people face are interesting. What’s not interesting? Extending the model of Yahtzee et al. (2001). Applying the Theory of Planned Behavior to yet another slightly different topic. Replicating Johnson’s (2019) paper. (Sorry, replicators – your work is important but it’s not particularly interesting. Yes, I know this is part of the problem.)
Any research can be made interesting if you write about it vividly, use realistic examples, and exploit the psychology of interest (being novel, paradoxical, or telling stories). If you don’t think your research can be made interesting – well, first, you’re wrong; but second, why are you doing it anyway? If it’s not too late, pick something that at least you find interesting. By the way, replicators: “replicating Johnson’s (2019) paper” isn’t interesting, but “challenging the empirical basis of Johnson’s (2019) groundbreaking work” is interesting. And in fact, if what Johnson did was on an interesting topic, your paper is automatically on an interesting topic too. Don’t lead with the fact that you are replicating the finding, but instead focus on why the finding is important. You can explain why it needs to be replicated later on.
By the end of the first paragraph, the reader should know what you’re talking about, why it’s important, and at least an inkling of what you’re going to say about it.
– Identifying the gap. A dissertation will be a permanent part of the cumulative scientific record, which is an awe-inspiring responsibility if you think about it. This means that not only does your topic need to be exciting, but there has to be at least something new that you can say about it. Even if you are replicating a study, explain why it’s necessary to do so. Have other, similar studies failed to replicate? Have subsequent findings cast doubt on the result? Has subsequent theory been hard to reconcile with the finding? Has tons of research built on the initial finding without double-checking whether it’s actually true? Is the finding from the Dark Ages of psychology that only ended in 2011? (Kidding, but not really.) This is, of course, easier if you are not replicating a previous finding. Then you just need to explain the two or three most closely related studies or the relevant theoretical gap, and explain how there’s a hole missing that you are going to fill. The gap is going to be your research question.
– Identifying your contribution. This is linked with the gap, but it’s not the same. A contribution is something important. Most gaps are not at all important and it’s for the better that they not get filled in. (Remember: “If it’s not worth doing, it’s not worth doing well.”) Science requires time and resources and we can’t study everything. Why should we study this specific gap? The style of argumentation here should combine the reasons why the topic is important with the reasons why the gap has not been filled. You may just have one central contribution – for a dissertation that’s fine, although a bit light for a journal article. You may have multiple contributions, possibly to different scientific literatures. In that case, explain what those are. What will we gain from closing the gap – by answering your research question? How will this advance our understanding of something that matters?
– Explaining your approach. How are you going to make this contribution and answer your question? Are you doing an experiment, conducting a survey, reviewing the literature, running focus groups, analyzing secondary data, performing a meta-analysis? You don’t necessarily have to explain why you are choosing the broad technique you are (e.g., everyone knows we do experiments because they are high in internal validity and meta-analyses because they summarize a lot of existing work), but you need to justify why the specific thing you are doing is a good idea for addressing your research question. Sometimes this paragraph will also explain how the rest of the paper is organized, especially if it’s a theory-paper that doesn’t have the standard Intro/Theory/Methods/Results/Discussion format.
19. Writing a good literature review. A good literature review prepares us for the rest of the paper. You identified a gap earlier on, but this was bought on credit. You didn’t explain the literature in enough detail for the reader to really verify that there’s a gap. Here you need to pay your debts – summarize the adjacent literature in enough detail that the reader is confident that you did your homework, know the relevant studies inside-out, and have thought carefully about how you are positioning the work relative to the rest of the field. At this point, I’m assuming you’ve spent a lot of time digging on Google Scholar, asked your advisor for suggestions, dug online some more, read through at least the abstracts of dozens of articles, looked at the references in those articles for more relevant articles, read at least 5 to 10 of these articles word-for-word, and thought about how all of this relates to your project.
So now you have all your references. First you have to decide which ones to talk about. Some of them are usually pretty obvious – the ones you are building on, that you rely on for your theoretical framework, ones that help to motivate the research, particularly seminal articles, or even just articles you like or think are interesting or fun. You should be scrupulous in also citing anything that is very closely related to what you are doing, even if you think that article “scooped” your project in the sense that it pre-empts your own research. Science, at its best, requires radical honesty and you should be upfront about this, while still explaining as best you can why your dissertation still makes an advance. This is usually possible because there are many choices researchers have to make in setting up a study, and it is astronomically unlikely that all of their choices were the same as yours. You can find some niche.
The step where students most consistently trip up, though, is not selecting the papers but organizing them.This is not something where you can just bang out an outline and follow it or, worse yet, just summarize each article in the order you found them. You need to read extensively before you can write an outline. You need to understand the key issues on your topic and how your research fits in. And, critically, you need to develop your own way of categorizing the previous literature that makes sense to you.
Which of the following do you think is easier to understand and remember? Seriously, try memorizing each set of words without looking at the next one, and see how many you can get on each list.
– Option 1:
Drive, Helmet, Legs, Pedal, Shoe, Walk, Ride, Key, Engine
– Option 2:
Ride, Helmet, Pedal
Key, Drive, Engine
Legs, Walk, Shoe
– Option 3:
Helmet, Pedal, Ride
Key, Engine, Drive
Shoe, Legs, Walk
Obviously 2 and 3 is easier than 1. Why is that? Options 2 and 3 divide the words into categories, so you can just remember that there are three different categories, each with three things. That’s easier than remembering an undifferentiated blob of nine things because it imposes structure on the information. But 3 is also easier than 2. That’s because option 3 also takes advantage of order within each category to impose further structure. The first item in each list is an object needed as a preliminary for an activity, the second is an object that is used to carry out that activity, and the third is the activity itself. Not only is this structured consistently, but it’s causally structured so that the first thing leads to the second thing, which leads to the third thing. Instead of having to remember 9 random things or 3 sets of 3 random things, you just have to remember 3 bigger things, each of which has the same underlying causal logic. Most people would not be able to successfully memorize the first list, but almost everyone can memorize the third one.
The same principles apply in your literature review. Find a scheme for organizing the articles. It could be by theoretical approach, it could be a few different parts of your own theory that each need separate support, it could be two different dimensions of something, it could be by method, it could be by which of several relevant variables they examined (either independent or dependent variables, or even mediators or moderators), it could be by whether the results were positive, negative, or mixed. Sometimes this is relatively easy and sometimes it is relatively hard. Paradoxically, the more you have read the easier this gets because you will have a better sense of the key dimensions along which the literature varies.
Probably 90% of the time in writing a literature review is reading the articles and figuring out how they fit together. Actually summarizing each article is comparably straightforward. For a dissertation-style literature review, one paragraph per article is often appropriate, although you can and should group multiple similar studies into the same paragraph if they don’t differ in any way you consider relevant (i.e., along the features you decided to use in your organizational scheme) and come to similar conclusions, serving the same feature of your argument.
Remember to write a one-paragraph summary at the end of your literature review that helps to integrate it all together and helps us to understand what the key knowns and unknowns are. If the evidence is conflicting, say that. If there’s evidence for X and Y, but no one has looked at important thing Z, say that. Some of this will be stuff you already said about the gap in the Introduction but didn’t have the evidence to back up. Now you do and we need a reminder.
20. Writing a good theory section. Coming up with a good theory is hard. But writing a good theory section is surprisingly easy if you wrote a good literature review. As a reminder, journal articles very often combine the literature review and theory sections, only reviewing the articles relevant for developing the theory. In a dissertation, you’ll probably be reviewing more than you would want to in a theory section, so it does make sense for you to keep them separated.
As I noted earlier, there are two kinds of theory sections. One proposes a series of deductive hypotheses, explaining the logic behind each one, whereas the other poses a series of questions, explaining why multiple different answers are plausible for each question. These are both ways of motivating your research. The advice is slightly different depending on which of these routes you are going down.
– Deductive theories. I’m not going to try to tell you how to come up with a good theory – although that would be a good topic for its own post – but I’m going to assume that you already have a model in mind, ideally something you came up with before collecting your data. Most social science theories have all or a subset of the following elements, although sometimes they are disguised:
a. Causal or explanatory factors. What are the variables that are either causing effects or explaining variation in other variables?
b. Effects or outcomes. What are the variables that the causes are affecting or that are being explained?
c. Mechanisms. What variables explain why the cause has the effect it does?
d. Moderators. What variables explain when the cause has the effect it does or when the explanatory variables are relevant to the variables being explained?
You’ll recognize these as the theoretical versions of familiar experimental design concepts – independent and dependent variables, mediators, and interactions. But since I want to distinguish between theoretical constructs (which are general and should apply beyond your specific experiment) and empirical methods (which are how you are operationalizing the constructs in your theory).
Think about what relationships your theory postulates. It’s a good idea to put them into a diagram with boxes and arrows like you often see in published papers. (For example, a large fraction of articles in the Journal of Consumer Research have figures of this kind.) Do this at least for yourself, but if you have more than a couple boxes it’s a good idea to put the diagram itself as a figure in your dissertation.
Do be careful with these – people naturally think in causal terms and it is natural to develop theories this way. But how strongly you can make causal claims is dependent on your design. Even if you have a causal theory, it is usually challenging to draw strong casual conclusions from correlational data (i.e., no experimental manipulation), especially using the statistical methods you are likely to have available to you if you are a Masters student. But all this is for your discussion – for now it’s okay to postulate cause and effect relationships.
Now, using your diagram, write out a hypothesis for each arrow. You can write things like: “H1: Lower levels of X will lead to higher levels of Y.” or “H3: People in situation X will do more Y than people in situation Z, but only when they are low on trait W.”
You are ready to write now. First write one or two paragraphs explaining the model as a whole. If there are a lot of variables, group them into different categories and talk about those, rather than talking about each individual variable. Now, state each hypothesis and explain the logic underlying it. Briefly state which hypotheses there is already direct evidence for, versus those for which you will be the first to test. Draw some connections to your literature review, but don’t harp on the prior literature here. I don’t want you to think this section is easy to write – it probably requires the strictest logical analysis of any section of writing – but it’s also not especially hard, if you have already thought through your model and know how it relates to existing empirical evidence and theories.
– Research questions. This approach is a bit more freeform. Here you need to put more effort into organizing each of your research questions. Generally, these questions will be about some of the potential arrows in a diagram like the above. They’ll be asking things like is there a positive or negative effect of X on Y, is it W or Z that is the mechanism of that relationship, and would this relationship hold even under conditions Q? You write a separate section for each one of these questions, probably headed by the question itself. Your job is to enumerate the plausible alternative answers to these questions and explain why, based on prior research, theory, and logic, they are unsettled – why multiple different answers are plausible. If you want, you can say which answer you predict to be right and explain why you think it’s more plausible, but we don’t elevate this to the lofty status of a hypothesis. It’s just a hunch.
21. Writing a good methods section. The methods section is by far the easiest one to write. It’s still very possible to screw it up. The basic things you need to cover are the participants, materials, procedure, and design. Read a bunch of methods sections in published papers to see what kinds of things get included or excluded. One of the most telling signs of someone who hasn’t read through the literature is including too much or too little detail about methods, that goes outside the normal range of what is covered. Your professors have read thousands of academic papers and they have a sixth sense for whether something should go into a methods section or not.
Overall, you need to give enough detail that another researcher could replicate the basic features of your study. How many participants and from what population? (Usually most researchers will also give basic demographics – median age and percentage female, and other demographics only if relevant to your research.) What are the conditions? How did the manipulation work? Was there a manipulation check? What were the dependent variables and how were they measured? Be sure to give lots of examples. For important things, like stimuli (unless very long and cumbersome) and dependent measures, give the exact wording that participants saw for at least one item. Consider putting your full questionnaire in an appendix (but not in the methods section!).
If your study is complicated, it’s a good idea to use sub-headings. You probably know that APA formatting asks for very specific sub-headings (participants, materials, etc.), and this can work. But I recommend against slavishly following these in most cases. In almost all Masters dissertations I read, these subheadings create confusion rather than clarity. Often it’s much easier to explain the materials while you are going through the procedure, rather than trying to explain them before they have been contextualized in terms of what participants are doing.
There is one big difference between the methods section and every other part of your dissertation. All of the other sections are argumentation – making assertions and backing them up with evidence. The methods section is a narrative. You are just trying to explain what you did and give readers a good mental model of your task. Narratives are easiest to process in temporal order of when things happened – unless you’re Christopher Nolan remaking Memento from a movie into an academic dissertation – and you want to preserve the subject’s-eye-view as much as possible. Readers need to understand, from the participant’s perspective, what the task was like. The tricky part of this is that they also need the god’s-eye-view, because in between-subjects experiments, the story is different for different participants. So you can pepper your methods section with sentences like, “In the P condition, participants did X, then Y, then Z. In the Q condition, they completed these tasks in the opposite order, Z, then Y, then X.” Then you would go on to explain X, Y, and Z in separate sentences or paragraphs.
If you introduce conventions such as names for your conditions (e.g., “the domestic brand condition” vs. “the foreign brand condition”), use these terms consistently both within the methods section and later on in your results and discussion. Descriptive names make your methods easier to read, but can confuse more than enlighten if the terms are not clear in the reader’s mind.
22. Writing a good results section. I have some controversial opinions about writing a good results section, but I am in good company because I learned these principles from the classic Herb Clark piece that I linked earlier. You should read his advice too, but here’s my version.
– Preliminaries. Before you can write your results section, you need to understand what your data say. Get all of your stats in one place before you start writing. A lot of students struggle with statistics, and I’m very sorry if you are in this camp but there’s not much I can do about it from here.
– Organizing your results. There are usually many ways of writing up results. It’s possible that you thought through your statistics carefully before you started your study (that’s good!) or that there’s only one or two sensible ways of analyzing the data (that’s also good). But if you have a complex dataset, there is a good chance you have cobbled together 100 pages of SPSS output of 500 lines of R code and torn your hair out by now. That’s fine and totally normal, but now I need you to stop doing that. I need you to think about whether you think on balance your hypotheses were supported, given the data you now have, and select the most relevant statistics for presenting that argument. If there are results that are contrary to your hypothesis, it is good to be scrupulous about them and include these robustness checks somewhere, but unless it is a real killer problem, it’s usually better to put robustness checks in an appendix rather than the main text. Gently refer readers to those appendices, which might include other model specifications that either support or contradict the models in the main text, and briefly describe whether the results are mostly consistent or not with those in the main text. This isn’t an attempt to bury contrary findings, but rather to make the paper readable while referring hard-core readers to the information they need to decide whether or not to accept your argument.
– Outline your results section. You should write your results section in terms of your hypotheses or research questions from earlier. You might also have other results that are interesting and which you want to talk about. That’s okay, but mark those results as exploratory so that readers know these were not the main focus and that you didn’t predict these in advance. Usually these should go after your main hypotheses, but if it makes sense to integrate one or more of these into the sections evaluating each hypothesis (e.g., if you found an unexpected moderation on a main effect), then go ahead as long as it’s clear.
– Argue, don’t spray us with numbers. A results section is written in ENGLISH, not numbers. The numbers are absolutely necessary – heck, that’s the whole point of doing quantitative research – but they should take a backseat to the argument you are making. By “argument” I don’t mean you should be like a lawyer, hammering away at your position as hard as you can. You should be more like a judge, carefully weighing the evidence and presenting your overall verdict. This does not in any way obviate the need to use logical, verbal argumentation that links back to your hypotheses.
– Introductory paragraph and topic sentences. A corollary of the above is that your results section overall and each individual paragraph should begin with an expression in English of what is going on. Begin your results section with an overview paragraph that sets the stage and summarizes the overall findings. If it’s too complicated to summarize the key findings in one paragraph, divide your results section into smaller sub-sections and begin each of those with its own summary paragraph.
In individual paragraphs, the topic sentence should tell us what we’re going to be learning about in this paragraph. For example, “What was the effect of red versus white wine on sleepiness?” or “Next, I tested the hypothesis that giraffes are perceived as cuter than hyenas” or “The results were strikingly different for participants in the Sad Movie condition.”
– Writing up the results of statistical tests. Statistics is hard and students are often taught that there is a formula to writing up the results of statistical tests. In a way this is true – you do need to include the correct statistics, such as the t-score, degrees of freedom, p-value, and Cohen’s d for a t-test, or the beta, standard error, and p-value for a regression coefficient. But what you don’t have to do is mindlessly reproduce the output of your statistics program. Report the results that are relevant and skip the parts that don’t matter for the question you are asking. (There are a few exceptions to this; for example, it would be unusual not to report an interaction effect if you report a two-way ANOVA.)
The protagonists of your results section should be people, not statistical tests. Don’t say things like “An independent-samples t-test revealed that the mean on satisfaction was higher in the Pepsi Condition than in the Coke Condition, t(99) = 1.96, p = .050, d = 0.42.” Bad! Instead, say “Pepsi participants were more satisfied than Coke participants, t(99) = 1.96, p = .050, d = 0.42.” These two sentences say exactly the same thing, yet the second one takes 10 times less effort to read. The people are the focus of attention, not the statistical tests.
You will have noticed that in the above sentence I didn’t even say what statistical test I ran. Isn’t that bad?! No, not in this case. What else could the second sentence possibly mean but the first? If I did some sort of weird t-test variant instead of the standard t-test, I would have said so, and you should do so too if you do something out of the ordinary. But for super-standard tools like t-tests and Pearson correlations, you do not need to say what statistical test you ran but should keep this in the background to preserve readability.
For more complicated analyses, you do need to explain a statistical model even if it is a standard technique. For example, if you fit a multiple regression model, you need to explain what the variables were and how they were entered (e.g., if you mean-centered them). This is because there is more than one possible way to do these things. However, you don’t need to explain what a regression is. If you do something out of the ordinary, like a Wilcoxon test (a non-parametric test to deal with weirdly distributed data) or multilevel model (a sophisticated regression approach that allows variables to exist at different levels, like students nested within classrooms nested within schools), then you need to add a sentence explaining what that test is and another one explaining why you used it. If you do something weird, it’s a good idea to also do the normal thing and include it in the appendix so that readers can see how robust your results are to different analysis choices. If the results are different, that’s not necessarily a problem but you need to explain why you think your approach is better (e.g., the data are non-normally distributed).
23. Writing a good discussion section. You’re almost done, and by comparison with everything else the discussion section is a joy to write. There’s no standard form of how this needs to be organized, but here are some sections that commonly appear. (These are my labels for these sections, not formal headings. You can use whatever headings make sense to you.)
– Reminder. In my opinion (but not everyone’s), it’s a good idea to use the first paragraph of your discussion to remind the reader what the point of your paper is. If you wrote an effective introduction, they knew this once, but they’ve just spent 8000 words with you in the weeds and they barely remember what the point of your study was anymore. Remind the reader what is at stake and the general idea of your theory or research question. Often this paragraph will end with a question. (Question marks are heavily under-used in scientific writing for an endeavor that’s supposed to be about asking questions!)
– Recapitulation. Summarize the key findings of your study and answer the question you asked in the previous paragraph. The answer could be “it’s complicated” – and it often is in a dissertation, since you often don’t have the time or resources to conduct the multiple studies often needed to reach a firm conclusion – but if so, you need to explain how it is complicated. What do your results say about your hypotheses or questions? What did we learn that we didn’t know before?
– Alternative explanations. You won’t always have this section, but often a series of studies proceeds by demonstrating a phenomenon and proposing an explanation, then systematically showing that other plausible explanations for that phenomenon do not hold. If you’re not doing this, don’t worry about this section. But if you are, you need to explain all these alternative explanations in one place and this is where to do it. What are these alternatives and why are they plausible? But given that they are plausible, why do we now think that they are not correct? (Hint: Evidence.)
– Limitations. This is related to the alternative explanations section, but subtly different. That section was about problems that you already directly addressed in your experiments. This section is about limitations to the conclusions you can draw from your study. I divide this into “real limitations,” “fake-out limitations,” and “boring limitations.” It’s good to have some of each, but at least not exclusively boring limitations.
* A real limitation is an actual problem that really does limit the conclusions you can draw. A confound, an alternative explanation you didn’t rule out, or some results being statistically weak are examples of real limitations.
* A fake-out limitation is something you want to talk about because you think someone might think of it, but you don’t actually think it’s a limitation. A confound that isn’t really a confound, an alternative explanation that doesn’t make sense based on prior literature, or inconsistent results that nonetheless all point in the same direction (if you can show this statistically) could be examples of fake-out limitations.
* A boring limitation is something that’s true of basically every study. You don’t have a perfectly representative sample of the entire Earth? You don’t have 10,000 participants? Your experiment isn’t a perfect encapsulation of the real world? These limitations can be real or fake, but they are always boring because they are always true and generating them does not require any critical thinking. The reader already knows about these problems so you don’t need to harp on them.
You can briefly mention some boring limitations if you genuinely think it is important to do so, but you should also try to say something that requires some thought – a real or fake limitation. (By the way, a usually boring limitation can be a real limitation if there is a good reason why it is especially problematic for your study in a way that isn’t true for every other study ever.)
– Theoretical Implications. What do your results say about scientific theory more broadly? This is often largely a repeat of the theory development section, listing the ways your dissertation adds to the existing scientific literature. If you include this section, please try to say something new here beyond what you said in the introduction.
– Future Directions. In my own papers, I often combine this with theoretical implications. What can other scientists do to build on your findings? Now that your question has been answered, what new questions does it suggest? What are some interesting follow-ups that could address your limitations? (Please do not say “this study should be replicated with a bigger sample.” That is always true [it's a “boring" limitations] yet no one will actually do this.) Examples of good future directions could be other methods of testing the same idea, moderators or boundary conditions (i.e., conditions under which your effect might be bigger or smaller or not exist at all), further tests of the underlying mechanism (i.e., further explaining why your findings happen), or applications of your study to other domains. The nice thing about this section is that it’s easy to make it either long or short because there are always lots of interesting directions to take research – if you can’t think of any, you haven’t thought hard enough.
– Practical Implications. A lot of science really doesn’t have any direct practical implications, and that is totally fine (unless you are studying business/management, in which case you probably want to have something serious to say here). The point of science is learning stuff, and we don’t know where that stuff will lead. Someday someone might find a practical application, or someone will build on your work and their work will have a practical application. That said, if you can think of practical implications of your work – e.g. for businesses, governments, NGOs, or individuals – then you can explain those implications here. However, it is also important to be realistic and not over-claim here. You can speculate – remember, the discussion is a manifesto and you can be opinionated here – but speculate responsibly. Don’t give advice that you wouldn’t follow yourself and think about reasons why it might not be a good idea to do what you are saying. You don’t have to harp on these limitations too much, but do show an awareness of any major ones.
A lot of academic writing – including published articles – is very bad. If you follow the suggestions above, your writing can be better than the typical academic article. Now, I don’t exactly expect that to be the case because there is a very slow learning curve on writing in general and technical writing in particular. But if you keep these principles in mind, you absolutely can write a better than average dissertation.
Some final words of encouragement. If I wrote this giant-ass blog post, you can write your dissertation. In fact, this post is longer than your dissertation will be – it’s almost 13,000 words. (That was a surprise when I put this into a word processor to check!) You’ve been thinking about your project for months and the hard part is already over. You came up with an idea (if you’re one of my 7 students writing a Summer 2020 dissertation, then it was a really good idea!). You designed an original experiment to test that idea, you came up with stimuli to test your idea, you programmed the study, and you’re a long way toward analyzing your data. The thing is: it doesn’t count until it’s written down. Science is a collective human activity – if it never gets written down, it doesn’t exist. All that’s left for you now is that last push. Grind away!