13  Anatomy of a Research Paper

(Modified on 2026-08-11)
3,967 words

In this reading, I give advice on how to layout a good research paper. I’ve also provided some more general writing guidelines and advice on our GitHub repo. In the Resources directory there is a writing style guide, as well as a document with advice/guidelines for writing in plain-text formats Please reference these guidelines in all your writing for this class.

For correlational and experimental studies, the research report typically contains the following components:

This layout is variable—not all research includes all the same details. These sections are not always laid out in the same order or explicitly separated into sections and sub sections; for example, the “background” is often incorporated into the introduction section. Longer papers might report multiple experiments, so you might have multiple methods/results/discussions.

Now in more detail.

Title

At least two issues are important in composing a good title. First, what is the size of the community of readers you are targeting? Some titles are obviously intended for a small group of readers: Coactivator and corepressor regulation of the agonist/antagonist activity of the mixed antiestrogen, 4-hydroxytamoxifen. Especially in interdisciplinary fields, it is important to have titles that make sense to a large number of readers. Always try to reach out beyond the narrow audience of professionals in your specialty.

The second consideration is whether the title identifies a topic or states a conclusion. A title such as The effect of scale degree on melodic accent identifies a topic, but reading the title does not tell you what the researchers discovered. By contrast, consider the title of a paper by Jane and William Siegel published in 1977: Categorical perception of tonal intervals: Musicians can’t tell sharp from flat (Siegel and Siegel 1977). In general, “conclusion” titles are better than “topic” titles. It is very helpful for readers if the title describes the main finding. Of course, not every study will produce a clear conclusion. Use a “topic” title when the main finding is too complicated to distill into a few words.

Good topical titles often pose a question—the question that motivated the research. Examples include: What is melodic accent? (Huron and Royal 1996), Is music an evolutionary adaptation? (Huron 2001a), What is a musical feature? (Huron 2001b) and Who enjoys listening to say music and why? (Vuoskoski et al. 2012). A good empirical paper will trace the research path: from question, to theory, to conjecture, to hypothesis, to protocol. The title may be a good place to present the motivating question.

Abstract

Provide a single paragraph that describes the essential elements of your study. Abstracts are typically 100–250 words in length and report in simple terms the hypothesis, a cursory description of the method, and a clear statement of the conclusions. Especially in the arts, scholars are sometimes tempted to organize the abstract like a film trailer—something intended to tantalize a reader into reading the full paper. This is the wrong approach. A bad film trailer is one that gives away the ending of the film. By contrast, a bad research abstract is one that doesn’t give away the ending. Good abstracts always report the main findings. If the results are negative, say so. If an abstract doesn’t report a conclusion then it suggests that the research may not have anything useful to say.

Introduction

The goals of an introduction are to:

  • Introduce and frame the topic area.
    • Help the reader (quickly) understand what the paper is about.
  • Motivate the topic/question/problem.
    • Convince the reader that the topic is interesting.
  • Contextualize your ideas.
    • Show the reader how your specific ideas relate to other ideas and literature.
    • Also, convince the reader that there are still unresolved issues in this area of research (which your study will help resolve).
  • Set up and frame the research question(s) that your project aims to answer.
    • Help the reader understand what the (open) questions you hope to answer are, and how they differ from, or build off of, other work.
  • Set up the idea of the paper: an answer to a research question which you will test.
    • The reader needs to know what your proposed solution/answer is and the gist of how you will test it. (Details of how your test works come in the methodology section.)

The introduction should start with “big picture” ideas that motivate and contextualize everything, then gradually introduce increasingly specific details until it arrives at your specific project idea. This can be conceptualized as a “funnel” or “inverted triangle”: start “wide”—talking about the big picture—then gradually “narrow” down to the main point. The funnel introduction will often retrace the main intellectual path of research: Question \(\rightarrow\) Theory \(\rightarrow\) Hypothesis \(\rightarrow\) Protocol.

In a research article, the introduction will typically be 3–5 paragraphs long. The breakdown of paragraphs in a typical introduction could look something like:

  • [1] Introduce and motivate topic.
  • [2] Raise/ask research questions.
  • (¶[...] Zero or more paragraphs, elaborating here.)
  • [-1] Propose research project which aims to answer the question(s).

The last paragraph often begins with a one- or two-sentence preview of your study. For example, the final paragraph might begin:

  • In brief, we carried out a study to determine…
  • In light of this question/debate/issue, three studies are described here…
  • In the first study…

It can also be effective to give a quick preview of your main conclusion(s) at the end of the introduction. So end the introduction:

To anticipate our results, this paper presents evidence that…

Hypothesis

The introduction should lead you right up to the statement of hypothesis. This often happens in the later parts of the introduction, but can also be in its own separate (short) section, especially if there are multiple hypotheses, and they are complicated to explain. So you might see a Hypotheses section like this:

In light of the Smith-Jones theory of music-induced foot-tapping, we might propose the following hypotheses:

  • H1. Listeners are more likely to tap their feet as the beat approaches 88 beats per minute.
  • H2. As the tempo increased, listeners will be more likely to tap their feet than bob their head.

By the time you get to stating your hypotheses, you should already have clearly explained and justified everything in it. This means defining specific terms used in the hypothesis, and justifying your choice of operational definition. The point is, by the time you state the hypothesis:

  1. The reader should be able to understand exactly what you are predicting, how you will measure it, and how it relates to the theoretical ideas behind the project.
  2. The reader should have been convinced that testing your hypothesis is valuable, and that your approach makes sense (especially your operational definitions of important variables).

Background

The “Background” section of a paper may also be called names like “Literature,” “Literature Review,” “Prior Work,” or “Related Work”—different journals may have different standards.

The goal of the Background section is to support your idea/project. This means that the Background section of a paper is not about the literature; Rather, it is about contextualizing your project. This means that a Background section is not just a list of relevant citations. Rather, all references to literature should relate to a point you are making that serves one or more of the following rhetorical purposes:

  • Clarifies the topic area.
    • Help the reader understand exactly what topic you are studying by elucidating how that topic is related to (or not related to) other topics and subtopics that have been studied.
  • Justifies your project.
    • Show the reader why your project fits into and/or extends previous work in a way that is sensible and valuable.
  • Justifies your methodology.
    • Show the reader how your methodological approaches compare to approaches that have been applied to this topic before—including good and bad approaches.
  • Demonstrates your knowledge.
    • Convince the reader that you know enough about your topic to make intelligent research decisions and conclusions. (This is the only way that literature that is not directly related to your project might appear. You might need to include it just to show the readers you know about it.)

The same logic applies at the level of sentences. Sentences should not be about a citation: citations should appear to support a useful point that the sentence is making. Since every sentence is about one idea, you can have multiple citations in each sentence to support the single idea—you don’t need one sentence for each citation. It can also mean that the same citation appears multiple times, in different sentences/paragraphs, because it supports different ideas. More specific advice on how to use citations can be found in the Appendix.

As in all writing, each paragraph in the Background section should focus on a single communicative/rhetorical goal, relevant to your research idea. Citations in each paragraph should serve to articulate or support the point of the paragraph. Structure your Background section by identifying the key points you want to support, and write your paragraphs around those points. For example, you might need one paragraph about the theoretical background of the project (citing all the associated literature), another paragraph about the mathematical model you are using (citing literature using that model, or related models), and a third paragraph talking about the experimental methodology (citing papers using the related methodologies). In this hypothetical example, the point of these paragraphs to explain and justify your (1) theory, (2) mathematical model, and (3) methodology. The citations in each paragraph are there to explain and justify these ideas.

Other notes:

  • Only include references to works you cite in the article.
  • Be sure that you have read everything you cite.
  • Don’t ignore the early and historical literature.
Independent section?

The Background section of a paper may exist as its own independent section, with header, after the introduction. However, it is also common—and can be very effective—to work the Background material directly into the Introduction, in the middle paragraphs. This obviously results in a much longer introduction. Generally, you should only put Background (e.g., references to other work) in the introduction that is important to the goals of the Introduction; Any other background can wait for the background section. If all or most of your background research is directly needed in the Introduction, you can end up with no independent Background section.

Write the Paper first

Since your Background section is organized based on ideas, you may identify ideas you need to support before you actually find the supporting literature. You can (and should) still Write the paper first. For example, during the early stages of working on a paper, you could write:

The average tempo of music is about 92(?) beats per minute (citation needed).

Of course, if you can’t find a supporting citation, you’ll need to go back and change that sentence (and anything else that depends on that point)!

Methodology

The goals of a Methodology section are to:

  • Clearly and unambiguously explain exactly what you did.
    • In theory, a reader should have enough information to exactly replicate your work!
  • Explain why you did it this way.
    • If a reader doesn’t understand why you did something, they may fail to appreciate the value of doing it the way you did it, or misunderstand the point.
  • Explain exactly how all variables are operationalized.
    • Your reader needs to understand what all your variables really are, and how they were measured.
  • Articulate the limitations of your approach.
    • No research is perfect—honestly pointing out flaws/problems (even outright mistakes) will help gain your readers’ trust.
  • Demonstrate your competence.
    • Show your readers that you are familiar with relevant design paradigms and why they are used, and that you have thought about every detail.

A Methodology section needs to be very detailed and precise. Use appropriate terminology throughout: e.g., “independent variable,” “levels,” “within-subject,” “operational definition,” “establishing construct validity”, “test-retest reliability”, etc. Define all terms precisely, especially your operational definitions for all variables.

Include as many specific details as you can—this demonstrates that you have thought things through, and establishes your credibility. Even if you aren’t sure about a detail, put in something—in early drafts, use markers like (?) to mark these uncertain spots.

It is not just enough to convey what your methodological decisions are: You should also explain why you make each methodological choice. Even if you make a design decision purely out of convenience or because previous research made the same choice (be sure to cite them), mention it.

Organization

The Methodology section usually has several subsections, which vary a lot depending on the nature of the project. These sections may, or may not, be separated into explicit subsections, with their own sub-headers. Examples include things like:

  • Sampling
  • Dataset
  • Recruitment
  • Participants
  • (Experimental) Design
  • Controls
  • Protocol / Procedure
  • Model
  • Evaluation

Before getting into these detailed subsections, it can be effective to start the Methodology section overview of the method in general. For example:

In brief, the study involved floor-level video capture of foot activity for seated patrons in a discotheque. The tempo of the music was manipulated and the videos coded in order to relate foot-tapping activity to musical tempo.

This overview is followed by the more detailed paragraphs/subsections. In some cases, there is no need to start with a general summary/overview, because you’ve already covered the general ideas in the Introduction and Background.

Citing methods

In many cases, you may base your methodologies on previous research. This takes the full burden of justifying methods off your shoulders. However, it does not remove the burden of explaining your methods. It is important that your readers understand the gist of and reasoning behind the methods you use without having to go read somebody else’s paper! Don’t just say things like “We used the same methodology as Wilson et al. (2018),” then move on. Tell us at least the most important details of Wilson et al.’s method.

Write the Paper first

The Methodology section is highly detailed and precise. This can be a challenge to our Write the paper first mantra, especially in the early stages of a project.

As you begin writing a paper, if any details of your methodology are still fuzzy, that’s okay. You can say things like:

  • I compared my new system to the existing state-of-the-art (?) system.
  • I used position on the Billboard (?) chart to operationalize popularity.

You can also fill in hypothetical details/numbers:

  • I recruited 32(?) participants…
  • I sampled 32,232(?) tracks from the Spotify API.
  • My genre variable was measured with eight(?) levels.

You can then search for these tags later in the writing process, and update them as needed.

Data / Sample

Most scientific research papers will need a subsection of the Methodology talking about the dataset used.

It is always important to describe the method you used to sample data.

  • What procedure did you use when selecting the music for an analysis? What composers? What genres? What part of the recording/score?
  • How did you recruit participants? Where? When? How?

Make sure to use appropriate terminology: e.g., convenience sample, quota sampling, stratified sampling, systematic sampling, etc. Also, make sure to identify any exclusionary criteria—that is, conditions for excluding particular people, performances, recordings, scores, etc. For example:

  • For the purposes of this study, we excluded musicians who reported possessing absolute pitch.
  • In analyzing the videos, we excluded dancers who were holding objects like drinks or purses, as well as people who were deemed to be merely walking across the dance floor.
  • We excluded tracks with multiple genre tags.

You should also provide basic information about your sample: how many songs/people/data-points in total? For research participants, provide some basic demographic information (average age, number of participants of each sex, musical training, etc.). For example:

  • Twenty-eight musicians participated in the study, 16 females and 12 males, with an average age of 21.6 years (range 18-27). Most (23) were instrumentalists; five were voice majors.
  • We analyzed 23,456 tracks in total, all released between 2003 and 2018, with 1,233–1,767 tracks sampled per year.

Design

Describe the design of your study. What variables and combinations of variables did you measure? (Within- or between-subject?) What controls or covariates were included?

Describe any experimental manipulations. This might be digital signal processing algorithms, musical stimuli played to participants, etc.

Stimuli

In behavioral psychology research, there is usually a separate “Stimuli” subsection, where the stimuli played to participants is described. If the stimuli were created by the researcher, what criterion were used in their construction? How long were the sounds? What timbres were used? Etc. Where possible, describe the stimuli in sufficient detail that another researcher might be able to duplicate them. Consider posting the stimuli on the web so that readers can hear them for themselves.

Procedure / Protocol

Describe how your study was conducted.

For behavioral research:

  • What were the precise instructions given to participants?
  • How long did the experiment last?
  • How many stimuli did they hear?
  • How were the data collected?
  • Etc.

Report the experiment instructions verbatim. For example: Participants received the following verbal instructions:

In this experiment you will view 20 photographs of people in everyday situations. Imagine that you are present in the scene. For each photograph, identify whether you think there is music playing in the background. If you think there is music, click on the “music” button. If you think there is no music present, click on the “no music” button. When you are finished, click on the NEXT button to view the next photo.

For computational research, how was your data divided into training/testing sets? What window sizes were used? Were arrays zero-padded? What DSP libraries were used? What ML or statistical algorithms were used?

Results

A “Results” section may also be called things like “Analysis” or “Data Analysis.”

Rhetorically, the goal of a Results section is to:

  • Present and justify the empirical (data-based) conclusions of the paper.
    • The reader needs to know what the main empirical takeaways are, and why they should believe them.
  • Clarify the scope of your claims:
    • The reader should know how generalizable (you think) your results are—in particular, they should understand any biases in the sampling strategy, and limitations of the analysis.
  • Earn your readers’ trust:
    • You must convince your readers that you made an honest, realistic effort to disprove your hypothesis—otherwise, they won’t trust you.

These high-level goals can be achieved numerous ways, but a specific pattern that is common in experimental research is to lay out the results like this:

  1. Describe the data.
    • Including exploratory/descriptive visualization.
  2. Visualize patterns in the data:
    • In particular, visualizations of any predicted patterns.
  3. Validate the data.
    • Argue that you have indeed measured what you wanted to measure, with the necessary precision.
  4. Argue that apparent patterns in the data cannot be explained by random noise.
    • This is the purpose of a formal statistical test of the primary null hypotheses.
  5. Test any supplementary hypotheses or post hoc observations.

Remember, you should never claim that your results prove anything! At best, your results are “consistent with” your hypothesis/theory.

Describe the data

Your reader should understand your data.

  • How much data is there (in total, and in different subgroups)?
  • How much data is missing?
  • How much error is there in the data?
  • How are different variables distributed?
  • Provide graphs which help the reader get familiar with the data.

(The procedure for how you got the data belongs in the Methodology section.)

Validate the data

One concern is always that measurement protocols we used turned out to be either too noisy or worse, biased. Test for reliability: are you reliably measuring what you think you are measuring? Present evidence of both intra-subjective reliability and inter-subjective reliability, or perhaps consistency/reliability of ML-model output. For example:

In order to determine intra-subjective reliability, we calculated the correlation between responses for duplicate stimuli. The mean correlation for all 30 participants was +0.38.

You may want to consider excluding data from unreliable participants or bad output. Generally, exclusionary criteria should be established a priori:

Prior to hypothesis testing, we resolved that data would be discarded for any subject whose test-retest correlation was less than +0.1. According to this criterion, data for two participants was eliminated.

However, you might find the need to make post hoc exclusionary decisions:

Although the mean test-retest correlation was +0.72, this number was depressed by two participants who were far less reliable than any other participants, at +0.31 and +0.22. Before testing our main hypothesis, we made the post hoc decision to exclude the data from these two participants. Having excluded this data, the mean test-retest correlation was +0.82.

The classifier’s output very frequently confused two classes: death metal and goth metal. Since these genre labels are nearly identical, we made the post hoc decision to exclude these categories during model performance evaluation.

Not just noise

In working with data, our main concern is that the apparent patterns that we think to see (worse, that we want to see) could actually be nothing but random noise. Our entire Results section should adopt the default, self-skeptical attitude that our analysis could be nothing but noise (the null hypothesis is true). Only if we can make as strong argument that there is more than just noise in the data, can we make such a conclusion. In order to make that argument (that the null hypothesis is false), we need to conduct and report formal tests: either statistical tests or cross validation.

Details of tests should be reported for each hypothesis. The American Psychological Association (2010, 34) recommends that all empirical papers also report effect sizes. In addition, in order to facilitate possible future meta-analyses, research reports should provide complete descriptive statistics (number of subjects, means, standard deviations), and report statistical values to three significant digits, including non-significant results.

Post-hoc tests

Often, once you see your data, you’ll see either problems or interesting (but unanticipated) patterns in the data. If these problems/patterns are extremely relevant to the main research questions, it is a good idea to report them. However, you must 1) first report your a priori results then 2) explicitly introduce your post hoc analyses as such. For example:

Having completed our main hypotheses, we next pursued one post-hoc analysis. We noted an unexpected correlation between participants’ musical genre preference and their performance on the experimental task. This correlation appears to be statistically significant (\(F^1_{32} = 5.44\), \(p<.05\)).

Conclusion

The conclusion is essentially an expanded abstract. Like the abstract, the conclusion will reiterate the question or hypothesis, recap the method, and state the results. Where the abstract may state only the main finding, the conclusion restates all of the findings. The conclusion is also a good place to state the big-picture “main takeaways” from the Results section in more general terms (the Results section itself should be extremely detailed). The conclusion should also assemble and restate all of the caveats and assumptions of the Methodology and Results. Typically, the conclusion is roughly three times the length of the abstract.

A good conclusion should be written as a “stand-alone” section. Having read the title and abstract, an interested reader should be able to skip directly to the conclusion and find a good summary of the research. In stating the conclusion, always use that circumspect language of empirical research. No “the results prove…, establish…, demonstrate…, or show…”. Instead—“the results are consistent with the theory that…” or “the results are consistent with the hypothesis that…”

Discussion

Only a minority of research articles include a Discussion section. Unlike the Conclusion, the Discussion section provides a more open forum to consider the repercussions of your work. The Discussion section allows you to speculate about what you think may be going on. You may also propose post hoc interpretations of your data. For example, post-experiment interviews might have alerted you to other possible phenomena:

In post-experiment interviews, several participants indicated that they had been singing-along with the stimuli. A post-hoc test showed that these participants performed better than the other participants. Moreover, when all of the trained vocalists were included, these participants collectively performed well above the chance level. It may be that “singing”—either overtly or clandestinely makes it easier for people to perform the task.

You may also make suggestions for future research.

Helping Readers

In general, an empirical research paper should be written to provide readers with an efficient way of accessing information. Everyone is busy and no one has time to read everything. Don’t expect scholars to read your entire paper. Some readers will be casually interested in your work. Others will be looking for results related to a particular interest they have. Yet other readers will be very interested in the subject and will read your report carefully—looking to ensure that the method is solid, that the research was done carefully, etc.

Different levels of interest are echoed in which parts of a research paper people read. Suppose that 100 people encounter your paper and read the title. Of that 100, perhaps 50 will flip through the article looking at the figures or illustrations. Perhaps 35 will read the abstract. Of those 35, 10 may also skip ahead and read the conclusions. Perhaps 10 people will begin reading the entire paper, but only 3 will finish it. A good research paper will cater to all of these people. Ideally, a good title will convey the essence of your work. This is not always possible since results are often messy. Nevertheless, readers appreciate titles that are informative. This is the reason why the titles for research papers are often so long. Similarly, the Abstract should be crafted so that readers have some idea of what question/hypothesis you addressed, the approach your study took, and what your main results were. The Conclusion should act like an expanded Abstract.

Figures and Captions

Newspaper, magazine, and book editors have long known about the importance of good photos and illustrations. Readers are attracted to the pictures, and that is just as true of research articles. Spend time creating interesting and instructive figures. However, the most under-valued aspect of a good research article is not the figures but the figure captions. Wherever possible, write a figure caption that tells the whole story. Imagine a reader who has read the title and perhaps skimmed the abstract. They have read nothing else in your paper, but now they are looking at a figure you have provided. Try to make your figure caption summarize the main result so that a casual reader will understand what you have achieved.

Mantras

Write the paper first

There is no simpler way to improve the quality and efficiency of research than writing the paper first—i.e., thinking through and planning everything in advance. That doesn’t mean it’s easy to make yourself do it! It’s not research until you write it down.

Err on the side of clarity

Using concise and precise technical terms is a good thing, if it enhances clarity. When in doubt, it’s better to over-define and explain than to under-explain, to make totally sure your readers/reviewers understand what you are saying.

Define acronyms and jargon. Don’t assume everyone will understand the words you use the way you understand them—explain what you mean by different terms. Don’t be afraid to repeat yourself.