Writing a thesis

Here is some advice for bachelor and master-students that write a thesis at the department of Biophysics at the Donders Institute for Brain, Cognition and Behaviour.

English

English is the scientific language, so it is essential that you learn to write in english.

Useful links & literature

Organisation of a manuscript

Section Purpose
Abstract/Summary summarizing experiment, results, conclusions
Introduction establish the topic, explain what has been done inf the field, clarify what interesting question remains, and how you will tackle this
Methods A description of the subjects, experimental set-up, paradigms and data analysis
Results A presentation of the results and analysis (both in words and graphics)
Discussion reporting main conclusions, comparison with other studies, putting your results in a bigger picture, (perhaps suggesting a model)

 

Front page

On your front-page, put:

  • title
  • name
  • student number
  • starting date
  • end date
  • internship duration
  • study credits / points / EC
  • study, e.g.:
    • Biology
    • Psychology
    • Cognitive neuroscience
    • Sciences
    • Biomedical sciences
    • Medicine
    • Physics
  • Bachelor / Master
  • supervisors

Introduction

The paragraphs in your Introduction should contain the following:

  • Explain the main issue/problem that you want to tackle.
  • Describe what others have done before you to put the issue in context
  • Describe what you will do in this study (you are allowed to use "we" or "in our study"! Nevertheless, don't do this too often), and make sure to explain why this study is better than other studies, and/or why it is interesting to tackle this scientific issue in this manner. Perhaps a figure detailing the rationale of the experiments might be useful.
  • Globally state what your main findings mean (e.g. they are surprising and interesting). You don't have to explain everything, yet: you want to have some surprises left for the reader.

Methods

The Methods-sections is usually straightforward, containing the following subsections:

  • Subjects/Listeners
  • Set-Up
  • Stimuli
  • Paradigms
  • Data analysis
  • Statistics

Reporting statistics

If you want to use the classical null-hypothesis significance testing, then here are some reporting guidelines from the journals of the American Physiological Society:

Table 1. Interpretation of P values

P Value Interpretation
\( P \nsim 0.1 \) Data are consistent with a true zero effect
\(0.05 \sim P \simeq 0.05 \) Data suggest there may be a true effect that differs from zero
\( 0.01 \nsim P \simeq 0.05 \) Data provide good evidence that the true effect differs from zero
\( P \simeq 0.01 \) Data provide strong evidence that the true effect differs from zero

 

Table 2.Guidelines for rounding P values to sufficient precision

P value range Rounding Precision
\( 0.01 \leq P \leq 1.00 \) Round to 2 decimal places: round P=0.031 to P=0.03
\( 0.001 \leq P \leq 0.009 \) Round to 3 decimal places: round P=0.0066 to P=0.007
\( P \lt 0.001 \) Report as P<0.001; more precision is unnecessary

It is much better to use Bayesian statistics. With current Markov Chain Monte Carlo sampling techniques it is easy to do Bayesian statistical analysis. For a course on Bayesian statistics, see "Bayesian Cognitive Modeling: A Practical Course" (which contains a free downloadable book). A very good book on Bayesian statistics is from John Krushke: Doing Bayesian Data analysis with R.

Bayesian null hypothesis testing is often done with the Bayes factor, which has the following interpretation.

Table 3. Interpretation of Bayes factor (Jeffreys 1961, Bayesian Cognitive Modeling: A Practical Course)

Bayes factor BF12 Interpretation
\( \gt 100 \) extreme evidence for model 1
\(30 \leq BF \leq 100 \) very strong evidence for model 1
\( 10 \leq BF \leq 30\) strong evidence for model 1
\( 3 \leq BF \leq 10 \) moderate evidence for model 1
\( 1 \leq BF \leq 3 \) anecdotal evidence for model 1
1 no evidence
\( 1/3 \leq BF \leq 1 \) anecdotal evidence for model 2
\( 1/10 \leq BF \leq 1/3 \) moderate evidence for model 2
\( 1/30 \leq BF \leq 1/10 \) strong evidence for model 2
\( 1/100 \leq BF \leq 1/30 \) very strong evidence for model 2
\( \lt 1/100 \) extreme evidence for model 2

Results

In the results-section you have to guide the reader along, explaining what results you have found and what kind of analyses you did. You should start with the basic data, and gradually increase the level of analysis.
So, usually the first figure in the results-section contains the raw data of one typical subject (such as a head movement trace, or a stimulus-response plot). With this figure you can introduce a more elaborate analysis, such as linear regression. Next you have to make the results quantitatively, so you want to have a measure for all subjects/conditions that you can easily plot in for example a histogram. This "first typical example"-"then quantification" abounds in scientific articles.

Don't: "Regression analysis was done, and the results of listener JB are plotted in figure X. "
Do: "listener JB accurately localizes the sounds, as evidenced by a high stimulus-response gain (gain = 0.9, see Methods, Figure X)".

The storyline is important. Each paragraph is preceded and followed by another paragraph. Link them together (use: therefore, however, next), and conclude.

Data visualization

A very important aspect of scientific reporting of data is data visualization. In your Matlab-script, you should usually see these commands:

  • axis square
  • box off
  • legend
  • set(gca,
    • 'LineWidth',2
    • 'MarkerFaceColor','w'
    • 'Color','k'

Your figures should be saved as vector-format eps-files:

 
print('-depsc','-painter',mfilename)
 

There is almost never a reason to save a figure in bitmap-format. The eps-files can be easily modified in Adobe Illustrator to make the figures more attractive (without distorting the data). And if you want to use Microsoft, Illustrator has an option to save your figure for Office-purposes.

Also very important: label your axes!

  • xlabel
  • ylabel

Also, you should use a readable font size (minimally 10 at the final stage).

Furthermore, remove as much "dead white space" as possible. For example, often you can plot several similar graphs in one figure, you can leave out the tick-labels for several of these graphs by correctly positioning them.

Discussion

One of the hardest part in any thesis to write is the Discussion. In the Discussion-section you have to put your results in perspective.

  • State your main findings that are important and relevant
  • What have others done, and how does this relate to your experiments
  • You may put things in a broader perspective
  • Propose a model
  • Can your data be explained by other mechanisms?

Concise

Students have the tendency to embellish, exaggarate and re-iterate (especially in the Introduction-section). Try to write in a concisive manner. Try to get your main point across quickly/immediately. Explain everyting what you need to explain, and nothing more. Shorten long sentences. Check your text for sentences like:

  • "It is clear that"
  • "Note that"

These sentence parts are superfluous, and can be easily removed (which often makes the text easier to read).

Action

Don't over-use the verb "to be". Replace it with action verbs.

  • "is dependent on" - "depends on"
  • "localization performance is good for listener JO" - "listener JO accurately localizes sounds"

Quantification

  • Be as precise as possible. When using terms like "a number of", "many", "highly", try to quantify them (e.g. "8 out of 10").
  • Differences are only differences when they are significant, and they are only significant when they are statistically significant (state the type of test and P-value).

Edit, edit, edit

It is always a good idea to check and re-check your text. Your supervisors will also heavily edit your manuscript drafts. Don't be disappointed when your first draft is returned completely covered in red changes. Expect this to happen!
It is impossible to write your thesis in one day.

Style

Write formally, so avoid:

  • colloquialisms
  • contractions (e.g. use 'do not' instead of don't)

I have heard from many students they have been taught also to avoid 'we' and that you should write in a passive form. Rubbish! The personal pronoun 'we' and writing in the active form is quite common in scientific papers. The active form actually speeds up reading. So, 'the listener localized sounds accurately' is better than 'the sounds were localized accurately by the listener'.

The most important rule is:

Don't write something absurd.

All other rules can be broken.

Certainty

Also, because you can never be 100% certain of your conclusions, you should express yourself cautiously, using expressions such as:

  • seems to
  • might
  • appears to
  • is likely to
  • can
  • apparently
  • could
  • probably
  • perhaps
  • may (well)
  • seemingly
  • tends to

Note that caution should also be used when discussing other people's results and conclusions (for example in introductions or reviews). Even though a statement is written down in stone (in Dutch "staat iets zwart op wit", in black and white), this does not make it true.

With a research article, you should convince your readers that your experiments are well-performed and interesting. You should lead the readers along, convince them of the mysteries or problems you want to tackle, and conclude with a solution. The goal is to report your findings and conclusions clearly and to the point. Think of an outline with a logical flow. Each paragraph should contain a clear topic. Importantly, you should communicate a storyline, having a specific purpose in mind for every section and paragraph. Describe, compare and argue.

Cohesion

To achieve cohesion of a text, connect sentences and paragraphs. Here are some examples of connectives.

type Purpose Examples
and
  1. listing
  2. transition
  3. summation
  4. apposition
  5. result
  6. inference
  1. Enumeration:
    first, furthermore, finally
    one, two, three
    first(ly), second(ly), third(ly)
    above all
    last but not least
    first and foremost
    first and most important(ly)
    to begin/start with, in the second place, moreover, and to conclude
    next, then, afterward, lastly/finally

    Addition:
    Reinforcement:
    also, again, furthermore, further, moreover, what is more, then, in addition, besides, above all, too, as well (as)
    Equation:
    equally, likewise, similarly, correspondingly, in the same way

    either, neither, nor, not only ... (but) also
    indeed, actually, in (actual) fact, really, in reality
  2. now, with reference/respect/regard to, regarding, let us (now) turn to, as for, as to
  3. in conclusion, to conclude, to sum up briefly, in brief, to summarise, altogether, overall, then, therefore, thus
  4. i.e., that is, that is to say, viz, namely, in other words, or, or rather, or better, and, as follows, e.g., for example, for instance, say, such as, including, included, especially, particularly, in particular, notably, chiefly, mainly, mostly
  5. so, therefore, as a result/consequence, the result/consequence is, accordingly, consequently, now, then, because of this, thus, hence, for this reason
  6. then, in other words, in that case, else, otherwise, if so/not, that implies, my conclusion is
or
  1. reformulation
  2. replacement
  1. better, rather, in other words, in that case, to put it (more) simply

    to be concise, don't use reformulation

     

  2. again, alternatively, rather, better/worse (still), on the other hand, the alternative is, another possibility would be
but
  1. contrast
  2. concession
  1. instead, conversely, then, on the contrary, by (way of) contrast, in comparison, (on the one hand) ... on the other hand
  2. besides, in any case, (or) else, at any rate, however, nevertheless, nonetheless, notwithstanding, only, still, (al)though, yet, for all that, in spite of, despite that, after all, at the same time, on the other hand, all the same, even if, though

 

Like versus such as

'Like' implies a comparison, 'such as' implies inclusion.