Academic Writing for STEM
Full Reference Document

1. How research actually works

1.1 Research is a conversation, not a report

The single most useful idea in this document is this: research is a conversation that has been going on for decades, and your job is to join it, not to restart it. Every field already contains people who have asked your question, disagreed with each other, built on each other's work and left gaps. A research report is your contribution to that conversation: here is what is known, here is where the accounts disagree, here is the gap, here is what I did about it, here is what it means.

This reframes everything. A literature review is not a list of summaries. It is a map of who says what, where they disagree, what the evidence supports, and where you stand. A citation is not decoration; it is a pointer to the person you are responding to. When you understand this, referencing stops feeling like a chore and starts feeling like the mechanism of the whole enterprise.

1.2 The research cycle

Two things follow from this cycle. First, published work is not automatically true; it is the best current candidate that survived scrutiny. Second, a single paper is never the whole story. This is why a good report cites several sources agreeing or disagreeing, and not one paper that happens to support the conclusion you wanted.

1.3 Primary, secondary and tertiary sources

1.4 Peer review: what it does and does not guarantee

Peer review means that independent experts in the field assessed the work before publication. It filters obvious errors, methodological weaknesses and unsupported claims. It does not guarantee that the findings are correct, that the statistics are sound, that the code works, or that the authors have no conflicts of interest. Retractions of peer-reviewed papers happen every year. Peer review raises the base rate of quality; it does not certify truth.

This is why you evaluate sources yourself (Part 2) rather than assuming that the word "journal" settles it. It also explains why preprints exist and why you must flag them: a preprint has not been peer reviewed, so it is a claim, not a finding.

1.5 How a paper is structured: IMRaD

Most STEM papers follow IMRaD: Introduction, Methods, Results, and Discussion. Knowing the structure changes how you read. The abstract tells you what the paper claims. The introduction tells you what gap it addresses. The methods tell you whether you should believe the results. The results are the data, not the interpretation. The discussion is where the authors interpret, admit limitations and say what should come next.

1.6 The literature review as a map

A literature review answers four questions. What is agreed? What is disputed? What is missing? Where does my work sit? Organise by theme or debate, never by source. A review that says "Smith found X. Jones found Y. Patel found Z." is a list. A review that says "The field agrees that X holds under condition A, but Smith and Jones disagree on condition B, and neither addresses small deployments, which is the gap this report fills" is a review.

1.7 Note-taking that saves you later

Every time you read something useful, capture five things in your notes immediately: the full citation, the page number, the exact claim or quote, your own one-line comment on why it matters, and the tag that links it to a theme in your outline. If you capture the citation at the moment of reading, the reference list writes itself. If you do not, you will spend the last night hunting for a page number and will be tempted to guess, which is how fabricated references happen.

2. What counts as a good source

2.1 The hierarchy of evidence

Not all sources are equal. In STEM the rough hierarchy, strongest to weakest, is:

2.2 The six-question source test

Before you cite anything, ask six questions. If a source fails two or more, find another.

2.3 Predatory journals: how to spot them

Predatory journals charge a publication fee and provide no real peer review. They exist because publication is a currency in academia. Citing one is not automatically misconduct, but it weakens your evidence and a marker who knows the field will notice.

2.4 Where to search for STEM sources

3. Reference dates: when old is fine and when it is fatal

3.1 The principle

An old source is not automatically a bad source. Age is a signal, not a verdict. What matters is whether the claim you are using it for is still valid. A 1948 paper by Shannon is still the correct source for information theory. A 2015 paper on the throughput of 802.11ac is now superseded for current design decisions but still valid as a historical benchmark. The question is always: has this claim been superseded, and does my reader need the current version?

3.2 Acceptable age by source type

3.3 The citation window

Many journals and examiners use a rule of thumb: in a literature review, a substantial proportion of your sources should be recent, often at least half within the last five years. This is not a law; it is a signal that you have engaged with the current frontier and not just with whatever came up first in a search. A report whose newest source is from 2016, submitted in 2026, reads as though you stopped reading halfway. A report whose sources are all from last year reads as though you do not know where the ideas came from.

3.4 When using an old source is a mistake

3.5 When using only new sources is a mistake

The opposite failure is equally common and less often noticed. If every source is from the last two years, you have read the surface of the field and not its foundations. You will not know why the current method exists, what it replaced, or what problem it was invented to solve. You will also be more likely to repeat a mistake the field already corrected ten years ago. Foundational sources earn their place in a reference list.

3.6 How to check whether a source is still valid

1. Check the publication date and ask what has changed in the field since then.
2. Search for the paper's title to see whether it has been retracted, corrected or heavily criticised. Publisher sites and Retraction Watch are both useful.
3. Look at who cites it. If the only citations are from the same authors, it may not be influential.
4. Check whether a newer edition, revision or superseding standard exists.
5. For standards, check the issuing body's site for the current edition and any amendments.

4. Harvard referencing, in detail

4.1 Why we reference

Referencing does four jobs. It credits the person whose work you used. It allows verification, so a reader can check whether you have represented the source fairly. It shows the evidence base for your claims. And it separates your contribution from everyone else's, which is the foundation of academic integrity. A reference list is not a formality; it is the audit trail of your argument.

4.2 The two halves

4.3 In-text citation forms

4.4 Reference list forms

Harvard has no single international authority; institutions publish their own variants. The examples below follow the Cite Them Right pattern, which most universities use. Check your module handbook and follow it exactly, because consistency with the required style is part of the mark.

Book

Tanenbaum, A.S. and Bos, H. (2015) Modern operating systems. 4th ed.
    Boston: Pearson.

Chapter in an edited book

Smith, J. (2020) 'Routing in dense networks', in Jones, A. (ed.)
    Wireless networking handbook. London: Springer, pp. 45-67.

Journal article

Smith, J. and Patel, R. (2021) 'Throughput analysis of 802.11ax in
    dense deployments', IEEE Transactions on Wireless Communications,
    20(4), pp. 1234-1248. doi: 10.1109/TWC.2021.1234567.

Conference paper

Smith, J. (2022) 'A comparison of Wi-Fi 6 and Wi-Fi 5 in campus
    networks', Proceedings of the IEEE International Conference on
    Communications, Rome, May. pp. 88-95.

Standard

IEEE (2021) IEEE standard for information technology: wireless LAN
    medium access control and physical layer specifications, amendment 1.
    IEEE Std 802.11ax-2021. New York: IEEE.

Report or white paper

NIST (2017) Digital identity guidelines: authentication and lifecycle
    management. NIST Special Publication 800-63B. Gaithersburg:
    National Institute of Standards and Technology.

Website or online document

Cisco (2023) Wi-Fi 6 and Wi-Fi 6E: technical white paper. Available at:
    https://www.cisco.com/example (Accessed: 12 October 2026).

Thesis or dissertation

Smith, J. (2020) Performance of dense wireless networks in urban
    campuses. PhD thesis. Botho University.

Dataset

Statistics Botswana (2024) Population and housing census 2022.
    Available at: https://www.statsbots.org.bw (Accessed: 12 October 2026).

Software or source code

Canonical (2024) Ubuntu 24.04 LTS [computer program]. Available at:
    https://ubuntu.com (Accessed: 12 October 2026).

4.5 The rules that are easiest to get wrong

4.6 Managing references without pain

Use a reference manager: Zotero, Mendeley or EndNote. Install the browser connector, and capture the citation the moment you read the source. The manager stores the metadata, formats the reference in Harvard and inserts in-text citations into your word processor. This removes the two failure modes that cost students most marks: a reference list that does not match the citations, and a page number guessed at midnight. A reference manager does not do the intellectual work, but it removes the clerical errors that hide good work.

5. Writing well in STEM: plain language and lean paragraphs

5.1 The principle

Academic writing is not complicated writing. It is precise writing. The goal is that a competent reader in your field understands your claim on the first read, and can check it. Complexity that does not add precision is a defect, not a virtue. The best technical writing is plain, short and specific.

The second version is shorter, names the three factors, and makes a checkable claim. The first version says nothing a reader can use.

5.2 Sentence-level rules

5.3 The lean paragraph

A paragraph is one unit of argument. It has four parts, usually in this order:

Four to eight sentences is a typical lean paragraph. One sentence is not a paragraph in academic writing (except occasionally for emphasis). Twelve sentences is usually two paragraphs that have not been separated.

5.4 Cohesion across paragraphs

Cohesion is what stops a report reading as a list of disconnected observations. Three techniques do most of the work. Signposting: tell the reader where you are going ("This section evaluates the hardware; the next section evaluates the software"). Repetition of key terms: in technical writing, repeating the exact term is better than elegant variation, because variation introduces ambiguity. Transitions used sparingly: "however", "therefore" and "in contrast" earn their place when they mark a real logical turn, and become noise when used in every paragraph.

5.5 Structure of a STEM report

5.6 Tables and figures

Every table and figure must have a number, a caption, a source line if it is not yours, and a sentence in the body that refers to it ("Table 3 shows that..."). A table that is never mentioned in the text is decoration. A table that duplicates the text word for word wastes space. The rule is: present the numbers in the table, interpret them in the text, and never do both jobs twice.

5.7 Editing pass

6. Plagiarism, similarity and Turnitin

6.1 What plagiarism is

Plagiarism is presenting someone else's work, words or ideas as your own. It does not require intent. Copying a paragraph and forgetting the citation is plagiarism whether or not you meant it. The offence is that the reader is misled about the origin of the work.

6.2 Types of plagiarism

6.3 Paraphrasing, quoting, summarising

6.4 What is common knowledge

Common knowledge is a fact that a reader in your field would accept without checking a source. It does not need a citation. In STEM, "water boils at 100 degrees Celsius at sea level" is common knowledge; "Wi-Fi 6 was standardised as 802.11ax in 2021" is arguably common knowledge in a computing report but should be cited in a general report; "this campus has 800 staff" is a fact about a specific case and needs a source. When in doubt, cite. Over-citing is a mild weakness; under-citing is misconduct.

6.5 What Turnitin actually does

Turnitin compares your text against a large database: published journal and book content, web pages, and previously submitted student papers. It reports a similarity score and highlights the matching strings. It does not decide whether you plagiarised. It flags matches for a human to review. This distinction is essential.

6.6 Reading the similarity score honestly

6.7 How to reduce similarity legitimately

1. Write from your own notes, not from the source text. Never have the source open while you draft a paraphrase.
2. Quote only when the exact words matter, and keep quotes short.
3. Cite everything you use, and cite it at the point of use.
4. Reduce the size of the reference list only if it is genuinely bloated, not to game the score.
5. Use standard terminology freely; it is not plagiarism to use the correct technical term.
6. Do not use "synonym swap" tools or AI paraphrasers. They produce awkward text and do not fool a human marker.

6.8 The integrity of your own prior work

Reusing your own earlier assignment in a new module without disclosure is self-plagiarism. It is not always wrong to build on earlier work, but it must be disclosed: "This section draws on my earlier report for module X" with a citation. The offence is the concealment, not the reuse.

6.9 Collusion versus collaboration

Collaboration is working together to understand material. Collusion is producing submitted work together when the task is individual. If the brief says individual, discussing ideas is usually fine, but sharing text, code or files is not. When in doubt, ask the lecturer before you share, not after.

7. Ethical use of AI tools

7.1 The distinction that matters

The useful question is not "did you use AI?" but "what did you use it for, and can you defend what you submitted?" There is a real difference between using an AI tool to understand a concept and using it to produce the work you submit. The first is closer to using a textbook or a tutor. The second is closer to submitting someone else's work, and is what most academic integrity policies prohibit when they ban AI tools.

7.2 The cardinal sin: fabricated references

Large language models generate text that looks like a reference. Author names, journal titles, years, volumes and DOIs can all be invented and all look real. A fabricated reference is academic misconduct even when it was produced by accident, because the reader is misled about evidence that does not exist. This is the single most damaging AI failure in academic work, and it happens constantly.

7.3 Disclosure

If your institution allows any AI use, it will usually require disclosure: what tool, for what task, and how you verified the output. A short statement in an appendix is enough. Disclosure is not a confession; it is part of the integrity record, and it protects you if a marker notices AI-like phrasing in your work. Undisclosed use is what turns a permitted use into misconduct.

7.4 AI detectors

AI detectors are unreliable in both directions. They produce false positives (flagging human writing, especially from non-native speakers, as AI) and false negatives (missing AI-generated text). Passing a detector is not evidence of integrity, and failing one is not proof of misconduct. Do not build a strategy around beating a detector. Build a strategy around being able to defend every sentence in a viva.

7.5 The viva test

Imagine a marker asking you to explain any sentence in your submission, justify any design choice, and modify any part of your code. If you can, the work is yours in the sense that matters. If you cannot, no detector result will save you, because the marker does not need a detector to notice that you cannot explain your own report.

7.6 Using AI without losing the skill

The reason institutions restrict AI is not suspicion for its own sake. The skills you are being assessed on, reading, synthesising, arguing, designing, debugging, are built by doing them. If AI does the doing, you do not build the skill. The students who get the most out of AI use it as a tutor and a sparring partner, not as a ghostwriter. Ask it to explain a concept, to quiz you, to point out weaknesses in an argument you wrote, to suggest what you might have missed. Then do the work yourself.

7.7 A practical AI policy for yourself

8. Checklists, glossary and self-test

8.1 Before you submit: the integrity checklist

1. Every claim of fact has a citation, and every citation is in the reference list.
2. Every reference in the list is a source I have personally opened and verified.
3. No reference was generated by an AI tool without verification.
4. Every quotation is in quotation marks with a page number.
5. Every paraphrase is in my own words and sentence structure, with a citation.
6. I have not reused my own earlier submitted work without disclosure.
7. I have not shared text or files when the task was individual.
8. I can explain every sentence, every design decision and every line of code.
9. I have run the similarity check, read the report, and resolved every match that is in my own prose.
10. I have checked the module handbook for any AI-specific rule and complied with it.
11. I have completed and signed the declaration of originality if one is required.

8.2 Before you submit: the writing checklist

1. The introduction states the gap, and the conclusion answers the introduction.
2. Every paragraph has one claim and a topic sentence.
3. Average sentence length is under 25 words.
4. Acronyms are defined at first use and used consistently.
5. Claims are hedged to match the evidence.
6. Units and significant figures are consistent.
7. Every table and figure is numbered, captioned, sourced and referred to in the text.
8. The reference list is alphabetical, in one style, with hanging indents.
9. No source is cited only because it is old, and none only because it is new.
10. The document has been read aloud once.

8.3 Glossary

8.4 Self-test

Answer these without looking back. If you cannot, reread the relevant part.

1. What are the four jobs of a reference?
2. What is the difference between a primary, secondary and tertiary source, with one STEM example of each?
3. Give three situations in which an old source is the correct source, and two in which it is a mistake.
4. What is the citation window, and why is it a rule of thumb rather than a law?
5. Write the in-text citation and reference list entry for a journal article by two authors, from memory.
6. What is the difference between a paraphrase, a quote and a summary, and when do you use each?
7. Why is a similarity score of 30 percent not automatically a problem, and a score of 5 percent not automatically fine?
8. What is the cardinal sin of AI use in academic work, and why is it so damaging?
9. What are the three questions of the viva test?
10. What does a lean paragraph contain, in order?