What Eight Theses Teach About Writing One
Eight examined master's projects, one of them with its whole apparatus attached, answer a question no single thesis can: which parts of the job survive contact with a deadline, and which do not.
What eight works show that one cannot
A single thesis teaches you what one person did. Eight written under comparable conditions let you see which decisions were forced by the design, which were forced by the form, and which were forced by the calendar. Three things become possible here that are impossible with one document.
- A full chain trace. One project arrives with its instrument, raw file, workbook, charts and schedule. Six reported figures can be walked from a question on a form to a sentence in the document.
- A displayed instrument and a displayed consent apparatus. Fourteen weeks of teaching material showed neither. This corpus supplies five instruments, one participant information sheet and one full consent text.
- Comparison across one template. Seven works carry the same section skeleton, which lets you see what the form accommodates and what it silently omits to ask for.
Structure: the form is a container, not a design
The shared skeleton runs: introduction, research background, aims and objectives, Research Boundaries, literature review, research method, discussion, conclusions and further research. Two works carry all eight headings and invert the order, putting method before literature review — and both are works in which the literature is the data rather than the background. That is the form breaking usefully.
Elsewhere it breaks less usefully. One work carries six of the eight, with a limitations section standing in the scope slot and no discussion heading at all. Two print a chapter headed "Main chapters" — the template's own placeholder, surviving into a passed document. One has no results heading anywhere, so its findings cannot be located from its contents page. And one work's contents page, typed by hand rather than generated, disagrees with the body in nine places and lists a numbered section the document does not contain.
Method: the smallest chapter, holding the largest gaps
PROPORTIONS ACROSS SEVEN WORKS ON ONE TEMPLATE (each figure is one work's own; none is a target)
| Chapter | Range across the seven | What the guidance says |
|---|---|---|
| Literature review | 13.9% – 27.9% of the document | "The majority" of the document should go in the literature review and method sections — unquantified |
| Method | 2.6% – 14.0% | As above; the guidance draws no distinction between the two |
| The two combined | 19.8% – 33.3% | Between a fifth and a third, against guidance that says the majority |
| Limitations | 0 – 569 words; 0% – 6.3% | The guidance places limitations in the discussion. None of the six works that have limitations does so — all sit in the conclusions |
Derived from measured section lengths. The smallest method chapter is 280 words, and its real method is distributed across three other chapters, so any comparison must say which figure it is using.
The more useful finding is not the length but the content. Across seven works, none names a sampling technique; none reports a pilot of any instrument; none supplies a participant information sheet; only one prints a denominator against which a response rate could be computed, and the arithmetic printed beside it is wrong. Several name an analysis procedure and never operationalise it — thematic analysis with no code, no frame, no coder and no theme.
Evidence: display is what makes a claim checkable
Where these works could be audited, the results are worth knowing. Twenty of twenty means recompute exactly from one printed dataset, and seventeen of twenty modes — the other three being ties a spreadsheet resolved silently. One review flow reconciles at every counted stage. Three percentage claims in a case study verify exactly against chart data recovered from inside the file, while its headline safety figure matches none of six candidate computations.
None of that could have been established from the printed pages. Displaying your instrument makes your design auditable; displaying your data makes your analysis auditable. The works that show the most are the works that can be criticised in most detail — an argument for showing more, not less.
Five things worth copying
Taken from work that was good, and specifically why
Audit a number before you repeat it
One work takes a widely quoted failure rate to the four sources most often given for it and reads each in the original. It finds an observation that became a statistic, an explicitly unscientific estimate of a range whose upper bound became the headline, a flat assertion with no data, and a cited survey whose findings are never supplied. It then names what a substantiated claim would have carried — data, tables, graphs, case studies, reference lists — and six study parameters none of the sources reports. This is the strongest analytical work in the corpus and it cost four papers' reading.
Run a group the way this one was run
Recruitment through a standing governance body rather than a list of individuals, with consent obtained twice. All participant contact routed through an intermediary, because the researcher belonged to the community and coercion — actual or perceived — had to be removed. A first sitting abandoned at two attendees and reported as such. A size decision made explicitly at the floor of a cited four-to-fourteen range, and said three times. A facilitation tool with branch instructions and consent-to-record language. And the best sentence in the corpus: that agreement cannot be inferred because "silence does not equate to agreeance".
Print the raw dataset
One work prints its entire dataset — twenty questions by sixteen respondents, 320 cells, four marked not applicable — and prints every mean and mode. The consequence is that all twenty means can be confirmed, three reported modes can be diagnosed as a spreadsheet's tie-breaking rule rather than a finding, and the work's undeclared ranking convention becomes visible. One sentence stating that rule would have made three of its claims correct. Nobody could have known that without the appendix.
Make the review flow reconcile
A systematic review here names its databases, prints twelve search strings with their result counts, tables five inclusion and six exclusion criteria, justifies every exclusion criterion in prose, and describes two-stage screening — title and abstract, then full text — correctly and with attribution. Its flow adds up: 27 plus 321 screened, 27 plus 288 excluded, 33 retained. It is the only reconciling review flow in this material.
Make the schedule's arithmetic true
One project's plan uses a five-day working calendar consistently: its 676-day project duration computes exactly, 22 of 23 task date ranges match their stated durations, all eight summary tasks roll up correctly from their children, no task starts or finishes on a weekend, no child outlives its parent, four milestones are used and a three-level work breakdown is applied throughout. The library's other worked schedule fails on most of those. Sound arithmetic is not a small virtue: it is what makes the rest of the plan worth arguing with.
The hardest lesson: everything failed at the last link
One project in this material can be checked from end to end, because the instrument as issued, the raw response file, the ten-sheet workbook with its eight tabulations and eight charts, and the finished document were all supplied together. Six reported figures were traced. Five survive intact.
- Instrument — sound enough to field
- Raw file — complete
- Tabulations — all correct
- Charts — all correct
- Sentence — both errors here
That inverts the usual anxiety. Most checking effort goes into the analysis — the coding, the formulas, the tabulations. In the one case where the whole chain is visible, the analysis was faultless and the damage was done retyping a correct number into a sentence, late, without looking back at the tabulation.
The second defect compounds the first in an instructive way. The chart carrying the four correct counts has no data labels, so a reader cannot recover the right pairing from the figure. Had the chart been labelled, the prose error would have been visible and harmless. That is the strongest available argument for labelling values on your charts, and it is a worked one.
The last fortnight
Eight checks, all mechanical, none requiring the author
- Trace every number in the results, discussion and conclusions back to the tabulation that produced it — reading the tabulation, not your memory of it.
- Search each headline quantity through the whole document and confirm every occurrence gives the same value on the same base.
- Read the conclusions' verbs against the evidence: disprove, prove and demonstrate commit you to far more than show, indicate or found no basis for.
- Reconcile the reference list in both directions, and run text-to-list first — that is the direction that costs a reader a source.
- For every variable in a hypothesis or a conclusion, name the instrument item that measured it.
- For every statistical technique named, confirm a value, an n and — where the design supports it — a measure of uncertainty appears near it.
- Read every in-text figure and table reference against the caption it points at, in one pass.
- Regenerate the contents page and confirm every heading in it exists in the body with the same wording.
What to take from the whole corpus
Each of these eight works gives something the others do not: a worked focus group and the corpus's best methodological sentence; a hundred-response instrument and seventy-five verbatim accounts of unsafe work; a complete audit of a number everybody repeats; the cleanest objective-to-structure tracking of its group; the only reconciling review flow; the only figures auditable against their own file; the only complete raw dataset and three disconfirmations of the researcher's own first round; and one entire research apparatus that made a chain trace possible for the first time.
Hundreds of individual defects sit alongside those contributions, and all eight passed examination. Both facts are true and the second does not cancel the first. The judgement worth forming is exactly that specific — what each work gives, what it does not, and the fact that all of them were written to a form that asked them for none of it.
If you are starting, the research project lifecycle sets the sequence and eight examined theses compared shows what the finished thing looks like. If you are in the middle, the examined thesis quick reference is the comparator page. If you are writing up, the challenge of writing up, research integrity in examined work and overreach in conclusions are the three to have open.
What to carry forward
- The template is a container, not a design. Decide what kind of study you are doing, then say in writing where the form does not hold it.
- The method chapter is where the standing absences live. Name your sampling technique, your pilot, your recruitment channel, your denominator and your coding procedure — five sentences nobody taught these writers to write.
- Display the instrument and display the data. Being checkable is not the same as being wrong, and every finding worth having in this corpus exists because something survived.
- Copy the five: audit a number before repeating it; run a group with recruitment, controls and failures reported; print the raw dataset; make the review flow reconcile; make the schedule's arithmetic true.
- Copy the strengths with their limits attached — every one of the five sits in a document that also has defects, and knowing both is what makes the example usable.
- In the one project traceable end to end, the instrument, the data, the tabulations and the charts are all sound, and every error is a writing-up error. Budget for the last link. Label your charts. Check backwards from the evidence, never forwards from the draft.
Frequently asked questions
If I copy one thing from this corpus, what should it be?
The habit of auditing a number before repeating it. One work took a widely quoted failure rate to the four sources most often given for it, read each in the original, and named the transformation in each case. It is reusable on any number your field repeats without attribution, it costs a few papers' reading, and it produced the strongest analytical work in the corpus from a design that collected no data.
How much of my thesis should be the method chapter?
There is no defensible target here. Across seven examined and passed works the method runs from 2.6% to 14.0%, against guidance that says only that the majority should sit in the literature review and method sections. Length is the wrong question: write enough that another researcher could repeat the study, which means naming the sampling technique, the recruitment channel, the pilot and the analysis procedure.
Should I really print my raw data in an appendix?
If you can do so without breaching confidentiality, yes. The one work here that prints its complete dataset is the only one whose every reported figure can be confirmed, and the exercise showed that most of its claims were exactly right. Publishing the data is what turned an unverifiable set of claims into a verified one.
Where do most errors actually happen in a thesis?
In the one project here that can be traced end to end, all of them are in the write-up: the instrument, the raw file, the tabulations and the charts are sound, and both errors occur when a correct number is retyped into a sentence. That is one project rather than a rule, but it is the only complete evidence available and it points the checking effort at the last link.
What is the single cheapest check before submission?
Take every number in your discussion and conclusions and walk it back to the tabulation that produced it, working from the evidence towards the draft rather than the other way round. Reading forward from your own text reproduces the reasoning that made the error; reading back from the tabulation does not.
Does this corpus tell me what a good thesis looks like?
It tells you what eight passed theses looked like, which is different and more useful. None is a model answer; each contains something exemplary and something that failed. The value is in reading them as a set — the same form, eight designs, and a visible pattern in which decisions were made well and which were never asked for.
References and source attribution
- Eight examined master's works in project management, supplied as student work and used here as observed practice rather than as model answers. Researchers, supervisors, institutions, employers, jurisdictions, industries, communities and programmes scrubbed. Seven are written to one section template; all eight were examined and passed.
- The complete research apparatus of the eighth project — survey instrument as issued and as a document, participant information sheet, raw response file, ten-sheet analysis workbook with eight pivot tables and eight charts, and a project schedule whose arithmetic was independently recomputed — from which six reported figures were traced from question to sentence.
- A complete printed raw dataset from a further work, twenty questions by sixteen respondents, recomputed mean by mean and mode by mode for this library; and chart values recovered from the chart objects and embedded worksheet of another.
- The search strategy, review criteria, written justifications and stage-by-stage flow of the corpus's one systematic quantitative literature review, recounted and reconciled for this library.
- The supplied teaching source: weekly study notes, slide decks and assessment activities for a master's-level research methods subject in project management, which places limitations in the discussion, names focus groups and the pilot study without teaching either, and supplies no inferential procedure. Author, institution and year not stated in the supplied files.
Suggested questions for Ask KEVOS
- What should I copy from examined master's theses when writing my own?
- Where do errors actually occur in a research thesis — the analysis or the write-up?
- What belongs in a method chapter that students most often leave out?
- Should I publish my raw dataset in an appendix?
- What checks should I run in the last fortnight before submission?
- How long should my literature review and method chapters be?