Overreach in Conclusions
The distance between what a study establishes and what its last two pages say opens late, in a chapter written when the evidence is already in and nobody is checking any more. Four examined works show the four shapes it takes.
Where the gap opens
The conclusions chapter is written last, is short, and is read more than anything except the abstract. In these works it is where claims lose their qualifications. Nothing in the four is invented — every conclusion has a real finding behind it. What changes between the results and the closing chapter is the strength of the verb, the base of the number, or the presence of a caveat. That matters because a reader takes a finding from the abstract or the conclusion: in the first work below, the careful formulation appears four times in the body and the overreaching one twice, in exactly those two places.
FOUR EXAMINED WORKS, FOUR SHAPES OF OVERREACH (findings of these four documents, not a claim about examined work generally)
| Work | What the study establishes | What the closing chapter says | The distance |
|---|---|---|---|
| A literature-based study of change-management models | The four sources most often given for a widely quoted failure rate do not evidence it; a published fifth author found the same across five more | "The findings disprove the fail rate"; practitioners may apply the models "with a view to success" | Absence of evidence for a rate is not disproof of it, and discrediting a failure statistic creates no success evidence. |
| A two-project case study of lessons-learned practice | Severity-4 safety events fell from 4 to 2; total recorded safety items rose from 26 to 31 | "the significant reduction in PL4 safety events (60%)"; "superior performance" on five deliverables | No computation from the work's own data yields 60%. The nearest candidate is 50%, on the narrowest of six bases. |
| A survey study of change management and entrepreneurship | Eight questions, fifty responses, eight correct frequency tabulations, and a literature review | "…findings together led to rejection of the null hypothesis as there is no benefit offered by the change management processes…" | No test, no criterion, and the clause after "as" restates the null hypothesis it claims to reject. |
| A survey-and-interview study of risk management in event projects | Among 46 respondents, risk management is endorsed, unevenly executed, rarely reviewed and almost entirely self-taught | A named standardised framework "will promote trust, credibility, professionalism, encourage industry growth and increase … maturity" | No respondent was asked about the framework. The recommendation is the stated aim, returned as a finding. |
One word, after the best evidence work in the corpus
The first work rewards close reading, because its overreach follows immediately after the strongest analytical passage in this student corpus. Its subject is a change-failure rate that is quoted constantly and attributed loosely, and it reads each of the four sources most often given for it in the original.
- The 1995 article most often credited with the figure states no rate at all. It reports watching more than a hundred companies attempt to remake themselves, and says most fell between success and failure "with a distinct tilt toward the lower end of the scale". An observation has become a statistic.
- A 1993 book gives "our unscientific estimate is that as many as 50 percent to 70 percent". A range's upper bound has become the headline, and an explicit estimate has become a measurement.
- A 2000 article states the rate flatly and supplies no data. A repetition has become a source.
- A 2009 consultancy article reports a survey of executives and a one-in-three success rate, and supplies no statistical findings for it. A survey is cited and its findings are not.
- A published prior audit is then used as corroboration: a 2011 author reviewed five publications professing the rate and found none carried "valid as well as dependable empirical substantiation".
The difference is not stylistic. Establishing that the sources given for a number do not evidence it says nothing about whether the number is true — the rate could be correct and badly sourced. The careful claim is defensible on the work's own evidence and is a real contribution from a design that collected no data. The strong claim is not, and it is the one a reader carries away.
A headline figure that no computation supports
The second work compares two near-identical maintenance projects on five deliverables. Its safety claim reads: "the most notable difference in safety performance was the significant reduction in PL4 safety events (60%) and identified PL4 hazards." The figures behind its charts survive inside the delivered file, so every candidate computation can be run.
EVERY CANDIDATE COMPUTATION FROM THE WORK'S OWN SAFETY DATA (recomputed from the file; the stated claim is 60%)
| Quantity | Project 1 | Project 2 | Change |
|---|---|---|---|
| Severity-4 events | 4 | 2 | 50.0% reduction — the nearest candidate |
| Severity-4 hazards | 7 | 6 | 14.3% reduction |
| Severity-4 events and hazards combined | 11 | 8 | 27.3% reduction |
| All events, all severities | 11 | 13 | 18.2% increase |
| All hazards, all severities | 15 | 18 | 20.0% increase |
| All recorded items, all severities | 26 | 31 | 19.2% increase |
Six candidate bases, none yielding 60%. Neither total is printed in the work. The largest single change is at severity level 1, from zero to fourteen, and severity 1 is never mentioned.
The closing chapter lists safety among five "critical key deliverables" on which lessons learned "helped produce superior performance", while the work's own data show more recorded items in the second project and its results section concedes the second project was not faster.
Rejecting a null hypothesis without a test
The third work is the one project here that arrives with its whole apparatus — instrument, consent sheet, raw file, workbook, schedule. That makes its hypothesis strand fully checkable, and it shows a failure with a single root.
The primary and secondary findings together led to rejection of the null hypothesis as there is no benefit offered by the change management processes in entrepreneurial success as per entrepreneurial perspective. This is because the findings of both strongly support a relationship between entrepreneurial success and the change management context.
The clause after "as" is the null hypothesis as printed in the work's own introduction. As written, the sentence rejects the null on the ground that the null is true; the next sentence supplies the intended reason. It is reproduced as printed and not repaired, because the repair is what a reader would do and no reader did.
Behind it sit two harder facts. No test was performed anywhere: the words correlation, regression, confidence, reliability, validity and p-value do not occur, and "test" appears three times, always as a claim that hypotheses "were tested", never as a procedure. And the dependent variable was never measured.
- Hypotheses about entrepreneurial success
- Framework leaves that variable in an unconnected box
- Instrument asks no question that measures it
- Analysis is eight frequency counts
- Conclusion rejects the null
When the aim was already the conclusion
The fourth work states its aim as: "The research aims to demonstrate that the implementation of standardised industry-wide management practices, would be beneficial for the events industry." An aim written as something to be demonstrated cannot be disconfirmed by the instrument that follows it, and here the instrument never asks about it.
- The aim states that the framework would be beneficial.
- The literature review argues the case, with four citations and no counter-evidence, before any data are described.
- The survey asks forty-six people how they currently practise risk management, and nothing about the framework.
- The conclusions recommend it, in the language of a finding, with five stated benefits.
Every step is visible and none is flagged. What makes this instructive rather than merely wrong is what the same forty-six responses do support — a modest, well-evidenced result the work very nearly states. Strip the framework recommendation and nothing of value is lost.
The caveat that does not travel
A qualification stated once, next to the finding, does no work. Three of these works show the point from different directions; the fourth shows what a properly carried caveat looks like.
WHERE QUALIFICATIONS SIT, AND WHETHER THEY SURVIVE (findings of four examined documents)
| The qualification | Where it is stated | Where it is not |
|---|---|---|
| "silence does not equate to agreeance" — stated of the study's own principal finding | Once, on the same page as the finding it undercuts | The discussion and conclusions, which restate the finding unqualified |
| "only the minimum suggested participants for a focus group, being four, so this number is not a true representation of the community" | Once, in the results | The conclusions, which state that "four themes emerged" — the themes were in fact carried in from the literature and confirmed |
| "the 35% of participants with less than 12 months … may not be able to provide accurate testimony" | The limitations, in the closing chapter | The discussion, which had already used those results |
| "lessons learned frameworks can be assumed with a high degree of confidence to not be the sole contributing factor to performance increases or decreases" | Three times, in three chapters, each naming a specific alternative cause — the right caveat, properly carried | Two paragraphs above it, in the same chapter, the causal claim is asserted across five deliverables anyway |
The first row is the best methodological observation in this student corpus: it names the central inferential hazard of group data and does not resolve it. It is then not carried forward.
There is a structural driver. Across the seven works written to one section template, every limitations section sits in the conclusions chapter and none in the discussion, so each limitation arrives after the claims it should have bounded. See where limitations are placed.
The repair, in five passes
Auditing your own closing chapter
Split every conclusion sentence into claim and warrant
Rewrite each as "X, because Y", where Y names specific evidence — a table, a figure, a count, an interview. If there is no Y, the sentence belongs in further research.
Check the verb against the warrant
Disprove, prove, demonstrate and establish commit you to far more than show, indicate, is consistent with, or found no basis for. In the first example above the whole overreach is one verb.
Trace every number back to a tabulation
A figure appearing for the first time in the conclusions has no source in the document.
Match each hypothesis to what would have rejected it
Name the variable, the item that measured it, the comparison and the criterion. If any is missing, you have a proposition rather than a testable hypothesis, and should report what the descriptive evidence supports.
Ask who was asked
For each recommendation, point at the item that put the question to a participant. If nobody was asked, the recommendation is your inference — legitimate, and to be written as inference rather than as finding.
What these four cases do and do not show
All four works were examined and passed. Every defect here is a defect in a document — a verb, a figure, a sentence, a chain of steps — not a judgement about a researcher or a mark. Four documents are also not a population. Two of the four are checkable only because raw material survived: chart data left inside a file, and a complete apparatus. Works that publish less show fewer defects without having fewer, an asymmetry set out in research integrity in examined work.
Name what was good in each. The first performs a source audit no week of teaching material asks for. The second repeats its confounding caveat in three chapters. The third supplies the apparatus that made its own hypothesis strand auditable. The fourth tracks objectives into discussion sub-headings so cleanly that any objective can be traced to the paragraph answering it. See eight examined theses compared, drawing conclusions from your results and the challenge of writing up.
What to carry forward
- Overreach is usually one word, one figure or one sentence — cheap to find and cheap to fix.
- Absence of evidence for a claim is not disproof of it, and discrediting a failure statistic supplies no success evidence.
- Every number in the conclusions must exist earlier in the document, on a stated base.
- A hypothesis needs a measured variable, a comparison and a criterion. Without them, report the descriptive evidence and say no test was performed.
- Write aims so the answer could be no. An aim stating what will be demonstrated has already spent the conclusion.
- A caveat stated once beside the finding does no work. Carry it forward, or drop the claim it qualifies.
Frequently asked questions
How do I tell the difference between a conclusion and an overreach?
Ask what would have to be true for the sentence to be wrong, and whether your study could have found that out. A conclusion is something your design could have failed to support. If no result you could have obtained would have changed the sentence, it is a position rather than a finding, and it should be written as one.
Is it wrong to recommend something my data did not test?
No, provided it is written as your inference and not as a result. Recommendations routinely go beyond findings — that is part of their job. The failure mode on this page is a recommendation presented in the grammar of a finding, when no participant was asked about it and no comparison was made.
What should I write if I stated hypotheses and never tested them?
Say so, and report what the descriptive evidence supports. "The descriptive results are consistent with H1; no inferential test was performed, so the hypothesis is not decided by this study" is honest, short and costs nothing. Claiming a rejection you cannot show is the only version that damages the work.
My conclusions repeat my discussion. Is that overreach?
Not by itself, but it is where overreach hides. Repetition tends to strip qualifiers: the second telling drops the base, the caveat or the hedge. Compare the two passages sentence by sentence and check that anything true in the first is still true in the second.
Where should limitations go so that they actually constrain my claims?
Before the claims, not after them. Every limitations section across these works sits in the closing chapter, by which point the discussion has already made its assertions. Naming a limitation in the discussion, next to the claim it bounds, is what makes it do work.
Does an overreaching conclusion mean the whole study is unsound?
No, and treating it that way misreads all four cases. In each, the overreach sits beside genuinely good work — a source audit, a repeated confounding caveat, a full apparatus, clean objective tracking. Separating a claim defect from a design defect is what lets you keep the parts that hold.
References and source attribution
- Four examined master's works in project management, supplied as student work: a change-management literature study, a two-project case study of lessons-learned practice, a survey study of change management and entrepreneurship supplied with its complete apparatus, and a survey-and-interview study of risk management in event projects. Researchers, supervisors, institutions, employers, jurisdictions, industries and cities scrubbed. Used as observed practice, not as model answers.
- The recovered chart data and embedded worksheet of the two-project case study, from which the six candidate safety computations on this page were recomputed; the values come from the chart objects' stored series, not from the appearance of the printed figures.
- The complete research apparatus of the entrepreneurship study — survey instrument as issued, participant information sheet, raw response file, ten-sheet analysis workbook and project schedule — against which its objective and hypothesis audit was performed.
- The consolidated extracts of seven examined works written to one section template, recording every claim ladder, every limitations section and every quantitative claim checked against the printed evidence.
- The supplied teaching source: weekly study notes, slide decks and assessment activities for a master's-level research methods subject in project management, whose report structure places limitations in the discussion. Author, institution and year not stated in the supplied files.
Suggested questions for Ask KEVOS
- How do I check whether my conclusions are supported by my findings?
- What should I write in my conclusions if I never tested my hypotheses?
- How do I phrase a recommendation that goes beyond what my data show?
- Where should limitations sit so they actually constrain my claims?
- What is the difference between disproving a claim and showing it is unevidenced?
- How do I audit a closing chapter sentence by sentence before submission?