Selection is necessary; outcome-led selection is not
No dissertation can cite everything ever written on a broad subject. A defensible selection focuses on the question, applies relevance and quality criteria, and states its limitations. It becomes selective in a problematic sense when the preferred result quietly controls inclusion: supportive studies receive detailed treatment, contrary work vanishes, weaknesses are stressed only on one side, or a review is cited without the caveats that qualify the claim.
Citation bias is related to but distinct from publication bias. Publication bias shapes which studies become publicly available; citation bias shapes which available studies receive subsequent attention. Both can distort the apparent balance of a field. A student cannot repair the entire evidence ecosystem, but can make search and selection visible and restrict conclusions to the evidence actually found.
Look for suspiciously tidy paragraphs
Audit sections where every reference appears to agree without exception. Phrases such as “research clearly proves” demand particular scrutiny when followed by a handful of individual studies. Other warning signs include search terms that encode only the expected direction, relying on abstracts, repeating a secondary citation without checking the original, and omitting non-supportive results reported in the same paper. Citation count is not a substitute for direct relevance or methodological strength.
Fair coverage does not mean giving identical weight to all findings. A large, suitable and carefully conducted body of work may deserve more weight than one weak counterexample. Apply the same criteria to both directions: design, sampling, measurement, uncertainty, context and transferability. Explain why evidence receives less weight rather than excluding it because its result is inconvenient.
Design a search capable of finding disagreement
Build concept blocks around the topic, population or setting and relevant methods. Alongside expected associations, consider terms for null findings, replication, criticism, limitation and alternative explanations where the discipline uses them. Use more than one route: specialist databases, references of central publications, forward citation searches and appropriate reviews. Record the date, database, complete query, filters and number of results.
Set inclusion and exclusion criteria before reading the detailed findings. Screen titles and abstracts against those criteria regardless of direction. Log reasons for excluding important full texts. A narrative dissertation review need not pretend to be a formal systematic review, yet it can still reveal how the main perspectives were located and where coverage remains incomplete.
Use reviews, primary studies and citation chains carefully
A rigorous review can summarise a field but may not replace the original source for a specific method, sample or numerical claim. Check whether the review makes the same interpretation you intend to cite and whether its search period excludes later evidence. Follow influential or surprising claims back to their primary studies. If the original could not be accessed, label the secondary route rather than writing as though it had been read.
Forward citation searching can reveal later replication, criticism or boundary conditions. Do not sort solely by citation count: newer challenges naturally have had less time to accumulate citations, while an old paper may be cited for historical rather than evidential reasons. Match every source to the precise sentence it supports.
Synthesis should represent direction and evidential weight
Open a paragraph with the proposition being assessed, not a procession of author names. Group findings by direction, method or context, and explain why they converge or diverge. Mention uncertainty, differences in operationalisation and limits on generalisation. Counting three studies for and one against is not a synthesis unless their design and precision are also considered.
Place conflicting evidence next to the claim it qualifies, rather than hiding it in a final limitations section. A calibrated conclusion might say that an association is reported under condition A but is smaller or uncertain under B, perhaps because the measures differ. This remains informative while avoiding a certainty that the evidence matrix cannot support. Null and ambiguous results help define the boundary of the claim.
Run a counter-reading audit
- Highlight every broad generalisation and reopen the sources attached to it.
- Confirm that at least one counter-evidence or replication search is recorded.
- Apply the same quality criteria to supportive and conflicting studies.
- Compare the evidence matrix, exclusion record and final bibliography.
- Narrow claims where uncertainty, population or context requires it.