How to compare conflicting findings in a literature review

Two papers can appear to disagree even when both are accurate within their own settings. One may study a different population, system boundary, baseline, outcome definition, or time horizon. A literature review should expose those conditions before it decides whether the findings are truly in conflict.

The Cochrane Handbook separates clinical diversity from methodological diversity and recommends examining study characteristics before synthesis. PRISMA 2020 also asks reviewers to report how they explored heterogeneity and to identify analyses that were not planned in advance. These principles are useful beyond health research because they make the reasoning behind a synthesis visible. See the official guidance in the Cochrane Handbook chapter on preparing for synthesis and the PRISMA 2020 expanded checklist.

The useful question is not which paper wins. It is which conditions explain the difference.

First decide whether the findings conflict

Start by rewriting each finding as a compact statement with four parts: the subject, the intervention or method, the measured outcome, and the conditions. Do not compare conclusions copied from abstracts. Record the result from the results section, together with the sample, measurement procedure, uncertainty, and source location.

Findings conflict only when comparable questions produce materially different answers. If one paper measures short-term efficiency and another measures lifetime reliability, the results may be complementary. If they use the same label for different outcomes, the conflict is semantic. If their confidence intervals overlap or both estimates are imprecise, the apparent disagreement may be stronger in the prose than in the data.

Align the review question before comparing answers

Build one row per study and use the same fields for every row. For intervention research, population, intervention, comparator, and outcome provide a useful starting frame. For engineering studies, replace population with system and add operating conditions, boundary conditions, input data, model assumptions, and evaluation horizon.

Normalize results without erasing context

Convert units and direction so the results can be read together, but keep the original values beside the converted ones. Record whether a larger value means improvement or harm. Separate absolute change from relative change. Align time windows and baselines where that is methodologically valid. Never convert a measure simply because the resulting table looks cleaner.

A narrative review still benefits from this discipline. You may not calculate a pooled estimate, but you can distinguish a difference in magnitude from a difference in direction. You can also flag results that cannot be compared and explain why.

Trace every disagreement to the source passage

Open the full text for each result and capture the smallest passage that supports your extracted finding. Then inspect the table or figure, method definition, and limitation that govern its meaning. This step catches common errors such as treating a subgroup result as the primary result, confusing simulation with field validation, or citing a discussion claim that is more confident than the data.

Preserve the source location in the comparison table. A coauthor should be able to reproduce your extraction without repeating the entire search. If the source is unavailable or the result cannot be located, mark it unresolved rather than filling the gap from memory.

Test explanations in a fixed order

Examine possible explanations from the most concrete to the most interpretive. Start with outcome definitions and units, then compare samples or systems, interventions and comparators, data collection, analysis choices, risk of bias, and random uncertainty. Record whether each explanation is supported by the study reports or remains a hypothesis.

Avoid inventing a post hoc story that makes every result fit. Sensitivity analyses and subgroup comparisons can test whether a conclusion depends on one study or one analytic decision, but unplanned analyses should be labeled as exploratory. The Cochrane Handbook recommends sensitivity analysis for influential decisions and careful interpretation when studies vary. See its official chapter on heterogeneity and meta-analysis.

Write a conditional synthesis

Organize the paragraph by claim, not by paper. Begin with the shared result, then state the condition under which findings diverge. Name the evidence for the most plausible explanation and end with the uncertainty that remains. This structure lets the reader see both consensus and limits without turning the review into a list of study summaries.

Example: compare two battery cooling studies

Consider an illustrative review of battery pack cooling. Study A reports that a liquid cooling design reduced peak cell temperature more than forced air. Study B reports no meaningful advantage. The conclusions look contradictory until the reviewer aligns the test conditions.

Study A used a high discharge rate, a compact pack, and peak temperature as the outcome. Study B used a lower discharge rate, greater cell spacing, and average temperature over a longer cycle. The studies do not provide two answers to one question. Together they suggest that the advantage of liquid cooling may depend on heat load, geometry, and the chosen thermal outcome. That conditional statement is more informative than reporting that the literature is mixed.

Use this final conflict check

1. Write each finding with its subject, method, outcome, and conditions.
2. Confirm that the studies address a comparable question.
3. Normalize units, direction, baseline, and time horizon where valid.
4. Save the exact source location for every extracted result.
5. Compare design, setting, measurement, uncertainty, and risk of bias.
6. Label supported explanations separately from reviewer hypotheses.
7. Write the synthesis by claim and condition, with residual uncertainty.

Rivul can keep study notes, source passages, claims, and citations close to the draft while you compare evidence. The researcher still decides which studies are comparable, whether a methodological explanation is credible, and how strongly the combined evidence supports a conclusion.

Start with the literature review evidence matrix to record the comparison, then use the claim audit to verify the final wording against its sources.