03 · What You Need to Know
The Problem Is Artificial Fragmentation, Not Productivity
One project can legitimately produce many papers
Publishing several papers from one research program is not inherently questionable. ICMJE explicitly recognizes that multiple manuscripts may be based on the same dataset and that different analytical approaches can be complementary and equally valid.
Its central expectation is that papers based on the same dataset should add substantially to one another to warrant separate publication. Previous publications from the dataset should also be appropriately cited so readers can understand their relationship.
ICMJE further notes that large trials may be planned from the beginning to generate numerous publications addressing separate research questions while using the same original participant sample.
Accordingly, salami slicing cannot be diagnosed merely by counting publications.
The defining concern is insufficient substantive separation
The term is generally used when researchers partition one coherent study into smaller papers whose distinctions are too thin to justify treating each as a substantial independent contribution.
COPE has discussed salami publication within the broader context of redundant publication. Its cases illustrate situations in which the same study generates papers with substantial overlap or separate papers reporting closely related endpoints that reiterate essentially the same findings.
The question is therefore not whether the papers can technically be distinguished. Almost any manuscript can be made to look different by changing its title, emphasizing another variable, or rearranging the analysis. The more important question is whether publishing the pieces separately improves scientific communication.
Legitimate multiple papers
Each publication addresses a meaningfully distinct research question or analysis and contributes substantial additional knowledge.
Salami slicing
A coherent body of closely related evidence is divided into minimally differentiated papers whose separate publication adds little scientific value.
Different outcome variables do not automatically justify different papers
Imagine a randomized study measuring anxiety, stress, depression, well-being, and quality of life. It would be possible to create a separate manuscript for every outcome.
Whether that would be sensible depends on the scientific questions. If the outcomes collectively address the same hypothesis about psychological well-being and were analyzed using closely related methods, separating them may obscure rather than clarify the overall evidence.
Conversely, one outcome may represent a genuinely different theoretical domain, require a distinct analytical strategy, and support conclusions that deserve independent treatment. The variable count does not decide the issue.
Subgroup analyses are not automatically separate studies
A dataset can often be partitioned by sex, age, location, disease severity, academic level, institution type, or almost any other characteristic. Creating a separate paper for each subgroup can rapidly produce an impressive publication list while contributing surprisingly little additional knowledge.
A separate subgroup paper is more defensible when the subgroup question has a substantive rationale, adequate statistical support, and an interpretation sufficiently distinct from the primary paper. It is less defensible when the subgroup merely repeats the main analysis in a smaller slice of the same sample.
Different statistical methods do not automatically create different contributions
A researcher might analyze the same relationship using regression in one paper, structural equation modeling in another, and machine learning in a third. These methods may genuinely reveal complementary aspects of the data.
But methodological difference alone is not enough. If all three papers answer essentially the same question and lead to the same substantive conclusion, readers may gain more from an integrated comparison than from three separate publications.
ICMJE's guidance is helpful here: different analytical approaches can justify multiple papers when they are complementary, but manuscripts based on the same dataset should add substantially to one another.
Ask whether the papers would be stronger together
A useful test is to imagine combining the proposed manuscripts. Would the result become scientifically clearer and more complete without becoming unmanageably broad?
If Paper A contains the primary outcome, Paper B contains a closely related secondary outcome, and Paper C contains another measure of the same underlying construct, combining them may allow readers to understand the pattern of evidence much better.
Fragmentation can force readers to reconstruct the study across several articles, sometimes without realizing that the papers come from the same participants. It can also separate findings that need to be interpreted together.
Salami slicing can exaggerate the apparent size of the evidence base
Fragmented papers can create problems beyond publication counts. If their shared dataset or participant sample is not transparent, readers may mistakenly treat them as independent studies.
ICMJE warns that duplicate publication of original research can cause data to be counted more than once or give the results of one study inappropriate weight. Similar problems can arise when overlapping publications are insufficiently transparent.
A systematic reviewer who encounters three papers from the same sample needs to know that they are related. Otherwise, one group of participants can acquire the statistical superpower of becoming several groups merely by appearing under different article titles.
Transparency does not automatically justify unnecessary fragmentation
Citing all related papers is essential, but cross-citation does not transform every possible subdivision into a worthwhile publication.
You can transparently disclose that Papers A, B, and C use the same sample and still face the question of whether B and C add enough to deserve separate publication. Transparency addresses concealment; it does not replace substantive justification.
Watch Out
Do not use cross-citation as a license to divide a study indefinitely. A paper should have a scientific reason to exist separately, not merely a transparent record of having been sliced from the same project.
Salami slicing is different from duplicate publication
The concepts overlap but are not identical. Duplicate publication typically involves substantial republication of work that has already appeared, particularly when the prior publication is not clearly referenced.
Salami slicing may involve different subsets of results from the same underlying study rather than exact republication of the same results. The concern is that the division is unnecessarily fine and the papers provide insufficient independent value.
The broader distinction between duplicate publication and legitimate secondary publication therefore addresses a different boundary.
Salami slicing is also different from legitimate secondary analysis
Secondary analysis can extract important new knowledge from existing data. A new theoretically motivated question, different endpoint, long-term follow-up, methodological comparison, or genuinely distinct analysis may warrant a separate paper.
ICMJE specifically provides for secondary analyses of clinical trial data, recommending that they cite the primary publication, clearly state that they contain secondary analyses or results, and retain the relevant trial and dataset identifiers.
The central distinction is added knowledge. Reusing data is not the problem. Producing papers that contribute too little beyond one another is.
The publication plan can reveal fragmentation before writing begins
For a large project, map the proposed manuscripts before drafting them. List each paper's research question, primary outcome, sample, analytical approach, expected contribution, and relationship to other manuscripts.
If several proposed papers have almost identical entries except for one variable, the fragmentation becomes easier to see. If each paper has a clearly different intellectual purpose, the rationale for separate publication becomes easier to defend.
This is particularly useful when deciding how many papers one dataset can legitimately support.
Related manuscripts should be disclosed to editors
ICMJE recommends that authors disclose submissions and previous reports that might be regarded as redundant publication and provide related material to the editor. The new paper should specifically reference such work.
This allows editors to evaluate whether the manuscripts are sufficiently distinct rather than forcing them to discover the relationship during peer review.
