Most manuscripts are not rejected for being wrong. They are rejected for claiming more than the data supports, for a method that cannot answer the question asked, or for being a poor fit with the journal. All three are visible before submission.
Whether the claims follow from the data
The first thing a referee does is compare the abstract against the results. A conclusion that is stronger than the analysis supports is the single most common reason for a request for major revision, and it is the easiest thing to fix before anyone else sees it.
Watch for causal language attached to a correlational design, for effects described as significant without an effect size, and for a discussion that quietly widens the population the finding applies to.
Whether the method can answer the question
A method review asks a narrow question: given this design, could the study have detected what it claims to have found? Underpowered samples, missing controls, and analyses chosen after seeing the data all surface here.
Reporting matters as much as design. A reader who cannot reconstruct what was done cannot assess it, and a referee who cannot assess it will ask for revisions rather than guess.
Whether the paper fits the journal
A desk rejection often has nothing to do with quality. The paper is simply outside the journal’s scope, or aimed at the wrong readership, or formatted for a different venue.
Naming your target journal when you submit for review lets the reviewer flag a mismatch early, which is a cheap save compared with a rejection weeks later.
Whether the reporting is complete
Missing ethics statements, absent conflict-of-interest declarations, unavailable data statements and incomplete reporting checklists are all avoidable, and all of them delay a submission.
Common questions
What is the most common fixable problem?
Overstated conclusions. The analysis is sound but the abstract and discussion claim more than it supports, which is a writing fix rather than a research one.
Do reviewers check the statistics themselves?
On a methods and statistics review, yes: the design, the analysis and whether the claims follow from the numbers. A standard review covers the claims and reporting rather than recomputing the analysis.