The analysis is finished and the tables are built, and then the manuscript stalls for days in front of the methods section.
The reason is usually the same. The decisions made along the way are scattered across your memory and a dozen files, so it is unclear what to report and in how much detail. Why four factors rather than three, why a particular item was dropped, what justified proceeding when fit fell short — all obvious at the time, all hazy some months later.
This article covers the order in which to report a scale development process, how to present statistical results in APA format, and how to disclose AI tool use.
A Methods Section Reports Decisions, Not Results
Writing the methods section as a summary of results fills the page and draws questions at review. What a reviewer is checking is not the numbers but whether the choices that produced them were defensible.
Take a threshold. "CFI was 0.93, meeting the criterion" is a results statement. The methods section needs the reasoning stated first: "CFI ≥ 0.90 was adopted as the acceptance criterion (Hu & Bentler, 1999)." The order should read as criteria set, then results obtained — never as results seen, then criteria fitted around them.
Reproducibility is the standard to write against. Ask whether another researcher following the same procedure would arrive at the same decisions from what you have written.
The Order to Report Scale Development
Following the development sequence itself reads most clearly.
Begin with how the construct was defined and the theory that definition rests on, along with the number of factors specified and the rationale for it.
Item generation comes next: how many items were written and how many survived, and how content validity was established. If an expert panel was used, report its size, the indices computed, and the criteria applied. CVR ≥ 0.99 with a five-member panel (Lawshe, 1975) and I-CVI ≥ 0.78 with six or more experts (Lynn, 1986) are the commonly applied criteria.
Describe the pilot and main samples separately: size, collection period, method, response rate, and exclusion criteria. How missing data were handled belongs here as well.
Analysis procedures close the section. Report how the sample was split between EFA and CFA, which estimator was used, and which fit criteria were applied with their sources. If items were deleted, report the deletion criterion and whether the model was re-estimated afterward.
Reporting CFA Results in APA Format
APA italicizes statistical symbols: sample size N, chi-square χ², degrees of freedom df, and the probability p. Established index abbreviations such as CFI, RMSEA, and SRMR are not italicized.
In-text reporting presents chi-square with its degrees of freedom, followed by the fit indices.
| Element | Format |
|---|---|
| Chi-square | χ²(df) = value, p = value |
| CFI, TLI | Two decimal places |
| RMSEA | Value with 90% confidence interval |
| SRMR | Three decimal places |
| Standardized loadings | Range per factor, or a full table |
RMSEA should be reported with its confidence interval; a point estimate alone hides the uncertainty around it. Drop the leading zero for any value that cannot exceed 1, which covers correlations, reliability coefficients, and p values.
Disclosing AI Tool Use
Disclose it. Major scholarly bodies and publishers hold that generative AI tools cannot be listed as authors and require that their use be reported along with its scope, because authorship is reserved for parties who can take responsibility for the content.
Where the disclosure goes varies by journal. Some ask for it within the methods section as part of the procedure; others maintain a separate declaration field. Checking the submission guidelines first is the reliable route.
Specificity helps. State what the tool was used for and how the researcher reviewed its output. If it drafted candidate items, say that the generated items were then reviewed by an expert panel; that completes the procedural account. The test is whether a reader can tell what the tool did and what the human judged.
What modidoc Does at the Methods Draft Stage
modidoc's methods draft stage assembles the decisions and statistical results accumulated across development into a draft methods section. Applied thresholds and their sources, the item deletion record, and the reasoning behind below-threshold judgments are carried into the prose, and statistical results are laid out in APA-format tables. The draft assumes researcher review and revision, and statistics not produced from real data are never cited in it. This stage is implemented internally as the AM-1 methods section engine.
Frequently Asked Questions
How do I report a scale development process in the methods section?
Follow the development sequence: construct definition and theoretical grounding, item generation and content validity, sample and collection procedure, then analysis methods and applied criteria. Reporting the judgment made at each step alongside its rationale is the core requirement, and the detail should be sufficient for another researcher to reach the same decisions from the same procedure.
How do I report CFA results in APA format?
Present χ²(df) = value, p = value first, followed by CFI, TLI, RMSEA, and SRMR. Report RMSEA with its 90% confidence interval, and italicize the statistical symbols (N, χ², df, p). Omit the leading zero for values that cannot exceed 1. Standardized loadings and reliability coefficients read better in a separate table.
How do I disclose the use of AI tools in a manuscript?
State the use and its scope. Generative AI cannot be an author, so the disclosure goes in the methods section or the journal's designated declaration field rather than the author list. Describing what the tool was used for and how you verified its output makes the chain of responsibility explicit. Placement and wording vary by journal, so check the submission guidelines first.
That completes the scale development pipeline series, from defining a construct through writing the methods section, with the judgment each stage calls for.
Previous: Final Checks Before Fielding a Questionnaire
First in the series: Construct Extraction and Factor Structuring