Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a well-structured study addressing a genuinely underexplored question, gender differences in physics attitudes among Latin American medical students, and the CLASS instrument is applied and reported carefully overall. A few points would strengthen the manuscript.

Kruskal-Wallis tests were run across the eight CLASS categories plus the Overall score, comparing men and women, but no correction for multiple comparisons is reported. The Real World Connections result (p = 0.048) sits close to the conventional significance threshold, so a Bonferroni or similar adjustment, or at least an acknowledgment of the multiple-testing issue, would help readers judge how robust this finding is.

There is also an inconsistency in the choice of statistical tests. The authors justify Kruskal-Wallis for category-level comparisons on the grounds that CLASS data are ordinal and non-normal, yet the item-level analysis in Section 3.4 switches to independent-samples t-tests for the same type of Likert-scale data, without explanation.

The three-tier favourability classification (high ≥57%, medium 46-56%, low <46%) is described as based on the observed distribution of scores in this sample rather than an external or previously validated benchmark. Since this classification shapes much of the interpretation, it would help to clarify whether these cutoffs follow prior CLASS literature or were derived post hoc, and if the latter, note this as a limitation.

The comparison table (Table 3) draws on a study cited as "Anonymous, under review" for the engineering student benchmark. Since this comparator has not completed peer review, its results should be treated as provisional, and this could be flagged more explicitly when drawing conclusions from the comparison.

Only three of the 42 CLASS items were selected for the item-level analysis in Section 3.4, described only as "representative" without stated selection criteria. Some justification for this choice would reduce the appearance of selective reporting.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The topic is worthwhile and underexplored in Latin American medical education, and the sample size and data collection procedure are reasonable. However, the statistical reporting contains a serious internal contradiction, the methodology mixes incompatible statistical approaches without explanation, and several of the paper's central claims rest on weaker evidence than the discussion implies. I recommend the authors address these issues before this is ready for publication.

1) Table 2 lists the Kruskal-Wallis result for the Overall category as H = 0.871, p = 0.351. Later, in Section 3.3, the text reports the same comparison as H = 0.351, p = 0.871. These two numbers appear to have been swapped somewhere between the table and the prose. This is more than a typo. It means at least one of these values is simply wrong, and as a reader I have no way of knowing which one to trust.

2) 

The Methods section is explicit that Kruskal-Wallis was chosen because CLASS data are ordinal and not normally distributed. That is a defensible choice. But Section 3.4 then reports independent samples t-tests on individual survey items (t(196) = 3.89, p < 0.001, and so on), which is the exact type of parametric test the authors said was inappropriate for this kind of data just a few pages earlier. Either the ordinality argument holds, in which case the item level analysis should use a non-parametric equivalent, or it doesn't, in which case it's unclear why the category level comparisons weren't done with t-tests or ANOVA to begin with. This inconsistency needs to be resolved and explained.

3) 

Gender differences were tested separately across eight CLASS categories plus the Overall score, nine tests in total, with no adjustment for multiple comparisons anywhere in the paper. This matters a lot here because one of the two "significant" findings, Real World Connections, has a p-value of 0.048, which is barely under the conventional threshold and would very likely not survive even a simple Bonferroni correction. Since this finding supports a good portion of the discussion and the curriculum recommendations, the paper's conclusions are more fragile than they're presented as being.

4) 

The paper sorts categories into "high," "medium," and "low" favourability using cutoffs of 57%, 46 to 56%, and below 46%. These cutoffs were derived from the observed range of the actual results (34.4% to 60.2%), which the paper states directly. This is circular: the bins exist only because the data happen to fall where they do, so calling something "low favourability" doesn't mean much beyond "lower than the other categories in this one sample." If the authors want to keep this classification, they should either use cutoffs from prior CLASS literature or drop the labels and just discuss the raw percentages.

5) 

The two significant gender differences have rank biserial correlations of 0.23 and 0.14, both small by standard benchmarks. Despite this, the discussion builds a fairly elaborate explanation involving cultural coding of physics as masculine, specialty stereotyping, and sense of belonging. None of that is actually measured by the CLASS instrument. The interpretation reaches well past what a small effect size in two out of nine comparisons can support, and this needs to be scaled back or explicitly flagged as speculative.

6) 

The comparison with Colombian engineering students, one of the three stated research questions, rests entirely on a citation labeled "Anonymous, under review." This is cited repeatedly throughout the paper and used as the entire basis for a major comparison in Table 3. As a reviewer I cannot check this data, verify the numbers, or confirm the methodology used to generate them. Any conclusion depending on this comparison should be held until that paper is actually published and available for scrutiny.

7) Data were collected in 2018, and this submission appears to be in 2026. Section 4.5.7 spends a full paragraph arguing that this gap is actually a strength, framing it as a "pre-pandemic baseline." That may be true, but it reads like a justification added after the fact rather than a genuine methodological rationale. If there's a real reason this analysis wasn't published closer to when the data were collected, that should be stated plainly rather than reframed as a design feature.

8) The three individual survey items highlighted to illustrate the gender gap (items 3, 25, and 28) are presented as representative, but there's no explanation of how they were chosen out of 42 total items, or whether other items showed a different or opposite pattern. Without that context it's hard to know whether these are genuinely representative or the strongest examples that happened to support the narrative already being told.

9) Adams et al. (2006) is introduced earlier in the paper as the original CLASS validation study. In Table 3 it is relabeled as data from "Medical Students, USA." If that 2006 sample wasn't actually medical students, this comparison is mislabeled and the cross national comparison in Section 4.3 built on it needs to be corrected or removed.

10) Since the eight CLASS categories aren't independent of each other (they come from the same 42 items on the same students), a MANOVA or a permutation-based multivariate test treating all eight categories jointly would better respect the correlation structure and reduce the inflation from testing each category separately. If they want to stay non-parametric, a permutation MANOVA in R would work well with ordinal composite scores.

11) Table 2 reports means and standard errors but no CIs. Given the small-to-medium effect sizes and borderline p-values, 95% CIs around each favourability percentage (and around the gender difference itself) would give readers a much clearer sense of the precision of these estimates than a bare p-value does.

Author Response

Please see attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have done an excellent job revising the manuscript and have addressed all of the reviewers' comments satisfactorily. In my opinion, the article is ready for publication.

Author Response

We sincerely thank the reviewer for their time, careful reading, and positive assessment of our revised manuscript. We are glad the revisions addressed all previous concerns satisfactorily, and we appreciate the constructive feedback provided throughout the review process, which has meaningfully improved the quality and rigor of this work.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors addressed nearly every comment substantively. But there are a few discrepancies between what the response letter claims and what the revised manuscript actually shows:

1) In v1, the authors stated they translated CLASS into Spanish themselves and had it reviewed by four physics-education experts. In v2, after swapping the anonymous citation for Miranda & Rada (2018), the translation-and-validation story changed: now they say they used Miranda and Rada's existing Spanish version rather than producing their own. That's not just a citation swap, it's a different methodological claim about who did the translation, and the response letter never mentions or justifies this shift. If accurate, fine, but it should have been called out as a substantive revision rather than folded silently into the "anonymous citation" fix. Also worth noting: the same "Anonymous, under review" placeholder was resolved into two different references (Miranda & Rada, 2018, and Rada-Crespo et al., 2022, for different claims), which the letter also doesn't mention.

2) The letter's "Additional Changes" section says Kruskal & Wallis (1952) was removed and Mann & Whitney (1947) was added. In the actual v2 reference list, Kruskal & Wallis (1952) is still there, even though it's no longer cited anywhere in the text since everything switched to Mann-Whitney. And Mann & Whitney (1947) is cited in-text (Section 2.7) but doesn't actually appear in the reference list. So this is essentially the opposite of what was claimed: the entry that should have been removed wasn't, and the entry that should have been added wasn't. Worth flagging back to the authors. Separately, "R Core Team, 2023" is also cited in-text (2.7) but has no reference-list entry.

 

Author Response

We thank the reviewer for this careful observation, and we apologize for not addressing it explicitly in our previous response letter.

The discrepancy arose as an unintended consequence of the blind-review anonymization: the placeholder "Anonymous, under review" concealed two of our own related publications rather than one. Once these were unblinded after the first review round, we corrected Section 2.4 to attribute the translation and validation of the CLASS survey to its actual source, Miranda & Rada (2018), rather than describing it as newly conducted for this study. The translation and four-expert validation process itself has not changed — only its attribution has been corrected now that the source can be properly cited. We should have flagged this explicitly as a substantive correction in our previous letter rather than folding it into the citation fix, and we appreciate the reviewer's diligence in identifying it.

Regarding the reference list issues (Kruskal & Wallis, 1952 no longer cited but still listed; Mann & Whitney, 1947 and R Core Team, 2023 cited but missing from the list), these have now been corrected in the revised reference list.