Gosselin_SCIREP_Data

A mixed sampling methodology was implemented (Figure 1) to collect journals and articles. First, a selection filter was applied within the Institute for Scientific Information (ISI) Journal Citation Report (https://jcr.clarivate.com) database to generate a list of 504 life science journals. Then, exclusion criteria were applied to the journal list and 245 periodicals were removed. Filters and exclusion criteria are given in Table 1. Using a pseudo-random sequence of 20 numbers between 1 and 259 generated using GraphPad QuickCalc (https://www.graphpad.com/quickcalcs/randMenu), a final shortlist of 20 journals among the 259 preselected ordered by decreasing 2018 Impact Factor were selected (the latest available impact factor at the time of designing this study). Four additional journals were finally excluded either because they were eventually found to be too clinical or because there was no online access granted to the author’s institution, leading to a final list of 16 periodicals (Table 3). Clinical journals were excluded although they may include publications with some preclinical experiments. This was justified to prevent the possible bias created by both the presumed small proportion of such articles in clinical periodicals which would have prompted a larger sampling and the supposed compliance of these studies with clinical guidelines whose standards may be different ^29,30. Fifteen articles per journal were collected by sampling the online contents of each journal, starting from the last issue released in 2019 and browsing backward. This time window was selected to avoid the abundant literature on Coronavirus disease 2019 (Covid-19) published since January 2020, which might show unusual statistical standards. Article inclusion and exclusion criteria are presented in Table 2. Studies using human data were acceptable when they used ex-vivo/in-vitro approaches for extracting tissues, cells or samples. From this intermediate list of 240 articles, 17 were finally excluded during the analysis due to previously unnoticed violations of inclusion criteria or for congruity with exclusion criteria, resulting in a final sample set that included 223 articles. <b> </b><b>Assessment of reporting</b>Each article was explored, and three types of statistical attributes were quantified (Table 4). Indicators of the transparency of study protocols were binary items coded as 0 (presence of all needed information in the text) or 1 (absence of information in the text for at least one figure or table) and were aggregated as proportions of articles that had an insufficiency (non-disclosure) for the given item. The indicators were chosen as the minimum set of information needed by a reader to replicate the statistical protocol: precise sample size (experimental units), well identified test, software and no contradiction. A contradictory information is defined as a mismatch between information provided in different parts of the manuscript although they refer to the same object, such as the disclosure of dissimilar statistical tests (in methods and figure legends) to describe the analysis in one figure or the disclosure of multiple sample sizes for one single set of data. The article structure was assessed using quantitative items, specified as total counts of given items as well as one binary outcome (presence of a statistical paragraph). Qualitative items represented the article content and have been summarised as an inventory of information of interest. In the sampled articles, supplemental methods and information were considered full-fledged methodological information, but supplementary figures and tables presenting results were not eligible for the quantification of statistical insufficiencies, even if they were used to report location tests.

本研究采用混合抽样方法（如图1所示）收集期刊与学术论文。首先在科学信息研究所（Institute for Scientific Information, ISI）期刊引证报告（Journal Citation Report, JCR，https://jcr.clarivate.com）数据库中设置筛选条件，生成包含504种生命科学期刊的列表。随后对该期刊列表应用排除标准，剔除245种期刊，具体筛选与排除标准详见表1。
借助GraphPad QuickCalc（https://www.graphpad.com/quickcalcs/randMenu）生成1至259之间的20个伪随机数序列，在按2018年影响因子降序排列的预筛选259种期刊中，最终选出20种期刊作为候选（该影响因子为本研究设计时可获取的最新数据）。后续又有4种期刊被排除：要么因最终发现其过于偏向临床领域，要么因作者所在机构无法获取其在线访问权限，最终确定16种期刊的最终列表（详见表3）。
尽管临床期刊可能包含部分临床前实验相关论文，但本研究仍将其排除，此举的合理性在于：一方面，临床期刊中此类文章的占比推测较低，可能导致抽样规模偏大；另一方面，此类研究多符合临床诊疗指南，其研究标准或存在差异，由此可能引入偏倚^29,30。
针对每本入选期刊，从2019年最新一期开始反向浏览其在线内容，每本期刊采集15篇论文。选择该时间窗口是为了规避2020年1月以来大量发表的新型冠状病毒肺炎（Coronavirus disease 2019, Covid-19）相关文献，此类文献可能存在非常规的统计标准。论文的纳入与排除标准详见表2。对于使用人类数据的研究，若采用离体（ex-vivo）/体外（in-vitro）方法获取组织、细胞或样本，则可纳入本研究。
从初步筛选得到的240篇论文中，最终因此前未发现的纳入标准违背情况或符合排除标准，剔除17篇，最终得到包含223篇论文的样本集。
**报告质量评估**
对每篇论文进行分析，量化三类统计属性（详见表4）。研究方案透明度指标采用二元编码：0代表文本中包含所有所需信息，1代表至少1幅图表存在信息缺失情况，并按比例汇总存在对应项目信息不完整（未披露）的论文占比。所选指标为读者复现统计分析方案所需的最低限度信息：明确的样本量（实验单位）、清晰标识的检验方法、所用软件，且无矛盾表述。矛盾信息指论文不同部分针对同一对象的表述不一致，例如：某幅图表的分析方法描述与方法学部分披露的统计检验方法不符，或针对同一组数据披露了多个不同的样本量。
论文结构采用量化条目进行评估，包括指定条目的总计数，以及1个二元结果指标（是否包含统计方法段落）。质性条目代表论文内容，已整理为目标信息清单。在采样的论文中，补充方法与补充信息被视为完整的方法论信息，但补充图表（即使用于报告位置检验）不计入统计不完整性的量化范围。