we introduce a framework for studying interactive agentic coding based on a privacy-preserving analysis of ~400,000 Claude Code sessions from between October 2025 and April 2026.
这个数据点表示研究基于约40万个Claude Code会话，时间跨度为7个月(2025年10月至2026年4月)。这是一个相当大的样本量，增强了研究结果的统计可靠性。然而，文章未明确说明这些会话是如何被筛选或分类的，以及是否代表了所有Claude Code用户群体的完整情况。40万个会话对应约23.5万用户，平均每位用户约1.7个会话，这可能表明用户参与度相对有限。
we introduce a framework for studying interactive agentic coding based on a privacy-preserving analysis of ~400,000 Claude Code sessions from between October 2025 and April 2026.
这个数据点显示了研究的样本规模为约40万次Claude Code会话，时间跨度为7个月。这是一个相当大的数据集，增强了研究结果的可靠性。然而，我们不知道这40万次会话是否代表了所有用户，或者是否存在样本偏差。此外，研究仅限于Claude Code的使用，可能无法推广到其他AI编码工具。
We present results from a survey of 1,260 social scientists about AI and coding agent use, fielded in February and March 2026.
这个样本量(1,260)对于社会科学研究来说相当可观，提供了足够的数据基础进行分析。然而，文章也提到这不是代表性样本，因为受访者是受邀参与AI工作流程研究的，可能导致结果偏向于对AI工具更感兴趣的研究者。这一数据点表明研究结果可能存在选择偏差。
Age at Enspryng Initiation 47 (32, 55) 47 (30, 52)
Makes sense to me. Looks to be similar across cohorts
case-driven trial was planned 205to accrue 130case
What was the rationale for accrual of 130 cases has not been explained here. Would be nice to have that explanation. When the study parameters (95% confidence interval, 60% vaccine efficacy, 1% baseline attack rate) were applied to the WHO calculator obtained from here, the sample size turned out to be 86000, and when 80% confidence interval, 60% vaccine efficacy, and 1% baseline rate among unvaccinated) was used to assess sample size, it turned out to be 21000; in any case, the study then would be underpowered unless the rationale of 130 case accrual is made.<br> Why was not the WHO sample size estimator used?
Lakens, D. (2021). Sample Size Justification. PsyArXiv. https://doi.org/10.31234/osf.io/9d3yf
This is a page note. I can write overall comments about the pre-print here.
Tags can also be added below.
Power Analyses
Power Analyses with SPSS, R, Stata, SAS
Althouse, A. D. (2020). Post Hoc Power: Not Empowering, Just Misleading. Journal of Surgical Research, 0(0). https://doi.org/10.1016/j.jss.2019.10.049
Maarten van Smeden on Twitter: “This is a kind reminder that most issues with data (e.g. measurement error, incomplete data, confounding, selection) do not disappear just because you have N = ginormous” / Twitter. (n.d.). Twitter. Retrieved July 19, 2020, from https://twitter.com/MaartenvSmeden/status/1283313496382373890