I continue to be interested in trying to wrap my head around how the way in which you extract coefficients and standard errors from other people’s tables can give false positives for p-hacking in aggregate. And that that false positive is weirdly enough correlated with the scale of the outcome and therefore the research design (some being more commonly employed for some outcomes that have been scaled than others wildly enough). I have a video walk through of a shiny app I made in this post. I dont think I’m really quite finished scratching this itch though. Because this is part of my Claude Code series, I include the conversation that we had for anyone who wants to see how I use it as a “thinking partner”.