Xiao-Li Meng

The soup analogy is excellent, but it requires three assumptions: (1) the soup is well mixed in the pot, (2) nothing is altered to the spoonful, and (3) the taster knows how to judge and report. (2) & (3) are critical with human review. scientificamerican.com/article/the-se… via @sciam

New podcast! For this month’s episode, we talk to @colbyhall & @LelandVittert for an-in-depth look at how the media uses #data to report & analyze U.S. presidential races past & present. Listen! hdsr.mitpress.mit.edu/podcast #2024PresidentialRace #Election2024 #ElectionPolls #voting

Or above their heads — teaching design without having carried out one is like teaching recipes without ever cooking a dish. I learned a lot by teaching sample surveys, but I failed Miserably when I got involved in an alumni survey — I couldn’t even get a pilot study going …

Unfortunately you are not alone. Providing timely data and examples requires instructors to stay current with practice and to invest time and energy ongoing-ly in pedagogy. Neither of which is incentivized suitably in many univ. Perhaps demand and funding from industry can help?

Yes your points are well taken. Well designed and implemented data collections involve a lot of “behavioral statistics”, which is a lot messier to formulate, investigate, and write articles about. Same for preprocessing, another critical area we statisticians stay away from :-(

Fully agree and they were not exciting or challenging because those of us who taught them were not excited or challenged ourselves. We taught them out of textbooks, not out of ours experience or headaches. But things are changing, e.g., I know you are excited & challenged! :-)

The best line for fundraising from @TimRitchieMOS “People give to what they value when they are asked by somebody they trust”. And of course a bit of luck also helps. So good luck! :-)

From the keynote speech by Nancy Potok at AI Day for Federal Statistics: CNSTAT Public Event, NASEM. The webinar on May 21-22 is free to all. Registration infor is in the post. Check out the special issue on Democratizing Data at HDSR site. Thanks!

Thank YOU @francescadomin8 and @IavorBojinov for all your effort and innovation! I gather by “muse”, you meant “amused”, i.e., by how I had managed to mistype “causal conversations” as “casual conversations”, causing some last-minute frantic changes before launching :-)

My first presentation inside a barrel, and hopefully it’s not the last one. How did I do? Well, it would depend on what’s in your glass or bottle …

I was often told that the common wisdom is that it’s unwise to launch an issue on Fridays. But I hope collectively we can prove that “wisdom of crowd” is not a part of HSI, and this issue on HI, AI, and HSI will provide a provocative and relaxing reading for the weekend. Thanks!

GPT-n is certainly a disruptive technology in many senses. It is therefore fitting that this special issue also kicks off the open call of submission for HDSR, which has been by invitation only (upon successful proposal screening). Please help to get the words out, and submit!

Great to be back in the loop, before the n in GPT-n becomes too large to catch up. :-) Writing a thematized editorial for HDSR is my quarterly intense intellectual exercise. I’m grateful to authors, reviewers, and editors for providing all the heavy equipments. More, pleases!

While we are at it, check out the podcast on the DP issue, and the special issue editors @RuobinGong, @EricaGroshen, and Salil Vadhan's great editorial hdsr.mitpress.mit.edu/pub/fgyf5cne Thanks again Robin, Erica, and Salil! Thanks much @Rmcleodb for putting everything together!

Thank you for not making this seminar private! :-) Data privacy is truly a topic for everyone. Abstract for those interested (see you 8:30am EST Dec 15!) dropbox.com/s/usdmpl5s81ja… Please also check out the special issue on DP for 2020 US census in HDSR: hdsr.mitpress.mit.edu/specialissue2

Fully agree since no new values are created by sampling. But subsampling can enhance DP. A lot more to be investigated; there are interesting/tough questions to answer — under what conditions it is the right tradeoff for prioritizing outliers over most states in a distribution?

Thank you! My wonderful coauthor Keli Liu put together this figure. He was an undergraduate student when we wrote that article. He’s so original that when I recommended a Ph.D student I had to write that he was my “best Ph.D student” in order to avoid a comparison with Keli. :-)

Thanks much for the invitation! As for TBA, perhaps TBW (to be written) or TBR (to be researched) would be more accurate. :-). In any case, here is the missing title: “Privacy, Data Privacy, and Differential Privacy” — wish me good luck to have it ready by 12/15 8am! :-)

Many thanks also to Nick Lindsay @bluenoser2 at @mitpress for bringing this topic to HDSR and for making the initial connections! Properly sharing and managing data is critical for a healthy evolution of the data science ecosystem. Grateful to all authors,editors and reviewers.

Thanks! Need data minding and data confession here rss.onlinelibrary.wiley.com/doi/abs/10.111…, not much data mining or AI. Sampling rate 0.00005% could be ok. The million -- or billion -- dollar question is whether there was a strict quality control of the sampling & human review processes ...