Fan Li

54 posts

Fan Li

Fan Li

@FanLiDuke

Professor@Duke, Statistician, Data Scientist, Causal Inference researcher

Durham, NC, USA Katılım Mart 2024
121 Takip Edilen1.1K Takipçiler
Sabitlenmiş Tweet
Fan Li
Fan Li@FanLiDuke·
Done another semester teaching causal inference🙂. Updated my course slides, added survival data, labs, corrected more typos this time. Close to 800 pages now. Always more to update next year. www2.stat.duke.edu/~fl35/CausalIn…
English
10
132
599
87.5K
Fan Li
Fan Li@FanLiDuke·
I was often asked by practitioners about power calculations for causal inference with observational data, a hard problem with little leads. Finally had a clean solution, thanks to my spectacular student Bo Liu. Here it is: arxiv.org/abs/2501.11181 cran.r-project.org/web/packages/P…
English
1
29
105
12.1K
Fan Li
Fan Li@FanLiDuke·
@sasilu6 What's wrong with lecture notes? This gives the most concise, update-to-date and complete survey of the field, with examples, HWs, codes. Any other book on market offers these?
English
1
0
2
348
prince_mishkin
prince_mishkin@sasilu6·
@FanLiDuke Market is competitive. This book is unfortunately more like lecture notes.
English
1
0
0
368
Fan Li
Fan Li@FanLiDuke·
Twitter academia: 1. I am happy to announce xx (whatever trivial) 2. I am thrilled/excited that xx (papers, grants, promotion) 3. I am honored that xx ("awards" in all senses) Adding to that list now: "How I publish xx papers in x years" What next? How I become god?
English
41
118
1.4K
188K
Fan Li
Fan Li@FanLiDuke·
@ssprickschuster Except that often "I am humbled/honored" is a simply thin-veiled bragging of trivial things or openly flattering relevant people (e.g. potential reviewers or letter writers)
English
1
0
9
2.5K
Steven Sprick Schuster
Steven Sprick Schuster@ssprickschuster·
@FanLiDuke The first three are just sharing good professional news. If a friend gets promoted, this is a great way to find out. What’s wrong with people doing this? The last one is totally different, since it often preys on people’s insecurities
English
1
0
13
2.8K
Fan Li
Fan Li@FanLiDuke·
@economeager "How I become god" is a even better click bait than "(non econ) how I write xx papers in x years” or "(econ) how I write one paper in less than 10 years."
English
2
0
16
5.1K
Rafe Meager (they/them)
Rafe Meager (they/them)@economeager·
@FanLiDuke This tweet goes so hard Fan that I now want to write a satirical substack piece called How I become god
English
2
0
10
5.4K
Fan Li
Fan Li@FanLiDuke·
@MarvinSchmittML and of course, wanting the whole world to know "I am humbled"
English
2
1
53
6.7K
Marvin Schmitt 👉🦋
Marvin Schmitt 👉🦋@MarvinSchmittML·
@FanLiDuke And of course the infamous "I‘m humbled by xx" (also awards) Which makes no sense at all because winning an award literally means you best everyone else how can that humble you lol
English
1
0
43
8K
Fan Li
Fan Li@FanLiDuke·
@chefjeffsf Except that Panda Express is not Chinese, it is safely American.
English
0
0
0
149
Jeff Tang
Jeff Tang@jefftangx·
Why isn't there a big Indian American restaurant chain? - Chinese: Panda Express (2398) - Korean: Bonchon (139) - Japanese: Benihana (70) - Chinese: P. F. Chang's (70) - Indian: Curry Up Now (22)
English
880
262
14.2K
2.2M
Frank Harrell
Frank Harrell@f2harrell·
@PWGTennant When I’m asked to review a paper that puts the methods in a supplement or appendix I just pretend not to be able to find them and say “Unfortunately the paper does not have a methods section” and reject.
English
5
6
121
12.5K
Peter Tennant has moved to Bluesky
Peter Tennant has moved to Bluesky@PWGTennant·
No. This is not acceptable. The methods section is the single most important part of a scientific paper. If those details are relegated to supplementary materials, then it's not a scientific paper in a scientific journal.
Peter Tennant has moved to Bluesky tweet media
English
114
469
4.5K
522K
Fan Li
Fan Li@FanLiDuke·
@BhramarBioStat Second this. Learning some history of statistics (started from survey sampling) also helps.
English
1
0
8
1.1K
Fan Li retweetledi
Peng Ding
Peng Ding@pengding00·
This is an interesting and useful trick. However, centering factors has some special restrictions on the estimated factorial effects when there are more than 3 factors (3 is the magic number there!). This motivates us to write this paper: academic.oup.com/biomet/article…
Matt Blackwell@matt_blackwell

A fun fact about regression that many know but maybe is new to you: If you have an interaction bw continuous X1 and binary X2, mean-centering X1 will make the coefficient on X2 be its marginal effect when X1 is at its mean level rather than 0 without changing the interaction

English
0
23
85
22.2K
Fan Li
Fan Li@FanLiDuke·
@gv_lazcano @5_utr I am not saying one should go to obs studies. I am saying, to analyze your holy grail RCTs properly, many causal inference techniques designed for observational studies are necessary. Again, blanket hostility is counter-productive.
English
2
0
1
158
Gabriel Lazcano
Gabriel Lazcano@gv_lazcano·
@FanLiDuke @5_utr And going for observational data isn't going to fix any of that. Also, another huge advantage of RCTs is the protocol by itself. For radiation oncology that could mean robust centralized QA, which also translates in better care for the trialed population.
English
2
0
3
98
NonsparseOncologist
NonsparseOncologist@5_utr·
Rarely have I seen an article miss the mark this badly 🤯 Misguided calls for more “real world” observational evidence should be replaced by calls for more efficient RCTs; patients and society deserve inference based on RCTs, nothing less
NonsparseOncologist tweet media
English
12
22
114
17K
Fan Li
Fan Li@FanLiDuke·
@lhan320 So sorry for your loss
English
0
0
1
166
Larry Han
Larry Han@lhan320·
Completely devastated. Grayson was my best friend growing up. He had a heart of pure gold. He was the most talented golfer, but he was humble, too. I hope Grayson's life can inspire many to come to know Jesus Christ. Looking forward to seeing you on the other side, my brother!
Todd Gibson@ToddGibsonWNCN

My first story with Grayson Murray was 18 years ago. I shot it for Chad Sokol and he voiced it. Just a young golfer having fun and winning. Murray & Han battled for a few years but were good friends. Sad to hear of Murray's passing. @WNCN @ChrisClark_ @PGATOUR @leesvilleroadhs

English
1
0
11
1.5K
Fan Li retweetledi
apoorva.lal
apoorva.lal@Apoorva__Lal·
what a vote of confidence [cc @pengding00 ]
apoorva.lal tweet media
English
0
2
45
16K
Fan Li
Fan Li@FanLiDuke·
@yudapearl @f2harrell @soboleffspaces @stephensenn @PWGTennant There is a subtle technical difference between the three versions of ATE, but I don't believe the technical purity of PATE and SATE makes them more superior. In practice, the numerical difference is little - that matters the most.
English
0
0
1
100
Boris Sobolev
Boris Sobolev@soboleffspaces·
"Some papers" is just a sleight of hand by @f2harrell. Badly done science doesn't invalidate the scientific method. If someone can’t perform Rach-4, it doesn’t mean that Rachmaninoff wrote bad music.
Judea Pearl@yudapearl

We are not talking "some papers", we are talking scientific imperatives. If you are trying to conclude effect E from data D, you got to have a logic leading from D to E. CI provides such logic. And it also proves that statistics alone does NOT, whenever effect E is Causal Effect. No escape.

English
1
1
10
8.9K
Fan Li
Fan Li@FanLiDuke·
@f2harrell @yudapearl @soboleffspaces @stephensenn @PWGTennant In fact, most CI practitioners estimate a third estimand, a mix of PATE and SATE, because of the conditioning on the X in the sample. My Bayesian causal review paper carefully discussed this. In practice, diff between P-, S-, M- ATE is small, not warrant much concern
English
2
0
3
148
Frank Harrell
Frank Harrell@f2harrell·
@yudapearl @soboleffspaces @stephensenn @PWGTennant 2/2 Many CI practitioners claim to be estimating PATE when there are really estimating SATE and there is no bridge for converting SATE to PATE due to lack of random sampling from the Population to get the Sample.
English
1
0
0
661
Fan Li
Fan Li@FanLiDuke·
@noah_greifer @stephensenn A caveat: the asymptotic equivalence holds for IPW, but for OW only with r=0.5 (r: trt proportion). We added a correction recently.
English
0
0
3
94

Keşfet

@sasilu6 @signalvoyager @Andrew_Akbashev @ssprickschuster @economeager @MarvinSchmittML @f2harrell @PWGTennant