Timeline for Exact matching + multiple regression on high-dimensional treatment-control study?
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 12, 2020 at 18:31 | comment | added | RobertF | @Noah I was looking up information on the American Causal Inference Conference for 2020 - it was going to be held in Austin but alas I see it's been cancelled: events.mccombs.utexas.edu/event/… likely for public health reasons. | |
| Jan 2, 2020 at 4:04 | history | edited | Noah | CC BY-SA 4.0 |
added 1210 characters in body
|
| Oct 23, 2019 at 3:48 | history | edited | Noah | CC BY-SA 4.0 |
added 452 characters in body
|
| Jul 29, 2019 at 12:48 | comment | added | Frank Harrell | We teach both traditional PSM, regression adjustment using PS, inverse probability weighting, and efficient match-while-randomizing methods. The latter increases efficiency of treatment comparisons. | |
| Jul 26, 2019 at 21:30 | comment | added | RobertF | @FrankHarrell Out of curiosity, are some of the techniques described in this answer being taught in advanced biostatistics classes at Vanderbilt? It seems that matching on propensity scores is no longer considered a valid approach. | |
| Jul 17, 2019 at 20:41 | vote | accept | RobertF | ||
| Jul 5, 2019 at 20:04 | comment | added | Frank Harrell | PSM with reduction of sample size due to the matching algorithm is likely to be statistically inefficient (higher standard error and lower power of treatment effect). | |
| Jul 2, 2019 at 16:43 | comment | added | Noah | I do know about FLAME, but it's primarily for massive data sets (i.e., that exist in databases) rather than typical samples. It's not ready for prime-time. The comparison of performance between the methods is tough and is a reason to prefer matching and weighting over regression-based, but you can also rely on the theoretical characteristics of the methods (e.g., that they tend to do well in arbitrary simulations). You could make your own simulation based on your data and see under what circumstances one method does better. | |
| Jul 2, 2019 at 16:40 | comment | added | Noah | There was a paper that attempted to refute K&N, but I can't find it anymore after some searching, but the success of PSM in many simulations is enough for me, I think. PSM is not completely out-of-date, but there have been many improvements on it that you should consider before going straight for it. Exact matching is still the gold standard and always will be, but it tends to be impossible in mot circumstances. CEM would be ideal too, but many data sets don't support it (see, e.g., Zubizarreta et al. (2014)). | |
| Jul 2, 2019 at 16:26 | comment | added | RobertF | BTW have you seen this paper on the "FLAME" (Fast Large-scale Almost Matching Exactly) model for causal inference? It's on arXiv so I can't tell you if it's been peer reviewed, but looks interesting: arxiv.org/abs/1707.06315 | |
| Jul 2, 2019 at 16:17 | comment | added | RobertF | Thank you, very helpful. I'll take a look at the papers you've cited - do you have any references for refutations of King & Nielsen's paper? Are propensity score matching and exact matching completely out-of-date, or just two of many approaches that ought to be tested & compared with more recent methods? Another question (which perhaps I should post separately) is how to compare a technique like exact matching with some of the machine learning models you've mentioned. For example, CEM and PSM can be compared via balance diagnostics on the covariates, but not with regression or ML models. | |
| Jul 1, 2019 at 21:41 | history | answered | Noah | CC BY-SA 4.0 |