Robin Blythe
@rbly.bsky.social
Assistant professor at Duke-NUS medical school. Mostly interested in health economics, biostats and clinical informatics.
Did you incorporate the model within the Shiny app, or just the output? I have tried to put a Gibbs sampler in a free shinyapps.io server before and it just runs out of compute :(
November 27, 2025 at 10:44 AM
Did you incorporate the model within the Shiny app, or just the output? I have tried to put a Gibbs sampler in a free shinyapps.io server before and it just runs out of compute :(
I like the idea of calibration though, you could use the values you get as some kind of representation of their perception of risk.
November 27, 2025 at 10:34 AM
I like the idea of calibration though, you could use the values you get as some kind of representation of their perception of risk.
Not sure about in behavioural economics, but in quality of life measurement it's common to have people do this implicitly through trade-offs of A vs B with probabilities P1 vs P2. I think a lot of studies start with the assumption that people suck at understanding probabilities.
November 27, 2025 at 10:34 AM
Not sure about in behavioural economics, but in quality of life measurement it's common to have people do this implicitly through trade-offs of A vs B with probabilities P1 vs P2. I think a lot of studies start with the assumption that people suck at understanding probabilities.
Cool study! Only briefly browsed it but, tangentially, looks like it would be quite useful in health economics for discounting unlikely future events from the patient's perspective rather than just slapping a 1/(1+r)^t NPV calculation on everything.
November 27, 2025 at 9:39 AM
Cool study! Only briefly browsed it but, tangentially, looks like it would be quite useful in health economics for discounting unlikely future events from the patient's perspective rather than just slapping a 1/(1+r)^t NPV calculation on everything.
It was, I confess, not very data-driven of me.
November 27, 2025 at 9:33 AM
It was, I confess, not very data-driven of me.
Yes, true - it's just more of a clearinghouse than a curated journal. Which is OK, but even arXiv and OSF have pretty good quality control!
November 27, 2025 at 9:19 AM
Yes, true - it's just more of a clearinghouse than a curated journal. Which is OK, but even arXiv and OSF have pretty good quality control!
I got a really funny complaint from reviewer when I was showing that TTE models can outperform LR with repeated measurements. Their complaint was that my LR was unrealistically good because I used cubic splines and pre-specified clinically important variables. First time I LOL'd reading peer review.
November 27, 2025 at 8:40 AM
I got a really funny complaint from reviewer when I was showing that TTE models can outperform LR with repeated measurements. Their complaint was that my LR was unrealistically good because I used cubic splines and pre-specified clinically important variables. First time I LOL'd reading peer review.
I've started going straight to the methods after the abstract. Can usually smell that the claims are overhyped by the time I get about 2 paragraphs into the analysis section. Honestly barely touch intro/discussion these days unless the paper is really well-written. It's just fluff most of the time!
November 27, 2025 at 8:36 AM
I've started going straight to the methods after the abstract. Can usually smell that the claims are overhyped by the time I get about 2 paragraphs into the analysis section. Honestly barely touch intro/discussion these days unless the paper is really well-written. It's just fluff most of the time!
Claudius would have loved this. Just fractionally too late for him to see it.
November 27, 2025 at 8:33 AM
Claudius would have loved this. Just fractionally too late for him to see it.
I have a paper published with SR on the suggestion of a prior supervisor. Didn't realise until after submission and review (I had to find my own editors lol) that it was such a poorly disguised mill.
November 27, 2025 at 8:30 AM
I have a paper published with SR on the suggestion of a prior supervisor. Didn't realise until after submission and review (I had to find my own editors lol) that it was such a poorly disguised mill.
Risk aversion seems like it also plays a role which, together with a low expected value, seems pretty rational.
November 27, 2025 at 8:28 AM
Risk aversion seems like it also plays a role which, together with a low expected value, seems pretty rational.
As in, people who have historically not had anything go their way will assume that some future decision will also not go their way (and vice versa)? That would be a rational decision if the expected utility of an outcome was low.
November 27, 2025 at 8:28 AM
As in, people who have historically not had anything go their way will assume that some future decision will also not go their way (and vice versa)? That would be a rational decision if the expected utility of an outcome was low.
I feel the same. Maybe it's the intuition that you may be, for once, fully in control of the model fitting process. Catching lightning in a bottle type stuff. Definitely beats the hours of project management every day to keep grants moving.
November 26, 2025 at 3:17 PM
I feel the same. Maybe it's the intuition that you may be, for once, fully in control of the model fitting process. Catching lightning in a bottle type stuff. Definitely beats the hours of project management every day to keep grants moving.
Last American conference I attended, they didn't feed us and there was no coffee (still a $1k+ attendance fee). Barbaric.
November 26, 2025 at 8:41 AM
Last American conference I attended, they didn't feed us and there was no coffee (still a $1k+ attendance fee). Barbaric.
Ahh, that's good. If I built the tool in R then I'll make a shiny app and check the logins, but it usually drops off dramatically after a few months. Hope you put the tool in your CV!
November 26, 2025 at 2:40 AM
Ahh, that's good. If I built the tool in R then I'll make a shiny app and check the logins, but it usually drops off dramatically after a few months. Hope you put the tool in your CV!
Actually, I've also heard about organisations using the same Excel-based survival model that a consultant built in 2001 and is now hideously out of date, but everyone is now too afraid to modify because nobody knows how it works.
November 26, 2025 at 1:29 AM
Actually, I've also heard about organisations using the same Excel-based survival model that a consultant built in 2001 and is now hideously out of date, but everyone is now too afraid to modify because nobody knows how it works.
This is so cool! I've developed tools for hospital decision-making before (also in Excel) but it's always impossible to learn whether your contributions are still being used just a few months later.
November 26, 2025 at 1:27 AM
This is so cool! I've developed tools for hospital decision-making before (also in Excel) but it's always impossible to learn whether your contributions are still being used just a few months later.
Quarto/Rmarkdown is genuinely a game changer for literate coding, especially if explaining your process to non-experts (or if, like me, your code is hideously inefficient and you need half a paragraph to explain the shit you're doing)
November 26, 2025 at 1:13 AM
Quarto/Rmarkdown is genuinely a game changer for literate coding, especially if explaining your process to non-experts (or if, like me, your code is hideously inefficient and you need half a paragraph to explain the shit you're doing)
That said, the worst DS area will always be Horsefuck Valley
November 25, 2025 at 1:06 PM
That said, the worst DS area will always be Horsefuck Valley
Urgh. I think this is what bounced me off the first time I played this game, had a woefully equipped dark magic build. Second time's the charm, but nearly gave up on Fume Knight (shudders)
November 25, 2025 at 12:31 PM
Urgh. I think this is what bounced me off the first time I played this game, had a woefully equipped dark magic build. Second time's the charm, but nearly gave up on Fume Knight (shudders)
This is great! I've had a paper on income losses due to HPV-attributable cancers tied up in peer review at PLOS One for almost a year but it would be fantastic to cite the updated reviews when I get the chance, will really strengthen the arguments in the discussion
November 24, 2025 at 2:37 PM
This is great! I've had a paper on income losses due to HPV-attributable cancers tied up in peer review at PLOS One for almost a year but it would be fantastic to cite the updated reviews when I get the chance, will really strengthen the arguments in the discussion
I agree, but it seems like governments think they can (and have) weaponise it, so they're uninterested in reining it in.
November 24, 2025 at 11:39 AM
I agree, but it seems like governments think they can (and have) weaponise it, so they're uninterested in reining it in.