James Steele
@jamessteeleii.bsky.social
I'm just a guy who's an academic for fun
Anyone have a copy of the following (pic because citation too long for character limit, also see alt text).
November 10, 2025 at 8:38 AM
Anyone have a copy of the following (pic because citation too long for character limit, also see alt text).
This is something I have been thinking about how to apply more in our work as in below (and explained in this talk: www.youtube.com/watch?v=39Aj...)
October 31, 2025 at 10:47 AM
This is something I have been thinking about how to apply more in our work as in below (and explained in this talk: www.youtube.com/watch?v=39Aj...)
This scene living rent free in my head 😳
October 27, 2025 at 6:33 AM
This scene living rent free in my head 😳
It only got more chaotic after that as everyone's arsenals were deployed... at one point I had a full 180 spin out and we had a head on collision take place.
Great fun, and we'll be playing again for sure.
🧵4/4
Great fun, and we'll be playing again for sure.
🧵4/4
October 22, 2025 at 10:32 AM
It only got more chaotic after that as everyone's arsenals were deployed... at one point I had a full 180 spin out and we had a head on collision take place.
Great fun, and we'll be playing again for sure.
🧵4/4
Great fun, and we'll be playing again for sure.
🧵4/4
Needless to say mayhem ensued... we had a stall on the start line and wrecks taking place immediately as the various ramming implements (e.g., the mechanical leg salvaged from a walker on my wagon) were put to use.
🧵3/4
🧵3/4
October 22, 2025 at 10:32 AM
Needless to say mayhem ensued... we had a stall on the start line and wrecks taking place immediately as the various ramming implements (e.g., the mechanical leg salvaged from a walker on my wagon) were put to use.
🧵3/4
🧵3/4
My friend had been long planning a game for "Orktober" and reskinned the game "Gaslands" where you have Mad Max style races to be a bit more... well, more Orky. So Daze of Funda was born! He also 3d printed us all the same chassis to use and we all kitbashed and kustomised our own Battlewagons.
🧵2/4
🧵2/4
October 22, 2025 at 10:32 AM
My friend had been long planning a game for "Orktober" and reskinned the game "Gaslands" where you have Mad Max style races to be a bit more... well, more Orky. So Daze of Funda was born! He also 3d printed us all the same chassis to use and we all kitbashed and kustomised our own Battlewagons.
🧵2/4
🧵2/4
Check out this cute ass copy of On Bullshit I found.
October 14, 2025 at 11:26 AM
Check out this cute ass copy of On Bullshit I found.
Meanwhile, Joslyn collected new data from 8 participants in a crossover design. We analysed his data using the priors from the meta-analysis. The result? His study barely shifted the posterior beliefs — the prior and posterior distributions almost perfectly overlapped.
5/10
5/10
![A two-panel forest-style plot showing results from an empirical study comparing the effects of ammonia inhalants, placebo, and control conditions on Peak Force (left) and Rate of Force Development (right). The x-axis represents Standardised Mean Difference (SMD), with dotted vertical lines marking the null region (–0.1 to 0.1). Each panel displays posterior distributions (gray density shapes) with mean estimates, 95% quantile intervals, and 2×log(BF) values indicating Bayes factors relative to the null region.
Peak Force panel:
Placebo vs Control: SMD = –0.08 [–0.28, 0.11], 2×log(BF) = 0.26
Ammonia vs Placebo: SMD = 0.22 [–0.01, 0.47], 2×log(BF) = 1.06
Ammonia vs Control: SMD = 0.14 [0.01, 0.30], 2×log(BF) = 0.91
Rate of Force Development panel:
Placebo vs Control: SMD = –0.08 [–0.28, 0.10], 2×log(BF) = –0.33
Ammonia vs Placebo: SMD = 0.30 [0.09, 0.52], 2×log(BF) = –0.53
Ammonia vs Control: SMD = 0.22 [0.09, 0.35], 2×log(BF) = 0.02
A note below the panels explains that 2×log(BF) values indicate the strength of evidence against the null, following the Kass and Raftery (1995) interpretation scale:
Negative = evidence for the null,
0–2 = weak,
2–6 = positive,
6–10 = strong,
10 = very strong evidence for an effect.
Overall, the posterior distributions overlap substantially, suggesting minimal change in beliefs from the prior evidence.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:evz2lw2i64j2p6qaawvfoohz/bafkreicl4mmfxkd626orcn5y2qf3imvb42jiraef75e66frvjkrvpx4yne@jpeg)
October 13, 2025 at 9:07 AM
Meanwhile, Joslyn collected new data from 8 participants in a crossover design. We analysed his data using the priors from the meta-analysis. The result? His study barely shifted the posterior beliefs — the prior and posterior distributions almost perfectly overlapped.
5/10
5/10
While most individual studies were mostly “null,” collectively they suggested small, possibly positive effects on some outcomes — though with modest precision.
4/10
4/10
![A three-panel forest-style plot showing results from a contrast-based network meta-analysis comparing the effects of ammonia inhalants, placebo, and control conditions on power, speed, and strength outcomes. Each panel (Power, Speed, Strength) shows standardized mean differences (SMDs) and 95% quantile intervals for each contrast (Placebo vs Control, Ammonia vs Placebo, Ammonia vs Control).
Power:
Placebo vs Control: SMD = –0.09 [–0.26, 0.09], Probability of null effect = 51%, meaningful effect = 2%.
Ammonia vs Placebo: SMD = 0.33 [0.19, 0.46], Probability of null = 0%, meaningful = 100%.
Ammonia vs Control: SMD = 0.23 [0.10, 0.38], Probability of null = 2%, meaningful = 98%.
Speed:
Placebo vs Control: SMD = 0.05 [–0.34, 0.45], Probability of null = 38%, meaningful = 41%.
Ammonia vs Placebo: SMD = 0.02 [–0.35, 0.40], Probability of null = 40%, meaningful = 34%.
Ammonia vs Control: SMD = 0.08 [–0.25, 0.40], Probability of null = 42%, meaningful = 44%.
Strength:
Placebo vs Control: SMD = –0.06 [–0.27, 0.15], Probability of null = 57%, meaningful = 6%.
Ammonia vs Placebo: SMD = 0.18 [0.01, 0.36], Probability of null = 19%, meaningful = 81%.
Ammonia vs Control: SMD = 0.12 [0.03, 0.28], Probability of null = 42%, meaningful = 57%.
Each panel shows posterior distributions as gray density shapes, centered around the estimated SMDs. Dotted vertical lines indicate the null region (–0.1 to 0.1). Text at the bottom notes that probabilities of null and meaningful effects were calculated as the proportion of posterior mass within or beyond this region, respectively.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:evz2lw2i64j2p6qaawvfoohz/bafkreig6dy4k5zxoxsdi27kbxtpbleyysp27pivevv62rs7hayeai7u654@jpeg)
October 13, 2025 at 9:07 AM
While most individual studies were mostly “null,” collectively they suggested small, possibly positive effects on some outcomes — though with modest precision.
4/10
4/10
Certainly from a Meehlian perspective, and even considering it's essentially retrodictive, that would tend to provide further corroboration for the theory that trained folks just don't really grow much. Especially when combined with our estimates from the latest Project Discover.
🧵6/6
🧵6/6
October 2, 2025 at 10:14 AM
Certainly from a Meehlian perspective, and even considering it's essentially retrodictive, that would tend to provide further corroboration for the theory that trained folks just don't really grow much. Especially when combined with our estimates from the latest Project Discover.
🧵6/6
🧵6/6
So I went back and pulled out the data from these studies to take a look, and lo and behold the estimates weren't that far off what we'd expect... the confidence interval still contains zero, but the estimate is fairly close with the power/precision boost from pooling the data.
🧵5/6
🧵5/6
![A forest plot-style graph titled “Degree of corroboration of theory derived predictions” with the subtitle "Grey band indicates the theory tolerance i.e., interval estimate derived from theory; Spielraum based upon 95% prediction interval from large scale meta-analysis (DOI: 10.1080/02640414.2023.2286748) of varied resistance training interventions". The x-axis is labeled “Standardised Mean Effect” ranging from approximately -0.25 to 1.00. A black point estimate with a narrow horizontal error bar is shown with the value 0.02 [95%CI: -0.01, 0.05] printed next to it, derived from individual participant-level modeling of previous studies (n=323). A much wider horizontal interval labeled “Spielraum” spans roughly -0.1 to 0.77, representing the 95% prediction interval from a large-scale meta-analysis. A shaded grey vertical band represents the predicted effect from theory ranging an interval of 0.042 to 0.057, indicating theory tolerance. Text annotations on the right report a Corroboration Index of 0.96 and a normalised Corroboration Index of 0.93. The plot caption reads "Estimate derived from individual participant (n=323) level modelling of data from previous studies; DOIs: 10.1139/apnm-2014-0162; 10.1519/JSC.0000000000001222; 10.1139/apnm-2016-0180; 10.1139/apnm-2018-0376; 10.1080/02701367.2022.2097625; Main effect of time (weeks converted to 12 week effect) from: Fat Free Mass Z-Score ~ Weeks + (1|Study/Participant)".](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:evz2lw2i64j2p6qaawvfoohz/bafkreid26gfe56f6djcbt6obhlrv7qzios23thdkewmg56w5cs6ja6b4xe@jpeg)
October 2, 2025 at 10:14 AM
So I went back and pulled out the data from these studies to take a look, and lo and behold the estimates weren't that far off what we'd expect... the confidence interval still contains zero, but the estimate is fairly close with the power/precision boost from pooling the data.
🧵5/6
🧵5/6
Excited to have now started new role as Head of Research with Macrofactor (macrofactorapp.com). Looking forward to getting stuck into some really cool projects and getting that out to all the users!
September 30, 2025 at 3:59 PM
Excited to have now started new role as Head of Research with Macrofactor (macrofactorapp.com). Looking forward to getting stuck into some really cool projects and getting that out to all the users!
#SportScience, especially in elite sports, has an issue that can make effects seem more impressive than they really are... selection bias on the sample (explanation🧵 and a simulation pictured to illustrate below).
1/12
1/12
September 28, 2025 at 6:59 PM
#SportScience, especially in elite sports, has an issue that can make effects seem more impressive than they really are... selection bias on the sample (explanation🧵 and a simulation pictured to illustrate below).
1/12
1/12
Me after finishing Expedtition 33...
September 22, 2025 at 10:13 AM
Me after finishing Expedtition 33...
How is it that time of year again already when everyone is "honoured" to be included in the top X% of bullshit citation metrics in the "Stanford List"?
Anyone I see posting such crap is getting blocked and reported 😊
Anyone I see posting such crap is getting blocked and reported 😊
September 22, 2025 at 9:50 AM
How is it that time of year again already when everyone is "honoured" to be included in the top X% of bullshit citation metrics in the "Stanford List"?
Anyone I see posting such crap is getting blocked and reported 😊
Anyone I see posting such crap is getting blocked and reported 😊
Me when speaking to sport and exercise science researchers who won't commit to a smallest effect size of interest.
September 15, 2025 at 7:08 PM
Me when speaking to sport and exercise science researchers who won't commit to a smallest effect size of interest.
Post a meme made by you.
August 31, 2025 at 6:46 PM
Post a meme made by you.
@brianhaig.bsky.social has a great book about abduction more broadly (not just inference to the best explanation) - direct.mit.edu/books/monogr...
I always tried to present things to students as a sort of holistic view of inference in the sciences; that each mode can have value for different aims.
I always tried to present things to students as a sort of holistic view of inference in the sciences; that each mode can have value for different aims.
August 26, 2025 at 9:06 AM
@brianhaig.bsky.social has a great book about abduction more broadly (not just inference to the best explanation) - direct.mit.edu/books/monogr...
I always tried to present things to students as a sort of holistic view of inference in the sciences; that each mode can have value for different aims.
I always tried to present things to students as a sort of holistic view of inference in the sciences; that each mode can have value for different aims.
I guess I should add then... this is how I tend to try to practice science, and where pre-registration fits in for me i.e., the part where I make specific statistical claims based on tests conducted in particular operational contexts.
August 23, 2025 at 8:38 AM
I guess I should add then... this is how I tend to try to practice science, and where pre-registration fits in for me i.e., the part where I make specific statistical claims based on tests conducted in particular operational contexts.
And now we return to our regularly scheduled programming... 🥰
August 13, 2025 at 5:27 AM
And now we return to our regularly scheduled programming... 🥰
Here it is osf.io/w2qpx
Plus the hot take that, at least above RDA (and based on reanalysis of the Nunes et al. 2022 data), protein intake barely matters for lean mass/muscle adaptation.
Plus the hot take that, at least above RDA (and based on reanalysis of the Nunes et al. 2022 data), protein intake barely matters for lean mass/muscle adaptation.
July 26, 2025 at 8:30 PM
Here it is osf.io/w2qpx
Plus the hot take that, at least above RDA (and based on reanalysis of the Nunes et al. 2022 data), protein intake barely matters for lean mass/muscle adaptation.
Plus the hot take that, at least above RDA (and based on reanalysis of the Nunes et al. 2022 data), protein intake barely matters for lean mass/muscle adaptation.
Casual (maybe juvenile trapdoor) spider strolling across the path on this morning's walk.
July 7, 2025 at 11:32 PM
Casual (maybe juvenile trapdoor) spider strolling across the path on this morning's walk.
For example, here we have the range prediction from theory, the background range of effects observed in other studies absent the theory (Meehl's "spielraum"), and the estimates from the study... I wonder if it's maybe that instead from a Bayesian perspective the spielraum should be the prior?
July 5, 2025 at 8:02 AM
For example, here we have the range prediction from theory, the background range of effects observed in other studies absent the theory (Meehl's "spielraum"), and the estimates from the study... I wonder if it's maybe that instead from a Bayesian perspective the spielraum should be the prior?
To clarify, I am thinking more of how the bayesian machinery works with it when you move from theoreticians problem to statisticians problem, whether or not that ends up operationalised as a "statistical hypothesis test" in the way that language is used e.g., some sort of Bayes Factor.
July 5, 2025 at 5:25 AM
To clarify, I am thinking more of how the bayesian machinery works with it when you move from theoreticians problem to statisticians problem, whether or not that ends up operationalised as a "statistical hypothesis test" in the way that language is used e.g., some sort of Bayes Factor.
Anyway, I guess I'm just a Meehl shill anyway... But hoping the t shirt I wore Tuesday at #MetaScience2025 influenced a few folks 😉
July 3, 2025 at 9:29 AM
Anyway, I guess I'm just a Meehl shill anyway... But hoping the t shirt I wore Tuesday at #MetaScience2025 influenced a few folks 😉