Joe Bak-Coleman
@jbakcoleman.bsky.social
Research Scientist at the University of Washington based in Brooklyn. Also: SFI External Applied Fellow, Harvard BKC affiliate. Collective Behavior, Statistics, etc..
If they’re really hoping to find an answer, not just launder known nulls through willing scientists, why not publish their results from 2019? They seemed to recognize they were of interest more broadly.
November 24, 2025 at 12:58 PM
If they’re really hoping to find an answer, not just launder known nulls through willing scientists, why not publish their results from 2019? They seemed to recognize they were of interest more broadly.
On teen mental health, the lawsuit is replete with indications that the content meta serves teens in the problem…
So when they engaged @cos.io for collaboration, they explicitly opted not to provide information about the type of content viewed. Chop off the mechanism and get a null.
So when they engaged @cos.io for collaboration, they explicitly opted not to provide information about the type of content viewed. Chop off the mechanism and get a null.
November 24, 2025 at 12:55 PM
On teen mental health, the lawsuit is replete with indications that the content meta serves teens in the problem…
So when they engaged @cos.io for collaboration, they explicitly opted not to provide information about the type of content viewed. Chop off the mechanism and get a null.
So when they engaged @cos.io for collaboration, they explicitly opted not to provide information about the type of content viewed. Chop off the mechanism and get a null.
As we discuss in this piece, withdrawal experiments will have downward bias in their effect sizes for things like polarization. If they got a small effect or a goose egg, jt may have made them feel a lot more confident about their collaborative experiments with academics.
arxiv.org/abs/2505.09254
arxiv.org/abs/2505.09254
November 24, 2025 at 12:46 PM
As we discuss in this piece, withdrawal experiments will have downward bias in their effect sizes for things like polarization. If they got a small effect or a goose egg, jt may have made them feel a lot more confident about their collaborative experiments with academics.
arxiv.org/abs/2505.09254
arxiv.org/abs/2505.09254
It would, of course, be great if the underlying results were made public but it looks like meta in 2019 ran withdrawal experiments to examine polarization and well being ahead of forging collaborations with academics. Seems like more than enough to design bias the collaborations towards nulls 🧵
November 24, 2025 at 12:46 PM
It would, of course, be great if the underlying results were made public but it looks like meta in 2019 ran withdrawal experiments to examine polarization and well being ahead of forging collaborations with academics. Seems like more than enough to design bias the collaborations towards nulls 🧵
“Are we the baddies?”
November 24, 2025 at 3:24 AM
“Are we the baddies?”
It’s really hard to defend industry academic collaborations with meta as earnest, if they’re internally burying evidence of harm.
www.reuters.com/sustainabili...
www.reuters.com/sustainabili...
November 23, 2025 at 6:22 PM
It’s really hard to defend industry academic collaborations with meta as earnest, if they’re internally burying evidence of harm.
www.reuters.com/sustainabili...
www.reuters.com/sustainabili...
Fantastic event @astoriabookshop.bsky.social for the launch of Lindsay Stuart Hill’s world of dew with readings from @emilyhockaday.bsky.social.
Two of my favorites so far when going through the books… it’s January still by LSH and River Clay by EH.
Two of my favorites so far when going through the books… it’s January still by LSH and River Clay by EH.
November 21, 2025 at 6:10 PM
Fantastic event @astoriabookshop.bsky.social for the launch of Lindsay Stuart Hill’s world of dew with readings from @emilyhockaday.bsky.social.
Two of my favorites so far when going through the books… it’s January still by LSH and River Clay by EH.
Two of my favorites so far when going through the books… it’s January still by LSH and River Clay by EH.
It all looks independent! So independent, in fact, that the authors defending the collaboration with Meta affirmatively declare no competing interests. Even when a core argument they're making is that they were great about declaring competing interests...
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
November 19, 2025 at 1:57 PM
It all looks independent! So independent, in fact, that the authors defending the collaboration with Meta affirmatively declare no competing interests. Even when a core argument they're making is that they were great about declaring competing interests...
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
It all looks like rigor but its design bias. From the jump, Meta's got big papers with likely goose-eggs in the pipeline. You're not paying the academics involved for this (but see below), so you can claim independence. Of course, they're all getting a lifetime of papers in top journals.
November 19, 2025 at 1:57 PM
It all looks like rigor but its design bias. From the jump, Meta's got big papers with likely goose-eggs in the pipeline. You're not paying the academics involved for this (but see below), so you can claim independence. Of course, they're all getting a lifetime of papers in top journals.
Let's take what (imo) is the focal question of the whole collaboration: Could it shift the vote? The work suggests a ~2.5% shift, enough to swing an election. Non-significant after multiple comparisons corrections. What do these multiple comparison corrections encode?
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
November 19, 2025 at 1:57 PM
Let's take what (imo) is the focal question of the whole collaboration: Could it shift the vote? The work suggests a ~2.5% shift, enough to swing an election. Non-significant after multiple comparisons corrections. What do these multiple comparison corrections encode?
www.pnas.org/doi/10.1073/...
www.pnas.org/doi/10.1073/...
Post Malört ergo propter malört
November 15, 2025 at 5:37 AM
Post Malört ergo propter malört
Daily dose of Fred
November 12, 2025 at 3:02 AM
Daily dose of Fred
Little guy got a clean bill of health.
November 8, 2025 at 5:53 PM
Little guy got a clean bill of health.
November 5, 2025 at 1:15 AM
I think that view is common. Here’s the center for open sciences take on why you need to preregister. It distinguishes your work as confirmatory which they assert a lot about.
www.cos.io/initiatives/...
www.cos.io/initiatives/...
November 2, 2025 at 8:31 PM
I think that view is common. Here’s the center for open sciences take on why you need to preregister. It distinguishes your work as confirmatory which they assert a lot about.
www.cos.io/initiatives/...
www.cos.io/initiatives/...
Meet Fred! You’ll be seeing more of him….
November 1, 2025 at 6:27 PM
Meet Fred! You’ll be seeing more of him….
October 29, 2025 at 10:02 PM
Assessing impact in this way and allocating funding is pretty: “yo dog I heard you like cumulative advantage in science so I added cumulative advantage to cumulative advantage”
October 29, 2025 at 12:56 PM
Assessing impact in this way and allocating funding is pretty: “yo dog I heard you like cumulative advantage in science so I added cumulative advantage to cumulative advantage”
It’s weird right? The old “but we make money on doing it anyway” loophole.
October 24, 2025 at 1:44 AM
It’s weird right? The old “but we make money on doing it anyway” loophole.
Solutions like open science, preregistration and registered reports fall apart in this space because much of the research isn't p-hacked... it's design biased.
For example, the thresholds industry studies use for significance when espousing benefits tend to be looser than identifying harm.
For example, the thresholds industry studies use for significance when espousing benefits tend to be looser than identifying harm.
October 24, 2025 at 12:12 AM
Solutions like open science, preregistration and registered reports fall apart in this space because much of the research isn't p-hacked... it's design biased.
For example, the thresholds industry studies use for significance when espousing benefits tend to be looser than identifying harm.
For example, the thresholds industry studies use for significance when espousing benefits tend to be looser than identifying harm.
They similarly tend to selectively fund and support research that aligns with their interests. We have no shortage of industry funded studies on labeling misinformation---but how often have they given independent access to into what they show users?
October 24, 2025 at 12:12 AM
They similarly tend to selectively fund and support research that aligns with their interests. We have no shortage of industry funded studies on labeling misinformation---but how often have they given independent access to into what they show users?
Selective Causal Focus: Research produced and funded by tech companies often either frames problems as user-driven, or solutions as the obligation of users (E.g. community notes). Distracting us from their design, business model, interface, and other causes steering us away from their profit model
October 24, 2025 at 12:12 AM
Selective Causal Focus: Research produced and funded by tech companies often either frames problems as user-driven, or solutions as the obligation of users (E.g. community notes). Distracting us from their design, business model, interface, and other causes steering us away from their profit model