Note: this is the first of several posts discussing the 2020 Meta Election research partnership and the resulting papers, in advance of a keynote panel at IC2S2 on the topic next Thursday July 18.
The thing about priors is that they are personal. So, the conclusion is something like "if you are open to the idea that social media influences politics, but not committed either way, this evidence should change your beliefs in the direct of believing this idea"
It's the absence of an impersonal, and therefore pseudo-objective, measure that accounts for the persistence of the classical hypothesis testing framework, even though it's mostly voodoo.
Great post. Do you think the effects that we can actually measure rigorously (for example with a RCT) may be always small (my prior), while the large effects follow these long term historical mood shifts? Something clearly changed with social media, but we can't really isolate the control group in a satisfactory manner.
To be clear, we can have huge effects when the cause is physical (e.g., vaccines). How feeds work isn't physical, feeds changes the way we interact normatively with others, how we communicate, how we give and take reasons.
It always comes back to effect sizes and scope conditions. We can measure the effect of the guillotine with a tiny sample.
On the point of “physical”—I think any attempt to come up with rules for what treatments have large vs small effects are fraught. Leaded gasoline is physical but more like the effect of social media. The video of trump’s attempted assissination is a short media treatment but might plausibly have a large effect. I think we need theory for this
The thing about priors is that they are personal. So, the conclusion is something like "if you are open to the idea that social media influences politics, but not committed either way, this evidence should change your beliefs in the direct of believing this idea"
It's the absence of an impersonal, and therefore pseudo-objective, measure that accounts for the persistence of the classical hypothesis testing framework, even though it's mostly voodoo.
Do you think we're better off with voodoo, or should we try to formalize our priors? I lean in the latter direction.
I’m all for formalising thinking but it won’t lead to common priors just clarified disagreement about them
Great post. Do you think the effects that we can actually measure rigorously (for example with a RCT) may be always small (my prior), while the large effects follow these long term historical mood shifts? Something clearly changed with social media, but we can't really isolate the control group in a satisfactory manner.
To be clear, we can have huge effects when the cause is physical (e.g., vaccines). How feeds work isn't physical, feeds changes the way we interact normatively with others, how we communicate, how we give and take reasons.
It always comes back to effect sizes and scope conditions. We can measure the effect of the guillotine with a tiny sample.
On the point of “physical”—I think any attempt to come up with rules for what treatments have large vs small effects are fraught. Leaded gasoline is physical but more like the effect of social media. The video of trump’s attempted assissination is a short media treatment but might plausibly have a large effect. I think we need theory for this