I'm open to these claims, but they ultimately need to generate testable hypotheses or we're not doing science. We have testable hypotheses about viruses and the immune system!
-
-
Right now claims about low-volume social media operations having big systemic effects lack credible supporting evidence or even specific claims that could be tested.
11 replies 15 retweets 79 likes -
Replying to @BrendanNyhan @NateSilver538
Many complex systems & past events can't have simple testable hypotheses and refusing to study them would not be good. This isn't even a scial science problem. Geology manages. Also, as I keep arguing, the DNC hack and aftermath is part of this so some impacts are face-value big.
1 reply 6 retweets 31 likes -
zeynep tufekci Retweeted
Plus, I'm increasingly wondering if it is that low-volume. (See here, which is, almost by definition an undercount since we only know what's been uncovered). At the moment, anything we can measure is a modest estimate for floor since it was covert. https://twitter.com/antoniogm/status/1075854442191175680 …)
zeynep tufekci added,
This Tweet is unavailable.1 reply 0 retweets 11 likes -
Once again, though, my MAIN argument is that Nate's metric: percent of social media posts, is not strong. Social media is characterized by interaction, feedback loops, mainstream media interaction, narrative shifting, meme-copying etc. % originating is not a decisive measure.
5 replies 4 retweets 56 likes -
Replying to @zeynep @BrendanNyhan
My quick-and-dirty metric is actually social media *impressions*, not posts. The Russian troll posts constituted a *lower* share of impressions than posts, which really cuts AGAINST your hypothesis (i.e. they punched below their weight rather than going viral).
9 replies 0 retweets 14 likes -
How about this as a **testable** theory: HYPOTHESIS: If Russian troll posts were more influential than the raw numbers imply, they should have higher-than-average engagement (retweets/replies/faves) from blue-checkmark journalist Twitter accounts than other 2016 content.
7 replies 0 retweets 11 likes -
Replying to @NateSilver538 @BrendanNyhan
That's just taking us back to Skinner, skipping over the whole cognitive revolution. People read and interpret, even if they don't react (especially to trolling). But, if we had full data, we could measure narratives they initiated -> media narratives. Theoretically measurable.
2 replies 3 retweets 23 likes -
Replying to @zeynep @BrendanNyhan
OK, it's pretty telling IMO that you've rejected an **actually** testable (in practice, not just in theory) hypothesis for ones that are either impossible or extremely cumbersome to test.
4 replies 0 retweets 11 likes -
Replying to @NateSilver538 @BrendanNyhan
Is it possible that it's not a good one? Why would people actively respond *more* to trolling that nevertheless impacted them? How would you ever measure impacts of abusive behavior online? We know: people tend to go silent. There really was a reason we moved beyond Skinner.
5 replies 0 retweets 18 likes
Yes my suggestion is maximal. You can look for bites. There might be natural experiments. Companies have longitudinal data—some of it may eventually be analyzed (if they are not already doing so). We should look back at panels that exist, etc. Weak, easy measures are still weak.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.