5 ms·

The paper can be downloaded from https://poseidon01.ssrn.com/delivery.php?ID=8201061010270271000930651041260741220370890890420530261070971240810731270640170181

The paper can be downloaded from https://poseidon01.ssrn.com/delivery.php?ID=820106101027027100093065104126074122037089089042053026107097124081073127064017018108060016059063040047117002119077004085103076007082044064086005008086000121064100092030015033096112070030080089119095026113074126068090099117115122106127084072103124120120&EXT=pdf https://poseidon01.ssrn.com/delivery.php?ID=8201061010270271... I have only skimmed through the paper, but I am a bit skeptical about their conclusions. The authors first showed that the attacks-on-refugees time series is strongly correlated to the anti-refugee-posts time series. This part is very basic and is OK. Then they tried to exhibit a causal relationship between the two series, and it is the trend of posts that drives the trend of attacks but not the other way around. This is where I have doubts. Apparently, the authors didn't try something basic first --- like a Granger causality test --- but applied a somewhat obscure regression procedure to establish causality. This procedure, which they called a Bartik-type approach, was invented only in 1991. Now I'm not familiar enough with time series analysis to know the merits of this Bartik-type approach and why Granger test is not useful in this case, but from the graph in the Economist article, it seems that the trend of attacks is leading the trend of posts, so I wonder if they were deliberately omitting a Granger causality test that might refute their conclusions.

RasputinsBro 9y ago
Just to point out that "Granger causality test" is a terrible name as is does not really test causality, it tests correlations between time-lagged series; time-lagged correlation is not causation. It's just a popular correlation test in the field of autocorrelated time-series.
roenxi 9y ago
It'd be more comforting to see something done by a psychologist. A correlation between physical violence and negative verbal sentiment doesn't seem that surprising. That talk on an online forum precedes violence is plausible. Claiming talk online /causes/ violence is a little surprising and has dangerous political implications, ie, makes a case to target speech of AfD, which polled a respectable ~10% in the last German federal election.
Const-me 9y ago
Another issue, they failed to investigate whether both things are caused by something else. E.g. what was the dynamic of migration inflow? If it has similar shape, this might explain both observations.

RasputinsBro 9y ago

Just to point out that "Granger causality test" is a terrible name as is does not really test causality, it tests correlations between time-lagged series; time-lagged correlation is not causation. It's just a popular correlation test in the field of autocorrelated time-series.

roenxi 9y ago

It'd be more comforting to see something done by a psychologist. A correlation between physical violence and negative verbal sentiment doesn't seem that surprising. That talk on an online forum precedes violence is plausible. Claiming talk online /causes/ violence is a little surprising and has dangerous political implications, ie, makes a case to target speech of AfD, which polled a respectable ~10% in the last German federal election.

Const-me 9y ago

Another issue, they failed to investigate whether both things are caused by something else. E.g. what was the dynamic of migration inflow? If it has similar shape, this might explain both observations.