Read on Twitter
#SwissCovidApp (preprint):
"Applying the rule used by the Swiss Covid-19
contact tracing app [...] we find that no exposure notifications would have been triggered despite the fact that all pairs of handsets were within 2m of one another for at least 15 mins."
"Applying the rule used by the Swiss Covid-19
contact tracing app [...] we find that no exposure notifications would have been triggered despite the fact that all pairs of handsets were within 2m of one another for at least 15 mins."
Now, what's going on here and why is it the best data so far?
First, it's the best data because it's the only data, Leith and Farerrell have made it freely available here:
https://github.com/doug-leith/dublinbus_gaen_dataset
So they tested this on a bus in Dublin
First, it's the best data because it's the only data, Leith and Farerrell have made it freely available here:
https://github.com/doug-leith/dublinbus_gaen_dataset
So they tested this on a bus in Dublin
Five participants sat with phones in the locations shown in Fig. 1 (b) and the participants were asked to hold the phones in their hands. After 15 minutes, the participants were asked to switch seats. These two runs were then repeated with 2 participants having been ...
replaced on the upper deck of the bus (yeah :-)). To get a baseline, the phones were placed in the center aisle and, starting at 0.5m, were moved by another 0.5m ever 5 minutes. Looking at this measurement, we see that between 1.5 and 2.5m the attenuation DECREASES.
Why is that? Well, the movement of radio signals in a metal tube with holes is complex (reflections, etc.). Now, this was just the baseline, now let's add some people. And now it starts to get interesting: According to BLE attenuation, the person furthest away is measured as
the closest. Oops. Why is that? BLE is not meant to be used to measure distances, especially not in a complex (for radiowaves) environment. It gets more interesting, as they repeated the experiment after shuffling the participants, the values changed, compare Figs 3 and 4...
Now, in this setting, what would actually have been detected using the #SwissCovidApp? None with a threshold of 15 minutes, because while checking signal attenuation 3 times, the other phones were never within the threshold (Fig. 7a). This could be tweaked of course, but what ...
false positives? Especially in the open... The authors conclude: "That is, distance between handsets has only a weak, if any, correlation with detection rate". And remember, the participants had their phones in their hands during the whole experiment.
Now what to make of this data? The paper could, IMHO, be better, however, it's the best there is. However, the results do not look good for the efficacy of #SwissCovidApp.
Link to the data: https://arxiv.org/pdf/2006.08543.pdf
(thanks to @s_______m______ for sharing the link)
Link to the data: https://arxiv.org/pdf/2006.08543.pdf
(thanks to @s_______m______ for sharing the link)
