Efficacy of contact tracing apps (part II)

I realize I haven’t posted here something that has informed my thinking for a while.

This paper is very interesting in assessing the efficacy of contact tracing apps. It does so by properly assessing the counterfactual (i.e. no app deployed) and thus decomposing the proposed interventions with more granularity.

In particular it shows that most of the utility of such app is derived not from the notifications, but from the other interventions put in place alongside, with the main benefit actually derived from the “isolation-upon-symptoms” policy. This is important, as many of the recipients of an app notification might be warned indeed before contact tracers find them, but - had the app not been deployed - would have started having symptoms a tad later anyways. So if you analyze efficacy of the app, you have to be very careful in setting your counterfactuals.

Once those other measures are separated, the paper also quantifies the utility of the app “on top” of those other measures. Essentially it says you can only hope for a reduction on the order of 0.05 in Reff, short of a completely unrealistic installation base in the population. The paper also points out their result is at first glance very different from much more influential papers on the efficacy of contact tracing apps, based on predictive models. They then go on to explain, in Section 6.4, that this is actually the result of the counterfactual mistake explained above: the results in those other papers are being misunderstood, as what appears useful is not just the app but the app and the “isolation-upon-symptoms” policy.

In any case, this formulation as a reduction of R_eff is very useful, since it matches how this other paper formulates its own results. The latter tries to quantify the reduction of R_eff associated to non pharmaceutical interventions. For instance, it finds out that closing nightclubs contributes to a reduction of R_eff of around 0.15 in Switzerland (compared to fully open, post lockdown). One (very preliminary) interpretation of this result is that the risk of opening of nightclubs would not be compensated by the benefit of SwissCovid.


This research validates what seems like a reasonable public policy posture:

  • Keep the public informed of risks, both in terms of what is known about how the virus spreads and in terms of what geographic areas are currently experiencing high community transmission
  • Invest in public health infrastructure for contact tracing
  • Invest in testing
  • Provide support to enable the population to manage their own risk - including a MyData friendly exposure notification app. This includes economic support.
  • Enforce sanctions on individuals who ignore reasonable directives and put others at risk.

I think Switzerland has so far failed on five out of those five points. It has done some in each direction, but way too little.

FYI, today:

The Federal Council decided today they wanted to avoid lockdown. They might meet on Sunday to discuss it further.

I don’t think “invest” is precise enough. You need to be “efficient” (input -> output). Throwing money at inefficient systems, with improper governance and not the right mindset will only create big but stupid machine. Peoples that are dependent on the administration for social assistance already had this kind of experience. Now everybody is concerned.

Today if you are tested positive, the slow administration needs almost a week to send an SMS to your contacts. Nobody can argue that can work someway.

In that case, you either taking on the wrong side, or you structure is not adapted to the problem. In both cases you failed big time.

So now democratic governments are going to blame individual behavior or specific groups like they always do when they fail or they are going to speak about technical details. Perhaps they are also going to point the finger to a foreign country that “do worst” or use “authoritarian force”.

In that regard it’s very close to the environmental problem: the one having the capacity and responsibility to govern blame the one with very limited capacity and possible choice and no governance responsibility.