Die Wissenschaft findet nie ein Ende. Wie geht es weiter mit Evaluationsstudien? Uns wurde häufig die Frage gestellt, ob denn die Effekte von Masken von anderen Maßnahmen unterschieden werden können. Deswegen haben wir „Figure S2“ im Anhang zu unserer PNAS Veröffentlichung erstellt. Die Frage ist, wie trennen wir Effekte vorheriger Maßnahmen von den Maskeneffekten? Und: zahlen sich die vorherigen Maßnahmen auch aus? „Do earlier measures pay off and where is the effect?” Hier ist meine Antwort.
As a theorist, I would like to draw your attention to figure 6, our SIR model. And I would offer an extended version of figure 6 (not included in the paper)
This extended version shows the green dashed curve. It represents the number of infectious individuals before any intervention. Then we have some earlier measures and their effect is shown by the vertical drop in the figure. Hence, the earlier measures put the number of infectious individuals on the red curve. Then the effect of masks is visible as of T+Dm = 40. As long as earlier measures are sufficiently far away in time from the introduction of masks, we measure the effects of masks.
To come back to your point, yes, earlier measures do pay off. They put infections from the green dashed onto the red curve. Earlier measures still have an effect. Our findings measures the effect of masks on top of the effects of earlier measures. So I really believe that there is no temporal confounding in the Jena setup.
Having said this, three further aspects: (i) We do not work with a regional SIR model. We should work with a SIR model that studies the pandemic in many regions. We should allow for measures being implemented in all of these regions (especially those that turn out to be in our synthetic control group) at different point in times. We should then study the effect of masks in this framework. Again, as long as masks became mandatory sufficiently far away from other measures (and this is true in Jean and Thuringia), this extension should lead to the same findings.
(ii) We assume effects of an intervention are visible 10.5 days after implementation (appendix A.3, table S3). Of course, the effect of an intervention follows a distribution (figure S3 and S4). This could also be modelled explicitly. We do not believe that this would lead to other results. But, yes, the proof is in the pudding. It should be done.
(iii) We (scientists) can only analyse what society offers us as data. We have stressed long ago (in German, and others even longer ago in other contexts) that we would know much more if we had a clearer study design. Let us organize experiments. But society (at least in Germany) is not willing to do this. Do not experiment with health is the standard criticism. Unfortunately, policy makers and the general public does not easily accept the idea that we experiment with health anyway all of the time. Then we could experiment such that we maximize our understanding of infection channels of CoV-2.