I'd personally really like to know what the metric would be if (hypothetically, just for analysis) the vaccine was a placebo. For example, if the people who get the vaccine are generally more able to work from home, then the metric would still be significantly above zero.
Just for clarity, I believe that the vaccine is effective, but I question the measurement methodology.
Many other confounders cannot easily be controlled for, but is there any reason to assume that they're correlated with the outcome? It could be that vaccinated people are more careful in terms of meeting other people.
At any rate, we had big RCTs at the beginning that demonstrated efficacy against the then prevailing variant (alpha, I assume).
It'll be tricky to repeat an RCT at this stage.
In some countries there's a formal labor law for doctors notes after X days that is known by all and which might mean your doctor checks and registers the case(?) in other places it is total discretion if a manager demands a note but rarely does in places that aren't shift work oriented.
Given two people who WFH the vaccinated one can conveniently eat in a restaurant and go to the gym while the unvaccinated needs to get take out or is getting regular tests. I think vaccinated people have much more interaction with the virus unmasked in an indoor environment but much less likelihood of ever being tested when having light or no symptoms.
In some countries (the US is an example); there's a social class aspect to vaccination; highly educated people are more likely to be vaccinated, and also more likely to work from home.
However, if this was a big factor you'd expect vaccines to appear dramatically less effective in near-universal-vaccination countries, like Portugal. That doesn't seem to be the case.
To overcome this, you'd need an RCT. But you're unlikely to find either a) unvaccinated people that want to be vaccinated, yet would be ok to get a placebo instead, or b) unvaccinated people that don't want to be vaccinated yet would be ok to get a vaccination instead.