You may well have guessed then what the key problem with this is – of course, we only experience one of these worlds. The Youth Guarantee has/ will be introduced (putting aside any implementation issues), so we now live in the world with the Youth Guarantee. Unless a portal allowing us to access an alternative reality is invented soon, we hong kong rcs data will never get to see what the outcomes of the young people that go on to participate in the Youth Guarantee would have been had it not been introduced. And therefore, we can’t see what their counterfactual outcomes would have been, and therefore what the causal impact of the policy being introduced is.
This is then where the family of impact evaluation techniques come in – they seek to identify a suitable counterfactual in our Youth Guarantee world, in order to estimate what causal impact it has had to doing this:
Introduce an experiment where you randomly assign individuals to either experience the intervention or not.
Identify a comparable group that does not experience the intervention, for example because they are ineligible or choose not to.
In order for the counterfactual produced by these two approaches to be of a high quality, the group used to compare outcomes to the treatment group (known as the control or comparison group) needs to have certain characteristics.