I don't know the source of the data, but there are only two possibilities:
1) It is self-reported, which is more likely. In that case, the percentage of people being able to hold a serious discussion will be lower than just for "conversation". How much is hard to say, since the definition of these things varies widely and the self-assessment of people is also questionable.
2) It is based on testing. But then the usefulness of the data is similar to (1), since we don't have the definitions the test used.