Survation has a bit of a problem with not having enough 16-24s on their panel, and this needs hugely upweighted, it leads to volatility in their polling (see the recent 7 point siwng to No that no one else saw and subsequently corrected) - if it's true, and I'm not saying it is, to see a huge drop in 16-24s but with the headline staying the same would possibly mean a drift to Yes in other groups, allowing for the uncertainty in that age group due to it's small sub sample size (and therefore greater uncertainty of it's representative nature). Also, every other pollster doesn't show 16-24 as being that No friendly.
Might be all bullshit mind, just going on comments I've seen elsewhere, not seen the datasets.