Honestly with that small of a sample size those numbers could mean anything.

Wed need to know things like:

• How many hours worth of new game content was released (maybe there were only enough new games to make up 20% of playtime)
• How mich do Games as a Service impact things (they get new content, but aren't "new" games)
• Are people playing games more or less (people could be playing new games the same amount, but increasing the amount of hours playing old games)

A small sample of averages is really hard to interpret.

That's just how the article phrased it, surprisingly it's hard to find historical "Steam Replay" stats so it's hard to confirm/expand on any of this.