Yeah, that thought struck me too, looking at the entire graph there can never be an absolute decrease. But then, which timescale is the salient one?

In context it would be one that would show that marijuana legalisation was probably the cause.

Indeed, it's probably not due to marijuana legalisation, unless the resulting decline had a four year lag period for some reason, which I won't attempt to argue.

I’d be more surprised if the effect was immediate. Addiction takes years to develop, and years to recover from. I would also expect marijuana legalization to have more impact on (potential) future drug users than people already addicted to opioids.