Pogue Mahone
Closet Gooner.
Add 1 SD error bars to that data and there is virtually 100% overlap. With SD error bars a significant gap (usually about half the size of the bars themselves), and not just no overlap between bars, will be present when there is a significant difference. In those cases the SD error bars are almost 100% overlapped for each data point so the chances of there being a significant difference is close to nil.
Jesus Wibs, you’re like a dog with a bone! You’re also wrong. You don’t know what size the SD bars would be on this graph. It’s a massive sample size, so they could be very small. Without doing the maths (which neither you nor I have the time, inclination or information to do) it makes everything else you say about these data redundant.
Statistical nit-picking aside, the most obvious flaw in that graph is the absence of numbers from years prior to 2017. That seems to be a really obvious way in which the tweet might be misleading. And the hashtag shows there’s an agenda at play here.