The myth of the "Earthquake Line" in China

There is an article widely cross-posted around the internet about a 'mysterious earthquake line' running from the estuary of Tumen River to the SW of Yunnan bordering Myanmar. Alleged this line connects The 1975 Haicheng quake, 1976 Tangshan quake, 1966 Xintai quake, 1695 Linfen Quake and even this year's Yinxiu quake on May 12th. (Further claim by "readers" even includes those 1000 years: 1037 Kaifeng, 1303 Zhaocheng, 1555 Weizhou, 1654 Xian and others)

Here is the "earthquake line"

Pretty inpressive, it seems. I happen to have scanned a "full" (recorded) earthquake map for China back in May. and also an "earthquake belt map" in the same post. Amazed by this map, I dig it out and draw a line and this is what I get.

The red dots are quakes since 1900 (size represents strength) and green dots are major quakes recorded in history. I am not sure if I would call this a correlation line. Yes, a line passing half a dozen major quakes, ok, but there are many dozens of such quakes and I can draw a hundred of these lines from this map. 

• Note some quakes, e.g. Haicheng is more than 100km away from "the line", the first map seems to have been distorted and the dot blurred
• The line does not pass Henan at all, so I do not know how the 1037 quake was claimed to be on the line, even the 1654 Xian quakes are more than 200 km from "the line". The original poster appears to be a bit more rigorous than those "readers"
• (A technical note: we are drawing line on a 'flatten' map while the earth is actually a sphere, so there are many ways to draw it. The geodesic (great circle line) is probably the most natural way to draw so the 'line' is only an approximation (though a reasonably good one).

Earthquakes do form patterns, as they are correlated with plate techtonics and fault lines, as indicated in, e.g., this chart, or this. If there really is a line, it would be the one running NS from Yinchuan to Kunming, which is, in fact 2 fault lines approxinately lined up by coincidence.

This charts show how one can create false correlation by selectively removing the data points. This is exactly what those earthquake predictors have done, selecting the a few data points from the hundreds such that they fit the "proposed" formula.