I don't have access to the data set, but it seems like a case of "lying with statistics." They could have chosen many other intervals over which to make a linear best fit that would have shown a zero or positive slope. Go back one more year, and it looks like the line would be flat. Eliminate 2009 and the line would be flat or positive. Start at 1999, the line would be very positive, but start at 1998 and the line would be very negative. Basically, the use of the linear best fit on small portions of these data amounts to bullsht.

__________________
1987 W201 190D