Comment by Supermancho

> ignoring all the actual data points

Well that's your problem.

The line is the predicted, not actual. How would you derive that line from plot of noise?

>> I drew an exponential through my noise.

The issue is that there was supposed to be a curve according to his reading, but the actual had no measurable trend. It's possible that the data was measured on the wrong scale. If you zoom out, those noise plots become a line segment. Then again, the predictable line is on the same scale (and we're assuming that it's correct according to his reading or the best he could fit) so zooming out would probably be a different form of lying with statistics via overfitting.