If you try to connect the upper and lower data points, that outlier down at zero really throws things off. So, if there is one data point that falls far away from the others, like that one at zero, you can ignore it. In this case, the original blue line is a good estimate of the slope—it is flat.