Errors in dependent variable
Suppose that you have two rulers, one graduated in inches and the other in cm. You use the latter one to assess how many cm go in 1, 2, 3 .... 10 inch. Reading errors are inevitable and show up in the graph because the data are not on a straight line anymore, some points being too high and others too low. It is likely that on average the error is nil, i.e. that points that are 1 mm too high balance points that are 1 mm too low. The straight line fitted by eye still gives a fair estimate of the relationship between cm and inch, and the slope is likely to be 2.54, or come very close to this.