The linear trendlines are computed using the method of least squares. By experiment I can see that very large values at the ends of the data can substantially, perhaps even, unnaturally affect the slope. Is there any rigorous mathematical discussion of this problem?
An examination of the alternatives would be a statistics course... Least squares is the easiest/entry level analysis.
ReplyDeleteIt you really need to do analysis, then you need something more substantial. I can't help you because the only programs I've used have been programing languages (MATLAB) and enterprise level systems like SAS. There is probably a freeware/open-source solution, but I have no idea what to recommend...