Sadly these trends can be significantly wrong as easily as they can be right.
One illustration of this is provided by the datasets crafted by Francis Anscombe known as ‘Anscombe’s Quartet’ (see Graphs in Statistical Analysis by F.J.Anscombe. American Statistician. 27 (February1973): 17-21)
The summary statistics for the four datasets (A, B, C, D) are identical:
But plotting the datasets shows that the four are really completely different
Now add Excel trendlines:
It is clear from the illustrations that three out of the four trendlines are completely wrong
Putting this another way: our visual inspection of the patterns is far more reliable than Excel's automatic calculation. We should trust ourselves.
In his book 'Show Me the Numbers: Designing Tables and Graphs to Enlighten' (Second Edition. Analytics Press. 2012, pp.224-8), Stephen Few shows how the trend shown can be affected significantly by the small differences in range of values selected.
But plotting the datasets shows that the four are really completely different
Now add Excel trendlines:
It is clear from the illustrations that three out of the four trendlines are completely wrong
Putting this another way: our visual inspection of the patterns is far more reliable than Excel's automatic calculation. We should trust ourselves.
In his book 'Show Me the Numbers: Designing Tables and Graphs to Enlighten' (Second Edition. Analytics Press. 2012, pp.224-8), Stephen Few shows how the trend shown can be affected significantly by the small differences in range of values selected.
So there are two big things to watch out for - the range of data that you decide to trend and the way Excel then attempts to fit a trend line
No comments:
Post a Comment