Computing curved trendlines
by Arthur Merrill
Data sometimes shows a tendency to level off with a downward curve. A curved trendline, in these
cases, would be more useful than a straight line.
Quite often, the data rises more and more rapidly, like a skyrocket. If this is the case, a curved trendline
would be more useful than a straight line.
The upward curving tendency may be caused by the distortion of the arithmetic scale, which makes a
given percent increase appear large at the top of a chart and small at the bottom. This distortion can be
cured by using a logarithmic scale and then a straight line might be appropriately used. This "fitting" can
also be done by adding another column to the spreadsheet introduced in the previous issue and computing
the logarithms of the data. The straight line is then fitted to the plotted logarithms.
Curved trendlines can be drawn freehand or fitted approximately with the aid of a french curve. They also
can be fitted mathematically using the least squares method which was described in my column last
month. A popular curve for this purpose is the second degree parabola or second degree potential series
with the formula: