Some common errors arise again and again
Here are seven to watch out for:
unclear basepoint for graphs
- A TV advert used to proudly proclaim:
- "X has 25% more active ingredient"
The screen however showed just the top of four test tubes. The words
may have been true, but it looked a lot more than 25% - a truthful advert?
%increases and changes in %difference
If 300 is 50% bigger than 200, is 200 50% less than 300?
In 1990 product A had 10% of market share and now it has 15%. Is that
a 50% increase or a 5% increase? Of course, the market may be only half
as big now, so there may be less of product A sold.
Lesson - be very careful with your language.
Interpolation - estimating unknown values between known values
- OK, but extrapolating - going outside the known limits is dangerous
(albeit sometimes necessary!).
Real, non-random effects may nevertheless be very small
Assuming that a non-significant result means no difference is like
Kate Winslett assuming she weighs nothing because there was no detectable
change in the waterline of the Titanic when she jumped off.
non-significant no effect
Big effects may not be significant if sample size is low or
There may be a common cause, or it may simply be a fluke!
don't do too many tests!!
5% significant means will happen by chance 1 time in 20. If
you do lots of tests, 1 in 20 will (on average) be 5% significant. So,
if you need to do lots of tests look for better significance values.