The detection of significance is affected by three factors: the strength of any true
trend, the number of data points examined and the background variability in the data.
We (Sparks & Menzel, 2002) have recommended 20 years as an appropriate length
of series to detect effects, and Sparks and Tryjanowski (2005) have given examples
of the problems of start year, end year and series length on the conclusions that may
be drawn. The background variability appears to be greater in early season species
than in later season ones (Figure 4.10), whichwould imply that trend detectionwould
be harder in early species. Figure 4.11 demonstrates this change in the flowering
date of daffodil. There is a clear change in flowering date, which may be more of
a step change than the straight-line fit suggests, but linear regression is still one of
the helpful tools in detecting significant change.