3.3 Trend classification

We compare the 95% credible intervals with a reference, upper and lower threshold to classify the strength of the effect into 10 classes. The change of a linear trend is converted into a change over the length of the data. The change of an index is the actual change between the two years. The reference is set to 0 (no change). The credible interval of a significant effect does not contain 0. We selected a change of -25% (3/4 of the initial value) as the lower threshold. We use the complement1 of that (+33% or 4/3 of the initial value) as the upper threshold. A -25% or +33% change over 5 years is equivalent to an average yearly change of -5.6% or +5.9% in case of a linear trend.

Below are the symbols, interpretations and rules for each of the 10 classes.

  • ++ strong increase: A significant positive trend and significantly stronger than the upper threshold.
  • +~ moderate increase: A significant positive trend and significantly weaker than the upper threshold.
  • + increase: A significant positive trend, not significantly different from the upper threshold.
  • ~ stable: A non-significant trend and significantly between the lower and upper threshold.
  • - increase: A significant negative trend, not significantly different from the lower threshold.
  • -~ moderate decrease: A significant negative trend and significantly weaker than the lower threshold.
  • -- strong decrease: A significant negative trend and significantly stronger than the lower threshold.
  • ?+ potential increase: A non-significant trend, significantly above the lower threshold.
  • ?- potential decrease: A non-significant trend, significantly above the upper threshold.
  • ? unknown: A non-significant trend, both the lower and upper threshold are probable.

One of the benefits is that we distinguish ~ (stable) and ? (unknown). Both are non-significant. The main difference between both cases is the uncertainty. We set the thresholds at important changes. If the uncertainty is large, then the credible interval contains both the lower and the upper threshold. So we have no clue what is happening, hence the unknown class. If the uncertainty is small, then the credible interval contains neither the lower nor the upper threshold. In this case we do known that the change is less extreme that the thresholds. So if there is a change, it will be smaller than important changes (the thresholds). This is informative, even though the change is not significant.


  1. \(\log(3/4) = -0.2877\) and \(\log(4/3) = 0.2877\)