1

I have several measurements of trees over time (different vegetation indexes).

Each measurement is in a different scale, some also include negative values (e.g. NDVI).

I need to compare the variation in time of all measurements, using a single statistical measure. Coefficient of variation is not relevant because of negative values. Can anyone recommend an approach?

Joel Galang
  • 148
Gilad Weil
  • 11
  • difficult. Can you give us more information about the different measures? – Ben Bolker Dec 21 '16 at 16:20
  • Thanks. the variables are based on various band(color) relations of images. See https://en.wikipedia.org/wiki/Vegetation_Index. The scale of the different variables differs; 0 to 1, -1 to 1 (as in NDVI), X to -X. Can skew help? maybe STD/range? – Gilad Weil Dec 21 '16 at 16:52
  • Even though I don't know why you were thinking of coefficient of variation negative values does not exclude as |mean| can be used in the denominator. The problem comes about when mean=0. – Michael R. Chernick Dec 21 '16 at 23:56
  • @MichaelChernick is absolute value of mean commonly used for coefficient of variation? I haven't seen that elsewhere and was curious if it's acceptable since I have a similar problem to the OP – qdread Jun 07 '18 at 16:05
  • 1
    Why do you need to use CVs at all? Why not employ a statistic that will be applicable to these data, such as their variances? – whuber Feb 21 '22 at 16:40

1 Answers1

0

When x- mean negative , change the origin to a particular point So that all the x values are positive and this will not affect the variance ,so also standard deviation.Hence you can talk about cv with respect to changed origin.

A K P C Swain
  • 1
  • 5
    Unfortunately, this recipe is arbitrary: it changes the problem to that of choosing this origin (as well as figuring out how to interpret the values) -- and the resulting CV is extremely sensitive to this choice. How should one proceed? – whuber Feb 21 '22 at 16:39