How do you choose among different data transformations(-1,0,0.5,1)?

[Jun Li]

Hi Tjalling and DebTox community,

I am using your BYOM, byom_tk_parent code to fit uptake/elimination rates (kin/kout) for earthworms based on a simple first-order kinetic model (dCi = kin* Cs - kout * Ci;). I forgot which type of data transformation I should choose when I input the earthworm concentration into Matlab (0 log-transform or 1 no transformation). I tried both ways and got different results for kin and kout. Which do you believe is the most appropriate and how to decide in other cases?

Looking forward to hearing from you.

Jun Li

[Tjalling]

Whether or not to transform depends on the (expected) distribution of the residuals around the mean. If you have replicated data, you may get an idea from the data. With body-residue measurements, I think it is safe to assume that the residuals will increase with the mean measured value. In general, you may expect a rather constant relative error (say 10%, irrespective of whether you have a high or low measured value), and that would scream for log-transformation. On the other hand, low values for body residues may have a larger error due to analytical issues, and log-transformation would focus too much on getting a good fit on these low values. Therefore, I would tend to suggest square-root transformation as something in between (so 0.5 in the top-left corner of the data matrix). However, your mileage may vary ...

[Jun Li]

Thank you for your reply, Tjalling. That is helpful.