It depends on your goals. For example, if you have a wide dynamic range for the measured signal and want to show a particular shape (or trend) at low values then log scale is better. Obviously, if you would present such a signal on a linear scale all low-intensity details would be hidden. If you want to show the fit along with your signal and low-intensity values are not important (you have high noise or low dynamic range etc.), then you won't gain anything by going to a log scale. Hence, I'd use a linear scale in this case. These are some general aspects pertinent to signal analysis but every specific case may be different...
Thanks Serge. I am measuring the level of expression of ulbp2 between control and treatment groups. In linear scale, the mean fluorescent intensity of control is higher than the mean fluorescent intensity of treatment group. However, when i ran again with log scale, the mean fluorescent intensity of control is lower than the mean fluorescent intensity of treatment group. Why this can happen?
Shouldn't happen! Say, all your values >1 (whatever units you use). Then, if A>B then log(A)>log(B). If your values log(0.1). However, if you try to present it in the absolute scale by losing "-" you'll get erroneous results! Is it something you're doing?...