logrithmic transformation

#1
Hi all,

Basically I have a positively skewed distribution and want to make a
transformation so it is normally distributed. Usually I would just
make a logarithmic transformation and this would be fine. However in
this instance some of the distribution I want to transform is
negative, and obviously I can't take the log of a negative value.

What I have tried so far is just adding a random constant to each
observation to force the distribution to be positive (I.e. shift it to
the right) and then make a logarithmic transformation. However im not
particularly happy doing this as when I come to use this as the
dependant variable in subsequent regression analysis I believe the
results will vary with my choice of the random constant I added to
force the distribution positive.

Does anyone know if the way i have approached this problem is just, or
have any further suggestions for what i could do?
 

TheEcologist

Global Moderator
#2
Hi all,

Basically I have a positively skewed distribution and want to make a

Does anyone know if the way i have approached this problem is just, or
have any further suggestions for what i could do?
I dont see any problem with your log transformation as long as you make sure that you add the same constant to each datapoint. Poeple often do log transformations like log(x + constant).

for example: if your minimum value is -100 just log transform all your datapoints with log(x + 101).

Also what question are you trying to answer? Do you really need your data to be normally distributed to answer your question? There are tonns of non-parametric analysis alternatives that dont require any normality assumptions.

good luck