Do I need to treat all variables in the same way when I transform values?


I want to carry out a Pearsons correlation to look relationships between methane flux from soil under different plantation treatments (i.e spruce, pine etc) and environmental conditions (i.e. level of water table, temperature etc).

I am comparing 50 weeks worth of measurments (one flux measurement for each week and mean temp (for example) for each week).

The flux values for some treatments are not normally distributed, I can transform them to log values to achieve a normal distribution however do I then need to use the log of the flux values from every treatment data set and log the enviromental variable values too? i.e. if I use log values for one variable do all other variable values need to be log too?

Hope that makes sense! Any help would be great I am going round in circles here trying to work out if it matters or not! Sorry for asking such a stupid question! But I can't seem to find a straightforward answer and I keep confusing myself!

It sounds like you should look at regression rather than correlation, since it sounds like you have multiple independent variables. A correlation doesn't give you a quantitative relationship between a dependent and independent variable(s).

You need to transform all of the dependent variable values the same way. If a transformation does not normalize them at all of the values of the independent variables, you need another transformation.


TS Contributor
I disagree - you do not need to transform all of the variables. However, be very careful about the practical interpretation of the results.....
I did not say you needed to transform all of the variables. I said you needed to transform all of the values of the variable. That is what I thought the question was. Sorry if I misunderstood. Obviously, dependent variables that are already normally distributed would need no transformation.