Hello,
To correct for a negative skew, I performed a reverse log transformation to my response variable, as such: log10(K + 1 - X), where K is the highest value of the variable X. If I understand correctly, to interpret the coefficients in my linear model, I now have to exponentiate them as such: 10^coefficient and... how do I correct for reversing the scores?
In a model log(y) = mlog(x) + b + err, I can interpret the coefficients in log-space or I can back transform and discuss y = emx + eb + eerr. It might be easiest to start by making a graph of the data. An audience that encounters and uses log transformed data frequently will probably prefer using the log transformed values. Other audiences might need more help.
Are you really getting a better model using log(K+1+X) versus a more simple log(X) or log(X+C) where C is some constant carefully chosen to avoid taking a log of values
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation