Regresion .... problem with transformation?

Dear Readers,

I have question on transformation in regression analysis. Both
independend and dependent (x=ln(y)) variables natural logarithmically
transformed. R-Sq-Adj >0.8, F and t- stat significant. Problem arises
when cross validation is done on regression equation with new data.
When the error (relative error=abs(actual-estimated)/actual) is
measured in logarithmically transformed data then is about 3-4% error
but when it's measured with raw actual and raw estimated data (exp(x))
error is 58-87%. Would you please tell me what i'm missing here?


Thank you in advance,
Jim

.

Relevant Pages

  • Re: CLT and regression
    ... Can I still use linear regression analysis? ... using a Box-Cox transformation to get them normal: ... positively skewed residuals, trying square root, cube root and log ... transformations often gets you close enough to normality. ...
    (sci.stat.consult)
  • Re: CLT and regression
    ... Can I still use linear regression analysis? ... To follow up Ray's post, you should fit the regression, and look at the residuals for normality. ... If they're not, e.g. if they're skewed, then you could look at using a Box-Cox transformation to get them normal: i.e. you use a power transformation if alpha=0 is indicated). ... Generally you don't have to be too precise with the transformation: for positively skewed residuals, trying square root, cube root and log transformations often gets you close enough to normality. ...
    (sci.stat.consult)