normality and data transformation
Posted: Thu Mar 04, 2010 5:41 pm
i'm doing research using data panel analyisis and i've read several literature that suggest test of normality in raw data as a must. but i've read in internet that test of normality in residual is a must, while normality of raw data is not necessary.
i do regression with one dependent variable and two independent variables, and all of them are not normally distributed.
my question is:
1. should i ignore that my data is not normally distributed, and focus on residual distribution? note that my first equation is y = c + bX1 + bx2, but the residual is not normal, so i've tried change it to log(y)=c+bx1+bx2, and the residual is normal. if i could ignore the normality of data, is there any literature that support it?
2. i already tried to transform the data, and so far the method that fits is johnson transformation on dependent variables,my questioin is, should i transform all variables with the same equation? because for independent the equation is different, or should i use diferent? and how does it affect on the regression model?
thanks
i do regression with one dependent variable and two independent variables, and all of them are not normally distributed.
my question is:
1. should i ignore that my data is not normally distributed, and focus on residual distribution? note that my first equation is y = c + bX1 + bx2, but the residual is not normal, so i've tried change it to log(y)=c+bx1+bx2, and the residual is normal. if i could ignore the normality of data, is there any literature that support it?
2. i already tried to transform the data, and so far the method that fits is johnson transformation on dependent variables,my questioin is, should i transform all variables with the same equation? because for independent the equation is different, or should i use diferent? and how does it affect on the regression model?
thanks