If for example we have a multiple linear regression as follows: $$hydrcarb=x_1+x_2tanktemp+x_3disptemp+x_4tankpres+x_5disppres+x_6tankpres^2+x_7dispres^2$$
And I am trying to do a backward elimination, I was told that in determining which variable is irrelevant/insignificant, we choose those variables which P(>|t|) is large. My question is what value is considered large? Is 0.02 large?
Also, I am supposed to transform the response variable, $hydrcarb$ to $log(hydrcarb)$, then do we log-transform the explanatory variables as well? Because if we don't, I realized that when I eliminate variables, the variables that I omit are different, which I think means that the regression model is different.
[The actual question is: Fit a multiple linear regression model with log(hydrcarb) as the response and all six explanatory variables.]