Thank you Mr. Ulrich, I recognized that lately(estimating b using model y=a+b*x). According to my understanding,in ancova we test model null hypothesis y=a+bx+e1 against alternative hypothesis y=a+b*x+z+e2. We simply compare the residuals of both models. So we need to adjust the dependent variable y using both estimated b's, since we don't know the which model is true. Then we have two groups of dependent variables. Thank you very much again Mr. Ulrich as all the time.
But, I have another problem also:) not related to that topic but related to ancova again. Let say, I want to make an anaylsis to compare income of working males and females. So my dependent variable is the income, independent grouping variable is the sex. I also want to partial out the effect of experience (worked years) from the dependent variable to test if the males and females have same experiance, will their income be the same? So experience year is the covariate. However, it is obvious that there is relation between covariate(year of experience) and sex. Females work less than males in general. So is using this covariate problem? Becouse it will also partial out some information from dependent variable related to sex. Is this a serious problem? What should I do? Thank you very much.