Expected coefficient of determination of a global linear regression

61 Views Asked by At

Suppose you have some ordered data, for instance average income for some people arranged by age. Suppose that you divide your sample in two subsets, say $P_1=$ people within 50 years old and $P_2=$ people above 50 years old.

If you perform two linear regressions on $P_1$ and $P_2$ you get the coefficients of determination $(R_1)^2$ and $(R_2)^2$, the regression coefficients $\beta_1$ and $\beta_2$ and the the intercept terms $c_1$ and $c_2$. Set also $|P_i|=N_i$ for $i=1,2$.

How to compute, as a function of the eight numbers $R_i$, $\beta_i$, $c_i$, $N_i$ ($i=1,2$), the expected value of the determination coefficient and the regression coefficient of the whole sample $P_1\cup P_2$ when performing a global linear regression?