Testing null hypotheses on regression

44 Views Asked by At

I have the following regression equation to predict house prices: $PRICE_i=\beta_1+\beta_2LIV_i+\beta_3LOT_i+\beta_4ROOMS_i+\beta_5BATHS_i+\beta_6YEAR_i+u_i$

I need to test the 5 different hypotheses that each individual explanatory variable is statistically significantly different from zero. i.e. $H_0: _ = 0$ where $ = 2, 3, 4, 5, 6$.

To do this, can I use the following equation $\frac{b\ -\ \beta}{SE_b}$ where b is the coefficient of the variable and $\beta$ in this case will be $0$. I have all of the coefficients and standard errors. I will use this equation to see whether the result lies within the acceptance/rejection region based on the t-test critical value.

My question is does the critical value remain the same for all of these tests? or do I need to find a new critical value for each test based on the degrees of freedom?

1

There are 1 best solutions below

0
On

One standard way to do this is to run reduced models. For example, if you want to test whether your $\beta_2=0,$ you would run the reduced model $$PRICE_i=\beta_1+\beta_3LOT_i+\beta_4ROOMS_i+\beta_5BATHS_i+\beta_6YEAR_i+u_i.$$ You get your standard squared error from the complete model, $SSE_C,$ and compare it with the (usually larger) $SSE_R$ from the reduced model, and then use an $F$ test for the final results as follows: $$F=\frac{(SSE_R-SSE_C)/(\#\beta_C - \#\beta_R)}{SSE_C/(\#\text{data points} - \#\beta_C)}.$$ The $F$ distribution has $\#\beta_C - \#\beta_R$ numerator d.o.f., and $\#\text{data points} - \#\beta_C$ denominator d.o.f.