Econ 488 – Applied Managerial Econometrics

Cameron Kaplan, Fall 2010

9/24/10

Lab 4 – Classical Assumptions and Multiple Regression

Use the lab4.gdt file, which is available on the course blackboard page to answer these questions.

1.  Using the ordinary least squares procedure in gretl, run the regression:

saleprici= β0 + β1sqfti +εi

  1. Report your results. In english, what is the meaning of your coefficient on sqft, β1? Pay attention to the units that the variables are measured in.
  2. What is the meaning of β0? In this model, is this a useful thing to know? Why or why not?

2.  Classical Assumptions. Most of the classical assumptions are not testable, but sometimes it is possible to see if the assumptions are likely to be true. Use the model from part 1 to answer these questions.

  1. Is this model linear in the parameters? (Assumption 1)
  2. Remember that the error term is a function of any variable that is not specified in the model. Name a few things that are correlated with salepric, that are not in the model.
  3. Are the variables you named in part b likely to be correlated with sqft? Why or why not? What assumption is violated if the answer is yes? Explain.
  4. Using gretl create a scatterplot of sqft (X-axis) and salepric (Y-axis). Based on your scatterplot do you think the homoskedasticity assumption is likely to be true in this model (Assumption 5). Explain.
  5. Create a new variable in gretl that is equal to 2*sqft, by selecting Add>Define New Variable… In the dialog box, type “sqft2 = 2*sqft”. Notice that there is a new variable in your variable window. Add sqft2 as an independent variable to your model (i.e. run saleprici= β0 + β1sqfti + β2sqft2i+εi). What happens? Why?

3.  Now we are going to modify to the model to add more variables. Run the following regression in gretl:

saleprici= β0 + β1sqfti + β2bedrmsi + β3bathsi + β4garagei + β5agei +εi

  1. Report your results. In english, what is the meaning of your coefficient on sqft, β1? Why is this different than your answer to question 1, part a?
  2. Does this model violate the “no perfect multicolinearity” assumption? (assumption 6).
  3. Imperfect multicolinearity (often just called multicolinearity) is when two or more of the variables in the regression are correlated with each other. Although it does not violate the classical assumptions, it can cause some problems, which we will discuss later. Does this model exhibit imperfect multicolineraity? Explain.