1 / 20

Multiple Regression

Multiple Regression. Lecture 13. Today’s plan. Moving from the bi-variate to the multivariate Looking at how the multivariate equation relates to the bi-variate equation Derivation The difference between true and estimated models. Introduction.

lanza
Download Presentation

Multiple Regression

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Multiple Regression Lecture 13

  2. Today’s plan • Moving from the bi-variate to the multivariate • Looking at how the multivariate equation relates to the bi-variate equation • Derivation • The difference between true and estimated models

  3. Introduction • In multivariate regressions, your number of X variables are restricted by (n-k) > 0, where k is the number of parameters in your model • In the bi-variate case we had If (n-k)  0 we wouldn’t be able to calculate test statistics • We will use an example where earnings is our dependent variable, years of schooling (YRS_SCH) is X1, and age is X2

  4. Derivation • The rules for the derivation of the parameters are the same as for the bi-variate world • Our g function will be g(a, b1, b2) • We will still want to minimize  e2 • Our model will be Y = a + b1X1 + b2X2 +e • We can rewrite this in terms from deviations from mean values (coded variables): y = b1x1 + b2x2 + e

  5. Derivation (2) • We can rearrange our model in terms of e: e = Y - a - b1X1 - b2X2 • Differentiating with respect to each of the parameters gives us:

  6. Derivation (3) • To get our estimate of we use the FOC that the sum of the errors equal zero. We substitute in for e and solve: • As we include more variables, we need more terms to calculate the intercept • Calculating is more complicated

  7. Derivation (4) • We have the first order conditions for • The multivariate case is much more complicated than the bi-variate case, but the pattern remains the same • Denominator still considers the variation in X • The numerator still considers the variation of X1, X2, and Y

  8. Derivation (6) • The multivariate case is much more complicated than the bi-variate case, but the pattern remains the same • Denominator still considers the variation in X • The numerator still considers the variation of X1, X2, and Y

  9. Matrix of products & cross-products • This will help us calculate b1 and b2, as well as other test statistics we’ll need • The matrix of products and cross-products is symmetric

  10. Example • On L12.xls there is an example of a matrix of products and cross-products that we’re interested in. • This spreadsheet also shows that LINEST can also accommodate a multivariate regression • From the spreadsheet we know:

  11. Example (2) • We can then calculate: • We can also calculate

  12. Example (3) • So now we can ask: What was the effect of including age? • Had we not included age, our bi-variate regression equation would be: Y = 4.53 + 0.097 X1,where X1 is years of schooling • Including age, the multivariate regression equation is: Y = 4.135 + 0.057 X1 + 0.023 X2 • By including age, we reduce the coefficient on education (X1) by nearly a half!

  13. True & estimated models • A true model can come from: 1) Economic theory • an example of this is the Cobb-Douglas production function Y=ALK • the form is provided by economic theory • we want to test if  +  = 1 2) Ad-hoc variable inclusion • The justification for the variables comes from economic theory, but we include variables on the basis of significance in statistical tests • An example: the Phillips Curve

  14. Omitted Variable Bias • Let’s go back to the returns to education example in L12.xls and examine Omitted Variable Bias: • Let’s assume that the true model is: Y = a + b1X1 + b2X2 +e • But what if we instead estimate the following model: Y = a + b1X1 +u where X1 is still years of education

  15. True & estimated models (3) • Reasons why we might not estimate the true model • we might not be able to collect the necessary data • we might simply forget to include other variables such as age in the regression • Let’s rewrite our equations in terms of deviations from the mean: True model: y = b1x1 + b2x2 + e Estimated model: y = b1x1 + u

  16. Omitted variable bias • Our estimate of the slope coefficient for the bi-variate model will be: • If we know the true model we can plug it into the above expression and take the expectation to get:

  17. This represents the omitted variable bias Omitted variable bias (2) • We can multiply out the terms and simplify the expression: • Recall that one of our CLR assumptions is E(x1 e) = 0, so

  18. This leads to a biased estimate of Omitted variable bias example • Returning to the L13.xls example, we have • If we think that then:

  19. Recap / what’s to come • We learned that deriving the multivariate regression equation is similar to deriving the bi-variate case • We worked with a matrix of products and cross-products • We looked at the difference between true and estimated regression models • We learned to calculate the omitted variable bias • In the next few lectures we’ll be doing some more with multivariate models and applications

  20. Unnecessary Variables • What happens if variables that are included in the estimated model, are not relevant under the ‘true’ model. Estimated model: y = b1x1 + b2x2 + e True model: y = b1x1 + u • If variables are unnecessary, they will not count in the estimated model. • How to detect that: t-ratio hypothesis tests/Joint hypothesis tests using the F-distribution. • Helps to make models parsimonious.

More Related