Lecture 5 MLR - 2
Lecture 5 MLR - 2
Lecture 5 MLR - 2
1 / 32
Outline
2 / 32
Hypothesis Tests and Confidence Intervals for a Single Coefficient
(SW Section 7.1)
3 / 32
Example: The California class size data
1. Single Regressor:
2. Multiple Regressors:
4 / 32
Standard errors in multiple regression in R
5 / 32
Tests of Joint Hypotheses (SW Section 7.2)
H0 : β1 = 0 and β2 = 0 vs H1 : β1 ̸= 0 or β2 ̸= 0 or both
▶ In general, a joint hypothesis will involve q > 1 restrictions. In the
example above, q = 2, and the two restrictions are β1 = 0 and β2 = 0.
6 / 32
Why can’t we just test the coefficients one at a time?
7 / 32
The F -statistic
8 / 32
Fq,∞ distribution
9 / 32
F-test example, California class size data:
10 / 32
F-test example, California class size data:
Then, test H0 : β1 = β2 = 0:
11 / 32
Testing Single Restrictions on Multiple Coefficients (SW Section 7.3)
Yi = β0 + β1 Xi1 + β2 Xi2 + ui , i = 1, . . . , n
▶ Consider the null and alternative hypothesis,
H0 : β1 = β2 vs. H1 : β1 ̸= β2
▶ This null imposes a single restriction (q = 1) on multiple coefficients – it
is not a joint hypothesis with multiple restrictions, e.g.,
H0 : β1 = 0 and β2 = 0.
▶ There are two methods for testing single restrictions on multiple
coefficients: (1) rearrange the regression (2) perform the test directly
12 / 32
Method 1: Rearrange (“transform”) the regression
▶ We start from
Yi = β0 + β1 Xi1 + β2 Xi2 + ui
H0 : β1 = β2 vs. H1 : β1 ̸= β2
▶ Add and subtract β2 Xi1 ;
13 / 32
Method 2: Perform the test directly
▶ Again, we have
Yi = β0 + β1 Xi1 + β2 Xi2 + ui
H0 : β1 = β2 vs. H1 : β1 ̸= β2
▶ Example:
14 / 32
Confidence Sets for Multiple Coefficients (SW Section 7.4)
Yi = β0 + β1 Xi1 + · · · + βk Xik + ui , i = 1, . . . , n
▶ What is a joint confidence set for β1 and β2 ?
▶ A 95% joint confidence set is:
▶ A set-valued function of the data that contains the true coefficient(s) in 95%
of hypothetical repeated samples.
▶ Equivalently, the set of coefficient values that cannot be rejected at the 5%
significance level.
▶ You can find a 95% confidence set as the set of (β1 , β2 ) that cannot be
rejected at the 5% level using an F-test (why not just combine the two
95% confidence intervals?).
15 / 32
Joint confidence sets (continued)
16 / 32
Confidence set based on inverting the F-statistic
17 / 32
Regression Specification: variables of interest, control variables, and
conditional mean independence (SW Section 7.5)
18 / 32
Regression Specification: control variables
19 / 32
Control variables: an example from the California test score data
20 / 32
Control variables example (continued)
21 / 32
Control variables (continued)
22 / 32
Control variables (continued)
23 / 32
Control Variables: Conditional mean independence
Yi = β0 + β1 Xi + β2 Wi + ui
24 / 32
Implications for variable selection and “model specification”
25 / 32
Digression about measures of fit...
2
It is easy to fall into the trap of maximizing the R 2 and R , but this loses sight
of our real objective, e.g., an unbiased estimator of the class size effect.
2
▶ A high R 2 (or R ) means that the regressors explain the variation in Y.
2
▶ A high R 2 (or R ) does NOT mean any of the followings;
▶ you have eliminated omitted variable bias.
▶ you have an unbiased estimator of a causal effect (β1 ).
▶ the included variables are statistically significant – this must be determined
using hypotheses tests.
26 / 32
Analysis of the Test Score Data Set (SW Section 7.6)
27 / 32
Analysis of the Test Score Data Set, continued
28 / 32
Test scores and California socioeconomic data ...
29 / 32
Digression on presentation of regression results
30 / 32
A Table to summarize estimation results
31 / 32
Summary: Multiple Regression
32 / 32