Chapter 2

Denition of Simple Regression Model Ordinary Least Squares Mechanics of OLS Units of Measurement and Functional Form Expected

Values and Variances of OLS Estimators Regression through the Origin

Chapter 2 The Simple Regression Model

Le Van Chon
University of Economics Ho Chi Minh City
June 2012 Based on Introductory Econometrics: A Modern Approach by Wooldridge
Le Van Chon
Applied Econometrics
Denition of Simple Regression Model Ordinary Least Squares Mechanics of OLS Units of Measurement and Functional Form Expected Values and Variances of OLS Estimators Regression through the Origin
Denition of Simple Regression Model

Applied econometric analysis often begins with 2 variables y and x . We want to study how y varies with changes in x . E.g., x is years of education, y is hourly wage. x is number of police ocers, y is a community crime rate. In the simple linear regression model: y = 0 + 1 x + u (1)
y is called the dependent variable, the explained variable, or the regressand. x is called the independent variable, the explanatory variable, or the regressor. u, called error term or disturbance, represents factors other than x that aect y . u stands for unobserved.
Le Van Chon Applied Econometrics
Denition of Simple Regression Model (cont.)

If the other factors in u are held xed, u = 0, then x has a linear eect on y : y = 1 x 1 is the slope parameter. This is of primary interest in applied economics. One-unit change in x has the same eect on y , regardless of the initial value of x . Unrealistic. E.g., wage-education example, we might want to allow for increasing returns.
Le Van Chon

An assumption: the average value of u in the population is zero. E (u ) = 0 (2) This assumption is not restrictive since we can always use 0 to normalize E (u ) to 0. Because u and x are random variables, we can dene conditional distribution of u given any value of x . Crucial assumption: average value of u does not depend on x. E (u |x ) = E (u ) (3) (2) + (3) the zero conditional mean assumption. This implies E (y |x ) = 0 + 1 x

Population regression function (PRF): E (y |x ) is a linear function of x . For any value of x , the distribution of y is centered about E (y |x ).
Le Van Chon
Ordinary Least Squares
How to estimate population parameters 0 and 1 from a sample? Let {(xi , yi ) : i = 1, 2, ..., n} denote a random sample of size n from the population. For each observation in this sample, it will be the case that yi = 0 + 1 xi + ui
Le Van Chon
Ordinary Least Squares (cont.)

PRF, sample data points and the associated error terms:
Le Van Chon

To derive the OLS estimates, we need to realize that our main assumption of E (u |x ) = E (u ) = 0 also implies that Cov (x , u ) = E (xu ) = 0 (4) Why? Cov (x , u ) = E (xu ) E (x )E (u ) = Ex [E (xu |x )] = Ex [xE (u |x )] = 0. We can write 2 restrictions (2) and (4) in terms of x , y , 0 and 1 E (y 0 1 x ) = 0 E [x (y 0 1 x )] = 0 (5) (6)
(5) and (6) are 2 moment restrictions with 2 unknown parameters. They can be used to obtain good estimators of 0 and 1 .

Method of moments approach to estimation implies imposing the population moment restrictions on the sample moments. 0 and 1 to solve the Given a sample, we choose estimates sample versions: 1 n 1 n
n n
0 1 xi ) = 0 (yi
i =1
(7) (8)
0 1 xi ) = 0 xi (yi
i =1
Given the properties of summation, (7) can be rewritten as 0 + 1 x y = (9) 0 = y 1 x or (10)


Drop 1/n in (8) and plug (10) into (8): n 1 x 1 xi ) = 0 y ] i =1 xi (yi [ n 1 n xi (xi x ) = ) i =1 i =1 xi (yi y n n 1 )(yi y ) = )2 i =1 (xi x i =1 (xi x Provided that
n
(xi x )2 > 0
i =1
(11)
the estimated slope is 1 =

n )(yi i =1 (xi x n )2 i =1 (xi x
y )
(12)
Le Van Chon
Summary of OLS slope estimate: - Slope estimate is the sample covariance between x and y divided by the sample variance of x . - If x and y are positively correlated, the slope will be positive. - If x and y are negatively correlated, the slope will be negative. -Only need x to vary in the sample. 0 and 1 given in (10) and (12) are called the ordinary least squares (OLS) estimates of 0 and 1 .
Le Van Chon

0 and 1 , dene a tted value To justify this name, for any for y given x = xi : 0 + 1 xi y i = (13) The residual for observation i is the dierence between the actual yi and its tted value: 0 1 xi u i = yi y i = yi Intuitively, OLS is tting a line through the sample points such that the sum of squared residuals is as small as possible term ordinary least squares. Formal minimization problem:
n 0 , 1 i =1 n
min
u i2
=
i =1
0 1 xi )2 (yi
(14)
Le Van Chon

Sample regression line, sample data points and residuals:
Le Van Chon

To solve (14), we obtain 2 rst order conditions, which are the same as (7) and (8), multiplied by n. 0 and 1 , we have the Once we have determined the OLS OLS regression line: 0 + 1 xi y i = (15) is also called the sample regression function (SRF) because it is the estimated version of the population regression function (PRF) E (y |x ) = 0 + 1 x . Remember that PRF is xed but unknown. Dierent samples generate dierent SRFs.
(15)

1 is of primary interest. It tells us the amount Slope estimate by which y changes when x increases by 1 unit. 1 x y = E.g., we study the relationship between rm performance and CEO compensation. salary = 0 + 1 roe + u salary = CEOs annual salary in thousands of dollars, roe = average return (%) on the rms equity for previous 3 years. Because a higher roe is good for the rm, we think 1 > 0. CEOSAL1 contains information on 209 CEOs in 1990. OLS regression line: salary = 963.191 + 18.501roe
(16)

E.g., for the population of the workforce in 1976, let y = wage , $ per hour, x = educ , years of schooling. Using data in WAGE1 with 526 observations, we obtain the OLS regression line: wage = 0.90 + 0.54educ Implication of the Only 18 people in education. the levels. Implication of the intercept? Why? the sample have less than 8 years of regression line does poorly at very low slope?
(17)
Fitted Values and Residuals
0 and 1 , we can obtain the tted value y Given i for each observation. Each y i is on the OLS regression line. OLS residual associated with observation i , u i , is the dierence between yi and its tted value. If u i is positive, the line underpredicts yi . If u i is negative, the line overpredicts yi . In most cases, every u i = 0, none of the data points lie on the OLS line.
Le Van Chon
Algebraic Properties of OLS Statistics

(1) The sum and thus the sample average of the OLS residuals is zero.
n
u i = 0
i =1
and thus
1 n
u i = 0
i =1
(2) The sample covariance between the regressors and the OLS residuals is zero.
n
xi u i = 0
i =1
(3) The OLS regression line always goes through the mean of the sample. 0 + 1 x y =
Algebraic Properties of OLS Statistics (cont.)

We can think of each observation i as being made up of an explained part and an unexplained part yi = y i + u i We dene the following:
n
(yi y )2 is the total sum of squares (SST),

i =1 n
(y i y )2 is the explained sum of squares (SSE),

i =1 n
u i 2 is the residual sum of squares (SSR).

i =1
Then SST = SSE + SSR

(18)
Algebraic Properties of OLS Statistics (cont.)

Proof:
n n n
(yi y )
i =1
=
i =1 n
[(yi y i ) + (y i y )] =
n i =1 n
[u i + (y i y )]2 (y i y )2
i =1
=
i =1
u i + 2
i =1 n
u i (y i y ) + u i (y i y ) + SSE
= SSR + 2
i =1
and we know that

n
u i (y i y ) = 0
i =1
Goodness-of-Fit
How well the OLS regression line ts the data? Divide (18) by SST to get: 1= SSE SSR + SST SST
The R-squared of the regression or the coecient of determination SSR SSE =1 (19) R2 SST SST It implies the fraction of the sample variation in y that is explained by the model. 0 R2 1
Goodness-of-Fit (cont.)
E.g., CEOSAL1. roe explains only about 1.3% of the variation in salaries for this sample. 98.7% of the salary variations for these CEOs is left unexplained! Notice that a seemingly low R 2 does not mean that an OLS regression equation is useless. It is still possible that (16) is a good estimate of the ceteris paribus relationship between salary and roe .
Le Van Chon
Units of Measurement on OLS Statistics

OLS estimates change when the units of measurement of the dependent and independent variables change. E.g., CEOSAL1. Rather than measuring salary in $000, we measure it in $, salardol = 1,000salary . Without regression, we know that salardol = 963, 191 + 18, 501roe . (20)
Multiply the intercept and the slope in (16) by 1,000 (16) and (20) have the same interpretations. Dene roedec = roe /100 where roedec is a decimal. salary = 963.191 + 1850.1roedec .
(21)
Units of Measurement on OLS Statistics (cont.)
What happens to R 2 when units of measurement change? Nothing.
Le Van Chon
Nonlinearities in Simple Regression
It is rather easy to incorporate many nonlinearities into simple regression analysis by appropriately dening y and x . 1 of 0.54 means that each additional year of E.g., WAGE1. education increases wage by 54 cents. maybe not reasonable. Suppose that the percentage increase in wage is the same given one more year of education. (17) does not imply a constant percentage increase.
Le Van Chon
Nonlinearities in Simple Regression (cont.)

New model: log (wage ) = 0 + 1 educ + u where log(.) denotes the natural logarithm. For each additional year of education, the percentage change in wage is the same. the change in wage increases. (22) implies an increasing return to education. Estimating this model and the mechanics of OLS are the same: log (wage ) = 0.584 + 0.083educ (23) wage increases by 8.3 percent for every additional year of educ .
(22)
Nonlinearities in Simple Regression (cont.)

Another important use of the natural log is in obtaining a constant elasticity model. E.g., CEOSAL1. We can estimate a constant elasticity model relating CEO salary ($000) to rm sales ($mil): log (salary ) = 0 + 1 log (sales ) + u where 1 is the elasticity of salary with respect to sales . If we change the units of measurement of y , what happens to 1 ? Nothing.
(24)
Meaning of Linear Regression

We have seen a model that allows for nonlinear relationships. So what does linear mean? An equation y = 0 + 1 x + u is linear in parameters, 0 and 1 . There are no restrictions on how y and x relate to the original dependent and independent variables. Plenty of models cannot be cast as linear regression models because they are not linear in their parameters. E.g., cons = 1/(0 + 1 inc ) + u
Unbiasedness of OLS
Unbiasedness of OLS is established under a set of assumptions: Assumption SLR.1 (Linear in Parameters) The population model is linear in parameters as y = 0 + 1 x + u where 0 and 1 are the population intercept and slope parameters. Realistically, y , x , u are all viewed as random variables. Assumption SLR.2 (Random Sampling) We can use a random sample of size n, (xi , yi ) : i = 1, 2, , n, from the population model.
(25)
Unbiasedness of OLS (cont.)

Not all cross-sectional samples can be viewed as random samples, but many may be. We can write (25) in terms of the random sample as yi = 0 + 1 xi + ui , i = 1, 2, ..., n (26)
To obtain unbiased estimators of 0 and 1 , we need to impose Assumption SLR.3 (Zero Conditional Mean) E (u |x ) = 0 This assumption implies E (ui |xi ) = 0 for all i = 1, 2, ..., n.

Assumption SLR.4 (Sample Variation in the Independent Variable) In the sample, xi , i = 1, 2, ..., n are not all equal to a constant. This assumption is equivalent to From (12): 1 =
n )(yi i =1 (xi x n )2 i =1 (xi x n i =1 (xi
x )2 > 0
n )yi i =1 (xi x n )2 i =1 (xi x
y )
Plug (26) into this: 1 =

n i =1 (xi
x )(0 + 1 xi + ui ) = 1 + n )2 i =1 (xi x
n i =1 (xi
x )ui SSTx

1 diers from 1 . ui s are generally dierent from 0. The rst important statistical property of OLS: Theorem 2.1 (Unbiasedness of OLS) Using Assumptions SLR.1 through SLR.4, 0 ) = 0 , and E ( 1 ) = 1 E ( The OLS estimates of 0 and 1 are unbiased. Proof:
n
(27)
1 ) = 1 + E [(1/SSTx ) E (
i =1 n
(xi x )ui ] (xi x )E (ui ) = 1
= 1 + (1/SSTx )
i =1
Le Van Chon

(10) implies 0 = y 1 x 1 x 1 ) = 0 + 1 x +u = 0 + (1 x +u 0 ) = 0 + E [(1 1 ) E ( x ] = 0 Remember unbiasedness is a feature of the sampling 0 and 1 . It says nothing about the estimate distributions of we obtain for a given sample. If any of four assumptions fails, then OLS is not necessarily unbiased. When u contains factors aecting y that are also correlated with x , it can result in spurious correlation.

E.g., let math10 denote % of tenth graders at a high school receiving a passing score on a standardized math exam. Let lnchprg denote % of students eligible for the federally funded school lunch program. We expect the lunch program has a positive eect on performance: math10 = 0 + 1 lnchprg + u MEAP93 has data on 408 Michigan high schools for the 1992-1993 school year. math10 = 32.14 0.319lnchprg Why? u contains such as the poverty rate of children attending school, which aects student performance and is highly correlated with eligibility in the lunch program.
Variances of the OLS Estimators

Now we know that the sampling distribution of our estimate is centered about the true parameter. How spread out is this distribution? the variance. We need to add an assumption. Assumption SLR.5 (Homoskedasticity) Var(u |x ) = 2 This assumption is distinct from Assumption SLR.3: E (u |x ) = 0. 0 and This assumption simplies the variance calculations for 1 and it implies OLS has certain eciency properties.
Variances of the OLS Estimators (cont.)

Var(u |x ) = E (u 2 |x ) [E (u |x )]2 = E (u 2 |x ) = 2 Var(u ) = E (u 2 ) = 2 2 is often called the error variance. , the square root of the error variance, is called the standard deviation of the error. We can say that E (y |x ) = 0 + 1 x Var(y |x ) = 2 (28) (29)
Le Van Chon

Homoskedastic case:
Le Van Chon

Heteroskedastic case:
Le Van Chon

Theorem 2.2 (Sampling variances of the OLS estimators) Under Assumptions SLR.1 through SLR.5, 1 ) = Var( 0 ) = Var( 2 2 = n 2 SSTx ( x x ) i i =1
1 2 n n 2 i =1 xi n )2 i =1 (xi x
(30) (31)
Proof: 1 ) = 1 Var( 2 SSTx

n i =1
SSTx 2 2 (xi x ) Var(ui ) = = 2 SSTx SSTx

2
Le Van Chon
(30) and (31) are invalid in the presence of heteroskedasticity. (30) and (31) imply that: j ). (i) The larger the error variance, the larger are Var( j ). (ii) The larger the variability in the xi , the smaller are Var( Problem: the error variance 2 is unknown because we dont observe the errors, ui .
Le Van Chon
Estimating the Error Variance

What we observe are the residuals, u i . We can use the residuals to form an estimate of the error variance. We write the residuals as a function of the errors: u i u i 0 1 xi = (0 + 1 xi + ui ) 0 1 xi = yi 0 0 ) ( 1 1 )xi = ui ( (32)
An unbiased estimator of 2 is 1 2 = n2
Le Van Chon
u i2 =
i =1
SSR n2
(33)
Estimating the Error Variance (cont.)

=
2 = standard error of the regression (SER). 1 ) = , sd ( SSTx
Recall that
if we substitute 2 for 2 , then we have the standard error of 1 : 1 ) = se ( = n SSTx )2 i =1 (xi x
Le Van Chon
Regression through the Origin

In rare cases, we impose the restriction that when x = 0, E (y |0) = 0. E.g., if income (x ) is zero, income tax revenues (y ) must also be zero. 1 x + u Equation y = (34) Obtaining (34) is called regression through the origin. We still use OLS method with the corresponding rst order condition
n i =1
1 xi ) = 0 xi (yi
1 =
n i =1 xi yi n 2 i =1 xi
(35)
1 is a biased estimator of 1 . If 0 = 0, then


Chapter 2

Uploaded by

Copyright:

Available Formats

Chapter 2

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 2

Uploaded by

Copyright:

Available Formats

Denition of Simple Regression Model Ordinary Least Squares Mechanics of OLS Units of Measurement and Functional Form Expected

Values and Variances of OLS Estimators Regression through the Origin

Chapter 2 The Simple Regression Model

June 2012 Based on Introductory Econometrics: A Modern Approach by Wooldridge

Denition of Simple Regression Model

Denition of Simple Regression Model (cont.)

Denition of Simple Regression Model (cont.)

Denition of Simple Regression Model (cont.)

Ordinary Least Squares

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Given the properties of summation, (7) can be rewritten as 0 + 1 x y = (9) 0 = y 1 x or (10)

Ordinary Least Squares (cont.)

the estimated slope is 1 =

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Ordinary Least Squares (cont.)

Fitted Values and Residuals

Algebraic Properties of OLS Statistics

Algebraic Properties of OLS Statistics (cont.)

(yi y )2 is the total sum of squares (SST),

(y i y )2 is the explained sum of squares (SSE),

u i 2 is the residual sum of squares (SSR).

Then SST = SSE + SSR

Algebraic Properties of OLS Statistics (cont.)

and we know that

Units of Measurement on OLS Statistics

Units of Measurement on OLS Statistics (cont.)

What happens to R 2 when units of measurement change? Nothing.

Nonlinearities in Simple Regression

Nonlinearities in Simple Regression (cont.)

Nonlinearities in Simple Regression (cont.)

Meaning of Linear Regression

Unbiasedness of OLS (cont.)

Unbiasedness of OLS (cont.)

Plug (26) into this: 1 =

Unbiasedness of OLS (cont.)

(xi x )ui ] (xi x )E (ui ) = 1

Unbiasedness of OLS (cont.)

Unbiasedness of OLS (cont.)

Variances of the OLS Estimators

Variances of the OLS Estimators (cont.)

Variances of the OLS Estimators (cont.)

Variances of the OLS Estimators (cont.)

Variances of the OLS Estimators (cont.)

Proof: 1 ) = 1 Var( 2 SSTx

SSTx 2 2 (xi x ) Var(ui ) = = 2 SSTx SSTx

Variances of the OLS Estimators (cont.)

Estimating the Error Variance

Estimating the Error Variance (cont.)

2 = standard error of the regression (SER). 1 ) = , sd ( SSTx

if we substitute 2 for 2 , then we have the standard error of 1 : 1 ) = se ( = n SSTx )2 i =1 (xi x

Regression through the Origin

1 is a biased estimator of 1 . If 0 = 0, then

You might also like