Transcript Bivariate Regression Analysis

Interpreting Bi-variate OLS Regression
• Stata Regression Output
• Regression plots and RSS
• R2 -- Coefficient of Determination
– Adjusted R2
• Sample Covariance/Correlation
• Hypothesis Testing
– Standard Errors
– T-tests and P-values
Week 4, 2007
Lecture 4
Slide #1
Data
• Use the “caschool.dat” file
• Data description:
– CaliforniaTestScores.pdf
• Build a Stata do-file as you go
• Model:
– Test score=f(student/teacher ratio)
Week 4, 2007
Lecture 4
Slide #2
Stata Regression Model:
Regressing Student Teacher Ratio onto Test Score
histogram str, percent normal
0
0
5
5
Percent
Percent
10
10
15
15
histogram testscr, percent normal
15
Week 4, 2007
20
Student Teacher Ratio
25
600
Lecture 4
620
640
660
Test Score
680
700
Slide #3
Regression Output
regress testscr str
Source |
SS
df
MS
Number of obs =
420
-------------+-----------------------------F( 1,
418) =
22.58
Model | 7794.11004
1 7794.11004
Prob > F
= 0.0000
Residual | 144315.484
418 345.252353
R-squared
= 0.0512
-------------+-----------------------------Adj R-squared = 0.0490
Total | 152109.594
419 363.030056
Root MSE
= 18.581
-----------------------------------------------------------------------------testscr |
Coef.
Std. Err.
t
P>|t|
Beta
-------------+---------------------------------------------------------------str | -2.279808
.4798256
-4.75
0.000
-.2263628
_cons |
698.933
9.467491
73.82
0.000
.
------------------------------------------------------------------------------
Week 4, 2007
Lecture 4
Slide #4
Regression Descriptive Statistics
cor testscr str, means
Variable |
Mean
Std. Dev.
Min
Max
-------------+---------------------------------------------------testscr |
654.1565
19.05335
605.55
706.75
str |
19.64043
1.891812
14
25.8
| testscr
str
-------------+-----------------testscr |
1.0000
str | -0.2264
1.0000
Week 4, 2007
Lecture 4
Slide #5
Regression Plot
600
620
640
660
680
700
twoway (scatter testscr str) (lfitci testscr str)
15
20
str
95% CI
testscr
Week 4, 2007
Lecture 4
25
Fitted values
Slide #6
Measuring “Goodness of Fit”
• Root of Mean Squared Error (“Root MSE”)
se 
RSS
, where RSS =  e2 , K = parameters
n K
– Measures spread around the regression line
• Coefficient of Determination (R2)
ESS   (Yˆi  Y ) 2 and TSS   (Yi  Y ) 2
“model” or explained sum of squares
R2 
Week 4, 2007
“total” sum of squares
ESS
RSS
and (1  R 2 ) 

TSS
TSS
Lecture 4
2
e

2
(
Y

Y
)
 i
Slide #7
Explaining R2
For each observation Yi, variation around the mean can
be decomposed into that which is “explained” by the
regression and that which is not:
Book terminology:
TSS = (all)2
RSS = (unexplained)2
ESS = (explained)2
unexplained deviation
explained deviation
Y
Yˆ
Week 4, 2007
Lecture 4
Stata terminology:
Residual = (unexplained)2
Model = (explained)2
Total = (all)2
Slide #8
Sample Covariance & Correlation
• Sample covariance for a bivariate model is
defined as:
 (Xi  X )(Yi  Y )
sXY 
n 1
• Sample correlations (r) “standardize”
covariance by dividing by the product of the
X and Y standard deviations:
sX Y
r
sX sY
Week 4, 2007
Sample correlations range from
-1 (perfect negative relationship) to
+1 (perfect positive relationship)
Lecture 4
Slide #9
Standardized Regression Coefficients
(aka “Beta Weights” or “Betas”)
• Formula:
sX
b  b1
sY
*
1
 1.892 
• In our example: - 2.28  
  0.226
 19.053 
• Interpretation: the number of std. deviations
change in Y one should expect from a onestd. deviation change in X.
Week 4, 2007
Lecture 4
Slide #10
Hypothesis Tests for Regression
Coefficients
• For our model: Yi = 698.933-2.279808*Xi+ei
• Another sample of 420 observations would lead to
different estimates for b0 and b1. If we drew many
such samples, we’d get the sample distribution of
the estimates
• We need to estimate the sample distribution,
(because we usually can’t see it) based on our
sample size and variance
Week 4, 2007
Lecture 4
Slide #11
To do that we calculate SEbs
(Bivariate case only)
se
SEb1 
, where TSSX   (Xi  X ) 2
TSSX
SEb0  se
Week 4, 2007
1
X2

n TSSX
Lecture 4
Slide #12
Interpreting Standard Errors
• For our model:
– b0 = 698.933, and SEb0 = 9.467
– b1 = -2.28, and SEb1 = .4798
Estimated Sampling Distribution for b1
Assuming that we estimated the
sample standard error correctly, we
can identify how many standard
errors our estimate is away from
zero.
The T-test reports the number of
standard errors our estimate falls
away from zero. Thus, the “T” for
b1 is 4.75 for our model. (rounding!)
b1 = -2.28
b1 - SEb1=-2.76
b1 + SEb1= -1.8
Week 4, 2007
Lecture 4
0
(which is 4.75 SEb1 “units”
away from b1)
Slide #13
Classical Hypothesis Testing
Assume that b1 is zero. What is the probability that your sample would have
resulted in an estimate for b1 that is 4.75 SEb1’s away from zero?
To find out, determine the cumulative density of the estimated sampling
distribution that falls more than 4.75 SEb1’s away from zero.
See Table 2, page 757, in Stock & Watson. It reports discrete “p-values”, given
the sample size and t-values. Note the distinction between 1 and 2 sided tests
In general, if the t-stat is above 2,
the p-value will be <0.05 -- which is
the acceptable upper limit in a
classical hypothesis test.
Note: in Stata-speak,
a p-value is a “p>|t|”
Assume that b1 = 0.0
(null hypothesis)
Week 4, 2007
Estimated b1 = 2.27
(working hypothesis)
Lecture 4
Slide #14
Coming up...
• For Next Week
– Use the caschool.dta dataseet
– Run a model in Stata using Average Income (avginc) to
predict Average Test Scores (testscr)
– Examine the univariate distributions of both variables
and the residuals
• Walk through the entire interpretation
• Build a Stata do-file as you go
• For Next Week:
– Read Chapter 8 of Stock & Watson
Week 4, 2007
Lecture 4
Slide #15