Bivariate Regression Analysis

The beginning of many types of

regression
 TOPICS
• Beyond Correlation
• Forecasting
• Two points to estimate the slope
• Meeting the BLUE criterion
• The OLS method
Purpose of Regression Analysis
• Test causal hypotheses

• Make predictions from samples of data

• Derive a rate of change between variables

• Allows for multivariate analysis

 Goal of Regression
• Draw a regression line through a sample of
data to best fit.

• This regression line provides a value of how

much a given X variable on average affects
changes in the Y variable.

• The value of this relationship can be used

for prediction and to test hypotheses and
provides some support for causality.
Perfect relationship between Y and X: X causes all
change in Y

  a  b
Where a = constant, alpha, or intercept (value of
Y when X= 0 ; B= slope or beta, the value of X

Imperfect relationship between Y and X

  a  b  e
E = stochastic term or error of estimation
and captures everything else that affects
change in Y not captured by X
 The Intercept
• The intercept estimate (constant) is where
the regression line intercepts the Y axis,
which is where the X axis will equal its
minimal value.

• In a multivariate equation (2+ X vars) the

intercept is where all X variables equal
zero.
 The Intercept

a    b
The intercept operates as a baseline for the
estimation of the equation.
 The Slope
• The slope estimate equals the average
change in Y associated with a unit change
in X.

• This slope will not be a perfect estimate

unless Y is a perfect function of X. If it
was perfect, we would always know the
exact value of Y if we knew X.
 The Least Squares Concept
• We draw our regression lines so that the
error of our estimates are minimized. When
a given sample of data is normally
distributed, we say the data are BLUE.

• BLUE stands for Best Linear Unbiased

Estimate. So, an important assumption of
the Ordinary Least Squares model (basic
regression) is that the relationship between
X variables and Y are linear.
 Do you have the BLUES?
The BLUE criterion
• B for Best (Minimum error)
• L for Linear (The form of the relationship)
• U for Un-bias (does the parameter truly
reflect the effect?)
• E for Estimator
 The Least Squares Concept
• Accuracy of estimation is gained by
reducing prediction error, which occurs
when values for an X variable do not fall
directly on the regression line.

• Prediction error = observed – predicted or

ˆ
 i  i
NOT BLUE
BLUE
 Ordinary Least Square (OLS)
• OLS is the technique used to estimate a line
that will minimize the error. The difference
between the predicted and the actual values of
Y

ˆ
Y Y  e
 OLS
• Equation for a population

Y    X  
• Equation for a sample

Y  a  bX  e
 The Least Squares Concept
• The goal is to minimize the error in the
prediction of b. This means summing the
errors of each prediction, or more
appropriately the Sum of the Squares of the
Errors.

SSE =  (i  i )
ˆ 2
The Least Squares and b coefficient
• The sum of the squares is “least” when

b
 (   )(   )
i i

• And  (   )i
2

a    b
Knowing the intercept and the slope, we can predict
values of Y given X.
Calculating the slope & intercept

b
 ( Xi  X )(Yi  Y )
 ( Xi  X ) 2

a  Y  bX
 Step by step
1. Calculate the mean X
of Y and X Y

2. Calculate the errors Xi  X

of X and Y Yi  Y

3. Get the product ( Xi  X )(Yi  Y )

(multiply)

4. Sum the products ( Xi  X )(Yi  Y )

 Step by step
5. Squared the ( Xi  X ) 2

difference of X

6. Sum the squared ( Xi  X ) 2

difference

7. Divide (step4/step6) b
 ( Xi  X )(Yi  Y )
 ( Xi  X ) 2

8. Calculate a
a  Y  bX
 An Example: Choosing two points
Y X

Log value Log sqft

5.13 4.02

5.2 4.54

4.53 3.53

4.79 3.8

4.78 3.86

4.72 4.17
Forecasting Home Values
LOG_VALU
5.2

5.1
2

5.0

4.9

4.8

4.7

4.6
Linear

1
4.5
3.4 3.6 3.8 4.0 4.2 4.4 4.6

LOG_SQFT
Forecasting Home Values
LOG_VALU
5.2

5.1

5.0 Y2 - Y1
_______
4.9

4.8 X2 - X1
4.7

4.6
4.54 – 3.53
Linear

4.5
3.4 3.6 3.8 4.0 4.2 4.4 4.6
__________ =.69

LOG_SQFT
5.2 – 4.5
 SPSS OUTPUT
Coefficientsa

Unstandardized Standardized
Coefficients Coefficients
Model B Std. Error Beta t Sig.
1 (Constant) 2.565 .929 2.761 .051
X .575 .232 .778 2.476 .068
a. Dependent Variable: Y

• The coefficient beta is the marginal impact of X

on Y (derivative)
• In other words for a one unit change of X how
much Y changes (.575)
 Stochastic Term
• The stochastic error term measures the
residual variance in Y not covered by X.

• This is akin to saying there is measurement

error and our predictions/models will not be
perfect.

• The more X variables we add to a model,

the lower the error of estimation.
 Interpreting a Regression
Coefficientsa

Unstandardized Standardized
Coefficients Coefficients 95% Confidence Interval for B
Model B Std. Error Beta t Sig. Lower Bound Upper Bound
1 (Constant) 797.952 45.360 17.592 .000 708.478 887.425
UNEMP -69.856 6.500 -.615 -10.747 .000 -82.678 -57.034
a. Dependent Variable: STOCKS

Model Summary

Adjusted Std. Error of

Model R R Square R Square the Estimate
1 .615a .378 .375 122.85545
a. Predictors: (Constant), UNEMP
 Interpreting a Regression

• The prior table shows that with an increase

in unemployment of one unit (probably
measured as a percent), the S&P 500 stock
market index goes down 69 points, and this
is statistically significant.

• Model Fit: 37.8% of variability of Stocks

predicted by change in unemployment
figures.
 Interpreting a Regression 2

• What can we say about this relationship

regarding the effect of X on Y?

• How strongly is X related to Y?

• How good is the model fit?

 Model Fit: Coefficient of
Determination
• R squared is a measure of model fit.

2
R
• What amount of variance in Y is explained
by X variable?

• What amount of variability in Y not

explained by X variable(s)?

R r 2 2
This measure is based on the degree to which
the point estimates of fall on the regression
line. The higher the error from the line, the
lower the R square (scale between 1 and 0).
 i
(    ) 2
= Total sum of squared deviations (TSS)
= regression (explained) sum of squared
 (ˆ i   ) deviations (RSS)
2

= error (unexplained) sum of squared

 i i deviations (ESS)
(   ˆ
 ) 2

TSS= RSS + ESS

Where R2 = RSS/TSS
 Interpreting a Regression 2
Coefficientsa

Unstandardized Standardized
Coefficients Coefficients
Model B Std. Error Beta t Sig.
1 (Constant) 3.057 .041 74.071 .000
UPOP 4.176E-05 .000 .133 13.193 .000
a. Dependent Variable: DEMOC

Correlations

DEMOC UPOP
Model Summary
Pearson Correlation DEMOC 1.000 .133
UPOP .133 1.000 Adjusted Std. Error of
Sig. (1-tailed) DEMOC . .000 Model R R Square R Square the Estimate
1 .133a .018 .018 3.86927
UPOP .000 .
a. Predictors: (Constant), UPOP
N DEMOC 9622 9622
UPOP 9622 9622
 Interpreting a Regression 2
• The correlation between X and Y is weak (.133).

This is reflected in the bivariate correlation

coefficient but also picked up in model fit of .018.
What does this mean?

• However, there appears to be a causal

relationship where urban population increases
democracy, and this is a highly significant
statistical relationship (sig.= .000 at .05 level)
 Interpreting a Regression 2
• Yet, the coefficient 4.176E-05 means that a
unit increase in urban pop increases
democracy by .00004176, which is tiny.

• This model teaches us a lesson: We need

to pay attention to both matters of both
statistical significance but also matters of
substance. In the broader picture urban
population has a rather minimal effect on
democracy.
 The Inference Made
• As with some of our earlier models, when
we interpret the results regarding the
relationship between X and Y, we are often
making an inference based on a sample
drawn from a population. The regression
equation for the population uses different
notation:

Yi = α + βXi + εi
 OLS Assumptions
1. No specification error
a) Linear relationship between X and Y
b) No relevant X variables excluded
c) No irrelevant X variables included

2. No Measurement Error
• (self-evident I hope, otherwise what would we
be modeling?)
 OLS Assumptions
3. On Error Term:
a. Zero mean: E(εi2), meaning we expect
that for each observation the error equals
zero.
b. Homoskedasticity: The variance of the
error term is constant for all values of X i.
c. No autocorrelation: The error terms are
uncorrelated.
d. The X variable is uncorrelated with the
error term
e. The error term is normally distributed.
 OLS Assumptions
• Some of these assumptions are complex
and issues for a second level course
(autocorrelation, heteroskedasticity).

• Of importance is that when assumptions 1

and 3 are met our regression model is
BLUE. The first assumption is related to the
proper model specification. When aspects
of assumption 3 are violated we may likely
need a new method of estimation besides
OLS