Scribd
Upload
15 views16 pages

2-04 Non-Linear Models - Quadratic and Polynomial Models

Uploaded by

MishalBD
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
15 views16 pages

2-04 Non-Linear Models - Quadratic and Polynomial Models

Uploaded by

MishalBD
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 16

Week 4, Tuesday

Feb 4
Topic for today
Non-linear models – Quadratic and Polynomial models
What we did last class
1. Should we care about getting our fitted values to be as close
as possible to observed values?
2. What happens when two or more independent variables are
correlated with each other?
3. What happens when we add irrelevant variables to our
model?
4. What should we be suspicious of when judging the results of a
regression analysis?
Questions we want to answer today
1. How do we estimate non-linear relationships between X and
Y?
2. How does the interpretation of coefficients change with non-
linear estimation?
3. How do we find ‘turning points’ in non-linear relationships?
Age (x-axis) and life satisfaction (y-axis)
Non-Linear Relationships in OLS
Though OLS is also known as linear regression, it can fit some
non-linear models :
• X values can be squared, cubed, etc.
• s cannot be subjected to a non-linear function
• Non-linearity in variables is feasible with OLS.
• Non-linearity in parameters is not feasible with OLS.

OLS can estimate : OLS can’t estimate :


Non-Linear Relationships in OLS
OLS can estimate non-linear effects via polynomial models.

1. A polynomial model includes X raised to power greater than 1.


The general form is
=+ + + +…

2. The most commonly used polynomial is the quadratic model


=+ + +
No need to ‘guess’ the
correct curvature.

Just specify a quadratic


term and OLS will yield
the curve that fits the
data!
Life expectancy and GDP
Extreme case
Example in which data is non-linear model but the following
model suggests no relationship:
Yi = b0 + b1Xi + ei

• This is “misspecification
bias”.
• It is a special case of
omitted variables bias.
• What is the omitted
variable in this case?
Coefficient interpretation
• In non-linear model, the effect of X changes
depending on the value of X
• Use calculus to calculate the effect (by taking derivative of Y
with respect to X)
• For Yi = b0 + b1Xi + b2Xi2 + ei the predicted effect of a one unit
increase in X is

b1 + 2b2Xi
S S
• Predicted change depends on
year:

• For Year = 1900:


= -0.166 + 2*0.000044*1900
= 0.0012 degrees
• For Year = 2000:
= -0.166 + 2*0.000044*2000
= 0.01 degrees
• Temperature is not just rising,
the pace of increase is
accelerating.
• Can’t be captured by a linear
model (constant slope!)
Finding key points with U-shaped fitted values
• For U shaped or upside down U shape relationship, the
minimum/maximum occurs when slope (first derivative) equals
zero (this is the flat part of the figure)
b1 + 2b2Xi = 0
Xi [min or max] = -b1 /2b2
≈ 0.072/(2 x 0.0009) = 40

Dependent variable: life satisfaction


Estimate Std. Error t value
(Intercept) 7.3609470 0.3866000 19.040
age -0.0721794 0.0179919 -4.012
ageSq 0.0008774 0.0001898 4.622
Income 0.4377058 0.0665792 6.574
Relig2 0.1309623 0.0263694 4.966
Child 0.0567545 0.0375899 1.510
Residual standard error: 1.738 on 1114 d.f.
Multiple R-squared: 0.08476
Recap
1. We estimate non-linear relationships between X and Y by
including squared or higher-order terms of X.
2. In non-linear regression, the effect of X on Y depends on the
value of X, i.e. the effect varies and is not constant and
hence the interpretation must incorporate that.
3. We find ‘turning points’ in non-linear relationships using the
fact that slopes at maximum and minimum points are 0.
What we will do next class
1. What other ways are there to estimate non-linear
relationships?
2. How does the interpretation of coefficients change when
estimating in logs?
3. How do we compare coefficients of variables when they are
using very different scales?

You might also like