unobserved heterogeneity - rob hicksrlhick.people.wm.edu/econ407/presentations/intropanel.pdf ·...
TRANSCRIPT
![Page 1: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/1.jpg)
Unobserved Heterogeneity
An Introduction to Panel Data
![Page 2: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/2.jpg)
The Population Regression Equation
• In an OLS assume COMPLETE PARAMETER HOMOGENEITY– Each and every cross-section unit is assumed to
share the behavioral parameters– Strong Assumption: Everyone reacts to a change in
gasoline prices in exactly the same way (in a strictly linear specification)
![Page 3: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/3.jpg)
Limitations of Cross Section Data
• In a linear regression setting, with such unobserved heterogeneity, perhaps difficult to estimate the correct slope coefficient, because we have a sample of N individuals but really need to estimate N constants plus coefficients on the other independent variables
• Unobserved heterogeneity not identified• Note:
– There are some advanced models that can introduce heterogeneity in a purely cross section setting (random parameter models)
![Page 4: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/4.jpg)
Partial Heterogeneity
• The classic case is that each observation i’s population regression function looks like
![Page 5: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/5.jpg)
y
x i 1 x1
OLS Estimated Relationship
True Relationship for individual red
True Relationship for individual blue
True Relationship for individual green
![Page 6: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/6.jpg)
What is ci?
• This unobserved, individual-specific effect does not vary across time– Ex: Labor market study: cognitive ability, drive,
and determination– Ex: Firm productivity study: each firm’s
organization or managerial structure
![Page 7: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/7.jpg)
Marginal Effects• We continue to maintain our focus on
marginal effects
• 3 Approaches:– Pooled OLS– Random Effects– Fixed Effects
![Page 8: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/8.jpg)
Method 1: Pooled OLS
• OLS (termed Pooled OLS) performs ok when– Independent Variables Exogenous
when
&
![Page 9: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/9.jpg)
Run this Regression Using OLS
![Page 10: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/10.jpg)
Pooled OLS
• The approach is unbiased but consistent• Standard errors are incorrect and the model is
inefficient• Instead, use robust standard errors – since we might
expect there to be different variances of the errors amongst cross section units.– Even if the method ignores possible correlations or errors
within cross section units.• This is a useful benchmark model, but usually we can
do better
![Page 11: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/11.jpg)
Method 2: Random Effects
• Exploiting what we know about the error structure– There must be correlation within cross section
units in the error structure:
![Page 12: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/12.jpg)
The Random Effects Model
• Fixes the standard errors• Puts all the unobserved heterogeneity in the
error term• For unbiasedness, relies on the condition
![Page 13: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/13.jpg)
OLS Bias and Partial Heterogeneityy
x i 1 x1
OLS Estimated Relationship
True Relationship for individual red
True Relationship for individual blue
True Relationship for individual green
![Page 14: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/14.jpg)
Method 3: Fixed Effects
Directly dealing with the unobservable ci
![Page 15: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/15.jpg)
A Two Period Example
• Now consider a different approach- we have information for each individual for two periods {t=1,2}
![Page 16: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/16.jpg)
First Difference
• The difference is
• Notice, that the difference is only a function of the xi (m and z) and ci is not in the model anymore
![Page 17: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/17.jpg)
For 2 period case, we now have a cross section
• One observation per cross section unit:
![Page 18: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/18.jpg)
We have the OLS Regression
• Under what conditions is the OLS estimate unbiased?
![Page 19: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/19.jpg)
And we are back in the OLS World
• Or are we? – Consider endogeneity and the proof of
unbiasedness:
![Page 20: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/20.jpg)
Need Strict Exogeneity
• Unobservable factors in one period can’t be related to the observable factors in any other period:
– For example, unexplained factors in period 1 can’t influence educational attainment in period 2.
![Page 21: Unobserved Heterogeneity - Rob Hicksrlhick.people.wm.edu/econ407/presentations/IntroPanel.pdf · • Unobserved heterogeneity not identified • Note: – There are some advanced](https://reader035.vdocument.in/reader035/viewer/2022070923/5fbb0e97f152fc5f2a3649da/html5/thumbnails/21.jpg)
Fixed Effects
• Deals with cases where unobserved heterogeneity is correlated with the independent variables
• Is unbiased, consistent, and efficient• Does NOT allow the inclusion of time invariant
independent variables in the analysis