Scribd
Upload
14 views

King 13

Uploaded by

Lance
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
14 views

King 13

Uploaded by

Lance
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Identification

Seemingly Unrelated Regression Models

Reciprocal Causation (Endogeneity)

Multinomial Choice Models

Reciprocal Causation (Endogeneity) 11 / 23 .


Model for Reciprocal Causation
• Stochastic component: (𝑌1𝑖 , 𝑌2𝑖 ) ∼ 𝑁 (𝜇1𝑖 , 𝜇2𝑖 , 𝜎1 , 𝜎2 , 𝜎12 )
• Systematic components

Vote: 𝜇1𝑖 = 𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 + 𝜇2𝑖 𝛽3


PID: 𝜇2𝑖 = 𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + 𝜇1𝑖 𝛾3
Where,
𝛽3 and 𝛾3 : QOIs
𝑥1 demographics (in both equations)
𝑥2 candidate characteristics (affecting vote but not PID)
𝑥3 parents PID (affecting PID but not vote)
• Likelihood
𝑛
𝑓 (𝑦∣𝜇, Σ) = ∏ 𝑁 (𝑦𝑖 ∣𝜇𝑖 , Σ)
𝑖=1
𝑛
1
= ∏(2𝜋)𝑁 /2 ∣Σ∣−1/2 exp {− (𝑦𝑖 − 𝜇𝑖 )′ Σ−1 (𝑦𝑖 − 𝜇𝑖 )}
𝑖−1 2
• What do we do with 𝜇𝑖 = (𝜇𝑖1 , 𝜇𝑖2 )?
Reciprocal Causation (Endogeneity) 12 / 23 .
How to substitute in the systematic component?
• Goal of reparameterization: Reduce parameter
dimensionality
• Standard models: reparameterized (substitute 𝜇 for 𝑋 𝛽)
• Time series models: recursively reparameterized
• M.E. Models: multiple equation reparameterization
𝜇1𝑖 = 𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 + 𝜇2𝑖 𝛽3
= 𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 + (𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + 𝜇1𝑖 𝛾3 )𝛽3
= 𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 + 𝑥1𝑖 𝛾1 𝛽3 + 𝑥3𝑖 𝛾2 𝛽3 + 𝜇1𝑖 𝛾3 𝛽3
1
=( ) [𝑥1𝑖 𝛽1 + (𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 )𝛽3 + 𝑥2𝑖 𝛽2 ]
1 − 𝛾3 𝛽3

𝜇2𝑖 = 𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + 𝜇1𝑖 𝛾3


= 𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + (𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 + 𝜇2𝑖 𝛽3 )𝛾3
= 𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + (𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 )𝛾3 + 𝜇2𝑖 𝛽3 𝛾3
1
=( ) [𝑥1𝑖 𝛾1 + 𝑥3𝑖 𝛾2 + (𝑥1𝑖 𝛽1 + 𝑥2𝑖 𝛽2 )𝛾3 ]
1 − 𝛽3 𝛾3
Reciprocal Causation (Endogeneity) 13 / 23 .
What Happens With no Variable Exclusions?

• Suppose we drop 𝑋2 and 𝑋3 from systematic components

1
𝜇1𝑖 = ( ) [𝑥1𝑖 + 𝑥1𝑖 𝛾1 𝛽3 ]
1 − 𝛾3 𝛽3
𝛽1 + 𝛾1 𝛽3
= 𝑥1𝑖 ( )
1 − 𝛾3 𝛽3

• Likelihood not identified


• {𝛽1 = 1, 𝛾1 = 30, 𝛽3 = 555, 𝛾3 = −30} ⟹ 𝜇1𝑖 = −𝑥𝑖 .
• {𝛽1 = 1, 𝛾1 = 1, 𝛽3 = −999, 𝛾3 = −1} ⟹ 𝜇1𝑖 = −𝑥𝑖 .
• As long as 𝛽1 = 1 and 𝛾1 = −𝛾3 , for any value of 𝛽3 , 𝜇1𝑖 = −𝑥𝑖 .
• ↝ High levels of model dependence wrt choices of 𝑋2 and 𝑋3
• What about other approaches?
• Lots of choices (IV, 2SLS, 3SLS, etc.)
• Should we believe them? (The girl in the mirror)
• My advice: wherever possible, choose a different project

Reciprocal Causation (Endogeneity) 14 / 23 .

You might also like