March 18, 2003

  
THIS IS FOR STUFF TO DO NEXT TIME I DO  A GEORGE MASON STATS FOR JUDGES

idea for frank:  a year later, send an email to all jduges, with names of people at											    
Institute they attended, emails, etc., and of instructors. and topics of days
lectures. Afte Henry Manne-- create networks. 

 Check out font sizes of STATA, EXCEL, beforehand.

Social data: take otu hte BLACK variable--too distracting. 

  
 Buy blank cube dice at a game shop. Paint them  +1,+2, +3. I
need some kids'
blocks.
 Paint a different die with +2, +4, +6. 

 

 Heteroskedasticity and Autocorrelation.  Put it in terms of the
disturbances.

Hetero: Use weighted least squares. Some observations have more
disturbance. If you
knew in advance which ones, you would weight them less.

Autocorrelation: If the disturbances are correlated, then you
have the same bias in
lots of observations.

Suppose you are trying to estimate an average, M. You observe
X1=M+U1.  Your best
estimate  is Mhat =X.
If you then observe X2=M+U2, you would average the two
observations together, to
get M+ (U1+U2)/2.  Since sometimes U1 has the opposite sign from
U2, the
disturbances can cancel out.  Roll dice. Can I get blank cubes
and paint on them?


Autocorrelation:  U1 and U2 are correlated. So it is like you
only have one
observation.
  Hetero:  U1 has much less variance than U2. So it should get
greater weight.

 



 

 DAY ONE: 
2. Regression to the mean examples.  Meaning of term Regression. 
  1. Law Associates example.

Correlation. A lot of time on slope vs. correlation. 

Hypothesis testing, with the applet. Explain the applet more than I did. 

 The Eye-drawing applet. 

Least squares vs. least deviations

Reducing the size of the dataset. 



Using Excel.  I did descriptive stats, correlation, and regression. I should do
graphing too.

 sample selection bias. If you pick 100 law firms, you will find 5 with apparent
discrimination.




DAY TWO
 Start with a long summary of day one. 
 THE Japanese judges example. Frank thinks I should skip it next time. He would
like more on expert witnesses. I could have ttranscripts of two good experts and of
one bad expert.
   Social data on states. From my overheads, and from STATA.
 

Maybe have as a theme, estimators vs. hypothesis testing (best guess vs. how good
that guess is)