Quant II
Duration exercise
Due: Mar 4 and 11

Continuous time duration methods 

Due Mar 4




A. Use durat.dta, which is the King et al coalition data set, already
stset.  The codebook and description of the data are in the
Box-Steffensemeier and Jones book.


1. Look at the KM plot of hazard and survivor. What does it tell you.

2. Now estimate an exponential model with some iv's of interest.
Interpret (briefly) the results. Make sure to use nohr if you want to
use the parameterization I find simplest; if you do not use nohr, make
sure you interpreations take that into account.

2a. For one iv (putting the others at reasonable values such as the
    mean or median or zero for dummies), by hand (that is, using
    spreadsheet like commands in stata or the like, you don't need to
    multiply by hand!), compute the expected duration of a govt for
    two different values of that iv. Briefly state results in English.

 2b. Have stata draw the hazard and survival plots at whatever values
     you chose for 2a. Compare the plots (and of course briefly discuss)

 2c. Redo using your own ml code in Mata - only thing that is at all challenging     is dealing with censoring - make sure your results match Stata's


Due Mar 11

3. Repeat (all of) 2 for a Weibull? How do the results change. Interpret 
   carefully the weibull parm. What does it tell you? Plot the hazard?

3a. Try another of the densities that Stata allows; plot the hazard? how do the results differ?

B. Cox

1. Take the parametric estimations of part A. Redo with a cox
   proportional hazards model. How do the results differ?

a. Play with the different means of breaking ties. Does it matter?

2. Use stata to capture the baseline hazard and graph it. What does it
tell you?