Homework - ml - week 4
Due - Wed, 2/18


1. Now you are going to do ordered probit. Use nes92nomiss. DV is v2,
attitude towards bush, a 4 point ordered scale. Starting with the
probit template, write down the likelihood function. It is hard (not
impossible, just annoying). So compare your  results
to what Stata gives you via the ordered command.

b. Compute predicted probabilities of being in each category and check
   with Stata predict.

c. Interpret the thresholds in some meaningful way.

d. Rewrite your code to allow the first threshold to be a parameter to
   be estimated. What happens?


2. . For the logit example in 2, use Clarify and the methods of MUS 14.7
to interpret the model in substantive terms. Do a few things, but no
more than a few things.

3. The next exercises use dutch election  data - the codebook is
 dutch271codebook.txt (4 main parties only). The Stata data are
 dutch271uncond.dta. The stata files
 themselves are also somewhat self documenting. The data are for the
 1971(!) Dutch parliamentary election. At this point we will only work
 with the "sociological" (uncoditional) model, all variables pertain
 to people, not people and choices.

  Take an unconditional (mlogit) model  that you
    like.  The depedendent variable, party,  is vote for one of the four
    biggest parties/ It  should have political variables (attitudes) as well as
    sociological variables. Make sure to set the baseline party
    yourself so Stata does not choose for you.
conditional mnl. Choose a baseline party and code it zero. Then
    do an unconditional logit analysis on the 4 major parties (these
    are the only ones in the data set)

a. Interpret A FEW interestng results.

b. Do a test of a hypothesis that one variable of interest has no
   effect on vote.

c. Redo with new baseline party. Check out that the results on one
   interesting condition change in the way discussed in class.

d. Calculate one interesting change in odds (for party 1 vs 0 and
   party 3 vs 4). Calculate both the estimates and their standard
   errors, using methods discussed in MUS or Clarify.3.