Further Statistical Methods -HT05
Practical - week 3
MIM and graphical models.
- Start MIM
- Download this file from a study of housing types
in Denmark and look at the file. The data is also available in MASS and
described there. Four variables: type of housing (4 different types), contact with neighbours (high/low), influence on decisions (low, medium, high),
and satisfaction (low, medium, high).
- Read the file into MIM
- Specify the saturated model and display its graph
- Which variables are ordinal?
- Specify these as ordinal in MIM (write eg ordinal i on command line)
- Test all conditional independences of any pair of variables, given the
others. Both by using a standard LR test and the appropriate test which takes
into account ordinality of some of the variables. Check also whether the
asymptotic results seem to conform with Monte-Carlo p-values. Decompose the
tests according to the values of the conditional variables (option z).
- Argue that, in light of the above, the graphical model with cliques
chs,cis, i.e. a single conditional independence of hi given cs, is a
reasonable model.
- MIM has different model selection procedures built in. Use global search
to find the best model using AIC and BIC. Look at their graphs and give an
interpretation.
- Make the model under 8 the base model and test the model found by BIC. Comment on the result.
- Try other model search methods.
- Study the marginal relationships between csh and cis respectively by
specifying the model as csh and cis respectively and look for independencies. Remember to take into
account ordinality.
- If time allows, study the odds-ratios in tables csh and cis for fixed
value of one of the other variables. How do the odds-ratios vary from one to
the other?
Back to course overview.
Last updated:
Friday, 18 February 2005 09:14.
Steffen L. Lauritzen