Nordic
Informatics Network in the Agricultural Sciences
Likelihood-based inference for hierarchical/mixed statistical mod
How to prepare I (necessary)
Prepare a dataset to use for the course
project (details below). The purpose of the course project is to analyse some real data, preferably your own,
by some of the methods covered in the course. As the focus of the summer school is on
hierarchical models, the dataset should preferably have a hierarchical structure (possibly spatial
or repeated measures).
It may be useful, and is certainly allowed, to have analysed the data before.
A few prepared
datasets will be available for course projects, and those not bringing their own data can also team up with
participants with data, but we do encourage bringing data of your
personal interest. Course projects will be presented in
(informal) lectures at the end of the summer school. The preparation of
your data/project (assuming that you bring your own data) should involve the following two steps:
- A brief (1 page) summary description of your project, including
- Title
- Short introduction - background to the data/study
- Description of your data where you
i) specify the hierarchical levels (e.g. cow, herd, county) and number
of units at each level,
ii) identify the key dependent variable(s) and the level at which they
exist,
iii) identify the key independent variables and the level at which they exist
- Purpose of the study/analysis - your thoughts on:
i) what is the most important hypothesis you want to study? (this should
fit with the dependent and independent variables identified above)
ii) what are your expectations? (based on literature or previous work
with the data)
These project descriptions will be photocopied and distributed to the
instructors and the participants.
- Data preparation - some guidelines/suggestions:
- structure the data with one record per observation at the lowest
level of the hierarchy (e.g. if the dataset contained data from cows within herds, the dataset should have 1 record per cow),
- make sure that each observation is uniquely identified (e.g. herd
and cow id's),
- identify the key variables of interest and create a dataset with
just those variables in it (rather than bringing the whole dataset if it is very large),
- if there are a lot of missing values, you might want to bring also a
version consisting of observations for which complete data are available,
- bring the data in a basic format (e.g., text file, csv-format or
Excel spreadsheet); you may additionally
bring the data in the format of a major statistical software package
(e.g., SAS, Stata or SPSS), but do include the basic format as well.
In case of problems or questions, contact
Henrik Stryhn (hstryhn@upei.ca)
for guidance on your choice of data set.
How to prepare II (recommended)
- if it has been some time since your last statistical analysis and you feel a little "rusty", take
some time to revisit your last work with regression analysis (and mixed models, if you have worked
with those before),
- if you have never worked with Bayesian methods before, browse
through one or a couple of the introductions to Bayesian methods on the
WinBUGS website,
- select a reference (journal article) from your field of interest in which mixed models
were used or discussed, and bring it with you to the summer school.
How to prepare III (optional)
Questions or comments about the content of this page to
Henrik Stryhn
(hstryhn@upei.ca).
Author:
phd@dina.kvl.dk. Updated:
18 juli 2005