A nested working dependence structure.
A working dependence structure that captures a nested hierarchy of groups, each level of which contributes to the random error term of the model.
When using this working covariance structure,
dep_data of the GEE instance should contain a n_obs x k matrix of 0/1 indicators, corresponding to the k subgroups nested under the top-level
groups of the GEE instance. These subgroups should be nested from left to right, so that two observations with the same value for column j of
dep_data should also have the same value for all columns j’ < j (this only applies to observations in the same top-level cluster given by the
groups argument to GEE).
Suppose our data are student test scores, and the students are in classrooms, nested in schools, nested in school districts. The school district is the highest level of grouping, so the school district id would be provided to GEE as
groups, and the school and classroom id’s would be provided to the Nested class as the
dep_data argument, e.g.
Labels lower in the hierarchy are recycled, so that student 0 in classroom 0 is different fro student 0 in classroom 1, etc.
The calculations for this dependence structure involve all pairs of observations within a group (that is, within the top level
group structure passed to GEE). Large group sizes will result in slow iterations.
The variance components are estimated using least squares regression of the products r*r’, for standardized residuals r and r’ in the same group, on a vector of indicators defining which variance components are shared by r and r’.
||Returns the working covariance or correlation matrix for a given cluster of data.|
||Solves matrix equations of the form |
||Called on the first call to update|
||Returns a summary string describing the state of the dependence structure.|
||Updates the association parameter values based on the current regression coefficients.|
© 2009–2012 Statsmodels Developers
© 2006–2008 Scipy Developers
© 2006 Jonathan E. Taylor
Licensed under the 3-clause BSD License.