Influence.ME is continuously being developed, the process of which is tracked on this page. The changes made in each new version are chronicled here. Also, changes to the development version of influence.ME is given. Although not yet publicly available, these changes give an idea about what’s coming in the next version. Although the code for these reported changes has all actually been written, some functionality may be withdrawn from a next version. In other words: no guarantee is given that the noted changes will actually be published in the next version of influence.ME. Any comments and / or feature request are highly appreciated.
Version 0.6.1: (published 2009-05-27)
Version 0.6: (published 2009-05-25)
- Added ME.pchange() and accompanying dp.ME.pchange() functions, to calculate and visualize the percentage change in the magnitude of fixed parameter estimates.
- Added the sort=TRUE parameter to the ME.cook() and dp.ME.cook() function to allow the output the be sorted.
- Added sort=TRUE and to.sort=… parameters to the ME.dfbetas() and dp.ME.dfbetas functions, to allow the output to be sorted, and -in the case of multiple variables- to indicate on which variable output should be sorted. These functions are present in the ME.pchange() and dp.ME.pchange() functions as well.
- Added abs=TRUE parameter to ME.dfbetas() and dp.ME.dfbetas() functions, to have the absolute value of DFBETAS returned. This parameter is present in ME.pchange() and dp.ME.pchange() as well.
- Changed procedure for data modification. During testing, we encountered quirky behavior regarding the update() function. As a result of which, the exclude.influence() under special conditions returned models with unreliable estimates. We only found this problem concerning generalized mixed models (untested with non-linear models). The variable order of the modified models returned by exclude.influence() was changed, which seems to have resolved this problem.
- Added documentation for the ME.pchange() and dp.ME.pchange() functions
- Modified documentation on the ME.cook(), dp.ME.cook(), ME.dfbetas(), and dp.ME.dfbetas() functions to reflect the new options on sorting the output.
- Changed title to: Tools for recognizing inï¬‚uential data in mixed models
- A few minor textual changes
Version 0.5: Initial Release (published 2009-04-11)
This is the initial release version of influence.ME. All basic functionality is present: excluding the influence of a grouping level from a mixed model, gathering the estimates excluding each of the grouping levels, and calculating Cook’s Distance and DFBETAS. Also, basic graphic representation of these measures is available.