gm.screening {gmvalid} | R Documentation |
Preliminary model search procedure for categorical data in three steps, using Goodman's and Kruskal's gamma and chi-squared tests.
gm.screening(data, conf.level = 0.95)
data |
Data frame or array. Variables need to be discrete and should have names. |
conf.level |
Confidence level (default is 0.95). |
The initial model screening divides into three parts:
(1) An edge is added to the main effects model,
if the marginal gamma coefficient or the chi-squared statistic is significant.
(2) An edge is added between two vertices, if either the conditional gamma
coefficient or the conditional chi-squared statistic is significant
for the two variables given any possible variable in condition.
(3) All triplets of variables whose vertices build
a complete subgraph are tested for removal using the conditional gamma coefficient and conditional chi-squared statistic.
An edge is removed if the corresponding p-value is larger than 1 - conf.level
.
mat |
Adjacency matrix of the screened model. |
model |
String of the graphical model. |
The model screening is not a model selection strategy! It 'just' searches for a good
start model to initialize a model selection. Common usage is to select
a model after screening based on a backwards and forwards selection
as implemented in gm.coco
.
Ronja Foraita, Fabian Sobotka
Bremen Institute for Prevention Research and Social Medicine
(BIPS) http://www.bips.uni-bremen.de
Kreiner S and Edwards D (1983) The analysis of contingency tables by graphical models Biometrika, 70(3):553-565
Kreiner S (2008) DIGRAM http://staff.pubhealth.ku.dk/~skm/skm/ last accessed 30.06.2008
Siersma V (2007) Studies in the interactions between disease development and interventions PhD thesis, Faculty of Health Sciences, University of Copenhagen.
backward
, gm.coco
, gm.chi
, gm.gamma
data(wam) model <- gm.screening(wam) gm.coco(wam,recursive=TRUE,criterion="aic",model=model$model)