Figure 3.

Conditional independence graph. The figure shows a conditional independence graph of the p = 20 variables (nodes) remaining after construction of indices based on the 2007 Swiss Health Survey estimated with GRaFo. Edges were selected with respect to an upper bound of 5 on the expected number of false positives <a onClick="popup('http://www.biomedcentral.com/1471-2458/12/655/mathml/M6','MathML',630,470);return false;" target="_blank" href="http://www.biomedcentral.com/1471-2458/12/655/mathml/M6">View MathML</a>. For example, <a onClick="popup('http://www.biomedcentral.com/1471-2458/12/655/mathml/M7','MathML',630,470);return false;" target="_blank" href="http://www.biomedcentral.com/1471-2458/12/655/mathml/M7">View MathML</a> is to be interpreted as an edge which is present in a graph in which we expect up to 2 false edges. We cannot set this bound to 0 as this would be equivalent to a graph in which can be certain that all shown edges are correct (consequently, the algorithm would suggest an empty graph). Five nodes (social network utilization, migration background, smoker, work restriction, and leisure time physical activity) were isolated (no edges) and thus neglected. Reproduced from [15].

Fellinghauer et al. BMC Public Health 2012 12:655   doi:10.1186/1471-2458-12-655
Download authors' original image