I am doing cluster analysis with a set of binary variables. Given the nature of variables and the research question, co-absence is not as much an indication of similarity as co-presence is. However, co-absence is still an indiciation of similarity (just not as much as co-presence is). For example, one of my variables indicates whether the person has punished for a criminal offense, and 10 percent of the persons in the dataset has been.
To me (and here I might be wrong) it seems as if there are three types of similarity measures available in Stata:
1) co-absence is as important as co-presence - e.g. (Zubin 1938, Sokal and Michener 1958) binary similarity coefficient - 'matching'
2) co-absence is not at all an indication of similarity - Jaccard (1901, 1908) binary similarity coefficient - 'jaccard'
3) matches - whether they come in the form of co-absence or co-presence - is given more weight than non matches - e.g. Sneath and Sokal (1962) binary similarity coefficient - 'sneath'
So what I am looking for here is a similarity coefficient which gives more weight to co-presence than co-absence, but still includes co-absence in measuring similarity. Maybe I am getting something fundamentally wrong here, but still, if I do I would much appreciate any help.
To me (and here I might be wrong) it seems as if there are three types of similarity measures available in Stata:
1) co-absence is as important as co-presence - e.g. (Zubin 1938, Sokal and Michener 1958) binary similarity coefficient - 'matching'
2) co-absence is not at all an indication of similarity - Jaccard (1901, 1908) binary similarity coefficient - 'jaccard'
3) matches - whether they come in the form of co-absence or co-presence - is given more weight than non matches - e.g. Sneath and Sokal (1962) binary similarity coefficient - 'sneath'
So what I am looking for here is a similarity coefficient which gives more weight to co-presence than co-absence, but still includes co-absence in measuring similarity. Maybe I am getting something fundamentally wrong here, but still, if I do I would much appreciate any help.
Comment