0
$\begingroup$

I have data that look like this in a 365*804 matrix:

 Sample     Variant1      Variant2        Variant3
 Person1      1/0            0/0             1/1
 Person2      0/1            1/1             0/0
 Person3      0/0            0/0             0/0

I want to see if any Variants are significantly associated with each other. By that I mean, if you see one how likely are you to see another? The possible values are either (0/0, 1/0, 0/1 or 1/1) with 1/0 and 0/1 being the same thing. If weighting is possible, 1/0 and 0/1 should be worth half what 1/1 is. I would prefer to do this in R.

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Are you saying there are 365 people and 804 variants? So that the output would be 804*803 = 645612 three-by-three tables? $\endgroup$
    – whuber ♦
    Commented Mar 29, 2013 at 16:35
  • 1
    $\begingroup$ Or do you mean to look at the sums across people, in which case you would have pairs of variants - either 804*803/2 or 365*364/2 pairs - but in either case, a vast number, many of which will be significant even if the data are totally random. $\endgroup$
    – Peter Flom
    Commented Mar 29, 2013 at 16:43
  • $\begingroup$ Yes 365 people and 804 variants. I may be able to the number of variants by pooling later on, so while I was aware that some will occur by chance, multiple testing correction and this pooling will help. I just want ideas of the best way(s) to do this. I want to see if some variants (two or more) usually occur together in different people. By occur I mean are given a non-zero score. $\endgroup$
    – cianius
    Commented Mar 29, 2013 at 18:25

1 Answer 1

Reset to default
0
$\begingroup$

The question is not entirely clear, and maybe that is because your goals are not very clear? In that case you should start with some descriptive analyses.

Maybe clustering the persons to to if there are groups of persons with similar profiles on the variables?. That could be a good start. Or clustering the variables for the same purpose. Maybe that could lead to some simplified representation of the data. Or this methods could be combined is some form of dual scaling, the following paper seems relevant SPSS white paper.

Taking want to see if any Variants are significantly associated with each other more literally, you could estimate some correlation matrix (of size $804\times 804$), but then we need some correlation measure for categorical variables. See for instance How to measure correlation between categorical variable? or Correlation and categorical variables.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.