0
$\begingroup$

Im am running a fishers exact test on some data but im unsure of how the input contingency table should be formatted.

Option 1

   1  2 3  4  5 6  7
1  0  5 0 18  1 0  7
2  5  2 0 15  1 0  9
3  0  0 0  2  0 0  1
4 18 15 2 46 25 0 27
5  1  1 0 25 10 0  7
6  0  0 0  0  0 0  0
7  7  9 1 27  7 0 12

or Option 2

   1  2  3  4  5  7
1  0  1  0 12  0  2
2  4  1  0  6  1  5
3  0  0  0  2  0  0
4  6  9  0 23 10  9
5  1  0  0 15  5  3
7  5  4  1 18  4  6

The difference is, that in option 1, all occurrences of [1,2] (five in total) are in both cells they meet. In option 2 there is one occurrence of [1,2] and four occurrences of [2,1] which are listed separately. The results of those tables are very different (Option 1 is significant, Option 2 is not). Which one is the right way to go?

EDIT: The rows and columns represent clusters of states and the values are conflicts between or within clusters. The rows and columns can be seen as who initiated the conflict, but this information does not matter for my analysis. I simply want to analyze how many conflicts there were between or within clusters and how significant the results are.

Improve this question
$\endgroup$
9
  • 3
    $\begingroup$ You need to tell us what the row/col categories represent. But no, there is no requirement that the table should be symmetric. What usually is required, is that in the table each event should be counted exactly once. $\endgroup$
    – kjetil b halvorsen ♦
    Commented Oct 25, 2022 at 12:07
  • $\begingroup$ The rows and column represent clusters of countries. The numbers in the cells represent conflicts that took place between countries in the clusters (actually, the level of analysis is the cluster). $\endgroup$
    – craszer
    Commented Oct 25, 2022 at 12:31
  • $\begingroup$ What distinguishes then row/cols? The cluster initiating a conflict, if that makes sense? Please add all new information as an edit to the post! $\endgroup$
    – kjetil b halvorsen ♦
    Commented Oct 25, 2022 at 12:36
  • 1
    $\begingroup$ OK, but then how do you decide where (below/above) diagonal to put a certain conflict? That seems to be unimportant ... $\endgroup$
    – kjetil b halvorsen ♦
    Commented Oct 25, 2022 at 12:59
  • 2
    $\begingroup$ That means you have a special kind of contingency table, needing special models, not the usual chisq test. Put all counts above the diagonal, and let the lower part be undefined. But you also has some counts on the diagonal, what does those mean? $\endgroup$
    – kjetil b halvorsen ♦
    Commented Oct 25, 2022 at 13:31

0

Reset to default

Your Answer

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