Internal consistency

[Attila Kovacs]

Hi,

I have to analyse a dataset in which there are drugs and used animals based on article review. The datasets looks like this:

rat mice dogs ...
Drug1 5 10 7
Drug2 3 2 9
..
In the above table numbers show article number in which the given animal was used by given drug. There are more drugs and animals, of course. The question is whether there is a difference between usage.

Perhaps chi-squared could check association, but it can't tell ranks. Spearman is not good either. I identified ranks using excel function for each drugs (top1,2,3,4,5) and calculated proportions with 95% CI-s.

Is there a better way to check ordered ranks? Do I need agreement (not rater but...). And sometimes top2 is the toplist, sometimes more than five. Do I have to weight rank by proportion?

[Carlo Lazzaro]

Attila:
welcome to this forum.
Perhaps something along the following lines can give you some clues:

. input drug rat mice dogs

          drug        rat       mice       dogs
  1. 1 5 10 7
  2. 2 3 2 9
  3. end

. egen usage=rowtotal(rat mice dogs)

. egen order=rank(usage), field

. sort order

. list

     +------------------------------------------+
     | drug   rat   mice   dogs   usage   order |
     |------------------------------------------|
  1. |    1     5     10      7      22       1 |
  2. |    2     3      2      9      14       2 |
     +------------------------------------------+

[Marcos Almeida]

If I understood correctly, you have counts. Maybe rebuilding the data set (1 column for drug, 1 column for anomal, 1 column for counts) and applying a Poisson model would do the trick.

[Attila Kovacs]

Hi,

Thanks for the replies, yes my table was not the best so these are numbers of articles, where the given animal was used by given drug,
like drug1 was used by rats in 5 articles, by mice in 10 articles and by dogs in 7 articles
etc
And the questions is whether there is a significant difference in the most frequently used animals related to different drugs.

Regards,
Attila

[Attila Kovacs]

so using "egen order_mice=rank(mice), field" can do it for each animals, and then I can compare each ordered ranks with total with some kind of test? Or can I transpose it and do it that way, as per drugs?

[Attila Kovacs]

How about transposing it, then having ranks as above mentioned (egen rank) then having spearman by pairs or vs grand total and/or having kap d1 total, wgt(w), etc or kappa d1 d2 ..?