attached is a (German...) bar chart which shows the five possible values of the one variable "Culture": Nix, BV, Candida, Mix, Unspecific.
The variable is measured at three different times, 1 (blue), 2 (green), 3 (something).

I am looking for some kind of test to determine that the number of measured individual values increase, or decrease. For example, I'd like to show that over time (i.e. from time=1 to time=3), the value "Nix" is measured more often, whereas "Candida" and possibly "Mix" go down.

I am a bit confused because we have three time points and five possible values, this makes it a bit unclear for me.

Bar chart

  • $\begingroup$ There is probably some information missing in your explanation: it seems unlikely that the absolute counts make sense over the 3 measurement times. Don't you want to compare (over the measurement times) the ratio of each value (i.e. the count divided by the number of items counted on that measurement time). In any case there are bound to be relations between the values that you don't specify. If not, you'll have to make grave assumptions to "assert" going up per value if you only have 3 numbers. $\endgroup$ – Nick Sabbe Aug 24 '11 at 11:02
  • $\begingroup$ Do I understand the data right in that you have 3 variables (columns time1, time2, time3) with a common list of 5 possible categorical values (Nix, BV, etc). Each row of the data is an observer; at time1 he observes, say, Nix, at time2, say, BV, at time3, say, Nix again. If that is the case than why does amount of observations differ for 3 times (50 in time1, 44 in time2, 41 in time3)? Is this because of missing data? Or your data is another structure? $\endgroup$ – ttnphns Aug 24 '11 at 11:10
  • $\begingroup$ ttnphns: Yes, there are some patients who apparently went missing while the study was executed. $\endgroup$ – Alexander Engelhardt Aug 24 '11 at 11:50
  • $\begingroup$ Nick Sabbe: Wait, you are right. There are more missing patients as time goes on. I should use ratios. What exactly do you mean with "relations between the values"? $\endgroup$ – Alexander Engelhardt Aug 24 '11 at 11:58

One simple test that at once come to my mind is marginal homogeneity test (it is available in SPSS: it seems to me, by the looks of your chart, that you were using SPSS). This is a repeated-mesures test (I understood from your comment that you have a total sample of 50 patients X 3 variables (times), but it will compare you only 2 variables at a time, not all 3 at once. It will tell you whether the frequency distribution of 5 categories is the same in both times. Thus, it is not category-by-category test but the simultaneous test for all the categories. Missing data is the problem: you should either delete it (then your sample will be 41 instead of 50) or fill in randomly.

Of course, other approaches and more complex analyses are possible with your data.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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