Hello, I have the following setting to an experiment:
There are N objects: Obj_1 can have the value either "red" or "black". Obj_2 can have the value either "1" or "2". Obj_3 can hold value "yes" or "no"... etc.
There are also M users. Each user picks k objects (as many as he wants, k < N) and asigns a value to them (k is not the same for all users).
I need to measure how diverse are the values given for each object (and overal) and somehow graphically depict it.
Ok these sounds like one distance or divergence to me. But the problem is tnat you do not have k=N and each of the M users chooses as many objects as he wants. But, listen an idea.
For any two users see how many questions they have in common. Say from 10 questions the first user responded 5 and the other user 3. If 2 of them are common say for these 2 in common what were their answers? I understood the responses are only 0 or 1, dichotomous. Then for every match assign 1 point, for every mismatch assign 0 points. Sum it up and divide by the common questions they have, for instance 3 in my example. If you do this for all users you end up with an MxM matrix full of divergences from one another. This is the idea for the divergences or distances between any two users. If you know SPSS this can bedone automatically and you get a plot also.
I think yes. It seems OK.
I want to somehow measure and represent the agreement between the users overall.
So, the closer to zero are the values on every object, the more users agree on their options.
Then I will plot all of them (any way to do that besides standard deviation ?) to have a general idea.
My knowledge on statistics is little, any deeper analysis or misconceptions on what I say would be appreciated
your knwoledge is little but your opinion is big. I told you my opinion. You think you are OK, I do not know what is it that you are trying to prove with your idea.
You don't need to be rude. I read your suggestion and I thank you for that. I just can't figure out how the plot will look like at the end. I don't use SPSS, I do my plots on matlab or excel.
My approach is more "object oriented". It measures the disagreement on every object and I wanted an opinion whether this is correct. Anyway.
Reply With Quote
