Reply With QuoteDoes anyone have any ideas how to approach this?
Ok so does everyone agree it is going to be a non-parametric problem?
If so how about using Friedman test?
Suggestions?
C2
Ok I will admit I am getting old, and out of practice so those graduate level classes might not have been useful :-)
I have 3 data sets. Within each set individual people ranked 12 photos 1-5 and 12th. So each individual within the set has 6 responses. Unfortunately the # of individuals per set is uneven.
I would like to compare the rankings between each set (3) and I would like to develop an overall ranking per set.
Anyone think I am making sense?
C2
Does anyone have any ideas how to approach this?
Ok so does everyone agree it is going to be a non-parametric problem?
If so how about using Friedman test?
Suggestions?
C2
Wondering if I am oversimplifying this question or if I am just not asking it right? Anyone have a suggestion regarding another site that would be able to give advice regarding this?
Id even pay someone at this point to run the stats
Maybe you should adher more to this:Originally Posted by Flashpoint
http://www.talkstats.com/showthread.php?t=185
People can impossibly help you if they dont know what you want.
The true ideals of great philosophies always seem to get lost somewhere along the road..
you don't think I gave enough information? I pretty much thought I adhere to those rules when I posted originally, ?
"I have 3 data sets. Within each set individual people ranked 12 photos 1-5 and 12th. So each individual within the set has 6 responses. Unfortunately the # of individuals per set is uneven.
I would like to compare the rankings between each set (3) and I would like to develop an overall ranking per set."
I had asked if it made sense figuring if it wasn't enough information someone would ask more specifics, so at least point me in the right direction of locating a proper test then I could come back and ask more questions regarding the specific test. Its the RANKING that is throwing me off. How to you account for that fact they are not a measurement but rather a rank (dependent on each other) and that rank is an arbitrary number not an actual measurement
Basically there are 3 data sets of groups of people. They are grouped by profession.
within each Profession individual people ranked 12 photos #1-#5 1 being the best thru #5 being the least best of the 5. They then ranked the overall worse of the 12 (= #12 in ranking). Each individual within a profession is independent. The photo rankings are obviously dependent upon each other ( 1 is better than 2 that is better than 3 that is better than 4 that is better than 5).
Hypothesis:
Each profession ranks the photos (1-5 and #12 (worst) ) the same.
or stated another way
Profession has no effect on the ranking of the photos.
Then if you remove groups (profession) and just evaluate the rankings overall what is the most highly ranked photo overall, and which is the least.
Am i getting anywhere with making more sense?
I'm not going to debate any rules with you as it is not your task to interpret them. You should show decency and respect, people are helping you for free. You wish people to reply to your questions? Then you should also reply to theirs. Suffice to say 'those rules' are guidelines meant to prevent people from posting cryptic statements like:
"I would like to compare the rankings between each set (3) and I would like to develop an overall ranking per set"
Consequently, you end-up elaborating mostly on what was already clear in your first post (though from your Null hypothesis I understand what you want).
There is a whole Scala of statistics devoted to preference testing, usually using ranking. Your problem is not unique. Here some studies to help you on your way:
http://www.sciencedirect.com/science...1fe6860ab88db0
http://www.jstor.org/pss/2985754
http://www.joe.org/joe/2003october/tt4.php
Also look for 'consumer preference', and studies evaluating the preferences of handicapped individuals.
[This is a long standing but still a tricky statistical field though]
Alternative: use a monte-carlo simulation to evaluate the distribution under your null hypothesis, this will help you evaluate whether your actual result is deviant.
The true ideals of great philosophies always seem to get lost somewhere along the road..
Wow that amount of hostility from you leads me to believe this isn't a FRIENDLY forum. Sorry you felt it was CRYPTIC but PERHAPS some of us ARENT SO WELL VERSED in Statistics as you. Oh wait maybe that is WHY we are going to an online forum for help. I didn't see any "disrespect" in my original post, lack of understanding maybe but disrespect? I think it is YOU who is disrespectful.
Unbelievable, I thought this was a question and answer forum to get ideas, discuss topics, get help, go through the progress, sort of like "over a cup of coffee" idea. Apparently not.
Oh wait after all the rhetoric from you regarding the questions you get down to actually discussing the problem at hand,
I'll begin by looking up those web sites. Thank you
Please do try to refrain from these comments Flashpoint. I simply asked you do give more information as specified by the posting guidelines. Your description of your study goals was cryptic and so cryptic that someone who you state is 'WELL VERSED' in statistics was not sure as to what you wanted to do. The lack of information was evident as your post remained unanswered after 3 bumbs! Thus it was my opinion that your post did not give enough information. The usual cause: people don't read the guidelines.. so we usually point it out.
Then instead of simply giving the information and being grateful that someone pointed this out. You question that person (a moderator) on his interpretation of the posting guidelines, implying that you did give enough information. A strange reaction to help to say the least. Now if that was not your meaning then you should have simply pointed that out, and you would have gotten an apology.
After helping hundreds of people on this forum I have not come across someone so quickly offended. We are trying to help people on this forum, we expect them to be respectful and appreciate the help. This is a friendly forum with thousands posts to prove it. In the future refrain from questioning people who help you.
I hope this end this, and that you can solve your statistical problems.
The true ideals of great philosophies always seem to get lost somewhere along the road..
your help was in the form of "maybe you should follow the rules"?
Somehow I didn't quite see that as "help" but rather admonishment of not following the rules.
Did you ask me any questions or suggest I state the hypothesis which apparently is what you were looking for, nope it was
basically
"read the rules"
that isn't a friendly discussion that is a hostile post.
first you accuse me of not following the rules and then having a discussion regarding the rules? UNBELIEVABLE this isn't GOD land, this is a stats forum! the number of replies per post is abysmal anyway with typically 0 to 2 per post. I had hoped to start a friendly discussion of this particular type of data analysis to develop a better understanding of the concepts behind the stats.
Your suggesting a Wilcoxon signed-rank test not Friedman?
I believe the potato test is probably the closest to my specific problem. Unfortunately I no longer have access to SAS so difficult to run it. Any suggestions of an online Wilcoxon calculator?
I'm now guessing we probably got off on the wrong foot. Miscommunication all the way, though I don't really agree with your reaction, I do now see where the misconceptions could of come from. Now I suggest we continue.
No, but the basic Wilcoxon method (first link I believe) should inform you on the theory behind preference test. I advise you to start your search for a solution there.
Friedman? Your situation is indeed a multiple groups problem though I didnt see that you have a repeated measures problem from your previous posts. Do you have repeated measures?
If not a Kruskal-Wallis may suffice. The only problems is (as I read from the info given) that you don't have a ranking for all photos (from 12 pics you have 6 rankings and you discarded the rest, atleast thats what i get from your post). This could be a problem. I have thus far only worked with preference test where the ranking was complete. This setup makes the test considerably more tricky. Though I know that sometimes in consumer tests not all 'preference subjects' (food types, car colors ect) get selected so you should be able to find solutions if you go through the literature.
My first step would however be to see how my null-hypothesis behaves under simulation. What do you want to know, that the median/mean rankings differed per group??? Though that only makes sense when you do the test on a per photo basis.... or are you looking for differences in the distribution of rankings over the 12 pictures?
Do you have the ability to run simulations? If you can work with SAS, then R is not that far off. R might be the tool to do this.
www.r-project.org
Additionally how many people in each of the 3 groups?
The true ideals of great philosophies always seem to get lost somewhere along the road..
Sorry for the delay i am out of town on business and wont be back until Monday. Hopefully I'll get an opportunity to look at the information and I am sure I'll have more questions! Thank you again for the helpful hints!
Posting Permissions
|
|