track data across quartile groups (if that's even a thing)

[PhD_student_needs_help]

Hello. I have responses to a questionnaire about fruit and veg consumption, and basically I want to look at how each individual's responses group into quartiles across each category.

The questionnaire consisted of about 100 questions, but I have summarised them into categories and calculated scores according to the response (and hence it is all quantitative data), e.g. Positive attitudes towards fruit & veg; frequency of F&V consumption; diversity of F&V consumption. I would like to analyse whether people are consistent across the categories, i.e. are those in the highest/lowest quartile for attitudes towards F&V also in the highest/lowest quartile for frequency and diversity of consumption. I have looked at the data on excel and colour coded each quartile for each category, so I can see visually that some people track consistently within a quartile, and others less consistently. But I would like a more mathematical way of doing this.

Any advice would be greatly appreciated! I've been searching all day and come up with nothing.

Thanks
Jen

 

[Stats fan]

Maybe Spearman's rho for the relationship between two ordinal variables can be useful here.

 

[PhD_student_needs_help]

Maybe Spearman's rho for the relationship between two ordinal variables can be useful here.

Thanks.
I've already used Pearson's correlation and all the categories are significantly correlated with each other. I don't want to look at the category as a whole however, but rather how each person sits within each category across the whole data set. I'm having trouble explaining it.

So the three categories I mentioned before, the first has a minimum score of -36 and maximum 36; the second has a min score of 0 and max 760; and the third has a min of 0 and max 38. So I want to look at whether someone who scored highly for their attitude to F&V also scored highly for frequency and diversity of consumption. I can see it visually in excel, but I don't know what statistical test to run to represent it numerically.

I probably still haven't explained it very well.

 

[Stats fan]

One the one hand, it sounds like all variables are ordinal (four ordered quartiles), and you want to know whether people scoring in lower quartiles on one variable, also score in lower quartiles on the other variable (and same for higher quartiles). Spearman's rho would be a useful measure in this case. On the other hand, you also say something about minimum scores and maximum scores in three different categories. I can't fit this information in the rest of the information you give. Can you be more precise about the measurement level of each of your variables?

 

[PhD_student_needs_help]

One the one hand, it sounds like all variables are ordinal (four ordered quartiles), and you want to know whether people scoring in lower quartiles on one variable, also score in lower quartiles on the other variable (and same for higher quartiles). Spearman's rho would be a useful measure in this case. On the other hand, you also say something about minimum scores and maximum scores in three different categories. I can't fit this information in the rest of the information you give. Can you be more precise about the measurement level of each of your variables?

Ah, so are you suggesting that I actually assign them a number 1-4 depending on which quartile they fall in to, and then do Spearman's rho?
I'll have a play about with that and see what comes up.
Thanks!

 

[Stats fan]

Maybe I get it now. You have three variables (which you call categories, that got me confused), the first ranging from -36 to 36, the second from 0 to 760, and the third from 0 to 38. You have transformed each variable into an ordinal variable consisting of four quartiles. Now you want to know if people scoring in lower quartiles on one variable, also score in lower quartiles on the other variable (and same for higher quartiles). I am still going for Spearman's rho ;-)

 

[Stats fan]

Ah I hadn't seen your last message when I posted my new response... Yes that is what I was suggesting. I assumed you had already done this

 

[PhD_student_needs_help]

YES!!!! You have explained it perfectly. Thank you! I will try soearman’s rho.
Can you tell stats isn’t my forte?!

 

[GretaGarbo]

Well, I believe that if you round a variable (like -36 to 36 to 1,2,34) you will lose a lot of information. First you collect information. Then you throw away a large part of it.

(It would be nice if someone did a simulation study about this <--- suggestion!)