Non-Parametric Analyses. Report medians and 95% CI in graphs?

Mimi

New Member
#1
Okay so,

I'm currently writing my thesis, and I have to do non-parametric tests. I know that I have to report the medians in text. However, I'm doing graphs. Just wanted to ask whether anyone knows if I report the 95% CI and the medians in the graphs? When i do this it looks super weird and doesn't really reflect my hypotheses (i.e. the CI on the graph makes it look like things are non-sig when they actually are).

Was hoping anyone could shed light on this.

Thanks,
Mimi
 

gianmarco

TS Contributor
#2
Can you describe your data a little bit, and what test you performed?
If you used Mann-Whitney or Kruskal-wallis, it may happen that you can have significance difference sport ed by the test and yet similar medians.

Gm
 

noetsi

Fortran must die
#3
Medians are just one of many approaches in non-parametrics. The use of pairs is at least as common or ranks. Regardless, advice I always give someone (form my own painful experience) is when you work on a thesis you should ask your advisor (and comittee) their views on this.

I don't think it is required to report CI even with mean tests let alone medians. Test of statistical signficance are more common from my observation.
 

bugman

Super Moderator
#4
You could consider plottin your data as boxplots. This will show the median, and other useful features of your distributions.
 

gianmarco

TS Contributor
#5
Just to supplement Bugman's reply, you could also use notched boxplots, to visually represent the 95% CI for the medians. But the problem the OP describes would linger. That's why, first, I would like to have some info about the data themselves and about the test used.