Question: At what point does does a departure from the normal constitute inflation (QQ plot)
0
gravatar for Alexander Skates
3.7 years ago by
United Kingdom
Alexander Skates340 wrote:

I'm trying to interpret a QQ plot I have from Matrix eQTL, to see if there is a factor I haven't accounted for. I know a departure from the x=y at low values can indicate a confounding factor, but at what point is a digression from the normal inflation, and when is it acceptable? Is there a "rule of thumb"?

Edit: uploaded the wrong QQ-plot:

eqtl qq-plot • 2.7k views
ADD COMMENTlink modified 3.7 years ago by Jean-Karim Heriche18k • written 3.7 years ago by Alexander Skates340
1

I refer to R-fortunes:

library(fortunes) 
fortune('chicken') 

A sufficiently trained statistician can read the vagaries of a Q-Q plot like a shaman can read a chicken's entrails, with a similar recourse to scientific principles. Interpreting Q-Q plots is more a visceral than an intellectual exercise. The uninitiated are often mystified by the process. Experience is the key here.
   -- Department of Mathematics and Statistics, Murdoch University
      StatsNotes

But because I don't want my comment to be completely useless, this thread might help you, in particular the latest 3 replies. Failing that, I am inclined to agree with Jean-Karim in that visually your plot suggests non-normality and normal distribution can be tested with the mentioned tests (though I have only used the Shapiro-Wilk test). 

ADD REPLYlink modified 3.7 years ago • written 3.7 years ago by A. Domingues2.0k

That is a fantastic quote.

ADD REPLYlink written 3.7 years ago by Alexander Skates340
1
gravatar for Jean-Karim Heriche
3.7 years ago by
EMBL Heidelberg, Germany
Jean-Karim Heriche18k wrote:

I don't know of any rule of thumb for deciding when a distribution is normal or not based on a QQ plot. Any visually strong departure from the diagonal line like in your plot is definitely a sign of non-normality. However, even small deviations from the line can indicate non-normality but in these cases normality can be a good approximation. If you need statistical motivation for deciding whether your data is normally distributed, there are a number of tests you could use, the most sensitive being Shapiro-Wilk and Anderson-Darling tests. 

ADD COMMENTlink modified 3.7 years ago • written 3.7 years ago by Jean-Karim Heriche18k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 2123 users visited in the last hour