quote:
Hello Nancy
Follow to our previous descussion regarding outlier,I have to tell you that the point at -200, -.06 is not the same person as 260,0.8.I have to state that after removing those two subject plus the other subject which sounds to be an outlier in the result of analysis without the first two subject, the result was exactly the same.
Now,I use the latest version of PLS to plot brain score with CI.I believe that the error bars for red and pink one looks strange and those conditions are exactly the same that contains point -200,-0.6 and 260,0.8 respectively.You can see the brain score with CI through:
ftp://ftp.physiol.umu.se/fysiologi/out/Alireza/
Is this a confident prove that those two subjects are outlier or .....?
Thanks,
/Alireza
Hi Alireza...
this is very interesting.... FYI: we are just now starting to provide Confidence Intervals for the DesignLV in TaskPLS..
First. What this indicates is that for those two conditions, you likely have a very skewed distribution for the DesignLV. We frequently see this behaviour in the CI in BehavPLS results. To help with interpretation in those instances, we have automatically included an 'adjusted' upper and lower confidence level, where we attempt to minimize the effect of the skew in the bootstrap distribution. These are boot_result.ulcorr_adj and boot_result.llcorr_adj. For the TaskPLS, these variables are boot_result.llusc_adj and boot_result.ulusc_adj (or something similar) . If you look at a barplot of the adjusted values, the CI should be more symmetric for those two conditions.
Second. I am not surprised that your results were the same after removing those two subjects.. they do not look like huge outliers.
Third. This skew in the sampling distribution of DesignLV values is not likely to be due to those two subjects
hope this helps,
Nancy