search

rondolab / MR-PRESSO

Performs the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) method.
81 stars 30 forks source link

Significance Threshold #6

Open jonnyrose opened 5 years ago

jonnyrose commented 5 years ago

Am I not capturing more low-level pleitropic effects if I increase signifThreshold? What would be the downside of say a signifThreshold > 0.1 or more, when I find no significant outliers at lower thresholds (ie default threshold 0.05), yet still my global test p-value is statistically significant. Thanks for any help and insight you can give me.

rondolab commented 5 years ago

Hi, I guess you could think that you are capturing more low-level pleitropic effects by increasing signifThreshold however you are more prone to false positives. As shown in the paper, the global test is well powered to detect a wide range of horizontal pleiotropic situations whereas the outlier test works only when < 50% of the instruments are submitted to horizontal pleiotropy. It is possible in your case that the global test detects some horizontal pleiotropy without the outlier test being able to pinpoint those pleiotropic outliers. I would not recommend to increase the threshold in that specific case. Best.