3

All the time, I have believed that MR analysis can provide the information such as "there is a causal effect of X on Y", until I saw a statement on this paper (https://www.nature.com/articles/s41588-020-0631-4):

A low P value from CAUSE (or any MR method) should not be regarded as proof of a causal effect. Instead, it is an indicator that the summary statistics for the two traits are consistent with a causal effect.

I am not quite sure what does this statement mean. Does that mean even I have a p-value which is lower than 0.05, but I still cannot say "there is a causal effect of X on Y"? If that is the case, then how should I interpret the result with low p-value?

Amy Chang
  • 99
  • 6

1 Answers1

1

The authors are simply noting the general fact that statistics provides evidence, not proof.

The causal interpretation of a Mendelian Randomization result relies on a number untestable assumptions. Even when we're willing to believe that the those assumptions hold, we have to consider the possibility that our sample is a poor representation of the population.

How should you interpret a low p-value? The authors you quote say, "it is an indicator that the summary statistics for the two traits are consistent with a causal effect."

While I generally agree with the authors, I also think they are being extremely careful about claiming causal inference from statistical data. It could be a sentence added to appease a reviewer, or maybe they're just hoping to mitigate criticism and doubt from experimental scientists.

abstrusiosity
  • 876
  • 1
  • 11
  • Oh! I see. MR analysis can provides the evidence of causal effect, but cannot 100% guarantee the causal effect exists. That's the reason why the authors add this sentence. Does that what you mean? – Amy Chang Nov 09 '20 at 23:51