search

josefin-werme / LAVA

51 stars 9 forks source link

Can rg be 1? #69

Closed dqq0404 closed 6 months ago

dqq0404 commented 6 months ago

Hi, I got a result which include r2 = 1.000000, r2_lower = 0.45880 ,r2_upper = 1,p = 6.02127e-07 Is this result right?

cadeleeuw commented 6 months ago

Yes, it is indeed possible for the (estimate of) the rG to be 1. The estimator of the rG itself is unbounded and as such can also exceed 1 (or -1). If this is exceeded by too much (by default, past +/- 1.25) it will be assumed that something is wrong with the estimate and it is set to NA. If it remains within those bounds however, it is just truncated to 1 (since if the true rG is close to 1m, then it is not uncommon for the raw estimate to be somewhat above 1 just due to stochastic noise).

Thus, an estimated rG of 1 can occur either a) if there is a large amount of noise making the estimate unstable, and it happens to end up above 1 (but not by so much that it exceeds that 1.25 bound). In that case the p-value will also be very high though, so this is clearly not the case in your example. The other possibility is that b) the true rG is genuinely very high, which seems to be the case in your result here. It is of course quite possible that the true rG isn't exactly 1, as the confidence interval also indicates, ultimately what the analysis gives you is still just an estimate. But it is still likely fairly close to 1.

dqq0404 commented 6 months ago

Thanks!!!

---- Replied Message ---- | From | @.> | | Date | 04/12/2024 17:46 | | To | josefin-werme/LAVA @.> | | Cc | dqq0404 @.>, Author @.> | | Subject | Re: [josefin-werme/LAVA] Can rg be 1? (Issue #69) |

Yes, it is indeed possible for the (estimate of) the rG to be 1. The estimator of the rG itself is unbounded and as such can also exceed 1 (or -1). If this is exceeded by too much (by default, past +/- 1.25) it will be assumed that something is wrong with the estimate and it is set to NA. If it remains within those bounds however, it is just truncated to 1 (since if the true rG is close to 1m, then it is not uncommon for the raw estimate to be somewhat above 1 just due to stochastic noise).

Thus, an estimated rG of 1 can occur either a) if there is a large amount of noise making the estimate unstable, and it happens to end up above 1 (but not by so much that it exceeds that 1.25 bound). In that case the p-value will also be very high though, so this is clearly not the case in your example. The other possibility is that b) the true rG is genuinely very high, which seems to be the case in your result here. It is of course quite possible that the true rG isn't exactly 1, as the confidence interval also indicates, ultimately what the analysis gives you is still just an estimate. But it is still likely fairly close to 1.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>