0
I’m using the Cox regression model, for a very large data series (I let it run overnight). And when it’s over, this message comes
Mensagens de aviso perdidas:
In fitter(X, Y, strats, offset, init, control, weights = weights, :
Loglik converged before variable 1 ; beta may be infinite.
Has anyone ever had this problem? Or knows how to solve?
Thank you!
Without the result of
summary(modelo)It is difficult to give an exact cause for this error. On the face, I would look at what were the estimated values for the Betas. I guess they got high. There, in doingexp(beta)they would get taller still, doing what the functionfitterinform you thatbeta may be infinite.– Marcus Nunes
I have an "array(NA,c(360,181,29))", which are completed in this very large loop. What even I would indicate is that it is only at some points and not at all that they are presenting this error. But even so, in this your kick, you would indicate from how much a high value? I am running again, and will save a table with all beta values, so I analyze in more detail.
– iara
Each case is a case when it comes to whether the Betas are high. If, for example, the estimate was
10, thenexp(10) = 22026. This means that the associated risk increases by 22,000 times, which may be exaggerated in this case. Since I don’t know what variable 1 is and I don’t know what risk is being calculated, it’s impossible to give an answer. If variable 1 is sex, is it possible that men are 22,000 times more likely to die than women? This is unlikely. On the other hand, if variable 1 is frequent use of heroin, perhaps the result makes sense. Information is missing for an accurate opinion.– Marcus Nunes
I am using for climate forecasts (e.g., http://www.cnpma.embrapa.br/climapest/workshop/anais/work/T050.pdf). Even so, from what I see of the generated beta values, they are between -1 and 1. I do not know, but maybe it is not in this line of 'problem'..
– iara