Su pregunta puede provenir del hecho de que se trata de Odds Ratios y Probabilidades, lo cual es confuso al principio. Dado que el modelo logístico es una transformación no lineal de la computación , los intervalos de confianza no son tan sencillos.βTx
Antecedentes
Recordemos que para el modelo de regresión logística
Probabilidad de : p = e α + β 1 x 1 + β 2 x 2(Y=1)p = eα+β1x1+β2x21+eα+β1x1+β2x2
Probabilidades de : ( p(Y=1)(p1−p)=eα+β1x1+β2x2
Log Odds of : log ( p(Y=1)log(p1−p)=α+β1x1+β2x2
Considere el caso en el que tiene un aumento de una unidad en la variable , es decir, x 1 + 1 , entonces las nuevas probabilidades sonx1x1+1
Odds(Y=1)=eα+β1(x1+1)+β2x2=eα+β1x1+β1+β2x2
- Por lo tanto, la Odds Ratio (OR) es
Odds(x1+1)Odds(x1)=eα+β1(x1+1)+β2x2eα+β1x1+β2x2=eβ1
Coeficientes de interpretación
¿Cómo interpretaría el valor del coeficiente ? Suponiendo que todo lo demás permanece fijo:βj
- Por cada unidad de aumento en la relación log-odds aumenta en β j .xjβj
- Por cada unidad de aumento en la razón de posibilidades aumenta en e β j .xjeβj
- For every increase of xj from k to k+Δ the odds ratio increases by eβjΔ
- If the coefficient is negative, then an increase in xj leads to a decrease in the odds ratio.
Confidence intervals for a single parameter βj
Do I just need to use 1.96∗SE? Or do I need to convert the SE using an approach described here?
Since the parameter βj is estimated using Maxiumum Likelihood Estimation, MLE theory tells us that it is asymptotically normal and hence we can use the large sample Wald confidence interval to get the usual
βj±z∗SE(βj)
Which gives a confidence interval on the log-odds ratio. Using the invariance property of the MLE allows us to exponentiate to get
eβj±z∗SE(βj)
which is a confidence interval on the odds ratio. Note that these intervals are for a single parameter only.
If I want to understand the standard-error for both variables how would I consider that?
If you include several parameters you can use the Bonferroni procedure, otherwise for all parameters you can use the confidence interval for probability estimates
Bonferroni procedure for several parameters
If g parameters are to be estimated with family confidence coefficient of approximately 1−α, the joint Bonferroni confidence limits are
βg±z(1−α2g)SE(βg)
Confidence intervals for probability estimates
The logistic model outputs an estimation of the probability of observing a one and we aim to construct a frequentist interval around the true probability p such that Pr(pL≤p≤pU)=.95
One approach called endpoint transformation does the following:
- Compute the upper and lower bounds of the confidence interval for the linear combination xTβ (using the Wald CI)
- Apply a monotonic transformation to the endpoints F(xTβ) to obtain the probabilities.
Since Pr(xTβ)=F(xTβ) is a monotonic transformation of xTβ
[Pr(xTβ)L≤Pr(xTβ)≤Pr(xTβ)U]=[F(xTβ)L≤F(xTβ)≤F(xTβ)U]
Concretely this means computing βTx±z∗SE(βTx) and then applying the logit transform to the result to get the lower and upper bounds:
[exTβ−z∗SE(xTβ)1+exTβ−z∗SE(xTβ),exTβ+z∗SE(xTβ)1+exTβ+z∗SE(xTβ),]
The estimated approximate variance of xTβ can be calculated using the covariance matrix of the regression coefficients using
Var(xTβ)=xTΣx
The advantage of this method is that the bounds cannot be outside the range (0,1)
There are several other approaches as well, using the delta method, bootstrapping etc.. which each have their own assumptions, advantages and limits.
Sources and info
My favorite book on this topic is "Applied Linear Statistical Models" by Kutner, Neter, Li, Chapter 14
Otherwise here are a few online sources: