/
Home |

S-Archive | Download Script |

pointwise.logit |
Pointwise confidence
intervals for logit predictions |

**DESCRIPTION**- Computes predicted values and pointwise confidence intervals for logistic regression.
**USAGE**`pointwise.logit(glm.obj,newdata,coverage=0.99)`**REQUIRED ARGUMENTS**`glm.obj`glm object with `family=binomial`and default link`newdata`list holding values of the covariates for which predictions are required. See `predict.lm`for details.**OPTIONAL ARGUMENTS**`coverage`confidence level for confidence intervals **VALUE**- List with components:
`upper`upper limits of confidence intervals `fit`predicted values `lower`lower limits of confidence intervals **DETAILS**- This routine is an improvement on the S-Plus function
pointwise for logistic regression objects. Confidence
intervals are computed using a normal approximation on
the linear predictor scale, and then mapped via the
logistic transform to the probability scale. The
generalized linear model dispersion parameter is assumed
to be 1.
Note that the S-Plus function

`pointwise`produces correct predicted values but its standard errors and confidence intervals are not reliable for logistic regression, or for any discrete generalized linear model. **EXAMPLES**`glm.obj <- glm(y~x,family="binomial")`

pw <- pointwise.logit(glm.obj,newdata="x")

S-Archive | Download Script |

Gordon Smyth.
Copyright © 1996-2016. *Last modified:
10 February 2004*