PARTIAL ODDS RATIO ESTIMATION

Share:

Description

Estimates odds ratio adjusted for a vector of covariates

Usage

1
partialOR(dd,ci=0.95)

Arguments

dd

Data frame with binary 0/1 response variables x,y and covariates z1,...,zm (in that order)

ci

Confidence level (default ci=0.95)

Details

partialOR() estimates the adjusted odds ratio OR(X,Y | Z1,...,Zm) between two binary variables X and Y adjusted for a vector (Z1,...,Zm) of m numerical covariates ("confounders"). It is based on fitting a multinomial logistic regression model to the data. In this model the categorical response variable corresponds to the four possible outcomes of (X,Y), namely (0,0), (0,1), (1,0) and (1,1). The program fits the null model (without covariates), the full F-model and the H-modelwith parameters restricted by the hypothesis of homogeneity of odds ratio's. The homogeneity hypothesis is tested by comparing the two models by the Likelihood Ratio test. The program reports OR estimates under the respective models, the standard errors of log(OR) and confidence intervals. Note: to include categorical covariates the user has to transform them into dummy variables.

Value

The program prints information about the convergence of the optimizer, the model deviances, the LR-test and the adjusted odds ratios. It calls the function fitOR() which, when called separatelly, returns detailed information on model fitting.

Author(s)

Vaclav Fidler and Nico Nagelkerke

References

Fidler, V. and Nagelkerke, N.J.D. (2012) The Mantel-Haenszel procedure revisited: models and generalizations. Submitted.

Examples

1
2
3
4
## simulate data from the H-model 
dd <- simData(n=50, m=2, rr=1.5, rseed=123) 
## estimate the OR
partialOR(dd)