matching can improve precision (efficiency) but it is not a measure to achieve or improveodds = proportion/1-proportionproportion = odds/1+oddsodds = e^b0+b1x so p = (1-p)*e^b0+b1x?Se+Sp = 1 = nullSe+Sp = 0 = inverse of real estimateSe+Sp = narrower confidence intervals, less study power, but still valid hypothesis testsGenerally more efficient than stratificationâ€basedmethods when data are sparsePrediction of a continuous outcomesSpecify continuous and categorical predictor variablesSpecify interactionsModel nonlinear relationships between exposure andoutcome and other covariateswe assume the log function is a linear relationshipLogistic regression uses a maximum likelihood method, which maximizes the probability b0+b1x + eb0 = log(odds) when x=0; b1 = log(odds ratio per unit change in x holding all other xs conThe odds of X among those with Y =1 is OR times greater thanthe odds among those with Y = 0, assuming no sources of bias.Unified: weighted average of stratum-specific measures of association similar to MH or inTraditional: standardize the rate in the exposed and standardize; calculate the ratio or dPr[Y a =1|A=1,L=l]=Pr[Y a =1|A=0,L=l]In a conditionally randomized experiment, within levels of stratification factor L, exposed unconditional: if the treated had been untreated, they would have been expected to haveconditional: same average outcome expected within each level of the covariates Lcohort = match exposed group to unexposed cohort groupcase-control = cases matched to controlsmatching --> exposed ---------------------->disease------------------------> selection

89238836a347e7e6e4bfddd6ab949c044ea64996Page 8the pairs for which a case is exposed and the control unexposed, or the case is unexposthose pairs where both the case and the control are exposed or both are unexposed - in appropriate: bias when not stratifiedunnecessary: more precise when not stratified but precision persists either wayover-matching: precise when stratified but biased when not stratifiedmatching on intermediate: only precise when not matched and not stratifiedtowards the null because matching makes the cases and controls more similar to each o

#### You've reached the end of your free preview.

Want to read all 8 pages?

- Summer '14
- FrancisCook