eg 7 year cumulative incidence Bias you got the answer right eg selection biase

# Eg 7 year cumulative incidence bias you got the

This preview shows page 6 - 8 out of 8 pages.

e.g. 7-year cumulative incidenceBias = you got the answer right (e.g. selection biase)Efficiency is the study/study design with the smallest varianceThe difference between the expected value of anestimation procedure and the true value that theprocedure is attempting to estimate. If this difference is zero, the estimate is unbiased.aka: the nullequal weightsstratum-specific person-timeinverse variancemantel-haenszelwhen the magnitude of the association between exposure and disease varies according and intrinsic phenomenon that cannot be eliminatedÂ from a study through clever desigHo:The rate ratio is the same across all I levels of the stratification variable(s)Ha: The rate ratio is not the same across all I levels of the stratification variable(s)H = test statistic (DF = # of strata - 1)1. they fail to summarize the data with respect to their consistency with any alternative h2. they do not give us any indication of the power of the data to detect any alternative hy3. a failure to reject the null hypothesis of no effect measure modification may often be eError in categorical data (either exposure or outcome)not to be confused with measurement error
89238836a347e7e6e4bfddd6ab949c044ea64996Page 7misclassification occuring in the same proportion in each group; equal amount of misclasmisclassification occurs in different proportions in each group; leads to a bias in either di1-sensitivitythe basic aim is to draw the least square line among the values of the outcome variable; 1-specificitypositive predictive value = TP/(TP+FP)negative predictive value = TN/(TN+FN)if all random variables in the sequence or vector have the same finite variance. - linear relinear, logistic, cox, poissonAllows the prediction of a continuous dependent (outcome) variable y from a continuous allows the analysis of dichotomous or binary outcomes with 2 mutually exclusive levels; pwhen analyzing binary outcomes (remember the s-curve graph); Linear regression assuma way of accounting for effect measure modification by creating weights that reflect the da comparison of crude rates can be misleading because the comparison can be biased dstandards are chosen to make inferences about a population with the distribution in eitheobserved rate in exposed/ expected rate in the exposed had they been unexposed - the if the unexposed population had been exposed, the outcome would have been x times thIPW = inverse-probability weighting; consistency, exchangeability, positivity, and no missfind stratum-specific probabilities; take inverse of the (A|L)-level probabilities and multiplythe pseudo-population created by IPW looks like a marginally randomized experiment; ethey are all conditionally exchangeable; there is no unmeasured confounding within leveMost useful in small studies with several confounding variables, matching blocks the patyes; we can evaluate the association between matching factor and exposure or outcomeimproves precision but does not eliminate bias; abolishes any association between matc

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

Want to read all 8 pages?

• Summer '14
• FrancisCook