MULTIVARIATE MODELING

mulivirateMultivariate regression models are typically used to predict or forecast an outcome (dependent) variable using multiple independent predictors or confounding variables.

• If outcome of interest is continuous quantitative variable, regression of choice is multivariate linear and for binary dependent outcome, multivariate logistic regression is the most popular technique of analysis.

• For discrete count data as outcome variable, poisson regression is method of choice.

Importantly, correlated outcome data is analyzed by linear mixed modeling for continuous variable or generalized estimating equations for binary or non-normally distributed dependent outcome.

• Other variations in regression analysis include robust regression (adjusting for extreme outliers), multinominal regression (outcome has multiple category), exact regression (for small sample size with few empty cells), conditional regression (for matched pair data), probit, ordinal and quantile regressions.

• Among survival or hazard analysis, Cox proportional hazard analysis, competing risks regression and Cox regression with time varying covariates are most commonly applied hazard models.