Applied biostatistics Francisco Javier Barn Lpez Dpto Medicina

Applied biostatistics Francisco Javier Barón López Dpto. Medicina Preventiva Universidad de Málaga – España baron@uma. es 1

Multivariate analysis n Generally used to study: ¨ the effect of one variable n Numerical n dichotomous, or n qualitative by using multiple binary variables. ¨ On another variable n n ¨ Numerical: Multiple linear regression Binary: Logistic regression Controlling for the effect of a few other variables n n n Control variables Covariates Confusion 2

The usual multivariate model in Health sciences Interesting variable Covariates Multivariate model Outcome age sex … Does [interesting variable] influence [Outcome variable] when [adjusting/controlling/taking into account] covariates 1, covariates 2, …? 3

Numerical outcome: Multiple linear regression Interesting variable Covariates Estimate±std. error; p Estimate; CI 95%; p age Multivariate model: Linear regression model Numeric outcome sex … • Estimate>0, Increasing effect • Estimate<0, Decreasing effect • Estimate=0, No effect We are NOT (very) interested in the significance of covariates. 4

Binary logistic regression Interesting variable Covariates OR; p OR; CI 95%; p age Multivariate model: Linear regression model Binary outcome 0/1 sex … The estimates now are: Odds Ratios (OR) • OR>1, Increased risk • OR<1, Decreased risk • OR=1, No effect 5

Dummy variables Qualitative interesting variable with 3+ levels Covariates Multivariate model Outcome age sex … We must encode the qualitatives non binary variables using only binary variables. How? 6

Encoding dummy variables Categoria laboral • Administativo • dummy. Seguridad=0 • dummy. Directivo=0 • Seguridad: • dummy. Seguridad=1 • dummy. Directivo=0 • Directivo: • dummy. Seguridad=0 • dummy. Directivo=1 7

Dummy variables Qualitative interesting variable with 3+ levels Multivariate model Dummy 1 … Dummy 2 … Covariates Outcome Multivariate model Outcome … Coding qualitative variables using dummy variables 8