1ZM31 - Pagina 2

Pagina 2 van de samenvatting

causation (proving)

experimental setup, longitudinal data, good theory

Missing values (options)

MCAR, MAR, NMAR

MCAR

Cells are missing completely at random, confirm using Little’s MCAR test / 2-sample t-test

What if MCAR

ignore missing values / imputation (mean substitution)

MAR

Missing cells depend on another cell, confirm using Little’s MCAR test / 2-sample t-test

What if MAR

use ML / Multiple imputation

NMAR

Missing cells have a distinct pattern, confirm through domain knowledge

What if NMAR

gain additional / new data

outlier (meaning)

case that is very different from other cases, indicate error in data collection

outliers (consequence)

disproportionate influence on statistical analyses

univarite outliers rule of thumb

assume normal distribution, calculate SD per case, if |SD| > 3 = outlier

multivariate outliers rule of thumb

use Mahalanobis Distance (D-square), or boxplot

Mahalanobis distance (D-square / MD)

determines center of data and draws region

Measurement error

degree to which observed values are not representative of true values

Validity

degree to which measure accurately represents what it is supposed to

Reliability

degree to which observed variable measures true value (error free)

multivariate measurement (summated scales)

use of 2 or more variables as indicators of single composite measure

indicator

single variable used in conjunction of variables

multicollinearity

degree of correlation among variables of variate

dimensional reduction

finding combinations of individual variables that captures multicollinearity among variables and allows for 1 single construct