URSI 403--Doing Urban Research -- Lecture Notes 8
VII.Parametric Tools
A. Correlation & Multiple Regression
- standardized covariance (covariance expressed in terms of standard deviation units in x and y)
- generally, interval measures required (may use ordinal if underlying interval relation can be assumed)
- correlation may be positive, negative, or zero
- square of correlation coefficient may be interpretted as PRE (proportional reduction in error) measure
- measure is fairly robust (tolerates violation of assumptions)
- highly affected by extreme values (outliers)
- correlations may differ, even if slope does not
- does not indicate causation: spurious correlation is possible, even likely
- multiple regression extends correlation to multiple variables with additive effect on dependent variable (y=a=bx1+bx2...)
B. Factor Analysis
- orderly simplification of interrelated variables
- uses correlation to group variables into common (highly inter-correlated) factors
© 1996 A.J.Filipovitch
Revised 1 September 96