Christopher Daniel
A Comparison of Missing Values Estimation Methods for Average Household Income in Canadian Census Data © 2005
This paper compares three methods of estimating missing values using the Canadian average household income census variable as a case study. The research incorporates census data from 6604 dissemination areas in the 2001 census for the Toronto census metropolitan area (CMA) but focuses on producing estimates for missing values for the dissemination areas in the Toronto census division. Building on the approach adapted by Antoniuk (1997), this research compares previously untested estimation methods in combination with some methods already used by Antoniuk. More specifically, this paper compares the use of ordinary least squares regression (OLS), inverse distance weighting nearest neighbor analysis (IDW), and geographically weighted regression (GWR) using a set of diagnostic statistics. The Pearson's correlation coefficient used as a goodness-of-fit measure indicates that the GWR method is most accurate of the estimation methods tested, although other statistics calculated suggest that the OLS method may be more appropriate in some circumstances.