• *

    This paper is based on Chapters 4 and 5 of my thesis Life-Cycle Saving, Inheritance and the Distribution of Income and Wealth in Canada presented to the University of London. I have benefited from the comments of A. F. Shorrocks, as well as the help of Gail Oja and Roger Love of Statistics Canada, Michael Wolfson of the Canadian Department of Finance, and John Whalley who provided the initial encouragement for this project. Responsibility for all errors and omissions is my own.


This paper is concerned with measurement of the size distribution of personal wealth in Canada. The only available estimates of this distribution are those provided on the occasions when Statistics Canada's Survey of Consumer Finance has surveyed assets and debts. Results of the latest “SCF” to do this, that of 1977, are not yet available. The paper shows that the previous study, conducted in 1970, indicated wealth-inequality as viewed by top quantile shares roughly of the same order as estimated by others for the U.S. and U.K. A comparison of asset and debt aggregates implied by the survey, however, with independent totals indicates that for almost all items the SCF likely under-estimated true holdings. The possible relative importance of sampling and non-sampling errors in explaining this distortion is considered, drawing on Monte Carlo evidence and American validation studies of survey response. It is concluded that sampling error is unlikely to provide the explanation for SCF discrepancies in aggregates, but that non-sampling error is capable of doing so. Finally the 1970 SCF distribution of wealth is re-estimated. First a correction is made for hypothetical differential response according to true net worth. Second an attempt is made to remove the effects of under-reporting by respondents. The “best-guess” re-estimated distribution exhibits mean net worth considerably greater than shown by the SCF but only a slightly greater degree of concentration. Under certain fundamental assumptions this result is surprisingly robust. The appropriate conclusion is not that survey estimates of the distribution of wealth are reliable, but that the strong non-sampling errors affecting the 1970 Canadian SCF wealth estimates may have been composed of almost completely offsetting sources of bias.