Introduction

The relationship between individual household income and the mean (or median) household income as defined using the smallest possible census areas is an issue which frequently comes up. Individual household income is not typically available in administrative data files and yet researchers wish to use household income, a variable shown to be important in many analyses (often using survey data).

This problem has been approached in two ways:


1. by presenting the relationship between individual and census area categories of income.
   
   The following table compares actual household income quintile (from the census) with MCHP small area income quintiles calculated from the 1986 census linked sample, including only Winnipeg residents (N=22185). The sample is described in Mustard et al. (1999) . The MCHP macro is based on postal codes. The following table displays the agreement on urban quintile rank (the percentages sum to 100% within MCHP ranks).

   MCHP Rank

   Household Income Rank (from Census)	U1	U2	U3	U4	U5
   U1	45.5%	24.8%	19.6%	11.1%	5.8%
   U2	23.8	27.8	24.1	18.5	10.2
   U3	16.4	21.6	21.5	22.1	17.7
   U4	9.1	15.2	19.4	26.7	25.1
   U5	5.2	10.7	15.4	21.7	41.2
   Total (weighted N)	100.0 (98087.2)	100.1 (88519.8)	100.0 (112212.7)	100.1 (130239.7)	100.0 (150564.8)

   Spearman correlation coefficient	0.435 (p<.0001)
   Exact match on quintile	190226
   Percent matching exactly	32.8%
   Matching within one quintile (includes exact match)	398139
   Percent within one quintile	68.7%

   
   
2. by using ecologic-level measures of income as proxies for individual-level measures
   
   Mustard et al. (1999) examine the effectiveness of using ecologic measures of income in predicting differences in population health when individual-level measures of income are unavailable. The study is based on a representative sample of 5% of the Manitoba population in 1986 (N=47,935). From the combination of information from administrative records of individual health care use, records of death, and 1986 census records containing information on household income and average neighborhood income, thirteen measures of health status were developed. The results from the study suggest that risk estimates from ecologic income measures are not weakened relative to estimates obtained from household income and provide evidence for the use of ecologic-level measures of income as proxies for individual-level measures (Mustard et al. (1999) ).

Limitations

This area is a controversial one, with studies conflicting as to the extent to which aggregate data provide a reasonable approximation of individual-level data. A list of studies (and a differing point of view) is provided by Southern et al. (2005) .

Differences among studies may be due to:


1. the accuracy with which neighborhood is assigned to individuals
   
2. the cutting points used to assign individuals to categories. Ideally, categories used (quintiles, for example) should contain the same number of individuals in each category.