Concept: Income Quintile Ranking Procedure

Introduction

Note: For a summary of the ranking method that is detailed in this concept, please see the Income Quintiles Based on the 1996 Census concept.

The postal code/municipality code to income quintile rankings are based upon census information from 1986, 1991, and 1996. The 1986 census information is used to rank the 1984 to 1988 Manitoba population. The 1991 census is used to rank the 1989 to 1993 population. The 1996 census is used to rank the 1994 to 1997 population.

The 1986, 1991 and 1996 censuses are organized as one record per enumeration area (EA). Enumeration areas are the smallest unit of geography for which census data are normally available. Enumeration areas can be grouped up into census tracts (CT), census subdivisions (CSD), and census divisions (CD). These and other census geography units are described more completely in: cnd.a9105.cdbk (internal access only) .

The method of attaching census average income values available at the enumeration area level to our Manitoba population databases involves several steps.

Step 1 - Exclusion of records which cannot be ranked to an Income Quintile

An observation is defined as not rankable for several reasons:


1. Postal codes which are associated with a personal care home (PCH) cannot be ranked. Postal codes where the majority of residents (>90%) are in a personal care home (PCH) must be excluded from ranking as the census does not collect income for institutionalized populations.
   
   For a description on how postal codes were identified as having a majority of personal care home residents please see:
   
   
   • pchres_bypc_byyear (internal access only)
     
     
   • pchres_bypc_byyear96 (internal access only)
     
     
   • pchres_bypc_byyear97 (internal access only)
   
   
2. Postal codes that are associated with other institutions cannot be ranked. Some examples of such postal codes are those belonging to the public trustee office, prisons, and mental health institutions. These postal codes were collected from various sources, most notably Charles Burchill, who, in collaboration with Fred Toll, provided a most comprehensive list of institutions that have their own postal code.
   
   For a complete list of these postal codes, please see:
   
   
   • Postal Codes and Institutions
   
   
3. Postal codes that are not present on the postal code conversion file (PCCF) cannot be ranked. The postal code conversion file provides a link from the postal code to the census enumeration area. If the postal code is not present on the PCCF there is no way to attach census income values required for the ranking procedure. (This is not entirely true as a portion of the data may be linked to the census by municipality code. More on this later.)
   
   
4. The postal code or municipality code is not rankable because no income value is provided on the census. The census will suppress income information for the EA if the non-institutional population of the EA is less than 250. I made an effort to impute an income value to these EA's as much as possible. Some EA's are left with a missing income value as the census defined no non-institutional population in that EA. Examples of this might be other institutions we did not define above, or industrial or business districts in the city with no homes present.

Step 2 - Preparing the Remaining Population for Ranking

After the observations that cannot be ranked have been removed, the population is divided into a group that can be ranked by postal code (strictly urban population) and a group that can be ranked by municipality code (mixed urban/rural and strictly rural populations).

The urban/rural status of a postal code is determined by the urban/rural designation of the EA linked to the postal code using the postal code conversion file. In addition several postal codes that we identified from the postal code book as rural routes link to strictly urban EA's according to the postal code conversion file. In reality, however, these postal codes are mixed urban/rural since they involve post office boxes located in urban areas, but the mail is delivered to populations residing in rural areas.

For a complete description of the method used to determine the urban/rural status of a postal code, please see:


• urban_pc (1986) (internal access only)
  
  
• urban_pc91 (1991) (internal access only)
  
  
• urban_pc96 (1996) (internal access only)

In addition, for a complete list of postal codes that are forced to mixed urban/rural status, please see:


• PC_List2

Preparing the Strictly Urban Population for Ranking

The strictly urban population is ranked by postal code. The postal code is linked to the census enumeration area via the postal code conversion file.

When a postal code links to 2 or more EA'S, the average income value assigned is the weighted mean of the average income values of each EA, using the total number of private households in each EA as the weight.

It is possible to link to EA's where the average income value is set to 0. These 0 values are re-set to missing so that they do not contribute to the weighted mean average income value assigned to the postal code.

This process defines the set of postal codes that cannot be ranked, described above as those postal codes that link only to EA's with 0/missing values of average income.

Preparing the Mixed Urban/Rural and Strictly Rural Populations for Ranking

After the strictly urban postal codes have been prepared, the mixed urban/rural and strictly rural postal codes remain. This population is ranked by municipality code.

Previously it was found that ranking rural populations by postal code was unsatisfactory because rural postal codes often encompass too wide a geographical area. For example, many times a rural town and the surrounding municipality share a common postal code. But these two populations can be separated by municipality code. By calculating the mean income value on the basis of municipality code, then, these two very different populations can be identified.

The municipality is linked to the federal enumeration areas belonging to the census subdivision that corresponds to the municipality. In most cases there is a one-to-one correspondence between municipality and census subdivision. The link between municipality and census subdivision was provided by some work done by Cam Mustard and Fred Toll. Please see:


• muncode_csd96 (internal access only)

As with the strictly urban postal codes, it is possible for a municipality to link to 2 or more EA's. When this happened, the weighted mean of the average income value of each linking EA is assigned to the municipality. Again, the weighted mean is calculated using the total number of private households as the weight.

Several municipalities did not link to a census subdivision for three reasons.


1. Several pre-1989 municipalities exist on the 1986 population file which were re-assigned in 1989. Since the municipality/census subdivision correspondence was developed in 1990, no corresponding census subdivision was provided to these municipality codes. I was able to determine the link between the pre and post 1989 municipalities and provide a weighted average income value for the "old" muncodes. Two old muncodes (309 and 314) could not be fixed using this technique, so it was decided that the population in these muncodes would be classified as strictly urban and ranked by postal code.
   
   For a complete description of the method used, please see: impute_old_mun_income (internal access only) .
   
   
2. Several First Nations bands boycotted the census and hence no average income value was provided for their municipalities. I was able to provide an imputed average income value to these municipalities by calculating the weighted average income value of all First Nations bands that did not boycott the census by north/south zone. The average income value is imputed to the boycotting bands on the basis of their zone.
   
   For a complete description of how this was done, please see:
   
   
   • impute_native_income (internal access only) .
     
     
   • impute_native_income91 (internal access only) .
     
     
   • impute_native_income96 (internal access only) .
   
   
3. One municipality code (A53) is defined as Inuit and out of province First Nations. This population is not rankable and is described previously as the not rankable population due to missing AVGINC value for the municipality.

Several municipalities linked to mixed urban and rural EAs. Because the municipality has to be defined as either urban or rural, the weighted mean of gurban (gurban=0 indicates rural EA, and gurban=1 indicates urban EA) was calculated for each municipality. The rounded value of this mean determined the urban/rural nature of the municipality.

Step 3 - Ranking by Urban and Rural Designation

At the end of all this we have strictly urban populations summed by postal codes, defined as urban (gurban=1 in all cases) and having an average income value attached. We also have mixed urban/rural populations summed by municipality codes, defined as urban or rural by a weighted majority rule and also having an average income value attached.

To rank postal codes and muncodes to a quintile, these two files are concatenated, and sorted by rural/urban and average income value. The total population by urban/rural is calculated. Within an urban/rural designation, postal code or muncode populations are summed into classes so that approximately 20% of the population is in each class. These classes form quintiles within the urban/rural designation.

After ranking each postal code or muncode, the population is again broken apart into the set ranked by postal code and the set ranked by muncode.

From these two datasets, a format that assigns an income quintile value to a postal code and a format that assigns an income quintile value to a muncode are produced.

Applying the Income Quintile Ranking to your Data

To use these formats on any SAS data set, all of the deletions and the division into strictly urban versus mixed and strictly rural groups described above must be performed on that data set. Once that is done, the group defined as strictly urban can be ranked into income quintiles by postal code value, and the remaining group, defined as mixed or strictly rural, can be ranked by the muncode.

A macro has been developed to perform the deletions and apply the formats correctly to a SAS dataset. For more information about the macro see:


• quint macro

Important consequences of this method of assigning income quintile

1. First Nations people with strictly urban postal codes will be assigned an income quintile on the basis of their postal code. All others will be assigned an income quintile on the basis of their home reserve, regardless of whether they actually live on the reserve. This cannot be overcome, because First Nations people always maintain their home reserve muncode whether they live there or not.
   
   
2. Portions of any event data set will not be ranked for various reasons. These reasons are described in the deletions above.
   
   
3. Populations living on the periphery of Winnipeg may be defined as rural by the census. The census defines the urban/rural nature of a population by a density rule. Populations living in East St Paul (muncode=304), for example, are defined as rural by the census. There are other examples of this.

Table 1 - Populations Excluded from Income Quintile Ranking By Year and Reason for Exclusion

Year	Pop of PC 90%+ in PCH	PC Belongs Other Instit.	PC not Present on PCCF	PC AVGINC Value Missing	MUNCODE AVGINC Value Missing
1984	3,773	2,235	447	2,473	179
1985	3,523	2,266	434	2,732	174
1986	3,549	2,187	964	2,667	184
1987	3,456	2,146	1,970	2,461	177
1988	3,687	2,058	3,641	2,166	187
1989	3,817	2,195	1,044	1,804	234
1990	3,787	2,343	874	1,808	223
1991	3,832	2,476	791	1,803	211
1992	3,947	2,543	1,603	1,840	246
1993	3,858	2,935	12,283*	2,530	236


Table 1 shows the population excluded from the quintile ranking procedure by year.

Note: A large number of new postal codes were introduced in 1993. These postal codes are not present on the 1991 PCCF file. However, a portion of this population can be assigned a quintile by municipality code. The macro ranks as much of this population as possible by municipality code.