3228.0 - Demographic Estimates and Projections: Concepts, Sources and Methods, 1999

ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 30/08/1999

Summary
Downloads
Explanatory Notes
Related Information
Past Releases

Page tools: Print

Print Page Print all pages in this product

Print All

Contents >> Chapter 3. Statistical Local Area population estimates >> Intercensal revision of SLA population estimates

3.41. When the census year population estimates are finalised for the SLAs (ie. 30 June) they can be compared with the preliminary estimates for the same date already produced by updating the previous census (using the method described above). The difference between the two estimates for each SLA is called the 'intercensal error'. (This estimate is updated later with an estimate of final intercensal error using final 1991-96 births, deaths and category jumping data, as described in Table 2.6, paragraph 2.62).

3.42. Of these two estimates, the first is customarily adopted as the true estimate, (ie. the true estimate for each SLA is the one based on the census at that date, not the one carried forward from the previous census).

3.43. To overcome the break in continuity that this would entail, all population estimates updated from the previous census are then recalculated. For SLA totals in the 1992 to 1995 intercensal period, this was done by first choosing the regression model which best estimated (in terms of producing the lowest SLA intercensal errors) the 1996 Census-based SLA population estimates. For each SLA this model was then used to obtain the SLA population estimates for 1992 to 1995 inclusive.

3.44. Recalculation of SLA age-sex estimates is done in a two stage process:

For each age and sex, SLA 1992-1995 levels are derived by interpolating between 1991 and 1996 Census-based estimates.

For each year and State, the results are then put through an IPF procedure to simultaneously constrain them to add to both State age-sex population estimates and the re-modelled intercensal SLA totals.

3.1: Intercensal Error at 30 June 1996, Statistical Local Areas (a)
Percentage Error (pe)	NSW	VIC	QLD	SA	WA	TAS	NT	ACT	Aust.
Number of SLAs

5 < pe	35	31	81	6	27	3	8	8	199
2 < pe <=5	42	34	77	13	17	4	7	20	214
0 < pe <=2	43	38	76	32	26	14	6	16	251
-2 < pe <= 0	30	34	72	27	32	8	8	16	227
-5 < pe <= -2	23	28	72	26	18	6	9	15	197
pe <= -5	10	29	52	14	18	6	22	12	163

Total	183	194	430	118	138	41	60	87	1,251

Percent

5 < pe	19.1	16.0	18.8	5.1	19.6	7.3	13.3	9.2	15.9
2 < pe <=5	23.0	17.5	17.9	11.0	12.3	9.8	11.7	23.0	17.1
0 < pe <=2	23.5	19.6	17.7	27.1	18.8	34.1	10.0	18.4	20.1
-2 < pe <= 0	16.4	17.5	16.7	22.9	23.2	19.5	13.3	18.4	18.1
-5 < pe <= -2	12.6	14.4	16.7	22.0	13.0	14.6	15.0	17.2	15.7
pe <= -5	5.5	14.9	12.1	11.9	13.0	14.6	36.7	13.8	13.0

Total	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
(a) Excludes SLAs with population less than 500 at 30 June 1996.

SLA totals

3.45. The preliminary estimated population at 30 June 1996 which was derived by updating the previous census fell short of the 1996 Census-based population by 21,572 persons. In other words, the preliminary intercensal error was -21,572. Despite this, as can be seen in Table 3.1, a larger number of SLA totals across Australia were over-estimated (positive intercensal error) than were under-estimated (negative intercensal error). The number of SLAs where the population was over-estimated was 651, while 569 SLAs were under-estimated (in 4 SLAs, estimated population exactly matched census population). As noted earlier in 3.41, the preliminary figure used to calculate the total Australia intercensal error quoted above is later updated using final 1991-96 births, deaths and category jumping data. Using this updated figure, the overall intercensal error is closer to -27,500. However, since only the national and State population figures were updated in this manner, and this chapter concentrates specifically on SLAs (which were not updated in this manner), preliminary national and State totals have been used in this analysis.

3.46. Similarly to total Australia, most States were under-estimated. However, Table 3.1 shows that six of the eight States had higher proportions of over-estimated SLA totals than under-estimated, while only South Australia and the Northern Territory had a higher proportion of under-estimated SLAs. The imbalance in proportions of under- and over-estimated SLAs was particularly apparent for New South Wales where 66% of SLAs were over-estimated. In comparison, 62% of SLAs in the Northern Territory were under-estimated.

3.47. An analysis of intercensal error at the LGA level is provided in Table 3.2. Similarly to the equivalent table produced for SLAs (Table 3.1), Table 3.2 shows that overall, a higher proportion of LGA totals were over-estimated than were under-estimated. New South Wales had the highest proportion of LGAs with a positive intercensal error (66%), while 75% of LGAs in the Northern Territory had negative intercensal error.

3.2: Intercensal Error at 30 June 1996, Local Government Areas
Percentage Error (pe)	NSW (a)	VIC (a)	QLD (a)	SA (a)	WA (a)	TAS (a)	NT (a)	Aust. (b)
Number of SLAs
5 < pe	33	3	24	3	25	2	0	90
2 < pe <=5	40	17	24	13	17	3	1	115
0 < pe <=2	43	20	22	31	24	8	0	148
-2 < pe <= 0	30	26	26	27	32	8	1	150
-5 < pe <= -2	21	10	17	26	18	4	2	98
pe <= -5	9	2	6	14	15	4	4	54

Total	176	78	119	114	131	29	8	655
Percent
5 < pe	18.8	3.8	20.2	2.6	19.1	6.9	0.0	13.7
2 < pe <=5	22.7	21.8	20.2	11.4	13.0	10.3	12.5	17.6
0 < pe <=2	24.4	25.6	18.5	27.2	18.3	27.6	0.0	22.6
-2 < pe <= 0	17.0	33.3	21.8	23.7	24.4	27.6	12.5	22.9
-5 < pe <= -2	11.9	12.8	14.3	22.8	13.7	13.8	25.0	15.0
pe <= -5	5.1	2.6	5.0	12.3	11.5	13.8	50.0	8.2

Total	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
(a) Excludes SLAs with population less than 500 at 30 June 1996. (b) ACT has combined State and Local Government functions and is excluded from this table.

3.48. Further insight into levels of intercensal error is obtained by comparing them for capital cities and the balance of the State.

3.49. The results of this are presented in Table 3.3. Overall, capital city populations were under-estimated by around 96,000 persons (0.8%), while the population residing outside capital cities was over-estimated by almost 75,000 persons (1.1%). This may help to explain why for most States, Table 3.1 showed higher proportions of over-estimated SLAs despite the fact that overall intercensal errors were negative, since in most States the majority of SLAs are located outside capital cities. The cities with the largest errors (in terms of percentage) were Darwin and Sydney, both of which were under-estimated. Outside the capital cities, the remainder of the Northern Territory and the remainder of Western Australia had the highest errors, with the remainder of the Northern Territory being under-estimated, while the remainder of Western Australia was over-estimated.

3.3: Intercensal Error by Part of State, 1996
	Persons	Intercensal Error
	('000)	Number	%
Capital city

Sydney	3,881.1	-59,732	-1.54
Melbourne	3,283.3	-34,467	-1.05
Brisbane	1,520.0	5,507	0.36
Adelaide	1,079.1	7,439	0.69
Perth	1,295.1	-12,309	-0.95
Hobart	195.7	-714	-0.36
Darwin	82.2	-1,325	-1.61
Canberra	307.9	-797	-0.26
Australia	11,644.5	-96,417	-0.83
Balance of State (a)
New South Wales	2,323.6	45,160	1.94
Victoria	1,276.9	15,315	1.20
Queensland	1,818.7	10,219	0.56
South Australia	395.1	-2,520	-0.64
Western Australia	470.2	9,786	2.08
Tasmania	278.7	-350	-0.13
Northern Territory	99.6	-2,971	-2.98
Australia	6,663.1	74,696	1.12
(a) As Canberra contains 99.9% of the ACT population, ACT is excluded from balance of State.

3.50. Table 3.4 presents a summary of the average absolute SLA and LGA intercensal error for each State, for both 1991 and 1996.

3.4: Average SLA Intercensal Error, 1991 and 1996
	Statistical Local Areas				Local Government Areas
	Number of SLAs (a)	Average SLA size (a)	Average absolute error	Average percentage error (b)	Number of LGAs (a)	Average LGA size	Average absolute error	Average percentage error (b)
	No.	No.	No.	%	No.	No.	No.	%
1991
NSW	184	32,058	627	3.2	176	33,507	652	3.1
VIC	231	19,136	348	3.2	210	21,050	363	2.7
QLD	425	6,966	280	5.4	135	21,931	556	4.4
SA	127	11,388	166	3.0	120	11,966	167	2.8
WA	145	11,283	342	6.1	138	11,856	351	5.7
TAS	62	7,468	168	4.7	46	10,145	150	2.1
NT	58	2,851	141	5.3	6	19,478	716	6.3
ACT	101	2,865	92	4.1	-	-	-	-
Australia	1,333	12,963	310	4.4	831	20,378	416	3.5
1996
NSW	186	33,359	779	3.4	177	35,048	812	3.3
VIC	198	23,031	972	6.3	78	58,462	885	2.1
QLD	448	7,469	272	4.5	125	26,710	517	4
SA	129	11,608	172	2.7	118	12,420	181	2.6
WA	150	11,768	305	5.1	142	12,431	254	4.2
TAS	43	11,034	229	3.2	29	16,360	307	2.6
NT	62	2,933	196	7.9	8	17,130	553	3.6
ACT	107	2,881	102	3.9	-	-	-	-
Australia	1,323	13,869	423	4.6	677	26,511	514	3.3
(a) Some SLAs combined to conform with published estimates. Excludes Off Shore Areas and Migratory SLAs. (b) Excludes SLAs with population less than 500.

3.51. Overall, the average absolute intercensal error for SLA totals was 4.6% in 1991 and 4.8% in 1996. In both years, South Australia was the State with the lowest error in percentage terms, while the Northern Territory was the highest. The errors for LGA totals were, in percentage terms, lower than those for SLAs. It can be assumed that this is partly due to LGAs having higher average population sizes. Despite this, errors for SLAs and LGAs follow generally the same pattern in that States with larger SLA errors, generally have larger errors for LGAs, and vice versa. Comparing 1991 and 1996, it can be seen that percentage errors for SLAs were fairly similar for New South Wales, South Australia, Western Australia, and the Australian Capital Territory, were lower for Tasmania and Queensland, but tended to be higher in 1996 for Victoria and the Northern Territory.

3.52. Between 1991 and 1996, SLAs in all States were affected by boundary adjustments. SLA boundaries in Victoria and Tasmania in particular underwent a series of major adjustments, while some significant boundary restructuring also took place in Queensland. This impacts on the continuity of the symptomatic indicators used in the regression phase and subsequently can be expected to lead to higher 1996 errors in these States, LGAs and SLAs than might otherwise have been the case. Another point to note when comparing State intercensal errors is that differences in factors such as the average population size, growth rates, state intercensal error and population distribution should also be taken into account, since all of these influence the degree of accuracy of population estimates (Demography Working Paper 98/1: Issues in Estimating Small Area Populations).

3.53. Table 3.5 presents a comparison of intercensal errors in a number of countries. As noted previously, the average population size of the areas being estimated has considerable effect on the accuracy of the estimates subsequently produced. For this reason, intercensal error figures for both SLAs and Statistical Subdivisions (SSDs) have been included for Australia. While SLAs are the basic units of estimation in Australia, SSDs, which are an amalgamation of SLAs, are closer in size, on average, to 'local areas' in other countries. Consequently, it is perhaps more meaningful, for comparative purposes, to use the intercensal error for SSDs in this type of analysis. It can be seen from Table 3.5 that average absolute percentage errors for Australian SSDs compare favourably with those for 'local areas' in other countries.

3.5: International Comparison of Intercensal Error
Country	Type of 'Local Area'	Reference year	Average population	Average absolute % error

Australia	Statistical Local Area	1996	13,300	4.6	(a)
	Statistical Subdivision	1996	99,500	2.2
Canada (b)	Census Division	1991	108,200	3.6
England & Wales (c)	Country District	1991	127,000	2.5
USA (b)(c)	County	1990	79,200	3.6
New Zealand (b)	Territorial Authority	1996	50,200	2.3
(a) Excludes SLAs with population less than 500. (b) Equates with error of closure which is the difference between population estimates produced prior to a census and corresponding census counts (no account is taken of variations in the undercount between censuses). (c) Census held at ten yearly intervals. Sources: Statistics Canada (1995); Office of Population Censuses and Surveys (1994); US Bureau of the Census (1994); Statistics New Zealand.