This chapter presents and analyses the undercount estimates derived from the 2001 Post-enumeration Survey. In addition to the aggregate results, the analysis features the age, sex, ethnic and geographic breakdowns. It is important to note that the 1996 PES results included here for comparison are based on revised figures. The revision was needed because of weaknesses in the original methodology, identified by the statistical procedures used to derive the 2001 PES results. The reasons for the revision are outlined briefly in the following section.
3.1 Revision to 1996 PES results
All results from the 1996 Post-enumeration Survey (PES) included for comparison in this chapter are revised net undercount estimates. The revision was necessary for two reasons, both relating to the methodology adopted for estimating undercount/overcount. Firstly, a small number of individuals were omitted from the 1996 PES calculations. This occurred where individuals matched to dwellings classed as unoccupied in the census were given a post-stratification factor of zero. Secondly, the 1996 eligibility criteria were overly restrictive, as they excluded all individuals in dwellings where there were some inadequate responses.
The use of more conservative eligibility criteria used in deriving results from the 1996 PES partly reflected the fact that the 1996 PES was the first such survey conducted in New Zealand. The revised 1996 PES figures adjust for both methodological factors.
3.2 Analysis of PES results
In section 2.2, it was pointed out that for practical reasons, temporary and non-private dwellings were excluded from the PES, as were dwellings in remote areas. Both the 1996 and 2001 Post-enumeration Surveys sampled individuals in permanent private dwellings only. Although the net undercount rate may be different in dwellings other than permanent private dwellings, the same net undercount rates have been assumed in deriving the undercount estimate for the total population, due to a lack of additional data.
The 'total population' as defined here is the 'true' number of New Zealand residents in New Zealand on census night. The total population is therefore equivalent to the 'census usually resident population count' plus the estimated census net undercount of New Zealand residents.
Table 1 carries the 2001 and 1996 net undercount estimates and the sample errors for these estimates. Because of the small sample size and the resulting large sample errors, it has been necessary to aggregate age groups and geographical areas.
Table 1
1996 and 2001 Post-enumeration Survey Results
Overall, the 2001 Census coverage was high, and this reflects the cooperation and support of the New Zealand public. The PES estimated that about 97.8 percent of New Zealand residents in the country on census night were enumerated in the 2001 Census. This represents a net undercount of about 2.2 percent or 85,000 people. This resulted from an estimated undercount of 107,000 people offset by 22,000 people being counted more than once. The 1996 PES gave an estimated net undercount of about 1.6 percent or 60,000 people.
While the 2001 Census enumerated 3.74 million residents in New Zealand, the PES results suggest that the number of New Zealand residents in New Zealand on census night was closer to 3.82 million. This figure excludes New Zealand residents temporarily away overseas on census night.
Given the small size of the PES sample, the estimates of net undercount yielded by the survey are subject to sizeable error margins, called the sample error. The sample error is a measure of the variability that occurs by chance because a sample, rather than the entire population, is surveyed. Therefore the undercount estimates need to be interpreted with caution.
The net undercount of 2.2 percent as estimated by the 2001 PES had a sample error of 0.3 percent. Thus there is a 95 percent probability that the true population undercount was between 1.9 and 2.5 percent, not allowing for non-sampling error. The corresponding interval for the 1996 estimate of net undercount was 1.4 and 1.8 percent. Thus the lower limit of the 2001 net undercount estimate is close to the upper limit of the 1996 estimate.
3.2.1 International comparison
It is inevitable that in an exercise of the scale of the census that some people will be missed. It is an experience shared by virtually all census-taking nations. As Barbara Bryant, a former director of the US Bureau of the Census aptly noted "it is not possible to count 100 percent of the population, although the public and many officials have an unrealistic expectation that this can be done..." (Bryant, 1993).
Given the strategic significance of the census data, many countries conduct special surveys to measure the coverage of their population census. The Australian Bureau of Statistics, for example, has run evaluation programs since 1966; the US Census Bureau since 1951; Statistics Canada since 1966; and the National Statistical Office of the United Kingdom since 1981. Net undercount rates for these four countries and New Zealand at recent censuses are compared in Table 2.
The web-based version of this report will be updated when the results from the latest post-enumeration surveys in Australia, Canada, and the United Kingdom become available.
Table 2
Estimated Census Undercount
New Zealand and Selected Countries
The undercount rate of 2.2 percent for New Zealand’s 2001 Census is similar to the rates for other countries. Differences in undercount rates by sex, age and ethnic minorities in New Zealand have also been found in other countries.
International comparisons should, however, be interpreted with caution because of the differences in sample size, in enumeration procedures and practices, and in the nature and size of sampling and non-sampling errors. The precise impact of these variations on the undercoverage levels cannot be readily ascertained.
3.3 Undercount differentials
Below the national level, there are significant variations in net undercount by age, sex, ethnicity and geographic area. Some population groups are more likely to be missed by the census than others. This section examines the findings of the 2001 PES on the undercount differentials in New Zealand, and compares these with the 1996 PES results.
3.3.1 Males and females
Table 1 shows that the 2001 Census missed fewer females than males – an estimated 37,000 females compared with 48,000 males. Males accounted for about 56 percent of the estimated 85,000 people missed in the census, although they make up under 49 percent of all New Zealanders.
The net undercount rate was 2.6 percent for males and 1.9 percent for females (Figure 2). In 1996, the net undercount was lower for both males and females, estimated at 1.9 and 1.3 percent, respectively. Given the sample errors, the 2001 net undercount ranged between 2.2 and 3.0 percent for males and between 1.6 and 2.2 percent for females.
Figure 2
Net Undercount by Sex
1996 and 2001 Censuses
Overseas studies also indicate that males are more likely to be missed in a census than females. In Australia in 1996, for example, the net undercount was 2.0 percent for males and 1.2 percent for females (Australian Bureau of Statistics, 1997). Similarly, in 1991 in England and Wales the adjustment factor for estimated undercount was 1.03 and 1.01 percent for males and females, respectively (Office of Population Censuses and Surveys, 1994).
3.3.2 Age groups
Demographic literature also offers a substantial body of evidence on the undercounting of young children and the youth in any census (Shryock and Siegel, 1973). A New Zealand study that compared birth registration forms of the babies born in the few months before the 1976 Census with the 1976 Census questionnaires, found that about 3 percent of babies were missed by the census (Yurjevich, 1982). This omission is significant from a statistical perspective as post-censal estimates based on the unadjusted, or inadequately adjusted base figures, clearly highlight the poorly enumerated birth cohorts, and are of limited value for planning, eg for pre-school education.
When the overall sample size is small, demographers face a difficult task in analysing the PES results by age and other characteristics, because sample errors are relatively large. For this analysis, it was decided to aggregate the age data. Only four broad age groups, ie under 15 years, 15–29 years, 30–44 years and 45 years and over, are analysed here.
In general, the age differentials in undercount are more pronounced than the gender differentials outlined in the preceding section, with significant variation in the completeness of coverage of various age groups (Figure 3). Based on the 1996 experience, young adults (aged 15–29 years), who are quite mobile and therefore have a relatively high risk of being missed in a census, were identified as a target group in the 2001 Census communication strategy. Table 1 shows that about 24,000 people missed by the 2001 Census were in this age group. They accounted for 28 percent of those missed, although they made up 20 percent of all New Zealand residents. As in 1996, they returned the highest undercount rate in the 2001 Census (3.1 percent).
The undercount rate for children (0–14 years) was also higher than average, at 2.7 percent in 2001. Those aged 45 years and over were better enumerated: their undercount rate (1.4 percent) in 2001 was less than half that of young adults.
Figure 3
Net Undercount by Age
1996 and 2001 Censuses
3.3.3 Ethnicity
Ethnic differentials in census undercount are marked. Ethnic minorities, in particular, are likely to be poorly enumerated. In Australia’s 1996 PES, the net undercount rate for Aborigines and Torres Strait Island people was considerably higher than that of non-indigenous Australians (Australian Bureau of Statistics, 1997). Similarly, in 1990 in the United States, the net undercount was low (0.7 percent) for non-Hispanic whites, and relatively high (between 4 and 5 percent) for Blacks, Hispanics and American Indians (US Census Bureau, 2001).
In New Zealand, the 1996 PES found that the net undercount was much higher for the Māori and Pacific peoples (3.7 and 3.9 percent, respectively) than for the majority European population (1.3 percent).
Consequently, for the 2001 Census "a more comprehensive communication strategy was developed to encourage the full cooperation of the public in the census, with a clear focus on reducing the undercount in particular groups of the population. Liaison with Māori and Pacific peoples was an important part of this consultative process, and included the development and distribution of the reorua (bilingual Māori/English) census forms" (Statistics New Zealand, 2001). A Māori census coordinator, kaitakawaenga (Māori liaison officers) and Pacific liaison officers were appointed to assist the promotion of the census to these groups. More detail on the 2001 Census communication strategy is available in the Statistics New Zealand report Introduction to the Census (2001).
Yet, the nature and magnitude of the ethnic differentials in undercount changed little between 1996 and 2001 (Figure 4). In line with the other results cited above, all ethnic groups recorded higher undercount rates in 2001 than in 1996. The net undercount was highest for Pacific peoples at 5.2 percent, followed by 4.4 percent for Māori, and 2.4 percent for the Asian ethnic group. The figure was again lowest for the European ethnic group (1.7 percent), which makes up three-quarters of all New Zealanders. Pacific peoples were thus about three times more likely to be missed in a census than their European counterparts. The Pacific undercount estimate in 2001 was also subject to the largest sample error of 1.6 percent. The range for the net undercount of this group was therefore 3.6–6.8 percent. For the European group the sample error was 0.3 percent, and the range of the net undercount was 1.4–2.0 percent.
Figure 4
Net Undercount by Ethnic Group
1996 and 2001 Censuses
Note: An estimate of the undercount of the Asian ethnic group is not available for 1996.
Ethnic differentials in net undercount may be partly attributed to the differences in the age structures of the various groups. Mäori and Pacific peoples have a more youthful population, and a higher percentage of young children.
3.3.4 Spatial variations
Besides age and ethnicity, area of residence can be an important determinant of census coverage. Overseas studies indicate significant variations both between and within areas. In 1991 in England and Wales, for example, outer London and non-metropolitan areas were better enumerated than those living in inner London and main metropolitan areas (Office of Population Censuses and Surveys, 1994). In the United States, successive surveys have found significant variations in the undercoverage estimate among states and between urban and rural areas (Documents, 1996).
Ideally, we would have preferred an analysis of net undercount by region, urban area or territorial authority. However, the sample size was not large enough to yield reliable estimates at these geographical levels. Following a preliminary evaluation, it was decided to focus on the broad areal groupings that were adopted for the 1996 analysis.
The rate of undercount for the northern half of the North Island, which consists of the Northland, Auckland, Waikato and Bay of Plenty Regions and is home to half of all New Zealanders, was 2.7 percent. It exceeded the national average of 2.2 percent. The net undercount in the southern half of the North Island (which consists of the Gisborne, Hawke's Bay, Taranaki, Manawatu-Wanganui and Wellington Regions) was 2.2 percent (Figure 5). The South Island, which is the larger of the country’s two main islands but houses just one-quarter of the total population, had better coverage than the North Island in both 1996 and 2001. The net undercount rate in the South Island was 1.3 percent, identical to that estimated at the 1996 Census. In contrast, the undercount rate increased in the North Island. The 1996 PES had estimated net undercounts of 1.9 percent and 1.3 percent for the northern and southern halves respectively.
Figure 5
Net Undercount by Area
1996 and 2001 Censuses
The higher undercount in the North Island, especially in the northern half, may be partly attributed to the differences in the age and ethnic structures of the population. Over half of Mäori and about three-quarters of Pacific and Asian peoples reside in the four northernmost regions, compared with less than half of Europeans. Also, while there are more Europeans in the South Island than in the southern half of the North Island, the opposite is true for other ethnic groups.
3.4 Undercoverage of dwellings
The estimated number of permanent private dwellings undercounted by the 2001 Census was about 2,400 or 0.2 percent nationally, compared with about 6,000 or 0.5 percent at the 1996 Census. The 2001 Census counted 1.5 million dwellings, and the 1996 Census counted almost 1.3 million dwellings.