|Page tools: Print Page Print All|
IMPROVED FAMILY ESTIMATES FROM THE LFS
The estimate of the number of families of each type (for example, couple families with dependent children under 15 years of age) will likewise increase, although the size of the increase varies according to family type. The largest proportional increase is for other families (that is, families which are not couple families or lone parent families), which increases on average by 5% (see graph below) albeit from a small base.
The graphs at the end of this publication show the estimated number of families of each type for the period August 2004 to May 2008 produced by the current estimation method compared with the improved estimation method.
The improved estimation method results in a large increase in the total number of families in the Northern Territory, and a small increase in the states and the Australian Capital Territory. The graphs at the end of this publication show the estimated number of families in each state/territory for the period August 2004 to May 2008 produced by the current estimation method compared with the improved estimation method.
In the Northern Territory, the improved estimation method produces an estimate of total families which is approximately 40% higher than the current estimate. This is primarily due to the number of families in the Northern Territory being considerably underestimated under the current method (see the previous section on the Current Estimation Method for details). The current method underestimates the number of families in the Northern Territory because:
The improved estimation method provides estimates of higher quality than the current method. The improved method includes households in which one or more members aged 15 years and over are permanent members of the Australian defence forces, and makes use of person and household benchmarks, which compensates somewhat for the lack of household relationship information in discrete Indigenous communities.
COMPARISON WITH OTHER ABS SOURCES OF FAMILIES DATA
Survey of Income and Housing
A key source of ABS data on families is the Survey of Income and Housing (SIH), which is conducted every two years. The SIH collects detailed information about income and personal and household characteristics of persons aged 15 years and over resident in private dwellings. The SIH sample of approximately 10,000 dwellings is drawn from all parts of Australia, except very remote areas, which are excluded from the scope of the survey.
Like the improved LFS family method, the SIH makes use of person and household benchmarks, however, those benchmarks exclude very remote areas to match the survey scope.
The table below compares families data from the SIH with families data from the LFS under the current estimation method and the improved estimation method. The LFS data for each month of 2005/06 have been averaged for comparison with SIH.
The scope of the LFS includes very remote areas of Australia, whereas the scope of the SIH excludes them. Therefore, LFS estimateS of the number of families would be expected to be higher than SIH, to account for the wider scope.
The table shows that the current LFS families estimates are generally lower than those from SIH, despite the wider scope. The largest difference is in the Northern Territory, where the current LFS estimate of the number of families is 9.4% lower than the SIH estimate.
The table also shows that the improved LFS families estimates are generally higher than the SIH estimates for all states/territories, except New South Wales, where the improved LFS family estimate is 2.9% lower than SIH. In the Northern Territory, the improved LFS family estimate is 31.1% higher than SIH. This difference is largely attributable to the large proportion of the population in the Northern Territory who live in very remote areas, which are in scope of LFS family estimates, but excluded from the scope of SIH.
The improved LFS family estimates compare favourably with those from SIH, and have the following advantages:
Family Characteristics and Transitions
Another source of ABS data on families is the Family Characteristics and Transitions Survey (FCTS), conducted in 2006-07. The FCTS was part of the Multi-Purpose Household Survey (MPHS), which is collected from a fraction of households who are in their final month of participation in the Monthly Population Survey (the survey vehicle which also includes the LFS).
The FCTS collected information on household and family composition, with a particular focus on families with children aged 0-17 years. The scope of the FCTS was usual residents of private dwellings in all parts of Australia except very remote areas. The estimates were derived from a sample of approximately 13,000 dwellings.
Like the improved LFS family method, the FCTS makes use of person and household benchmarks, however, those benchmarks exclude very remote areas to match the survey scope.
The table below compares families data from the FCTS with families data from the LFS under the current estimation method and the improved estimation method. LFS data for the 10 months of FCTS enumeration in 2006/07 have been averaged for comparison.
The scope of the LFS includes very remote areas of Australia, whereas the scope of the FCTS excludes them. Therefore, LFS estimates of the number of families would be expected to be higher than FCTS, to account for the wider scope.
The table shows that the current LFS families estimates are generally lower than those from FCTS, despite the wider scope. The largest difference is in the Northern Territory and the Australian Capital Territory, where the current LFS estimate of the number of families is 6.8% and 6.6% lower than the FCTS estimate.
A possible contributing factor to the differences for the Northern Territory and the Australian Capital Territory is the fact that they have a higher proportion of households containing a person outside the scope of the LFS (for example, permanent members of the Australian defence forces) than the states. Since the current LFS family estimation method excludes families with persons out of scope of the LFS and the method makes no compensation through benchmarking, the underestimation of the number of families is higher in the territories.
In the states, the Australian Capital Territory, and at the Australia level, the improved LFS families estimates are similar to the FCTS estimates, despite the LFS having a wider scope. This appears to be related to the higher proportion of family households (as opposed to non-family households such as single-person households and group households) in the FCTS sample compared to the LFS sample under the improved method. In the FCTS, a family did not contribute to the estimates if one or more members of the household did not contribute to LFS person-level estimates (for example, the person failed to meet LFS selection rules). Non-family households are more likely to have one or more members of the household who did not contribute to the LFS person-level estimates, thus increasing the proportion of family households in the FCTS sample.
In the Northern Territory, the improved LFS family estimate is 27.6% higher than FCTS. This difference is largely attributable to the large proportion of the population in the Northern Territory who live in very remote areas, which are in scope of LFS family estimates, but excluded from the scope of FCTS.
The improved LFS family estimates compare favourably with those from FCTS, and have the following advantages:
Another important source of ABS families data is the Census of Population and Housing, conducted every five years. The main advantages of Census data are the fine level of geographic detail available, the complete coverage of all geographic areas, including very remote areas, and the increased reliability resulting from a complete Census of the Australian population, rather than a sample survey.
While the Census aims to collect data on every person and household in Australia, there are a number of limitations to families data from the Census, including:
The table below compares families data from the Census with families data from the LFS under the current estimation method and the improved estimation method. LFS data are for August 2006 for comparison.
The table shows that the current LFS families estimates are generally higher than family counts from the Census, except in the Northern Territory, where the current LFS family estimate is 9.3% lower than Census.
Under the improved LFS family estimation method, the difference between the LFS family estimates and Census family counts is slightly greater than under the current method, except in the Northern Territory where the LFS family estimate is 27.9% higher than the Census family count. This is likely to be related to the limitations of Census families data, listed above. Census underenumeration is higher in the Northern Territory (7.6%). Other limitations of Census data may also be more pronounced in the Northern Territory than elsewhere, due to its large Indigenous population.
The improved LFS estimates of the number of families are considerably higher than the family counts from the Census. Although Census counts provide data at a very fine level of geographic detail, they appear to understate the number of families due to:
These limitations understate the number of families, especially in the Northern Territory.
BREAK IN TIME SERIES
The improved estimation method and consequent increases in the size of the estimates will cause a break in time series. To enable comparisons to be made over time, revised historical data for the period August 2004 to August 2008 will be made available at the same time as data for September 2008 are released (16 October 2008). Data will be released monthly thereafter. Data for months prior to August 2004 cannot be revised using the improved estimation method, as insufficient family information was collected at that time.
The graphs above, and at the end of this publication, show that the improved family estimates produce a time series which exhibits less volatility than the current time series. This is primarily due to the use of independent population benchmarks.
From October 2008, family estimates will be released monthly, instead of annually. The data will be released in the families datacubes (Labour Force Status and Other Characteristics of Families - Electronic Delivery, cat. no. 6224.0.55.001). Each of the five datacubes (FA1 to FA5) will be split into two separate datacubes, one for the current methodology and one for the improved methodology. Using FA1 as an example, the FA1_jun94 datacube will contain annual data as currently published from June 1994 to June 2004, and the FA1_aug04 datacube will contain monthly data from August 2004 onwards calculated using the new methodology. These two datacubes will replace the current FA1 datacube, which will be withdrawn.
Care should be taken when comparing families data from month to month. Some series appear to have a strong seasonal pattern and no seasonal adjustment process has been applied to the data. Changes from month to month should be interpreted in the context of year to year change, to ensure that normal monthly fluctuations are interpreted in the context of any seasonal pattern.
An example is couple families with dependants (see graph below). The definition of 'dependant' includes all family members under 15 years of age; family members aged 15-19 attending school or aged 15-24 attending a tertiary educational institution full time (except those classified as husbands, wives or lone parents).
The graph shows the estimate of the number of couple families with dependants drops considerably in January of each year, which is more noticeable under the current LFS family method than under the improved method. January enumeration of the LFS falls during the holiday period for both school and tertiary students, increasing the likelihood of a family with dependants staying away from their usual residence at the time of LFS enumeration. Families with dependants are therefore less likely than other family types to contribute to family estimates in January. Without the use of benchmarks to compensate for these non-responding families, the underestimation of couple families with dependants under the current LFS family method is especially noticeable in January each year.
The drop in January each year is also noticeable under the improved LFS family method. This may be partly related to the increased likelihood of couple families with dependants to be staying away from their usual residence at the time of LFS enumeration, but also to uncertainty surrounding the full time status of students aged 15-24 years. To be classed as a 'dependant' a person aged 15-24 years must be either at school or attending a tertiary educational institution full time. In December, January and February, a number of people aged 15-24 years are reported as having a full time student status of 'not known/unclear', and a lower proportion are reported as studying full time. This results in a lower estimate of couple families with dependants in these months.
'NOT DETERMINED' LABOUR FORCE STATUS
The new method of producing family estimates uses all LFS households where sufficient family information has been provided. A small proportion of respondents in those households are outside the scope of LFS person-level estimates, for example because they are permanent members of the Australian defence forces, or because they fail to meet normal LFS selection rules. No labour force information is collected for persons out of scope of LFS. Therefore there will be a small proportion of persons included in LFS family estimates for whom no labour force information is available. In the published data, these persons will be assigned to one of two new categories: 'Not determined, defence force personnel' or 'Not determined, other'.
'Not determined, defence force personnel' will denote that a particular characteristic (such as labour force status) was not determined, because the family member was a permanent member of the Australian defence forces, and therefore outside the scope of the LFS. 'Not determined, other' will denote that a particular characteristic could not be determined for other reasons, such as the family member failed to meet selection rules, or did not respond fully to the survey.
Within the families datacubes (Labour Force Status and Other Characteristics of Families - Electronic Delivery, cat. no.6224.0.55.001) the 'Not determined' categories will apply to a small proportion of records when classifying by labour force status, status of employment and duration of unemployment.
The 'Not determined' categories will also apply to other items collected which are not currently included in the families datacubes, but are available on request, such as hours worked and occupation.
The 'Not determined' categories give users the flexibility to decide how they are to treat defence force families and other types of previously excluded families.
EFFECT ON SAMPLING ERROR
The estimates produced from the LFS, as from all surveys, are subject to sampling error; that is, they may differ from the estimates that would have been produced if all dwellings had been included in the survey. The most common way of quantifying sampling error is to calculate the standard error for the estimate. The standard error indicates the extent to which an estimate might have varied by chance because only a sample of dwellings was included in the survey.
The improved estimation method is expected to reduce the standard error on family estimates by up to 60%, compared to the current estimation method. The decrease in standard error is mainly attributable to the use of independent population estimates (benchmarks). The size of reduction in the standard error for a given estimate is largely dependent on the correlation between the estimate itself and the independent population estimates. For example, the standard error on estimates of the total number of families for Australia and for each State/Territory will decrease considerably, because the total number of families is highly correlated to the number of persons and households, which are used as benchmarks. There is less benefit in terms of standard error for estimates of each type of family, because the correlation to the benchmarks is less strong.
Independent of the improved estimation method, however, is a reduction in the size of the LFS sample from July 2008. Further information about the sample reduction can be found in the April and May releases of Labour Force, Australia (cat. no. 6202.0). From July 2008 the sample size of the LFS was reduced by 24% when compared with the June 2008 sample. The sample reduction increased standard errors on all LFS estimates, including family estimates from July 2008 by approximately 15%, compared to the standard errors which would have been observed without the sample reduction. The relative change in sample size varies across the States and Territories, with corresponding impacts on standard errors.
STANDARD ERROR MODEL
Due to technical limitations, it is not possible to publish the standard error associated with every possible cross-tabulation in the LFS datacubes. Instead, the datacubes annotate those estimates where the standard error is 25% or more of the estimate itself. The standard errors used for the annotations are based on a standard error model, which provides an indication of the standard error for any estimate without the need to access unit record data. The standard error models are produced by statistically modelling the standard errors as a function of the estimate itself. Standard errors derived from a statistical model are an approximation of the standard errors calculated directly from the data.
Currently, the standard error model underlying the families datacubes is the model used for LFS estimates of employed persons. Since the introduction of composite estimation in May 2007 (for details, see the 2007 issue of Forthcoming Changes to Labour Force Statistics, cat. no. 6292.0) and other changes, the employed persons model does not give a very accurate indication of the true standard errors associated with the family estimates. From October 2008, the families datacubes will use an improved standard error model, derived specifically for use with family estimates. The improved standard error model will also be made available to users in the spreadsheet Labour Force Survey Standard Errors, Data Cube (cat. no. 6298.0.55.001).
The table below shows the point at which the standard error is 25% or more of the estimate itself for June 2008 (before the sample reduction) and July 2008 onwards (after the sample reduction). Estimates which are smaller than these estimates are associated with standard error too high for most practical purposes.
For example, in June 2008, the LFS estimate of the number of same sex couple families with dependants in Australia is 2,700. This figure is below the Australia-level cut-off of 5,831, and therefore associated with standard errors too high for most practical purposes.
These documents will be presented in a new window.