Stats NZ has a new website.

For new releases go to

www.stats.govt.nz

As we transition to our new site, you'll still find some Stats NZ information here on this archive site.

  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+
Ongoing impact of Canterbury earthquakes on the Household Labour Force Survey

This article discusses the longer-term impact of the Canterbury earthquakes on the collection and results of the Household Labour Force Survey (HLFS). Specifically, it looks into what impact the creation of the red zones and the population shifts in Canterbury as a result of the earthquakes, have had on the survey.

Despite the earthquakes and the associated changes, the HLFS will continue to provide unbiased estimates of the key survey variables. Statistics NZ will continue to monitor the impacts of the earthquakes on the HLFS.

The article details:

For background information on the effects of the February 2011 Christchurch earthquake, see: 

Structure of the Household Labour Force Survey

There are several stages of selection in the HLFS. Firstly, geographic areas are selected, secondly dwellings, and finally, respondents from these dwellings.

The first stage consists of choosing small clusters of dwellings, called primary sampling units (PSUs). PSUs are geographically compact regions, and those used for the HLFS are based on the 2001 Census of Population and Dwellings. Most PSUs have between 50 and 100 dwellings, with an average of 67 dwellings. New Zealand is made up of 20,394 PSUs.

The PSUs are grouped into 119 strata of broadly similar PSUs, based on geography, the percentage of Māori, and other characteristics.

We select PSUs from all of these strata for the HLFS. In every quarter, we conduct HLFS interviews in a total of 1,768 PSUs.
In any given quarter we take roughly 10 dwellings from each selected PSU, and from these dwellings get information from those aged 15 years and over. Selected dwellings remain in the sample for eight quarters.

In any one quarter, 7/8 of the sampled dwellings were in the survey the previous quarter, and 1/8 of the dwellings from the previous quarter leave the sample. This 1/8 of the sample is replaced by different dwellings from the same PSUs of the departing sample.

Periodically, after the Census of Population and Dwellings, we reform the pattern of PSUs and select and phase in a new sample of PSUs.

top

Sampling variability

The sample we draw is just one out of a very large number of other potential samples that could be drawn, using the HLFS sample design. The estimates from one of these other samples would probably be different from the estimates we get from the sample we have.

Any given sample may over or under estimate a characteristic of the population, to a greater or lesser extent. Sampling error is a measure of this sampling variability. The sample was designed to keep the sampling variability small, subject to keeping costs down, that is, to keep the sampling errors on the key estimates small.

For example, retaining 7/8 of the sampled dwellings between quarters reduces the sampling variability of the changes between quarters.

Clustering addresses into compact geographic groups, and selecting clusters, is called cluster sampling. It is done as a cost-saving measure to reduce travelling cost.

Clustering makes it harder to pick up geographically localised events – something that happens in only one PSU will show up in the sample only if that PSU is selected. In general, clustering will increase the sampling error in the survey.

Another way we can reduce sampling variability is through the use of auxiliary data, which is strongly related to the key estimates from the HLFS. For example, population age and sex totals are strongly related to labour force estimates, like the total working-age population or the number of persons employed.

From the sample we have, we can make estimates of the national age-sex totals, and compare those with the known national totals from demographic projections. We then adjust the sample weights so that our sample gives the correct values for these national age-sex totals. This adjustment will also improve estimates strongly related to these totals.

Most of the auxiliary data we have that is suitable for use in this way is at the national total level.

The sample in Canterbury

In the Canterbury region there are 2,778 PSUs. At the 2001 Census, these contained about 187,000 occupied dwellings, and that number now exceeds 200,000.

The current HLFS selected 184 of those PSUs. From these, we approached 2,241 dwellings for the HLFS in the June 2010 quarter.

For this quarter, the achieved sample of responding dwellings in Canterbury was large, at 1,701.

Red zone addresses in Canterbury

Some PSUs will have dwellings in the red zone. To preserve respondent confidentiality, we will not identify in which Canterbury PSUs we are operating the HLFS, and where exactly they are located.

Of the 184 sampled Canterbury HLFS PSUs, 13 contained red zone addresses, (using the definition of red zone current at 17 August 2012). This number is 7.1 percent of the sampled Canterbury PSUs.

The same percentage for all of the Canterbury PSUs is 6.4 percent. The sample proportion agrees well with the actual proportion.

To get a rough idea of one impact of the earthquakes, note that at 17 August 2012, there were 8,294 addresses in the red zone.

At 9 October 2012, 4,775 of these had been settled (and presumably will be vacated). We can expect continuing reduction in the sizes of some PSUs because of the red zone.

Note that addresses (sourced from Terralink International) include non-private and commercial buildings, in addition to private dwellings.

Population movements in Canterbury are of course, larger than those vacating red zone homes. They include those moving either temporarily or permanently after the earthquake, those moving out while homes are repaired or to follow relocated businesses, or workers moving in for the reconstruction.

top

Robustness of the Canterbury sample

As mentioned above, the sample in Canterbury is a clustered sample, but a reasonably-sized one, and well spread throughout the region. It should cope fairly well with the changes that are occurring, and the estimates we get should reflect what is really happening.

The estimates should pick up, for example, red zone changes and the impact of the population changes.

The changes and movements in the Canterbury population will mean that the sampling errors will be larger than they would have been before the earthquakes.

The estimates of changes will probably be less affected than the estimates of levels.

At the moment, we do not have good auxiliary data for Canterbury that could be used to adjust the sample in the ways we discussed earlier. The survey itself is probably the largest ongoing data collection in Canterbury.

The next sections will discuss in more detail some factors – in addition to the red zones –that may affect the results. These are the impact of changes in PSU sizes and in private and non-private dwellings and their uses.

Changes in primary sampling unit sizes

Sampling variability discussed the use of auxiliary data to adjust the survey results to be more representative.

We cannot perfectly control the size of PSUs, even when we first form them. This can therefore increase sample variability, that is, higher sampling error. We adjust the survey weights to mitigate this. We adjust the sample weights so that the sample produces the correct estimates for the number of dwellings in each stratum at the time the PSUs were formed.

Growth in PSUs since the time we formed the PSUs makes this adjustment less effective as time passes. The changes in Canterbury PSUs, because of demolitions and new construction, are likely to be bigger than in most of the country.

We normally readjust the weights using census information. However, a small deterioration in quality may occur because of the postponement of the 2011 Census to 2013.

This deterioration will have some impact on the results, but we do not expect it to be large. In particular, the estimates of change should be less affected.

Private and non-private dwellings

Currently, the HLFS contacts only private dwellings. People living in non-private dwellings make up about 2 percent of the total population. We rate up survey results to the full population, assuming the main estimates are distributed in the non-private dwellings in the same way as in the private dwelling sample.

Temporary housing and housing for reconstruction workers may mean the proportion of non-private dwellings changes in Canterbury, and may have different characteristics from the private dwellings in Canterbury.

Even ideally where the number of non-private dwellings remained the same, there might be changes the survey would not pick up, because we do not survey non-private dwellings.

For example, some non-private accommodation might change from taking very short-term visitors to taking workers for the reconstruction. If we could apply our survey rules to these people, we would record short-term visitors in their place of origin, while new visitors would be recorded as working in Canterbury. These are different outcomes which we cannot distinguish because we do not survey in non-private dwellings.

top

Further work – exploring sources of auxiliary data

We will continue to monitor developments in Canterbury and to explore sources of auxiliary data that we could use to improve the estimates.

One important source of auxiliary data that we know will be available is the data from the 2013 Census. This will give us detailed regional information on the sizes of the PSUs in the whole country and enable us to update our demographic projections.

We will update the survey once the 2013 Census data is available.

Conclusion

Despite the Canterbury earthquakes and the associated changes, the HLFS will continue to provide unbiased estimates of the key survey variables.

Statistics NZ will continue to monitor the impacts of the Canterbury earthquakes on the HLFS, and in particular, whether there is auxiliary data that could help improve the Canterbury estimates. Most likely, we will make no changes to the survey methodology until the results of the 2013 Census are available and we revise the whole sample.

For more information contact:
Daniel Griffiths
Phone: 04 931 4600
Email: Info@stats.govt.nz

Published 30 October 2012  

  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+
Top
  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+