Stats NZ has a new website.

For new releases go to

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+
Imputation study for Household Labour Force Survey March 2011 quarter

The Household Labour Force Survey (HLFS) technical notes for the March 2011 quarter release said that Statistics New Zealand would investigate revising the methodology used to account for non-response in the March 2011 quarter once data was available from the June 2011 quarter. The usual methodology was temporarily altered in the March 2011 quarter because of the unusual high non-response in the Canterbury region, due to the earthquake on 22 February 2011. The response rate for the Canterbury region fell to 55 percent in the March 2011 quarter.

Analysis of time series for various estimates showed no evidence of introduced bias in the estimates calculated under the changed methodology for the March 2011 quarter. However, the published March 2011 quarter estimates did not necessarily reflect the effect of the earthquake as there was little data collection in the Canterbury region after the earthquake. Canterbury respondents had their selection weights increased to account for the region's non-respondents, however most of these respondents had been interviewed before the earthquake. If the labour force information received from respondents before the earthquake was likely to have changed after the earthquake then this method might not show the impact the earthquake had on the Canterbury region in the March 2011 quarter.

Once data was available from the June 2011 quarter, we began investigating whether we could revise the methodology using imputation methods to reflect earthquake-related changes in the Canterbury labour force. Imputation is a type of error treatment that deals with data that is unlikely to be correct, or as in this case, where data is missing.

Imputation method

The focus was to use data from the December 2010, March 2011, and June 2011 quarters to inform us of likely responses for those Canterbury dwellings that we did not interview post-earthquake in the March 2011 quarter. The December and June quarters data built a better picture of whether respondents were likely to have been interviewed under normal circumstances. Additionally, the June 2011 quarter data for the ‘unemployed’ and ‘not in the labour force’ respondents gave us information about how long since they had last worked, how long they had been looking for work, and what their main activity was last week. Comments they provided also gave us an insight into whether they had left their most recent job because of the earthquake. This type of information helped to inform us of the likely labour force status to impute for those respondents with missing March quarter information.

Studying how Canterbury respondents labour force status changed over the December 2010, March 2011, and June 2011 quarters allowed us to use these transition probabilities to impute a labour force status for all the remaining Canterbury respondents who we had labour force information for in the June 2011 quarter, but no labour force status imputed yet for the March 2011 quarter.

The imputed data was only retained in our analysis where we had managed to impute some information for the whole household. Using this approach we imputed information for 57 percent of the Canterbury dwellings that we did not interview in the March 2011 quarter.

Impact on estimates and sample errors

We investigated the impact of imputed March 2011 data on the key unadjusted and seasonally adjusted published estimates, the sample errors, and the quarterly and annual movements for both the March and June 2011 quarters. All the differences in level estimates for the imputed and original data were within sample error. The sample errors had minimal or no change.

At this point, given that the imputed data was not impacting significantly on the estimates and movements, we decided that there was no benefit in extending the study to impute more of the Canterbury dwellings using other methods, such as donor techniques. We will not be incorporating the imputation methodology into any of our published or available HLFS outputs.

The imputation study has not identified any significant change in the Canterbury labour market immediately after the earthquake. The published June quarter 2011 estimates will have captured any ongoing impact, as the methodology for non-response returned to the usual method, and the response rate for Canterbury in that quarter increased to 81 percent.

For more information contact:
Emma Bentley
Wellington 04 931 4600

Published: October 2011

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