|Page tools: Print Page Print All|
Introduction of Standardised Imputation Methodology for the Building Activity Survey
Investigations have shown that relatively simple methods perform to the same standard as more involved methods - including the current method, which uses a complex regression model not able to be performed in standard corporate infrastructure. In particular, the Respondent Mean method was recommended. This method calculates imputes based on the mean values of respondents with similar characteristics to the non-respondents, initially at a very fine level, and falling back to broader classes if required. Only two data items (probability of commencement, and proportion of work done) are imputed, and all other data items are able to be derived from these together with other information already known about the non-respondent. This method results in revisions between preliminary and final estimation that are distributed very similarly to the current method. Parallel estimates produced over a number of quarters show that the resulting estimates closely aligned (within two standard errors) with no evidence of any systemic bias or change in bias.
Recently, Business Statistics Methodology, the Construction Business Statistics Centre, and Technology Applications Delivery have been working together to refine, specify, test and implement the new methodology, which has resulted in some minor changes, such as introducing an additional fallback stage to make the imputes more robust. The new method is expected to go live for the September 2015 cycle providing that final testing is passed. The transition to the new method will be closely monitored to ensure it is robust to any shift in respondent behaviour and that published estimates continue to be of the same high quality.
These documents will be presented in a new window.