Identifying and adjusting for under reporting in the Survey of Freight Movement
by
Paul Sutcliffe
Australian Bureau of Statistics
Coauthors: Brett Frazer, Andrew Sutcliffe

The Survey of Freight Movement collects trip data over a 2 week period from a random sample of owners of articulated vehicles. While every effort was made by the survey practitioners to reduce non-sampling error under reporting in the second week was identified, measured and a correction made. This paper presents the TRAMO/SEATS time series regression techniques used to analyse the 12 months of daily data used to determine that the difference between the data reported in week one was statistically significant from week 2. The magnitude of the under reporting varied for different data items and ranged between -4.6% for the number of laden trips to -10.2% for weight carried. A number of methods for making the adjustments to the survey outputs were considered. The final choice of a direct imputation method for missing trip records was chosen because of it appropriateness to account for the identified non-sampling error, the time available prior to publication, and the ability to incorporate the method into existing systems. In order to optimize the impact of the imputation method an analytical approach was used assess the best choices of imputation classes. The final outcomes was the hot deck imputation of 9,647 trip records (5%) used to account for the under reporting which accounted for the assessed differences of four of the key data items.

Date received: April 23, 2002