Technical Report: Changes to coding processes for industry and occupation in the Labour Force Survey
The ABS has changed the processes used to code industry and occupation data in the Labour Force Survey (LFS). These changes were introduced for the May 2005 survey.
Although no changes have been made to the classifications used, and the underlying coding methodology is unchanged, the changes to coding processes will result in an improvement in the quality of estimates classified by industry and occupation from May 2005 onwards. Aggregate estimates of employment and unemployment are unaffected, and there is no meaningful change in the level of employment classified to any industry Division or occupation Major Group.
The following describes the changes in coding processes.
LFS industry data are classified according to the Australian and New Zealand Standard Industrial Classification (ANZSIC), 1993 (cat. no. 1292.0) and are coded at the ANZSIC Group (3 digit) level. These data are collected in the mid month of each quarter (i.e. February, May, August, and November).
The LFS currently asks each employed person to provide, for their main job, a description of the industry, business or service where they work, and the name and address of their employer or business.
In the LFS, occupation data are collected for all employed people in the mid-month of each quarter. Responses are classified according to the ASCO - Australian Standard Classification of Occupations, Second Edition (cat. no. 1220.0), at the ASCO Unit Group (4 digit) level.
The LFS asks each employed person to describe, for their main job, the job title and their main tasks or duties.
Computer assisted coding
The computer assisted coding (CAC) system was used for coding all records prior to May 2005. In this system, the basic and qualifying words from the details supplied are typed in and a computer searches a list to find a match to an industry or occupation. If no unique match can be made, the coder will be presented with a list of possible matches. The computer also indicates the steps required to make a match. The coder then selects the best match and the computer assigns the appropriate industry or occupation classification code.
From May 2005, industry and occupation codes are coded automatically by a computer matching the survey responses to an industry or occupation index. Where the autocoding (AC) system is unable to allocate a valid code to a record, the record is then passed on to the CAC system for coding. The AC system saves time, and improves accuracy and consistency.
IMPACT ON ESTIMATES
This change in procedure has the potential to affect estimates of employment and unemployment classified by industry or occupation.
In order to analyse the effect of the change, dual coding using both methods was conducted on the August 2004, November 2004, and February 2005 surveys. Results of analysis conducted on these surveys suggest that AC produces a valid code in about 63% of records for occupation and 58% of records for industry. The remainder of the records are coded using CAC.
Of those records coded by AC during the dual coding exercise, approximately 86% of occupation codes and 90% of industry codes had been allocated the same codes as CAC. Of those records where AC and CAC had allocated different codes, AC achieved lower error rates (1.8% for industry and 3% for occupation) compared to the CAC system (8% for industry and 11% for occupation). Records were treated as being incorrect if the same code would not have been obtained had the coding rules been strictly followed.
Analysis of estimates of employed people at the industry Division and occupation Major Group level showed that in a small number of cases there were statistically significant differences. These differences were inconsistent across the months analysed, and were so small and variable, that application of adjustment factors is not warranted.
'Not further defined' codes
Where there is insufficient detail collected from the survey respondent to allocate the most detailed level of occupation or industry code, 'not further defined' codes (also known as 'nfd' codes) are used. The incidence of occupation and industry nfd codes was reduced by the use of the AC system over the three months in the dual coding analysis. Industry nfd code use was reduced by approximately 15% under AC, while the use of occupation nfd codes was reduced by approximately 23% under AC.
Between February and May 2005, the incidence of nfd codes in estimates of employed people classified by industry decreased by 28% from 588,200 to 424,900 people. Over the same period, the incidence of nfd codes in occupation estimates decreased by 39%, from 234,600 to 142,100 people.
For more information about Labour Force Survey estimates classified by industry or occupation, contact the Assistant Director, Labour Household Surveys on Canberra 02 6252 6565. For email enquiries, please contact Client Services on client firstname.lastname@example.org.