Presentation on theme: "WP2 workshop, NIESR, November 24-25, 2005 Volume measures of labour input."— Presentation transcript:
WP2 workshop, NIESR, November 24-25, 2005 Volume measures of labour input
Reconciling data from different sources Which source to use: establishment surveys, labour force surveys, other (social security statistics) Issues for discussion Options were to impose the same type of source on all partners or allow each to decide on the best source for their own country? In the latter can we adjust data to ensure definitions are comparable across countries?
Reconciling data from different sources UK example Large number of sources – Employment Census [AES] (establishment), Annual Business Inquiry [ABI] (establishment), Labour Force Survey [LFS] (individual), Social security data [SS] (individual) Data availability AES – from 1978; LFS – from 1984; ABI – from 1998; SS – 1970-1978 All series at least 2 digit SIC – some 3 digit Sufficient detail to generate full EUKLEMS series
Comparison LFS (primary jobs) and AES, ratio AES/LFS, average 1996-01 – 41 industries
Comparison LFS (primary jobs) and AES, ratio AES/LFS, average 1996-01,– 20 industries
Reconciling data from different sources: UK Attempt to redefine in terms of common definitions LFS allocate second jobs to industry where labour is employed - Mostly in services
Reconciling data from different sources: questions To what extent have consortium members found similar discrepancies between sources? Which source should be used? As control totals – NA if available, but what is this? To divide by industry – small sample sizes implies more variation LFS coefficient of variation significantly negatively correlated with sample size Should we combine data sources – one as control total for broad sectors and use shares of sub-sectors in broad sectors from another source to disaggregate How do we decide what is a small sample
Industry concordances Options for concording Optimal – get NSI to do it Consistent – construct weights based on data for an overlapping year Fudge – When data are not available for an overlapping year. Use whatever information is available to get an approximate concordance between industry, then use growth rates in another series to construct an overlapping year, to ensure no jumps
Industry concordances Consistent – construct weights based on data for an overlapping year Simplest case X Old SICNew SIC Y ZY Z X = Y + Z, so weights are Y/ (Y+Z) and Z/(Y+Z)
Industry concordances Consistent – Often more complicated X Old SIC New SIC Y ZY Z Set of simultaneous equations but may need interative procedures if sufficiently complicated As long as overlapping year data exist there should not be jumps in the data T W
Illustration of fudge method Time series for industry x break year t-1t
Industry concordances UK example – three SICs, 1968, 1980, 1992 LFS – no overlapping year AES some overlapping years, e.g. 1990-93 on both SIC80 and SIC92, but for detailed (3 digit industries) data only available for GB. Fudge – for LFS could use growth in AES for overlapping year to infer an overlapping year in LFS. (note levels in AES and LFS differ so cannot use AES weights applied to LFS)
Industry concordances Issues for discussion To what extent are industry concordances an issue? What methods have colleagues used to overcome problems? Can prodsys help?
Historical data – how to fill gaps Look for additional data – censuses, surveys If not available what are the options If earlier data are more aggregated then can assume growth in sub-industries equal growth in aggregate If no historical data available then what do we do? Assume growth rates the same as for aggregate economy? Assume growth rates the same as other variable in EUKLEMS dataset? Assume growth rate same as similar industry in similar country?
Data delivery Deadline for revised data January 15 Require prodsys readable form Important source of information Crucial for productivity calculations DOCUMENTATION Sources Assumptions