Presentation is loading. Please wait.

Presentation is loading. Please wait.

Producing monthly estimates of labour market indicators exploiting the longitudinal dimension of the LFS microdata R. Gatto, S. Loriga, A. Spizzichino.

Similar presentations

Presentation on theme: "Producing monthly estimates of labour market indicators exploiting the longitudinal dimension of the LFS microdata R. Gatto, S. Loriga, A. Spizzichino."— Presentation transcript:

1 Producing monthly estimates of labour market indicators exploiting the longitudinal dimension of the LFS microdata R. Gatto, S. Loriga, A. Spizzichino Istat, Italian Statistical Institute Labour Force Survey unit Silvia Loriga NTTS 2009 - Bruxelles 19-20 February 2009

2 NTTS 20092 LFS is a continuous survey; main results are produced on quarterly basis Eurostat currently releases monthly unemployment estimates based on LFS data (national and EU level); methodologies are chosen by each NSI Italy is still not producing monthly unemployment estimates (Italian figures are quarterly estimates) From 2007 a study project is being conducted in Istat (supported by an Eurostat grant). The object is: to study a proper methodology to produce robust monthly estimates based exclusively on LFS data (no administrative sources on unemployment are available in Italy) to deal with the complexity of producing monthly estimates on regular basis, no later than 3 weeks after the end of each month (partial sample)

3 NTTS 20093 Methods used by Member States Monthly LFS estimates Germany, Finland, Sweden 3-months moving averages LFS estimates Netherlands, UK, Norway Quarterly LFS estimates + monthly registered unemployment with Chow Lin model Portugal Linear extrapolation of registered unemployment + benchmark to LFS Belgium, Bulgaria, Czech Republic, Denmark, Estonia, Ireland, Spain, France, Latvia, Lithuania, Hungary, Poland, Slovenia, Slovakia, Cyprus, Malta, Austria, Luxembourg, Croatia

4 NTTS 20094 Estimators comparison in the Italian study CONTEXT External sources on monthly unemployment not available (use of registered unemployed or Chow Lin)… COMPARISON 1.The direct monthly estimator from LFS (the LFS sampling design has a monthly stratification) 2.The regression composite monthly estimator from LFS 3.The 3-months moving average estimator from LFS (only if 1 and 2 are not satisfactory) CRITERIA Sampling error, robustness, time series decomposition, coherence with quarterly estimates

5 NTTS 20095 The Italian sample rotation scheme 2-(2)-2 households participate to the survey 4 times during a 15 months period 50% overlap between following quarters 50% overlap between a quarter and the same quarter of the previous year The sample follows a monthly stratification: 50% overlap between months t and t-3 50% overlap between months t and t-12 Q1_Y1Q2_Y1Q3_Y1Q4_Y1Q1_Y2Q2_Y2 JanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJun

6 NTTS 20096 The regression composite estimator (RCE)1 It is a kind of composite estimator Direct estimators: design based, only observed sample information Composite estimators: model based, observed sample + additional information It may be applied to repeated surveys with partially overlapping samples The idea: as the employment status observed in a previous point in time is correlated with the current employment status, using it as auxiliary variable will improve the estimation It is based on the regression of the usual cross-sectional estimator on a set of predictors computed on the overlapping sub-sample from previous time points Level and changes estimates are improved

7 NTTS 20097 The regression composite estimator (RCE)2 Singh et al. developed for the Canadian LFS a kind of RCE in which micro-level past information are used as predictors, through calibration Calibration estimator: weights w k are computed in two steps: 1: initial weights d k are obtained as the inverse of the inclusion probabilities 2: final weights w k are obtained solving the following minimization problem under constraints: solved through an iterative procedure

8 NTTS 20098 The regression composite estimator (RCE)3 The calibration estimator is the estimator currently used to produce quarterly LFS estimates in Italy In the regression composite version of the calibration estimator developed to produce monthly estimates for the Italian LFS additional auxiliary variables have been introduced: the employment status observed at the individual level 3 months ago and 12 months ago (only for individuals in the overlapping sub-sample) constraints are derived from the previous estimates referred to 3 months ago and 12 months ago obtained using this same estimator (X overlapping %) No longitudinal auxiliary variables (no constraints) for the non overlapping sub-sample (no imputation as Canadian LFS estimator)

9 NTTS 20099 The regression composite estimator (RCE)4 Monthly estimator similar methodology as the quarterly estimator; good for coherence Based on weights computation; good to produce consistent estimates of different variables Design based through the use of inclusion probabilities in weights computation Micro-level model based through the use of individual auxiliary variables It exploit the longitudinal dimension of the sample: Considering 3 months ago and 12 months ago information we are adding information on both short and long term period Estimation of changes at t-3 and t-12 improves More robust estimation of both trend and seasonality

10 NTTS 200910 Dealing with partial sample1 According Reg. 577/98 interviews can be done until 5 weeks after each reference week Quarterly data have to be transmitted to Eurostat within 12 weeks of the end of the quarter Monthly estimates have to be produced and transmitted no later than 3 weeks after the end of each month Monthly estimates have to be computed over a partial sample Higher sampling errors Not at random: mode (capi/cati), reference week within the month, household typologies, employment status Bias Variability over time Higher distance with final monthly and quarterly estimates

11 NTTS 200911 Dealing with partial sample2 1st quarter 2008 weekly data

12 NTTS 200912 Dealing with partial sample3 2nd quarter 2008 weekly data

13 NTTS 200913 Dealing with partial sample4 From 4th quarter 2007 to 3rd quarter 2008 monthly data

14 NTTS 200914 Dealing with partial sample5 From 4th quarter 2007 to 3rd quarter 2008 monthly data

15 NTTS 200915 Dealing with partial sample6 To correct for bias due to the partial sample, additional constraints have been included in the calibration procedure: Households by 4 rotation groups Households by mode (capi-cati) Households by reference week (4 or 5 in each month) Constraints are derived from the theoretical sample By this way bias due to the partial sample is partially corrected and estimates over partial samples are closer to final estimates over the whole sample

16 NTTS 200916 Legenda QCE(Q): Quarterly calibration estimator - per quarter (the usual quarterly estimator) QCE(M): Quarterly calibration estimator - per month (the usual quarterly estimator applied to single months) MCE(M): Monthly calibration estimator (a calibration estimator similar to the usual one, computed for each month) MRCE(M): Monthly regression composite estimator (a calibration estimator similar to the usual one + constraints on condition at t-3 and t-12, computed for each month)

17 NTTS 200917 Employment estimates

18 NTTS 200918 Unemployment estimates

19 NTTS 200919 Erratic component Employment estimates

20 NTTS 200920 Erratic component Unemployment estimates

21 NTTS 200921 Main references As the regression composite estimator Survey Methodology – Volume 27, Number 1, June 2001 Special section on Composite Estimation A.C. Singh, B. Kennedy and S. Wu Regression Composite Estimation for the Canadian Labour Force Survey with a Rotating Panel Design and following papers by W.A. Fuller and J.N.K. Rao P. Bell J. Gambino, B. Kennedy and M.P. Singh As the calibration estimator J.C. Deville and C.E. Särndal (1992) Calibration Estimators in Survey Sampling JASA, vol. 87

22 NTTS 200922 Annex 1 Quarterly calibration constraints

23 NTTS 200923 Monthly calibration constraints

24 NTTS 200924 Annex 2 QCE(M)_emp

25 NTTS 200925 MCE(M)_emp

26 NTTS 200926 MRCE(M)_emp

27 NTTS 200927 QCE(M)_une

28 NTTS 200928 MCE(M)_une

29 NTTS 200929 MRCE(M)_une

Download ppt "Producing monthly estimates of labour market indicators exploiting the longitudinal dimension of the LFS microdata R. Gatto, S. Loriga, A. Spizzichino."

Similar presentations

Ads by Google