Katrin Baumgartner Angelika Meraner Alexander Kowarik Statistics Austria
Katrin Baumgartner Angelika Meraner Alexander Kowarik Statistics Austria Rome 16 May 2014 www. statistik. at Longitudinal Weights for the Production of Transitions and Flow Estimates 9 th Workshop on Labour Force Survey Methodology We provide information
Introduction • Longitudinal Dimension of the Austrian LFS • Bias Analysis • Imputation • Weighting • Summary www. statistik. at slide 2 | 10/30/2020
Longitudinal Dimension of the Austrian LFS • Is not in focus so far • Depends on the survey design • Only available for a subset of LFS sample • Every quarter: exchange of 1/5 of the survey • Household stays 5 consecutive quarters in the survey www. statistik. at slide 3 | 10/30/2020
Longitudinal Dimension of the Austrian LFS • Cross sectional point of view: quarter consists of 5 waves (rotation numbers) • Longitudinal point of view: a rotation number appears in 5 quarters • Q-Q-Changes: approximately 4/5 of the LFS sample can be used • Y-Y-Changes: approximately 1/5 of the LFS sample can be used www. statistik. at slide 4 | 10/30/2020
Longitudinal Analysis Can be based on different subsamples: • One removes all incomplete cases for the flow analysis, i. e. flows are based on the subsample of persons who are successfully surveyed in both quarters, q(t) and q(t+1 ) (immobile persons). • All persons who are successfully surveyed in one quarter, q(t) or q(t+1) and do not regularly rotate in or out, are used (mobile + immobile persons). Potentially missing information of the second q(t+1) and first quarter q(t) respectively is imputed. www. statistik. at slide 5 | 10/30/2020
Bias Analysis Are there any differences between mobile and immobile persons in their sociodemographic structure? • Linking to administrative data from Central Population Register • Information of mobile persons is only available for one quarter: first q(t) or second q(t+1). www. statistik. at slide 6 | 10/30/2020
Bias Analysis (Q 1 2012) Demographic Characteristics in % Mobile Immobile Total Male Female 40. 5 59. 5 52. 0 48. 0 31. 8 68. 2 49. 5 50. 5 32. 4 67. 6 49. 7 50. 3 AT Non-AT Total (in 1 000) 83. 2 16. 8 1, 759 90. 5 9. 5 21, 766 89. 9 10. 1 23, 525 Age Sex 15 -34 35 -64 Nationality S. : Microcensus-LFS 2012 Q 1 2012. – Unweighted sample population in Q 4 2011 and Q 1 2012 (aged 15 to 64). – Persons with administrative identifier (b. PK). www. statistik. at slide 7 | 10/30/2020
Bias Analysis (Q 1 2012) Demographic Characteristics Age Sex Nationality in % Mobile Outflux Influx 15 -34 35 -64 40. 5 59. 5 45. 1 54. 9 34. 5 65. 5 Male Female 52. 0 48. 0 52. 4 47. 6 51. 3 48. 7 AT N-AT 83. 2 16. 8 1, 759 82. 8 17. 2 1, 005 83. 7 16. 3 754 Total (in 1000) S. : Microcensus-LFS 2012 Q 1 2012. – Unweighted sample population in Q 4 2011 and Q 1 2012 (aged 15 to 64). - Persons with administrative identifier (b. PK). www. statistik. at slide 8 | 10/30/2020
Bias Analysis (Q 1 2012) S. : Microcensus-LFS 2012 Q 1 2012. – Unweighted sample population in Q 4 2011 and Q 1 2012 (aged 15 to 64). Register labour status for mobile and immobile persons with administrative identifier (b. PK). www. statistik. at slide 9 | 10/30/2020
Bias Analysis – Flows (Q 4 2011 + Q 1 2012) 100% 1. 8 2. 9 3. 3 3. 9 13. 8 19 80% 51. 3 60% 95. 4 40% 50 93. 6 88. 8 92. 9 → Inactive → Unemployed →Employed 20% 34. 9 31 0% Immobile Mobile Employed Immobile Mobile Unemployed 1. 8 4. 6 3. 1 8. 1 Immobile Mobile Inactive S. : Microcensus-LFS 2012– Unweighted sample population in Q 4 2011 and Q 1 2012 (aged 15 to 64). Register labour status for mobile and immobile persons with administrative identifier (b. PK). www. statistik. at slide 10 | 10/30/2020
Imputation Proportions of imputed missing values for longitudinal data sets: Longitudinal Data Q 4 2011 & Q 1 2012 Q 4 2011 Q 1 2012 & Q 2 2012 Q 1 2012 Q 2 2012 & Q 3 2012 Q 2 2012 Q 3 2012 & Q 4 2012 Q 3 2012 Q 4 2011 & Q 4 2012 Q 4 2011 Q 4 2012 www. statistik. at Imputed missing values in % 3. 9 4. 4 3. 9 4. 5 3. 2 4. 5 3. 4 3. 8 7. 6 9. 6 slide 11 | 10/30/2020
Imputation Random hot deck imputation of important labour market characteristics Selection of domain variables (max. 7) Based on bias analysis • Gender, age, nationality • Administrative labour status (if available) of both quarters/years • ILO labour status of `complementary´ quarter/year (if administrative labour status not available) Longitudinal concept Currently imputed variable pertaining to `complementary´ quarter/year Multinomial logit model (forward selection) 7 th (last) domain variable for cases with no administrative labour status www. statistik. at slide 12 | 10/30/2020
Weighting: Longitudinal Weights • www. statistik. at slide 13 | 10/30/2020
Weighting: Longitudinal Weights Two versions of weights: 1. 2. Reducing the bias: not calibrating against the ILO labour market status Providing consistency between stocks and flows: additionally calibrating against the ILO labour market status (LMS adj) Key figures for ILO labour market status stem from published quarterly results of the microcensus -> projected data www. statistik. at slide 14 | 10/30/2020
Weights 1: Reducing the bias Base weights calibrated against marginal totals for q(t) and q(t+1) consecutively 1. Population by NUTS-2 region, sex and age 2. Population by NUTS-2 region and nationality 3. Weights corresponding to people born, deceased, immigrated or emigrated in q(t) are calibrated against the natural population change and the migration statistics www. statistik. at slide 15 | 10/30/2020
Weights 2: Consistency of stocks and flows Base weights calibrated against marginal totals for q(t) and q(t+1) consecutively 1. Population by NUTS-2 region, sex and age 2. Population by NUTS-2 region and nationality 3. Population by nationality, sex, age and ILO labour status 4. Weights corresponding to people born, deceased, immigrated or emigrated in q(t) are calibrated against the natural population change and the migration statistics www. statistik. at slide 16 | 10/30/2020
Comparison Q 4 2011 - ILO Labour Status 80 71. 8 72. 3 70 60 50 weights 1 40 30 24. 8 24. 5 weights 2 (LMS adj) 20 10 3. 4 3. 3 00 Employed Unemployed Inactive Comparison of weighting options 1 and 2 for cross-sectional data Q 4 2011 of population living in private households aged 15 -64 without persons doing their military or civilian service according to ILO labour status. www. statistik. at slide 17 | 10/30/2020
Comparison 100. 0000 90. 0000 4. 9 1. 6 5. 2 2. 0 23. 2 20. 9 21. 9 28. 1 80. 0000 70. 0000 60. 0000 50. 0000 40. 0000 93. 5 78. 0 80. 4 92. 8 → Inactive → Unemployed 30. 0000 54. 9 20. 0000 51. 0 10. 0000 → Employed 4. 2 3. 7 17. 8 16. 0 weights 1 weights 2 (LMS adj) Employed Unemployed Inactive Comparison of weighting options 1 and 2 for flows corresponding to longitudinal data comprising Q 4 2011 and Q 4 2012, i. e. the population living in private households at both time points, aged 15 -64 and not doing their military or civilian service according to ILO labour status. www. statistik. at slide 18 | 10/30/2020
Summary • Obvious differences between immobile and mobile persons for demographic characteristics and administrative labour status • Random hot deck imputation of missing data Ø Ø Longitudinal concept incorporated Use of administrative labour status • Weighting option 1 preferred Ø Ø Bias reducing Not adjusted to ILO Labour Market Status -> not providing consistency between stocks and flows www. statistik. at slide 19 | 10/30/2020
Please address queries to: Katrin Baumgartner Angelika Meraner Alexander Kowarik Contact information: Guglgasse 13, 1110 Vienna phone: +43 (1) 71128 -7211 katrin. baumgartner@statistik. gv. at www. statistik. at Thank you very much for your attention slide 20 | 10/30/2020
- Slides: 20