Download presentation

Presentation is loading. Please wait.

Published byAubrie Dolphin Modified over 2 years ago

1
Federal Planning Bureau Economic analyses and forecasts Federal Planning Bureau Economic analyses and forecasts On weights in dynamic- ageing microsimulation models Some work in progress Gijs Dekkers 1 and Richard Cumpston 2 1. Federal Planning Bureau and Katholieke Universiteit Leuven 2. Australian National University Paper presented at the Ministero dell'Economia e delle Finanze, Rome, February 15 th, 2011

2
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models This work is confidential and under embargo until June 8 th, 2011

3
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

4
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

5
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Why weights? Datasets are often used to assess trends of aggregated units. So, they need to contain unbiased and credible sample estimators on population parameters. This need for representativeness is however hampered by bias caused by differential cross-sectional selection probabilities non-response

6
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

7
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models An obvious solution: transform the probability weights in frequency weights and expand the dataset...

8
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models

9
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Drawback: -Expanding is inefficient, because it ultimately means simulating the entire population. -Use standardized weights, but: -Can one expand using standardized weights? -I have my doubts on the way in which standardized weights are derived. - Sampling to round the weights introduces sampling variance, which may be more important than the rounding error (this certainly is the case with standardized weights).

10
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

11
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models An alternative strategy: using weights as a simulation variable in the model The method presented in this paper involves the partial expansion or “splitting up” of individual weighted households in case of moves of individuals in between households of different weights.

12
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models An example: suppose two households X and Y. Both households consist of two individuals, denoted X1, X2, Y1 and Y2. Suppose that individuals X2 and Y2 fall in love and form a new household, say, Z. What frequency weight should this household get? Case a: the frequencies of households are equal Case b: the frequencies of households are unequal: F(1)=2 and F(2)=3

13
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models When the frequency weights of the two ‘donating’ households differ, the household with the highest frequency is expanded to two households. And then the merge is done with equal frequency weights. ‘Donating’ household 1 (F1)=3 household 2 (F2)=2 ‘Donating’ household 1 (F1)=1 ‘Donating’ household 1 (F1)=2 MERGE household 1 (F1)=1 Merged household 3 (F3)=2

14
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models

15
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

16
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Weights and alignment: Rank according to risk 3. Select the first # individuals, #=S x auxiliary proportion

17
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Weights and alignment: some solutions Strategy 1: split up the last household Strategy 2: select a household for alignment so that there is no mismatch Strategy 3: iteratively reduce mismatch - the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration.

18
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models

19
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Overview of this presentation What is the problem? A simple solution (which does not really work) A proposed method of using weights in dynamic-ageing MSM’s Weights and alignment Some empirical results on Australian data

20
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Unweighted unit records: 2001 Australian Household Sample Survey (HSF), Unweighted sample size of about 175,000. Weighted unit records: Australian Survey of Income and Housing Costs (SIHC), These files covered 16,824 persons, grouped into 6,786 households. Household weights in the SIHC sample were intended to replicate the Australian population of about 19.4m. To give an unweighted sample size of about 175,000, the weights were multiplied by and rounded to the nearest integer. household microsimulation model (Cumpston 2009). Using the aforementioned datasets HSF and SHIC as the starting point, the Cumpston model was ran in its original and weighted form for the years Alignment was done using random selection, using strategy 3.

21
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models

22
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models The total efficiency gain depends on the average initial size of the weight, and the speed of the convergence process.

23
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

24
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

25
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

26
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

27
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

28
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

29
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Conclusions So far, there are no efficient ways in which dynamic MSM’s can include weights. This method uses weights as ‘just another’ variable in the model. It prevents losses in efficiency involved in expanding the starting dataset. This paper proposes three methods for alignment of weighted data It applies an iterative procedure where the ‘overshooting’ individual in each iteration is taken out of the pool before starting the next iteration. Efficiency gains may be quite considerable, though limited to the first few decades.

30
Federal Planning Bureau Economic analyses and forecasts On weights in dynamic-ageing microsimulation models Thank you

Similar presentations

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google