1. MODELING of SWIFT & Survey to Survey Imputation

2. SWIFT modeling Cross-validation to decide the optimal p-value
Run the stepwise regression using the optimal p-value determined by the cross-validation
Check the coefficients of the final model
Simulate household expenditures using PovMap or MI
Estimate poverty rates using the simulated expenditures and MI’s formula using “mi estimate”
To check stability, conduct backward imputation analysis

3. Steps for SWIFT modeling
Objective
Program
1. Cross Validation
Choose optimal p-value
crossvalidation.do
2. Finalization of models
Find a model with the optimal p-value
Simulation and estimation.do or
PovMap
3. Simulate household expenditure/income
Simulate household expenditures based on the above estimation Or
4. Estimate poverty statistics
Using the simulated expenditures or income, poverty rates are estimated using “mi estimate”
mi estimate

4. Data preparation You need to have y0 and y1
y0: data for developing models
Need both consumption and regression variables
GLSS6 with lnrpcexp
y1: data for estimating poverty rates using models
Need only regression variables that have the same definitions as Y0
SWIFT data
GLSS6 without lnrpcexp
Both data need to have all variables used in cross-validation, simulation, and estimation stages

5. crossvaliation2.do Choose a location of the dataset properly
If you want to change variable sets, you can change global macros for variable groups, such as location, etc.

6. What level do you want to do the modeling?
If you want to create a model for only urban areas or only one region, you need to modify the program here.
Note that you need to be consistent between cross-validation and simulation/estimation

7. Defining 10 folds
Randomly define 10 folds (subsamples)

8. Loop
Randomly define 10 folds (subsamples)

9. Result matrix A pe fold_ poor_ r2 mse pred_poor absdiff 0.01 1 0.348
0.426
0.382
0.302
0.046 2
0.316
0.427
0.243
0.248
0.067 3
0.225
0.434
0.245
0.310
0.085 4
0.106
0.442
0.228
0.233
0.126 5
0.238
0.439
0.274
0.281
0.043 6
0.344
0.460
0.273
0.249
0.095 7
0.358
0.271
0.307
0.131 8
0.154
0.452
0.264
0.244
0.090 9
0.456
0.279
0.318
0.010 10
0.304
0.436
0.252
0.052
0.02
0.386
0.321
0.027
0.437
0.240
0.294
0.022
0.466
0.250
0.293
0.068
0.450
0.221
0.236
0.129
0.445
0.270
0.303
0.065
0.392
0.286
0.284
0.155
0.462
0.278
0.098
0.467
0.299
0.009
0.232
0.300
0.004

10. Results of cross validation
Average of absolute differences between actual and projected poverty rates (absdiff)
Mean Squared Errors (mse)
2 percent is chosen since the massive increase in MSE after 2%

11. Simulation and estimation.do

12. Simulation and estimation.do

13. Simulation results
Over | Mean Std. Err. [95% Conf. Interval]
Actual |
Simulation |
Number of observation = 1016
Area = Region 8 Urban

14. Thank you! If you have any question, please let me know at nyoshida@worldbank.org