Online Appendix to

Estimating Poverty Transitions Using Repeated Cross-Sections:

A Three-country Validation Exercise

Guillermo Cruces, Peter Lanjouw, Leonardo Lucchetti, Elizaveta Perova,

Renos Vakis and Mariana Viollaz

1. Additional details on data sets

Chile (CASEN Survey)

We use the 1996, 2001, and 2006 CASEN Panel survey. The CASEN survey is carried out jointly by the Foundation for Overcoming Poverty (FSP), Ministry of Planning (Mideplan) and Social Observatory of the Universidad Alberto Hurtado (OSUAH) and its main objective is to study poverty dynamics and vulnerability.The first round of the CASEN panel survey interviewed 5,209 households between November and December 1996 in the Third, Seventh, Eighth, and Metropolitan regions of Chile, representing approximately 60 percent of total population. The second and third rounds interviewed 4,648 and 3,769 households respectively between November and December of 2001 and 2006. The survey includes information mainly on education, employment, income, health, labor history, participation, and housing.

Nicaragua (ENMV Survey)

To study transitions in and out of poverty in Nicaragua we use the 1998, 2001, and 2005 rounds of the EMNV survey which includes a rotating panel. The survey was developed by the National Institute of Statistics and Censuses with the technical and financial assistance of the World Bank, the United Nations Development Programme (UNDP), the Inter-American Development Bank (IDB), and the government of Nicaragua. The main objective of the EMNV survey is to study the socio-economic characteristics and the living conditions of the population of Nicaragua.

The first round of the rotating panel component of the EMNVsurvey interviewed 3,136 households and has national coverage. The survey was fielded between May and July 1998 and provides information on family relationships, education, health, economic activity, time, housing, consumption, household enterprise, and agro-pastoral activities. The second round of the EMNV was fielded between May and July 2001. The rotating panel component includes3,112households; while the last round was fielded between July and October 2005 and has a panel sample of 3,735 households (it includes panel households from all three rounds as well as for the 2001-2005 period). The sample of the rotating panel that includes information for all three years is 2503 households, which is the sample we use for the validation work.

Peru (ENAHO Survey)

In order to estimate poverty transitions in Peru we use the 2008 and 2009 ENAHO survey, which was developed by the Peruvian Statistics Bureau (INEI). The ENAHO is a nationally representative survey yielding rich information on education, employment, income and expenditure, health, participation, social programs, housing, and perceptions. The survey’s main objectives are to measure poverty evolution and households’ living conditions.

The first round of the panel survey interviewed 7,560 households from January to December 2008, while the second round of the survey interviewed 7,546 households from January to December 2009.

2. Models used for the upper and lower bound estimates

Table A1: Log Income Estimates
Chile 1996
Dependent Variable: Log Income / [1] / [2] / [3]
Male / -0.394 / -0.369* / -2.627
[0.297] / [0.211] / [1.637]
Age in 1996 / 0.033 / 0.025 / 1.918***
[0.091] / [0.081] / [0.599]
Age Squared in 1996 / 0.000 / 0.000 / -0.022***
[0.001] / [0.001] / [0.007]
Years of Education / 0.124*** / 0.102*** / -0.026
[0.018] / [0.017] / [0.169]
% of Female Head of HHs / -1.057 / 4.578
[0.646] / [12.862]
% of Head of HHs With Incomplete Primary Education / -0.722 / 21.283
[1.048] / [15.864]
% of Head of HHs With Incomplete Secondary Education / -2.398** / 32.029**
[1.015] / [13.693]
% of Head of HHs With Complete Secondary Education or More / -0.213 / 58.139***
[1.205] / [21.883]
Region Fixed Effects / Yes / Yes / Yes
Interaction Terms / No / No / Yes
Number of HHs / 824 / 824 / 824
Data source: CASEN, 1996.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Column [1] shows a simple model with household time invariant characteristics. Column [2] adds sub-national controls from census and region of birth fixed-effects. Column [3] adds interactions between household time invariant characteristics and sub-national controls.
Table A2: Log Consumption Estimates
Nicaragua 2008
Dependent Variable: Log Consumption / [1] / [2] / [3]
Male / -0.078 / 0.035 / -1.834
[0.060] / [0.057] / [2.955]
Age in 2008 / -0.058** / -0.059*** / -0.075
[0.025] / [0.022] / [1.040]
Age Squared in 2008 / 0.001*** / 0.001*** / 0.003
[0.000] / [0.000] / [0.012]
Years of Education / 0.093*** / 0.073*** / 0.450
[0.007] / [0.007] / [0.352]
Language: Spanish and Foreign / -0.155 / -0.326
[0.119] / [0.210]
% of Female Head of Households in 2001 / 0.924*** / 1.314
[0.338] / [7.619]
% of Head of HHs With Complete Primary Education in 2001 / 0.324 / -0.854
[0.351] / [8.151]
% of Head of HHs With Complete Secondary Education in 2001 / -0.461 / 14.252
[0.970] / [21.168]
% of Head of HHs With Complete Tertiary in 2001 / -0.148 / -11.455
[0.854] / [16.747]
% of HHs with Access to Electricity in 2001 / 0.402* / 2.389
[0.209] / [4.683]
% of HHs with Access to Water in 2001 / 0.146 / -2.883
[0.172] / [3.892]
% of HHs with Toilet in 2001 / 0.333 / 9.135*
[0.252] / [5.382]
Average Household Size in 2001 / -0.005 / 2.905***
[0.051] / [0.996]
Department of Birth Fixed Effects / Yes / Yes / Yes
Interaction Terms / No / Yes / Yes
Number of HHs / 685 / 685 / 685
Data source: EMNV, 2001.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Column [1] shows a simple model with household time invariant characteristics. Column [2] adds sub-national controls from census and region of birth fixed-effects. Column [3] adds interactions between household time invariant characteristics and sub-national controls.
Table A3: Log Consumption Estimates
Peru 2008
Dependent Variable: Log Consumption / [1] / [2] / [3] / [4]
Male / -0.222*** / -0.137*** / -0.865 / -0.789
[0.037] / [0.034] / [0.528] / [0.513]
Age in 2008 / -0.008 / -0.01 / 0.121 / 0.111
[0.012] / [0.011] / [0.219] / [0.223]
Age Squared in 2008 / 0.000** / 0.000** / -0.001 / -0.001
[0.000] / [0.000] / [0.002] / [0.002]
Years of Education / 0.085*** / 0.058*** / 0.105** / 0.077*
[0.003] / [0.004] / [0.049] / [0.047]
Language: Spanish and Foreign / 0.275*** / 0.196*** / 2.474 / 2.168
[0.033] / [0.042] / [3.521] / [3.887]
Percentage of Male Head of HHs / -1.497*** / 9.581* / 4.834
[0.309] / [5.805] / [5.617]
% of Head of HHs With Incomplete Primary Ed. / -0.630*** / -3.494 / -2.956
[0.198] / [3.760] / [3.547]
% of Head of HHs With Incomplete Secondary Ed. / 0.883 / 10.748 / 6.995
[0.567] / [10.819] / [10.286]
% of Head of HHs With Complete Secondary Ed. or + / 0.804*** / -5.71 / -3.3
[0.249] / [4.534] / [3.860]
% of HHs with Access to Electricity / 0.274*** / -0.159 / 0.112
[0.086] / [1.621] / [1.596]
% of HHs with Access to Water / 0.065 / 1.046 / 0.415
[0.078] / [1.329] / [1.236]
Department of Birth Fixed Effects / Yes / Yes / Yes / Yes
Interaction Terms / No / No / Yes / Yes
Asset Ownership in 2008 / No / No / No / Yes
Number of HHs / 2,280 / 2,280 / 2,280 / 2,280
Data source: INEI - Peruvian National Household Survey (ENAHO), 2008.
Note: Results are constrained to the panel sample of HHs whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Column [1] shows a simple model with household time invariant characteristics. Column [2] adds sub-national controls from census and region of birth fixed-effects. Column [3] adds interactions between household time invariant characteristics and sub-national controls. Column [4] adds retrospective variables (asset ownership).

3. Testing two transition lines – Chile and Nicaragua

Table A4: Transition Matrices With Two Lines- Repeated Cross Sections vs. Panel Data, Moderate and Twice the Moderate Poverty Lines
Chile 1996 and 2006
Status in 1996, 2006 / Lower Bound Estimates / Truth / Upper Bound Estimates
[1] / [2] / [3] / [4] / [3] / [2] / [1]
Poor, Poor / 7.46 / 6.68 / 5.35 / 4.64 / 2.61 / 2.66 / 3.04
(3.19, 6.07)
Poor, Vulnerable / 7.70 / 9.49 / 9.65 / 10.38 / 8.57 / 8.65 / 9.11
(8.293, 12.46)
Poor, Middle Class / 0.00 / 0.86 / 1.44 / 9.21 / 12.93 / 12.00 / 16.21
(7.23, 11.1)
Vulnerable, Poor / 0.13 / 0.92 / 1.25 / 2.31 / 1.70 / 2.38 / 2.41
(1.28, 3.34)
Vulnerable, Vulnerable / 19.09 / 11.90 / 9.67 / 9.61 / 6.60 / 7.02 / 9.00
(7.599, 11.62)
Vulnerable, Middle Class / 11.90 / 10.60 / 10.94 / 20.22 / 13.43 / 15.77 / 18.31
(17.47, 22.96)
Middle Class, Poor / 0.00 / 0.00 / 1.00 / 0.65 / 3.29 / 2.55 / 2.14
(0.097, 1.19)
Middle Class, Vulnerable / 0.61 / 6.00 / 8.07 / 7.40 / 12.22 / 11.71 / 9.28
(5.60, 9.18)
Middle Class, Middle Class / 53.11 / 53.56 / 52.64 / 35.58 / 38.65 / 37.24 / 30.50
(32.31, 38.85)
R-squared / 0.21 / 0.31 / 0.43 / . / 0.43 / 0.31 / 0.21
Residual Correlation / 0.26 / 0.22 / 0.22 / . / 0.22 / 0.22 / 0.26
Observations / 823 / 823 / 823 / 823 / 823 / 823 / 823
Data source: CASEN, 1996-2006.
Note: R-squared is calculated for opposite halves of the total 1996 sample. Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Columns [1] show a simple model with household time invariant characteristics. Columns [2] add sub-national controls and region fixed effects. Columns [3]add interactions between household time invariant characteristics and sub-national controls. Results in column [4] show actual panel mobility. 95 percent confidence interval between parentheses. Upper bound estimations are based on 50 repetitions. The poor are those living on less than the official poverty line, the vulnerable are those living between the official and twice the official poverty line, and the middle class are those living with more than twice the official poverty line.
Table A5: Transition Matrices With Two Lines- Repeated Cross Sections vs. Panel Data, Moderate and Twice the Moderate Poverty Lines
Nicaragua 2001 and 2005
Status in 2001, 2005 / Lower Bound Estimates / Truth / Upper Bound Estimates
[1] / [2] / [3] / [4] / [3] / [2] / [1]
Poor, Poor / 40.63 / 39.44 / 37.34 / 35.68 / 31.09 / 30.97 / 31.64
(32.08,39.26)
Poor, Vulnerable / 0.00 / 0.00 / 3.18 / 3.24 / 8.64 / 8.06 / 9.77
(1.91, 4.57)
Poor, Middle Class / 0.00 / 0.00 / 0.13 / 0.11 / 1.80 / 1.82 / 2.21
(0.13, 0.353)
Vulnerable, Poor / 21.17 / 22.13 / 23.00 / 22.56 / 21.08 / 21.17 / 20.77
(19.4, 25.6)
Vulnerable, Vulnerable / 17.27 / 16.17 / 16.52 / 11.10 / 10.42 / 9.87 / 9.50
(8.74, 13.4)
Vulnerable, Middle Class / 0.00 / 0.04 / 1.13 / 3.16 / 4.39 / 4.28 / 4.41
(1.85, 4.47)
Middle Class, Poor / 0.00 / 0.23 / 1.46 / 3.57 / 9.64 / 9.66 / 9.39
(2.17, 4.95)
Middle Class, Vulnerable / 8.27 / 9.37 / 5.85 / 11.20 / 6.49 / 7.61 / 6.28
(8.83, 13.5)
Middle Class, Middle Class / 12.65 / 12.61 / 11.40 / 9.38 / 6.47 / 6.55 / 6.04
(7.19, 11.5)
R-squared / 0.30 / 0.45 / 0.53 / . / 0.53 / 0.45 / 0.30
Residual Correlation / 0.58 / 0.53 / 0.52 / . / 0.52 / 0.53 / 0.58
Observations / 684 / 684 / 684 / 684 / 684 / 684 / 684
Data source: EMNV, 2001-2005.
Note: R-squared is calculated for opposite halves of the total 2001 sample. Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Columns [1] show a simple model with household time invariant characteristics. Columns [2] add sub-national controls from census and region fixed effects. Columns [3] add interactions between household time invariant characteristics and sub-national controls. Results in column [4] show actual panel mobility. 95 percent confidence interval between parentheses. Upper bound estimations are based on 50 repetitions. The poor are those living on less than the official poverty line, the vulnerable are those living between the official and twice the official poverty line, and the middle class are those living with more than twice the official poverty line.

4. Sensitivity analysis using different poverty lines

Figure A1: Poverty Dynamics – Repeated Cross Sections vs. Panel Data for Alternative Poverty Lines Chile 1996 and 2006
Data source: Chile CASEN 1996 and 2006.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Upper bound estimations are based on 50 repetitions.
Figure A2: Poverty Dynamics – Repeated Cross Sections vs. Panel Data for Alternative Poverty Lines
Nicaragua 2001 – 2005
Data source: EMNV 2001 and 2005.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. Upper bound estimations are based on 50 repetitions.

5. Sub group Poverty Dynamics Estimates

Figure A3: Poverty Dynamics – Repeated Cross Sections vs. Panel Data by Sub-population Groups
Chile 1996 and 2006
Data source: CASEN, 1996-2006.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. The 45 degree line shows actual panel mobility. Upper bound estimations are based on 50 repetitions.
Figure A4: Poverty Dynamics – Repeated Cross Sections vs. Panel Data by Sub-population Groups
Peru 2008 and 2009
Data source: INEI - Peruvian National Household Survey (ENAHO), 2008-2009.
Note: Results are constrained to the panel sample of households whose heads are between 25 and 65 years old. Results are weighted using household-level survey-sampling weights. The 45 degree line shows actual panel mobility. Upper bound estimations are based on 50 repetitions.

1