ADePT Automated DECs Poverty Tables

1
ADePT Automated DECs Poverty Tables

• Michael Lokshin, Zurab Sajaia and Sergiy Radyakin
• DECRG-PO
• The World Bank

2
Step 1 Data and Output
3
Step 2 Household variables
4
Step 3 Individual variable
5
Step 4 Tables and Graphs
6
Why to automate?

• To free resources for more meaningful and
  interesting tasks.
• Minimize human errors.
• Significantly speed-up production of basic
  results.
• Produce print-ready tables/graphs/reports
• Easily introduce new cutting-edge techniques and
  methods of poverty analysis.
• The automation tools could be used as valuable
  research instruments, tools for sensitivity
  analysis and educational tools.
• Might be helpful in situation of a limited data
  access
• Simple checking of the previous reports/results

7
Main Components of Poverty Assessment
Welfare Indicators
Poverty Lines
Poverty Assessment
8
Possibilities for automation Welfare indicator

• Low automation (even if standardized)
• High degree of subjective of the algorithm
• Should reflect country-specific characteristics
• Different countries require different algorithms
• Possible to automate some tasks, not the whole
  process
• Hedonic price regressions (housing prices)
• Flow of services from durable good consumption
• The economies of scale
• Imputation of expenditures from consumption of
  home-produced goods.

9
Possibilities for automation Poverty Line

• Moderate-to-high automation
• There is a standard World Bank methodology for
  deriving the poverty lines.
• Some subjective decisions need to be made, but
  most of them could be programmed as options in
  the algorithm.
• Could be an important sensitivity analysis,
  research and educational tool would allow fast
  comparison of poverty profiles under various
  assumptions.
• But the new poverty lines are calculated only
  once in several years.

10
Possibilities for automation Poverty Update

• High automation
• It is possible to define almost an exhaustive set
  of tables/graphs that are commonly used for
  poverty updates.
• Minimal requirements on the data
• Possibility to introduce an extensive set of
  controls and sensitivity tools.
• It is easy to integrate the latest methods into
  the report
• Production of print-ready tables/reports in very
  short time.
• Substantial budget savings

11
ADePT Data and Variables

• Accepts individual or household level data
• One or more years of data
• Required variables
• Household ID
• Consumption aggregate per person or per
  equivalent adult
• Poverty line up to two lines, numbers or
  variables
• Urban-rural indicator
• Optional variables
• Regions
• Weights
• Land-ownership
• Income
• Relation to the head
• Age
• Gender
• Education
• Employment Status
• More could be added

12
ADePT Checks and filters

• All variables are checked
• Correct type of variables
• Correct values (e.g., gender has only 2 values).
• Presence of a variable in all data files.
• Variable consistency over the years of data
• All the constructed variables are generated
  automatically household size, shares of
  different age/gender groups, etc.
• The program produces report with basic statistics
  on all variables.
• Possible control for influential outliers in
  terms of values or observations.

13
ADePT Tables and Graphs

• Tables and graphs are selected based on PA from
  Bulgaria, Bangladesh, Honduras, Georgia, Jordan,
  Mongolia, Nepal, Sri Lanka, Ukraine
• The program automatically generates the list of
  tables/graphs that could be produced based on the
  defined variables.
• Three versions of each table actual table, table
  with standard errors, table with frequencies in
  each cell.
• Users can apply IF conditions and change titles
  of the tables/graphs.
• ADePT was tested on datasets from Georgia,
  Jordan, Serbia, Ukraine, Montenegro.

14
ADePT Tables and Graphs

• Report on variables in every dataset
• Report on possible errors in variables,
  inconsistencies between the datasets, other
  warnings and notes
• Overall Poverty, Expenditure Inequality
• Decompositions of poverty changes
• Poverty profiles by socio-demographic categories
• Consumption regressions
• Poverty simulations
• Sensitivity analysis

15
Table 2.1 Original
Table 2.1 with Standard Errors
Table 2.1 Frequencies
16
ADePT What to expect in the nearest future?

• Testing on data from other countries
• More tables
• More graphs
• Extended set of variables for analysis
• Smart Graphs/Tables program can automatically
  format graphs, control for outliers, generate
  warning messages
• Ability to save and load predefined program
  configurations

17
ADePT Directions for future development

• ADePT Public Release mid-June 2007
• Multiple extensions of ADePT that can cover other
  areas of the typical PA Labor, Health,
  Education, etc.
• Automated Poverty Lines (expected in fall 2007)
• Set of tools to simplify the construction of
  consumption aggregates.