Not Separate, Not Equal: Poverty and Inequality in postApartheid South Africa

1
Not Separate, Not EqualPoverty and Inequality
in post-Apartheid South Africa

• Berk Özler
• January, 2007

2
Introduction
3
Introduction
4
Introduction

• These figures are interesting and powerful for
  South Africa, however
  
• If both your data and your analysis are not
  credible, your results may never see the light of
  day.
  
• If they do, they may be assailed and shoved aside
  by those who have an incentive to do so

5
Data Can we afford to let perfect be the enemy
of good?

• Surveys used for our analysis of the South
  African expenditure distribution suffer from
  various problems
  
• Sampling frames biased and outdated
• Expenditure, not consumption surveys (with
  implications for the consumption aggregate)
  
• No quantity information (i.e. no unit values) and
  no community price surveys
  
• No rural price data
• Low quality data on home production of food

6
Analysis Blunders at the Stats Office

• A rapid descriptive analysis of the results
  disseminated too quickly to the national media
  caused immediate concerns with the data.
  
• On the other hand, the primary data were not made
  available to the public for a long time,
  preventing the replication or debunking of the
  results.
• Multiple revisions to the data (especially to
  sampling weights) only increased data concerns.

7
How to do poverty work that receives wide-spread
acceptance?

• Under ideal circumstances, the data collection
  effort should be planned very carefully.
  
• Sampling frame issues
• Panel vs. cross-sectional data collection
• Comparability over time vs. improvements in
  methodology
  
• Diary vs. recall
• Etc.

8
How to do poverty work that receives wide-spread
acceptance?

• Under less than ideal circumstances, do every
  sensitivity analysis possible under the sky,
  exploit every data source available, and
  anticipate all the criticism ahead of time and
  prepare.
• The work on South Africa is an example of this
  latter approach.

9
Basic data work

• Carefully document all of your adjustments to the
  data
  
• Cleaning
• Merges
• Trimming
• Carefully examine survey design to understand
  sampling weights, stratification, clustering, etc.

10
Building a consumption aggregate

• Decide what will be included in your consumption
  aggregate, which will be your welfare indicator
  
• What should be included, what can be included?
• What is the comparability of survey design over
  time
  
• Make these decisions based on careful arguments
  that are based in theory and empirics
  
• If the choice ends up being somewhat subjective
  (which will, to a certain extent, always be the
  case), then test the robustness of your results
  later to your inclusion/exclusion of certain
  items.

11
South Africas consumption aggregate

• The consumption aggregate includes the following
  expenditure categories
  
• food, beverages, and cigarettes (excluding
  home-grown foods)
  
• housing (imputed rental value of residence and
  utilities)
  
• compensation for domestic workers personal care,
  household services, and other household consumer
  goods
  
• fuel (excluding firewood and dung)
• clothing and footwear transport (excluding cost
  of purchased vehicles)
  
• Communication, education, reading matter, cost of
  licenses and other rental charges, and cost of
  insurance.

12
South Africas consumption aggregate

• Important categories of expenditures we have
  excluded from the consumption aggregate are
  
• water firewood and dung
• health
• imputed value of household durables
• food consumption from home production
• lobola/dowry, funerals, religious or traditional
  ceremonies, gambling
  
• lumpy expenditures, such as furniture,
  appliances, vehicles, sound and video equipment,
  etc.

13
Poverty Lines

• For South Africa, we chose to draw normative
  poverty lines for our analysis, using the
  cost-of-basic-needs method.
  
• This method stipulates a consumption bundle
  deemed to be adequate for basic consumption
  needs, and then estimates its cost for each
  province (Ravallion, 2001).
• The basic needs bundle is typically anchored to
  food-energy requirements consistent with common
  diets in the specific context.

14
Poverty Lines

• The food basket we have selected consists of the
  mean per capita quantities of each food item
  consumed by the third quintile of the (nominal)
  expenditure distribution in 2000.
• Using the nutritional value information for each
  food item obtained from the Medical Research
  Council (MRC) in South Africa, we calculated that
  this bundle would provide the average household
  with roughly 1927 kilocalories per capita per
  day.

15
Prices

• To calculate quantities from expenditure data
  (which was not asked in our surveys), we need
  price data.
  
• Nor did our surveys collect information on prices
  of various food items from the markets in the
  sampled communities.
  
• However, STATS SA has been collecting monthly
  price data for practically all the items in the
  food module of the IES surveys from metropolitan
  and urban areas of the nine South African
  provinces.

16
Prices

• The Laspeyres Food Price Index was calculated
  using these prices from January 2001.
  
• To derive an overall price index, we had to also
  derive a price index for non-food and housing
  (Lanjouw et al., 1996)
  
• We derived a housing price index by predicting
  the rental value of a house in an urban area
  that has 4 rooms, brick walls, a flush toilet,
  and access to electricity and street lighting in
  each province.
• Finally, we took a weighted average of the food
  and the housing index to estimate our non-food
  (non-housing) price index.

17
Food Poverty Line

• The average representative bundle of the third
  quintile
  
• costs 180 Rand in 2000 prices, and
• provides 1927 kilocalories per person.
• Using recommended average energy allowances, we
  calculated that the consumption in kilocalories
  recommended for an average South African
  household per capita is 2261.
• Linearly adjusting the 180 Rand figure by
  2261/1927, we arrived at a food poverty line of
  211 Rand the amount necessary to purchase
  enough food to meet the basic daily food-energy
  requirements.

18
Upper- and Lower-Bound Poverty Lines

• To derive the overall poverty line, we set a
  lower bound and an upper bound for
  cost-of-basic-needs poverty lines in South
  Africa, following Ravallion (1994).
• We calculate the mean non-food expenditure of
  those households whose total consumption
  expenditures lie in small, but increasing
  intervals around the food poverty line.
• The simple average of these mean non-food
  expenditures plus the food poverty line yields a
  lower bound poverty line of 322 Rand.
  
• The basic idea here is that if a households
  total expenditure is equal to the food poverty
  line, then any non-food expenditure for that
  household must be absolutely necessary as the
  household is giving up basic food needs for those
  non-food consumption goods.

19
Upper- and Lower-Bound Poverty Lines

• Using the same technique, but this time
  calculating the mean total expenditure of
  households whose food consumption expenditures
  are equal to the food poverty line, we derive an
  upper bound poverty line of 593 Rand.
• If the basic needs norms that are anchored to
  food-energy requirements of South African
  households are deemed reasonable, then the
  poverty line for South Africa must lie between
  322 and 593 Rand in 2000 prices.
• We also briefly discuss results using two more
  poverty lines 87 Rand and 174 Rand per capita
  per month. These are equivalent to the commonly
  used international poverty lines of 1/day and
  2/day, adjusted for purchasing power parity. The
  2/day poverty line is close to that used by
  Deaton (1997), and the food poverty line of 211
  Rand described above. In this sense, the 2/day
  poverty line can be thought of as an extreme
  poverty line.

20
Poverty Lines Will people buy into them?

• First, the poverty lines should be internally
  consistent.
  
• Second, it helps if the poverty lines are
  consistent with popular benchmarks (for example
  in South Africa, earning 1000 Rands per month)
  
• Third, it does not help if your poverty line
  defines everyone (or very few) as poor.
  
• Finally, in any case, all of your distributional
  analysis should include stochastic dominance
  analysis (as in the figures shown above).
  
• For analysis over time, simply inflate/deflate
  your poverty line using your inter-temporal price
  indices (do not reconstruct it!)

21
Potential Areas of Concern

• Sampling weights due to problems with sampling
  frame
  
• Lack the necessary information to impute a
  comparable value for consumption of home-grown
  products
  
• The lack of rural price data
• Excluded items from the consumption aggregate

22
Sensitivity Analysis Sampling Frame

• Following the end of apartheid, internal
  migration and emigration might have led to rapid
  shifts in demographic composition in South
  Africa, possibly making the 1996 Population
  Census the sampling frame for the 2000 IES
  somewhat outdated.
• Comparing respective population shares of racial
  groups in the IES with those from the recent 2001
  Census (STATS SA, 2003) confirms that this may be
  true to some extent.

23
Sensitivity Analysis Sampling Frame
24
Sensitivity Analysis Rural/Urban Price
Differentials

• Go to another data source to get an idea of
  rural/urban cost-of-living differences.
  
• Go outside the country to get an idea regarding
  whether urban/rural price differentials are
  increasing or decreasing.
  
• Go to other parts of your data to see where
  people purchase their products and whether they
  purchase bulk.
  
• Finally, if you dont have data, devise scenarios
  that would overturn your main conclusions. Then,
  discuss whether those scenarios are realistic.
  If not, you are safe.

25
Sensitivity Analysis Rural/Urban Price
Differentials
26
What Do Other Data Sources Tell Us?

• The results were challenged by the government and
  other researchers.
  
• Data from the Quarterly Bulletin Time Series of
  the South African Reserve Bank shows that the
  total Final consumption expenditure by
  households grew by approximately 3 per year
  between 1995 and 2000.
• Census figures from Statistics South Africa put
  the annual population growth rate in South Africa
  at exactly 2, i.e. the per capita growth rate at
  approximately 1.
• In comparison, using household survey data to
  carefully construct comparable expenditure
  aggregates in this study, we find that the annual
  per capita growth rate of per capita household
  expenditures is 0.5.

27
What Do Other Data Sources Tell Us?
28
What would we expect given what is known about
this period in South Africa?

• Low growth rate (National Accounts Data)
• Rising unemployment (Klasen Woolard, 2000)
• Job losses for lower-paid, less-skilled persons
  (Whiteford and van Seventer, 2000)
  
• An increase in poverty is not unexpected, given
  the large increase in the number of unemployed.
  (Meth Diaz, 2003).

29
Final Lessons?

• Plan ahead
• Do not skimp on questionnaire development,
  training and pre-testing.
  
• Put great care into documentation of survey
  design
  
• Dont do it yourself it is a lot of work!
• Cooperate with (and be nice to) all the people
  you might need data from.

30
Final Lessons?

• Be honest
• Admit the possibility that results could be
  different under certain assumptions
  
• Be open
• Make sure the data are available and your results
  are replicable (make your codes available to
  others)
  
• Be thorough, leave no holes in your work
• Hire someone to independently review your work
• Have written agreements regarding data release,
  release of reports, etc.
  
• Be patient, dont lose your cool, but also be
  determined and do not back down.