<<

. 6
( 13)



>>




137
Michael R. Carter and Munenobu Ikegami


Mid-skill, no risk
Mid-skill, risk
12
High skill, no risk
High skill, risk




8
Income




4




0


0 10 20 30 40 50 60
Time

Figure 6.5. Vulnerability hurts ˜average™ individuals most


sequence of capital stocks. To isolate the pure ex ante effect of risk,
no shocks are actually realized in this simulation (despite the fact that
individuals fearfully behave as if shocks will occur).
r A Full Stochastic Simulation, in which the risk-adjusted optimal accumula-
tion rule, ir— (kt |·, ), is applied, but after each application, the individual
receives a random shock generated in accordance with the probability
structure . This simulation permits us to isolate the full effect of random
events (both ex ante and ex post) on the time path of capital, income,
and consumption for an individual.

Figure 6.5 illustrates the generated time paths when the Non-Stochastic and
the Full Stochastic simulations are applied to the mid-skill and high-skill
individuals whose initial endowment positions are shown in Figure 6.4 by the
circle and triangle, respectively. The simulation was run for sixty time periods.
As can be seen, in the non-stochastic simulation, both individuals move
smoothly toward the high equilibrium. 9 Autarchic accumulation provides a
pathway from poverty for both individuals. Both become non-poor around

9
Note that the high-skill individual has a higher level of steady state capital stock because
the marginal productivity of capital is boosted by her skill level.




138
Looking Forward

Table 6.1. Simulations for archetypical individuals

Non-stochastic Risk without shocks Full stochastic

Low skill (· = 0.94, k0 = 4.9)
Discounted stream of utility 2.7 2.7 1.2
Discounted stream of income 26 26 25
Dynamic asset gap k (·) ’ k0 ∞ ∞ ∞
˜
Calvo“DerconP fta (0) 60 60 60
Calvo“DerconP fta (1) 10.2 10.3 14.3
Middle skill (· = 1.10, k0 = 2.1)
Discounted stream of utility 4.0 3.8 3.3
Discounted stream of income 37 29 28
Dynamic asset gap k (·) ’ k0
˜ 0 2.3 2.3
Calvo“DerconP fta (0) 8 60 60
Calvo“DerconP fta (1) 0.6 2.8 3.9
High skill (· = 1.22, k0 = 1.1)
Discounted stream of utility 5.5 5.5 4.7
Discounted stream of income 44 43 40
Dynamic asset gap k (·) ’ k0
˜ 0 0 0
Calvo“DerconP fta (0) 4 5 6
Calvo“DerconP fta (1) 0.5 0.6 0.6




year 10 of the simulation as their achieved capital stocks exceed the asset
poverty line.
In the stochastic simulation, the high-skill individual is occasionally buf-
feted about by shocks and must continually rebuild her assets in order to
reattain the desired steady state capital stock. 10 In sharp contrast, the middle-
skill agent undertakes a fundamental shift in strategy in the presence of risk.
As seen in Figure 6.4, this individual has fallen below the Micawber Frontier
once risk is taken into account. While this individual suffers ¬‚uctuations akin
to those suffered by the high-skill individual, the more fundamental effect
results from her retreat from trying to reach the high equilibrium (i.e. a
pathway from poverty is no longer attainable or sustainable).
Table 6.1, which presents statistics related to all three simulations, further
substantiates this latter point. The table includes results for the low-skill
individual whose initial asset position is indicated by the triangle in Figure
6.4. For each of the three individuals, the table displays the discounted
stream of utility which is obtained under each simulation. In addition, the
discounted value of income produced by each individual over the simulation
is also listed. In contrast to the mid-skill individual, the ˜Risk without Shocks™
simulation barely perturbs the time paths and outcomes of both the low-
skill and high-skill agents as neither of these agents shifts strategy in the face



10
Note that the desired steady state level of capital is reduced by the presence of risk.




139
Michael R. Carter and Munenobu Ikegami

of risk. 11 When shocks are actually realized (the Full Stochastic Simulation),
then these agents suffer more substantial losses, especially in terms of
utility. 12
In contrast, for the mid-skill individual, the ˜Risk without Shocks™ simula-
tions brings a major drop in production (from 37 to 29) and utility (from 4.0
to 3.8) compared to the Non-Stochastic simulation. While the Full Stochastic
simulation brings some additional income losses for this individual, they are
modest compared to the losses occasioned by the risk-induced strategy shift.
Among other things, these simulations show that in the presence of critical
asset thresholds, risk takes on particular importance for those individuals
subject to multiple equilibria.


6.3. Forward-Looking Measures of Chronic Poverty
and Vulnerability

The theoretical analysis in the prior section has used dynamic economic
theory to elucidate the multiple dimensions of chronic and persistent poverty,
and to demonstrate how vulnerability to economic shocks further increases
chronic poverty. Building on those ideas and insights, this section puts for-
ward two types of chronic poverty measures. The ¬rst generalizes a suggestion
in Carter and Barrett (2006) and uses information on the Micawber Frontier
to create a asset-based chronic poverty measure. The second uses the asset
dynamics implied by the BCI model to create a forward-looking income
stream that can then be used to calculate any of the income-based chronic
poverty measures discussed by Calvo and Dercon (this volume). Both the asset
and income-based measures can be utilized to create explicit chronic poverty
vulnerability measures, where vulnerability is understood as the increase in
the chronic poverty measure induced by risk and shocks.
The measures put forward in this section rely on structure of the standard,
backward-looking Foster“Greer“Thorbecke (FGT) class of poverty measures
de¬ned as:
N
z ’ fi „
1
P („) = Ii
M z
i=1

where M is the total population size (poor and non-poor), i indexes individual
observations, c is the scalar-valued poverty line, ci is the ¬‚ow-based measure
11
Their desired steady state values do diminish and hence production and consumption
fall modestly, generating the changes shown in the table.
12
Realized shocks affect utility more strongly than income. For example, for the low-skill
agent, the discounted stream of the utility of consumption falls from 2.7 to 1.2, while the
discounted stream of income only falls from 26 to 25. The proportionately larger drop in util-
ity occurs because individuals will end up spending more of their income on reaccumulating
assets destroyed by shocks.




140
Looking Forward

of welfare (income or expenditures) as measured retrospectively at the
time of the survey, Ii is an indicator variable taking value one if fi < z and
zero otherwise, and „ is a parameter re¬‚ecting the weight placed on the
severity of poverty. Setting „ = 0 yields the headcount poverty ratio P (0)
(the share of a population falling below the poverty line). The higher-order
measures, P (1) and P (2), yield the poverty gap measure (the money-metric
measure of the average ¬nancial transfer needed to bring all poor households
up to the poverty line) and the squared poverty gap (an indicator of severity
poverty that is sensitive to the distribution of well-being amongst the poor).

6.3.1. Chronic poverty measures based on the Micawber Threshold
As suggested by Carter and Barrett (2006), information on asset dynamics
that permits identi¬cation of the Micawber Frontier opens the door to a
forward-looking poverty measure. The standard, money-metric poverty line is
frequently criticized as an arbitrary construct which has no behavioural foun-
dation. In contrast, the Micawber Frontier is an empirical construct whose
foundation is observed behaviour. Conceptually, the Micawber Frontier can
separate households expected to be persistently poor from those for whom
time is an ally that promises better standards of living in the future. Poverty
measures based on the Micawber Frontier thus promise to help identify the
long-run health of an economy as judged by its ability to facilitate growth in
living standards amongst its least well-off members.
Generalizing the Carter and Barrett measure to allow for heterogeneous
ability yields the following expression:
M „
˜ (·i ) ’ ki
1 k
,
Iik
Pk („) = (6.3)
˜
M k(·i )
i=1

where ki is asset stock of household i and the binary indicator variable Iik = 1
if ki < k(·i ) and re¬‚ects whether the household i™s asset stock is below the
˜
Micawber Frontier. When „ = 1, we can use the core part of this measure,
Iik (k(·i ) ’ k), to de¬ne the ˜dynamic asset poverty gap™. Table 6.1 reports this
˜
measure for three prototypical individuals whose initial asset positions are
illustrated in Figure 6.4. The normalized asset poverty gap for the high-skill
individual is always zero. For the mid-skill individual, the gap is zero in the
absence of risk, but rises to 1.3 units of capital (or about 33 per cent of
the Frontier value of 4) when the discouraging effect of risk shifts out the
Micawber Frontier. For the low-skill person, the Micawber Frontier and the
dynamic asset poverty gap are in¬nite as there is no level of capital stock
from which this individual will ¬nd it desirable to sustain the high-level
equilibrium.
The existence of the intrinsically chronically poor individuals (for whom
the dynamic asset poverty gap is in¬nite) renders the poverty measure (6.3)



141
Michael R. Carter and Munenobu Ikegami

mathematically problematic as the portion of the expression in parentheses is
unde¬ned for these individuals. We thus modify the measure as follows:

k(·i ) ’ ki
˜
1
,
˜
P k („) = Iik
˜
M k(·i )
˜
iµ M

˜ ˜
where M is the subset of the total population for whom k(·i ) is ¬nite. Setting
˜
„ = 0, this modi¬ed measure ( P k (0)) gives the headcount ratio of all non-
intrinsically poor individuals who are below the Micawber Threshold and
who would therefore be expected to be chronically poor in the sense of
being trapped at the low-level equilibrium. Information on the fraction of the
˜
population that is intrinsically chronically poor ( M’ M ) can be used to create a
M
˜
˜
complete chronic poverty headcount measure, CPHC = P k (0) + ( M’ M ).
M
˜
When „ = 1, P k (1) yields a normalized measure of the asset transfers that
would be necessary to place multiple equilibrium chronically poor households
in a position from which they can grow and sustain a non-poor standard
˜
living in the future. In the language of Carter and Barrett (2006), this P k (1)
measure would indicate the resources needed for the cargo net transfers
required to eliminate multiple equilibrium chronic poverty. Note that there
are no asset transfers that will sustainably eliminate the chronic poverty of
the intrinsically chronically poor.
To illustrate these ideas, we used the BCI model to conduct a sixty-year
simulation of poverty and its evolution for an imaginary community of
100 households. We performed two sets of simulations using the procedures
described in the prior section. First we performed the Non-Stochastic simu-
lations (in which households follow the optimal accumulation rule de¬ned
by the non-stochastic version of the BCI model). Second we utilized the Full
Stochastic in which each household follows the optimal accumulation rule
de¬ned by the stochastic version of the BCI model and each received their
own random (idiosyncratic) shock in each time period.
For the simulation, it was assumed that 25 per cent of the households
had suf¬ciently low skill endowments that they were in the intrinsically
chronically poor category in the presence of risk (that is, ·i < · L ). Another
50 per cent were in the mid-skill (multiple equilibrium) range, · L < ·i < · H ,
while the ¬nal 25 per cent were in the high-skill (intrinsically upwardly
mobile) range (·i < · H ). While these assumptions are arbitrary, they do match
the empirical ¬ndings of the Santos and Barrett (2006) study of East African
pastoral households. Finally, initial asset endowment for each household was
randomly distributed (using a uniform distribution) over the range of 0.1 to
10 units of capital. 13

13
In the real world, one would not expect to ¬nd ˜initial™ endowments uncorrelated with
skill. However, for illustrative purposes, this egalitarian assumption permits us to more fully
see the operation of the model.




142
Looking Forward

Table 6.2. Simulated chronic poverty measures

Non-stochastic Stochastic Vulnerability
simulation simulation

˜Backward-looking™ measure
FGT P (0) at time t = 1 0.34 [0.07] 0.34 [0.07] ”
FGT P (0) at time t = 60 0.21 [0.02] 0.55 [0.08] ”
Forward-looking chronic poverty measure
Carter“Barrett threshold measures
Intrinsic headcount, M’ M 3% 24% 88%
M
˜
P k (0) measure at time t = 1 18% 24% 25%
Complete headcount, CPHC 21% 48% ”
˜
P k (1) measure at time t = 1 10% 15% ”
Calvo“Dercon income stream Measures
Pifta (0) (average) 13 [22%] 28.5 [48%] 54%
Pifta (1) (average) 1.5 3.7 59%




Included in the table are the standard (backward-looking) Foster“Greer“
Thorbecke (FGT) poverty measures for both the initial period (t = 1) and ¬nal
period of the simulation (t = 60). Under the scenario that assumes away eco-
nomic shocks, the standard poverty headcount drops over the period of the
simulation from 34 per cent to 21 per cent of the population. This latter ¬gure
exactly equals the period 1 dynamic asset poverty threshold headcounts of the
chronically poor (both the intrinsically and multiple equilibrium chronically
poor). As this simple example shows, the forward-looking threshold-based
measure captures the dynamics of the system and thus provides a more infor-
mative portrayal of the expected long-run evolution of poverty. 14 Combining
the two pieces of information would permit us to say (in period 1) that 34
per cent of the population is currently poor and that we would expect (under
existing dynamics) to see 13 per cent of the population escape poverty, and
˜
the other 21 per cent to remain chronically poor. The P k (1) measure of the
size of the dynamic asset poverty gap shows that, on average, the chronically
poor have assets that are 10 per cent below the Micawber Frontier.
When shocks (and risk) are brought into the model, the results change
rather signi¬cantly as shown in the second column of Table 6.2. Over the
sixty-year period of the simulation, the FGT headcount rises from 34 per cent
to 55 per cent. In this case, the backward-looking FGT measure overstates the
long-run health of the economy. In contrast, the year 1 Carter“Barrett asset-
based CPHC indicates that 48 per cent of the population is chronically poor
(with this fraction split evenly between the intrinsically chronically poor and
the multiple equilibrium chronically poor). This 48 per cent ¬gure is in fact
an understatement of the functioning of the economy as it fails to account

14
The BCI model assumes that the underlying structural dynamics of the economic do not
change over the period of the simulation, a stricture unlikely to be met in the real world.




143
Michael R. Carter and Munenobu Ikegami

for multiple equilibrium households that are knocked below the Micawber
˜
Threshold over the time period of the simulation. 15 The P k (1) measure rises
to 15 per cent, indicating that the depth of dynamic asset poverty rises for
the multiple equilibrium poor. The cargo net transfers needed to lift these
individuals over the Micawber Frontier have thus increased. As in the non-
stochastic case, the combination of the FGT and the dynamic asset poverty
measures provides a more comprehensive view of the nature of poverty and
its likely future evolution.


6.3.2. Using asset dynamics to create forward-looking income-based
chronic poverty measures
In addition to underwriting chronic poverty measures based on the Carter“
Barrett dynamic asset poverty gap, information on asset dynamics can be
used to project future asset and income levels. When combined with the
income-based chronic poverty measures of Calvo and Dercon (this volume),
these projections open up another class of forward-looking chronic poverty
measures.
Calvo and Dercon suggest a number of ways of consistently analysing a
sequence of income levels for a given household over T time periods. While
they are primarily thinking of sequences of past incomes, they suggest that
their methods can be applied to estimated future income streams. The analysis
here follows this suggestion.
For purposes here, we will limit our attention to what Calvo“Dercon call the
FTA (Focus“Transformation“Aggregation) chronic poverty measure. Letting
z denote the standard income poverty line, and fit denote the income of
household i in period t, we can write the FTA measure (P f ta ) as:
T
z ’ fit „
f ta f ta
,
‚T’t Iit
Pi („) = (6.4)
z
t=1

fta
where Iit is an indicator variable that takes the value of one if fit < z, and
‚ is the discount factor. Note that measure (6.4) is speci¬c to a particular
individual and does not aggregate across individuals. For illustrative purposes
here, we set ‚ = 1, so that all poverty spells are treated identically (see Calvo
and Dercon for more discussion on the desirability of this assumption). In
the special case when „ = 0, (6.4) simply counts the number of poverty spells
experienced by the individual.
The various simulations of the BCI model used in the prior section can
be used to illustrate our forward-looking use of the Calvo“Dercon measures.
f ta f ta
Table 6.1 presents the Pi (0) and Pi (1) measures for the low-, medium- and
15
In principle, the Carter“Barrett measure could be adjusted to account for the likelihood
that some individuals will receive shocks that will knock them under the threshold.




144
Looking Forward

high-skill individuals under the various simulation scenarios introduced ear-
lier. The low-skill individual is poor all sixty periods under all scenarios, as
shown by the degree zero FTA measure. The increase in the degree 1 FTA
poverty gap measure under the Full Stochastic simulation re¬‚ects the impact
of realized shocks.
The FTA measures for the high-skill agent shows that she escapes poverty
rather quickly under all scenarios. In contrast, the degree zero FTA measure
jumps from 8 to 60 for the mid-skill individual once risk is brought into
the picture. As this example illustrates, the forward-looking Calvo“Dercon
measure captures the chronic poverty impacts of risk that are overlooked by
standard FGT measures. In addition, while not explicitly established to cap-
ture threshold effects, the Calvo“Dercon family measures are quite sensitive
to their impacts. 16
Table 6.2 similarly presents FTA measures for the stylized 100 individual
economy analysed in the prior section. Results are shown for both the Non-
Stochastic and the Full Stochastic simulations. While measure (6.4) is individ-
f ta
ual speci¬c, Table 6.2 reports the simple average of the Pi („) measures across
the 100 individuals in the simulation. To ease comparability with the other
f ta
measures, the ¬gures in square brackets divide the Pi (0) by the total number
of periods and thus yield a measure of the fraction of time that the average
individual spends below the poverty line during the course of the sixty-period
simulation.
f ta
As can be seen in Table 6.2, the average value of Pi (0) when there are
no economic shocks is 13, indicating that the average household was below
the income poverty line 13 out of the 60 total time periods, or 22 per cent
of the time. Interestingly, this ¬gure corresponds closely to the period 60
FGT measure, as well as to the dynamic asset poverty measure. Similarly,
for the stochastic simulation (in which households anticipate and are subject
f ta
to economic shocks), the Pi (0) averages 28.5 poverty spells across the 100
households, indicating that households are poor roughly 48 per cent of the
time. However, it should be stressed that the equivalence of the FTA ¬gure
to the Carter“Barrett CPHC measure is somewhat coincidental. The former
re¬‚ects the fact that the intrinsically upwardly mobile may have poverty spells
as they accumulate assets and/or recover from shocks. Similarly, the long-
term chronically poor may have spells of non-poor income if they fortuitously
begin life with an ample (but unsustainable) asset endowment. 17 But despite
these differences from the threshold-based measure, the Calvo“Dercon FTA
measures capture the intrinsic dynamics of the system and provide a more
informative, forward-looking picture than does the standard FGT family of
16
This same comment would also apply if the Calvo“Dercon measures were used to look
backwards to evaluate the degree of poverty in a past realized income history.
17
Variants on the Calvo“Dercon measures that more heavily weigh ¬nal outcomes would,
however, present information that is closer in spirit to the dynamic asset poverty measures.




145
Michael R. Carter and Munenobu Ikegami

measures. Again it should be stressed that these forward-looking measures
are in principle estimable in time 1, 18 though their accuracy depends on the
stability of the underlying dynamics in the economy.



6.3.3. Vulnerability as increased chronic poverty
While there is debate over how best to conceptualize and measure vulnera-
bility (compare Calvo and Dercon, 2005, with Ligon and Schechter, 2003),
one natural approach would be to de¬ne vulnerability as the increase in
chronic poverty that results when individuals are exposed to shocks. Linking
vulnerability to increases in chronic poverty captures the sense of drastic
and irreversible harm that Calvo and Dercon (2005) identify as the common
thread that unites various concepts of vulnerability. In addition, the ability
to de¬ne vulnerability in terms of increased chronic poverty provides a very
compelling policy focus, indicating the fraction of chronic poverty that can
be remedied through social protection programmes.
The far-right column in Table 6.2 de¬nes vulnerability using both the
Carter“Barrett and the Calvo“Dercon chronic poverty measures. In both cases,
vulnerability is de¬ned as the fraction of total chronic poverty revealed by
the full stochastic simulation that is created by risk and shocks. That is,
vulnerability is the difference between chronic poverty in the stochastic and
the non-stochastic simulations, normalized by the chronic poverty in the
stochastic simulation.
As can be seen in Table 6.2, nearly 60 per cent of total chronic poverty
in the simulation analysis is the result of vulnerability under both the Carter“
Barrett and the Calvo“Dercon measures. Social protection policies would have
an enormous impact on chronic poverty in this case. This large increment in
chronic poverty created by vulnerability results from the three forces discussed
earlier. First, realized shocks sometimes push individuals below the income
poverty line. 19 Second, increased chronic poverty also results when realized
negative shocks knock individuals below the Micawber Frontier, rendering
infeasible a pathway from poverty, and indeed spelling ruin in the language
of Calvo and Dercon (2005) cited above. Third and ¬nally, the prospect that
ruin can occur has a discouraging effect on accumulation strategies, shifting

18
They are estimable if the accumulation rule can be estimated as well as the error distri-
bution that generates deviations between expected and actual accumulation. With those two
pieces of information, a set of forward-looking projections could be generated using either
stochastic or non-stochastic simulations.
19
Note that unlike the Ligon and Schechter vulnerability measure that increases with any
¬‚uctuation in income, the vulnerability measure based on the Calvo“Dercon FTA measure has
a poverty focus and only increases for ¬‚uctuations that drive individuals below the poverty
line. Note that the Carter“Barrett measure will not increase for individuals pushed below the
income poverty line, but who remain above the Micawber Frontier.




146
Looking Forward

the Micawber Frontier beyond the reach of some individuals, driving yet
additional increases in the measured (multiple equilibrium) chronic poverty.
Calculation of the vulnerability measures in Table 6.2 is feasible because
the BCI model allows us to straightforwardly simulate how individuals would
counterfactually behave in the absence of risk. However, the real world does
not offer data on how individuals would (counterfactually) behave in the
absence of risk. For example, we do not have data that could be used to
directly identify what the Micawber Frontier would be in the absence of risk as
we do not observe individuals behaving in the counterfactual, risk-free world.
Empirical implementation of this type of vulnerability measure would there-
fore be far from straightforward. 20 Nonetheless, it would in principle be possi-
ble to obtain estimates of the parameters that shape behaviour and then simu-
late what behaviour would counterfactually be in the absence of risk. It might
also be possible to take advantage of naturally occurring variation of risk (as
Rosenzweig and Binswanger, 1993, do) in order to gain insight into how the
Micawber Frontier shifts with risk. Signi¬cant future work will be required to
empirically implement the type of vulnerability measures shown in Table 6.2.



6.4. A First Application to South Africa

The prior sections of this chapter have laid out an ambitious agenda, showing
how economic theory can be used to underwrite a suite of theoretically
grounded, forward-looking chronic poverty measures. This section uses data
from South Africa to illustrate the use of these measures, employing the
KwaZulu-Natal Income Dynamics Study (KIDS) data that cover the KwaZulu-
Natal province that is home to roughly 25 per cent of South Africa™s popu-
lation (see Aguero et al., forthcoming, for thorough discussion of the KIDS
data).
The top half of Table 6.3 displays standard FGT poverty measures for the ¬rst
two rounds of the KIDS data (1993 and 1998) as reported in Carter and May
(2001). As can be seen, this period was characterized by substantial downward
mobility as the headcount measure of poverty rose from 27 per cent to 43 per
cent, while the FGT poverty gap measure (P (1)) held steady at 33 per cent. 21
20
Note however that the partial impact of vulnerability can be recovered rather straight-
forwardly by doing an empirical analysis that is akin to the middle column of Table 6.2
(risk without shocks). Using some of the methods of Schechter, it should be possible to
simulate the impact that shocks have on individuals, holding the Micawber Frontier ¬xed.
Such information could be quite useful from the perspective of designing a social safety net.
21
The Carter and May (2001) FGT measures are based on poverty line estimates using the
household subsistence line (HSL). The HSL became unavailable after 1998 and subsequent
analysis (such as that reported in Agüero, Carter, and May, forthcoming) has relied on the
poverty line standard suggested by Hoogeveen and –zler (2005). Using this latter poverty
line, the poverty headcount in the KIDS data rose from 52% to 57% over the 1993 to 1998
period.




147
Michael R. Carter and Munenobu Ikegami

Table 6.3. Backward- and forward-looking
poverty measures for South Africa

FGT measures
P (0) in 1993 27%
P (0) in 1998 43%
Carter“Barrett threshold measures
˜
P k (0) in 1998 59%
˜ k (1) in 1998
P 11%




These same data can be used to recover the underlying asset dynamics.
In a recent paper, Adato, Carter, and May (2006) use these KIDS data to
estimate the pattern of asset dynamics under the assumption that k(·) = k∀·.
˜ ˜
In other words, Adato, Carter, and May assume that the Micawber Frontier is
the same for all agents, irrespective of the individual™s skill level. In terms of
Figure 6.2, the Micawber Frontier would appear as a horizontal line under the
assumptions used by Adato, Carter, and May.
As detailed in that paper, Adato, Carter, and May ¬rst estimate an asset
index for each individual, and then use non-parametric regression techniques
to recover the pattern of asset dynamics. 22 Interestingly, they estimate the
Micawber Frontier to be at a level of assets expected to generate a living stan-
dard almost twice the poverty line. Individuals below that estimated frontier
would be predicted to slide backwards over time towards a sub-poverty line
standard of living, while those above it would be predicted to achieve a living
standard well above the poverty line. Note also that not everyone who is
observed to be poor by standard consumption measures will be predicted to be
poor in the longer term. In particular, households that have assets in excess of
the Frontier and are ˜stochastically poor™ (in the language of Carter and May,
2001) would not be predicted to be poor in the long term.
While the Adato, Carter, and May analysis rests on several strong assump-
tions, it does permit us to illustrate the use of asset threshold-based poverty
measures. As shown in the bottom half of Table 6.3, fully 58 per cent of KIDS
households were below the estimated Micawber Threshold in 1998 and are
therefore expected to be chronically poor. Because of the assumption that
the threshold is the same for all households, it is not possible to partition
this group into the intrinsically chronically poor and the multiple equi-
librium chronically poor. Nonetheless, the fact that this measure is above
the 1998 backward-looking poverty headcount indicates that the underly-
ing asset dynamics predict future increases in poverty. Put differently, the
˜
P k (0) measure indicates that the South African economy was not offering a
22
The asset index includes human capital variables as well as tangible physical assets such
as land and business equipment. Related methodological approaches to recovering a critical
asset threshold can be found in Lybbert et al. (2004), Barrett et al. (2006), and Carter et al.
(2007).




148
Looking Forward

favourable environment for asset accumulation and income growth for the
less well-off.
The reliability of such a prediction depends on the stability of the under-
lying asset dynamics, as well as on the quality of the actual estimation. The
KIDS 2004 round of data indicates a decline in the standard poverty head-
count, rather than further increases as would be expected based on the asset
threshold-based measure (see Aguero et al. for results from the 2004 data). The
predictive failure of the asset-based measure may re¬‚ect an underlying change
in asset dynamics (that is, the prospects for accumulation and income growth
improved dramatically between 1998 and 2004). 23 It may also re¬‚ect the
simplifying assumption used by Adato, Carter, and May, that the Micawber
Frontier is the same for all households. The fact that the asset poverty gap
˜
measure ( P k (1)) was a modest 11 per cent in 1998 indicates that the typically
asset-poor household was not too far below the estimated frontier. Either
a modest improvement in asset dynamics (or a modest overestimation of
the threshold for mid-skill agents) may have led to the predictive failure
of the threshold-based chronic poverty measures. Future efforts are clearly
needed to help the empirical measures catch up with the sophistication of the
theoretically derived measures discussed earlier.



6.5. Chronic Poverty Measurement and Policy

This chapter began with the challenge of understanding how much poverty
is chronic in the sense that it would be expected to persist into the future.
The microeconomic theory of asset accumulation and poverty traps suggests
a way of approaching this problem and estimating future asset accumula-
tion and income growth. This information can in turn be used as the basis
for two families of theoretically grounded, forward-looking, chronic poverty
measures.
While there is still much to be done to improve the chronic poverty
measures put forward here, they are ultimately intended to complement, not
replace, conventional, ˜backward-looking™ poverty measures. While the latter
are meant to give us a sense of the current (or at least recent) economic
status of people at the bottom of the income distribution, the former use
information on patterns of asset accumulation to project forward who is likely
to remain poor in the future. Together, the two classes of measures provide a
more complete description of the groups for whom the economy is not well
functioning.
23
The BCI model used in the theoretical analysis here assumes that the income generation
process does not change over time. In principle, the model could be modi¬ed to re¬‚ect
growth in productivity and wages (or cycles of macroeconomic boom and bust). The impact
on behaviour would depend on how individuals anticipated these changes.




149
Michael R. Carter and Munenobu Ikegami

As in any area of economics, looking forward into the future is fraught
with dif¬culties. The information that can be gleaned from the chronic
poverty measures suggested here is probably most valuable over a medium-
term time horizon when the structure of the economy is relatively stable.
But even within these limits, the capacity of the asset-based chronic poverty
measures to provide information on the intrinsically chronically poor and
the multiple equilibrium chronically poor is potentially quite valuable from a
policy perspective. Moreover, while empirical calculation of the vulnerability
measures discussed in section 6.3.3 is probably fraught with dif¬culty, the
theoretical analysis put forward makes clear that vulnerability to economic
shocks is potentially an important part of chronic poverty. This is especially
true in economies where large numbers of agents ¬nd themselves in the
multiple equilibrium category, facing a positive but ¬nite Micawber Frontier.
The theory reviewed here suggests that the provision of social protection
measures will lower the Micawber Frontier for average individuals, crowding
in private accumulation and rendering feasible new pathways from poverty
for at least some. While there is still much to ¬nd out about whether social
protection can in practice really have these twin effects on reducing chronic
poverty, further efforts to more sharply conceptualize and measure chronic
poverty will move us in the direction of being able to explore these ideas and
pilot new social protection programmes.


Appendix 6.1: Details of Theoretical Model

This section provides additional detail on the formal model used to generate the results
discussed in sections 6.2 and 6.3. For additional details, see Barrett et al. (2007).


Non-stochastic model
Under the technological speci¬cation given in section 6.1, we assume that individuals
face the following in¬nite horizon model as they make the decision about how best to
divide income ( f (·, kt )) every period t between consumption (ct ) and investment (it ):

‚t’1 u(ct )
max
t=1
ct + it ¤ f (·, kt )
s.t.
f (·, kt ) = max{·k„ L , ·k„ H ’ E }
kt+1 = it + (1 ’ °)kt
k1 given

where ‚ is the discount factor, ct is consumption, u(·) is utility function, it is investment
and ° is depreciation rate. Note that there is neither saving nor borrowing and that the
household is assumed to live forever. Solution of this problem, using the parameters
˜
given below, generates the Micawber Frontier, k(·), illustrated in Figure 6.2 in the main
body of the text.



150
Looking Forward

Stochastic model
Households face a number of risks. These risks can be classi¬ed as (i) asset shocks,
(ii) income shocks, and (iii) marginal utility shocks. While all three types or risk are
important, the analysis here focuses on the relatively simple case of asset shocks. In
particular we assume that assets evolve according to the following speci¬cation:

kt+1 = Ët [it + (1 ’ °)kt ]

where Ët ∈ (0, Ëmax ] is the asset shock. Note that this multiplicative speci¬cation makes
the magnitude of risk increase as accumulated capital stock increases.
We assume that the probability distribution of Ët is known and that the individual
decision maker allocates income between consumption and investment in order to
solve the following problem:

‚t’1 u(ct )
max E 1
t=1
ct + it ¤ f (·, kt )
s.t.
f (·, kt ) = max{·k„ L , ·k„ H ’ E }
kt+1 = Ët [it + (1 ’ °)kt ]
k1 given

where E 1 is expectation at period 1. Solution of this problem, again using the parame-
ters outlined below, yields the results summarized in Figure 6.3.



Parameters and other details for numerical simulation
The functional speci¬cation for the utility function u(·) is
1’Û
ct ’ 1
u(ct ) =
1’Û

The probability density of Ët is assumed to be:
§ «
⎪ 0.90 if Ët = 1.0 ⎪
⎪ ⎪
⎪ ⎪
⎨ 0.05 if Ë = 0.9 ¬
t
density of Ët =
⎪ 0.03 if Ët = 0.8 ⎪
⎪ ⎪
⎪ ⎪
© ⎭
0.02 if Ët = 0.7

The other structural parameter values are assumed to be as follows: Û = 1.5, ° = 0.08,
‚ = 0.95, „ L = 0.3, „ H = 0.45, E = 0.45.
We discretize continuous variables k and · as follows: k = {0.1, 0.2, . . . , 15.0} and
· = {0.94, 0.96, . . . , 1.22}.
For the simulation of the stylized economy of 100 individuals we draw · from
N(1.08, 0.0742 ). Parameter values of mean and variance are chosen so that ex ante
proportion of low-, middle-, and high-type individuals (de¬ned relative to the stochastic
Micawber Frontier) would be 25 per cent, 50 per cent, and 25 per cent, respectively. We
draw k1 from Uniform [0.1, 10.0] and assume that k1 and · are statistically independent
from each other.



151
Michael R. Carter and Munenobu Ikegami

We specify the asset poverty line as the asset level that satis¬es the following
equation:

y p = f (·, k p ).

where y p is the income-based poverty line. Note that the asset poverty line depends
on · and we denote it by k p (·). We assume that the income poverty line, y p , is 1.62,
the level of income that an average individual (· = 1.12) would produce in equilibrium
using the low technology.




References

Adato, M., Carter, M. R., and May, J. (2006), ˜Exploring Poverty Traps and Social Exclu-
sion in South Africa Using Qualitative and Quantitative Data™, Journal of Development
Studies, 42(2): 226“47.
Agüero, J., Carter, M. R., and May J. (forthcoming), ˜Poverty and Inequality in the First
Decade of South Africa™s Democracy: What can be Learnt from Panel Data?™, Journal
of African Economies.
Barrett, C. B., Carter, M. R., and Ikegami, M. (2007), ˜Social Protection Policy to Over-
come Poverty Traps and Aid Traps: An Asset-Based Approach™, working paper.
Marenya, P. P., McPeak, J. G., Minten, B., Murithi, F. M., Oluoch-Kosura, W., Place,
F., Randrianarisoa, J. C., Rasambainarivo, J., and Wangila, J. (2006), ˜Welfare Dynamics
in Rural Kenya and Madagascar™, Journal of Development Studies, 42(2): 248“77.
Buera, Francisco (2005), ˜A Dynamic Model of Entrepreneurship with Borrowing
Constraints™, working paper, Northwestern University.
Calvo, C., and S. Dercon (2005), ˜Measuring Individual Vulnerability™, Oxford, Depart-
ment of Economics Discussion Papers Series No. 229, March.
Carter, M. R., and Barrett, C. B. (2006), ˜The Economics of Poverty Traps and Persistent
Poverty: An Asset-Based Approach™, Journal of Development Studies, 42(2): 178“99.
and May, J. (2001), ˜One Kind of Freedom: Poverty Dynamics in Post-Apartheid
South Africa™, World Development, 29(12): 1987“2006.
Little, P., Mogues, T., and Negatu, W. (2007), ˜Poverty Traps and the Long-Term
Consequences of Natural Disasters in Ethiopia and Honduras™, World Development.
Chronic Poverty Research Centre (CPRC) (2004), ˜The Chronic Poverty Report 2004“05™,
Manchester.
Foster, J., Greer, J., and Thorbecke, E. (1984), ˜A Class of Decomposable Poverty Mea-
sures™, Econometrica, 52: 761“5.
Hoddinott, J. (2006), ˜Shocks and their Consequences across and within Households in
Rural Zimbabwe™, Journal of Development Studies, 42(3): 301“21.
Hoogeveen, J. G., and –zler, B. (2005), ˜Not Separate, not Equal: Poverty and Inequality
in Post-Apartheid South Africa™, William Davidson Institute Working Paper No. 739,
Ann Arbor: University of Michigan.
Jalan, J., and Ravallion, M. (2004), ˜Household Income Dynamics in Rural China™, in
S. Dercon (ed.), Insurance against Poverty, Oxford: Oxford University Press, 108“24.
Ligon, E., and Schechter, L. (2003), ˜Measuring Vulnerability™, Economic Journal,
113(486): 15“102.



152
Looking Forward

Lybbert, T., Barrett, C., Desta, S., and Coppock, D. L. (2004), ˜Stochastic Wealth
Dynamics and Risk Management among a Poor Population™, Economic Journal,
114(498): 750“77.
Rosenzweig, M., and Binswanger, H. (1993), ˜Wealth, Weather Risk and the Composition
and Pro¬tability of Agricultural Investments™, Economic Journal, 103(416): 56“78.
Santos, P., and Barrett, C. B. (2006), ˜Heterogeneous Wealth Dynamics: On the Roles of
Risk and Ability™, Cornell University working paper.
Schechter, L. (2006), ˜Vulnerability as a Measure of Chronic Poverty™, paper presented
at the Workshop on Concepts and Methods for Analysing Poverty Dynamics and
Chronic Poverty, University of Manchester, 23“5 October.
Zimmerman, Fred, and Carter, Michael (2003), ˜Asset Smoothing, Consumption
Smoothing and the Reproduction of Inequality under Risk and Subsistence Con-
straints™, Journal of Development Economics, 71(2): 233“60.




153
7
Poverty in Time
Exploring Poverty Dynamics from Life History
Interviews in Bangladesh—

Peter Davis




7.1. Introduction

This chapter discusses the usefulness of a life history method for analysing
poverty dynamics in developing country contexts, using ¬ndings from a study
in Kushtia district in Bangladesh. A high level of contextual and historical
detail can be collected in life history interviews enabling the exploration of
people™s perceptions and understandings of the complex and dynamic realities
of their lives. Life history methods uncover a number of phenomena that
tend to be concealed from more usual quantitative methods. These include
events with multiple causation, ˜last-straw™ threshold effects (events which
lead to catastrophe due to a series of previous events), other cumulative
trends, outcomes based on the ordering of a sequence of events, and complex
interactions. Variable-based research where the household is used as the unit
of analysis also produces the problem of masking events associated with
household breakdown and intra-household effects. Life history approaches
can help to explore these.
However in case-based research, the numbers of cases studied are usually
quite small, limiting the ability to generalize across larger populations. Here a
larger-than-usual number of cases are analysed and categorized, along with
the patterns of crisis, coping, and opportunity that emerged. Eight main




I would like to thank Bob Baulch, David Hulme, Andrew Shepherd, Geof Wood, and
Zul¬qar Ali for helpful comments. For research assistance I am particularly indebted to
Saidozamman Jewel, M. Shahidur Rahman, Ro¬kul Islam, Jahangir Alam, Amit Chakrabarty,
and Habibur Rahman.




154
Poverty Dynamics in Bangladesh

life-trajectory patterns are identi¬ed from the life histories studied and
selected respondents™ stories used to illustrate each pattern.



7.2. Methodological Lessons Learned

In the research outlined here, people from three towns and six villages in
Kushtia district in western Bangladesh were interviewed during 1999“2001
(see Davis, 2005, 2006). A life history interview was conducted for twenty
people randomly selected from each site. In addiction a pro¬le was also
constructed collecting information on household members, extended family,
skills and education, religion, economic resources (income, assets, debts),
household facilities, coercive power, prestige, and networks and relationships.
This approach drew from the resource pro¬le approach (Lewis and McGregor,
1992; Lawson, McGregor, and Saltmarshe, 2000; McGregor, 2004) and other
livelihoods approaches (e.g. Carney, 1998), but deliberately took a power-
resource-focused perspective. 1 This allowed a number of ¬elds of power (Bour-
dieu, 1990) to be examined that are an important part of the informal social
protection system. These include: economic resources, access to bureaucratic
resources, the means of violence, social network resources, and social prestige.
Household pro¬les were constructed before the life history interview and
provided an initial platform of information allowing the life history interview
to be guided towards relevant areas.
The life history interviews were in bangla and took 2 to 4 hours, sometimes
over multiple sessions depending on the availability of respondents. As a
way of visualizing the life history at the end of each interview we mapped
out signi¬cant events in the history of Bangladesh on a chart. The vertical
axis was used to indicate the respondent™s well-being and the horizontal axis,
time. Life trajectory patterns were based on the perception of a person™s obosta
(loosely ˜life condition™) as it changed over time, constructed by myself and
the respondent. The term obosta was chosen because it is vague, is in common
everyday usage, and roughly translates as ˜life condition™ which seemed to be
a suitable proxy for well-being in bangla.
We began the interview by setting up a chronological framework of major
life events and avoided general questions about overall circumstances at the
initial stage of the interview. Other details could then be written onto the
template referring back to these dates to triangulate data. It also helped
to ˜warm up™ people™s memories. We started the interview by working out
1
The semi-structured interview drew from livelihoods (e.g. Carney, 1998) and resource
pro¬les approaches (e.g. Lewis and McGregor, 1992; Lawson,McGregor, and Saltmarshe, 2000;
McGregor, 2004) but extended the scope of interest beyond livelihood resources to what I
called ˜power-resources™ drawing from Korpi (1980, 1983, 1985) and Korpi and Palme (1998),
Bourdieu and Nice (1977), and Bourdieu (1986, 1990) in a more power-focused approach.




155
Peter Davis

concrete details to do with age, marriages, births, and deaths. Very few poor
people in rural Bangladesh are able to tell you their age accurately, so age was
often estimated with reference to national or local events that most people
remember.
If married, the respondent™s date of marriage was worked out. This was
followed by recording the births of children. If children™s birth dates had
been worked out with reference to date of marriage, it was often necessary to
work back from the present to check their ages. This was followed by working
out marriage dates of any married children. Dates of deaths of parents, close
relatives, and children were also asked about and recorded, including children
who had died around or shortly after birth.
With each marriage I asked how much dowry (joutok) was given and
received including other wedding costs, which were often quite considerable
relative to income. Often this led to discussions of loans taken out, land mort-
gaged, livestock sold, contributions from kin, and community collections. I
explored how loans were arranged and who had organized village collections.
This often revealed interesting information about patronage relations and
other signi¬cant network connections. I also asked about marriages of sisters
and other relatives as contributions to dowry are often also made in such sit-
uations. When others had helped, expectations of reciprocity were discussed.
Signi¬cant sicknesses were also investigated: who raised money to pay
for medical care? Who devoted labour for care? And where was treatment
sought? I started with the respondent™s parents and grandparents, as events
leading up to the death of parents often led to large expenditures and sale of
assets. As with other issues, asking sweeping questions about illness yielded
poor responses, but situating questions within a slowly emerging history
often allowed memories to be jogged. A number of interesting issues to do
with selective memory emerge, particularly when it comes to remembering
episodes of ill health. 2 The structure and sequencing of the interview was
important in helping to stimulate memory; it was particularly useful to map
out more memorable life transitions ¬rst so that links could be explored.
I also asked about events relating to land being divided among brothers
(usually), although sometimes sisters received a share. This time in life cycles
often corresponded with the formation of new households and had a sig-
ni¬cant bearing on individual well-being. This then led to the history of
employment and business. Attention was paid to periods of unemployment
and re-employment. Who helped out during unemployment? Who helped to
get another job? How were loans secured to start and maintain a business? I
asked about education and work. Are children supporting the parent? I asked


2
This has been a fruitful area of research among cognitive psychologists with important
implications for memory-based life histories (see Becker and Mahmud, 1984; Rubin and
Baddeley, 1989; Conway, 1990).




156
Poverty Dynamics in Bangladesh

about land ownership. When was land bought and sold? Had any land been
lost due to river erosion? Why was land sold? Had land been mortgaged?
Court cases were also a common cause of crisis. What were the cases over?
How long had they continued? What costs were involved? Were they asso-
ciated with any violence or intimidation? I asked about theft and any other
forms of violent con¬‚ict including police extortion or intimidation, which
seemed to be a common feature of poor people™s experience.
While I was aware of the topics and issues I wanted to cover, I usually
allowed the interview to follow its natural course in a conversational manner
without imposing too much structure. Topics that were clearly irrelevant were
ignored; unusual, signi¬cant, and interesting events were pursued in detail.
After my interviewing skills had developed it was possible to cover most
important issues with very little reference to the listed topics on my clipboard.
By keeping the interview conversational I found that information was slowly
revealed in a way that would not happen in a more formally structured
interview.
At the end of the life history interview a graph was drawn with the
respondent showing the various trends and the effect of crisis and oppor-
tunity episodes on well-being. The episodes were ranked from the worst crisis
labelled as (C1) followed by the second (C2) and third (C3). As signi¬cant
crisis episodes were identi¬ed in the course of the interview the (often com-
plex, multifaceted, and cumulative) causes were explored, followed by the
coping strategies employed during the crisis and the in¬‚uence on more long-
term well-being and security. When people™s circumstances had improved, the
main causes of these periods of opportunity were also identi¬ed (O1, O2, O3)
with relevant details noted.
Once a number of crises had been identi¬ed and discussed during the life
history interview, including a discussion of sources of insecurity and crisis,
the range of crisis coping strategies employed during particular episodes was
examined in detail. The relationship between various ˜power resources™ avail-
able to the individual or household was investigated, including economic
assets and income, social prestige and status, access to of¬cials and bureau-
cratic resources or to the means of violence, membership of organized factions
or parties, and ascribed identity (gender, regional identity, etc.). The analysis
of the use of these power resources aided in the understanding of the social
mechanisms which led to the differentiation between highly insecure people
and not-so-insecure people.


7.3. Terminology

With an explicitly dynamic focus, a number of de¬nitions become useful.
First, I used the terms ˜event™, ˜episode™, and ˜trajectory™ deliberately. An ˜event™



157
Peter Davis

referred to a short discrete period of time (up to about a month) in which a
crisis of single or multiple causation can occur. ˜Episode™ was used to describe
a longer period of time (up to about a decade) which is characterized by a
particular state of affairs (such as a chronic illness or a long-drawn-out court
case) and within which a number of ˜events™ could occur. The word ˜trajectory™
was reserved for still longer periods of time such as a person™s entire life or
large part of a person™s life, and could span a number of ˜episodes™.
Second, I used the words ˜transition™, ˜passage™, and ˜life stage™, drawing
from Dewilde™s (2003) usage, to connect events, episodes, and trajectories to
more predictable phases of a life course. A ˜transition™ referred to a socially
de¬ned change of state in a person™s life, which is to some extent predictable
and is usually abrupt. For example, this may be marriage, the death of a
parent, division of paternal property, or the birth of a child. The word ˜passage™
was reserved for a transition, or series of transitions, which results in a new
life stage. A ˜life stage™ is a form of identity which places people within
socially constructed phases in expected life courses, such as: being a child,
student, married person, parent, or elderly person. Life stages affect social
roles, responsibilities, prestige, power, and household structure.
Third, in order to situate life histories within community and national
contexts, I referred to groups of people in similar life stages at similar periods
of national history as a ˜cohort™ and the periods of national history as ˜eras™:
the 1971“4 war“famine nexus in Bangladesh is a good example of an ˜era™.



7.4. Categorizing Trajectory Patterns

The following discussion is based on 90 of the 242 interviews carried out. 3
The simplest way of categorizing the trajectories of respondents is (following
Lawson et al.™s 2000 groupings and the approach set out by Hulme and
Shepherd, 2003) to see people in declining, level, or improving trajectories,
where the variable changing is some measure of a person™s well-being. In
this section I attempt to build on such approaches but expand the number
of trajectory types, using not only the present trajectory direction, but also
the trajectory pattern observed over a longer period. I do this because much
more can be learned from life trajectories than merely whether the person™s
condition is at present level, declining, or improving. Trajectory patterns over
signi¬cant periods of time help us to piece together more interesting and
complex relationships and provide better scope for improving the ¬t between
patterns of crisis and social policy.

3
The chosen interviews were those which had yielded the most comprehensive life stories
and only those which I had conducted myself. The total of 242 interviews also included pilot
interviews.




158
Poverty Dynamics in Bangladesh

Table 7.1. Current trajectory direction of all respondents

Declining Level Improving

Total 91 47 104
Male 43 22 54
Female 48 25 50
Jibonpur 22 7 15
Haripur 16 4 18
Gopalpur 9 12 14
Kamalpara 6 6 8
Teliapara 5 10 5
Kumarpara 6 3 12
Mirpur 8 3 9
Goshpur 9 0 11
Dukhipur 8 2 10
Rural non-remote 31 23 47
Rural remote 25 5 30
Urban 17 19 29
Hindu 8 9 12
Muslim 82 37 89



Of the 242 respondents, 91 were judged to be at present declining, 47 were
level, and 104 were in improving trajectories. Overall, and in nearly all sub-
categories of respondents examined in Table 7.1, the numbers of improving
respondents tended to be slightly higher than the numbers of those declin-
ing. 4 This ¬nding is compatible with the overall trend of poverty reduction
in Bangladesh reported most recently by Sen and Hulme (2004). Differences
in numbers of those deemed to be ˜level™ should not be seen as signi¬cant
because the choice between seeing a trajectory as level or not was somewhat
arbitrary and subjective, differing signi¬cantly between interviewers and how
much detail the interview yielded. Broad variables such as religion, gender, or
urban/rural, remote/non-remote location did not show signi¬cant differences
in numbers of life trajectories declining or improving. What this approach
does offer, however, is a high level of detail in individual stories. The challenge
is then to analyse and aggregate this complex detail so that generalizations
can be made and policy lessons drawn.


7.5. Using Trajectory Patterns as Heuristic Tools in Poverty
Dynamics Studies

In order to analyse trajectory patterns I created a small number of categories
(or fuzzy sets (Ragin, 2000) ) or stylized patterns or types of life trajectory
which seemed to recur in the data. Such an approach takes a preliminary step
towards making generalizations over time in individual life trajectories and
4
Village names and respondent names have all been changed to ensure anonymity.




159
Peter Davis

Table 7.2. Ideal typical trajectory patterns

Trajectory Trajectory Depiction Number of cases
direction pattern (out of 90)


level smooth 6


improving smooth 3


declining smooth 6


level saw-smooth 17


improving saw-tooth 17


declining saw-tooth 14


declining single-step 13




declining 14
multi-step



across numbers of individuals. These ideal types of trajectories (or parts of a
life trajectory) are represented diagrammatically in Table 7.2. The following
discussion is framed using these eight trajectory categories.
A number of observations emerge as overall life trajectories were considered:
r Improvements in people™s life conditions tend to happen only gradually,
whereas sudden declines were much more common. People rarely win a
lottery, but they can frequently and suddenly become ill, lose their land,
spouse, or income.
r Crises are more likely to produce serious and sudden declines when the
crisis either directly affects something constitutive of a person™s well-
being, 5 such as their health, or when a person has very few ˜buffers™ (e.g.
low resilience due to previous crises, no insurance resources, few assets
or savings, poor network resources). Most poor people have few buffers
and are therefore more likely to translate a crisis into a serious decline in
well-being. 6

5
See Sen (1998) for a useful discussion on the distinction between ˜constitutive™ and
˜instrumental™ determinants of well-being.
6
See Room (2000) for a useful conceptual framework using ideas of snakes, ladders,
passports, and buffers to describe a dynamic view of processes of social exclusion.




160
Poverty Dynamics in Bangladesh

r Life trajectories and parts of life trajectories (episodes) can be categorized
into a fairly small number of patterns. For example, some trajectories were
marked by one big crisis that overshadowed the rest of the person™s life
(what I call a ˜single-step decline™); others resemble more the teeth of a
single action saw (˜saw-tooth™), gradual improvements interspersed with
more abrupt declines; others are fairly smooth all the way (˜smooth™). In
further research it may be useful to use categories such as these as fuzzy
sets, as a way forward in bridging the qual“quant divide in studies of
poverty dynamics (see Ragin, 2000).




7.6. Trajectory Direction

7.6.1. Improving trajectories
There were only two categories which described improvement: (i) ˜improv-
ing smooth™ and (ii) ˜improving saw-tooth™. The lack of other patterns
re¬‚ects the difference between decline and improvement in general: declines
are often steep but improvements are not. Because of this, sudden single-
and multi-step improvements did not appear among this sample of non-
metropolitan Bangladeshis interviewed. Long-term improvements are either
slow and smooth (usually for the more resource rich) or they consist of
slow improvements interspersed with sudden declines, which nevertheless
are not serious enough to undermine an overall upward trend. The ˜improv-
ing saw-tooth™ pattern which results is the most common trajectory type
for poor people on an improving trajectory. An understanding of this is
relevant for conceptualizing social policy interventions: if life trajectories,
even among those which are improving, are interspersed with setbacks,
anti-poverty policy can have two aims: to support processes which allow
gradual periods of improvement to occur on the one hand, and to pre-
vent or mitigate events which cause inevitable sudden declines on the
other.



7.6.2. Declining trajectories
In contrast with the two forms of improvement there were four patterns of
decline. Obviously in many actual cases it was dif¬cult to decide which type
(or combination of types) the idiosyncratic and complex real-world patterns
corresponded most closely to. However, the distinctions are still useful for
heuristic purposes because ideal types, created by abstracting out from the
real-world cases, can in turn be re¬‚ected back onto real-world events and
processes, as tools for further questioning and analysis.



161
Peter Davis

(I) DECLINING SAW-TOOTH
It seemed to me that many trajectories of the poor people interviewed
resemble the teeth of a saw. Periods of slow improvement were commonly
interspersed with more sudden downward falls. When the falls outweighed
the improvements an overall downward trajectory resulted. This ˜declining
saw-tooth™ pattern has some resonance with Chambers™s idea of a downward
ratchet (Chambers, 1983). However, it is useful to distinguish between tra-
jectories where there is scope for improvement between downward steps and
where no scope for improvement or recovery occurs. This distinction is not so
clear in the ˜downward ratchet™ analogy.

(II) SMOOTH DECLINE
Some trajectories decline smoothly rather than in sudden steps. Smooth
decline patterns are less common and tend to appear when crisis episodes
are long term (drawn-out court cases, chronic illness, underemployment) or
when there is some long-term underlying or structural cause of disadvan-
tage which precludes improvement (such as ˜adverse incorporation™ (Davis,
1997, 2001; Wood, 1999) within constraining or exploitative patron“client
dyads).
r Single-step decline. Single-step declines describe a trajectory which is char-
acterized by a crisis event (which may be of composite causation) which
has an overshadowing and de¬ning signi¬cance due to its catastrophic
impact. This may be due to a relative™s death, an accident or serious
illness, a court case, or a catastrophic combination of adverse events
(double and triple whammies) occurring at around the same time. For
many poor women, their abandonment, divorce, or widowhood, if it had
happened, often constituted such a de¬ning event. Also people can be
more vulnerable to such declines during transitions or passages between
life stages such as the death of parents and the associated division of
parental property, the beginning and end of marriages, and during or after
migration.
r Multi-step decline. Multi-step decline was similar to a declining saw-tooth,
the difference being more in terms of a small number of serious crises (2“
5) with little improvement or recovery in between. The lack of recovery
between crises often suggested a lack of resilience, particularly associated
with more vulnerable individuals.


7.6.3. Level trajectories
Two categories of level trajectory appeared: what I refer to as (i) ˜level saw-
tooth™ and (ii) ˜level smooth™. These can occur at relatively high, medium,
or low levels of ˜life condition™ (obosta). Saw-tooth trajectories at a fairly



162
Poverty Dynamics in Bangladesh

low level of ˜life condition™ are the most common type experienced by the
chronically poor. For the individuals who were extremely poor, a low but ˜level
saw-tooth™ type trajectory re¬‚ected the way that these people were barely
surviving, avoiding a declining trajectory largely because there was little scope
for further decline in their life condition without total destitution and death.
It was usual, at this very low level, for moral-economy norms corresponding
with Scott™s ˜subsistence ethic™ to appear, which compel relatives, patrons, and
neighbours to provide for these individuals in a number of ways crucial for
their survival (Scott, 1976). Some of these processes, particularly those involv-
ing extended kinship relationships, have also been identi¬ed in Bangladesh
by Indra and Buchignani (1997). While in their highly vulnerable states,
however, these people were often beset by regular crises, for example: illness,
unemployment, and various forms of violence, domination, and coercion.
This suggests that low levels of informal social protection are not suf¬cient to
prevent long-term harm.
˜Level smooth™ trajectories were more associated with the better off. The
smooth trajectory usually re¬‚ects access to resources which effectively buffer
against regular crises. For these people health problems were dealt with
through relatively expensive private clinics, legal disputes resolved due to
access to a number of social and economic power resources, and dowries
paid for from savings and sale of assets without damaging quality of life. In
addition to the better off, however, a smaller number of individuals without
signi¬cant assets and low income also had fairly smooth and level trajectories.
These tended to be younger respondents, sometimes married couples with
young children who had not yet faced dowry problems, with their parents
still relatively young and healthy, and with property not yet divided between
brothers”an event which usually occurred after the death of parents. How-
ever, the level trajectories enjoyed by these poorer young people tended to
be short-lived once they moved on in their life cycles when parents became
old and daughters needed dowries. A smaller number of single older people
were also living fairly crisis-free lives, re¬‚ected in the ˜level smooth™ trajectory
pattern. In these cases the level smooth trajectory was due more to good
fortune than to resilience or a lack of vulnerability. It occurred while elderly
people were healthy, had children and neighbours caring for them, and had
few responsibilities for others.


7.7. Trajectory Patterns

In the following section I draw attention to the range of trajectory patterns
identi¬ed above using selected cases from actual life history interviews. The
diagrams used are a distillation of information from actual life history inter-
views and the life trajectory diagrams drawn with respondents. For the sake



163
Peter Davis

Table 7.3. Examples of causes of declining smooth patterns

Main cause Gender Current age

1. A combination of low household income M 63
and several chronic health problems
2. Chronic illness M 51
3. Husband™s chronic illness F 35
4. Mother-in-law™s chronic illness (TB) F 32
5. Low income and chronic illness M 22
6. Low income and brother™s chronic illness M 20




of clarity, only the most relevant information is included in the diagrams.
Accompanying discussions are based on my interpretation of relationships
and processes emerging from a much wider set of life history interviews. I use
the eight trajectory types to organize the discussion.



7.7.1. Smooth trajectories
Smooth trajectories were not common. Approximately ¬fteen smooth trajec-
tories (17 per cent) could be identi¬ed from the ninety most comprehensive
life histories but this ¬gure should be used with caution. Life history inter-
views varied greatly in the quality of information gathered, and a smooth
trajectory could be erroneously drawn when an interview did not go well
because there was insuf¬cient information available to describe the ups and
downs in a person™s life. In interviews where a rich and accurate level of detail
was recalled, a more accurate trajectory was drawn, and this tended not to be
smooth. However, in some cases, the smoothness of the trajectory was not due
to a lack of detail: it re¬‚ected either a period of relatively few serious crises, or
suf¬cient resources to deal with them”particularly in the ˜level smooth™ and
˜improving smooth™ cases. In the declining smooth cases the trajectory usually
re¬‚ected chronic and long-term downward pressures with few countervailing
opportunities.


(I) DECLINING SMOOTH
Most cases of smooth decline in individual life condition were associated
with chronic illness of the respondent, or someone close to the respondent.
Table 7.3 provides examples of six cases which could clearly be identi¬ed as
˜declining smooth™. For these six, it was clear that the trajectory was actually
fairly smooth rather than an interview which had failed to yield adequately
detailed information. A number of other cases were categorized as ˜declining
smooth™ but, due to the paucity of information from the interviews, I had
much less con¬dence in drawing conclusions from these.



164
Poverty Dynamics in Bangladesh

The following case of Sukur Ali (number 1 in Table 7.3) illustrates a
smoothly declining life trajectory caused mainly, but not exclusively, by the
chronic health problems suffered by other members of his household.
Sukur Ali is a 63-year-old man who has worked as a village chowkidar,
employed by the Union Parishad, since 1979. His wife Momataz is 50 and
works as a day labourer in a local tobacco-processing plant. They have a 16-
year-old daughter living with them in the same ghor (room). In addition Sukur
Ali™s brother eats with them but sleeps in a separate ghor in the same bari or
homestead. The brother has TB and has not been able to work for the last
three years. There are three separate rooms in the bari. Sukur Ali earns 700
Tk. per month as a chowkidar but is not usually paid on time. The recurring
delay in wages exacerbates his dependent relationship on the Union Parishad
chairman who oversees his employment.
Before 1979 Sukur Ali had worked as a rickshaw driver for twenty-nine
years. When he was married in 1965 his father owned 15 bighas 7 (5 acres)
of land. In 1962 his father became ill and over the following ¬fteen years
(which also spanned the war of independence) they sold all of their land to
pay for their father™s treatment (8 bighas before the war and 7 after). During
the 1974 famine they also sold their rickshaw to buy food. At that time also
Momataz was ill with an abscess which persisted for a year. Sukur Ali™s father
died in 1977 after ¬fteen years of illness. Momataz was pregnant four times
between 1974 and 1984 but all of the children died around the time of birth.
In 1987 they arranged the marriage of their eldest daughter and negotiated
the dowry payment of Tk. 3,000 to be paid over one year.
Figure 7.1 shows a simpli¬ed version of that drawn during the interview
and attempts to clarify the relative impact of these events and episodes on
Sukur Ali™s well-being over time. The combination of chronic illness (father,
wife, and brother) and other stresses (dowry, crop losses) produced a smoothly
declining trajectory. Low and intermittent income and a denuding asset base
do little to mitigate these long-term downward pressures.
In these cases the smoothness of the decline often re¬‚ects chronic problems

<<

. 6
( 13)



>>