Professional Documents
Culture Documents
Dorosh 2016
Dorosh 2016
xxx–xxx, 2016
0305-750X/Ó 2016 Published by Elsevier Ltd.
www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2016.08.014
Gollin, 2016; Dorosh & Thurlow, 2013, 2014). There are fewer includes statistics for all low-income countries in Sub-
studies, however, that explicitly estimate PGEs within non- Saharan Africa (SSA).
agriculture. One study by Thirtle et al. (2003) disaggregates The five case studies were selected because detailed social
non-agriculture and finds that, unlike for agriculture, there is accounting matrices (SAMs) are available for these countries
no significant relationship between poverty reduction and for a common base year. Although our small sample of coun-
growth in either industry or services. This suggests that sepa- tries accounts for only a fifth of low-income Africa’s total pop-
rating non-agriculture may be unnecessary. However, Loayza ulation, it still reflects the subcontinent’s diversity. Malawi and
and Raddatz (2010) disaggregate industry, and find a signifi- Zambia offer the widest contrast for many of the indicators
cant relationship between growth and poverty for labor- shown in the table. Malawi’s population is overwhelmingly
intensive manufacturing and construction, but not for mining. rural, whereas Zambia is one of Africa’s most urbanized coun-
Similarly, Suryahadi, Suryadarma, and Sumarto’s (2009) tries. GDP per capita is seven times higher in Zambia (US
study of Indonesia finds that while rural agricultural growth $950) than it is in Malawi (US$133), and is still more than
is strongly linked to poverty reduction, so too is growth in three times higher after adjusting for purchasing power parity
urban services. These heterogeneous outcomes at the subsec- (PPP). Zambia has a large mining sector, which produces three
toral level confirm the need for more detailed analysis of quarters of the countries’ exports. In contrast, agriculture
non-agriculture and its linkages to poverty. accounts for three quarters of Malawi’s exports and a third
In this paper we complement the expanding literature on of its GDP. Agro-processing, which falls under manufacturing
agricultural growth and poverty by estimating PGEs for differ- and includes the production of food, beverages, and tobacco,
ent nonagricultural subsectors. This is done using dynamic is also more important for Malawi than it is for Zambia. 2
computable general equilibrium (CGE) models of five low- Overall, the sample includes two predominantly agrarian
income African countries, which reflect a range of initial con- countries (Malawi and Tanzania) and two countries where
ditions, including a varying importance and composition of mining and heavy industry play more important roles
agriculture, industry, and services. The models allow us to (Mozambique and Zambia). Finally, Uganda is more depen-
experiment with alternative patterns of growth and to link dent on construction and services.
growth to poverty using a consistent macro–micro framework. The table also reports poverty headcount rates, which mea-
In order to permit cross-country comparisons, the model’s sure the share of the population whose daily expenditures fall
databases are constructed for a common base year and pov- below the US$1.25 and US$0.75 poverty lines (measured in
erty is defined using a common poverty line. Changes in pov- 2005 PPP-adjusted dollars). Again, the case studies reflect a
erty are measured using micro-simulation techniques. The next range of initial conditions. Uganda has the lowest poverty rate
two sections present our case study countries and describe the in our sample (50.3%) due to its higher per capita consumption
models. We then present our simulation results and conclude spending (US$632 per year according to survey data). Con-
by summarizing their implications. versely, Tanzania has a higher poverty rate (69.9%) and lower
per capita consumption (US$441 per year). Average consump-
tion measures hide differences in income inequality. Zambia
has the most unequal expenditure distribution, i.e., the highest
2. CASE STUDY COUNTRIES Gini coefficient. This explains why Malawi and Zambia have
similar poverty rates, despite Malawi’s much lower per capita
Our analysis is based on five African countries: Malawi consumption.
(MAL), Mozambique (MOZ), Tanzania (TZA), Uganda The structural heterogeneity discussed above is central to
(UGA), and Zambia (ZAM). Table 1 reports these countries’ our analysis of growth–poverty linkages. Differing composi-
economic and demographic characteristics for 2007, which is tions of non-agriculture across countries can cause their aggre-
the base year for the economy-wide models. The table also gate growth–poverty relationships to vary. Mozambique and
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
BEYOND AGRICULTURE VERSUS NON-AGRICULTURE 3
Zambia, for example, have large mining and metals industries, this approach are relevant here. First, it assumes that economic
which are typically associated with weaker employment and growth does not alter macroeconomic structure, i.e., there are
poverty linkages (see Thurlow & Wobst, 2006). In contrast, proportional changes to the trade balance (X M), indirect
agro-processing plays a larger role in Malawi and Uganda, taxes (T), and the composition of total absorption (C + I
and this subsector tends to be more labor intensive and less + G). Secondly, the focus on household consumption precludes
skill intensive than metals or mining. We might therefore an explicit examination of growth patterns, i.e., uneven changes
expect smaller aggregate non-agricultural PGEs in Mozam- in sectoral GDP at factor cost. Survey-based PGEs are often
bique and Zambia than in Malawi or Uganda. The differing only decomposed in general terms, such as by treating urban
compositions of non-agriculture in our five case study coun- consumption as a proxy for nonagricultural GDP (see
tries both motivates and strengthens the analysis. Ravallion & Datt, 1996). This constrains the level of sectoral
Differences in a sector’s production and consumption link- analysis and has prompted most studies to overlook inter-
ages can also cause PGEs to vary across countries. Production sectoral spillover effects, such as when agricultural growth ben-
linkages arise when producers use intermediate inputs from efits both rural and urban consumers. 3
other sectors. For example, when the transport sector A purely survey-based approach is therefore not ideal for
expands, it uses inputs like fuels and repair services, thereby decomposing PGEs. Instead, we use a combination of CGE
stimulating production in other sectors or demand for and micro-simulation models. The CGE models provide a
imported goods. The size of production linkages depends on consistent structural framework that explicitly traces the
the technologies and input-intensities of sectors. Consumption impact pathways between sector-level growth and
linkages on the other hand occur when workers’ incomes are household-level consumption. The models also allow us to
used to purchase a range of goods and services beyond those control for changes in macroeconomic structure. The micro-
produced by the sector in which the workers themselves are simulation module retains one of the major strengths of
employed. The size of consumption linkages depends on the survey-based approaches, i.e., it captures the heterogeneity
share of labor and capital incomes distributed to households, of household consumption patterns.
the composition of consumption baskets, and the share of Following the approach in Dervis, De Melo, and Robinson
domestically-supplied goods in consumer demand. These dee- (1982), our country CGE models simulate the functioning of a
per structural characteristics are captured in our five country market economy, including markets for land, labor, capital,
SAMs and hence in the economy-wide models that we use to and products, and offer insights into how accelerations in sec-
measure sector-specific PGEs. toral growth are mediated through prices and resource reallo-
cations. All resource and macroeconomic constraints are
respected, which is an advance over fixed price and semi-
3. MEASURING POVERTY–GROWTH ELASTICITIES input–output multiplier models that are often used to measure
agricultural and nonagricultural growth linkages (Haggblade
The main PGE used in this study is defined as the percentage et al., 2007; Khan, 1999; Thorbecke & Jung, 1996). Below
change in the poverty headcount rate given a one percent we describe the CGE models’ structure and behavior. 4
change in GDP per capita. This definition of the PGE is sen-
sitive to the initial magnitude of the poverty rate. For instance, (a) Production, employment, and trade
economic growth in two countries may lift the same share of
population out of poverty, but the PGE will be higher in the The contribution of a subsector to national growth is deter-
country with the lower initial poverty rate. A ‘‘semi- mined by its initial share of GDP and its linkages to the rest
elasticity” overcomes this by defining the PGE as the absolute of the economy. If a subsector has strong production and con-
or percentage point change in the poverty rate given a one per- sumption linkages (i.e., if it uses inputs from local producers
cent change in GDP per capita (see Misselhorn & Klasen, or generates incomes that support domestic consumption) then
2006). We report both types of PGEs, but we focus on the for- growth in this sector could generate positive inter-sectoral spil-
mer, since this is the definition most commonly used in the lover effects. For example, agriculture is said to have strong
empirical literature on growth–poverty linkages. growth-linkages because it supplies raw materials to down-
A further complication in measuring PGEs is that poverty stream sectors (i.e., forward production linkages); uses
calculations are based on household consumption rather than locally-produced trade and transport services (i.e., backward
GDP. As shown in the national accounting identity below, production linkages); and generates incomes for households
GDP includes not only private consumption C, but also who are more likely to purchase domestic goods (i.e., consump-
investment I, government spending G, and the net flow of tion linkages). These linkages are country- and sector-specific
exports X and imports M. Moreover, consumption is mea- because they depend on local production arrangements and
sured at market prices (mp) and therefore includes indirect household income and expenditure patterns. Our models cap-
taxes T. This implies that economic growth, measured using ture growth linkages through their calibration to SAMs.
GDP at factor cost (fc), might not lead to proportional The models include detailed breakdowns of industries and
changes in household consumption. For example, changes in production technologies. Agriculture is divided across subna-
the growth pattern might alter marginal savings and tax rates, tional regions in order to capture differences in agro-
and thus affect investment and public spending differently ecological conditions and livelihood patterns. Labor markets
from private consumption. are segmented by education levels, with labor in each group
able to migrate between sectors. One exception is lower-
GDP fc þ T ¼ GDP mp ¼ ðC þ I þ GÞ þ ðX MÞ educated workers, whose annual time allocation between farm
Most studies in the literature avoid making an explicit link and nonfarm work is fixed or predetermined. This reflects the
between GDP and consumption. One approach that is often seasonal labor constraints facing farm households and ensures
used is to estimate PGEs directly from household surveys by that sufficient family labor is available during the growing sea-
assuming that changes in total reported consumption can proxy son. Note that, outside of the growing season, farmers can
for changes in total GDP (see, for example, the studies in Besley allocate their time to nonfarm activities. We initially assume
& Cord, 2007; Grimm, Klasen, & McKay, 2007). Two aspects of that the size of a country’s total workforce is unaffected by
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
4 WORLD DEVELOPMENT
productivity growth, i.e., total labor supply is inelastic and mate PGEs assume that changes in total GDP are propor-
wages adjust to equate demand and supply. However, given tional to changes in total private consumption. We select
possible underemployment outside of the growing period closure rules that mimic this assumption in order to allow
and in urban centers, we will later drop this assumption for for more direct comparisons between ours and other studies.
less-educated nonfarm workers (discussed in the next section). First, we assume that the real exchange rate adjusts in order
In all scenarios, workers are able to move between sectors in to ensure that the trade balance (X M) is held fixed in for-
order to maximize the marginal returns to labor. The models eign currency. Secondly, we assume that changes in total nom-
therefore capture how changes in sectoral productivity can inal absorption (C + I + G) are distributed proportionally
lead to changes in sectoral employment patterns (i.e., struc- across its three aggregate components. This is achieved via
tural change). proportional adjustments to marginal savings and direct tax
Linkages between domestic sectors and the rest of the world rates, with indirect tax rates (T) remaining constant. Together
are important for determining how growth affects prices and these closure rules maintain existing macroeconomic struc-
incomes. One argument against the need for agricultural tures, such that our estimated PGEs are primarily determined
growth in Africa, for example, is that food imports are now by micro-structure rather than macro-behavior. We replicate
more readily available than they were at the time of Asia’s the assumption underpinning survey-bas7ed PGEs, but with-
Green Revolution (see Diao et al., 2010). Conversely, sectors out losing the ability to isolate growth in particular sectors
with larger export market opportunities may generate higher and capture inter-sectoral spillover effects.
returns but fewer indirect linkages to other sectors. The
CGE models capture import competition and export opportu- (d) Data and limitations
nities by allowing producers and consumers to shift between
domestic and foreign markets/goods depending on changes Most of the CGE models’ parameters are derived from
in relative prices. We assume that domestic and foreign goods SAMs. A SAM is a consistent economy-wide database that
are imperfect substitutes. Finally, world prices are fixed under tracks all income and expenditure flows within an economy
a small country assumption. during a given year (see Breisinger, Thomas, & Thurlow,
2009). The SAMs used in this study were produced by the
(b) Household incomes and poverty International Food Policy Research Institute. 5 They draw
on a range of datasets and this process revealed various incon-
Changes in the poverty headcount rate is determined by sistencies between sources. In our case studies, as in most low-
households with per capita consumption levels close to the income countries, national accounts are not fully reconciled
poverty line. This implies that GDP growth can raise total with government financial statements or the balance of pay-
consumption without necessarily affecting the poverty rate. ments. There are also differences between total consumption
This highlights the importance of capturing distributional levels in national accounts and in national household surveys.
change. Income changes in CGE models can vary across In order to reconcile these differences, cross-entropy
households depending on the factor intensity of a particular estimation techniques were used to systematically minimize
growth pattern and on the claims that households have on fac- adjustments to the original data (see Robinson, Cattaneo, &
tor earnings. Drawing on information on employment and El-Said, 2001). If national accounts and household surveys
factor endowments from household surveys, the CGE models both contain useful information, then the reconciliation pro-
trace the impact of GDP growth on the returns to different cess will have improved the information base of our study.
factors and on the level and distribution of incomes and con- Nevertheless, like most studies in the growth–poverty litera-
sumption of different household groups. ture, our analysis depends on the accuracy of sectoral GDP
National surveys are used to define representative house- estimates from national accounts and poverty estimates from
holds in the CGE models, each of which is an aggregation household surveys.
of survey respondents. We group households according to The models also include behavioral elasticities that influence
rural and urban areas, subnational regions, farm and nonfarm how consumers and producers respond to changes in incomes
status, and per capita expenditure quintiles. In order to retain and relative prices. Import and export substitution elasticities
as much information as possible on household heterogeneity, are based on cross-country estimates from Dimaranan (2006).
the CGE models are linked to survey-based micro- Income elasticities were econometrically estimated using
simulation modules following the approach described in national household surveys, which also provided information
Arndt et al. (2012). In this specification, each household group on sectoral employment and wages. When presenting our
in the CGE models is mapped to their corresponding house- results, we conduct sensitivity analysis on the models’ calibra-
holds in the surveys. The CGE model simulations estimate tion of labor markets.
changes in real commodity-level consumption spending for Finally, it is worth reflecting on the difference between PGEs
each household group. This information is used to update con- estimated using econometrics versus simulation models. The
sumption for individual survey households, assuming that all former have the advantage of being grounded in observation,
households within each representative group experience the but, as discussed earlier, data limitations make it difficult to
same proportional change in commodity-level spending. The decompose sector-level growth–poverty relationships. Struc-
total level of consumption for each survey household is recom- tural models, on the other hand, allow us to isolate sectoral
puted and compared to the poverty line in order to determine growth, but the link to poverty is partly achieved through
if their poverty status has changed. assumptions about macro and micro behavior. We apply the
same behavioral assumptions to each country such that any
(c) Macroeconomic closures cross-country variation in PGEs is entirely due to differences
in countries’ underlying data. Structural models also control
One feature of CGE models is that they can specify macroe- for external factors that introduce noise into econometric
conomic behavior. This is done using ‘‘closure rules” that studies, such as weather shocks. However, excluding these
determine how macroeconomic balances are maintained. As external factors means that our estimated PGEs cannot be
discussed earlier, studies that use household surveys to esti- empirically tested using historical data. Recognizing these
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
BEYOND AGRICULTURE VERSUS NON-AGRICULTURE 5
caveats, we still compare our PGEs to historical trends and GDP. For example, agriculture accounts for about one third
results from other studies. of total GDP in Malawi and so these sectors need to expand
by 7.5% in the agriculture-led growth scenario in order to gen-
erate a 2.5% increase in total GDP. Ultimately, the required
4. SIMULATION RESULTS sectoral productivity growth in each scenario depends on the
lead sectors’ initial shares of GDP and the strength of their
(a) Baseline and simulation design economy-wide growth linkages. 6 Finally, CGE models con-
tain non-linear functions that can constrain rapid growth in
We first establish a baseline scenario for the period 2007–15. small sectors. As such, we do not report elasticities for lead
This provides a counterfactual for alternative growth scenar- sectors that need to more than double in size by 2015 in order
ios. We adopt the same baseline assumptions for each country. to achieve the 2.5% increase in total GDP.
Labor and capital supplies grow at three percent per year, cul-
tivated land expands at half a percent per year, and total fac- (b) PGEs under a balanced growth scenario
tor productivity (TFP) grows at a uniform two percent per
year in all sectors. This generates a fairly balanced growth Poverty is initially measured using the US$1.25 per day pov-
path in which total GDP expands at roughly five percent per erty line. The estimated PGEs are reported in Table 3. The All
year (or two percent in per capita terms). Sectors scenario reflects a balanced growth path with a uni-
In our analysis we accelerate economic growth by increasing form increase in TFP in all sectors relative to the baseline.
productivity in specific sectors, e.g., agriculture or manufac- In Malawi, a one percent increase in total GDP led by all sec-
turing. Our general equilibrium framework takes into account tors causes a 0.76% reduction in the poverty rate. 7 The mag-
how growth originating in one sector may generate additional nitude of the PGE varies across countries—it is largest in
growth in other sectors via indirect production and consump- Uganda (1.20).
tion linkages. For example, increased agricultural production By design, the variation in PGEs across countries in the All
may expand downstream food processing within the manufac- Sectors scenario is not due to the rate or composition of GDP
turing sector. Increasing agricultural productivity may there- growth. Moreover, given our closure rules, the 2.5% increase
fore increase both agricultural GDP (direct effect) and in total GDP generates a similar 2.5% increase in total private
manufacturing GDP (indirect effect). We therefore refer to consumption in each country. The variation in PGEs in the All
our simulations as being ‘‘led” by the sector where the initial Sectors scenario therefore reflects differences in the level and
productivity growth occurs (e.g., ‘‘agriculture-led”), even distribution of private consumption. Figure 1 shows cumula-
though some of the increase in national GDP may occur out- tive distribution functions (CDFs) for per capita consumption.
side of this sector (e.g., in downstream manufacturing). Our In Uganda, for example, 50.3% of the population consume at
estimated PGEs therefore incorporate the general equilibrium a level of US$1.25 or less per day, which is equal to the pov-
effects that are implicitly captured in econometric-based esti- erty headcount rate reported in Table 1. If all households’ con-
mates (see Christiaensen et al., 2011; Suryahadi et al., 2009). sumption increases, then the CDF shifts to the right and
The effect on poverty of a given percentage change in sec- poverty rates decline.
toral productivity is expected to fall as the share of that sector It is evident from Table 3 and Figure 1 that the variation in
in total GDP declines. In order to control for the different size PGEs across countries is strongly related to base-year poverty
of each sector, we target the same percentage increase in total rates (see discussion in Ravallion & Huppi, 1991). This is
GDP in all sector-led growth scenarios, i.e., a 2.5% increase because the PGE reports a percentage change in the poverty
relative to the baseline in 2015. Table 2 reports the implied rate. Countries with lower poverty rates are more likely to
increase in total GDP per capita in each country (in US dollar achieve larger percentage changes. Accordingly, the PGE in
terms and relative to baseline values in 2015). The table also the All Sectors scenario is largest for Uganda, where the
reports by how much the lead sectors in each scenario are base-year poverty rate is lowest. The opposite is true for
required to grow in order to achieve a 2.5% increase in total Malawi. Therefore, given the way these PGEs are calculated,
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
6 WORLD DEVELOPMENT
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
BEYOND AGRICULTURE VERSUS NON-AGRICULTURE 7
reducing poverty than nonagricultural growth, whereas it is for example, generates demand for raw tobacco inputs, which
only 1.4 times more effective in Zambia. This is due to differ- is mostly supplied by smallholder farmers near to the poverty
ences in countries’ structural characteristics, rather than the line. Similarly, Zambia’s grain milling sector uses
definition of poverty, which is the same across countries. 8 domestically-grown maize and wheat. In contrast, almost a
Consistent with Loayza and Raddatz (2010), we find that third of Uganda’s agro-processing is in meat and dairy, which
the distribution of the benefits of growth is determined by generates demand for livestock that tend to be owned by
the factor intensity of production and the factor endowments better-off farmers. Finally, agro-processing in Malawi, Tanza-
of households. Changes in relative prices also affect real con- nia, and Zambia tends to favor lower-educated workers more
sumption levels. Table 6 shows the ratio of the lowest to the than it does in Mozambique and Uganda, where processing is
highest household quintile’s real per capita consumption relatively more capital- and skill-intensive.
growth. A value greater than one implies a narrowing of the Trade and transport services also have stronger growth–
gap between top and bottom quintiles. In all countries, this poverty linkages. In Zambia, for example, the PGE for these
gap narrows under agriculture-led growth and widens under services is very close to agriculture’s. Two reasons explain
non-agriculture-led growth. This is because agricultural this strong link. First, increasing the productivity of trade
growth generates higher returns for those factors that low- and transport in the CGE models reduces transaction costs
income households are more endowed with, and reduces the for all marketed products. This is particularly beneficial for
real prices of products that low-income households consume sectors whose products have high margins, such as agricul-
more intensively. ture and food. Increasing traders’ productivity therefore also
As expected, when nonagricultural growth is decomposed, increases the profitability of farmers and agro-processors.
we find considerable variation across subsectors. As shown This is consistent with Adam et al. (2016), who find that
in Table 3, the PGEs for growth led by manufacturing are reducing transaction costs increases incomes for lower-
much closer to agriculture’s in Malawi and Tanzania, and they skilled workers in Tanzania. Secondly, most trade services
exceed agriculture’s in Zambia. This is mainly due to agro- in countries like Zambia are produced by relatively low-
processing, which directly employs poorer workers and has paid informal traders (Resnick & Thurlow, 2016). Trade
strong backward linkages to the types of agriculture that poor and transport services therefore have strong direct and indi-
households are engaged in. Malawi’s tobacco curing sector, rect linkages to the poor.
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
8 WORLD DEVELOPMENT
We find that growth led by construction and government (d) Sensitivity analysis
services tend to have lower PGEs. This is partly due to produc-
tion technologies. While government services usually has a We conclude our analysis by testing the robustness of our
high labor value-added share in total GDP, the sector also findings to changes in the poverty line and the flexibility of
tends to be one of the most skill-intensive sector in African nonfarm labor markets. Table 7 reports PGEs estimated using
economies, i.e., it employs doctors, teachers and bureaucrats. a US$0.75 per day poverty line.
Most of the additional labor value-added from an expansion Virtually all PGEs increase when the poverty line is lowered.
in government services therefore accrues to relatively few As discussed earlier, in the All Sectors scenario, this increase is
well-paid workers who are less likely to live close to the pov- mainly due to having lower initial poverty rates. For example,
erty line. Moreover, government services, like health and edu- when we switch from the US$1.25 to the US$0.75 poverty line,
cation, do not appear as household consumption spending, there is a smaller fall in the poverty rate in Zambia than in
since they are often freely provided by the government. As Uganda. As such, while Zambia’s PGE for the balanced
such, expanding public services may not translate into higher growth scenario remains virtually unchanged, Uganda’s
consumption and hence lower monetary poverty for house- increases substantially. Again, our estimated PGEs using a
holds. This is true at least over the medium term, before US$0.75 poverty line in the All Sectors scenario are consistent
improved health and education have had the chance to raise in magnitude with the observed PGEs reported in Table 4.
household earnings. Changing the definition of poverty can alter the factor
Our finding that construction has a low PGE differs from endowment profile of households living near the poverty line,
Loayza and Raddatz (2010). One explanation comes from which is important in determining how effective growth in a
the way in which growth is modeled in our analysis, i.e., as particular sector is at reducing poverty. In Tanzania, for
a supply-side increase in productivity rather than an increase example, agricultural growth becomes even more effective at
in demand. Construction is almost exclusively used for invest- reducing poverty relative to nonagricultural growth when
ment, which is in turn determined by the level of savings. A the poverty line is lowered. This is because households near
significant portion of the total savings in our case study coun- the lower poverty line are even more dependent on farm
tries come from external borrowing and foreign aid, which we incomes and on consuming agricultural goods than house-
assume are fixed in foreign currency. Demand for construction holds near the higher poverty line. This is consistent with
is therefore constrained by the level of domestic savings. Christiaensen et al. (2011) and Ligon and Sadoulet (2014),
Increasing productivity in construction leads to large reduc- who find that agricultural growth has a larger beneficial effect
tions in its price, or an increase in the real price of other goods on poorer households. Regardless of these changes, however,
and services, including those that are consumed by poor the rank order of sectoral PGEs remains virtually unchanged
households. The low PGE for construction therefore partly compared to those reported in Table 3. Our conclusions are
reflects our decision not to finance economic growth through therefore robust to lowering the poverty line.
additional foreign borrowing or development assistance. This We also consider whether our results are sensitive to our
caveat does not apply to the same extent in our other sectoral assumption that the size of the nonfarm workforce is fixed
growth scenarios. at baseline levels. If there is substantial underemployment in
In summary, we find that agricultural growth is more effec- Africa’s urban centers, as studies like Roubaud and Torelli
tive than aggregate nonagricultural growth in reducing pov- (2013, pp. 70–73) suggest, then faster productivity growth
erty in our five country case studies. However, when may lead to increased labor force participation rather than just
nonagricultural growth is decomposed, we find that certain higher wages. We therefore introduce a flexible labor supply
subsectors, such as manufacturing and trade, often have PGEs function in the CGE models. Labor supply for less-educated
that are closer to, and sometimes exceed, agriculture’s. The nonfarm workers is now driven by changes in the wage rate
composition of nonagricultural growth is therefore an impor- with a unitary elasticity. An increase in economic growth
tant determinant of its overall effect on poverty. Consequently, and labor demand will lead to lower underemployment in
different perspectives on what constitutes ‘‘non-agriculture” towns and cities. Note that relaxing resource constraints nor-
and its future sources of growth may underpin divergent views mally increases the size of growth effects, as demonstrated by
on the sector’s importance vis-à-vis agriculture. fixed price multiplier models. However, PGEs control for the
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
BEYOND AGRICULTURE VERSUS NON-AGRICULTURE 9
NOTES
1. Own calculations using data from UNSD (2013) for 31 low-income 4. See the Appendix Tables 9–11 and Diao and Thurlow (2012) for a full
African countries for the period 2000–12. Eritrea, Somalia, and Zimbabwe specification and description of the CGE models.
are excluded due to data limitations. The definition of ‘‘low-income” is
according to 2007 classifications. 5. The SAMs and documentation can be downloaded at www.ifpri.
org/datasets.
2. The level of processing required for an agricultural product to be
reclassified as a manufactured product may vary between countries. For 6. When increasing average productivity in a sector, we are not
example, grain milling by farmers for their own consumption may be necessarily asserting that all firms/farms become more productive. We
treated as agriculture, whereas milling for urban consumers is manufac- are, however, assuming that average and marginal intermediate input use
turing. Our analysis is consistent with countries’ own classifications as is the same.
reflected in national accounts.
7. Our PGEs can also be interpreted as a percentage change in the
3. The cross-country econometric analysis by Christiaensen et al. (2011) number of poor people in 2015 since the size of the total population in
finds significant inter-sectoral spillover effects between agricultural growth 2015 does not vary across baseline and sector growth scenarios.
and poverty reduction.
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
10 WORLD DEVELOPMENT
8. Our results are less favorable for agriculture than those of 9. We also tested sensitivity to changes in factor substitution elasticities,
Christiaensen et al. (2011), who find that agricultural growth is 11 times since this is the only behavioral parameter in the models without survey-
more poverty reducing than nonagricultural growth for resource-poor based or cross-country estimates. Reducing the elasticity from 2.00 to 0.75
low-income countries in Sub-Saharan Africa (using a $1-a-day poverty in all agricultural sectors generally lowers all PGEs, but does not change
line). the rank order of sectors within countries.
REFERENCES
Adam, C., Bevan, D., & Gollin, D. (2016). Rural–urban linkages, public Hazell, P. B. R. (2013). Options for African agriculture in an era of high
investment and transport costs: The case of Tanzania, Working paper food and energy prices. Agricultural Economics, 44(Supplement),
WPS/2016-01. Oxford, UK: Center for the Study of African 19–27.
Economies, Oxford University. Hazell, P., Poulton, C., Wiggins, S., & Dorward, A. (2010). The future of
Arndt, C., Benfica, R., Tarp, F., Thurlow, J., & Uaiene, R. (2010). small farms: Trajectories and policy priorities. World Development, 38
Biofuels, growth and poverty: a computable general equilibrium (10), 1349–1361.
analysis for Mozambique. Environment and Development Economics, Khan, H. A. (1999). Sectoral growth and poverty alleviation: A multiplier
15(1), 81–105. decomposition technique applied to South Africa. World Development,
Arndt, C., Hussain, M. A., Jones, E. S., Nhate, V., Thurlow, J., & Tarp, 27(3), 521–530.
F. (2012). Explaining poverty evolution: The case of Mozambique. Ligon, E., & Sadoulet, E. (2014). Estimating the relative benefits of
American Journal of Agricultural Economics, 94(4), 854–872. agricultural growth on the distribution of expenditures. Unpublished
Besley, T., & Cord, L. J. (2007). Delivering on the promise of pro-poor manuscript.
growth: Insights and lessons from country experiences. Washington DC, Loayza, N. V., & Raddatz, C. E. (2010). The composition of growth
USA: The World Bank. matters for poverty alleviation. Journal of Development Economics, 93
Bezemer, D., & Headey, D. (2008). Agriculture, development, and urban (1), 137–151.
bias. World Development, 36(8), 1342–1364. Misselhorn, M., & Klasen, S. (2006). Determinants of the growth semi-
Breisinger, C., Thomas, M., & Thurlow, J. (2009). Social accounting elasticity of poverty reduction. In Proceedings of the German develop-
matrices and multiplier analysis: An introduction with exercises. ment economics conference, Berlin 2006.
Washington DC, USA: International Food Policy Research Institute. Ravallion, M., & Datt, G. (1996). How important to India’s poor is the
Christiaensen, L., Demery, L., & Kuhl, J. (2011). The (evolving) role of sectoral composition of economic growth?. World Bank Economic
agriculture in poverty reduction – An empirical perspective. Journal of Review, 10(1), 1–25.
Development Economics, 96(2), 239–254. Ravallion, M., & Huppi, M. (1991). Measuring changes in poverty: A
Collier, P., & Dercon, S. (2014). African agriculture in 50 years: methodological case study of Indonesia during an adjustment period.
Smallholders in a rapidly changing world?. World Development, 63 World Bank Economic Review, 5(1), 57–82.
(10), 92–101. Resnick, D., & Thurlow, J. (2016). The political economy of Zambia’s
De Janvry, A., & Sadoulet, E. (2010). Agricultural growth and poverty recovery: Structural change without transformation?. In M. McMillan,
reduction: Additional evidence. World Bank Research Observer, 25(1), D. Rodrik, & C. Sepúlveda (Eds.), Structural change, fundamentals and
1–20. growth. Washington DC, USA: International Food Policy Research
Dercon, S. (2009). Rural poverty: Old challenges in new contexts. World Institute.
Bank Research Observer, 24(1), 1–28. Robinson, S., Cattaneo, A., & El-Said, M. (2001). Updating and
Dercon, C., & Gollin, D. (2014). Agriculture in African development: estimating a social accounting matrix using cross entropy methods.
Theories and strategies. Annual Review of Resource Economics, 6, Economic Systems Research, 13(1), 47–64.
471–492. Roubaud, F., & Torelli, C. (2013). Employment, unemployment and
Dervis, K., De Melo, J., & Robinson, S. (1982). General equilibrium models working conditions in urban labor markets of Sub-Saharan Africa:
for development policy. New York, NY, USA: Cambridge University Main stylized facts. In P. de Vreyer, & F. Roubaud (Eds.), Urban labor
Press. markets in Sub-Saharan Africa. Washington DC, USA: World Bank.
Diao, X., Hazell, P., & Thurlow, J. (2010). The role of agriculture in Suryahadi, A., Suryadarma, D., & Sumarto, S. (2009). The effects of
African development. World Development, 38(10), 1375–1383. location and sectoral components of economic growth on poverty:
Diao, X., Thurlow, J., Benin, S., & Fan, S. (Eds.) (2012). Strategies and Evidence from Indonesia. Journal of Development Economics, 89(1),
priorities for African agriculture: Economywide perspectives from 109–117.
country studies. Washington DC, USA: International Food Policy Thirtle, C., Lin, C., & Piesse, J. (2003). The impact of research-led
Research Institute. agricultural productivity growth on poverty reduction in Africa, Asia
Diao, X., & Thurlow, J. (2012). A recursive dynamic computable general and Latin America. World Development, 31(12), 1959–1975.
equilibrium model. In X. Diao, J. Thurlow, S. Benin, & S. Fan (Eds.), Thorbecke, E., & Jung, H.-S. (1996). A multiplier decomposition method
Strategies and priorities for African agriculture: Economywide perspec- to analyze poverty alleviation. Journal of Development Economics, 48
tives from country studies. Washington DC, USA: International Food (2), 279–300.
Policy Research Institute. Thurlow, J., & Wobst, P. (2006). Not all growth is equally good for the
Dimaranan, B. V. (2006). Global trade, assistance, and production: The poor: The case of Zambia. Journal of African Economies, 15(4),
GTAP 6 database. Center for Global Trade Analysis, Purdue Univer- 603–625.
sity. UNSD (United Nations Statistics Division). (2013). National accounts
Dorosh, P., & Thurlow, J. (2013). Agriculture and small towns in Africa. main aggregates database. <https://unstats.un.org/unsd/snaama/Intro-
Agricultural Economics, 44(2013), 435–445. duction.asp>, last accessed December 2013.
Dorosh, P., & Thurlow, J. (2014). Can cities or towns drive African World Bank. (2013). World Development Indicators. Washington DC,
development? Economywide analysis for Ethiopia and Uganda. World USA. <http://data.worldbank.org/data-catalog/world-development-
Development, 63(10), 113–123. indicators>, last Accessed December 2013.
Grimm, M., Klasen, S., & McKay, A. (2007). Determinants of pro-poor World Bank. (2014). PovcalNet: An online poverty analysis tool. Wash-
growth: Analytical issues and findings from country cases. New York, ington DC, USA. <http://iresearch.worldbank.org/PovcalNet>, last
NY, USA: Palgrave McMillan. Accessed June 2014.
Haggblade, S., Hazell, P., & Dorosh, P. (2007). Sectoral growth linkages
between agriculture and the rural nonfarm economy. In S. Haggblade,
P. Hazell, & T. Reardon (Eds.), Transforming the rural nonfarm
economy: Opportunities and threats in the developing world. Washington
DC, USA: Johns Hopkins University Press.
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
BEYOND AGRICULTURE VERSUS NON-AGRICULTURE 11
APPENDIX A
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014
12 WORLD DEVELOPMENT
P
QT ct ¼ c0 ðcd cc0 QDc0 t þ cmcc0 QM c0 t þ cecc0 QEc0 t Þ 18
IncomesPand expenditures
YF ft ¼ Pc WF ft WDfct QF fct 19
YH ht ¼ f xhf ð1 tff Þ ð1 rff Þ YF ft þ ghh popht cpi þ whh X 20
P
PQct QH cht ¼ PQct cch þ bch ð1 shh Þ ð1 thh Þ YH ht c0 PQct0 cc0 h 21
QI ct ¼ IAt qinvc 22
Incomes and expenditures continued
QGct ¼Pgat qgovc P P 23
YGt ¼ h thh YH ht þ f tff YF ft þ c ðtmc pwmc QM ct X þ tqc PQct QQct Þ 24
Equilibrium conditions
P
qfsft ¼ Pc QF fct P 25
P ct ¼ c0 cacc QN
QQ P c0 t þ h QH cht þ QGct þ1QI ct P þ QT ct 26
0
P
c pwmP c QM ct þ f ð1P tff Þ rff YF ft X t ¼ c pwec QEct þ h whh þ cabt 27
YGt ¼ c PQct QGct þ h ghh popht cpi þ FS t 28
P P
h shh ð1 thh Þ YH ht þ FS t þ cabt X t ¼ c PQct QI ct 29
Capital accumulation and allocation
YF
ARft ¼ qfsft 30
QF fct WF ft WDfct P
ft
P
QK fct c0 PQc0 t cic0 ¼ qfs ARft c0 PQc0 t QI c0 t 31
ft
ScienceDirect
Please cite this article in press as: Dorosh, P., & Thurlow, J. Beyond Agriculture Versus Non-Agriculture: Decomposing Sectoral
Growth–Poverty Linkages in Five African Countries, World Development (2016), http://dx.doi.org/10.1016/j.worlddev.2016.08.014