Jump to ContentJump to Main Navigation
Growth and Poverty in Sub-Saharan Africa$

Channing Arndt, Andy McKay, and Finn Tarp

Print publication date: 2016

Print ISBN-13: 9780198744795

Published to Oxford Scholarship Online: May 2016

DOI: 10.1093/acprof:oso/9780198744795.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (oxford.universitypressscholarship.com). (c) Copyright Oxford University Press, 2021. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use. date: 17 June 2021

Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle

Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle

Chapter:
(p.89) 5 Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle
Source:
Growth and Poverty in Sub-Saharan Africa
Author(s):

Karl Pauw

Ulrik Beck

Richard Mussa

Publisher:
Oxford University Press
DOI:10.1093/acprof:oso/9780198744795.003.0005

Abstract and Keywords

Poverty reduction is a primary development goal in Malawi. While a variety of policy interventions and strategies have sought to reduce poverty, the most significant policy intervention in recent years has been the Farm Input Subsidy Programme (FISP). FISP provides heavily subsidized fertilizer and other farm inputs to poor smallholders. Although the programme is primarily designed to address food security through increasing maize yields, its significant burden on public resources has meant that the programme’s wider impact on growth and poverty reduction has also been closely scrutinized as it is believed to crowd out other government interventions. Malawi’s recent growth performance has been remarkable. Official national accounts data reveal strong growth across all economic sectors during 2005–11. However, the authors find that the extreme poverty rate has increased marginally, suggesting that the most vulnerable continue to be excluded from the benefits of economic policy and growth.

Keywords:   Malawi, Farm Input Subsidy Programme, poverty, economic policy, economic growth, smallholders

5.1 Introduction

Poverty reduction is a primary development goal in Malawi. While a variety of policy interventions and strategies have sought to reduce poverty, the most significant policy intervention in recent years has been the Farm Input Subsidy Programme (FISP). FISP provides heavily subsidized fertilizer and other farm inputs to between 1.5 and 2 million poor smallholders at a cost of around 3 per cent or more of annual GDP. Although the programme is primarily designed to address food security through increasing maize yields, its significant burden on public resources has meant that the programme’s wider impact on growth and poverty reduction has also been closely scrutinized as it is believed to crowd out other government interventions.

Malawi’s recent growth performance has been remarkable. Official national accounts data reveal strong growth across all Malawi’s economic sectors during 2005–11, with per capita GDP growth averaging 3.5 per cent per annum (NSO 2012b). Growth was particularly robust in the agricultural sector, which government attributes to FISP (GoM 2012). Since a large majority of the poor are rural smallholder farmers, and since about half of these farmers were explicitly targeted by the subsidy programme, the expectation was always that Malawi’s growth trajectory during this period would be highly pro-poor, even if poverty reduction was not necessarily a primary goal of the subsidy programme.

Prior to the release of official poverty estimates in 2012, a heated debate had already developed over the desirability, sustainability, and overall impact of (p.90) input subsidy programmes in Malawi and elsewhere in the continent (see, for example, Chirwa and Dorward 2013; Jayne et al. 2013; Lunduka et al. 2013; Arndt et al. 2015). Although initially praised globally as a bold and resoundingly successful initiative, emerging survey evidence was beginning to cast doubt on the ability of FISP to raise maize yields enough for direct benefits to outweigh programme costs, while official crop production estimates which underlie GDP estimates were also being questioned (see Pauw and Thurlow 2014).

The FISP ‘success story’ unravelled further when official poverty estimates suggested that purported economic growth had in fact not been accompanied by rapid poverty reduction as expected: using the latest two rounds of Integrated Household Surveys (IHSs), the National Statistical Office (NSO 2012a) reported that the poverty headcount rate declined only marginally from 52.4 per cent in 2004/5 (IHS2) to 50.7 per cent in 2010/11 (IHS3). Even more disconcerting was the fact that rural poverty had apparently increased marginally over the period despite the introduction and scaling up of FISP during this time.

This growth–poverty puzzle justifies further investigation. One possibility is that economic growth, and particularly agricultural growth, is overstated in national accounts; another is that the structure of growth was such that it did not benefit those close enough to the poverty line to be lifted out of poverty. However, it is equally possible that poverty estimates understate the reduction in poverty. We focus on the latter and undertake our own consumption-based poverty analysis using the same datasets as in the official poverty analysis conducted by Malawi’s National Statistical Office (NSO).

Several methodological advances distinguish our approach from that of the NSO. First, we allow the consumption bundles from which poverty lines are derived to change over time as consumers near the poverty line adjust their spending patterns in response to relative price changes. Second, rather than relying on a national poverty line only, we construct regional poverty lines that account for variations in consumption preferences, prices, and non-food consumption shares. In doing so we return to the tradition of using regional poverty lines in Malawi in the late 1990s (see Mukherjee and Benson 2003).

Third, we adopt a slightly revised version of Ravallion’s (1998) iterative procedure to devise consumption bundles and poverty lines that more closely represent actual consumption preferences of the poor, as well as an entropy-based method by Arndt and Simler (2010) that ensures that poverty bundles are utility-consistent. Finally, we use a revised set of conversion factors (Verduzco-Gallo et al. 2014) for converting non-standard food consumption into standard units. This may influence unit costs of foods and hence the estimate of household consumption spending or the poverty line.

(p.91) In contrast to the official poverty estimate, our analysis reveals large declines in regional and national poverty in Malawi between 2004/5 and 2010/11. Poverty declined by 7.5 and 10.3 percentage points in rural and urban areas respectively, which is more consistent with the economic growth trajectory reported in the national accounts as well as reported improvements in several non-monetary dimensions of welfare. Although weaknesses in Malawi’s economic statistics compel us to remain cautious in our conclusions, and while we do not explicitly test causality, our results ultimately corroborate a much more positive narrative about FISP and its impact on growth and poverty reduction compared to the more pessimistic view held by many.

The chapter is structured as follows. Section 5.2 further explores the official evidence on economic growth and poverty trends in Malawi. Section 5.3 describes, in simplistic terms, the method for constructing utility-consistent regional poverty lines for Malawi and contrasts this with the method used by the NSO. Section 5.4 presents our consumption-based poverty results as well as trends in non-monetary poverty; and finally, section 5.5 concludes and highlights areas for further research.

5.2 Recent Evidence of Growth and Poverty Trends in Malawi

Rain-fed smallholder maize production accounts for around one quarter of agricultural GDP in Malawi. The agricultural sector, in turn, accounts for 30 per cent of the economy (Benin et al. 2012). This high dependence on a single, large agricultural subsector is a major reason for Malawi’s highly erratic economic growth path. Since around 2006, however, favourable weather combined with the introduction of the Farm Input Subsidy Programme (FISP) has contributed to robust and more stable agricultural growth led by rapidly rising maize yields (see Figure 5.1). The initial success of Malawi’s FISP created renewed interest globally in fertilizer subsidies as a tool for promoting growth while at the same time ensuring food security and reducing poverty.

Increased yields achieved under FISP led to a significant increase in maize production. Although FISP was also associated with relative declines in maize prices, these were found to be fairly small (Ricker-Gilbert 2014), and as a result agricultural GDP still grew strongly. The first three columns of Table 5.1 report average sector-level and national GDP growth rates for different sub-periods (i.e. 2005–7, 2005–11, and 2007–11). Agricultural growth was particularly rapid during the first years of FISP (2005–7) but then slowed down as the potential for further yield gains through input subsidies was exhausted. The years 2010 and 2011 were exceptionally troublesome years for the important tobacco sector, which faced large declines in international prices and domestic sales volumes. However, despite these setbacks, the average agricultural (p.92)

Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle

Figure 5.1. Maize yields by region: Smallholder summer harvest, 2000/1–11/12

Source: Authors’ calculations using data from MoAFS (2013)

Table 5.1. Sectoral GDP growth rates and contributions to change in GDP, 2005–11

Average growth rates (%)

Value added share 2005–11 (%)

Contribution to change (%)

2005–7

2007–11

2005–11

2005–7

2007–11

2005–11

Agriculture

15.9

7.3

10.1

30.0

54.7

26.4

34.2

Mining, industry, and construction

3.0

11.4

8.5

16.7

6.3

22.7

18.2

Trade and transport

5.5

6.1

5.9

21.2

15.3

15.2

15.2

Private services

7.6

11.7

10.3

18.8

16.6

25.8

23.3

Government services

4.1

6.4

5.6

13.3

7.2

9.9

9.2

National GDP

6.2

7.5

7.1

100.0

100.0

100.0

100.0

Source: Authors’ calculations based on data from NSO (2012b)

growth rate over the whole FISP period remained above 10 per cent per annum. Malawi’s growth path can be described ‘broad-based’, with strong growth also in non-agricultural sectors. The final columns show sector-level value added shares and average contributions to national GDP. As expected, given its size and high rate of growth, agriculture remained a major driver of growth during the FISP period, with a contribution of 34.2 per cent to overall GDP expansion.

These results paint a very positive story about Malawi’s recent economic performance and the role of agriculture. Growth far outstripped population growth of around 3.3 per cent per annum. Moreover, growth was driven to a large extent by smallholder maize production increases, a crop cultivated by the vast majority of farmers. Since around 90 per cent of Malawians live in (p.93)

Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle

Figure 5.2. Official poverty headcount rates, 1997/8–2010/11

Source: Authors’ calculations based on NSO (2005, 2010, 2012a)

predominantly poor, rural farm households (Benin et al. 2012), one would very legitimately expect that the level and structure of growth would have had a significant impact on rural poverty.

The most recently released poverty trends in Malawi are based on the comparison of per capita consumption welfare aggregates from two rounds of household surveys (i.e. the IHS2 in 2004/5 and the IHS3 in 2010/11). During the inter-survey period, the NSO also published a series of poverty estimates based on imputed consumption from several rounds of Welfare Monitoring Surveys (WMSs). The imputations, which were necessary because the WMSs did not include a detailed expenditure module, were done on the basis of an econometric model fitted to the IHS2 data, with right-hand side variables including household demographics, health, education, employment, housing conditions, and amenities that are included in both the IHS2 and WMSs. The poverty trend from these WMSs showed a sharp decline in poverty to around 39 per cent in 2009 (NSO 2010) (see Figure 5.2), which at the time created the expectation that the imminent poverty rates from the IHS3 would be of a similar magnitude.

The WMS poverty trend also seemed entirely consistent with national accounts data available at the time. For example, using a general equilibrium model for Malawi, Pauw et al. (2011) replicate the reported agricultural production and economic growth trends for 2005–10. The model yields a poverty rate of about 40 per cent by 2010, which is remarkably close to the estimate in the WMS 2009. However, contrary to expectations, the official poverty figures for 2010/11, which were finally released in 2012 after much delay, were a major disappointment. Although urban poverty declined sharply from 25.4 to (p.94) 17.3 per cent, the biggest concern was the increase in rural poverty from 55.9 to 56.6 per cent. National poverty declined only marginally from 52.4 to 50.7 per cent over the period.

One possible explanation for this unexpected result is that the production effect of FISP and hence estimates of economic growth may have been vastly overstated (see Jayne et al. 2008; Chirwa and Dorward 2013). Another is that growth did not trickle down to the rural poor, which is unlikely given overwhelming evidence that growth is generally associated with declines in poverty (Dollar et al. 2013), and that agricultural growth in particular is strongly pro-poor (Diao et al. 2010). However, the rising rural–urban income divide and an overall increase in inequality as measured by the Gini coefficient (i.e. from 0.39 to 0.45) (NSO 2012a) does seem to suggest that growth may have disproportionately benefitted wealthier sections of the population. Yet another possibility is that the decline in poverty has been understated in the official estimates. The remainder of this chapter focuses on this third possible explanation for the growth–poverty puzzle in Malawi; particularly, we explore how the NSO estimated their poverty rates and provide alternative estimates of our own that are based on a method that introduces some advances over the one used by the NSO.

5.3 Constructing Regional Poverty Lines for Malawi

In this section we briefly, and in largely non-technical terms, explain the method adopted to estimate regional poverty lines for Malawi. As far as possible, we compare this against the NSO method to provide a simplistic explanation of possible sources of difference between our estimates and those of the NSO. For a more comprehensive and technical discussion please refer to Pauw et al. (2014).

5.3.1 Poverty and Prices

In its official assessment of poverty in 2004/5 the NSO followed a cost of basic needs approach to estimate national food and non-food poverty lines. However, in coming up with a new set of poverty lines for 2010/11, they opted not to follow this approach, arguing that the standard of living implied by any newly estimated consumption bundle would not necessarily be utility-consistent with the IHS2 bundle (NSO 2012a). This may be true, but as we explain in this section there are approaches in the literature that allow us to overcome this problem.

NSO’s approach was to estimate an inflation rate they believed to be representative of price changes faced by the poor. The 2004/5 poverty line(s) were (p.95) then multiplied by the inflation rate to arrive at the new lines for 2010/11. Rather than using the implicit prices contained in the two household surveys to estimate inflation, the NSO opted to derive an inflation estimate on the basis of a revised consumer price index (CPI). The official CPI traditionally used to monitor rural and urban inflation was thought to understate the true extent of price increases over the analysis period, and hence this revision was carried out with technical assistance from South Africa’s statistics agency and the World Bank. The original and revised CPI series are shown in Table 5.2.

One particular oddity about the NSO’s approach is that although the revised CPI suggests significantly higher inflation for food products compared to non-food products, both the food poverty line (representing ‘ultra-poverty’) and the overall poverty line (food plus non-food) were adjusted by the same ‘poverty line inflation rate’ (i.e. by 128.9 per cent). Their own data also show large differences in rural and urban food and non-food inflation rates, but in the interest of maintaining a single national poverty line this difference was disregarded.

The last section of Table 5.2 shows a ‘true’ estimate of inflation based on the revised price indexes for urban and rural food and non-food items. Essentially, in doing this exercise, we assume that the food and non-food indexes as reported, and hence their reported inflation rates, are correct; but we then use actual food/non-food consumption shares of relatively poor households—defined for simplicity as those below the median per capita consumption

Table 5.2. Official and revised CPI and inflation estimates

National

Urban

Rural

All items

Food

Non-food

All items

Food

Non-food

All items

Food

Non-food

(a) Official estimates

CPI 2004/5

178.0

161.1

201.3

192.1

183.9

196.6

171.8

155.9

205.4

CPI 2010/11

315.6

276.3

370.0

379.7

425.9

354.7

287.8

242.7

383.5

Inflation (%)

77.3

71.5

83.8

97.6

131.6

80.4

67.5

55.6

86.7

(b) Revised estimates

CPI 2004/5

212.7

220.5

201.9

199.5

202.5

197.9

218.4

224.5

205.3

CPI 2010/11

488.1

543.8

411.0

457.9

549.6

408.1

501.4

542.5

413.7

Inflation (%)

128.9

146.7

103.6

128.9

171.4

106.2

128.9

141.7

101.5

(c) ‘True’ inflation

CPI 2004/5

213.4

220.5

201.9

200.7

202.5

197.9

217.2

224.5

205.3

CPI 2010/11

499.0

543.8

411.0

494.8

549.6

408.1

499.5

542.5

413.7

Inflation (%)

133.8

146.7

103.6

146.5

171.4

106.2

130.0

141.7

101.5

Note: CPI 2004/5 represents the average index value for the period March 2004–March 2005, which is the period during which the IHS2 was conducted. Similarly, CPI 2010/11 represents the average index value for March 2010–March 2011. The ‘true’ inflation rates are obtained by averaging food and non-food inflation rates using actual consumption shares of relatively poor households.

Source: Official CPI estimates from NSO website (<http://www.nsomalawi.mw/>); revised CPI estimates obtained from NSO as supplementary material to IHS3 report (NSO 2012a)

(p.96) level—to construct new weighted average indexes for ‘all items’. This is done separately for rural and urban as well as at national level. As can be seen, these index values, and hence their associated inflation estimates, are quite different from the 128.9 per cent estimate of NSO. We are only able to (roughly) replicate the NSO’s national ‘all item’ index values of 212.7 in 2004/5 and 488.1 in 2010/11 and hence their national inflation rate of 128.9 per cent using the 2004/05 consumption weights for all households (i.e. poor and non-poor) in both years 2004/05 and 2010/11. Our contention is that the NSO’s inflation rate is not necessarily representative of the inflation faced by the poor and fails to account for differences in food and non-food inflation in urban and rural areas.

5.3.2 Regions, Preferences, and Utility Consistency

Although official poverty estimates have always been reported at a regional level in Malawi, region-specific poverty lines were last constructed using the first Integrated Household Survey (IHS1) of 1997/8 (Mukherjee and Benson 2003; NSO 2010). Following the IHS1 approach we include four regions of Malawi: rural areas are divided along the three administrative regions, namely ‘south rural’, ‘centre rural’, ‘north rural’, while a fourth region comprises all major urban centres, including Mzuzu in the north, Lilongwe in the centre, and Blantyre and Zomba in the south. The rationale for having a regional disaggregation is to capture regional differences in consumption preferences, relative prices of commodities or their changes over time (i.e. regional inflation rates), and demographic composition which affects calorie needs. Taking account of regional and temporal differences in prices and preferences tends to have a significant impact on estimated poverty lines (Tarp et al. 2002).

As is common in the literature we use consumption expenditure rather than income in our poverty estimation, mainly because consumption expenditure provides a smoother, less lumpy measure of welfare through time. The consumption aggregate used in the official poverty assessment is publicly available and consists of food and non-food components, with the latter consisting of expenditure on non-durable goods, estimated use value of durable consumer goods, and the rental value of housing. Whereas we adopt the NSO non-food consumption measure for our own analysis, we construct a new food consumption component.

Total quantity of food consumed in a household is the sum of purchased food, own production, and gifts. Food consumption is based on a seven-day recall period. In order to carry out meaningful analysis, quantity units of measurement—these include standard (metric) and non-standard units (e.g. plates, cups, bags, or pails)—need to be converted into grams using conversion factors typically supplied with household expenditure survey data. Conversion (p.97) factors take into account the volume of the measurement unit but also the weight density of the particular food product. The quality of conversion factors is crucial for determining unit costs of food consumption baskets.

We use versions of the official conversion factors for IHS2 and IHS3 that were systematically checked for errors and cleaned by Verduzco-Gallo et al. (2014) to construct both our consumption aggregate and estimated poverty lines. This explains at least some of the differences between our consumption aggregate and the one used in the official poverty assessment. The differences in conversion factors also affect poverty lines through the composition and cost of the poverty line food bundle, although since we use regional poverty lines as opposed to a national one, regional differences in prices and consumption patterns also influence the final results.

Since respondents provide estimates of both the cost and quantity of purchased food, this data can be used to estimate unit values for different food items. The valuation is carried out in the same way as the NSO describe their valuation. If a household reports to have consumed a food item not purchased in the last week (e.g. gifts, own production, or food purchased earlier), the median unit value from its cluster is used to value that consumption. If no other household in that cluster consumed the same item, or if there were not enough observations to obtain a reliable unit value, the median unit value from the next geographical level within which the household resides was used to estimate the value of that consumption. Total food expenditure for each household is calculated by multiplying the unit values by the quantity consumed. In principle, the use of revised conversion factors should be the only factor that may cause our food consumption estimate to differ from the official one.

Once a consumption aggregate has been estimated, it is necessary to construct a poverty line, or in our case, several regional poverty lines. The total poverty line for each region is the sum of the food and non-food poverty lines. The poverty line and its subcomponents reflect value judgements about basic food and non-food needs, and are set in terms of a level of per capita consumption expenditure that is deemed consistent with meeting those basic needs at prevailing prices. The level at which the food poverty line is set is crucial to our understanding of poverty since poor households allocate such a significant proportion of their spending to food. The non-food poverty line, on the other hand, recognizes that the poor also allocate a non-trivial proportion of their total consumption to non-food items.

For each of the four regions a food poverty line is constructed by determining the food energy (caloric) availability for the reference population (i.e. the poor), the caloric content of the typical diet of the poor in that region, and the average cost (at local prices) of calories when consuming that diet. Thus, the food poverty line, expressed in this instance in Malawi Kwacha (MWK) per (p.98) person per day, is the area-specific cost of meeting the minimum caloric requirements when consuming a food bundle comprised of goods that the poor in a particular area typically consume. Essentially, the caloric requirement is a way of anchoring the estimated poverty line at a specific welfare level. We therefore opt to use the same caloric requirement as the NSO, namely 2,400 kilocalories per person per day (NSO 2005, 2012a).

The first step in our poverty line estimation is to establish reasonable cost estimates of commonly consumed food commodities. Following Arndt and Simler (2010), we assume the most common food items consumed by the poor to be those that account for 90 per cent of their food expenditure. On average, we find that these bundles represent about 95 per cent of caloric availability. The values of these area-specific food bundles are then scaled to equal 100 per cent of calorie requirements. Each household’s food consumption bundle therefore essentially excludes less common (and often more expensive) sources of calories. A set of household-specific prices is then calculated (i.e. amount spent divided by quantity in grams). The median unit values of these price distributions will eventually be used in valuing regional bundles.

In order to ensure that our poverty lines reflect preferences and prices faced by poor people, we adopt and modify an iterative process described by Ravallion (1998). In the original procedure, households are first ranked by consumption per capita, and then the bottom x1 = 60 per cent of households is at first arbitrarily identified as the relatively poor. This means x1 may be regarded as a preliminary estimate of the poverty headcount rate. These ‘poor’ households’ food quantities and prices are then evaluated to estimate the cost of the calories they obtain. Preliminary poverty line calculations are made using the minimum caloric requirement as the basis, and the nominal consumption values are converted into real terms by taking into account region-specific differences in the cost of acquiring the basic needs bundle. This gives a preliminary poverty headcount ratio x2. Households are then re-ranked using this first approximation of consumption per capita in real terms, and the bottom x2 per cent of this ranking identified as the relatively poor. Observed consumption patterns and prices in this subsample are calculated, producing a second estimate of poverty lines. This gives another preliminary poverty headcount ratio x3. Again households are ranked by their real expenditure, and the iterative process continues until the poverty line converges, meaning that the same, or nearly the same, subsample of households are identified as the poor.

Applying this procedure to Malawi yielded a very low poverty line and hence a very low poverty rate. Therefore, we adopt a slightly modified version of the iterative procedure in that the poorest 60 per cent of the population was used in each iteration. Between each iteration the consumption aggregate was deflated by the estimated poverty lines only. Since poverty lines were (p.99) estimated at the regional level, this procedure allowed the final consumption baskets to be calculated based on the consumption of the 60 per cent poorest in terms of real consumption. This is, however, still a deviation from the approach taken by the NSO where the poverty line was based on the fifth and sixth consumption deciles.

The above procedure should reduce poverty rate bias; however, there is no guarantee that the estimated poverty lines are consistent across time and space. Arndt and Simler (2010) found that in both Mozambique and Egypt poverty lines based on such traditional iterative methods produced biased poverty estimates. This study therefore adopts the methodological advancement first proposed by Arndt and Simler (2010) to ensure that poverty lines are utility-consistent. The method is not described in detail here, but essentially entails use of an entropy-based approach that makes small adjustments to budget shares until imposed constraints—in this instance, utility consistency—are satisfied. The method ensures that the information content in the original budget shares is preserved to the greatest degree possible. Spatial utility consistency is ensured if the regional bundles satisfy revealed preference conditions. This means that the 2010/11 bundle of domain A is not manifestly of higher quality than the 2010/11 bundle of domain B (and vice versa); similarly, the 2004/5 bundle of domain A is not manifestly of higher quality than the 2010/11 bundle of domain A (and vice versa).

The next step is to estimate a non-food poverty line. Similar to the approach adopted by the NSO, this is taken as a weighted average of non-food expenditure for people with food expenditure at between 80 and 120 per cent of the food poverty line. A triangular weighting scheme is used where the closer a household’s food expenditure is to the poverty line, the higher the weights. The sum of the non-food and food poverty lines is the total poverty line, which is the basis for calculating initial poverty rates.

Having derived the poverty lines, the next step is to estimate poverty rates. Since expenditure data are collected at the household level, we assume a uniform intra-household distribution such that each household member has the same per capita expenditure level. In keeping with the popular poverty literature, we use poverty measures proposed by Foster et al. (1984):

P(z,α)=1Ni=1N(z-yiz)αI(yi<z)
(5.1)

In equation (5.1), yi is per capita consumption expenditure of person i drawn from a sample of size N, z is a poverty line, α‎ is a measure of poverty aversion, and I(.) is an indicator function equal to one if the condition yi < z holds, and zero otherwise. The parameter α‎ takes on values of 0, 1, or 2 to measure the poverty headcount rate, poverty gap, and squared poverty gap respectively. (p.100) Specifically, when α‎ = 0, we have the poverty headcount index. This gives the percentage of the population who are, based on consumption, poor. The headcount is easy to interpret; however, it has some limitations. It neither takes into account how close or far the consumption levels of the poor are to the poverty line, nor the distribution among the poor. When α‎ = 1, we have the poverty gap index. It measures the extent of the difference between the poverty line and the average consumption of poor households. This measure captures changes in poverty that the poverty headcount index does not detect. For instance, if the consumption of the poor increases without necessarily crossing the poverty line, the headcount will not detect this change, while the poverty gap will. For α‎ = 2, we have the squared poverty gap index. It measures the severity of poverty, and also takes into account inequality changes among the poor. A transfer from a poor person to somebody less poor may leave unaffected the headcount or the poverty gap but will increase this measure. Our analysis focuses on the poverty headcount rate (α‎ = 0).

5.4 Results and Analysis

5.4.1 Monetary Poverty Analysis

In this section we present our estimates of changes in monetary (or consumption-based) poverty in Malawi between 2004/5 and 2010/11. Throughout we compare our estimates with the official poverty estimates of the NSO. Differences between our poverty results and those of the NSO reflect differences in poverty lines as well as differences in the distribution of the consumption aggregates used in the respective poverty analyses.

Our estimated poverty lines are reported in Table 5.3 and compared with the official food, non-food, and national poverty lines used by the NSO. All poverty lines are converted to Malawi Kwacha (MWK) per person per day. In order to facilitate the comparison, we present, in addition to our regional poverty lines, population-weighted national average poverty lines, which can be compared against the NSO’s national poverty lines. We also estimate an average rural poverty line in the same manner. Given the approach adopted here, our poverty lines may vary by region, while, more importantly, the implied inflation rate faced by the poor—we refer to it here as the ‘poverty line inflation rate’—may also vary by region and for food or non-food items.

While both food and non-food poverty lines of the urban and rural northern regions as well as the non-food poverty line for the central rural region are above the official estimates, the food poverty lines of rural south and rural central as well as the non-food poverty line of rural south are below the official estimates. We estimate a national poverty line of MWK 43.2 per person per day, which is about 2.4 per cent below the official estimate of MWK 44.3. In (p.101)

Table 5.3. Food, non-food, and overall poverty lines for 2004/5 and 2010/11

Poverty lines (Malawi Kwacha/person/day)

Poverty line inflation rates

IHS2 (2004/5)

IHS3 (2010/11)

Food

Non-food

Total

Food

Non-food

Total

Food

Non-food

Total

National

26.0

17.2

43.2

59.4

33.1

92.5

128.2

92.5

114.0

Urban

30.3

26.7

57.0

65.4

62.1

127.5

116.1

132.5

123.8

Rural

25.5

16.0

41.5

58.3

28.0

86.2

128.8

74.8

108.0

North rural

29.7

16.3

46.1

64.7

31.0

95.4

117.5

89.5

108.0

Centre rural

26.1

17.6

43.8

59.8

28.7

88.4

128.8

62.6

102.1

South rural

23.8

14.4

38.2

55.0

26.4

81.3

131.0

83.5

113.1

NSO poverty line (national)

27.5

16.8

44.3

62.9

38.5

101.4

128.9

128.9

128.9

Note: Poverty lines are estimated for each of the four regions. Estimates for ‘national’ and ‘rural’ poverty lines are population-weighted averages of the region-specific poverty lines.

Source: NSO (2012a) and authors’ estimates based on IHS2 and IHS3

2010/11 all food and non-food poverty lines of the rural regions are lower than the official lines, except the food poverty line in the rural north, which is slightly higher. As in 2004/5 the urban food poverty line and in particular the non-food poverty line are above the official lines. The overall poverty line of MWK 92.5 per person per day is about 8.8 per cent lower than the official poverty line of MWK 101.4. Our estimated national poverty line inflation rate is 114.0 per cent, which is somewhat lower than the 128.9 per cent estimated by the NSO. We therefore concur that the official inflation estimate of 77.3 per cent (see Table 5.2) understates actual inflation faced by the poor, but at the same time inflation faced by the poor may not have been as high as 128.9 per cent.

Importantly, there is some regional variation in our inflation estimates. Consistent with the ‘true’ inflation estimates shown in Table 5.2, our urban poverty line inflation rate (123.8 per cent) is significantly above the rural poverty line inflation rate (108.0 per cent). While there is some variation among rural areas, there is a much more pronounced difference between the urban and rural inflation rates, particularly as far as non-food inflation rates are concerned: whereas urban non-food inflation was 132.5 per cent, the rural inflation rate is only 74.8 per cent. This may reflect significant welfare effects associated with the subsidy programme: the commercial value of the full subsidy package of two bags of fertilizer, distributed to half of rural farm households, is equivalent to over 20 per cent of our estimated non-food rural poverty line (see Arndt et al. 2015). Another likely factor is a sharp reduction in clothing and footwear prices linked to a dramatic rise in sub-Saharan Africa’s importation of inexpensive used clothing from developed countries (see Baden and Barber 2005). The NSO’s revised CPI data shows that (p.102) clothing and footwear prices increased only about half as much as the overall consumption basket, allowing households to drastically reduce spending—the CPI weight for clothing and footwear items have recently been adjusted downwards from 20.3 to 6.4 per cent of non-food expenditure—yet there is an equally sharp rise in the proportion of households that feel their clothing is ‘adequate’ (see Table 5.5 and related discussion in section 5.4.2).

Perhaps the most important methodological difference between our approach and that of the NSO is that our inflation rate is based on flexible consumption bundles. This means that we account for observed changes in the non-food share of expenditure, whereas the NSO fix the non-food expenditure share in both survey years at 38 per cent. Since the non-food poverty line is estimated on the basis of non-food expenditures of households close to the food poverty line, different choices of food poverty lines will mean that non-food poverty lines are estimated on the basis of the observed expenditures of different subsets of the population; hence, the level and share of non-food expenditures may be sensitive to the choice of poverty line.

We find several interesting results in this regard. Firstly, if Engel’s Law holds, the estimates of non-food expenditure shares would rise as we move to higher poverty lines, simply because given the estimation procedure we would then be evaluating non-food expenditures of slightly wealthier households. It appears this only holds for urban households in 2004/5. In all other instances the non-food share declines or is constant as we evaluate non-food expenditure at higher notional food poverty lines, which suggests extra income earned by the poor is initially spent on more (or better-quality) food rather than non-food expenditures. We can reasonably assume that many households with food expenditure at or below the food poverty line are not satisfying their caloric needs; for example Verduzco-Gallo et al. (2014) estimate a calorie deficiency rate of 40 per cent in 2004/5. Hence, it is quite plausible to think that additional income will be spent on food until they reach a desired level and quality of food intake. Non-food shares therefore only start to rise at relatively high food poverty lines. The relatively low rural non-food inflation rate also means that poor households are able to divert household income normally reserved for essential non-food items to food spending because of significant savings on non-food items.

Secondly, while NSO assumed a constant non-food expenditure share of 38 per cent, we find this rate to be reasonably close to our own non-food shares in 2004/5 only in central and southern rural areas. The rural north has a lower non-food share, while in urban areas the share is well above 38 per cent. This is consistent with the literature where urban households are often found to consume fewer and more expensive calories (Tarp et al. 2002). This finding should therefore also be reflected in the estimated poverty line for the urban areas. In summation, the regional and time-specific approach to poverty line (p.103) estimation appears to be important in the present setting: consumption patterns, even those represented by crude non-food shares, differ substantially across regions.

Table 5.4 reports our estimated poverty headcounts and compares them with official numbers. We calculate and report 95 per cent confidence intervals to accompany the percentage point changes in poverty rates between 2004/5 and 2010/11. If the confidence interval value exceeds the absolute value of the percentage point change in poverty, the change is considered statistically insignificant.

Whereas we estimate a slightly lower rural poverty rate in 2004/5 (48.2 per cent) to that of the NSO (55.9 per cent), our urban poverty rate (37.6 per cent) is substantially higher than the official rate (25.4 per cent). Since our estimates of rural poverty are at a somewhat lower level than those of the NSO for both surveys and our estimates of urban poverty are somewhat higher, we will focus our discussion on the changes in poverty over time rather than their absolute levels. By 2010/11 our rural poverty rate (40.6 per cent) is substantially lower than in 2004/5 (48.2 per cent), i.e. we estimated a significant 7.5 percentage point decline in rural poverty compared to the NSO’s estimate of a small 0.7 percentage point increase. We also report substantial declines across all rural

Table 5.4. Poverty headcount rates and changes in poverty between 2004/5 and 2010/11

Poverty headcount rate (%)

Percentage point change (2004/5 to 2010/11) and 95% confidence intervals

Change in poverty (fixed non-food share)

2004/5 (IHS2)

2010/11 (IHS3)

Regional poverty estimates

NSO poverty estimate

Regional poverty estimates

NSO poverty estimate

Change in poverty

NSO change in poverty

Normal poverty

National

47.0

52.4

38.6

50.7

−8.4

± 2.8

−1.7

± 2.4

−4.1

Urban

37.6

25.4

27.3

17.3

−10.3

± 9.4

−8.1

± 6.8

−11.6

Rural

48.2

55.9

40.6

56.6

−7.5

± 2.9

0.7

± 1.4

−2.3

North rural

59.4

56.3

48.0

59.9

−11.4

± 6.8

3.6

± 6.5

−8.8

Centre rural

40.0

46.7

33.7

48.7

−6.3

± 4.4

2.0

± 4.2

0.6

South rural

53.1

64.4

45.1

63.3

−8.0

± 4.5

−1.1

± 3.7

−3.5

Extreme poverty

National

17.1

22.3

17.9

24.5

0.8

± 2.0

2.1

± 2.2

Urban

9.0

7.5

4.7

4.3

−4.2

± 3.6

−3.2

± 3.4

Rural

18.1

24.2

20.2

28.1

2.1

± 2.2

3.9

± 2.4

North rural

30.1

25.9

25.6

29.0

−4.5

± 6.7

3.0

± 6.4

Centre rural

11.8

16.1

16.1

21.5

4.3

± 2.8

5.4

± 3.2

South rural

21.1

31.5

22.6

34.2

1.6

± 3.8

2.7

± 4.0

Note: The confidence interval is used to determine the statistical significance of the difference in the poverty rate between 2004/5 and 2010/11. Since the distribution of the poverty rate is unknown we follow Simler and Arndt (2007) in defining the confidence interval as plus or minus twice the standard error. The change in the poverty headcount rate is considered statistically insignificant if zero is within the confidence interval around the estimate.

Source: NSO (2012a) and authors’ estimates based on IHS2 and IHS3

(p.104) regions, which stands in contrast to the NSO’s estimates of increases in poverty in the rural north and centre. On the other hand, consistent with NSO’s estimate, we find a substantial decrease in urban poverty, i.e. by 10.3 percentage points compared to the 7.4 percentage points estimated by the NSO. Overall, we estimate a statistically significant 8.4 percentage point decline in national poverty, which stands in sharp contrast to the NSO’s statistically insignificant decline of 1.7 percentage points.

Table 5.4 also presents results for changes in extreme poverty. Households are deemed extremely poor or ‘ultra-poor’ when their total per capita consumption falls below the food poverty line. Consistent in direction with the NSO estimates, we find that urban extreme poverty declines by 4.2 percentage points, while rural extreme poverty increases by 2.1 percentage points. At the national level we estimate a small and insignificant increase in extreme poverty (0.8 percentage points) compared to the NSO’s increase of 2.1 percentage points, which is also statistically insignificant. It appears Malawi’s economic policies, while benefitting those close to the poverty line, may have failed to benefit the rural ultra-poor.

At least one important methodological question raised earlier is whether an assumption of fixed non-food consumption shares is reasonable. To investigate how such an assumption would affect our poverty estimates, we estimate regional food poverty lines as before, but then assume that essential non-food spending in 2010/11 is a fixed share of food spending, with the share determined by the 2004/5 non-food shares. The final column of Table 5.4 reports the resulting change in poverty. The results now show a slightly higher reduction in urban poverty (i.e. by 11.6 percentage points), but a much smaller decline in rural poverty (i.e. by 2.3 percentage points). Poverty estimates are therefore sensitive to the choice of flexible or fixed non-food shares, and our estimates are closer to that of the NSO when we fix the non-food budget share.

Figure 5.3 provides a visual picture of how the cumulative distribution functions of our consumption aggregates compare with the official welfare aggregates. The solid lines represent our newly calculated consumption aggregates, while the dashed line represents that of the NSO. Poverty lines (logarithmic values) are indicated by vertical lines, which are also dashed or solid to match the cumulative distribution function. The point at which a poverty line intersects the relevant cumulative distribution function indicates the associated poverty level (vertical axis). These values are comparable with those shown in Table 5.4.

While our consumption aggregates for urban households are similar to those of the NSO in both 2004/5 and 2010/11, our urban poverty lines are much higher, and hence our estimated poverty rates in urban areas are higher. The 2004/5 panel for rural areas shows that our cumulative distribution (p.105)

Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle

Figure 5.3. Consumption distribution functions and poverty estimates

Source: Authors’ estimates based on IHS2 and IHS3

function lies slightly to the right of the NSO’s function, which implies that our consumption estimate is higher over most of the distribution; however, our poverty line is slightly lower, which explains why we get a lower estimate of rural poverty compared to that of the NSO. In 2010/11, our rural poverty line is again lower than that of the NSO, but our consumption estimate is once again to the right of that of the NSO, thus yielding a much lower poverty estimate than the NSO. These visual inspections lead us to conclude, with some circumspection, that differences in our poverty line estimates are the most important factor in explaining differences in poverty rates and their changes over time between the respective studies.

5.4.2 Non-monetary Poverty Analysis

The preceding discussion has focused on estimates of monetary poverty whereby consumption expenditure is used as an indicator of welfare. The rationale for a consumption-based approach to identifying the poor is that there is a strong correlation between ‘means’ (i.e. income or consumption levels) and ‘ends’ such as adequate levels of health, education, or freedom. An individual above the monetary poverty line is thought to possess sufficient purchasing power to acquire the bundle of attributes yielding a level of (p.106) well-being sufficient to function in society, thus providing a rationale for a consumption-based approach to poverty analysis.

However, given the often weak relationship between income and welfare—this may be due to incomplete markets, presence of externalities, or provision of public goods—money-metric measures are not always good indicators of welfare. Besides, there is no guarantee that households with incomes at or above a poverty threshold would use their incomes to purchase those ‘basic needs’ considered necessary for achieving an acceptable level of well-being. Decision-makers in the household may, for example, instead choose to satisfy wants for, say, alcohol and tobacco at the expense of satisfying the minimum caloric requirements of their children. In the money-metric approach such households would be classified as non-poor when in reality at least some of their members are deprived of some basic needs (Thorbecke 2005).

Consequently, Sen (1985) argues that the measurement of poverty should, where possible, go beyond income or consumption and look at other dimensions of well-being such as health, education, empowerment, and freedom of association, among others. Income and consumption expenditure are instrumentally important as a means of achieving the other dimensions of well-being, but the other dimensions of well-being are in and of themselves intrinsically significant. Thus, these dimensions are equally important and deserve recognition and measurement in their own right. Moreover, trends in non-monetary dimensions of well-being can be used to validate or challenge observed trends in monetary poverty.

This section looks at levels and trends of non-monetary dimensions of well-being in Malawi between 2004/5 and 2010/11. Although we omit some results due to space considerations—a full set of results can be found in Pauw et al. (2014)—we find overwhelming evidence of improvements in non-monetary welfare measures. Specifically, we note significant improvements in housing quality, access to clean water, and ownership of consumer durables, which are consistent with the finding of rising incomes and declining poverty. School enrolment has increased, while Malawians appear to have become better educated over time in terms of academic qualifications achieved. Nutritional outcomes reveal a somewhat mixed picture, with declines in stunting and underweight children, but an increase in wasting (see Verduzco-Gallo et al. 2014 for a more in-depth discussion with regards to food and nutrition security outcomes).

The IHS2 and IHS3 also include several questions related to subjective well-being vis-à-vis adequacy of food consumption, housing, clothing, and health care. We treat households that report access to be ‘less than adequate’ as being deprived, while those that report ‘just adequate’ or ‘more than adequate’ are (p.107)

Table 5.5. Changes in subjective well-being, 2004/5–10/11

Prevalence of food inadequacy

Prevalence of housing inadequacy

2004/5 (IHS2) (%)

2010/11 (IHS3) (%)

Percentage point change (%)

2004/5 (IHS2) (%)

2010/11 (IHS3) (%)

Percentage point change (%)

Urban

48.1

26.7

−21.5

44.2

31.5

−12.6

Rural

58.5

40.4

−18.1

56.8

44.5

−12.3

North

35.3

29.8

−5.5

36.4

39.0

2.6

Centre

59.1

33.8

−25.3

60.3

40.4

−19.9

South

63.8

49.2

−14.6

58.7

49.9

−8.8

Total

57.3

38.5

−18.9

55.4

42.8

−12.5

Prevalence of clothing inadequacy

Prevalence of health care inadequacy

2004/5 (IHS2) (%)

2010/11 (IHS3) (%)

Percentage point change (%)

2004/5 (IHS2) (%)

2010/11 (IHS3) (%)

Percentage point change (%)

Urban

54.5

49.3

−5.2

51.8

28.1

−23.7

Rural

74.9

58.5

−16.4

62.0

34.5

−27.5

North

52.3

55.3

3.0

43.1

27.8

−15.3

Centre

84.2

54.7

−29.5

65.4

39.3

−26.1

South

71.8

62.1

−9.7

63.6

31.1

−32.5

Total

72.6

57.0

−15.5

60.9

33.4

−27.5

Source: Authors’ computation using IHS2 and IHS3

considered not deprived. The corresponding prevalence rates of deprived households are reported in Table 5.5. We note substantial reductions in the percentage of households reporting to be deprived in terms of access to food (–18.9 per cent), housing (–12.5 per cent), clothing (–15.5 per cent), and health care (–27.5 per cent), which is largely consistent with our corresponding finding of a large decline in national poverty. Equally significant declines in deprivation rates are observed in most regions of Malawi.

5.5 Conclusion

Malawi experienced rapid economic growth during the period 2005–11. Although growth was broad-based and originated from several sectors, the most important contribution came from the large and rapidly growing agricultural sector, which enjoyed significant support in the form of fertilizer input subsidies during the period. On average, and somewhat remarkably, this sector purportedly grew at an average annual rate of 10.1 per cent per year, while overall GDP expanded at a rate of 7.1 per cent per year, far exceeding (p.108) population growth, and thus allowing per capita GDP to grow at roughly 3.5 per cent per year on average during the period.

A reasonable expectation was that rapid smallholder-led agricultural growth would have a significant impact on poverty. However, the official poverty estimates based on the IHS3, which were released in 2012, suggested national poverty had declined only marginally by 1.7 percentage points between 2004/5 and 2010/11, while rural poverty increased by 0.7 percentage points. This raised several questions, including whether reported maize production and agricultural GDP growth estimates were in fact accurate. Others speculated about whether or not the poor were excluded from the benefits of rapid growth. The sharp rise in inequality measured by the Gini coefficient (0.39 to 0.45) supported the ‘no trickle-down effect’ hypothesis to some extent, but the puzzling question remained: how could Malawi have had no reduction in rural poverty after investing so heavily in poor smallholder agriculture and seemingly reaping rewards from the subsidy programme in terms of economic growth?

While recognizing that poverty measurement is challenging and poverty results are extremely sensitive to the assumptions made, this study notes two major concerns about the official poverty analysis conducted by the National Statistical Office (NSO). The first relates to the revision of the official consumer price index (CPI). Our analysis concurs that the official 2004–11 inflation rate (77.3 per cent) is indeed an underestimation, and that revision was justified. However, our estimate of the national average poverty line inflation rate (114.0 per cent) is somewhat lower than that of the NSO (128.9 per cent). The NSO approach of applying the same inflation rate to the food and overall poverty lines in both urban and rural areas is another source of significant bias in their estimated poverty rates. Our analysis suggests that urban poverty line inflation was higher (123.8 per cent) and rural poverty line inflation somewhat lower (108.0 per cent) than the national average. This, coupled with significant shifts in consumption patterns, makes an approach that adopts regional poverty lines and flexible but utility-consistent consumption bundles more appropriate.

Second, as pointed out by Verduzco-Gallo et al. (2014), there are several inconsistencies contained in the official sets of consumption conversion factors. Even minor adjustments to conversion factors, especially for important food commodities, may significantly alter estimates of poverty. There is need for a consultative process to agree on a final set of conversion factors that can be used in all ongoing and future poverty or nutrition analyses.

In contrast to the official poverty estimate, our own analysis reveals large and statistically significant declines in poverty between 2004/5 and 2010/11. The poverty rate declines by 7.5 percentage points in rural areas, with a particularly large decline in the rural north. Urban poverty also declines by (p.109) 10.3 percentage points, which is slightly higher than the official estimate of 8.1 percentage points. At national level, we find that poverty dropped by 8.4 percentage points. These results are consistent with the purported economic growth trajectory of Malawi, as well as evidence of significant improvements across a range of non-monetary dimensions of welfare in recent years.

While these results corroborate a much more positive narrative about FISP and its impact on growth and poverty reduction, it still appears as if FISP, and possibly other economic policies in Malawi, continue to neglect the ultra-poor. We find a borderline statistically significant increase in rural extreme poverty (2.1 percentage points), and a more sizable and statistically significant decline in urban areas (i.e. by 4.3 percentage points). The direction of these changes is consistent with the findings of the NSO. Many among the ultra-poor are landless or labour-constrained households, and would make ideal candidates for other forms of social support such as cash transfers, which are currently being scaled up in Malawi.

There are some areas that require further analysis. We concur that the official CPI series understates inflation in Malawi. The implication is that the GDP deflator currently used to estimate real GDP levels may also be understated, leading to economic growth estimates that are too optimistic. At present, GDP estimates are available up until 2012, but all estimates beyond 2007 are subject to revision. In order to truly understand the growth–poverty puzzle, more up-to-date national accounts data is needed, while supporting data for estimation of agricultural GDP, such as crop estimates, need to be strengthened. A more likely growth–poverty narrative for Malawi is probably that growth was lower but that the poverty outcome was significantly more optimistic than what official estimates suggest, particularly in rural areas. Extreme poverty, however, may have increased slightly, suggesting that the most vulnerable in Malawi’s society have been excluded from the benefits of economic policy and growth.

References

Bibliography references:

Arndt, C., K. Pauw, and J. Thurlow (2015). ‘The Economy-wide Impacts and Risks of Malawi’s Farm Input Subsidy Program’, American Journal of Agricultural Economics, forthcoming, doi: 10.1093/ajae/aav048.

Arndt, C. and K. Simler (2010). ‘Estimating Utility Consistent Poverty Lines’, Economic Development and Cultural Change, 58: 449–74.

Baden, S. and Barber, C. (2005). The Impact of the Second-hand Clothing Trade on Developing Countries. Oxford: Oxfam.

Benin, S., J. Thurlow, X. Diao, C. McCool, and F. Simtowe (2012). ‘Malawi’, in X. Diao, J. Thurlow, S. Benin, and S. Fan (eds), Strategies and Priorities for African Agriculture: (p.110) Economy-wide Perspectives from Country Studies. Washington, DC: International Food Policy Research Institute, 245–79.

Chirwa, E. and A. Dorward (2013). Agricultural Input Subsidies: The Recent Malawi Experience. Oxford: Oxford University Press.

Diao, X., P. Hazell, and J. Thurlow (2010). ‘The Role of Agriculture in African Development’, World Development, 38(10): 1375–83.

Dollar, D., T. Kleineberg, and A. Kraay (2013). ‘Growth Still Is Good for the Poor’, Policy Research Working Paper No. 6568. Washington, DC: World Bank.

Foster, J., J. Greer, and E. Thorbecke (1984). ‘A Class of Decomposable Poverty Measures’, Econometrica, 52: 761–6.

GoM (Government of Malawi) (2012). ‘The Malawi Growth and Development Strategy II (MGDS II), 2011–2016’. Lilongwe: Government of Malawi.

Jayne, T. S., A. Chapoto, I. Minde, and C. Donovan (2008). ‘The 2008/09 Food Price and Food Security Situation in Eastern and Southern Africa: Implications for Immediate and Longer Run Responses’, International Development Working Paper. East Lansing: Michigan State University.

Jayne, T., D. Mather, N. Mason, and J. Ricker-Gilbert. (2013). ‘How Do Fertilizer Subsidy Programs Affect Fertilizer Use in Sub-Saharan Africa? Crowding Out, Diversion, and Benefit/Cost Assessments’, Agricultural Economics, 44: 687–703.

Lunduka, R., J. Ricker-Gilbert, and M. Fisher (2013). ‘What are the farm-level impacts of Malawi’s farm input subsidy program? A critical review’, Agricultural Economics, 44: 563–79.

MoAFS (Ministry of Agriculture and Food Security) (2013). ‘Agricultural Production Estimates Survey (APES), 2001–2012’. Lilongwe: Government of Malawi.

Mukherjee, S. and T. Benson (2003). ‘The Determinants of Poverty in Malawi’, World Development, 31: 339–58.

NSO (National Statistics Office) (2005). ‘Integrated Household Survey 2004–2005, vol. I: Household Socio-economic Characteristics’. Zomba, Malawi: National Statistics Office.

NSO (National Statistics Office) (2010). ‘Report Welfare Monitoring Survey 2009’, July. Zomba, Malawi: National Statistics Office.

NSO (National Statistics Office) (2012a). ‘Integrated Household Survey 2004–2005. Household Socio-economic Characteristics Report’. Zomba, Malawi: National Statistics Office.

NSO (National Statistics Office) (2012b). ‘Quarterly Statistical Bulletin’, September. Zomba, Malawi: National Statistics Office.

Pauw, K., U. Beck, and R. Mussa (2014). ‘Did Rapid Smallholder-Led Agricultural Growth Fail to Reduce Rural Poverty? Making Sense of Malawi’s Poverty Puzzle’, WIDER Working Paper No. 2014/123, October. Helsinki: United Nations University—World Institute of Development Economics Research.

Pauw, K., O. Ecker, and J. Mazunda (2011). ‘Agricultural Growth, Poverty, and Nutrition Linkages in Malawi’, Malawi Strategy Support Program Policy Note 8. Washington, DC: International Food Policy Research Institute.

Pauw, K. and J. Thurlow (2014). ‘Malawi’s Farm Input Subsidy Program: Where Do We Go from Here?’, Malawi Strategy Support Program Policy Note 18. Washington, DC: International Food Policy Research Institute.

(p.111) Ravallion, M. (1998). ‘Poverty Lines in Theory and Practice’, Living Standards Measurement Study Working Paper No. 133. Washington, DC: World Bank.

Ricker-Gilbert, J. (2014). ‘Wage and Employment Effects of Malawi’s Fertilizer Subsidy Program’, Agricultural Economics, 45: 337–53.

Sen, A. K. (1985). Commodities and Capabilities. Amsterdam: North Holland.

Simler, K. and C. Arndt (2007). ‘Poverty Comparisons with Absolute Poverty Lines Estimated from Survey Data’, Review of Income and Wealth, 53(2): 275–94.

Tarp, F., K. Simler, C. Matusse, R. Heltberg, and G. Dava (2002). ‘The Robustness of Poverty Profiles Reconsidered’, Economic Development and Cultural Change, 51(1): 77–108.

Thorbecke, E. (2005). ‘Multi-Dimensional Poverty: Conceptual and Measurement’. Paper prepared for the Many Dimensions of Poverty International Conference, UNDP International Poverty Centre, Brasilia, 29–31 August.

Verduzco-Gallo, I. E. Ecker, and K. Pauw (2014). ‘Changes in Food and Nutrition Security in Malawi: Analysis of Recent Survey Evidence’, MaSSP Working Paper 6, July, Malawi Strategy Support Program, International Food Policy Research Institute.