METHODOLOGY FOR POVERTY MEASUREMENT IN MALAWI (2016/17) The National Statistics Office of Malawi and The World Bank, Poverty and Equity Global Practice 2018 Zomba, Malawi 1. Introduction ............................................................................................................................. 3 2. Welfare indicator ..................................................................................................................... 4 2.1. Consumption aggregate .................................................................................................... 4 2.1.1. Food component........................................................................................................ 4 2.1.2. Nonfood component.................................................................................................. 6 2.1.3. Durable goods ........................................................................................................... 7 2.1.4. Rent for housing ........................................................................................................ 8 2.2. Adjustment for household size and composition ............................................................. 9 2.3. Adjustments for cost of living differences ..................................................................... 10 2.3.1. Between the Third and Fourth Integrated Household Surveys ............................... 10 2.3.2. Price adjustment within the Fourth Integrated Household Survey ......................... 12 3. Poverty line ............................................................................................................................ 14 4. Poverty ................................................................................................................................... 15 4.1. Poverty measures............................................................................................................ 15 4.2. Poverty estimates............................................................................................................ 16 4.3. Robustness of poverty results......................................................................................... 17 4.3.1. Density of consumption .......................................................................................... 18 4.3.2. Survey to Survey (S2S) ........................................................................................... 18 Appendix ....................................................................................................................................... 20 Appendix A: List of items included in the food index .............................................................. 20 Appendix B: Adjustment for household composition ............................................................... 23 Appendix C: Survey-to-survey imputation ............................................................................... 25 Appendix D: Classification of individual consumption by purpose (COICOP) ....................... 25 References ..................................................................................................................................... 30 1. Introduction The fourth Integrated Household Survey (IHS4) is a multi-topic survey implemented by the National Statistical Office (NSO) of Malawi between April 2016 to April 2017. Like its predecessor, IHS3 (2010/11), this survey provides socioeconomic indicators that are representative at district level. IHS3 and IHS4 are comparable because similar questionnaires were implemented to collect the data, and the sample selection is based on a sampling frame using the 2008 population census. The sampling frame used in both surveys includes the three regions (North, Central, and South) and is stratified into rural and urban strata. The urban stratum includes the four major cities (Lilongwe, Blantyre, Mzuzu, and Zomba). In both surveys, a stratified two-stage sampling design is used. The IHS4 and IHS3 collected information from 12,271 and 12,447 households, respectively. About 88 percent of the sample in both surveys came from 28 districts in rural areas. A key difference between IHS4 and IHS3 is the data collection mechanism: IHS4 was conducted using the Computer-Assisted Personal Interviewing (CAPI), while IHS3 was implemented using a paper questionnaire. Despite this difference in the platform of data collection, IHS3 and IHS4 are comparable and hence can be used to monitor changes in welfare and the progress toward achieving some of the targets/goals in the Malawi Growth and Development Strategy (MGDS). The purpose of this note is to describe the methodology used to measure monetary poverty using IHS4 data and steps taken to ensure comparability with the poverty estimates generated using IHS3. To conduct poverty analysis, we need two main sets of information: (a) a welfare indicator that ranks individuals and (b) a threshold welfare level (that is, poverty line) below which an individual is considered as poor. To rank the population from the person with the lowest to the highest welfare level, the total expenditure on food and nonfood items is used. Individuals with the welfare indicator below and above the poverty line are classified as poor and nonpoor, respectively. The welfare indicator is constructed using the same methodology as IHS3 to guarantee comparability over time. The rest of the report is organized as follows. Section 2 outlines the steps in the construction of the nominal consumption aggregate and adjustments for living cost differences and household size. Section 3 describes the derivation of the poverty line, and finally Section 4 presents the poverty measures used in this report and the poverty estimates. 3 2. Welfare indicator Previous poverty measurement studies have reached some consensus on the use of monetary values as an indicator of welfare/living standard, and this approach is regularly employed for poverty monitoring and analysis. Although the monetary indicator of welfare does not cover all aspects of human well-being, it captures a central component of any assessment of living standards. In developing countries such as Malawi, it is also a common practice to use consumption expenditure as the preferred welfare indicator because it is likely to be a more accurate measure of living standards than income. The following subsections describe the construction of aggregate consumption expenditure by component: food, nonfood, durables, and rent expenditures. 2.1. Consumption aggregate 2.1.1. Food component Measurement of food consumption is critical for poverty analysis as food is basic for human survival and standard of living. The food module of IHS4 collects data on the food consumed in the household over the past one-week recall period. More specifically, consumption information was collected on 136 food items from the most knowledgeable member of the sampled household. To make the data collection and analysis easier, these food items were organized into 11 categories: cereals, grains, and cereals products; roots, tubers, and plantains; nuts and pulses; vegetables; meat, fish, and animal products; fruits; cooked food from vendors; milk and milk products; sugar, fats, and oil; beverages; and spices and miscellaneous. During the construction of the food component of total consumption, several considerations and adjustments were made. First, all major sources of food consumption are accounted for. These include purchases, own-production, gifts, and other sources. Second, the survey has focused on actual consumption of food items as opposed to total purchases or total own-production. This distinction is important as not all purchased and/or own-produced items get consumed over the same period by all households. Indeed, many farm households cultivate crops not just for own consumption but for the market as well. Third, to get aggregate food consumption, monetary values of both purchased and non-purchased items were calculated. Because the survey collects information on both quantity and expenditure 4 on purchased food items, unit values were constructed by dividing expenditure with quantity consumed. These unit values are then used to calculate monetary values of non-purchased food items. However, adjustments must be made on unit values as they reflect not only price differences between different items but could also capture quality differences for the same item. This is particularly relevant if the item has many varieties and the IHS survey instrument did not capture these varieties separately. In this regard, IHS4 has some improvements over IHS3 in that consumption information on fish items are disaggregated not only on how they are made—that is, sundried, fresh, or smoked as in IHS3—but also on their size (small, medium, and large as in IHS4). In contrast to IHS3, consumption information on sweet potatoes and groundnuts were also further disaggregated by their types in IHS4. To deal with remaining quality differences in unit values in IHS4, median unit values were calculated for each item at several levels with both geographical and time dimensions. Geographical disaggregation includes district, urban and rural areas, and national. In these disaggregation, the survey month and year are taken into consideration. Thus, if a sampled household consumed a food item that was not purchased, the median unit value from its district and matching survey time would be used to value that consumption. If no other household consumed the same item in that district during the same survey month, or if there are not enough observations to obtain a reliable unit value, the median unit value from the immediate upper level (in this case urban or rural areas) during the same survey month and year would be used to estimate the value of that consumption. Fourth, to reduce cognitive and informational burdens on surveyed households, respondents were allowed to report their quantity consumption in nonstandard and local units such as heaps and pails. These units were transformed into kilograms using conversion factors that were collected from a supplementary survey conducted alongside Malawi’s 2013 Integrated Household Panel Survey (IHPS) and IHS3. This standardization of consumption information is necessary before unit values were calculated and expenditure on food was aggregated. Finally, to deal with exceptionally low or high levels of quantity of consumption and/or unit values, both automated and subjective trimming methods were implemented. The automated method includes measuring how far the unit value or quantity consumed for each item is from the mean or median of its distribution. For some items, the extremely high unit values are replaced by the 95th percentile and the extremely low unit values are replaced by the 5th percentile, and this decision 5 is taken on an item-by-item basis after extensive consultation between experts at the NSO and the World Bank.1 2.1.2. Nonfood component The nonfood consumption modules (Modules I–K) of IHS4 have detailed information on household expenditure on various nondurable nonfood goods and services. We include household expenditure on all nonfood items as described in the international standard for Classification of Individual Consumption by Purpose (COICOP). Appendix D shows the COICOP classification of items and the respective questions in the IHS4 questionnaire. Therefore, parts of the total nonfood expenditure is made up of the value of expenditure on nonfood nondurable item groups such as education; health services, including prescription and nonprescription drugs; housing utilities such as water, electricity, gas, firewood, charcoal, and others; clothing and footwear; transport service including operation cost of private vehicle/bicycle/motorbike, but not the actual purchase of these durable items, and public transportation; communication services such as mobile phone services; recreation and cultural services, except the purchase of durables such as televisions; hotel and lodging; and miscellaneous goods and services such as personal care like soap and personal effects such as umbrella. Expenditures on these goods and services are reported/collected in different reference periods (past 7 days, 1 month, 3 months, and 12 months). For those items with a reference period shorter than 12 months, the corresponding expenditure is annualized. The total annual household expenditure on these goods and services is compiled to calculate the total expense on nondurable nonfood items and matched with durable goods and rental/housing expense in the corresponding COICOP code. Some expenditures that are sporadic in nature such as wedding, funerals, and births are excluded from consumption aggregate, which is intended to capture households’ regular expenditure, to avoid overestimation of well-being. Remittance to others is excluded from consumption aggregate as it does not imply welfare improving consumption. Expenditure to repair or upgrade dwelling such as purchase materials and labor cost is also excluded from consumption as the housing/rental expenditure, discussed below, captures the value gains from this repair/upgrade. 1 The list of items and their respective trimming threshold is available upon request. 6 Finally, it is important to note that we rely on total expenditure values and that there is no unit value data for nonfood goods and services. The diversity of nonfood items, both in quality and unit price, makes it difficult to compute a standard price for these nonfood items. For instance, the type, quality, and unit of measurement of prescription medicines are so diverse that it is not possible to calculate their unit values. 2.1.3. Durable goods The ownership and utilization of durable goods is a crucial component of consumption aggregate as these goods improve the well-being of households. However, these goods are often purchased occasionally and used over extended periods. To properly account for the welfare of households, it is important to impute the use value of (or utility derived from) these goods in each year of service—not at the time of purchase. The utility derived from the use of these goods could be imputed using the purchase value and the expected lifetime of the goods. Estimation of the use value derived from these durable goods is based on the information collected in the data and certain assumptions outlined below. The durable goods module (Module L) of IHS4 collects information on 22 home appliances and other durables used by households to improve their daily lives.2 The information collected about these items include their age, estimated current value, and number of each item owned by the household. Using the information on current value, age, and number of goods and the following three important assumptions, we estimate the use value.3 First, the purchase of these durables is assumed to be uniformly distributed over time. This assumption allows us to estimate the lifetime of each durable good, except car and motorcycle, as twice the current age of the item. For car and motorcycle, ownership of which are recent phenomenon in rural Malawi, the distribution is likely skewed and hence we calculate lifetime of these two durables as three time their current age.4 Table 1 presents the average estimated lifetime of each durable good and the number of households that own at least one of these items. Second, the remaining service years left for each durable good is calculated as its current age minus the 2 This module also collects information on durable assets used for productive purposes. However, these goods are not directly used to improve welfare and hence are not included in the consumption aggregate. 3 Due to lack of data on purchase value, the estimated current value is used for approximating the total value of the durable goods. 4 The decision to use a different approach for car and motorcycle gives a more reasonable estimated lifetime: 11.8 years and 9.0 years, respectively. However, if we decide to adopt uniform distribution assumption, the estimated lifetime for car and motorcycle becomes only 7.9 years and 6.0 years, respectively. 7 estimated lifetime of the good. For goods that are very old, the estimated remaining service left might be negative. In such cases, the remaining service year is replaced by two years. Finally, the ratio of the current value and the remaining lifetime of services is used to approximate the annual use value of each durable good. Table 1: Estimated average lifetime of consumer durables 2010 2016 Lifetime Households Lifetime Households (years) reporting (years) reporting 1 Mortar/pestle (mtondo) 12.3 1,497 15.4 5,600 2 Bed 12.5 1,441 13.8 4,689 3 Table 12.4 1,282 13.3 3,605 4 Chair 10.2 1,389 11.2 4,577 5 Fan 6.6 173 7.7 586 6 Air conditioner 9.7 9 7.5 12 7 Radio ('wireless') 7.5 1,589 8.2 4,536 8 Tape or CD/DVD player 6.2 472 8.1 1,274 9 Television 7.3 424 9.3 1,583 10 VCR 8.5 46 10.5 106 11 Sewing machine 13.8 114 18.6 297 12 Kerosene/paraffin stove 10 34 15.4 42 13 Electric or gas stove 6.8 180 8.1 487 14 Refrigerator 9.6 193 9.7 750 15 Washing machine 6 8 10.1 18 16 Bicycle 11.4 1,224 12.4 4,551 17 Motorcycle/scooter 10.7 14 9.0 218 18 Car 7.7 79 11.8 259 19 Computer equipment 5.2 66 5.9 343 20 Satellite dish 6.2 136 8.2 606 21 Solar panel 4.4 43 5.5 905 22 Generator 4.6 36 8.8 91 Note: Households reporting refers to the total number of households that have reported owning these goods. 2.1.4. Rent for housing Like durable goods, only the service derived from dwellings, not the construction or repair expenses, needs to be included in the consumption basket. The residence of a household could 8 either be owned by the household itself or rented from others. The rental expenditure on dwellings rented from other owners could be a good estimate of the service value of housing if the rental market is competitive. The IHS4 housing module (Module F) collects rental expense for households that rent their residences from others. However, most households, especially in rural areas, own their dwellings. For these households, self-reported rental values are collected. The self-reported rental data might, however, be inaccurate. To improve the accuracy of self-reported rent, information on actual rental expense is used. To improve the accuracy of self-reported rental expenses (as well as actual rent), a hedonic regression is estimated using logarithm of rent (for those who are renting) and a hedonic rental value is imputed for both renter and non-renter households. The estimation takes into account types of dwelling (number of rooms and type of wall, roof, and floor), services available in the dwelling (source of drinking water, type of toilet, and availability of electricity in the home and in the village/town), and region and survey time fixed effects (urban, region, district, and survey year and month fixed effects). Based on the regression coefficients and the characteristics of the dwellings, the predicted rental value of the dwelling is estimated.5 These estimates are used to replace outliers in the rent and self-reported rent data. 2.2. Adjustment for household size and composition The next step in the construction of the welfare indicator requires adjusting consumption to account for household size and demographic composition of households to make welfare comparisons across individuals, not across households. This involves converting the standard of living defined at the household level to an indicator defined at the individual level. Adult equivalence scales are sometimes used to convert household consumption into individual consumption by correcting for differences in the demographic composition and household size. In this report, consumption expenditure per capita is used as indicator of individual welfare. A sensitivity analysis of poverty estimates to applying a different scale for children is presented in Appendix B. 5 The predicted rent in logarithm is converted into normal scale. 9 2.3. Adjustments for cost of living differences 2.3.1. Between the Third and Fourth Integrated Household Surveys To estimate poverty rates using IHS4, the poverty line from the previous round (IHS3) must be updated to the latest round. This is accomplished by inflating the national poverty line from March 2010 (the start of IHS3 fieldwork) to April 2016 (start of IHS4 fieldwork). To do this, price indexes need to be used to express the poverty lines from both rounds in the same constant real prices. The Consumer Price Index (CPI) is often used to adjust for cost-of-living differences between the two periods. However, the NSO has rebased its CPI in 2012. The CPI series before and after the rebasing are not directly comparable because they use different weights and price data between before and after the rebasing are not fully comparable. The quality of food price data in the CPI before the rebasing has been particularly weak and not directly comparable to price data after the rebasing in 2012. In such cases where a continuous CPI series between two consecutive household surveys is not available, a food price index is often developed using the unit values (expenditure divided by quantity) of items in the two surveys (IHS4 and IHS3). This approach has several advantages under this circumstance, including the use of actual transactions the households conducted and price data coming from several locations, especially for items such as maize that is commonly consumed. Also, unit values are used to value consumption of own production. However, unit values are also known to include both price effects and quality effects (see Deaton and Tarozzi 2000). Therefore, the price adjustment relies on a survey-based food price index by using unit values of food items in IHS3 and IHS4. To adjust for changes in prices of nonfood items, the official nonfood CPI is used. The food and nonfood indexes are then combined using their respective budget shares to calculate overall inflation rate between March/April 2010 and April/May 2016. The unit values of 57 food items, which account for 88.2 percent of the total food consumption in IHS4, are used to construct the food price index. Relatively rare food items that are consumed by less than 20 households in IHS4 and/or IHS3 are excluded from the analysis. The resulting food inflation rate was found to be almost identical if food items consumed by fewer than 50 households 10 were excluded from the food price index calculation.6 The food price index was calculated over the first two survey months of IHS3 and IHS4 (that is, between March/April 2010 and April/May 2016) as opposed to calculating the food price index between the first survey month of the two rounds. This is done to increase the number of common items between IHS3 and IHS4 and to reduce the impact of outlier unit values. The list of food items, their weight, median unit values, and number of sampled households that have reported their consumption are presented in Table A.1. The food price index is, therefore, estimated based on the food module of IHS3 and IHS4. The median food unit values are used rather than mean values as the former is less likely to be affected by extreme values. The food index is calculated as follows: 𝐴�𝑟𝑖𝑙/𝑀𝑎𝑦,2016 �𝑗 Food Index = ∑� 𝑗=1 𝑤𝑗 × 𝑀𝑎𝑟�ℎ/𝐴�𝑟𝑖𝑙,2010 × 100, �𝑗 where 𝑤𝑗 is the national weight for item 𝑗. It is the average food budget share for item j in IHS4. 𝑡 The national median unit value for item 𝑗 in month and year 𝑡 is represented by �𝑗 . The averages 𝑀𝑎𝑟�ℎ/𝐴�𝑟𝑖𝑙,2010 of these median values over the first two survey months are taken to represent �𝑗 𝐴�𝑟𝑖𝑙/𝑀𝑎𝑦,2016 and �𝑗 for item 𝑗. The nonfood index is calculated from the official nonfood CPI. The official nonfood CPI is used because it is less affected by revisions in data collection and handling before and after the 2012 rebase. In addition to this, IHS4 did not collect price data on nonfood items as in the price module of the community survey in IHS3.7 This precluded the possibility of estimating nonfood index using the community surveys of IHS3 and IHS4. The food and the nonfood price indexes are combined using the food and the nonfood consumption shares: 50.2 percent and 49.4 percent, respectively. The resulting overall inflation is 271.4 percent between March/April 2010 and April/May 2016. The corresponding food and nonfood inflations 6 Note that rare food items have low weight in the food basket, and hence including/excluding them has minimal effect on the estimated food inflation. 7 This approach was adopted to estimate the nonfood inflation rate between the 2010 and 2013 Integrated National Panel Surveys. 11 are 319.2 percent and 222.5 percent, respectively. Thus, an inflation of 271.4 percent is used to adjust the poverty line expressed in March 2010 prices to a poverty line expressed in April 2016. 2.3.2. Price adjustment within the Fourth Integrated Household Survey For poverty analysis using household surveys, the nominal consumption must also be adjusted for temporal and spatial differences in cost of living observed within the survey period and across survey locations. The temporal adjustment deals with differences in cost of living over time (April 2016 to April 2017). For example, MWK 1,000 in April 2016, or at the start of the fieldwork for IHS4, may not be worth the same value in April 2017, or at the end of the fieldwork for the survey. The spatial adjustment deals with differences in cost of living over locations. For example, MWK 1,000 in a rural district may not be worth the same in a large city such as Blantyre. Because temporal price variations can differ significantly across areas, a region-specific temporal adjustment is implemented by using a combination of the unit values of food items from IHS4 and the official nonfood CPI. These itemized unit values are combined with their respective average food budget shares in the household survey to calculate the regional monthly food price index. The food price index is then combined with the nonfood CPI to calculate the overall regional monthly price index. The food price index is calculated using unit values from the household survey— consistent with the price adjustment across surveys described earlier. At the end of this exercise, consumption in IHS4 is adjusted to regional prices of April/May 2016. Figure 1 shows these regional monthly price indexes by region. Figure 1: Regional monthly price indexes by region and month of fieldwork (2016/17) 120 115 Regional CPI 110 105 100 2016m4 2016m8 2016m12 2017m4 Urban Rural North Rural Centre Rural South Source: Authors’ calculation using IHS4. 12 In addition, adjustments were also made for spatial cost-of-living differences across regions. To do this, a spatial Paasche price index is estimated. Similar to the temporal price adjustments above, food prices come from unit values from IHS4, while the price data for nonfood items come from the price data used to calculate official nonfood CPI. Following the source of the prices, the weights of the items in the price index come from IHS4 for food items and the weights for nonfood items comes from the regional weight of the official nonfood CPI. Both correspond to average budget shares at the regional level. The food and nonfood price indexes are then combined using the average budget shares of the two consumption aggregates at the regional level. The base for spatial price index is All-Malawi for April/May 2016, which are the beginning months of fieldwork for IHS4. Average national prices are compared with average regional prices for the same period. By having the same reference period at the national and regional levels, the difference in prices in this calculation is attributable only to spatial differences. Table 2 shows the spatial price indexes. As expected, urban areas are the most expensive followed at a distant second by Rural North. Rural Center and Rural South regions are similarly expensive. Table 2: Paasche spatial price indexes by region Location April/May 2016 All-Malawi 100.0 Urban 110.7 Rural North 100.2 Rural Central 90.5 Rural South 92.8 Source: Authors’ calculation using IHS4. After combining the spatial and temporal differences in prices, the trends in regional price indexes for the IHS4 survey period are presented in Figure 2. The basis here is national prices for April/May 2016. 13 Figure 2: Temporal and spatial indexes 130 120 110 100 90 2016m4 2016m8 2016m12 2017m4 Urban Rural North Rural Centre Rural South Source: Authors’ calculation using IHS4. 3. Poverty line The cost-of-basic-needs approach is most commonly used to establish a poverty line. In this approach, the cost of acquiring enough food for adequate nutrition—in the case of Malawi 2,400 calories per person per day—is first estimated and then an allowance for the cost of other basic needs is added (Haughton and Khandker 2009; Ravallion 1998). The poverty line in Malawi was set in 2005 using the basic needs approach. Therefore, the total poverty line is the cost meeting basic nutritional needs (that is, food poverty line) and the allowance for other basic needs (that is, the nonfood poverty line). The nonfood allowance was estimated as the average nonfood consumption of the population whose food consumption is close to the food poverty line. 8 If a person’s total expenditure is below the poverty line, the person is considered poor. An individual with consumption below the food poverty line is considered ultra-poor. The poverty line is in essence absolute, and it also needs to be expressed in constant prices (that is, real poverty line). In other words, the poverty line is absolute because it fixes the same standard 8 Further details on the construction of the poverty line is available in the 2005 NSO report (NSO 2005). 14 of living throughout Malawi—two persons with the same welfare level will be treated the same way regardless of the location of their residence. Similarly, to ensure proper comparison of well- being over time, the real poverty line is used. The 2004/05 (IHS2) poverty line has been adjusted to reflect 2010 prices (IHS3) in the previous poverty measurement. The inflation rate (271.4 percent) calculated in Section 2.2.1 is used to adjust the March 2010 poverty line to reflect April 2016 prices. Table 3 shows the poverty line expressed in March 2010 and April 2016 prices. Table 3: Poverty lines per person per year (in 2010 and 2016 prices) 2010 prices 2016 prices Food 22,956 85,259 Nonfood 14,046 52,167 Total 37,002 137,425 Source: IHS3 poverty line and price index 4. Poverty 4.1. Poverty measures To measure poverty, the class of poverty measures proposed by Foster, Greer, and Thorbecke (FGT) are used (Foster et al., 1984). In addition to the poverty headcount index, the FGT provides poverty gap and severity indexes. This family of poverty indexes can be summarized by the following equation: � 1 𝑧 − 𝑦𝑖 𝛼 𝑃𝛼 = ∑ ( ) ∗ 𝐼(𝑦𝑖 < 𝑧), � 𝑧 𝑖=1 where � is a nonnegative parameter that takes value 0, 1, or 2; 𝑧 is the poverty line; 𝑦𝑖 denotes consumption of individual 𝑖 ; and � is the total number of individuals in the population. 𝐼(𝑦𝑖 < 𝑧) is an indicator function which is equal to 1 when individual 𝑖′s consumption is below the poverty line and 0 when the consumption is above the poverty line. The poverty headcount index (� = 0) is the percentage of population whose consumption is below the poverty line. This simple and easy-to-interpret index is the most widely used poverty measure. However, it has some limitations in that it does not capture how close/far the poor are from the poverty line and the distribution of consumption among the poor. Two other poverty indexes address these limitations. The poverty gap (� = 1), which is the average consumption shortfall of 15 the poor relative to the poverty line, addresses the first limitation by accounting for extent consumption shortfall. Finally, the poverty severity (� = 2), which is also called poverty gap squared, accounts for the inequality among the poor. For instance, redistribution of consumption among the poor will not be captured by both poverty headcount and poverty gap. However, such a transfer, for example, transfer from a poor person to somebody less poor, increases poverty severity but might not affect headcount or poverty gap. In the poverty severity index, larger poverty gaps carry higher weight (Haughton and Khandker 2009). 4.2. Poverty estimates The poverty estimates for 2016/17 (IHS4) are presented, alongside the poverty estimates for 2010/11 (IHS3), in Table 4. The results show that the national poverty headcount has not changed much since 2010/11. By 2016/17, about 51.5 percent of Malawians had consumption level below the national poverty line, and this is after a 0.8 percentage point increase since 2010/11. This small increase in national poverty rate is not statistically significant. The results also show that the poverty headcount rate remains much higher in rural areas (59.5 percent) than urban centers (17.7 percent). The poverty headcount rate increased in rural areas by 2.8 percentage points between 2010/11 and 2016/17. Comparison of the poverty rate across rural regions shows that the poverty rate remains the highest in Rural South, followed by Rural North. However, poverty rate has increased in both Rural South and Rural Central, while it did not change in Rural North. The poverty gap, which captures the average shortfall of consumption of the poor as a percentage of the poverty line, has decreased slightly between 2010/11 and 2016/17. Nationally, the poverty gap decreased from 18.9 percent to 16.8 percent. This decrease is observed both in urban and rural areas. 16 Table 4: Poverty indexes (2010/11 and 2016/17) 2016/17 2010/11 Std. [95% Conf. Std. Std. [95% Conf. Coef. Coef. Err. Interval] Err. Err. Interval] Poverty headcount rate National 51.5 1.1 49.3 53.7 50.7 1.2 48.4 53.0 53.0 Urban 17.7 1.8 14.2 21.3 17.3 2.6 12.2 22.4 22.4 Rural 59.5 1.0 57.5 61.4 56.6 1.1 54.5 58.8 58.8 Rural North 59.9 2.1 55.7 64.1 59.9 2.4 55.3 64.6 64.6 Rural Center 53.6 1.5 50.5 56.6 48.7 1.7 45.3 52.0 52.0 Rural South 65.2 1.3 62.7 67.8 63.3 1.6 60.2 66.4 66.4 Poverty gap National 16.8 0.5 15.9 17.8 18.9 0.6 17.7 20.1 20.1 Urban 4.5 0.5 3.5 5.5 4.8 0.8 3.3 6.4 6.4 Rural 19.7 0.5 18.7 20.6 21.4 0.6 20.2 22.6 22.6 Rural North 18.9 0.9 17.1 20.7 22.2 1.2 19.8 24.6 24.6 Rural Center 16.4 0.7 15.0 17.7 17.3 0.9 15.6 19.0 19.0 Rural South 23.2 0.7 21.8 24.5 25.1 0.9 23.3 27.0 27.0 Poverty gap squared National 7.4 0.3 6.8 7.9 9.3 0.4 8.6 10.0 10.0 Urban 1.6 0.2 1.2 2.1 2.0 0.4 1.3 2.8 2.8 Rural 8.7 0.3 8.1 9.2 10.6 0.4 9.8 11.3 11.3 Rural North 7.9 0.5 7.0 8.9 10.7 0.7 9.2 12.1 12.1 Rural Center 6.8 0.4 6.0 7.5 8.3 0.5 7.2 9.3 9.3 Rural South 10.8 0.4 9.9 11.6 12.8 0.6 11.6 14.0 14.0 Source: Authors’ calculation using IHS3 and IHS4. The poverty rate at district level is presented in Table A.2. The results indicate that poverty rate is much higher in districts in South region. 4.3. Robustness of poverty results As a robustness check, two further analysis are conducted. First, a comparison of the kernel densities of per capita consumption in IHS3 and IHS4 is presented along with moderate and ultra- 17 poverty lines. Second, a survey-to-survey (S2S) imputation of consumption in IHS4 is conducted using IHS3. 4.3.1. Density of consumption The kernel densities of per capita consumption in IHS3 and IHS4 are presented in Figure 3. The consumption per capita and poverty lines are expressed in April 2016 prices. The densities show that there is no change in the share of the population under the moderate poverty line. The area between IHS3 and IHS4 densities before their second intersection (area A) is almost equal to the area after their second intersection and the moderate poverty line (area B+C). This implies that moderate poverty did not change much. On the other hand, ultra-poverty has declined: the area before their second intersection (area A) is greater than the area between them after the second intersection and ultra-poverty line (area B). Figure 3: Kernel density of consumption per capita (in April 2016 prices) C B A 4.3.2. Survey to Survey (S2S) An S2S imputation approach is used to estimate a reliable poverty trend across surveys. An S2S imputation methodology estimates consumption models in one survey and applies the models to another survey to estimate household expenditures and poverty rates. This method is often used when there is a debate on comparability of CPIs and/or consumption data. Often, as in the case of Malawi, the CPI series might be discontinuous and/or it might deviate from price index imputed using unit value. In such cases, the S2S technique is used to adjudicate the choice the right price 18 adjustment (Christiaensen et al. 2012). This approach was used for Russia, Morocco, Lesotho, Gabon, and Malawi (2010/11–2013). This method, however, assumes that the estimated relation between consumption and its predictors is stable over time (Christiaensen et al. 2012). There are several variants of S2S approach. For this analysis, we adopted a methodology developed by Yoshida et al. (2015). This methodology was developed on a small area estimation method proposed by Elbers, Lanjouw, and Lanjouw (2002, 2003) while incorporating multiple imputation technique developed by Rubin (1987) and Schafer (1999) and a machine learning technique developed by James et al. (2013). For this analysis, to reflect regional specific consumption patterns, we estimated models separately for Urban, Rural North, Rural Central, and Rural South regions using IHS3 (2010/11). The final models are presented in Appendix C. Applying these models to IHS4 data, household expenditures and poverty rates are estimated. The comparison between the poverty estimates presented in Section 4.2 and the S2S implied poverty estimates is presented in Figure 4. The result shows that the two estimates are statistically identical. Figure 4: Comparison between S2S imputation and IHS4 poverty estimations 80 S2S Poverty estimates 60 64.4 65.2 59.9 56.6 53.6 50.3 49.7 51.5 40 20 17.0 17.7 0 Urban Rural North Rural Center Rural South National 19 Appendix Appendix A: List of items included in the food index Table A.1: Food items used in the food price index calculation IHS4 IHS3 Households Median unit Median unit Households Weight (%) reporting value in value in reporting in in April/May March/April March/April April/May 2016 2010 2010 2016 1 Maize ufa mgaiwa 26.2 218.8 395 48.6 374 (normal flour) 2 Maize ufa refined (fine 13.6 265.0 169 49.8 210 flour) 3 Maize ufa madeya (bran 0.7 175.0 21 52.5 31 flour) 4 Green maize 2.4 166.3 52 36.8 44 5 Rice 2.4 806.5 264 150.0 254 6 Bread 1.5 600.0 308 232.7 275 7 Buns, scones 0.7 875.7 243 234.3 300 8 Biscuits 0.1 817.1 146 487.6 255 9 Cassava tubers 1.0 170.8 187 41.4 210 10 White sweet potato 1.4 116.9 443 38.3 171 11 Orange sweet potato 0.5 110.5 190 36.2 60 12 Irish potato 0.9 272.4 183 77.4 189 13 Plantain 0.1 192.8 41 13.1 28 14 Bean, white 0.8 725.0 142 165.0 123 15 Bean, brown 3.1 733.4 422 190.9 374 16 Pigeon pea (nandolo) 2.1 691.2 22 160.0 34 17 Groundnut flour 0.6 1452.9 279 390.8 130 18 Soybean flour 0.7 450.0 67 85.0 24 19 Cowpea (khobwe) 0.5 706.6 61 113.5 58 20 Onion 0.8 803.6 522 174.2 510 21 Cabbage 0.4 128.8 171 36.8 101 22 Tanaposi/rape 1.6 178.4 592 63.4 322 23 Nkhwani 3.3 367.8 225 90.8 219 24 Chinese cabbage 0.2 200.8 147 48.8 43 25 Tomato 4.3 493.8 989 105.7 791 26 Cucumber 0.3 246.3 77 56.4 92 27 Pumpkin 1.9 82.3 137 15.4 152 28 Okra/Therere 1.1 495.0 235 130.0 198 29 Eggs 1.7 1133.8 410 431.0 276 30 Beef 1.3 1500.0 135 500.0 127 31 Goat 2.0 1500.0 184 500.0 163 32 Pork 0.8 1275.0 131 375.0 81 20 IHS4 IHS3 Households Median unit Median unit Households Weight (%) reporting value in value in reporting in in April/May March/April March/April April/May 2016 2010 2010 2016 33 Chicken 2.6 1897.9 94 535.3 97 34 Banana 0.6 274.7 266 73.5 406 35 Citrus - naartje, orange, 0.1 187.6 123 46.9 91 and so on 36 Guava 0.2 143.6 77 45.0 98 37 Avocado 0.3 173.4 137 41.9 247 38 Fresh milk 0.5 463.6 144 90.5 151 39 Powdered milk 0.4 1930.0 150 1400.0 105 40 Margarine - blue band 0.2 1850.0 65 875.0 88 41 Sugar 3.7 650.0 842 145.0 832 42 Sugarcane 0.5 49.1 411 10.8 254 43 Cooking oil 5.2 1366.5 1025 469.6 804 44 Salt 2.2 462.3 1267 166.6 1295 45 Maize - boiled or roasted 0.1 283.2 32 55.2 22 (vendor) 46 Chips (vendor) 0.2 687.1 171 245.3 128 47 Meat (vendor) 0.1 1317.2 45 488.7 32 48 Fish (vendor) 0.1 1159.9 56 241.6 48 49 Mandazi, doughnut 1.2 947.2 497 247.0 369 (vendor) 50 Samosa (vendor) 0.1 638.5 75 229.1 43 51 Tea 0.2 1800.0 429 1000.0 481 52 Coffee 0.0 5000.0 27 2225.0 21 53 Squash (Sobo drink 0.3 750.0 73 200.0 136 concentrate) 54 Fruit juice 0.3 750.0 78 400.0 28 55 Freezes (flavored ice) 0.1 275.0 134 100.0 58 56 Soft drinks (Coca-Cola, 0.2 833.3 99 185.7 183 Fanta, Sprite, and so on) 57 Thobwa 1.4 130.0 65 40.0 35 Note. Households reporting refers to the total number of households that have reported consuming the food item. Food items that are consumed by less than 20 households are excluded from the analysis. 21 Table A.2.: Moderate Poverty and Ultra-Poverty by District Ultra- District Poverty Std.Error poverty Std.Error Chitipa 73.8 2.2 33.8 2.4 Karonga 57.1 2.5 22.7 2.1 Nkhatabay 57.7 2.5 16.3 1.9 Rumphi 53.6 2.5 17.3 1.9 Mzimba 42.9 2.5 16.8 1.9 Likoma 31.4 3.4 4.4 1.5 North Mzuzu City 9.7 1.5 2.0 0.7 Kasungu 53.0 2.6 14.5 1.8 Nkhotakota 53.4 2.6 25.1 2.2 Ntchisi 53.5 2.6 22.8 2.1 Dowa 48.8 2.6 15.6 1.9 Salima 58.4 2.5 26.6 2.3 Lilongwe 47.9 2.1 13.9 1.4 Mchinji 50.5 2.6 17.4 1.9 Dedza 63.1 2.5 25.6 2.2 Ntcheu 54.1 2.5 19.2 2.0 Central Lilongwe City 18.0 1.6 4.7 0.9 Mangochi 59.5 2.5 22.8 2.1 Machinga 72.4 2.3 28.5 2.3 Zomba Non-City 55.9 2.5 19.3 2.0 Chiradzulu 66.4 2.4 28.0 2.3 Blantyre 38.9 2.5 11.5 1.6 Mwanza 53.6 2.6 16.0 1.9 Thyolo 67.3 2.4 29.3 2.3 Mulanje 69.2 2.4 35.8 2.5 Phalombe 83.2 1.9 50.6 2.6 Chikwawa 63.2 2.5 34.6 2.4 Nsanje 74.3 2.2 37.0 2.5 Balaka 61.3 2.5 21.5 2.1 Neno 46.9 2.6 16.6 1.9 Zomba City 15.8 1.9 3.9 1.0 South Blantyre City 8.0 1.4 1.0 0.5 22 Appendix B: Adjustment for household composition As noted in Section 2.3, one of the key steps in the construction of the welfare indicator is moving from a measure of standard of living defined at the household level to a welfare indicator at the individual level as the ultimate objective is to make comparisons across individuals and not across households. Even if per capita consumption is frequently used, sometimes adult equivalence scales are used to convert household consumption into individual consumption by correcting for differences in the demographic composition as well as household size. The use of an adult equivalence scale accounts for the demographic composition of households as it allows us to rescale to reflect the fact that members might have different needs based on their age and gender. It is commonly assumed that children need only a certain fraction of adult consumption needs. As a result, during a comparison of two households that have the same total consumption and household size, but different composition of children and adults, the one which is composed mainly of children is thought to have higher welfare as the weights assigned for children is lower. However, there is no consensus on the appropriate weight applicable to children. Some are based on nutritional grounds, a child may need only a fraction of the food requirements of an adult, but it is not clear why the same scale should be carried over to nonfood items. It may very well be the case that the same child requires more in education expenses or clothing. As a sensitivity analysis, this report presents poverty estimates using incremental scale for children under the age of 16 and a weight of one for adults (Table B.1). Table B.1: Adult equivalent scale Age (years) Equivalence scale 0–1 0.33 1–2 0.47 2–3 0.55 3–5 0.63 5–7 0.73 7–10 0.79 10–12 0.84 12–14 0.91 14–16 0.97 >16 1.00 The comparison of welfare status based on adult equivalent and per capita consumption is conducted by investigating the correlation of these two welfare indicators and the implied poverty. 23 The correlation coefficient is 0.99 and hence the two welfare indicators are strongly correlated. These indicators also correctly classify the poor and the nonpoor individuals in the same way. For example, the comparison of 50 percent of the population ranked in terms of per adult equivalent consumption and another 50 percent of the population ranked in terms of per capita consumption shows that both indicators classify 95 percent of the individuals in the same way. In other words, these two welfare indicators correctly identify the same population as poor if a fixed poverty is chosen (Table B.2). Table B.2: Poverty using alternative consumption aggregates Per capita consumption Poor Nonpoor Total consumption Poor 47.4 2.6 50 equivalent Per adult Nonpoor 2.6 47.4 50 Total 50 50 100 24 Appendix C: Survey-to-survey imputation This appendix is redacted. Readers who are interested in this section could access it by contacting the NSO. Appendix D: Classification of individual consumption by purpose (COICOP) Table D.1: Classification of items by COICOP and the repetitive modules in IHS4 questionnaire Module (M), question COICOP (Q) and label/code Description code (L): in this sequence MQ-L 01 Food and nonalcoholic beverages 01.1 Food Cereals, tubers, nuts, vegetables, fruits, oil, sugar, and so on G02-101 to G02-818 01.2 Nonalcoholic beverages Tea; coffee; cocoa, Milo; squash; thobwa; fruit juice; freezes; soft G02-901 to G02-907, drinks; bottled water; maheu; and other G02-909 to G02-G910, G02-912, G02-916 02 Alcoholic beverages and tobacco 02.1 Alcoholic beverages Bottled or canned beer, traditional beer (masese), wine or G02-G908, G02-G911, commercial liquor, locally brewed liquor (kachasu), and chibuku G02-G913 to G02- (commercial traditional-style beer) G915 02.2 Tobacco Cigarettes or other tobacco I02-103 03 Clothing and footwear 03.1 Clothing Infant clothing J02-301 Baby nappies/diapers J02-302 Boy's trousers J02-303 Boy's shirts J02-304 Boy's jackets J02-305 Boy's undergarments J02-306 Boy's other clothing J02-307 Men's trousers J02-308 Men's shirts J02-309 Men's jackets J02-310 Men's undergarments J02-311 Men's other clothing J02-312 Girl's blouse/shirt J02-313 25 Module (M), question COICOP (Q) and label/code Description code (L): in this sequence MQ-L Girl's dress/skirt J02-314 Girl's undergarments J02-315 Girl's other clothing J02-316 Lady's blouse/shirt J02-317 Chitenje cloth J02-318 Lady's dress/skirt J02-319 Lady's undergarments J02-320 Lady's other clothing J02-321 Cloth, thread, other sewing material J02-326 Laundry, dry cleaning, tailoring fees J02-327 03.2 Footwear Boy's shoes J02-322 Men's shoes J02-323 Girl's shoes J02-324 Lady's shoes J02-325 04 Housing, water, electricity, gas and other fuels 04.1 Actual rents for housing Actual rent payment F04 04.2 Imputed rents for housing Estimated the rent for non-renters F03 04.4 Water supply Water for cooking, bathing, and so on F37 04.5 Electricity, gas and other fuels Value of the firewood used in the past week F18 Electricity F25 Charcoal I02-101 Paraffin or kerosene I02-102 Candles I02-104 Matches I02-105 Light bulbs I02-209 Solar panel L02-531 Generator L02-532 05 Furnishings, household equipment, and routine home maintenance 05.1 Furniture, furnishings, carpets, and other floor coverings House decorations J02-338 Carpet, rugs, drapes, curtains K02-401 Mat - sleeping or for drying maize flour K02-403 Mosquito net K02-404 26 Module (M), question COICOP (Q) and label/code Description code (L): in this sequence MQ-L Mattress K02-405 Bed L02-502 Table L02-503 Chair L02-504 05.2 Household textiles Linen - towels, sheets, blankets K02-402 05.3 Household appliances Repairs to household and personal items (radios, watches, and so I02-218 on) Fan L02-505 Air conditioner L02-506 Sewing machine L02-511 Kerosene/paraffin stove L02-512 Electric or gas stove; hot plate L02-513 Refrigerator L02-514 Washing machine L02-515 05.4 Glassware, tableware, and household utensils Bowls, glassware, plates, silverware, and so on J02-328 Cooking utensils (pots, stirring spoons, whisks, and so on) J02-329 05.5 Tools and equipment for home Batteries I02-220 Recharging batteries of cell phones I02-221 Torch/flashlight J02-331 Paraffin lamp (hurricane or pressure) J02-333 Mortar/pestle (mtondo) L02-501 05.6 Goods and services for routine home maintenance Milling fees, grain I02-201 Wages paid to servants I02-215 Cleaning utensils (brooms, brushes, and so on) J02-330 06 Health 06.1 Medical products, appliances, and equipment Expenditure for nonprescription medicines (Panadol, Fansidar, D12 cough syrup, and so on) 06.2 Out-patient services Expenditures for illnesses and injuries (medicine, tests, D10 consultation, and in-patient fees) Expenditure not related to an illness (preventative health care, pre- D11 natal visits, check-ups) Stay(s) at the traditional healer or faith healer D19 Stay(s) at the traditional healer or faith healer, transport costs D20 27 Module (M), question COICOP (Q) and label/code Description code (L): in this sequence MQ-L Stay(s) at the traditional healer or faith healer, food costs D21 06.3 Hospital services Hospitalization(s) or overnight stay(s) in a medical facility D14 Hospitalization(s) or overnight stay(s) in a medical facility, D15 transport costs Hospitalization(s) or overnight stay(s) in a medical facility, food D16 costs 07 Transport 07.1 Purchase of vehicles Bicycle L02-516 Motorcycle/scooter L02-517 Car L02-518 07.2 Operation of vehicles Petrol or diesel I02-212 Motor vehicle service, repair, or parts I02-213 Bicycle service, repair, or parts I02-214 07.3 Transport services Public transport - bicycle, taxi I02-107 Public transport - bus, minibus I02-108 Public transport - other I02-109 08 Communication 08.1 Postal services Postage stamps or other postal fees I02-210 08.3 Telephone and fax services Cell phone F35 09 Recreation and culture 09.1 Audio-visual, photographic and information processing equipment Music or video cassette or CD J02-336 Film, film processing, camera K02-407 Radio (wireless) L02-507 Tape or CD player; HiFi L02-508 Television L02-509 VCR L02-510 Computer equipment and accessories L02-529 Satellite dish L02-530 09.2 Durables for recreation and culture, including repairs Sports and hobby equipment, musical instruments, toys K02-406 09.3 Other recreational items and equipment, gardens and pets 28 Module (M), question COICOP (Q) and label/code Description code (L): in this sequence MQ-L Expenditures on pets I02-219 09.4 Recreational and cultural services Tickets for sports / entertainment events J02-337 09.5 Newspapers, books, stationery Newspapers or magazines I02-106 Stationery items (not for school) J02-334 Books (not for school) J02-335 10 Education 10.1 Education, all levels Tuition, including any extra tuition fees C22A After school programs and tutoring C22B School books and stationery C22C School uniform and clothing C22D Boarding school fees C22E Contributions for school building or maintenance C22F Transport C22G Parent association and other school related fees C22H Other school expenses C22I 11 Restaurants and hotels 11.1 Vendors, cafes, restaurants Vendor consumption: maize (boiled or roasted), chips, cassava, G820-G830 eggs, chicken, meat, fish, mandazi, samosa, meals eaten at restaurants, other 11.2 Accommodation services Night's lodging in rest house or hotel J02-339 12 Miscellaneous goods and services 12.1 Personal care Bar soap (body soap or clothes soap) I02-202 Clothes soap (powder) I02-203 Toothpaste, toothbrush I02-204 Toilet paper I02-205 Glycerine, Vaseline, skin creams I02-206 Other personal products (shampoo, razor blades, cosmetics, hair I02-207 products, and so on) 12.3 Personal effects Umbrella J02-332 29 References Christiaensen, L., P. Lanjouw, J. Luoto, and D. Stifel. 2012.“Small Area Estimation-based Prediction Methods to Track Poverty: Validation and Applications�. The Journal of Economic Inequality, 10(2) 267–297 Deaton, A., and A. Tarozzi. 2000. Prices and Poverty in India. Princeton University. Princeton, NJ Elbers, C., J. O. Lanjouw, and P. Lanjouw. 2002. “Micro-Level Estimation of Welfare.� World Bank Policy Research Working Paper Series 2911, Washington, DC. ________. 2003. “Micro-Level Estimation of Poverty and Inequality.� Econometrica 71 (1): 355–64. Foster, J., J. Geer, and E. Thorbecke. 1984. “Notes and Comments: A Class of Decomposable Poverty Measures.� Econometrica 52 (3) 761-766. Haughton, J., and S. Khandker. 2009. Handbook on Poverty +Inequality. World Bank. Washington, DC. James, G., D. Witten, T. Hastie, and R. Tibshirani. 2013. An Introduction to Statistical Learning with Applications in R. New York: Springer Science+Business Media. NSO (National Statistics Office). 2005. “Note on Construction of Expenditure Aggregate and Poverty Lines for IHS2.� National Statistics Office, Malawi Ravallion, M. 1998. “Poverty Lines in Theory and Practice�. LSMS Working Paper 133, World Bank, Washington, DC. Rubin, D. B. 1987. Multiple Imputation for Nonresponse in Surveys. New York: Wiley. Schafer, J. 1999. “Multiple Imputation: A Primer.� Statistical Methods in Medical Research 8: 3–15. Yoshida, N., R. Munoz, A. Skinner, C. Kyung-eun Lee, M. Brataj, W. Durbin, D. Sharma, and C. Wieser. 2015. “SWIFT Data Collection Guidelines Version 2.� World Bank Group, Washington, DC. https://hubs.worldbank.org/docs/imagebank/pages/docprofile.aspx?nodeid=24620058 30