Who Uses Electricity in Sub-Saharan Africa? Findings from Household Surveys

Analysis of household expenditure surveys since 2008 in 22 Sub-Saharan African countries shows that one-third of all people use electricity. As expected, users are disproportionately urban and rich. In communities with access to electricity, lack of affordability is the greatest barrier to household connection. Lifeline rates enabling the poor to use grid electricity vary in availability, with six countries allowing 30 kilowatt-hours or less of electricity usage a month at low prices. Affordability challenges are aggravated by sharing of meters by several households -- denying them access to lifeline rates -- and high connection costs in many countries, made worse by demands from utility staff for bribes in some countries. Collection of detailed information on residential schedules enabled calculation of the percentage of total household expenditures needed for electricity at the subsistence and other levels. Affordability varied across countries, with grid electricity even at the subsistence level being out of reach for the poor in half the countries and even more so once connection charges are considered. Examination of the gender of the head of household shows that female-headed households are not disadvantaged in electricity use once income and the place of residence (urban or rural) are taken into account. However, female-headed households tend to be poorer, making it all the more important to focus on helping the poor for the goal of achieving universal access. Installing individual meters and subsidizing installation, encouraging prepaid metering so as to avoid disconnection and reconnection charges, reformulating lifeline blocks and rates as appropriate, and stamping out corruption to eliminate bribe-taking can all help the poor.

Analysis of household expenditure surveys since 2008 in 22 Sub-Saharan African countries shows that one-third of all people use electricity. As expected, users are disproportionately urban and rich. In communities with access to electricity, lack of affordability is the greatest barrier to household connection. Lifeline rates enabling the poor to use grid electricity vary in availability, with six countries allowing 30 kilowatt-hours or less of electricity usage a month at low prices. Affordability challenges are aggravated by sharing of meters by several households-denying them access to lifeline rates-and high connection costs in many countries, made worse by demands from utility staff for bribes in some countries. Collection of detailed information on residential schedules enabled calculation of the percentage of total household expenditures needed for electricity at the subsistence and other levels. Affordability varied across countries, with grid electricity even at the subsistence level being out of reach for the poor in half the countries and even more so once connection charges are considered. Examination of the gender of the head of household shows that female-headed households are not disadvantaged in electricity use once income and the place of residence (urban or rural) are taken into account. However, female-headed households tend to be poorer, making it all the more important to focus on helping the poor for the goal of achieving universal access. Installing individual meters and subsidizing installation, encouraging prepaid metering so as to avoid disconnection and reconnection charges, reformulating lifeline blocks and rates as appropriate, and stamping out corruption to eliminate bribe-taking can all help the poor.

Introduction
Sub-Saharan Africa (Africa hereafter) lags behind all other regions of the world in household access 2 to electricity, per capita power consumption, and installed generation capacity. Africa accounted for 13 out of the top 20 countries in the world with the largest numbers of people without access to electricity identified in the Global Tracking Framework report published in 2015 (World Bank and IEA 2015). The report found that the percentage of people living in households with access to electricity in 2012 was about 35 percent. However, disparity between urban and rural areas was stark-69 percent of urban residents but only 15 percent of rural residents were estimated to have had access to electricity. The region with the second lowest access rate, South Asia, had more than double the percentage of people living with electricity (79 percent), and nearly five times the percentage share of rural residents with electricity (70 percent).
The global initiative, Sustainable Energy for All, brings diverse partners together to work toward universal access to electricity by 2030. Africa is likely to be the last region of the world to reach this target. Years of underspending, serious shortcomings in operational efficiency, underpricing, high costs of small-scale operation, over-reliance on expensive oil-based power generation, and the inability of many customers to pay for electricity services have contributed to much lower generation capacity per capita than in other regions. Existing customers face frequent power outages, and new connections in some countries have barely kept with population growth.
Benefits of electrifying households are many. Electric lighting is safer, more convenient, and far more efficient than kerosene lamps. Electricity enables households to use electronic forms of communication, and refrigeration makes food safer as well as reduces the frequency of grocery shopping and cooking. Studies analyzing data from household surveys have found positive effects of electrification on income (Kumar and Rauniyar 2011;Khandker et al. 2012;Khandker, Barnes, and Samead 2013;Van de Walle et al. 2013), employment (Dinkelman 2011; Khandker et al. 2012;Grogan and Sadanad 2013;Van de Walle et al. 2013), and education (Kumar and Rauniyar 2011;Khandker et al. 2012;Khandker, Barnes, and Samead 2013;Van de Walle et al. 2013).
In order to understand better why so many households in Africa do not have access to electricity despite its many benefits, some have studied possible constraints facing households. Komives et al. (2005) analyzed electricity access by quintile and by poverty status. The study covered about 40 countries, and predictably found that access was much lower for poor households. However, only three countries in Africa were included in the study, and most other countries had much higher rates of access. The study assessed the effectiveness of subsidies embedded within the tariff structure and the potential improvements that could be made by redesign of the tariff structure.
Komives et al. broke down access into two factors in assessing whether subsidies could benefit low-income households: the "community access rate" and the "connection rate" for those living in communities with 2 In international initiatives to achieve universal access, access has come to mean having electricity at home, and this paper uses the word in that sense. However, some literature has defined access to mean availability of electricity in the community and uses the term "coverage" to mean having electricity at home. The difference between percent coverage and percent access is the percentage of people or households who live in communities with electricity but have chosen not to have it at home (for example, have chosen not to connect to the grid), most usually for financial reasons. To the extent that anyone can have a generator at home, private generators are excluded from access in the latter sense. electricity. Together these factors defined the "access" rate. 3 For example, 88 percent of all households in Cabo Verde lived in communities that had connection to the grid, while 63 percent of these households were actually connected, leading to an overall access rate of 55 percent. The difference between community access and household access was much greater for poor households-72 percent of the poor had community access in Cabo Verde, but only 34 percent was actually connected, giving an access rate of 25 percent. The situations in São Tomé and Príncipe and Rwanda were similar-community access was only slightly lower for the poor, but household connection of those with community access was much lower. The distinction between "community access" and "household access" was shown to be important in the study by Van de Walle et al. (2013). This showed that non-connected households, when their neighbors were connected, may have also received benefits from the electrification. This conclusion suggested that the overall benefits from providing access could be greater than those accruing just to the households connected.
Briceño-Garmendia and Shkaratan (2011) analyzed spending on electricity by quintile in Africa. They found that the share of total household expenditure allocated to electricity was below 3 percent in most countries, and that the share was relatively stable across expenditure quintiles. The authors used this information to check whether households could afford to pay the full economic cost of consumption at the subsistence level. Affordability was defined as spending being in the range of 3-5 percent of total household expenditure, while consumption at the subsistence level was defined as 50 kilowatt-hours (kWh) per month. The study found that on existing tariffs consumption at the subsistence level is affordable for the large majority of those already connected. However, for those not already connected consumption at the subsistence level would be affordable at existing tariffs for only one-quarter of such households. Were tariffs to be raised to cost recovery levels, far fewer of the unconnected would be able to afford to purchase even consumption at the subsistence level. An even larger barrier to access is the existence of a connection charge. Golumbeanu and Barnes (2013) described the way in which connection charges can be so large that they are unaffordable for many low-income households. Evidence from a number of countries indicated that countries with higher connection charges, controlling for the level of gross domestic product (GDP) per capita, tended to have lower access.
Surveys of household access to electricity and its use have highlighted the issue that some households pay nothing or only part of the appropriate charge for the use of electricity. Smith (2004) pointed out that electricity theft could be in the form of fraud (meter tampering), stealing (illegal connections), billing irregularities, and unpaid bills. Calculations based on data from 102 countries showed that average transmission and distribution losses, which include losses from theft, had increased between 1980 and 2000 for all regions except Western Europe and North America. Antmann (2009) provided an analysis of methods of reducing technical and non-technical losses in the power sector drawn from experience in several countries. The use of pre-payment meters has been identified as one way of dealing with delinquent bill payment by households. Tewari and Shah (2003) describe the South African experience, starting in 1989, of the introduction of pre-paid meters and their benefits, costs, and problems. A study by Jack and Smith (2015) of purchasing patterns of electricity in South Africa through pre-payment meters found that the bottomquintile households tended to buy electricity three times as often but in amounts that are only one-fifth of those households in the top quintile. The pattern of small frequent transactions is consistent with the existence of liquidity constraints and difficulties in smoothing income, suggesting that the pay-as-you-go approach of pre-paid meters is preferable for low-income households to a monthly billing cycle. In Uganda bulk metering was introduced for micro and small enterprises in 2009 and pre-payment meters for households in 2012. An analysis of the bulk metering program by Never (2015) indicated that although this approach improved bill collection, there were still problems of theft and inability to pay. This paper is part of a broader study examining the financial viability and related aspects of the power sector in Africa. The broader study focuses primarily on grid electricity for reasons of data availability at the regional level. The study examines quasi-fiscal deficits-difference between the total cost of providing electricity and total cash collected by utilities-in 39 countries (Trimble et al. 2016), tariff structures in 39, and household use of electricity in 22. 4 The component on household use of electricity, which is the subject of this paper, takes household expenditure surveys to gain a better understanding of who uses electricity, what type of electricity, and how much. It asks the following questions:  How widely is electricity used?
 What are the barriers to making electricity the main source of energy in Africa for lighting and powering appliances?  Is electricity affordable?
 Is there potential evidence suggesting that female-headed households are disadvantaged in some ways with respect to electricity use?
Although household expenditure surveys capture spending on all forms of electricity, as with the broader study, detailed analysis is carried out primarily on issues associated with grid connection.

Scope and Methodology
The household expenditure surveys analyzed in this paper were carried out in 2008 or later. The survey period and the sample size are summarized in Table A.1 in the annex. This paper reports results calculated using household weights or population weights, depending on whether percentages of households or people are being examined. The only exception is when the sample size is discussed, primarily to point out that the number of households being interviewed was too small for meaningful results (for example, the number of rural households in the bottom 40 percent connected to the grid). In each country, people are divided into quintiles based on per capita expenditures, with quintile 1 being the poorest and quintile 5 being the richest. Per capita expenditures are derived from consumption aggregates used to construct official poverty measures. 5 Poverty measurements are based on real expenditures, either per capita or per adult equivalent, except in Botswana where nominal values are used. Each quintile has the same number of people, not households. Because household size typically declines with increasing wealth, and because the rich tend to be concentrated in urban areas, upper quintiles have more households than lower quintiles and are more urban than rural. The results are analyzed by consumption quintile, households' poverty status (poor or nonpoor based on the official poverty measures), location (urban or rural), and gender of the head of household. Where official poverty is based on per capita expenditure, the bottom quintile is always poor, but where poverty is based on expenditure per adult equivalent, the overlap between the poor and the bottom quintile is not complete. Households that had spent more than 30 percent of their total expenditures on electricity are considered outliners and dropped.
All questionnaires except the one in South Africa asked for the primary sources of energy for lighting and cooking; the South African survey asked if the household was connected to the grid, and the response to that question substitutes the question about the primary energy source for lighting in the rest of this paper.
Some surveys listed off-grid electricity separately (typically solar and backup generators), while some had only one category called electricity. "Solar energy"-the term used in Botswana, Burkina Faso, Ethiopia, Ghana, Senegal, Swaziland, and Uganda-could mean solar home systems only or could include solar lanterns. This lack of distinction is unfortunate because solar lamps would not fall under the category of electricity access, whereas solar home systems would. Mali, Mozambique, Rwanda, Sierra Leone, Tanzania, and Zambia asked specifically about solar panels. Seventeen surveys asked specifically about grid electricity, of which sixteen asked if the household was connected to the grid, and Togo asked about paying electricity bills over the previous two months (thereby excluding those connected to the grid but had not made payments during the recall period); several asked respondents if they had their own meters or were sharing meters with others.
Spending on electricity in this paper excludes expenditures on items not considered as part of access to electricity: batteries, candles, kerosene, and liquefied petroleum gas. While fuels for generators would be part of spending on electricity, only Rwanda and Zambia enabled isolation of spending on fuel for generators; all other countries bundled spending on gasoline and diesel for vehicles or lawn mowers with spending on either fuel for generators. For consistency across countries, spending on diesel used in generators in these countries was excluded from expenditure on electricity.
The surveys did not enable systematic analysis of access at different levels according to the recently formulated multi-tier access framework for the Sustainable Energy for All initiative. The framework defines energy access as the ability to obtain energy that is adequate, available when needed, reliable, affordable, legal, convenient, healthy, and safe for all required energy applications (World Bank 2015). For household connection to electricity, there are five tiers in the framework, with grid electricity corresponding to tiers 3-5 depending on appliances being used and kWh consumed per year and per day. Information on appliance use was not available in any survey, and plausible data on kWh consumed were available in only two to three surveys. In the face of these data limitations, this paper instead examined access at five different levels: 1. People living in households that reported connection to the grid 2. Adding to 1 people living in households that reported using electricity-excluding generators and solar energy if they are separately counted-as the primary source of energy for lighting, cooking, or both 3. Adding to 2 people living in households that reported using generators or diesel as the primary source of energy for lighting, cooking, or both, or reported owning generators 4. Adding to 3 people living households that reported using solar panels as the primary source of energy for lighting, cooking, or both, or reported owning solar panels; or solar energy in the place of solar panels if only information on solar energy was available 5. Adding to 4 people who reported non-zero expenditures on electricity.
Definition 2 is the same as definitions 3 and 4 in Malawi and São Tomé and Príncipe where the survey did not specifically asked about use of generators and solar panels or solar energy. Access to solar panels is preferred whenever information is available so as to avoid including people living in households using one or two solar lanterns. In countries where survey questionnaires asked only about solar energy (Botswana, Burkina Faso, Ethiopia, Ghana, Senegal, South Africa, Swaziland, Togo, and Uganda), definition 4 could well over-estimate the number of people with access, if households with solar lanterns had replied that they were using solar energy for lighting.
All surveys asked about spending on electricity, most commonly in the last 30 days. There are limitations with this approach. Payment arrears may be common, but no household surveys in this paper probed this point. By contrast, the 2012 National Survey of Household Income and Expenditures in Mexico asked a series of questions to understand arrears, including when the last payment for electricity had been made, while the 2005 Integrated Sample Household Budget and Labor Survey in the Kyrgyz Republic asked for kWh of electricity consumed, the amount billed, and the amount paid for three successive months as well as the amount of subsidy received. Most household surveys, however, simply ask how much the household paid last month for electricity. Asking just one question about how much the household paid over a fixed period of time could under-or over-estimate (the latter if past debts are being repaid to utilities) monthly expenditures on electricity. Cross-checking household survey data against data from the utilities could indicate the magnitude of these problems and ways of adjusting data for a more accurate picture. Lampietti and Junge (2006) combined billing and payment records from the utility and merged them with household survey data to address recall errors, under-and over-reporting, and the presence of arrears, which enabled more accurate estimation of current and historical electricity consumption as a function of household income and other characteristics. However, obtaining utility data for such cross-checking was outside the scope of the study.
The survey in Senegal asked if the household had spent anything on electricity over the last 12 months, which would more likely capture whether the household was paying something for electricity. If the question is about spending only in the last month only, no payment might have been made for a variety of reasons and the household would be registered as having zero expenditure. Similarly, some households did not answer the question (they might have found it difficult to recall the amount in the short time given to answer this question during the interview), and all missing responses were recorded as zero in the analysis. In all 22 countries, not all who cited electricity as the primary source of energy for lighting reported positive expenditures in electricity. Legitimate reasons for having no expenditures for grid electricity, the focus of this study, include electricity included as part of the rent, bundling of utility services (such as combined water and electricity bills), or free electricity being part of the compensation package for employment. It is also possible that the household happened not to have made payments during the recall period-for example, the household could have missed one bill payment, making up for it later.
Examination of spending on electricity was necessarily confined to those who reported positive expenditures on electricity. Among those who did, spending on electricity was computed as a share of total household expenditures as one measure of affordability. Because power tariffs in many countries are pan-territorial and do not have large regional differences that are observed with food and other items, nominal expenditures on electricity and nominal total household expenditures were used to compute expenditure shares, except in Angola and Mozambique where nominal expenditures were not available and where regionally adjusted expenditures were used.
Where enough information was available on how consumption aggregates were derived for official poverty statistics, expenditures on food were disaggregated into those paid for by cash and freely acquired food that had been assigned imputed values. The purpose of doing so was to estimate the total amount of cash available to pay for electricity. Imputed values are also assigned to items other than food, such as collected firewood. However, separating cash expenditure on food from imputed values was considered adequate for the purpose of assessing affordability of electricity among low-income households, because by far the greatest share of non-cash expenditures is food consumption, the share of which increases with decreasing income.
To examine the affordability of grid electricity, monthly consumption corresponding to tiers 3-5 in the multitier matrix for access to household electricity supply was taken-30 kWh, 100 kWh, and 250 kWh, respectively-and corresponding electricity bills were computed. The bills are inclusive of energy charges, fixed charges, other charges (such as a rural electrification fund fee), and applicable value-added tax (VAT) and any other tax. They do not include charges unrelated to electricity use (specifically public television and radio license fee) even if they are always added to every residential electricity bill. In addition, 50 kWh a month was considered for comparison with the AICD and because several countries set 50 kWh as the cap on highly subsidized lifeline rates.
The monthly electricity bills for these amounts were then expressed as shares of total household expenditures for all households, whether or not they were connected to the grid. The multi-tier access framework defines electricity as being affordable if households spend less than 5 percent of total expenditures on monthly consumption of 30 kWh. This paper similarly considers grid electricity to be unaffordable if a household has to spend more than 5 percent of its total expenditures on electricity. As part of the broader study, detailed information on the tariff schedule in effect in July 2014 was collected in 39 countries, including all the 22 countries with household survey data. In many countries, more than one schedule existed, and the least-cost option was taken for the purpose of examining affordability. If the survey was undertaken when a different tariff schedule was in effect, then monthly nominal per capita expenditures were increased at the same rate as nominal per capita GDP in local currency to the year when the tariffs prevailing in July 2014 first came into effect and total household expenditures were then computed from the adjusted per capita expenditure. If the tariff schedule in effect in July 2014 had been first introduced before the survey date, no adjustment of expenditure was necessary-Burkina Faso, Ethiopia, Mali, Mozambique, Senegal, Sierra Leone, and Togo fell under this category and did not require any adjustment to total household expenditures. Connection charges were available for all but Mali and São Tomé and Príncipe. They were computed as multiples of total monthly household expenditures. If information on when the connection charges came into effect was not available, the dates of effectiveness were assumed to be the same as those for tariff schedules (see Table A.8 in the annex for more detail).
This paper computes the poverty gap for grid electricity in the same way the poverty gap is defined in economics as follows: where the required monthly payment is the monthly bill inclusive of taxes and other charges that a household has to pay to consume the corresponding amount of electricity, P is the total population living in households for whom the monthly payment exceeds 5 percent of their total monthly household expenditures (inclusive of freely acquired food and other items), and N is the total population of the country. Where the monthly electricity bill exceeds the 5-percent share, electricity is deemed unaffordable, and the degree of unaffordability for a household is the size of the gap between the bill and the 5-percent share when this is positive, zero otherwise. Although monthly consumption of 30 kWh is the basis for defining affordability, the poverty gap is also computed for 50, 100 (multi-tier framework tier 4), and 250 kWh (multitier framework tier 5) a month to see how many people can afford higher consumption. For these calculations, the lowest-cost tariff schedule for each consumption level is taken where there are two or more possible schedules. This paper also takes the numerator in the above equation and aggregates the affordability gap (where it is positive) across all households using household weights. The sum is the amount of subsidy needed to enable every household to keep spending on electricity at or below 5 percent of its total household expenditure.
In addition to the poverty gap, this paper also computes the poverty headcount for grid electricity. The poverty gap takes account of the depth of poverty and is 100 percent only if every person has zero total household expenditure. The poverty headcount, by contrast, is simply the percentage of people who find consumption of a certain amount of electricity unaffordable according to the household-expenditure-share threshold of 5 percent. If every person has to spend 5.05 rather than 5.00 percent of total household expenditures to purchase electricity, the poverty headcount would be 100 percent but the poverty gap would be 1 percent, which is the difference between 5.05 and 5.00 divided by 5.00.
Lastly, in additional to analysis of supplementary questions related to grid electricity, the study also investigated differences in electricity use between female-and male-headed households. Simplified regression analysis was carried out country by country to see if, after accounting for total expenditures and location (urban or rural), female-headed households were any more likely to use electricity than maleheaded households, and whether spending on electricity showed differences. Probit regressions (for the first three below) and ordinary least squares (for the last) were carried out on the following dependent variables:  1/0 "expenditure dummy" for having positive expenditures on electricity (1 if positive, 0 if zero or missing)  1/0 "electricity dummy" for citing electricity of all forms, including generators and solar, as the primary source of energy for lighting or cooking (1 if electricity was used for lighting or cooking, 0 otherwise)  1/0 "grid dummy" for connection to the grid, or if that information was not available, for citing electricity excluding generators and solar as the primary source of energy for lighting or cooking where generators, solar, or both were separately listed  logarithm of expenditures ("log expenditure") on electricity for those households that reported positive expenditure, and in addition repeating the regression confining the sample only to those connected to the grid The following explanatory variables were tried and their statistical significance was tested using at 5-percent significance test 6 (that is, the probability that the coefficient for the independent variable is actually zero when the regression shows a non-zero value is less than 5 percent):  logarithm of total household expenditures  logarithm of per capita expenditure and logarithm of household size as an alternative to the above  1/0 dummy for female-and male-headed households (1 for female, 0 for male)  1/0 dummy for urban and rural (1 for urban, 0 for rural) The dummy for urban/rural may be viewed as a proxy for the level of infrastructure development, and more specifically a crude proxy for the availability of grid electricity. This variable was always statistically significant, raising the question of whether the independent variable of interest, female/male dummy, may be correlated with it. To examine this question, another probit was tested with the urban/rural dummy as the dependent variable and the logarithms of per capita expenditure and household size and the female/male dummy as explanatory variables. With the exception of Angola, the female/male dummy was always statistically significant. Therefore, the sample was split into urban and rural in every country, and regressions were run separately, eliminating the urban/rural dummy as an explanatory variable. 7 In every case where the coefficient for female/male dummy was statistically significant, the probability of increasing the dependent variable by switching from male-to female-headed households was computed. For that purpose, the values of the remaining two variables assumed the weighted averages in urban and rural areas, respectively. For ordinary least squares, which drops all households with zero expenditure on electricity and regresses the logarithm of spending on electricity on the three explanatory variables, the percent increase in spending on electricity (not logarithm) was computed, again setting the values of the two remaining variables at average values of the households in the sample.
For cross-regional comparison of household access to electricity, presented first, access rates in 1990 and 2012 in different regions from the 2015 Global Tracking Framework report (World Bank and IEA 2015) were compared as a function of two measures of the poverty gap ($3.10 and $1.90 per day per person at purchasing power parity in 2011 international dollars) and of logarithms of per capita GDP (at the market exchange rate in 2005 U.S. dollars and at purchasing power parity in 2011 international dollars). 2012 was selected because it is the last year for which access rates are available in the Global Tracking Framework report, three other years being 1990, 2000, and 2010. 1990 was selected as an alternative because the poverty gap in South Asia, the region with the second lowest rate of access to electricity, was close to that in Africa in 2012. Unlike GDP, the availability of the poverty gap data is sporadic, reducing the sample size by more than half. For access in 2012, corresponding poverty gap data were taken from 2012 whenever they were available, and otherwise from 2011 or 2013, and if the data were not available from any of these years, they were taken from 2010 (the poverty gap in 2014 was not available for any country). For 1990, in order to capture India, which had the poverty gap data in 1987 and 1993 but no other year in between, the range of years was extended to 1987-1992. For regression equations, in addition to access, per capita GDP, and the poverty gap, 1/0 regional dummies for Africa, South Asia, East Asia and Pacific, Latin American and the Caribbean, and "other" for the countries in the remaining regions were also used.
The next section entitled "Cross-Regional Perspective" is the only one that does not use data from recent household surveys in the 22 countries covered in the rest of the paper. It is also the only section that defines the poverty gap according to the poverty lines of $3.10 and $1.90 per person per day. Elsewhere, the poverty gap refers to the grid-electricity poverty gap defined in the equation above. To distinguish different measures of the poverty gap, "poverty gap" is followed by threshold values: $3.10, $1.90, 30 kWh, 50 kWh, 100 kWh, and 250 kWh.

Cross-Regional Perspective
Broadly, access expansion mirrors economic development. Does access in Africa as a function of income or depth of poverty fall in line with the rest of the world? To answer this question, access in different countries was compared as a function of the poverty gap at $3.10 and $1.90 and per capita GDP. Correlation coefficients between access and the two measures of the poverty gap and per capita GDP are statistically significant at 1 percent, with correlation coefficients ranging from -0.72 for the logarithm of GDP per capita at the market exchange rate to -0.89 for the poverty gap at $3.10. Figure 1 plots the relationship between access and the poverty headcount at $3.10 for 2012. There is a downward sloping line on which most countries in regions other than Africa lie, whereas several countries in Africa lie markedly below, as do Cambodia and Vanuatu. When access is regressed on any one of the measures of the poverty gap or per capita GDP together with 1/0 regional dummies, Africa is the only region for which the coefficient for the regional dummy is consistently negative and statistically significant at 1 percent. The predictive power of the poverty gap at $3.10 was the highest, followed by the poverty gap at $1.90, the logarithm of per capita GDP at purchasing power parity, and finally the logarithm of per capita GDP at the market exchange rate.

Descriptive Statistics
Household surveys in the 22 countries show that, on average, two-thirds of all people lived in rural areas (see Table A.2 in the annex). In terms of numbers of households, one-third of households lived in rural areas in South Africa, but in all other countries the percentage was 42 or higher. In 16 countries, the average per capita expenditure in every rural quintile (calculated using population weights) was lower than in the corresponding urban quintile, and in seven countries household expenditure in every rural quintile (calculated using household weights) was similarly lower. The median for the percentage of people who were officially classified poor across the 22 countries was 46 percent. In rural areas, the poor constituted a majority, with a median of 56 percent of all people and a weighted average of 50 percent. About one-fifth of all people lived in households headed by a woman. The median of the percentage of people living in femaleheaded households was 23 in urban areas, 20 in rural areas, and 21 percent of the total population. At the highest end of the spectrum, Botswana stands out with half of all people living in female-headed households; at the opposite end is Mali with only 4 percent reporting living in female-headed households. Across the 22 countries studied, the median monthly per capita expenditure in 2014 U.S. dollars 8 was $73 in urban areas and $37 in rural areas, giving a national median of $49 per capita per month.

Measuring Access in Household Surveys
What information about access do household surveys provide? How many surveys provide enough information about kWh of electricity consumed by each household-essential information for designing targeted subsidies? How do access rates differ by income, between urban and rural, between the poor and 8 Expenditures were adjusted for inflation to 2014 and converted to U.S. dollars using the exchange rate in 2014. non-poor? As explained on page 6, this paper uses five measures to calculate access. Access statistics by location, quintile, and poverty status using the most expansive definition (definition 5) are shown in Table 1; the statistics using the first definition (grid connection) are reported in Table A.3 in the annex. The results are consistent with the general observations made elsewhere that access rates in rural Africa are significantly lower than in urban areas, the poor are far less likely to have access to electricity than the rich, and in some countries access of the rural poor to electricity is essentially non-existent. Overall, the poor in nine out of 22 countries had an access rate of less than 5 percent. Source: World Bank staff analysis of household surveys a. The statistics for Togo are not for those who are connected to the grid because that question was not asked, but for those who reported positive expenditures on electricity utility bills.
Differences in access rates according to the metrics as defined on page 6 are shown in Figure 3. The difference in the rate of access between definition 1 and definition 5 represents largely people who use electricity other than grid electricity and people who failed to report grid connection (possibly those connected to neighbors' grid and did not consider themselves connected to the grid). The difference was striking in Mali, especially in rural areas and among the poor where the differences reached 50 percentage points. Two factors accounted for this difference. For rural residents, the most important cause of the difference was extensive use of solar panels. For the urban poor, it was non-zero expenditures on electricity by those who did not cite electricity as the main source of lighting or cooking. These expenditures were small in urban areas. However, in rural areas, spending on electricity by those who did not cite electricity as the main source of lighting or cooking, did not indicate connection to the grid, or did not own a generator or a solar panel was much higher when expressed as a share of total household expenditures, and significantly higher among the rural poor. It was not possible to infer from the questionnaire what explains this odd result. The second largest difference was found in Zambia, where the difference was due largely to use of generators in rural areas. Lastly, in Niger the difference in access between definitions 1 and 5 was larger than 10 percentage points for urban households as well as for the urban poor. The greatest contributing factor was use of electricity as the primary source of energy by those who did not report connection to the grid, followed by use of generators (even among the urban poor); the survey did not ask questions about solar energy. The difference also exceeded 10 percentage points in rural Nigeria and rural Ghana. In rural Nigeria, it was due to generator ownership: nearly one-fifth of all rural households own generators; the questionnaire did not ask about solar energy. In rural Zambia, the difference was due three-fifths to generator use and twofifths to solar panels. Only four surveys-Madagascar, Mozambique, São Tomé and Príncipe, and Togo-asked about quantities of electricity consumed. Recalling quantities of electricity consumed is even more challenging than recalling the amount spent, and the evidence of this challenge was clear in São Tomé and Príncipe, where 15 surveyed households reported electricity consumption of 1 kWh a year, 11 of them in the top quintile. The results were even more striking in Togo, where average monthly consumption was only 5 kWh, reaching only 6 kWh even in the top urban quintile. Such a consumption pattern seems more indicative of a lack of understanding of kWh as a unit of electricity than actual consumption. In addition, there was one country-Rwanda-that enabled back-calculation of kWh consumed from spending on electricity because the tariff schedule consisted of a single block with a unit charge applicable to all residential consumers and no fixed charges, and the survey asked about the last electricity bill. Taking the tariff schedule in effect at the time of the survey, kWh consumed was calculated. The results for those reporting non-zero quantities (and non-zero spending on grid electricity in the case of Rwanda) are presented in Figure 4 for all these countries other than Togo. Source: World Bank staff calculations using household survey data. Note: The survey in São Tomé and Príncipe had many households reporting exceptionally low monthly consumption, including those in the top quintile. They are reported to illustrate difficulties encountered in obtaining information on electricity consumption. Rwandan consumption is calculated from expenditures and the unit price in effect at the time of the survey.
The results show the predictable pattern of increasing consumption with increasing quintile, and significantly higher consumption by the top quintile in the first three countries where the results seem more plausible than in São Tomé and Príncipe or Togo. Rwanda, the only country in which kWh consumed was calculated from reported billed amounts, shows low consumption compared to Madagascar or Mozambique. In addition to seeming confusion about units for electricity consumption, another reason for caution in interpreting the results is the very small sample size in lower quintiles, especially in rural areas, leading to results that are not meaningful. For example, the bottom three quintiles in rural Mozambique had a total of only 11 households reporting kWh consumed. The bottom two quintiles in Rwanda similarly had only 7 and 12 households, respectively, connected to the grid with positive expenditures.

Spending on Electricity
The share of household expenditures spent on electricity among those with non-zero spending on electricity provides an indication of how much households are willing to spend, balancing affordability with the perceived value of electricity. The results are summarized in Table 2 for the share of household expenditures that include imputed values of freely acquired items, largely home-grown food. The corresponding results by quintile are given in Table A.4 in the annex. The median expenditure share is about 3 percent, and not markedly higher among the poor. The findings are consistent with households not spending much more than 5 percent of their total expenditures on electricity. However, there was considerable variation across the countries, ranging from a mere 0.3 percent in Malawi to 9.3 percent in Swaziland. Notable exceptions include the poor in Botswana and Swaziland. Urban households spent a larger share in 13 out of 22 countries. São Tomé and Príncipe Rwanda However, Sierra Leone stands out with the average rural expenditure share being twice as high. Among the poor, urban expenditure shares were higher in 15 countries, although the numbers of households in the sample are too small for meaningful results in many countries. Expenditure shares were higher for femaleheaded households in 19 out of 22 countries. Uganda was exceptional in that there were no households among the urban poor who had purchased electricity; 0.8 percent of the urban poor reported receiving free electricity. Because the amount of cash available is an important determinant of the household's ability to pay for electricity, Table 3 re-computes the shares by excluding these imputed values from household expenditures in the 18 countries where cash expenditures were computed. The corresponding results by quintile are given in Table A.5 in the annex. As an indication of how much rural households rely on freely acquired food, the urban expenditure share was larger in only 5 out of 13 countries (38 percent), in contrast to 59 percent found in Table 2. Sierra Leone and Uganda stand out with rural expenditure shares being about two-and-a-half times higher. Among the poor, in addition to these two countries, the rural expenditure share is three-and-ahalf times higher in Tanzania. In absolute terms, the expenditure share appears exceptionally high in rural Sierra Leone, Botswana, and Swaziland, but this is in part because of the small sample sizes reducing representativeness of the households reporting positive expenditures. The pattern with respect to the gender of the head of household remains the same, with female-headed households showing higher expenditure shares. When the shares in Table 2 and Table 3 are compared, they increase more in rural areas than in urban areas when imputed values are excluded from household expenditures. The differences also predictably decrease with increasing quintile. Both demonstrate greater reliance on home-grown food by the poor and rural households. Nationally, the largest increase was found in Nigeria (53 percent increase), followed by Ethiopia (21 percent), and Burkina Faso (18 percent). There was little difference between female-and male-headed households, indicating comparable dependence on home-grown food.

Charges for Grid Electricity Use and Connection
How expensive is it to buy electricity? Are tariffs designed to make electricity affordable to the poor? Do connection charges take the ability to pay into account, or is every residential customer charged the same?
This study collected information on residential tariffs for grid electricity in effect in July 2014 in 39 countries, including the 22 countries with household survey data. Table 4 summarizes tariff concessions available to those consuming little electricity in the 39 countries. The column labeled kWh shows the size of the lifeline block (the first block in a tariff schedule with a low price per kWh, intended to help the poor consume electricity) or, in the absence of an explicit lifeline block, the first block if there are two or more blocks. The column labeled "ratio" calculates the ratio of the unit energy charge (price per kWh) in the second block to that in the first block, or if there are two or more schedules of a single block each with increasing levels of installed capacity, the ratio of the second level of service to the schedule corresponding to the lowest unit energy charge. The unit energy charges used in computing the ratio include ad valorem taxes but exclude fixed charges, which are spread over the entire consumption and the unit values of which vary with consumption. If the first block is exempt from fixed charges and higher blocks are not, the ratio will be larger what the table shows, but if all blocks are subject to the same fixed charge, the ratio will be smaller. The comment column describes concessions offered to the lifeline rate or the first block, such as exemption from tax and fixed charges. The comment column also indicates whether increasing block tariffs apply in going from the first to the second block (different unit energy charges are applied to corresponding segments of consumption, with the unit price for the first block applying to kWh consumed up to the limit of the first block, the unit price for the second block applying to kWh above the limit of the first block up to the limit of the second block, and so on) or if volume-differentiated tariffs apply (a single unit charge, determined only by the total consumption, is applied to the entire consumption). Note: Names of countries with household survey data are shown in bold. "Cannot exceed so many kWh a month" means that the lifeline rate does not apply if monthly consumption exceeds the limit, and the entire consumption is charged a higher tariff, instead of increasing block tariffs. n.a. = not applicable. a. The monthly size of a tariff subject to a lifeline rate, or the size of the first block when there are two or more blocks. Some countries have multiple schedules, each of which has a single block. Senegal defines the size of the consumption blocks over two months. The first block is 150 kWh over two months, shown here as 75 kWh a month. b. The ratio of the effective energy charge, inclusive of ad valorem tax such as VAT but exclusive of fixed charges and taxes applied to fixed charges, between the second block (or the second level of service if there are several schedules of a single block each with increasing installed capacity). c. These numbers are for Johannesburg. Depending on eligibility, poor households are provided with 50, 100, or 150 kWh of free electricity a month in Johannesburg.
To the extent that subsidies for the poor are offered, all countries use some measure of consumption (kWh a month, amperage, or kilovolt-amperes) as a proxy for income. One exception is Johannesburg in South Africa, where the city calculates a poverty index for each household based on income and other indicators and offers three different levels of free electricity-currently 50, 100, and 150 kWh a month-depending on the poverty index.
The most common lifeline block size is 50 kWh (8 countries), followed by 25, 75, and 100 kWh (3 countries each). Eight countries have lifeline blocks up to 40 kWh, and five are 25 kWh or smaller. In these five countries, the lifeline block is not sufficient to meet the daily electricity need in tier 3 of the multi-tier access framework; all five except Benin have increasing block tariffs. Increasing block tariffs at least allow poor households to enjoy a large price subsidy for the first block. South Africa, represented by the tariff schedule in Johannesburg, may appear as having an exceptionally large first block, followed by The Gambia. However, many municipalities in South Africa offer free electricity to the poor-such as 25, 50, 60, 100, and 150 kWh a month-depending on eligibility criteria (for example, prepaid lifeline customers consuming less than 250 kWh a month based on a 12-month average can receive 60 kWh of free electricity a month in Cape Town, falling to 25 kWh for consumption of up to 450 kWh a month on average), which differ by municipality. An interesting case is Cameroon, which has increasing block tariffs but where the unit energy charge in the upper four of the five blocks are explicitly subsidized, the largest unit subsidy reserved for the block covering 801 to 2,000 kWh a month.
Subsidized lifeline rates are limited only to those consuming less than the cap in nine countries (Benin, Cameroon, Cabo Verde, Gabon, Ghana, Mozambique, Nigeria, São Tomé and Príncipe, and Togo), shifting households to tariffs for the next tier for the entire consumption if the cap is exceeded. This has the same effect as volume-differentiated tariffs in Gabon, Mozambique, and Nigeria. Not having access to the subsidy by exceeding the limit by even 1 kWh makes it more difficult for the poor when the size of the block size is relatively small (say less than 50 kWh). The median increase in the effective unit energy charge for consuming more than the limit on the first block is 68 percent. There is large variation across countries, however, ranging from 4 percent in The Gambia and 7 percent in Senegal to 340 percent in Madagascar to 450 percent in Kenya and Zimbabwe. In Mali, pre-paid customers do not lose access to the low rates for the first two blocks and tax exemption even if they consume more than 100 kWh, but post-paid customers lose both benefits if 100 kWh is exceeded. Table A.6 in the annex provides additional information, including the number of schedules and blocks and the type of tariffs (increasing block or volume-differentiated) in each of the 39 countries. Among the 22 countries with household survey data, only Rwanda had a single tariff schedule with one block and no fixed charges. All other countries had more than one schedule, more than one block, or both, making back-calculation of kWh consumed from spending on electricity not possible. 9 Increasing block tariffs were the most common tariff type, but there are several with volume-differentiated tariffs, of which three have volume-differentiated tariffs only between the first and second blocks, with increasing block tariffs above the second block. In countries with volume-differentiated tariffs, exceeding the lifeline block increases the unit cost of the lifeline volume by at least 47 percent (São Tomé and Príncipe), and as much as 268 percent in Nigeria, although many customers falling in the lifeline category are not being metered there. Eleven countries list separate schedules for prepaid customers. Table A.7 compares effective unit charges (US$/kWh) for consuming 30, 50, 100, and 250 kWh a month, inclusive of taxes and all fixed charges. Fixed charges punish low-consumption households. For example, unit tariffs (price per kWh) are higher for monthly consumption of 30 kWh than 50 kWh in 14 countries due solely to fixed charges. The median rises from US$0.12/kWh for 30 and 50 kWh to US$0.14 for 100 kWh and US$0.17 for 250 kWh, reflecting generally progressive tariff schedules. The table also shows the payment that needs to be made to connect to grid electricity for the first time. Some countries do not charge for connection or charge very little (US$2 in Cabo Verde, US$13 in Swaziland, and US$31 in Mauritius), but others have high fees, the highest of which is $680 in Botswana. Botswana gives an option of paying over 18, 60, or 180 months, but the first payment is the same irrespective of the payment period and is high at US$140. The household survey data show that about two-fifths of customers select 18 months (no interest charged), one-quarter settle the connection charge in a single payment, one-fifth pay over 60 months (prime interest rate minus 0.5 percent), and one-tenth pay over 180 months (prime interest rate). The percentage of households in each quintile selecting a particular payment plan is surprisingly independent of quintile for the one-off payment. The share is twice in the bottom two quintiles as in the top quintile for the 18-month option and about the same difference for the 60-month option. The largest difference is seen with the 180month option, selected by 13 percent of the households in the bottom quintile but only 5 percent of those in the top quintile. In Angola and Cabo Verde, the cost of reconnection after disconnection following nonpayment is much higher than the initial connection cost.
To the extent that subsidized connection is offered to the poor, consumption is used as a proxy for income. An example is South Africa, where connection is free for 20 amperage but there is a fee for 60 amperage. In Mauritania, there are six connection fees for residential customers depending on kilo-volt amperes subscribed, ranging from US$128 to US$950. There is one country that takes the ability to pay into account: The Gambia, where the connection fee is twice as high in the Greater Banjul area because of higher income in the capital than in the rest of the country. However, this segmentation does not target the poor specifically.

Affordability of Electricity
How affordable is electricity to households? More specifically, what proportion of households would not be able to afford the subsistence level of electricity, having to spend more than 5 percent of total household expenditures? Among those who cannot afford 30 kWh, what is the gap between the percentage of household expenditures and the affordability threshold of 5 percent? How big a subsidy would it take to make monthly consumption of 30 kWh affordable to every household in each country? How do the calculated subsidies compare to utilities' quasi-fiscal deficits? What if, instead of 30 kWh (tier 3), monthly consumption is 50, or 100 (tier 4), or 250 (tier 5) kWh? How well are the measures of "affordability" at different consumption levels correlated with access? This section probes these questions. Because it is much more costly to connect rural households to the grid, it is most unlikely that current tariffs can be maintained for large-scale grid electrification in rural areas without adversely affecting quasi-fiscal deficits. The results related to rural areas should therefore be treated with caution.
Grid-electricity affordability at the subsistence level of consumption Using tariff schedules from the utilities, this paper calculated hypothetical expenditure shares of grid electricity when consuming varying amounts, starting at 30 kWh a month, for all households, whether or not households were connected to the grid. Table 5 presents the results. On average, this share exceeds 5 percent in five out of 22 countries, but varies markedly both within and across countries. Electricity consumption at this level is least affordable in Madagascar, followed by Rwanda, Burkina Faso, Togo, and Sierra Leone. Barring provision of free electricity in South Africa, the amount of which varies by municipality and total monthly consumption, grid electricity is most affordable in Angola, with an exceptionally low lifeline rate, followed by Nigeria. 10 Electricity consumption at the subsistence level is not affordable to the poor on average in seven countries, nor to the bottom 20 percent in ten countries. Burkina Faso and Madagascar stand out for grid electricity being marginally unaffordable even to the top quintile. The result from Madagascar may at first seem inconsistent with Figure 4. However, large bill collection losses may explain in part why households in the top quintile reported monthly consumption of 70 kWh: the utility fails to collect 40 percent of the billed amounts (Trimble et al. 2016). Senegal is the only country where the electricity share of household expenditures is smaller for female-headed households; in all other countries, the share is the same or greater than for male-headed households. Lastly, the expenditure share is greater in every country in rural areas than in urban areas. Source: World Bank staff calculations based on utility information and household survey data. a These numbers are for Johannesburg. Depending on eligibility, poor households are provided with 50, 100, or 150 kWh of free electricity a month in Johannesburg. Table 6 presents results that are more easily compared to those of Briceño-Garmendia and Shkaratan (2011), except that, thanks to significant progress in energy efficiency improvement, power consumption at the subsistence level has declined from 50 kWh a month to 30 kWh in the intervening years. The table shows the hypothetical expenditure shares for those with and without actual expenditures on electricity as well as for those with and without access to the grid (definition 1 on page 6). It also calculates the percentage of households for which the expenditure share lies below 5 percent. The expenditure shares for those who reported non-zero expenditures on electricity lie close to the expenditure shares of those with access to the grid due to a considerable overlap between these groups. As expected, with the exception of Angola and Nigeria, a greater percentage of those not connected to the grid found electricity unaffordable than gridconnected households. Source: World Bank staff calculations based on utility information and household survey data. The definition of affordability in the multi-tier framework is based on the share of household income, arguably making cash expenditures more suitable as the basis for computing the expenditure share. Data on cash expenditures excluding home-grown food were available in 18 countries. The results are shown in Table  7. Source: World Bank staff calculations based on utility information and household survey data.
Predictably, electricity becomes markedly less affordable. The 5-percent threshold is exceeded for the poor in two-thirds of the countries (against one-third previously), and for four-fifths of the bottom quintile (against half previously). These results are consistent with much greater reliance on home-grown food by the poor than the rich. Table 8 reproduces Table 6 based on total household cash expenditures. In Burkina Faso and Rwanda, only about one out of every eight households not connected to the grid would find 30 kWh of electricity a month affordable, compared to one out of every four in Table 6. Out of the 15 countries for which full information was available, more than half of non-connected households would not find electricity affordable in six countries.

Affordability at higher levels of consumption
Corresponding results for 50, 100, and 250 kWh are shown in Table A.9- Table A.12, Table A.13- Table A.14, and Table A.15-Table A.16, respectively, in the annex. Although this paper uses monthly consumption of 30 kWh to define affordability, higher consumption levels were also tested to see how many households could "afford" (that is, pay no more than 5 percent of total household expenditures to purchase) more electricity. The number of countries in which electricity consumption becomes unaffordable (exceeding the 5-percent threshold) to the poor, based on total household expenditures (including imputed values), nearly doubles from 7 to 13 at 50 kWh. At 100 and 250 kWh a month, electricity is unaffordable to the poor in all but Angola and South Africa (where the poor are entitled to varying amounts of free electricity, depending on the municipality). The number of countries in which electricity becomes unaffordable to urban households doubles from three at 30 kWh to six at 50 kWh, and more than doubles to 14 at 100 kWh a month. Table A.10 in the annex enables a limited degree of comparison with the findings of Briceño-Garmendia and Shkaratan (2011). That study covered 18 countries, 14 of which are also in this study. Briceño-Garmendia and Shkaratan found that monthly consumption of 50 kWh would be affordable to about one-third of unconnected households on average, whereas this study found that the median for 50 kWh being affordable would be two-thirds of unconnected households and the average would be three-fifths. These results suggest that electricity tariffs became more affordable in the intervening years.

Grid-electricity poverty measures
The electricity poverty gaps for different levels of consumption are shown in Figure 5. In the figure, each rectangle for different monthly consumption is incremental. For example, the poverty gaps in Rwanda for 50, 100, and 250 kWh are 46, 67, and 85 percent, respectively. Table A.17 in the annex provides the results by location (urban and rural). Figure 6 contrasts the poverty gap and headcount. The highest poverty gap for 30 kWh is less than 30 percent, but the corresponding poverty headcount is in excess of 70 percent.  substantial relative to household expenditures, more than doubling the poverty gap in 17 out of 22 countries including, predictably, every country where the lifeline block is smaller than 50 kWh a month. Going from 50 kWh to 100 kWh and from 100 kWh to 250 kWh a month more than double the poverty gap in 16 countries each. A rapid rise in the poverty gap between 30 and 50 kWh raises concerns. At the opposite end of the spectrum, there is no compelling reason to make 250 kWh of monthly consumption so affordable in Angolawhich unfortunately could not be covered in the analysis of quasi-fiscal deficits for lack of data. Subsidy to make the subsistence level of electricity affordable to all Given limited resources, and in line with the affordability criterion for the multi-tier access framework, enabling households to consume 30 kWh a month would arguably be an appropriate goal for the foreseeable future in Africa. Table 9 lists the grid-electricity poverty gap and headcount for 30 kWh, the access rate (according to definition 1, or definition 2 in the absence of data for grid connection), the percentage of the total household expenditures needed to pay for 30 kWh of grid electricity a month, and the subsidies needed to make up the difference between the amount in excess of 5 percent of household expenditures and the monthly electricity bill, expressed as percentage of the quasi-fiscal deficit, total revenue collected by the utility, and GDP. These subsidy calculations assume a perfect world where precise targeting of individual households based on income is possible. The countries are listed in order of decreasing grid-electricity poverty gap. The calculation procedures and detailed results for quasi-fiscal deficits are available in Trimble et al. (2016).  The results show that the subsidies needed to enable all households to consume 30 kWh a month are generally not large. In seven countries, the total (perfectly targeted) annual subsidy is less than US$1 million. As a percentage of the cash collected by utilities, the highest is 6 percent in Madagascar and Rwanda. However, it should be noted that such assistance is unlikely to make sense for rural households-there are many rural households for whom connection to the grid network would be too costly for the utility, and they would be better served by off-grid electricity. A lower bound would be the total subsidy required to make grid electricity affordable to every urban household. Table A.19 in the annex provides additional results for annual subsidies for monthly consumption of 30, 50, and 100 kWh, splitting subsidies into urban and rural areas. The highest annual subsidy to make grid electricity affordable to all urban households is US$14 million in Burkina Faso, followed by US$10 million in Madagascar. The annual subsidy for this purpose is less than US$1 million in 15 countries. Given the relatively small levels of subsidies required, cross-subsidies-by better-off residential customers as well as commercial and industrial customers-are likely to be a reasonable source of funding.
Further analysis of the top ten countries in Table 9 is informative. Among them, only Togo, Uganda, and Niger do not suffer from underpricing, which is a measure of how much lower tariffs are compared to the tariffs required to eliminate quasi-fiscal deficits if utilities improve operational efficiency by reducing network losses, bill collection losses, and overstaffing (Trimble et al. 2016)-in other words, they have the "fiscal space," as it were, to cross-subsidize needy households. All other countries will continue to have underpricing as a problem even after attaining benchmark performance in operational efficiency. Aside from Sierra Leone, where underpricing is less than 0.1 percent of GDP, other countries have underpricing of at least 0.4 percent, and Zambia, Botswana, and Senegal have underpricing in excess of 1 percent of GDP, with 3.3 percent in Botswana even at benchmark performance being the highest. The current quasi-fiscal deficits are not small in the top three countries: 2.2 percent of GDP in Madagascar, 1.0 percent in Rwanda, and 1.1 percent in Burkina Faso. The presence of underpricing does not mean there is no scope for making the subsistence level of electricity consumption more affordable. Two examples of mitigation steps are to introduce a lifeline rate (Rwanda, Botswana) and consider volume-differentiated in the place of increasing block tariffs (Madagascar, Sierra Leone, Zambia, Senegal). But these observations point to the challenges facing the countries with high electricity poverty gaps and low access rates.

Relationships between access, poverty, and quasi-fiscal deficits
Which is the best predictor of access, grid-electricity poverty at 30 kWh? Do quasi-fiscal deficits tend to rise with increasing affordability of electricity? Or are deficits high precisely in countries where people are too poor to pay for electricity? To answer these questions, relationships between access, poverty gap, poverty headcount, and quasi-fiscal deficits at current performance as well as benchmark performance (eliminating operational inefficiencies) were examined. Poverty measures at 30, 100, and 250 kWh, and nine measures of quasi-fiscal deficits-deficit at current utility performance, that at benchmark performance (which is identical to underpricing, the degree to which tariffs fall short of cost recovery at efficient operation), and deficit due to operational inefficiencies, each as a percentage of GDP, bills sent out by the utilities, and cash collected by the utilities-were examined for this purpose.
Correlation coefficients between access (using definition 1 where the information is available, and definition 2 otherwise) and the two measures of grid-electricity poverty at different electricity consumption levels are all negative and statistically significant using a 5-percent significance test, with the exception of the poverty headcount at 250 kWh. The poverty gap at 250 kWh and the poverty headcount at 100 kWh have the highest correlation coefficients and remain statistically significant at 1 percent. Correspondingly, when access is regressed on different measures of grid-electricity poverty, the poverty gap at 250 kWh and headcount at 100 kWh, shown in Figure 7, are the two best predictors of access. However, although countries do not overlap, the correlation coefficients are not as high as between access and the economic poverty gap at $3.10 or $1.90. The share of total household expenditures required to consume different amounts of electricity is another predictor of access, but is not as good as the two measures of grid-electricity poverty. As mentioned above, because extending grid electricity to rural households is costly and off-grid electricity will be cheaper for low-density, low-consumption areas, and because the grid network is nonexistent in many parts of rural areas, these grid-electricity poverty statistics are arguably more relevant for urban households. Repeating the above exercise for urban households only delivers similar results, with the urban poverty gap at 250 kWh and urban poverty headcount at 100 kWh having the highest correlation coefficients with urban access rates. The highest correlation coefficients between urban access and the measures of grid-electricity poverty are obtained when the latter are calculated nationally, and different measures of national gridelectricity poverty are better correlated with urban access than with national access. There is no obvious explanation for these results.
The poverty gap and headcount at 30 kWh have the smallest correlation coefficients and have the weakest predictive power for access for both national and urban statistics. This may seem counterintuitive at first glance-should the affordability to the poor not determine the rate of access? Several reasons may be offered for this finding. First, the total household expenditures used include freely acquired food and other items, making cash available much smaller for some households, especially in rural areas. Yet excluding freely acquired goods and services is not necessarily a more compelling approach, because two households with the same cash income but one of which does not have to pay for food are clearly not in the same financial situation. Second, a household that can barely pay for 30 kWh a month of electricity may have other pressing demands on its budget. Third, connection costs are high in many countries (see the next paragraph), and a household that can just afford 30 kWh a month may have no financial resources to pay for the initial connection. Lastly, depending on how many higher-consumption, better-off consumers are in the same area, it may not be financially viable for the utility to extend the grid to poor households for whom even 30 kWh a month poses an affordability challenge. By contrast, correlation coefficients between access with different measures of quasi-fiscal deficits are not statistically significant at 5 percent as long as South Africa (which has the highest rate of access and high quasi-fiscal deficits) is omitted. This finding would suggest that access expansion is not an important driver of quasi-fiscal deficits. The poverty gap and headcount at 30 kWh are not correlated with any of the nine measures of quasi-fiscal deficits using a 5-percent significance test, and quasi-fiscal deficits due to operational inefficiencies are not correlated with any measure of poverty at any monthly consumption level. The highest correlation coefficients are for the poverty headcount at 250 kWh for quasi-fiscal deficits at benchmark performance. Using a 1-percent significance test, the correlations that remain are those between the three measures of quasi-fiscal deficits at benchmark with the poverty headcount at 250 kWh and the poverty gap at 250 kWh. All coefficients are negative. This finding seems reasonable: the larger the underpricing-that is, the more subsidized the tariffs compared to cost-recovery levels at efficient operation, the lower the grid-electricity poverty.

Affordability of connection costs
For the poor, compounding the challenge of paying for minimal monthly consumption of electricity is the upfront cost of getting connected to the grid. Mozambique and South Africa offer free connection to those who consume little electricity, and Senegal offers free connection to those living within 40 meters of the grid. Nigeria has prohibited connection fees, although customers pay for the materials needed for the connection. Table 10 summarizes the lowest costs charged for connecting to the grid as multiples of monthly household expenditures. The costs shown typically do not apply to cases where the distribution line needs to be extended, requiring the household to be located close to the distribution network. Table 11 recalculates the same costs as multiples of monthly household cash expenditures. Source: World Bank staff calculations based on household survey data. Note: n.a. = information on connection costs not available. a. The connection payment is the first payment in an installment plan (same first payment irrespective of the number of months over which payments are spread). A one-off payment is 4.8 times higher. b There is no connection fee but customers buy the materials needed and present them to the utility. There is supposed to be no exchange of money between the customer and the utility. c. The connection payment is the first payment in an 18-month installment plan. A one-off payment is twice as high.   The countries in which monthly consumption of 30 kWh is already unaffordable to the poor tend to have high connection charges. The correlation coefficient between the expenditure share of monthly electricity consumption and multiples of monthly household expenditures for the lowest connection costs is 0.58, statistically significant at 5 percent. Botswana and Sierra Leone offer installment plans to make it easier to pay for the upfront cost of connection. However, the costs remain relatively high, and the subsequent monthly payments make electricity even at the subsistence level too expensive for the poor. In Botswana, the total payment even when payments are spread over 180 months make the monthly bill for 30 kWh 28 percent of monthly cash expenditures and the corresponding percentage is 19 in Sierra Leone. These would be unaffordable by any measure.
It is informative to compare the share of total expenditures required for the initial connection between those with and without positive expenditures on electricity or grid connection. The results are tabulated in Table  12. It is clear that those not connected to the grid are much more income-constrained than those connected to the grid. Corruption in the form of bribes extracted add to the cost of using electricity. Surveys in Malawi and Nigeria asked households connected to the grid if they had to pay an "informal fee" over and above the official connection charge to get the connection (free since August 2012 in Nigeria), while the survey in Senegal asked about the general perception of corruption in the government and specifically about the power sector.
The results for Malawi and Nigeria are shown in Figure 8. Bribery was more prevalent in Nigeria. In Malawi, there were too few households in the bottom quintiles for meaningful statistics, but in Nigeria every quintile had enough households sampled. The percentage of households reporting informal payments was lowest in the bottom quintile and highest for the top quintile-that is, those who are most able to pay bribes are also most likely to be asked and pay-although there was no consistent pattern with increasing quintile.
Recognizing the pervasive nature of this problem, the Nigerian Electricity Regulatory Commission in August 2012 prohibited connection fees and required that new customers be given a list of materials to purchase and provide the utilities with the materials, so as to avoid a "direct exchange" of money between the customer and the utility. The only variable is the length of the connection wire. After the materials are presented and checked by the utility, the utility has 48 hours in which to connect the new customer (NERC 2012). The regulator announced that this new regulation was intended to "rid the power sector of corruption and protect electricity consumers from fraud and extortion often committed against them by electricity companies." As one potential indication of progress in this area, the percentage of households reporting informal payments declined from those in the 2010 survey by 3.8 percentage points nationally, 2.0 percentage points in urban areas, and 5.8 percentage points in rural areas. The decline was primarily among lower-decile households, while the top decile statistics remained largely the same. In Senegal, 86 percent of households said that corruption had increased over time. However, only 1 percent of those connected to the grid said that they had been asked by utility officials to pay bribes. The question asked in Senegal is not comparable to that in Malawi and Nigeria. It is possible that this more positive outcome in Senegal was due in part to the fact that the question concerned only the past 12 months, and if bribes are associated primarily with the initial connection, those households that had obtained grid connection more than 12 months earlier would not have reported bribe payments.

Community versus Household Access
In Malawi, Niger, and Nigeria, the survey asked whether their communities had electricity. 11 The data can be used to examine electrified communities and compute the percentage of households that have chosen to connect to the grid. In urban Nigeria, 92 percent of all households living in electrified communities had connected to the grid, whereas less than half of such urban households in Niger and Malawi had grid electricity. In rural areas, the percentages were much lower in the latter two countries, as shown in Figure 9. Predictably, the percentage increased with quintile, but remained one-third or lower in rural Malawi and Niger.
Figure 10 examines households without grid connection. Most urban households without grid connection in Malawi and Niger lived in electrified neighborhoods, whereas a lower percentage in urban Nigeria lived in neighborhoods connected to the grid, pointing to lack of community electrification as an important barrier. Most rural households without grid connection lived in communities without grid electricity. 11 In Malawi and Nigeria, the survey asked each household, whereas in Niger the survey asked this question in the community questionnaire. In Ghana the survey also asked in the community questionnaire, but the data were not available to the World Bank. In Senegal the survey asked if the community had been newly electrified over the previous five years, but that would not identify all households living in communities with electricity.  For households living in communities connected to the grid, in Malawi and Nigeria the survey asked households without grid electricity why they had not connected (Table 13). The reasons were specified in the questionnaire and households selected from a range of options. Not surprisingly, unaffordability of the connection charge was the most common reason. In Malawi, the payment for the initial connection to the grid is large, costing about US$160 even when no grid extension is required and about US$450 if one-pole extension is required. The second most common reason in urban Nigeria was poor reliability of grid electricity; all but one of those respondents who cited poor reliability had a private generator.  -Source: World Bank staff calculations using household survey data. Note: No need = no need for electricity; dwelling unsuited = dwelling unsuited for connection to the grid; pending = application for connection pending; disconnected = disconnected for non-payment; -= reason not among the options cited in the questionnaire.

No spending on grid electricity
Survey questionnaires are generally not sufficiently detailed to quantify how many households do not pay for electricity. The data collected should ideally be able to separate the absence of reported household spending on electricity for legitimate reasons-such as inclusion of electricity in the rent, or in the compensation package for employment-from evasion of payment, such as theft of electricity or non-payment without disconnection. No household survey enabled probing of this question at that level of detail.
 In Mali, Niger, and Zambia, the survey asked if the rent included electricity. In South Africa, the survey asked if rent or free rental included "electricity, water, etc." without isolating electricity. In South Africa, the survey also asked how much the household had spent on utilities if water and electricity bills were bundled.  In Mali and Niger, the survey asked if the household received free "water-electricity" as part of employment compensation, without isolating electricity.  In South Africa, the survey asked if the household received free electricity and its value. Lowconsumption or poor households (depending on the municipality) can obtain certain amounts free of charge and pay for the rest. Therefore, answering yes to the question does not mean no payment to the utility was required. It is also possible that some households would report stolen electricity as free electricity. In Uganda and Zambia, the survey did not ask about free electricity, but the household was asked to value electricity received without making payment. Neither survey asked why free electricity was available.  In Sierra Leone, the survey asked if the household had paid electricity bills (either bills addressed to the household or shared with others) over the past 12 months.
Asking if the household had made payment at any point over a long time period, such as 12 months in Sierra Leone, is far more likely to catch payments than asking about the past 30 days, when the household may have missed a payment, pre-paid enough in advance, or pay less frequently than once a month. Households with missing values (missing answers) may not have answered the question despite having paid (for example, the respondent may have had trouble recalling the amount paid), but all missing values were assigned zero for the purpose of this study.
It is important to note that national household expenditure survey data cannot be used to estimate bill collection losses (percent of the total billed amount not collected by the utility) from residential consumers because no survey asks what should have been paid. It is not possible to tell from non-zero expenditure whether the payment was made in full, or in part, or whether it was to pay for arrears. Nevertheless, a large percentage of grid-connected households reporting zero expenditures could signal many households not paying fully, if at all, for electricity.
The results, summarized in Figure 11, show that Angola, Botswana, Ethiopia, Malawi, South Africa, Uganda, and Zambia had one-fifth or more of households connected to the grid reporting zero expenditures on electricity, with Angola (46 percent), Zambia (38 percent), and Botswana (31 percent) leading. By contrast, Nigeria-where bill collection losses have been documented to be very high-significantly under-reported non-payment, showing a non-payment rate of only 4 percent. The survey in Senegal, which asked for expenditures over the last 12 months, found no households connected to the grid not having made payments. The results by quintile are shown in Table A.20 in the annex. Generally, the share of gridconnected households reporting zero expenditures declined with quintile, but Ethiopia and Uganda showed the opposite trend, with the top quintile having the highest proportion of grid-connected households reporting zero expenditures. There are several potential explanations. The rich may be more likely to get jobs with generous benefits, including free electricity. They may be more likely to rent houses with electricity included in the rent. Or they are able to manipulate meters or bribe utility employees to avoid bill payment, in which case they should be targeted for reducing commercial and bill collection losses. Among the countries with one-fifth or more of grid-connected households reporting zero expenditures, only South Africa and Zambia enable further analysis. Of the two, only Zambia isolates electricity for inclusion in rent, whereas South Africa includes water in addition to electricity, making disaggregation of electricity from other items not possible. If every appearance of combined electricity and water bills paid is assumed always to include electricity, and if everyone receiving free electricity is considered to rely only on free electricity, which is unlikely-the percentage of households receiving free electricity increases from 2 percent in the bottom two quintiles to 3 percent in the top three quintiles-the share of households not paying for grid electricity without justification declines from 29, 11, and 24 percent for urban, rural, and national to 2 percent each in South Africa. The reduction is primarily from inclusion of electricity in rent or electricity and water bills being bundled.
In Zambia, the number of grid-connected households who had zero expenditures on electricity but whose rent included electricity was significant in urban areas. The share of non-paying households decreases from Rural Total 33, 65, and 38 percent for urban, rural, and national to 17, 59, and 23 percent, respectively, once inclusion of electricity in rent is taken into account.

Multiple Connections
Progressive tariffs to protect the poor would be effective only if each household is individually metered. Unfortunately, shared metering is all too common in Africa. Multiply connected households cannot take full advantage of lifeline rates, and end up paying higher effective unit prices charged to the bulk of the consumption (see Table 4 for more detail). Four surveys asked questions from which the prevalence of multiple connections can be estimated:  In Côte d'Ivoire, the survey asked if the household had its own meter, or shared a meter with others, or had no meter.  In Ethiopia, the survey asked if the household had its own meter or shared a meter.
 In Sierra Leone, the survey asked if the household shared an electricity bill with others.
 In Senegal, the survey asked the household to report the amount paid to the neighbor over the last 12 months if the household did not have a contract with the utility and was using its neighbor's electricity.
The results show that multiple connections are widespread in the first three countries. The share of households connected to the grid that had their own meters in Côte d'Ivoire and Ethiopia or did not share a utility bill with others in Sierra Leone was less than half, 42 percent, 40 percent, and 44 percent, respectively. In Côte d'Ivoire, this percentage did not vary much by quintile, but in Sierra Leone the share increased from the bottom quintile (26 percent) to the top quintile (46 percent), signaling that shared metering was dominant among low-income households. By contrast, in Ethiopia, the share fell markedly in the top quintile, meaning shared meters were, unexpectedly, most common among the top quintile. To the extent that progressive tariffs are intended to help the poor, shared meters among the richest 20 percent do not defeat the purpose of progressive tariff design.
Perhaps not surprisingly, in Senegal-which does not charge the connection fee if a household is within 40 meters of the nearest power line-a much lower percentage reported not having a contract with the utility and paying their neighbors instead. The actual percentage of households with shared meters depends on the assumption about how many households are connected to a single meter on average, with the highest percentage of shared meters corresponding to only two households per meter. If only two households are assumed to be connected to the same meter on average, then 84 percent of households would have had their own meters. This percentage was lowest in the bottom quintile, 53 percent, and highest in the top two quintiles, 90-91 percent. If three households are connected to a meter on average, the percentage increases to 88 percent of all households having their own individual meters.
If there is only one distribution company and the number of residential consumers is separately reported by that company, comparison of the number of households connected to the grid found in the survey and the number of residential customers reported by the utility gives another indication of the prevalence of multiple connections. Such comparison is possible only in Malawi, where the utility data are available for the survey year. If every shared meter is assumed to be connected to two households, the percentage of grid-connected households with their own unshared meters is 57 percent. If more than two households are connected, this percentage increases, for example to 68 percent if three households on average are connected to a single shared meter.
In Mali, the monopoly distribution company reports the total number of low-voltage (but not residential) customers in the survey year. The number of households reporting connection to the grid exceeds the number of low-voltage customers by 30 percent. If all voltage customers are assumed to be residentialwhich they are not-and an average of two households are assumed to be connected to a shared meter, the percentage of households with their own unshared meter is calculated to be 54 percent. The percentage increases to 65 percent if the assumption is changed to three households for every shared meter on average.
Lastly, in Ethiopia, the monopoly distribution company reports the total number of residential customers in the year before the survey year. The number is 1.65 million customers in 2012, against 1.63 million households that reported having their own private meters out of a total of 4.2 million households reporting connection to the grid in 2013. These numbers strongly suggest that many households that replied that they had private meters were actually sharing their meters with their neighbors, with the distinction understood by survey respondents to be between having meters installed on their premises (private meters) versus connected to neighbors' meters (shared meters). This in turn would mean that the percentage of households sharing meters was much greater in Ethiopia than the survey finding of 60 percent.

Outages
Seven surveys-Ethiopia, Madagascar, Malawi, Mali, Niger, Nigeria, and Senegal-asked questions about power outages. Because questions asked varied from country to country, it is not possible to compare them on a common basis. The results are summarized in Table 14 for households connected to the grid who replied to the questions on the frequency, duration, or both of blackouts. Power outages were common in all the countries, and serious in Malawi (six out of every seven households reporting daily blackouts over the last 12 months), Niger, Nigeria, and Senegal. 55% of urban and 59% of rural households reported daily blackouts.

Senegal
Over the last 4 weeks, households connected to the grid reported blackouts 15 days on average, or 53% of the time.
The number of days with outages decreased slightly from the second to the top quintile.
The average number of days with blackouts over the last 4 weeks was 14 in urban and 18 in rural areas.
Source: World Bank staff analysis of household surveys Note: All percentages are percentages of households connected to the grid who replied to the questions on blackouts.

Gender of Head of Household
Is there evidence that female-headed households tend to be poorer? If they are, examining the relationship between the gender of the head of household and access could be misleading, confusing poverty with gender. Are female-headed households less likely to have access to electricity, or buy less electricity, than their male-counterparts, after accounting for per capita and household expenditures?
Female-headed households are over-represented in lower quintiles in Botswana, Malawi, Rwanda, South Africa, and Swaziland. The worst is South Africa, where only 10 percent of all people living in female-headed households is in the top quintile. Conversely, female-headed households are over-represented in upper quintiles in Burkina Faso, Ethiopia, Mali, Niger, and Senegal. In Mali, only 3 percent of all people living in female-headed households was in the bottom quintile and 40 percent in the top quintile. Generally, per capita expenditures in each quintile were comparable between female-and male-headed households. However, with the exception of South Africa and Swaziland, female-headed households were considerably smaller, making their household expenditures smaller than their male counterparts.
In the regression analysis, the logarithms of per capita expenditure and household size as independent variables consistently gave a better fit than the logarithm of total household expenditure. Probit regression of the urban/rural dummy on the female/male dummy and the above two logarithms found that the gender of the head of household was insignificant only in Angola, and was negative only in Malawi and Swaziland. Being female-headed increased the probability of residing in urban areas by as much as 23 percent in Senegal (see Table A.21 in the annex); the median increase was 7 percent. This may be in part because rural areas tend to be more traditional, making it more difficult for women to head households.
Probit regression of the female/male dummy on three dependent variables-urban/rural dummy, logarithm of per capita household expenditure, and logarithm of household size-showed that the coefficients for both per capita expenditure and household size were significant and negative, except in Senegal where femaleheaded households had a positive significant coefficient for per capita expenditure and Botswana where the coefficient for household size was positive. (The coefficient for expenditure was insignificant in Mali and Sierra Leonne and that for household size was insignificant in South Africa and Swaziland.) The coefficient for the urban/rural dummy was usually statistically significant and positive. While the equation specification tested was overly simplified, these results suggest that, after accounting for the location (urban/rural), female-headed households may have tended to be poorer on average in most countries.
Statistically significant results found in the regression analysis relating the gender of the head of household to electricity use are shown in Table 15. For probit equations, the results show the difference in the probability between female-and male-headed households that the dependent variable (for example, electrification) would be 1 at mean values of per capita expenditure and household size. For ordinary least squares, the table shows the percentage by which spending on electricity increases, again at mean values of per capita expenditure and household size. In total, the gender dummy was statistically significant in at least one equation in 15 countries. The coefficients for probit equations were negative in Ghana (all urban), Sierra Leone (rural expenditure and rural grid dummies), Togo (all urban), and Zambia (rural grid), and positive elsewhere. While the gender dummy was negative for rural households in Sierra Leone, these were the only equations in which the logarithm of per capita expenditure was not statistically significant, casting doubt on the results. Among those who reported positive expenditures on electricity, the gender of the head of household affected the amount spent only in four countries, with negative coefficients in Ghana and positive elsewhere. On balance, female-headed households did not differ from male-headed households once per capita expenditure and household size were taken into account, and if anything evidence pointed to femaleheaded households being more likely to use electricity, with urban Ghana and urban Togo being notable exceptions. However, when the logarithm of expenditure on electricity is examined among grid-connected households only, female-headed households had spent considerably less in rural São Tomé and Príncipe and Sierra Leone, as shown in Table A.22 in the annex, although the opposite was observed in rural Angola.
Source: World Bank staff calculations. a. 1 if spending on electricity is positive, zero otherwise. b. 1 for electricity (grid as well as off-grid, such as solar and diesel generators) as the primary source of energy for lighting or cooking, 0 otherwise. c. 1 for households connected to the grid or electricity (excluding off-grid, such as solar and diesel generators) as the primary source of energy for lighting or cooking, 0 otherwise; see footnotes d and e and associated results in Table A.21 for more detail. d. Logarithm of spending on electricity retaining all households with positive expenditure. The results for the dummy represent the increase in the probability of the dependent variable being 1 as result of changing the gender dummy from 0 (male) to 1 (female). The results for log expenditure on electricity represent the percent increase in expenditure on electricity (not logarithm) as a result of changing the gender dummy from 0 to 1. -= statistically insignificant at 5; n.d. = no distinction made for different sources of electricity; n.a. = not available (data not collected); NA = not applicable.
The surveys in Malawi, Niger, and Nigeria enabled isolation of households who lived in communities connected to the grid. The probit regression for grid connection for households that could have connected was performed by taking these households as a single group as well as by separating urban from rural households. Statistically significant results were obtained when urban and rural households were combined only in Niger-female-headed households were more likely to connect-and separating urban and rural households similarly returned statistically significant results only in urban Niger-again female-headed households were more likely to connect.
Given the data limitations, these findings should be viewed only as preliminary, calling for more detailed investigation. A recent detailed analysis of the gender of the head of household in Africa since the 1990s suggests that the share of female-headed households has been rising, while that poverty has been declining faster among female-headed households than among their male counterparts (Milazzo and van de Walle 2015). These changing dynamics can inform future research and affect policy direction.

Conclusions
It is worth highlighting the limitations of this study-in scope and data availability-before summarizing the key findings and possible policy implications:  With respect to household access, there is much less information on the supply side compared to the demand side. Connection charges tend to increase rapidly with increasing distance from the power line, and yet information on the distance between each household and the nearest distribution line is not available. Grid connection may not be a practical option even for many living in electrified communities.  Information on income, let alone disposable cash income, is not available and is substituted by household expenditures used to calculate official poverty statistics. Those using kerosene for lighting have some cash-based expenditures to redirect to electricity purchase, but no survey separates spending on kerosene for lighting from that for cooking. Trying to pinpoint determinants of access beyond affordability based on national household expenditure surveys requires more data collection and analysis in individual countries.  It would not be appropriate to extrapolate the conclusions and recommendations in this paper to offgrid electricity where different factors will be at play. There are many areas in Africa-areas with low population density and low household expenditures, and far from cities and towns-where it would be too costly to supply grid electricity. Where the unit cost of grid electricity supply markedly exceeds what would be considered reasonable to pay for grid electricity, off-grid solutions are likely to offer more practical solutions. This paper confirms many widely accepted observations, such as that the percentage of population using electricity remains low in Africa, that electricity use is far more prevalent in urban than in rural areas, and that adoption of electricity and quantities consumed rise sharply with income. There are also relatively new findings. The rate of household access to grid electricity is only weakly correlated with affordability of tariffs applicable to the subsistence level of electricity consumption. There is little evidence of a trade-off between utilities' deficits and household access, but there is evidence of trade-offs between affordability at relatively high residential consumption levels and underpricing. There is no marked difference between female-and male-headed households with respect to access to electricity and spending on electricity, once income (per capita as well as household, captured through the combination of per capita expenditures and household size) and the place of residence are accounted for. The remainder of this section is largely about grid electricity.
The findings of this study suggest several measures that can enable use of electricity by more households:  Avoid multiple connections. They substantially reduce, if not eliminate altogether, the intended benefits of the subsidies provided by progressive tariffs. Shared connections also degrade the quality of service to those so connected because the combined demand of the households can easily exceed the installed capacity.  Making connection charges affordable is one of the top priorities in a number of countries. The difference in the share of household expenditures needed to pay for the initial connection between those with and without grid connection is stark, with medians differing by a factor of two when total household expenditures are considered and a factor of three when only cash expenditures are considered. In some countries, connection charges are beyond the reach of low-income households, arguably an important reason for multiple connections. Easing payment terms for new connection and building connection charges in general tariffs of all-and especially high-consumption consumers in all sectors-would help. In easing payment terms, attention should be paid to balancing tariffs and monthly installments for connection charges, because the combined effect could make monthly payments unaffordable.  Utilities can do more to stamp out corruption and eliminate bribe payments demanded by their employees. In the two surveys where grid-connected households were asked if they had paid a bribe to get connection, many replied "yes." Bribes for connection further encourage multiple connections.  Lifeline rates for blocks in the vicinity of 30-50 kWh a month may help strike a balance between affordability on the one hand and utility financial viability on the other. The first block of 25 kWh or less per month in some countries seems too small. If the country has a single schedule of a single block, progressive tariffs may be considered. For full benefits to be reaped, however, households need to be individually and accurately metered.  Prepaid meters help low-income households. The ability to pay in small increments when cash is earned, and not having to face the prospect of disconnection and a high reconnection charge, help especially the poor. However, electricity services are unreliable-in every one of the seven countries in which households were asked about service quality, blackouts were common-and this could mean that prepaid customers pay in advance for electricity they cannot get when needed. For this reason, it may be unfair to mandate prepaid metering on all customers.  Consideration may be given to basing lifeline rates on a moving average of consumption over several months. Volume-differentiated tariffs with a large increase in unit price from the first to the second block in particular may pose financial hardships for the poor if they are applied strictly on monthly consumption. To address this concern, in Cape Town, South Africa, free electricity is based on monthly consumption averaged over the past 12 months. Using moving averages and issuing warning messages in advance to customers as they approach the limit are two ways of alleviating such hardships.  High appliance efficiency further helps reduce monthly bills. Recent advances in energy efficiency improvement are enabling households to meet basic electricity needs for as little as 15 kWh a month. For example, using four light-emitting diode light bulbs of 800 lumens each for four hours a day, charging a cell phone, running a large fan for five hours a day, and watching one small efficient television for three hours a day consume about 15 kWh over a month. That said, light-emitting diode light bulbs and other efficient appliances are relatively expensive, making it difficult for the poor to pay the upfront costs-one of the main reasons energy is more expensive for the poor than for the rich, exacerbating energy poverty. Financial assistance may be needed.  For gender equality, helping the poor is likely to go a long way in helping female-headed households.
Once both per capita and total household expenditures are accounted for, there is little evidence that female-headed households are disadvantaged with respect to electricity use, except in urban Ghana and Togo. That said, these are preliminary findings, and the gender dimension should be investigated further to see how robust this finding is.
Utilities can take steps to provide more targeted assistance to low-income households without increasing their deficits markedly:  Rolling out prepaid meters helps utilities. Utilities are assured of prompt payment for every kWh consumed.
 Tariffs do not have to be exceptionally low. Where consumption at the subsistence level is a very small share of monthly household cash expenditures, consideration may be given to raising tariffs. In the sampled countries, Angola and Nigeria have exceptionally low unit tariffs in absolute terms and relative even to the income of the poor.  Volume-differentiated tariffs at the high end may be considered. Several countries have volumedifferentiated tariffs between the first and the second block, followed by increasing block tariffs thereafter. But if households are individually metered so that high monthly consumption signals a household with many appliances, their ability to pay can be exploited more to cross-subsidize the low end of monthly consumption by switching the order: increasing block tariffs between the first and the second block, followed by volume-differentiated tariffs.  Reallocation of cross-subsidies can help make electricity affordable in most countries. Targeted subsidies needed to make 30 kWh a month of grid electricity affordable to all urban residents, are not large in most countries. While recognizing that perfect targeting is not possible, the subsidies calculated in this study point to the potential for achieving universal access in urban and peri-urban areas.  Eliminating operational inefficiencies may not eliminate underpricing, but can make subsidies to the poor more financially viable to utilities.  Many Latin American countries treat assets associated with new connection no differently from all other assets-high-and medium-voltage transmission lines, for example-and capture them in the regulatory assets for tariff setting. Such an approach spreads the costs of new connection and helps reduce multiple connections by making the initial connection much more affordable in countries where connection fees are currently high.
Lastly, this paper found areas where questions in household surveys delivered ambiguous or questionable answers. A standardized approach to data analysis would also be helpful. Some examples follow.
 Quantitative information on kWh consumed seems difficult to obtain. Enumerators should be better trained to spot implausible answers if this question is asked. Without information on kWh consumed, it is not possible carry out the distributional analysis of subsidies outlined by Komives et al. (2005). In this paper, multiple schedules in many countries and the widespread practice of sharing meters meant that kWh consumed could not be back-calculated (except in Rwanda), because there is no way of telling which tariff schedule was applicable to each household.  Given the policy implications of pervasive use of shared meters, the wording of questions concerning shared meters can be tightened, for example by distinguishing between "private" meters on one's premises to which other households connect and unshared meters on one's premises.  It would be useful to distinguish between solar lanterns and solar panels, and avoid using the term "solar energy."  In countries where generator ownership is prevalent, it would be useful to separate spending on automotive fuels from spending on fuels for generators.  In countries with large commercial and bill collection losses, it might be worth asking questions about nonpayment in a way that would enable more robust conclusions. Separating electricity from water where the questionnaire currently reads "free electricity/water" as employment benefits and inclusion in the rent are two examples.
 National household expenditure surveys are useful because they enable identification of the poor.
They are less useful for calculating access to electricity. Agreeing how to standardize analysis across divergent household surveys among energy practitioners would help compare computation of access rates across studies.  More systematic collection of freely acquired basic goods and services to separate cash expenditures from own consumption would help estimate disposable cash income, which is needed to assess the ability to pay for electricity.  A national household expenditure survey cannot substitute a dedicated energy survey. One problem with energy surveys is that information on total household expenditures is not collected because that requires a long list of questions. One option is to launch the energy survey in parallel with the national household expenditure survey, visiting the same households. Further, examination of household survey data side by side with data on residential customers from utilities would enrich analysis considerably. This paper has examined access largely from the point of view of demand. But supply constraints are equally important. Indeed, where data on community access and household connection to the grid were available, most unconnected households did not live in electrified communities, a result dominated by lack of community access in rural areas. This points to the importance of developing a financially viable strategy for rural electrification, while making grid electricity affordable in urban areas.

Annex
The surveys used in this study are listed in Table A.1. Total expenditures used to measure poverty are in real terms (accounting for regional and sometimes temporal differences in prices), except in Botswana, where the government's decision to use nominal values was followed. Poverty is based on per capita or per adultequivalent expenditures, as indicated in the last column.  Mozambique  70  71  29  62  52  28  23  24  54  23  32  Niger  83  82  18  55  48  14  6  7  65  33  39  Nigeria  64  60  28  64  51  14  8  10  79  45  58  Rwanda  86  85  22  49  45  22  23  22  72  24  31  São Tomé and Príncipe  49  50  64  59  62  39  34  37  73  77  75  Senegal  57  51  33  57  47  36  15  24  78  51  63  Sierra Leone  62  61  31  66  53  30  24  26  42  26  32  South Africa  39  33  31  69  45  36  54  43  269  83  197  Swaziland  76  63  31  73  63  38  53  50  124  51  69  Tanzania  74  69  6  26  21  24  21  22  62  29  38  Togo  62  56  35  73  58  20  15  17  73  36  50  Uganda  77  73  10  22  19  32  26  27  76  38  47  Zambia  65  64  28  77  60  20  19  20  69  22  39   Median  67  63  26  56  46  23  20  21  73  37  49 Access to grid electricity expressed as the percentage of all people who live in households reporting connection to the grid is shown in Table A.3. The statistics in the table correspond to the lowest measure of access rate.    Table A.6. A schedule has its own unique energy and fixed charges, and therefore two schedules with identical energy charges but different fixed charges are considered distinct and counted as two. The table shows the number of schedules for single-phase and three-phase electricity connections. Some countries do not make this distinction, but where the schedule depends on installed capacity (kilowatts or kilovolt-amperes), the inferred number of phases is used to construct the entries in Table  A.6. If no distinction is made and the schedule is independent of installed capacity, the number of schedules appears under single phase only in the table below, with no entries under three phases.
Sources: Utility and regulatory agency websites. Note: "Volume differentiated for block 2, increasing thereafter" means the tariff schedule is volume-differentiated between block 1 and block 2, but assumes a increasing block tariff structure from block 2 to block 3 and higher blocks. # of single-phase schedules = number of schedules for single-phase connection or all connections in countries with no distinction between single and three phases; # of blocks = number of blocks in the schedules preceding the column; # of 3-phase schedules = number of schedules explicitly associated with threephase connection, or with power capacity that is associated with three phases; prepaid schedule = Y (yes) if a schedule for pre-paid is separately listed, N (no) otherwise; n.a. = not available (information could not be obtained); TOU = time of use.  a. US$141 is the first payment in an installment plan of 18, 60, and 180 months. For the 180-month plan, the monthly spending increases by US$4.98 in the remaining 179 months if the customer chooses to pay over 180 months. US$141 is used to examine the connection-charge share of expenditures. b. US$118 is the first payment in an installment plan of 18 months. The monthly spending increases by US$6.72 in the remaining 17 months if the customer chooses this payment plan. US$118 is used to examine the connection-charge share of expenditures. c These numbers are for Johannesburg. Depending on eligibility, poor households are provided with 50, 100, or 150 kWh of free electricity a month. Because free electricity cannot be generalized to the rest of the country, monthly spending without free allocation of electricity is shown.  Table A.12 reproduce Table 5-Table 8 for monthly consumption of 50 kWh. Table A.13 and Table  A.14 present results for monthly consumption of 100 kWh, and Table A.15 and Table A.16 do the same for  250 kWh. Table A.17 shows poverty-gap calculations for grid electricity using the lowest-cost tariff schedule for each consumption amount. Table A.18 shows the corresponding poverty headcount, which is the percentage of people (rather than households) who have to spend more than 5 percent of their total household expenditures (inclusive of freely acquired food and other items) on grid electricity. For South Africa, where different municipalities with their own eligibility criteria provide varying amounts of free electricity, the tariff schedule in Johannesburg without free electricity was used for this purpose. Table A.19 summarizes the levels of subsidies needed to enable every household to consume 30, 50, and 150 kWh of electricity a month, expressed in millions of U.S. dollars as well as a proportion of the total revenue collected by the utilities. Source: World Bank staff calculations based on utility information and household survey data. a. 50 kWh a month of free electricity may be provided to the poor in some municipalities, including Johannesburg. Source: World Bank staff calculations based on utility information and household survey data. Note: -= Grid connection status not available in the survey. a. 50 kWh a month of free electricity may be provided to the poor in some municipalities, including Johannesburg. Source: World Bank staff calculations based on utility information and household survey data. Source: World Bank staff calculations based on utility information and household survey data. Note: -= Grid connection status not available in the survey. Source: World Bank staff calculations based on utility information and household survey data. a. 100 kWh a month of free electricity may be provided to the poor in some municipalities, such as Johannesburg. Source: World Bank staff calculations based on utility information and household survey data. Note: -= Grid connection status not available in the survey. a. 100 kWh a month of free electricity may be provided to the poor in some municipalities, such as Johannesburg. Source: World Bank staff calculations based on utility information and household survey data. a. Where the poor are entitled to varying amounts of free electricity (as much as 150 kWh a month for some poor families in Johannesburg), the expenditure share would be lower. Source: World Bank staff calculations based on utility information and household survey data. Note: -= Grid connection status not available in the survey. a. Where the poor are entitled to varying amounts of free electricity (as much as 150 kWh a month for some poor families in Johannesburg), the expenditure share would be lower.  30  50  100 250 35  85  100  10  31  78  98  Nigeria  0  0  18  62  0  0  39  83  0  0  31  76  Rwanda  34  53  71  90  68  87  97  100  63  82  93  98  São Tomé and Príncipe  1  4  28  93  1  5  33  96  1  4  30  95  Senegal  3  8  33  94  11  29  75  100  8  20  57  98  Sierra Leone  13  43  85 100  45  82  98  100  33  67  93 100  South Africa a  1  3  14  47  1  6  29  79  1  4  20  59  Swaziland  2  7  28  62  4  12  40  80  3  11  37  76  Tanzania  1  2  42  95  3  13  75  99  3  10  66  98  Togo  16  55  81  97  53  88  97  100  39  76  91  99  Uganda  11  33  69  94  25  58  90  99  22  52  85  98  Zambia  5  8  14  48  34  42  57  90  24  30  42  75  Median  2  7  29  87  5  19  60  96  4  16  47  94 Source: World Bank staff calculations based on utility information and household survey data. a. Tariffs in Johannesburg are used. No allowance for free electricity is assumed for the purpose of these calculations.  The last two tables show the results of probit and least-squares regression analysis of the gender of the head of household. Note: -= statistically insignificant at 5%; n.d. = no distinction made for different sources of electricity; n.a. = not available (data not collected); NA = not applicable. a. 1 for urban, 0 for rural. b. 1 if spending on electricity is positive, zero otherwise (spending is zero or missing). c. 1 for electricity (grid as well as off-grid, such as solar and diesel generators) as the primary source of energy for lighting or cooking, 0 otherwise. d. 1 for electricity (excluding off-grid, such as solar and diesel generators) as the primary source of energy for lighting or cooking, 0 otherwise. e. 1 for grid connection, 0 otherwise, for the 17 countries where the survey asked about the status of grid connection. f. Logarithm of spending on electricity retaining all households with positive expenditure. g. Logarithm of spending on electricity retaining only grid-connected households with positive expenditure.  Table A.21. Footnotes a-d correspond to d-g in Table  A.21.