Expanding Social Protection Coverage with Humanitarian Aid: Lessons on Targeting and Transfer Values from Ethiopia

Abstract While social protection programmes have multiplied over the last two decades across sub-Saharan Africa, these coexist alongside humanitarian assistance in many places, calling for better integration of assistance delivered through the two channels. Progress on this front is hampered by limited evidence of whether and how these historically siloed sectors can work together. Using quantitative and qualitative data from districts covered by Ethiopia’s Productive Safety Net Programme (PSNP) and where humanitarian food assistance (HFA) was delivered, we assess differences in targeting and transfer values. We find that the PSNP and HFA were targeted to households with different characteristics. PSNP transfers did, on average, reach those households that were chronically food insecure. HFA, while delivered through the PSNP systems, was targeted to households that were acutely vulnerable. These are promising findings as they suggest that social protection systems are able to effectively deliver a continuum of support in response to different types of vulnerability and risk. On transfer values, we find that the value of PSNP transfers is greater than those for HFA. One reason for this may be due to the social pressure on local officials to distribute support more widely across a drought-affected population when faced with acute needs.


Introduction
Social protection policies and programmes are now firmly established across much of sub-Saharan Africa (Beegle, Coudouel, & Monsalve, 2018). In 2016, Cirillo and Tebaldi (2016) mapped and profiled 127 non-contributory social protection programmes from 39 African (i) strengthening the linkage between the PSNP and HFA; and (ii) supporting the application of a common set of operational procedures to the provision of the PSNP transfers and transfers to the non-PSNP households in response to drought. Using mixed methods, we consider three entry points for systems strengthening along the social protection delivery chainadministrative coordination, targeting and payments (Seyfert, Barca, Gentilini, Luthria, & Abbady, 2019). We pay particular attention to: (i) learning about the effectiveness of an integrated system when targeting households affected by different stresses and shocks; and (ii) understanding whether transfer values were equivalent for social protection and humanitarian assistance clients, given that divergences in these might create social tension or attempts to move back-and-forth between benefits.
Drawing on evidence from Ethiopia during the period 2017-2018when the system supporting the PSNP was used to deliver humanitarian support in an aftermath of a severe widespread drought that occurred in 2015/2016allows us, uniquely, to analyse both social protection provision and humanitarian assistance in the same study. A striking finding is that humanitarian transfers enabled expansion of coverage to those most affected by the shock, even though they are not the poorestas the poorest were largely covered by social protection interventions. The distinct objectives of humanitarian and social protection provision help explain the observed differences in average transfer levels between the PSNP and HFA beneficiaries.

Context
Ethiopia has a long history of devastating droughts that have led to famines or localised food shortages (De Waal, 2017;Webb & von Braun, 1994). Major famines leading to hundreds of thousands of deaths occurred in 1974 and 1984, and throughout the 1990s, there were recurrent requests for emergency food aid to address localised food shortages. Emergency assistance following a major drought in 2002 narrowly averted another mass famine, but, by that time, it was clear that more sustainable types of responses were needed. Beginning in 2005, the Government, together with a consortium of international donors, began implementing a new response to chronic food insecurity in rural Ethiopia, the PSNP. Unlike annual emergency appeals, the PSNP was conceived as a multi-year programme to provide recipients with predictable and reliable transfers. The ambition was to improve food security and to prevent asset depletion at the household level, and through public works, to create assets at the community level. 2 Between 2006 and 2014, food security improved among households that took part in the public works component of the PSNP, and about 80 per cent of this improvement can be attributed to the programme . As of early 2021, the PSNP operated in more than 300 chronically food insecure woredas (districts) 3 with about eight million beneficiaries.
PSNP woredas can apply for additional contingency funds to respond to a broad range of shocks to protect PSNP investments and support the livelihoods of programme beneficiaries. Contingency planning is a design feature that has been embedded in the PSNP from the programme's outset (Wiseman, Van Domelen, & Coll-Black, 2010). To avoid short-term household food needs turning into chronic food insecurity, the fourth phase of the PSNP (2015-2020) included two design features: a woreda contingency budget and a federal contingency budget. The contingency budget is calculated as 16 per cent of the base transfer budget: 11 per cent is held at the federal level while 5 per cent is managed by woredas. 4 Contingency funds have two uses: (1) to cover the needs of existing beneficiaries who require additional 'top-up' transfers for a limited period of time in order to maintain their food consumption and protect their productive assets during a shock; and (2) covering non-beneficiaries in PSNP woredas who require support on a temporary basis until the shock passes. The two contingency budgets are meant to enable the PSNP to provide an early response, before the full effects of the emerging shock are felt, so that people do not have to resort to harmful coping mechanisms such as selling assets.
Expanding social protection with humanitarian aid 1983 But the PSNP is only able to expand up to the level of available resources in the woreda and federal contingency budgets. Any transitory needs that exceed these additional sources are covered through the national emergency response systemtypically through HFA. The geographical overlap between the PSNP and HFA is considerable. In 2018, more than 90 per cent of the PSNP woredas were also receiving HFA (Figure 1), indicating that despite the successes of the PSNP, the need for HFA has persisted. In non-drought years, approximately five million Ethiopiansnot included in the PSNPneed emergency assistance (National Disaster Risk Management Commission [NDRMC], 2018). In the period covered by this study (2015)(2016)(2017)(2018), poor rainfall during the main cropping seasons (the short rains, the belg and the long rains, the meher) in 2015 led to the worst drought in decades resulting in an additional 10.2 million people (that is, in addition to those covered by the PSNP) requiring food assistance (NDRMC, 2017). While the rains in 2016 were better, the 2017 agricultural year was characterised by below average and erratic belg rains, conflicts in Oromia and Somali regions and localised weather shocks in the south, resulting in 9.35 million people needing food assistance, including 850,000 internally displaced people (NDRMC, 2018). Rainfall and conflict shocks have continued since then; for example, in addition to those households covered by the PSNP, it is estimated that the 2021-2022 drought, together with civil conflict has driven HFA requirements to record levels (FEWSNET, 2022).
The PSNP uses a mix of geographic and community-based targeting to identify chronically food insecure households in chronically food insecure woredas. Initially, data on past receipt of food aid were used to determine the number of eligible beneficiaries in each region and woreda (Wiseman et al., 2010). Woreda administrators then selected the chronically food insecure kebeles, assigning the woreda's 'PSNP quota' among these areas. Within programme kebeles, community-based targeting is used to identify eligible households, which are then assigned to public works or direct support (Government of the Federal Democratic Republic of Ethiopia [GFDRE], 2004). Community-based targeting is based on common programme criteria, including household's chronic food security status, recent shocks, household assets (particularly livestock), non-farm income, specific vulnerabilities (such as households with chronically ill or elderly members), as well as the needs of poor and vulnerable pregnant and lactating women. Household re-targeting in the PSNP is conducted every two to three years, with minor adjustments taking place annually. PSNP transfers are in the form of cash or food and in most woredas their distribution is managed by the Government of Ethiopia. PSNP beneficiary households receive transfers in return for participating in public works over a six-month period during the dry season when agricultural activities are not taking place. Food insecure households with limited labour capacity are exempt from public works and receive direct support. Food payments are supposed to consist of 3 kg of cereals (usually in the form of wheat) and 0.8 kg of pulses for each day worked (GFDRE, 2014). Cash payments are calculated based on the cost of buying 3 kg of cereal and 0.8 kg of pulses per day. The wage rate is adjusted annually to take account of changes in grain prices, though these increases have not always been sufficient to maintain the real value of cash payments (Hirvonen & Hoddinott, 2021;Sabates-Wheeler & Devereux, 2010). Each household is entitled to five days of payments per person, to a maximum of five persons or 25 days of payments. Thus, for households with 5 or fewer members, the per person per month PSNP transfers equal to 15 kg of cereals and 4 kg of pulses. However, during the period of this study, pulses were included in the payment package sporadically and these were later abandoned altogether due to funding constraints (World Bank, 2017).
The geographical allocation of HFA is guided by a semi-annual classification of woredas according to their food security status. This 'hotspot classification' is based on a set of indicators that include nutrition, agriculture, market, water, and hygiene, health, and education. Food insecure woredas are scaled from priority 1 to 3 with priority 1 woredas categorised as being most in need of assistance. In the period covered by this study, out of the 396 woredas classified as priority 1, 2, or 3 in July 2016, only 4 per cent (17 woredas) graduated out of HFA by January 2018. 5 In woredas selected to receive HFA, kebele officials select households eligible for the support. While the PSNP has specific targeting guidelines, there is no stated target group or guidelines outlining household eligibility for HFA. Household targeting for HFA is typically conducted twice a year, after the seasonal assessment of the humanitarian situation.
Until recently (and covering the period 2017-2018 when data for this paper was collected), HFA transfers were directed by NDRMC and the World Food Programme (WFP), with a consortium of non-governmental organisations (NGOs) helping to deliver. HFA households receive transfers from 3 to 12 months, depending on the results of needs assessment and the availability of resources. Unlike the PSNP, about 90 per cent of the HFA transfers are made in the form of food: 15 kg of cereals, 1.5 kg of pulses, and 0.45 l of cooking oil, per person per month (World Bank, 2017). These are unconditional transfers, although households were encouraged to participate in PSNP public works if they exist in their locality.

Data
We use quantitative and qualitative data covering the period 2017-2018. These data were collected as a part of the midline evaluation of the fourth phase of the PSNP (PSNP-4). Our analysis is based on woredas that were selected for both PSNP and HFA programmes in six regions (Afar, Amhara, Oromia, Somali, Southern Nations, Nationalities, and Peoples' Region (SNNP), Tigray).
Quantitative surveys, carried out in June-July 2018, collected data at the woreda, community and household levels. They were based on a stratified sampling strategy. First, 112 woredas were randomly selected from the list of woredas in which the PSNP was operational in the 6 regions. Three kebeles from each woreda were randomly selected; within these, one Expanding social protection with humanitarian aid 1985 enumeration area was randomly selected. A woreda level capacity survey focussed on understanding how PSNP and HFA operated at the woreda level. The primary respondents were woreda level officials with operative knowledgeable of the PSNP and HFA. The community level surveys were carried out at the kebele level, again targeting respondents who were in charge of managing PSNP and HFA at this level.
Household level data is based on combining two samples. One used the sample for the main PSNP-4 evaluation. The underlying sampling frame included all PSNP households, with the inclusion of non-PSNP households depending on their subjective welfare status. During listing, households were asked to place themselves on to a poverty ladder that had seven rungs. The first rung represented the very poorest households in the kebele and the highest (seventh) rung the very richest households in the kebele. Non-PSNP households were chosen from the bottom four rungs of the ranking because these households were used as a control group in the main PSNP evaluation. 6 While well suited for the impact evaluation of the PSNP, it limits our ability to study the targeting of the HFA and PSNP as the richer strata in these communities are missing from the sample.
We addressed this limitation by drawing a supplement sample using another sampling frame. In this sampling frame, only relatively richer households (that is, those in the top three rungs of the subjective poverty ladder) were eligible to be included in the additional sample. The additional sample was drawn from 56 randomly selected PSNP woredas (out of 112 included in the main evaluation) that (1) were in the main PSNP-4 evaluation sample and (2) were selected to receive HFA in 2018. Interviews of the additional household sample took place at the same time as the main PSNP evaluation surveys and was administered in the same kebeles and enumeration areas that were selected to the PSNP evaluation from that woreda. Three households in each enumeration area were added to supplement the main evaluation sample.
By combining these two sampling frames we have a household sample that represents the full welfare distribution in the kebeles. We restrict our quantitative analyses to these 56 woredas and 166 kebeles from which we have household level data based on both sampling frames. The household sample consists of 5006 households, 4510 from the sample used in the PSNP-4 evaluation and 496 households from the supplemental sample. Note that from the evaluation sample, we have 30 households from the lower rungs of the subjective poverty ladder in each kebele 7 while the additional survey provides us three households from the top rungs. Based on the listing data, these original 30 households drawn from each kebele represent 88.8 per cent of the population in the sampled areas while the 3 households at the top three rungs represent 11.2 per cent of the total population (Table A1 in the Supplementary Materials). As households at the bottom rungs were slightly oversampled, we apply sampling weights to correct for this. 8 Payment data are taken from households' self-reports of payments received between June 2017 and May 2018. For each month, respondents were asked the type of payment they received (cash, cereals, pulses, oils) and the amount. We have 2803 households reporting PSNP or HFA payments over 13,125 household-payment months. About half were made in food, and half in cash. To enable comparability between cash and food payments, we converted all food payments into Ethiopian birr. We valued food payments using price data collected from local markets in June 2018. We adjust cash payments made prior to June 2018 by computing regionspecific monthly cereal price indices using the monthly cereal price data collected at the kebele level. After applying the price index on cash payments, all paymentsirrespective of the payment modality or payment monthwere expressed in June 2018 prices.
Our qualitative data includes key informant interviews carried out at regional, woreda, and kebele levels with government officials responsible for implementing the PSNP and coordination with HFA. These included the Regional Food Security Task Force (RFSTF) and Regional Transfers and Resource Management Technical Committee (RTRMTC), members of the Woreda FSTF and Woreda TRMTC, members of Kebele FSTFs and Development Agents. Four focus groups were held at the community level that included men and women, the elderly, and the young, different types of programme beneficiaries and non-beneficiaries. Twenty woredas were selected for qualitative work (12 in the highland regions of Amhara, Tigray, Oromia and SNNP and 8 in the lowland regions of Afar and Somali) with one kebele visited per woreda.

Coordination processes
At the federal level, coordination between PSNP and HFA operations was supported by the National Disaster Risk Management Committee (NDRMC), created as a stand-alone institution following the adoption of the Sendai framework in 2015. The NDRMC implements, leads, and coordinates multi-sectoral responses to disasters. Until recently it was responsible for implementing HFA; however, this was transferred to the Food Security Coordination Directorate (FSCD) in the Ministry of Agriculture, the government agency responsible for implementing the PSNP. This reform at the federal level is meant to consolidate implementation of HFA with that of the PSNP (World Bank, 2020a). Yet, even before the shift in institutional arrangements for HFA, at the subnational level (at the woreda and kebele levels), a range of mechanisms and practices exist to coordinate PSNP implementation with HFA delivery. Responsibility for the PSNP cascades down to the regional level, where the Head of the Bureau of Agriculture (BoA) is responsible for its management while also chairing the Regional Food Security Taskforce (RFSF). The Woreda Food Security Task Force (WFSTF) reviews kebele annual plans and budgets, ensures that contingency plans for contingency budgets are in place, participates in monitoring and evaluation activities, and provides assistance to kebeles. The Kebele FSTF oversees all planning and implementation of PSNP 4 activities at kebele level. Finally, the Community FSTF is responsible for identifying programme clients while also participating in mobilising communities for participatory planning of public works. Development agents play a facilitating role in both PSNP and HFA implementation; they support the preparation of annual plans, help to ensure targeting lists are updated and verified, prepare PSNP payment lists, and so forth.
Qualitative fieldwork at the regional, woreda and kebele levels explored the degree of coordination between the PSNP and HFA. 9 A fundamental point is that the PSNP operates according to five-year plans. These specify the programme's coverage, budget and implementation guidance over a five-year period, or 'phase'. While the contingency budgets can be used to address food insecurity arising from localised shocks, regional and woreda Food Security Task Force (FSTF) officials pointed out that there is no mechanism to make large-scale adaptations to PSNP in response to either an increase in levels of humanitarian need or to the inflow of humanitarian aid.
No, we did not change the PSNP plan. Because PSNP and HFA are different things-they have their own guidelines and mechanisms. [TIG-RFSTF] No, the region has never changed plans to respond to humanitarian situations as regards to budget, targeting, complaints and grievance mechanism. [AMH-RFSTF] One exception to this was in the implementation of public works. While not uniform, officials in some regions and woredas explained that they followed federal guidance in cancelling public works requirements on beneficiaries during droughts. For instance, in Afar the region moved to unconditional transfers in response to the severe drought conditions that developed in 2016: Two years ago (in 2016), at the beginning of PSNP 4, due to the existence of a big droughtwhen grazing lands dried up, livestock died, and even people themselves were exposed to deaththe PSNP budget shifted to respond the situation. In this regard, PSNP transfers were given freely to PSNP beneficiaries with no requirement of performing public works. [AFA-RFSTF] Officials shared their views on coordinating implementation of PSNP and HFA support. While there was no common approach to institutionalise coordination, some regional officials referred to efforts to coordinate targeting, aligning transfer values, and reporting mechanisms: Expanding social protection with humanitarian aid 1987 We are developing one master list of clients, master list A for PSNP clients and master list B for humanitarian beneficiaries to avoid overlap and duplication. [ORO-RFSTF] R2: The operational linkages are manifest with the support of administration budget. Whenever relief is delivered from the Federal Government, the administrative budget is always covered from the regional management budget of the food security (PSNP). R3: Since HFA doesn't have technical capacity, the safety net accountant and coordinator are assisting them to do their transfer by PASS (software, used for PSNP systems). In this way, the two institutions are supporting each other. [AMH-RFSTF] Both PSNP contingency plans and HFA are delivered through one delivery mechanism using PASS software. Currently, the extended support of the 11 percent contingency budget is fully handled by the Regional Early Warning and Disaster Management Bureau. We support them in data assessment when the federal Disaster Response Management-Agriculture Task Force conducts a study. [AMH-RTRMTC] There is an operational linkage between PSNP and humanitarian assistance in all structures, from the regional to woreda and kebele levels. First, the beneficiary lists are identified differently for PSNP and for emergency/humanitarian assistance. As we are working in coordinated manner care is taken place to avoid duplications. [AFA-RFSTF] Interviews with woreda level officials uncover a similar picture of considerable efforts to integrate PSNP and HFA operations, even in the absence of a common approach for doing so. The result is a patchwork of practices and mechanisms to support the integration of PSNP and HFA, reflecting uneven implementation capacities both between and within regions. Woreda and kebele FSTF structures cover operations for both the PSNP and humanitarian responses in most places, even if targeting mechanisms and payments differ. As the quotations above highlight, years of PSNP implementation have contributed to building an infrastructure for delivery that is also useful for implementing HFA. Regional officials in Oromia pointed to numerous challenges for HFA distribution, which PSNP systems and structures were able to address: Among the challenges for delivering HFA are a lack of experience in the management of cash transfers, the absence of administration costs for the allocated HFA (transfer) resources … Coordination and logistics problems at the woreda level are many. [ORO-RFSTF] Somali Region has gone further than most regions in integrating PSNP and HFA delivery systems. According to the Somali RFSTF, a single system has been developed to align assistance. It incorporates harmonisation of planning, joint committee structures at different levels, aligning the timing as well as the value of transfers, the use of PASS, and issuing of client cards. Officials explained that, to manage the two channels of support (PSNP and HFA), they used HFA to increase the caseload and extend the period of public works payments from 6 to 12 months. While the funding is from different sources, the same principles used for PSNP targeting are applied and the same committee targets PSNP and HFA beneficiaries.
In Afar, regional officials explained that PSNP and HFA beneficiaries were different as was the distribution period, which reduced the challenge of simultaneously managing the two channels of support. In the past, the PSNP and HFA transfers in Afar were mixed together (which was possible given that they are both provided in food in Afar, whereas in Somali region PSNP transfers are made in cash and food). However, in response to the 2016 drought, PSNP and HFA support was kept separate, and different populations were targeted. Whereas PSNP beneficiaries were selected using the programme's targeting procedures, joint annual assessments conducted with UN agencies and NGOs were used to determine levels of humanitarian need.

Targeting
Our discussion of processes identified several areas of harmonisation and operational linkages between the PSNP and regional/woreda structures covering humanitarian assistance, including proactive support for aligning targeting across the sectors. So, for instance in Tigray, while 'PSNP has its own targeting mechanism and HFA also has its own mechanism' the two processes are harmonised so that 'HFA beneficiaries were those seasonally affected by drought and PSNP was for chronically food insecure' [TIG-SAE-KFSTF].
In principle, this harmonisation could take three non-mutually exclusive forms, using HFA to: (a) provide additional support to existing PSNP clients; (b) provide support to households that should have been included in the PSNP but were not because of budget constraints; (c) including households affected by shocks but who are not considered eligible for the PSNP. Most respondents in our qualitative surveys indicated that non-PSNP households were prioritised in HFA targeting processes. However, there were variations across and within regions. In Tigray, regional officials indicated that the HFA was intended only for non-PSNP households, so that there would be no overlap. As reported by a woreda official in SNNP, 'families suffering from entrenched chronic food shortage are covered under PSNP, while those facing seasonal food shortage are covered in HFA'. Our qualitative data suggest that in woredas where the PSNP is operational, while PSNP households are also eligible for HFA, non-PSNP households are prioritised in HFA targeting processes.
In HFA we look for those suffering the impacts of drought, internally displaced due to conflict, diseases, or any other risk. While PSNP goes to those who were critically food insecure for more than 3 years, HFA and contingency resources address those who normally have resources but have transitory needs due to shocks. [SOM-GUR-WTRMTC] Some woredas saw the HFA as an opportunity to increase the coverage of the needy households within the communities: The additional food aid helped us to increase our beneficiary number to cover those who are in need but were excluded because of the fixed quota system. [AFA-ELI-WFSTF] We have considered it as an opportunity rather than a challenge because many of the people in our woreda were not targeted by the PSNP. [AFA- ABA-WFSTF] The question as to whether PSNP households are eligible for HFA transfers was included in the quantitative community surveys. About 26 per cent of the kebele informants reported that PSNP households could be eligible to receive HFA with regional estimates ranging from 21 per cent in Tigray to 31 per cent in Afar, though in practice this rarely occurred in any of these six regions (see below). These surveys also included a series of 'forced choice' questions about kebele officials' attitudes towards targeting. Table 1 summarises the responses given by the kebele officials, contrasting views regarding PSNP and HFA targeting. 10 These indicate that attitudes of those with decision making power at local levels are pro-poor. Responses to questions 1 and 5 show that across both programmes fairness corresponds to provision of programme transfers and benefits to poor households. That said, there is a noticeable difference in strength of perceptions across the two programmes, with fewer officials reporting that poverty is the main criteria of fairness for the HFA programme. This may reflect the fact that drought shocks affect both non-poor and poor households. Responses to question 2 suggest that regarding HFA targeting, it is more difficult for kebele level officials to distinguish between the poor and less poor compared to the PSNP targeting. This may reflect an absence of clarity in the official eligibility criteria for HFA beneficiaries as well as the fact that during the bi-annual needs assessment the HFA focusses more on transiently poor instead of the chronically poor households targeted by the PSNP (World Bank, 2020a). Responses to the fourth forced choice question suggest that identifying the poor is significantly more difficult for HFA transfers (63 per cent of respondents said they could identity the poor for targeting HFA, compared to 84 per cent for PSNP targeting). Fuzziness in eligibility for HFA transfers is further illustrated in the response to question 3, where expectation of tension within the community (option 3a) is higher for HFA transfers.
Expanding social protection with humanitarian aid 1989 Using our quantitative household data, Table 2 shows the share of households benefitting from each programme in 2017. The PSNP provides considerably more coverage than HFA in these localities. About 46 per cent of the households benefitted from the PSNP compared to 7 per cent from HFA (Somali region being an outlier). Only 50 households reported receiving  both PSNP and HFA benefits, indicating that the HFA was rarely used to provide additional support to existing PSNP clients. (To avoid double counting, we consider these 50 households as PSNP beneficiaries only in our analyses.) We consider targeting performance by assessing how the likelihood of selection varies by household asset levels. Following previous targeting analyses in Ethiopia (Berhane et al., 2013, Berhane, Gilligan, et al., 2016, our primary measure of wealth is households' livestock holdings. Livestock is a widely used measure of wealth in rural Ethiopia and has been often used by the PSNP to target households at the local level (Coll-Black, Gilligan, Hoddinott, Kumar, & Wiseman, 2012). We measure livestock holdings in terms of tropical livestock units (TLU). Table A3 in the Supplementary Materials shows the TLU distribution by region. The average household in our sample owns 3.7 TLUs (median 2.3 TLU). We categorise households into zone-specific quintal ranges based on their TLU holdings. 11 Households in the bottom quintile own 0.01 TLUs (equivalent to 1 chicken) on average while the households in the top quintile own more than 10 TLUs on average. Households in the lowland agro-pastoralist regions of Afar and Somali regions having significantly more livestock than the highland regions; Amhara, Oromia, SNNP, and Tigray. Figure 2 shows the share of households benefitting from each programme by TLU quintile. These results are consistent with earlier work showing that PSNP targeting is generally propoor (Coll-Black et al., 2012;World Bank, 2020b): households in the poorest two quintiles are nearly three times more likely to be selected into the programme than households in the richest quintile. By contrast, the likelihood of selection into the HFA increases steadily as we move from the poorest to the richer quintiles. 12 Households in the richest quintile are nearly two times more likely to be selected to receive HFA than the households in the bottom two quintiles. These findings are robust to using alternative asset measures: TLU per capita (Supplementary Materials, Figure A2); lagged TLU levels ( Figure A3); and ownership of consumer durables ( Figure A4). Next, we estimate a multinomial logit regression where the dependent variable (Y ir ) equals zero if the household did not benefit from either programme in 2017, one if the household benefitted from the PSNP and two if the household benefitted from HFA. Choosing non-beneficiaries as the reference category (Y ir ¼0), we estimate: where k ¼ 1 if the household benefits from PSNP, and k ¼ 2 if from the HFA. The unit of analysis is a household i located in region r. Vector X captures various household characteristics that we hypothesise to be associated with selection into the PSNP or HFA. These include household head's characteristics (sex, age, and level of schooling), household asset levels (TLU, land size, and dwelling characteristics), duration of residence in the kebele, and whether the household head holds an official position in the kebele (see Lind, Sabates-Wheeler, Hoddinott, & Taffesse, 2018;Sabates-Wheeler, Lind, & Hoddinott, 2013). We also consider recent self-reported drought shocks as a predictor of programme selection. The term R contains a set of binary indicator variables for each administrative region. We estimate this equation using a weighted multinomial logit model where the weights are based on the survey correction weights described above. Standard errors are clustered at the woreda level. Coefficients are expressed as relative risk ratios; these quantify the probability of being selected to the PSNP (or HFA) relative to the probability of not being selected to either programme per unit increase in each independent variable (that is, variable in vector X). A relative risk ratio greater than 1 indicates an increased probability of being selected into the PSNP (or HFA). Table A4 in the Supplementary Materials lists these variables and their summary statistics. Across many of the descriptive indicators, HFA beneficiaries are more similar to PSNP nonbeneficiaries than to PSNP beneficiaries. Compared to the PSNP beneficiaries, the HFA households are better off, more likely to be male headed, and better educated. Table 3 shows the results of estimating equation (1). The estimated relative risk ratios are consistent with the narrative above that described how HFA is used to extend the safety net in PSNP localities to those who are otherwise not included. After controlling for differences in other household characteristics, households with larger livestock holdings are less likely to be selected into the PSNP. A unit increase in household's TLU holdings is associated with a lower probability of being selected into the PSNP (as indicated by the statistically significant coefficient of 0.93). Relative to non-beneficiary households, female-headed households are more likely to be selected in to the PSNP while the corresponding estimate for HFA is not statistically significant. Households with corrugated metal roofs are less likely to get selected into PSNP, but roofing material does not seem to play no role in selection into the HFA. Poor dwelling condition is also correlated with selection into both programmes (compared to not selected to either programme). Recently arrived households are more likely to be selected into the PSNP. The relative risk ratio in the case of HFA is also greater than one but not statistically significant. Finally, households reporting to have suffered from a drought are more than two times more likely to be selected into the HFA (as opposed to not selected to either programme). The same estimate is not statistically significant in the case of PSNP suggesting that household's exposure to recent localised droughts does not alter selection into the PSNP.
In summary, HFA was used to provide support to non-PSNP households that have experienced a drought shock. Because the PSNP is well-targeted (PSNP beneficiaries are, on average, poorer (as measured by TLUs), with less educated heads, older heads, more likely to be female headed, and have poorer quality housing) and because HFA was not used to provide additional assistance to existing PSNP households, this meant that HFA went to households that were relatively better off than the average PSNP beneficiary.

Transferspreferences and equivalence values
An interesting observation that falls out of the 'attitudes to targeting' data suggests that the ability to distinguish who is eligible for HFA transfers is more difficult than for PSNP, and that there is an expectation that HFA transfers should be more widely available than PSNP transfers. As it is possible that this leads local level officials to dilute HFA payments across more households than officially allocated, we look at the transfer preference and the level of transfer values across the PSNP and HFA payments.
Our qualitative work revealed that there has been some efforts to harmonise the value/wage rate of the transfer across the PSNP and the HFA. However, this is not applied in all woredas. Notes: Coefficients are relative risk ratios. Standard errors (in parentheses) are clustered at the woreda level. The null hypothesis is that the relative risk ratio equals 1. ÃÃÃ p < 0.01, ÃÃ p < 0.05, Ã p < 0.1. 0/ 1 ¼ binary variable. Estimates based on survey weights.
When asked whether the provision of HFA created any challenges (or opportunities) for PSNP implementation, regional and woreda level officials described concerns relating to the nonequivalence of the food value of the two transfers; the perceived inequality in the number of family members included in the calculation of the transfer amount; and the difference in the duration of receipt of the transfers.
The challenge is that the food for HFA includes pulses and oils, and this creates some jealousy for PSNP beneficiaries. They say: 'we are poor, but we did not get the pulses and oils'. [TIG-RFSTF] Under HFA the whole family is targeted, but PSNP transfers are limited to five members. It causes complaints: when some PSNP beneficiaries see their neighbours getting full transfers they complain. [SOM-SHI-WFSTF] We [the region] are concerned that the higher wage rate and household cap for food aid will cause dependency syndrome by discouraging people from becoming self-reliant through the support of PSNP development concept. [AMH-RFSFT] Woreda officials indicated that they have tried to deal with the complaints from PSNP beneficiaries towards the HFA provision.
We have been trying to help the PSNP clients to understand the difference, as HFA is a temporary intervention as opposed to PSNP. Besides, HFA does not have other program components like those included in the PSNP. [AMH-DESS-WFSTF] Regional, woreda and kebele officials gave mixed views on whether households preferred to receive their support through HFA or PSNP. The PSNP was seen to offer regular support over a longer period. However, the value of the in-kind payments received under HFA is higher than those received from the PSNP, particularly after the pulses were dropped from the PSNP payment package.
They prefer HFA, because the food transfer norm for PSNP has been reducedthe cash does not buy the same amount of grain. But for HFA, the food package includes pulses and oil. [TIG-RTMTRC] Generally, the community prefers to have transfers from the PSNP. We think it is because PSNP is for 6 months while HFA is mostly for 1 to 3 months. [ORO-Kuy-WFSTF] Notes: Ordinary least square method. Dependent variable is (log) per capita payment level (in birr). Unit of observation is household-calendar month. Standard errors (parentheses) clustered at the woreda level ÃÃÃ , ÃÃ , and Ã indicate significance at the 1, 5, and 10 per cent critical level.
A regional official from Amhara described the tensions around preferences for HFA versus PSNP: Some people at the community level prefer transfers through HFA. This is because of two reasons. If the transfer is in food, it has a full package (oil, cereals, and pulses) and fulfils the dietary requirement of the family. And, if the transfer is in cash, it exceeds the PSNP by 10 birr per person (43 birr vs 33 birr). For these reasons, people prefer HFA. In contrast, PSNP is preferred by the administration as well as many other community members. This is because, unlike HFA transfer, PSNP has a management budget which supports the delivery expenses of the transfer. Some in the community prefer PSNP because it is predictable and, so, people can plan ahead. [AMH-RTMTRC] A woreda FSTF official explained the confusion felt by community members: There is confusion about the difference between the transfers given through PSNP and HFA. The PSNP clients look at the transfer payment given in food items as a privilege. There are some PSNP clients who prefer the HFA more than the PSNP support. This is probably because of the recent deductions of transfer payments and skyrocketing food grain prices that have discouraged PSNP clientsas the transfer payment does not purchase grains equivalent to the amount of grains given by HFA. [AMH-DESS-WFSTF] In Afar, where both transfers are made in food, the preferences are not as straightforward.
As you know the PSNP is implemented from January to June and HFA is implemented based on needs assessments. In terms of predictability, the community prefers PSNP because the community are aware of the timing of PSNP transfers; but, in terms of volume and quality they prefer HFA because they get additional food items included like oil and pulses. [AFA-RTRMTC] These qualitative data show mixed views about the equivalence and preferences between PSNP and HFA payments. The differences in payment modalities (while HFA payments come mainly in the form of food, PSNP has largely shifted to cash transfers) makes it difficult for administrators and beneficiaries to compare the payment levels between the two programmes. Our previous work has shown that PSNP beneficiaries overwhelmingly prefer food payments, mainly because food payments maintain their value in the context of high food inflation (Hirvonen & Hoddinott, 2021). This explains the preference towards HFA payments in the form of food, especially in areas where PSNP payments come in the form of cash.
These considerations motivate us to assess the equivalence of transfer values across the HFA and PSNP recipients. We model (log) per capita payment received in month m by household i residing in woreda w on a binary variable obtaining value 1 if the payment came through the PSNP (and zero if HFA): where X and H are vectors capturing time-varying and time-invariant variables that may affect payment levels. Previous work on PSNP has noted that because of persistently high food inflation, the value of cash payments is typically considerably lower than the value of food payments (Hirvonen & Hoddinott, 2021;Sabates-Wheeler & Devereux, 2010). To account for this and the fact that cash payments are more common in the PSNP than in the HFA, we control for the difference in the payment value through a binary variable equalling one if the payment was in cash and zero if it was in food. As PSNP payments were characterised by delays that result in situations where several months of payments coming at once (that is, payment lumpiness), we also include a variable capturing the number of payments the household received in the previous five months. Time-invariant variables include household size (a set of binary variables for different number of household members) that we expect to influence payment levelseven after expressing Expanding social protection with humanitarian aid 1995 payments in per capita terms. We include a binary variable capturing female-headed households to control for possible differences in payments received by male and female-headed households as well as woreda (l) and month (h) fixed effects. The former controls for all observed and unobserved time-invariant characteristics fixed to the woreda (for example, administrative capacity, infrastructure) while the latter controls for macro shocks and seasonal changes occurring in each month and affecting all households. e is the error term. We cluster standard errors at the woreda level.
We restrict the data to months in which a payment occurred. Consequently, our regression model attempts to answers the following question: 'What is the difference in the transferred amount when two similar households residing in the same woreda receive a payment in the same month with only difference being the source of the payment (PSNP or HFA)'. The coefficient b quantifies this estimated difference in payment levels. A positive and statistically significant coefficient indicates that the PSNP payments are larger than the HFA payments, while the opposite is true if b is negative and significant. Table A5 in the Supplementary Materials provides summary statistics. The unconditional differences in payment levels are not statistically different from zero. Cash payments are more common in the PSNP than they are in the HFA. Beneficiary households received 2.3 payments in the previous five months, on average. The difference between PSNP and HFA households is not statistically different from zero. Female-headed households are more likely to receive PSNP payments than HFA households, as do smaller households.
Results shown in Table 4, column 1, tell us that monthly PSNP payments are, on average, 25 per cent larger than HFA payments and once we control for payment modality, the average estimated difference increases to 38 per cent (column 2). 13 The coefficient decreases slightly when we add further controls capturing payment lumpiness and household demographics to the regression model (column 3 in Table 4). Column 4 in Table 4 reports our preferred estimates. Controlling for payment modality and lumpiness, head's sex, and household size as well as calendar month and woreda fixed effects, payments through the PSNP are, on average, 28 per cent higher than those through the HFA. These results are robust to replacing calendar month fixed effects with calendar month times region fixed effects (Supplementary Materials, Table A6, column 1) as well as replacing woreda fixed effects with kebele fixed effects (Supplementary Materials, Table A6, column 2). They are also robust to restricting the data to the last six months (that is, January 2018 to June 2018) (Supplementary Materials, Table A6, column 3); this specification addresses concerns related to respondents having difficulties in accurately recalling transfers that occurred in more distant past.
These results are surprising given the generally held assumption and qualitative findings that indicate that HFA is a more valuable transfer. One way to explain this is that the differences in payment modalities between the two programmes coupled with relatively high food inflation makes it difficult for administrators and beneficiaries to compare the actual payment levels between the the PSNP and HFA. It is also possible that the pressure of local demand on local officials in the context of an acute and unanticipated shock may lead them to distribute the total HFA resources more widely than the officially prescribed transfer amount, diluting the total amount of transfer across more households.
Coefficients on the control variables reported in column 4 indicate that cash payments are about 17 per cent lower in value than food payments, consistent with our previous work (Hirvonen & Hoddinott, 2021;Sabates-Wheeler & Devereux, 2010). Households who have received more payments in the last six months receive smaller payments. This suggests that delays in payments results in payment lumpiness, a finding also consistent with previous work . We also see that larger households are receiving smaller per capita payments than smaller households are. We also explored whether the difference between PSNP and HFA payments varies by household size, but our results do not support this hypothesis (Supplementary Materials, Table A7).

Conclusions
Provision for basic needs in the wake of sudden and unanticipated shocks traditionally sits within the remit of humanitarian response and is typically provided as short-term (often oneoff) support. Social protection, by contrast, is usually provided as a regular transfer allowing clients to cope with and overcome the more predictable risks to their livelihoods. Ideally, a continuum of support and response would be offered to help households manage risk and uncertainty across a range of circumstances and contexts. Innovative programming in recent years has enabled social protection in different contexts to scale-up assistance in response to large covariate shocks, facilitated by targeting systems and contingency funding that provides programmes with the ability to respond more quickly to acute needs in a crisis than conventional humanitarian responses. The PSNP is one such programme.
Using data from 2017 to 2018, we find that in the context of a drought crisis, the institutional architecture and processes established by the PSNP provided a useful and effective system to also deliver humanitarian assistance. Furthermore, while the humanitarian assistance was channelled through the same structures, this did not necessarily constrain the provision of it to the same target group. Few households (50 out of approximately 5000) reported receiving both PSNP and HFA transfers. We find that the safety net support and humanitarian support were targeted to households with different characteristics. The PSNPmeant for the poorest and most food insecuredid, on average, reach those households that were characterised as poor. Households in poorest two quintiles have a higher likelihood of benefitting from the PSNP than the richest quintile. The humanitarian support, while delivered through the same system for PSNP, did not cater for households with poverty-identifiers, but responded to households that had reported a shock experience in the preceding 12 months. This is what one would expect from humanitarian support that is provided in response to an acute, rather than chronic, problem. Given that non-PSNP beneficiaries are often marginally better off, but also still prone to shocks, it is not surprising that the likelihood of selection into the HFA increases as we move to richer quintilesconsistent with the view that HFA is used to expand the total safety net. These findings suggest that social protection systems can effectively deliver support in response to different vulnerabilities and shocksto both chronic poverty as well as to acute vulnerability. The PSNP is a safety net programme with long-term objectives whereas HFA is designed to respond to ad hoc emergency needs. This is precisely what a shock responsive system should be designed to do.
A concern that continues to vex practitioners working to facilitate linkages across the humanitarian and social protection sectors relates to the non-equivalence in transfer values and the possible adverse effects this can have on recipient behaviour (double-dipping, social tension, or strategic movement between benefits). While the broader literature often indicates that humanitarian support is, on average, higher value per capita (Ghorpade & Ammar, 2021;McLean, Carraro, Barca, & Alfers, 2021), careful analysis shows that monthly PSNP payments support are higher than HFA. This may be due to the social pressure on local officials to distribute a fixed amount of support much more widely across a drought-affected population, diluting support across more beneficiaries than was officially planned for, thus allowing for expansion of coverage (horizontally and vertically) while maintaining different programme objectives (graduation and a provision for survival).
Our findings are based on the analysis of mixed-methods data from a well-known, relatively mature social protection programme. While generalisable at the level of learning for systemwide improvement, the findings are specific to the combined provision of social and humanitarian assistance in the context of a weather-related shock (drought), where a well-developed national social protection system exists. In (at time of our study) a relatively stable governance setting, the investment in national and decentralised infrastructure for social assistance reaped benefits for a continuum of response. However, ongoing conflict in Ethiopia, and the ensuing disruption to PSNP implementation in the affected areas, illustrates how the nature of a shock will Expanding social protection with humanitarian aid 1997 determine whether social protection and humanitarian provision can (or should) align and be harmonised. Conflict and crises that disrupts systems of provision justify a continued role for humanitarian support. then a regional targeting analysis would indicate that humanitarian support is targeted towards wealthier households. To address this issue, we use zone-specific TLU quintal ranges; each household's standing in the quintile is based on its relative ranking within a zone, not within a region. 12. Supplementary Materials, Figure A1 shows the beneficiary incidence graph; the share of beneficiaries by TLU quintile. In the PSNP, nearly 52 per cent of the households in the poorest two quintiles benefit from the programme. In the HFA, the corresponding share is 29.5 per cent. 13. This increase in the estimate is consistent with our earlier finding that, due to high food inflation, cash transfers are less valuable than food transfers (Hirvonen & Hoddinott, 2021). Since PSNP transfers are made mainly in the form of cash and the HFA mainly in the form of food, the estimated difference in the value of PSNP and HFA payment widens further once we control for the payment modality of the transfer.

Funding
Initial work on this paper was funded under World Bank Award No. 8005441. We acknowledge funding from the CGIAR Research Programme on Policies, Institutions, and Markets (PIM).