The Coping Strategies Index (CSI) Baseline Survey: World Food Programme (WFP) Assisted Refugees in Western Tanzania

prepared by Greg Collins for WFP Tanzania June – July, 2004

ACKNOWLEDGEMENTS Given the size and geographic distribution of the refugee population covered by the Coping Strategies Index Baseline Survey, it should be obvious that its success was due in large part to the coordination efforts of WFP Tanzania. Although it is impossible to name and thank each contributor, some deserve special mention. •

Tobias Flaemig (WFP Nutritionist) was responsible for overall coordination of the CSI baseline survey.

•

Fe Guevarra was responsible for initial planning and design of the CSI in Western Tanzania.

•

Odasi Mudogo, Abubakar Mdathiru, Rose Msafiri, Caroline Robert, Sixtus Baragovia, and Happygod John (WFP Program Staff) were actively involved in data collection and analysis.

•

Evaline Dianga (ODK VAM Officer) facilitated the May 2004 CSI training workshop and provided technical feedback on drafts of this report.

•

Robin Wheeler (ODK VAM), together with Evaline Diana, played a critical role in initial planning and conceptualization.

•

Keith Ursel (Emergency Coordinator) provided the management directive that ensured the survey’s success. 18 enumerators were hired to conduct the household survey1.

• •

WFP’s partners were also actively involved in the initial CSI workshop and form an important audience for this report.

Finally, I would like to acknowledge the contribution of the refugees that participated in the Focus Group Discussions and the 780 refugee households that participated in the survey who have helped WFP better understand how they cope with acute food insecurity.

Greg Collins, M.P.H. Consultant to WFP Tanzania July 21st, 2004 TABLE OF CONTENTS

1

Members of this enumeration team (list available through WFP Kigoma sub-office) should be used during subsequent surveys.

ii

List of Tables Figures and Boxes

iv

List of Acronyms

iv

Executive Summary

v

Introduction

1

Background Purpose

1 1

Data Collection and Methodology Creating the Coping Strategies Index and Deriving the CSI Score Household Survey Methodology Data Analysis

2 4 6

Coping Strategies and the CSI CSI Comparisons between Groups Multivariate CSI Analysis Conclusions and Recommendations Recommendations for Time Series CSI Comparisons Program Recommendations Based on the CSI Baseline Appendices Appendix 1: Appendix 2: Appendix 3: Appendix 4: Appendix 5:

2

6 6 12 16 16 17 19

Sampling Survey Protocols and Special Sampling Instructions CSI Questionnaire Refugee Camps by District, Nationality, and Market Access Statistical Output for Models 1 And 2

20 21 23 25 26

LIST OF TABLES, FIGURES, AND BOXES

iii

Tables 1. Frequency Responses: Description, Categories, and Code 2. Camps by Market Stratification 3. Percentage of Households using CSI Coping Strategies by Severity Figures 1. Consumption Coping Strategy Severity 2. Mean CSI Score by District 3. Market Access by Nationality 4. Mean CSI Score by Market Access and Nationality 5. Mean CSI Score by Head of Household’s Educational Attainment 6. Educational Attainment of the Head of Household by Nationality 7. Mean CSI Score by Income Sources 8. Income Sources by Nationality 9. Model 1 – Net Effect of Explanatory Variables on CSI Score 10. Model 2 – Net Effect of Explanatory Variables on CSI Score Boxes - Boxes provide additional guidance for readers who are unfamiliar with statistical analysis and interpretation. 1. Interpreting Inter-Group Comparisons: Who is Food Insecure 2. Using Multivariate Analysis to Understand Why Households are Food Insecure LIST OF ACCRONYMS CSI FGD ODK UNHCR VAM WFP

Coping Strategies Index Focus Group Discussion WFP Regional Office, Kampala United Nations High Commissions for Refugees Vulnerability Assessment and Mapping World Food Programme

EXECUTIVE SUMMARY

iv

The Coping Strategies Index (CSI) was designed as a rapid household food security assessment and is well suited to WFP’s desire to monitor changes in the food security status of refugee populations in western Tanzania in response to shocks such as market closures, movement restrictions, and reductions in ration size. The CSI is also a food aid monitoring tool and will allow WFP Tanzania to gauge the impact of its programming. The CSI generates an index score based on severity and frequency of the strategies refugee households use to cope with acute food insecurity. The CSI scores established during this baseline survey provide a relative measure of food insecurity that can be used to compare subgroups within the refugee population. More importantly, these scores represent the initial data point that will be used to track changes and monitor trends in the food security status of these sub-groups and the refugee population as a whole. There is a strong association between lack of access to external markets and comparatively higher CSI scores, meaning those without access to markets outside of the camps are more food insecure. Although trends can not be established with a single data point, this suggests that restrictions on movement and the closure of markets in recent months have led to a deteriorating food security situation among households who have been cut-off from external markets. To the extent possible, WFP and its partners should advocate for the timely reopening of markets that allow refugees to complement the food aid they receive with other foods and non-food items. Households that depend on agricultural labor as a source of income are more food insecure than households that do not. Although this finding appears at first to be counter-intuitive, a probable explanation for this association is that working as a laborer is the only available income generating option for those lacking access to capital and inputs (e.g. poorer households that are already more food insecure). By contrast households that earn income from livestock production, petty trade, and working as incentive workers for NGOs are more food secure than those without income from these sources. The CSI baseline survey also suggests that those households originating from the Democratic Republic of Congo (DRC) are less food insecure than households originating from Burundi. Further analysis shows this relationship is partially driven by the fact that Lugufu 1 and Lugufu 2, camps exclusively occupied by Congolese refugees, are the only camps with current access to external markets. Future survey rounds should provide clarity in determining the degree to which this association is caused by market access, by an unspecified factor associated with nationality, or some combination of both. The results outlined above and elaborated upon in this report establish the baseline for future comparisons. To this end, it is recommended that CSI surveys be scheduled at regular 6 month intervals during the first two years so that seasonal trends in food security status can be established. In addition to scheduled surveys, the timing and frequency of CSI surveys should be adjusted in response to perceived shocks or improvements that are likely to have an effect on refugee livelihoods and food security. Each subsequent survey should examine whether or not the comparatively food insecure sub-groups identified in this report are disproportionately affected by a particular shock or improvement. They should also be used to identify emergent sub-groups that are being disproportionately affected.

v

INTRODUCTION Background There are approximately 460,000 refugees, primarily from Burundi and the Democratic Republic of Congo (DRC), receiving WFP food aid in 13 refugee camps in western Tanzania. The ration previously took into account opportunities for self-reliance as refugees were able to move within a 4 km radius around the camps, allowing them to access commodity markets for trade, land for production, and agricultural and casual labor opportunities with Tanzanian farmers. At times during the last few years breaks in the food pipeline have led to temporary reductions in the food aid ration, affecting household access to food and the ability of households to sell food to meet their essential non-food needs. Furthermore, the enforcement of movement laws in 2003 by the Government of Tanzania has restricted access outside the camps and constrained refugee access to food and income sources that had previously complemented the assistance they receive. Purpose The degree to which these, and other shocks have impacted the livelihoods and food security of refugee households is largely unknown. Nutritional surveys that have been conducted do not provide a responsive and sensitive measure of acute food insecurity2. Recognizing this critical information gap, WFP Tanzania identified the need to develop a family stress index to monitor changes in the food security of the refugee population3. In their search for an appropriate methodology to fill this need, the Country Office liaised with staff members of the Vulnerability Assessment and Mapping (VAM) regional unit in Kampala, Uganda who had been involved with the initial Coping Strategy Index (CSI) pilot studies in Kenya. The CSI was designed as a rapid household food security assessment and food aid monitoring tool for use in emergencies. It makes use of the way in which households cope with acute food shortages as a means of comparatively assessing changes in household food insecurity between and within groups over time. Although an initial investment is required to tailor the CSI to a particular local context, once this is done the CSI is relatively quick and easy to administer and straightforward to analyze4. A decision was made to conduct a pilot study to gauge the appropriateness of using the CSI to monitor changes in the food security of refugees in western Tanzania. The survey is also intended to serve as a baseline for comparison with future CSI survey

2

The stability of prevalence indicators for weight-for-height (wasting) at < 5% over the last few years attests to the inability of anthropometric indicators to capture almost certain seasonal and shock driven fluctuations in household food security. Nutritional surveys are not well suited to the task of monitoring food security as they only measure the end result of severe food insecurity and are confounding by nonfood factors (health, water, environment). 3 The June 2003 WFP/UNHCR Joint Assessment Mission (JAM) also cited a need to monitor trends in the food security of refugees and measure the impact of WFP’s food aid program. 4 This is particularly true when viewed in comparison to other commonly used methods for quantifying changes in food insecurity (CARE/WFP, 2001).

1

rounds. To support this effort, the Country Office, together with VAM ODK, identified an external consultant5 to provide technical support to the CSI baseline. This report begins with a description of the CSI methodology as it was applied to the refugee population resident in 12 camps in western Tanzania6. Following this brief overview, an analysis and interpretation of the baseline data are presented. In addition to establishing baseline measurements, the data collection and analysis process that took place in June and July 2004 were used as training opportunities to build WFP staff capacity in these regards7. A separate document entitled Guidelines for CSI Data Collection and Analysis: WFP Assisted Refugees in Western Tanzania provides detailed guidance for WFP staff on how to conduct future survey rounds using the CSI tools developed during the baseline survey and how to analyze time-series CSI data. DATA COLLECTION AND METHODOLOGY Creating the Coping Strategies Index and Deriving the CSI Score Refugee households in Western Tanzania use a range of strategies to cope with shocks that affect their livelihoods and food security. These include both longer-term changes in income earning or food production activities (livelihood coping strategies) and shorter-term consumption coping strategies8 that are an immediate response to acute food insecurity. The CSI utilizes this second subset of coping strategies to derive a sensitive measure of acute food security. Severity The CSI methodology uses focus group discussions (FGDs) to establish a locally relevant set of consumption coping strategies and rank them in terms of perceived severity. Thirty-four FGDs were conducted among various refugee groups in western Tanzania. A description of FGD methodology (e.g. participant selection and discussion formats) is available as an annex to this report9. A list of approximately 30 potential consumption coping strategies was developed during the FGDs. This list was reduced to 13 strategies for inclusion in the household questionnaire. Seventeen strategies identified during the FGDs were rejected due to a) too few FGDs providing a severity ranking for the strategy and/or b) the variance in 5

Greg Collins was selected due to his involvement in developing the survey tools and sampling strategies for the CSI pilot studies in Kenya. 6 The small caseload of Rwandese/Congolese/Burundians housed in a camp for protection cases was excluded from the survey population. 7 WFP program staff members were actively involved in questionnaire development, sampling design, and management of data collection teams. In addition, a two day data analysis training was held in Kigoma and covered how to derive mean CSI estimates and make comparisons of mean CSI estimates among different groups within the refugee population. 8 Consumption coping strategies include changes in diet, increasing short-term food access, reducing the number of people to feed, and rationing of available food (CARE/WFP, 2001). 9 The Focus Group Discussions were conducted as an independent data collection exercise prior to the consultant’s arrival.

2

severity ranking among the 34 FGDs was greater than 1.00 (e.g. there was no consensus on how severe the strategy was in the FGD results)10. For these 13 strategies the severity weighting was derived by summing the results of all FGDs and dividing by the number of FGDs providing a response. The severity weights for each consumption coping strategy are provided in figure 1.

Figure 1 - Consumption Coping Strategy Severity (1 = least severe, 4 = most severe) 4.00

3.50

Severity Rank

3.00

2.50

2.00

1.50

1.00 i.

e.

j.

g.

b.

a.

f.

k.

d.

c.

l.

h.

m.

Consumption Coping Strategy

Coping Strategy Key11 a. b. c. d. e. f. g. h. i. j. k. l. m.

Borrow food or money (which you have to repay) from neighbors, friends, or relatives Purchase food on credit Send household members to eat elsewhere Send household members to beg Limit portion size at mealtimes Restrict consumption of adults in order for small children to eat Reduce number of meals eaten in a day Skip entire days without eating Sell high value, preferred foods to purchase larger quantity of less expensive foods Exchange your labor for food (work for food) Sell Household Assets or the NFI's the household owns Engage in prostitution or theft of food (illegal activities) Have some members of the household migrate elsewhere or repatriate

10

Two exceptions were made in which removal of the 2 highest and 2 lowest extreme values resulted in the variance being reduced to less than 1.00. 11 Order in key corresponds to the order in which the respondents were asked about each coping strategy.

3

Frequency Following the establishment of a locally relevant consumption coping strategies list and ranking, a household survey was used to quantify how frequently, during the last 14 days12, households had resorted to using each strategy because they ‘did not have enough food or enough money to buy food’13. The mid-point of each range was then entered into the database and used as the ‘frequency measure’ in the calculation of the CSI score for each household14. The frequency categories and codes are provided in table 1. Table 1 – Frequency Responses: Description, Categories, and Code Descriptiopn

All the time

Pretty often

Once in a while Hardly at all

Never

Number of Days (range)

13 - 14 days

6 - 12 days

2 - 5 days

1 day

0 days

Code

13.5

9

3.5

1

0

Calculating Household CSI Scores For each coping strategy, the frequency measure indicated by the household was multiplied by the relevant severity weight of that strategy. The sum of this product for all 13 coping strategies is equal to the CSI score. Household Survey Methodology Stratification Table 2 - Camps by Market Stratification

WFP and its partners expressed a desire to have separate CSI estimates for camps with good market access and those with poor market access. Accordingly, the twelve refugee camps included in the survey were divided into two strata according to market access15 and a separate sample was taken from each (table 2).

Strata 1- Good Market Access 1. Lukole A 2. Lukole B 3. Nduta 4. Kanembwa 5. Mtendeli 6. Karago 7. Mtabila 1

Strata 2 - Poor Market Access 1. Mtabila 2 2. Muyovozi 3. Nyarugusu 4. Lugufu 1 5. Lugufu 2

12

A 14 day recall period more accurately reflects the cycle of coping in the refugee camps (corresponds with food distribution every 2 weeks) and was more appropriate than the 30 day recall period recommended by the CARE/WFP CSI Field Methods Manual. An early pre-test of the questionnaire in May, 2004 suggested that recall over a period containing two food distributions led to confusion. 13 Respondents were prompted to quantify the number of days within the ranges given in table 1 if this information was not directly discernable from their initial response. 14 The choice to collect the data categorically, rather than as a continuous variable, is designed to speed up the time spent in data collection during this and subsequent surveys. Because the data is recalled over a 2 week period and meant only to serve as a proxy of coping stress, the loss of useful information is negligible and outweighed by the reduction in time spent during data collection. 15 Each camp was classified into one of four categories by WFP program staff during the CSI Training Workshop in May, 2004: Very Good = External markets with good supply, Good = internal markets with good supply, Poor = internal markets with limited supply, Very Poor = no markets. These categories were collapsed into 2 categories, good and poor, for use in stratifying the sample.

4

Sampling and Household Selection A sample of 390 households was taken from each stratum for a total sample of 780 households. The parameters used in determining the sample size requirements are provided in appendix 1. For each stratum, households were selected for inclusion in the sample from a sampling frame16 of all refugee households. A random starting household was chosen to select the first household for inclusion in the sample. Additional households were then selected systematically by calculating a sampling interval equal to the total number of households within the stratum divided by the number of households required from the stratum. The sampling interval was added to the random starting household to select the second household and then added to the second household to select the third and so on until 390 households were selected in each stratum17. Due to inconsistencies18 in the address data provided by UNHCR for Nyarugusu, Mtabila 1, Mtabila 2, and Muyovozi, an alternative household sampling method was devised for these camps. A description is provided in the survey protocol (appendix 2). Selecting Respondents and Replacement Strategy The desired respondent for the questionnaire was the head of household; defined as the primary decision maker within the household concerning food and income use decisions. When this person was unavailable, the spouse of the head of household was interviewed. If the spouse was unavailable, any other adult age 16 or above in the household was interviewed. If no respondents meeting these criteria were available the household was replaced by selecting the next closest plot in any direction as described in the survey protocol (appendix 2). The total number of replacement households in both strata was 183. This large number is attributable to data collection corresponding with food distribution days in some locations. Questionnaire The questionnaire used in the household survey is included in this report as appendix 3. The translated versions of this questionnaire (Kirundi and Kiswahili) used in data collection are available upon request. Software and Dataset Four data entry clerks, supervised by the CSI Coordinator, entered the household questionnaires into Microsoft Access. The data was exported to Statistical Package for Social Sciences (SPSS) version 11.5 for cleaning and analysis. The dataset is available in SPSS format by contacting Tobias Flaemig, WFP Kigoma Sub-office (tobias.flaemig@ wfp.org). 16

The sampling frame was constructed using data provided by UNHCR and is available through WFP Tanzania, Kigoma sub-office. 17 The total number of households per strata, random start numbers, and sampling intervals are provided in appendix 1. A list of address for the 780 households selected is available from WFP Kigoma 18 Missing street numbers prevented the use of addresses provided as unique identifiers in these camps.

5

DATA ANALYSIS Coping Strategies and the CSI As indicated in the CSI Methodology section, focus group discussions were used to tailor the list of consumption coping strategies and their perceived severity to reflect what refugees in Western Tanzania do when they ‘do not have enough food or enough money to buy food’. Information on the frequency of use of each consumption coping strategy was then collected during a household survey in order to calculate a CSI score for each household19. The coping strategies used in the index are listed in table 3 by perceived severity (least to most severe) and percentage of households that indicated using each strategy at least once during the two week baseline recall period. Almost all households (98%) indicating one or more coping strategies during this period. Table 3 – Percentage of Households using CSI Coping Strategies by Severity Consumption Coping Strategies Used to Derive the CSI

Percentage of Households Using Strategy

Sell high value, preferred foods to purchase larger quantity of less expensive foods

37%

Limit portion size at mealtimes

81%

Exchange your labour for food (work for food)

45%

Purchase food on credit

52%

Reduce number of meals eaten in a day

81%

Borrow food or money (which you have to repay) from neighbors, friends, or relatives

75%

Restrict consumption of adults in order for small children to eat

70%

Send household members to beg

24%

Sell Household Assets or the NFI's the household owns

31%

Send household members to eat elsewhere

20%

Engage in prostitution or theft of food (illegal activities)

8%

Skip entire days without eating

44%

Have some members of the household migrate elsewhere or repatriate

12%

CSI Comparisons between Groups The mean CSI score among refugee households during the baseline recall period is estimated at 53.02 +/- 2.8 CSI points20. Given that the CSI monitoring tool is a comparative, rather than absolute, measure of food insecurity, this score alone has no inherent meaning21. However, it establishes a baseline measurement from which 19

CSI = sum (frequency of each coping strategy * severity of each coping strategy) The estimated overall, weighted mean CSI score (weights correspond to actual proportion of population within each Market category used for stratification during sampling) is 52.61 +/- 2.3 21 There is no CSI threshold for definitively classifying a household as food insecure. 20

6

changes in food security among refugee households can be monitored over time. Furthermore, cross-sectional (e.g. one point in time) comparisons of food security status can be made between sub-groups within the refugee population. Box 1 – Interpreting Inter-Group Comparisons: Who is Food Insecure? The analysis below uses mean CSI scores to compare the relative food insecurity between groups defined by one or more household characteristics. These group comparisons describe associations between household characteristics and food insecurity (e.g. who is comparatively food insecure). However, it is critical that readers understand that these comparisons alone do not provide strong evidence concerning causal relationships (e.g. why these households are comparatively more food insecure).

A comparison of district CSI estimates suggests that Kibondo has the highest level of food insecurity and that Kigoma has the lowest level of food insecurity22. Kasulu and Ngara have approximately the same level of food insecurity23. The estimated mean CSI scores for each district24 are presented in figure 2. Figure 2 - Mean CSI Score by District 70

65

60

55

Mean CSI Score

51 50 40

37

30 20 10 0 Kigoma

Ngara

Kasulu

Kibondo

District

Refugees from the Democratic Republic of Congo appear to have lower levels of food insecurity than refugees from Burundi. Their estimated mean CSI scores were 41 and the 60 respectively25. Perhaps one explanation for this association is the relationship between market access and nationality of origin.

22

Both differences are statistically significant at p<.05 The difference between Kasulu and Ngara districts is not statistically significant 24 A list of refugee camps by district, market access, and ethnicity is provided in appendix 4 25 The difference is statistically significant at p<.05 23

7

Market access was divided into four descriptive categories on the basis of type of markets (external vs. internal) and a subjective quality assessment based on available commodities26. The category of ‘very good access’, defined as access to external markets with a wide variety of commodities, is limited to Congolese households. The categories ‘poor access’, defined as limited internal markets, and ‘no access’ are comprised solely of Burundian households. Figure 3 depicts market access by nationality. Figure 3 - Market Access by Nationality 70%

61%

60%

Percentage

60% 50% 40% 30%

Congolese

39%

Burundian 23% 17%

20% 10% 0%

Very Good Good Access Poor Access No Access (no Access (external (internal markets)* (limited supply external or internal and internal internal markets) markets) markets) Market Access Description (*The difference is statistically significant at p<.05)

Comparing groups defined by market access suggests that those households with ‘very good market access’ are less food insecure than those whose market access is limited to ‘poor access’ or ‘no access’27 (figure 4). However, given the fact that households with ‘very good access’ to markets are exclusively Congolese and those with ‘poor access’ and ‘no access’ are exclusively Burundian, it is only possible to make CSI comparisons by nationality of origin and market access for those households with ‘good access’ to markets. Within this group, Congolese households are less food insecure28. Given that market access is subject to change and nationality of origin is not, it is likely that future CSI surveys will help to clarify the discrete relationship of each of these factors to household food security status29. 26

Each camp was classified into one of these four market access categories by WFP program staff during the CSI Training Workshop in May, 2004. Although market access indicators were collected during the household survey and those indicating access to external markets were less food insecure than those who did not (p<.05), these indicators are more likely to reflect a combination of market access and utilization. WFP program staff deemed that market access is approximately uniform within each refugee camp. Therefore, defining this variable at the camp level is intended to isolate access, regardless of whether or not households choose to utilize these markets. 27 The difference is statistically significant at p<.05 28 The difference is statistically significant at p<.05 29 Both market access and nationality are defined at the camp level (e.g. each camp was assigned one value for market access and one value for nationality such that these variables are constant for households within each camp). The high level of covariance depicted in figures 3 and 4 (Pearson’s correlation = .786) between nationality and market access variables makes it statistically impossible to

8

Figure 4 - Mean CSI Score by Nationality and Market Access 70

61

61

58

Mean CSI Score

60 46

50 40

37

Congolese Burundian

30 20 10 0 Very Good Access Good Access (external and (internal markets)* internal markets)

Poor Access (limited supply internal markets)

No Access (no external or internal markets)

Market Access Description (*The difference is statistically significant at p<.05)

Female and male headed households had approximately the same level of food insecurity. However, the level of food insecurity varies by the educational attainment of the head of household. Households whose head of household has either no or primary level educational attainment are more food insecure than those with secondary or higher educational attainment30. Figure 5 displays the means for these two groups.

Figure 5 - Mean CSI Score by Head of Household's Educational Attainment Level

Mean CSI Score

Interestingly, however, there is no association between head of household educational attainment levels and the CSI food insecurity measure when Congolese and Burundian households are analyzed separately. This is due to the fact that Congolese households, which we have already shown to be less food insecure, also have much higher head of household educational attainment levels than Burundian households (figure 6). Therefore, it is the relationship between nationality and head of household educational attainment that drives the

60 50 40 30 20 10 0

56 43

*None or Primary

*Secondary or Higher

Educational Attainment (*The difference is statistically significant at .05)

quantify the independent effect of each on CSI score when both are considered together except for a very small sub-group with ‘good access’ to markets. 30 The difference is statistically significant at p<.05

9

superficial relationship between head of household educational attainment and CSI score.

Percentage

Household size is also associated with household CSI score: the larger the household, the more food insecure the household31. Households with less than six members32 had an estimated mean CSI score of 49 and households with six or more members had an estimated mean CSI score of 5733. This association holds true for both Congolese and Burundian households.

Figure 6 - Educational Attainment of the Head of Household by Nationaltiy 60%

Congolese

50%

Burundian

40% 30% 20% 10% 0% None

Primary

Secondary Secondary +

Sources of household income34 is Educational Attainment also an important factor for differentiating food security status (figure 7). Households that sold crops that they produced and households that sold livestock or livestock products that they produced were both less food insecure

Mean CSI Score

Figure 7 - Mean CSI Score by Income Sources 80

Yes

70

No

60 50 40 30 20 10 0 Selling Crops*

Selling Livestock*

Petty Trade*

NGO Incentive Earner*

Ag. Labor Ag. Labor for for Refugees* Tanzanians*

Casual labor*

Income Source (*The difference is statistically significant at p<.05) 31

The association between the continuous variable for number of household members and CSI score is statistically significant at p<.05 32 The mean number of household members (5.88) was used as a cut-off point for transforming the continuous ‘number of household members’ variable into a dichotomous variable. 33 The difference is statistically significant at p<.05 34 The ‘sources of income’ variable reflects whether or not a household indicated income from each source, but not which income sources were the ‘main’ income sources for the household.

10

than households that did not. Similarly households involved in petty trade and those with members working as incentive earners with NGOs were less food insecure than those without income from these sources. Conversely, those household’s that derived income from agricultural or casual labor were more food insecure than those that did not indicate these as a sources of income. As with other household characteristics, the relationship between income sources and food insecurity is not uniform for Burundian and Congolese households. This is due primarily to the fact that Burundian and Congolese households utilize the sources of income depicted in figure 7 to different degrees. A higher percentage of Burundian households rely on agricultural labor, whereas a higher percentage of Congolese households rely on selling crops, petty trade, and selling part of the ration35. Figure 8 illustrates sources of income by nationality of origin. Figure 8 - Income Sources by Nationality

Percentage of HH

60% Burundian

50%

Congolese

40% 30% 20% 10% 0% Ag. Labor*

Causal Casual Labor Labor

Selling Crops*

Selling Livestock/ Livestock Products

Petty Trade*

IGAs

Selling Part NGO of the Incentive Ration* Worker

Income Sources (*The difference is statistically significant at p<.05)

Given these differences in income sources, separate analyses of the association between income sources and the CSI food security indicator were conducted for Congolese and Burundian households. Petty trade, involvement in income generating activities, and being an NGO incentive earner, are associated with lower levels of food insecurity among Congolese households36. For Burundian households, selling livestock and being an NGO incentive earner are associated with lower levels of food insecurity37. The association between involvement in agricultural labor and casual labor and higher levels of food insecurity described earlier only holds true among Burundian 35

Selling part of the ration for income has no association with mean CSI score. The difference in mean CSI for Congolese households involved in Petty trade or NGO incentive workers versus those not involved in these activities is statistically significant at p<.05. The difference in mean CSI for Congolese households between those involved in IGAs and those not involved in IGAs is statistically significant at p<.05. 37 All differences are statistically significant at p<.05. 36

11

households. Applying a similar analysis to household food sources suggests that there is no association between food sources and food security when Congolese and Burundian households are considered separately. Multivariate CSI Analysis The following section examines the predictive relationship between multiple household characteristics and the CSI measure of food insecurity. Box 2 – Using Multivariate Analysis to Understand Why Households are Food Insecure The previous section used simple comparisons of mean CSI scores to contrast the relative levels of food insecurity between groups defined by one or more characteristics. For example, on average households whose household head had an educational attainment level of none or primary are more food insecure than those with secondary or higher educational attainment levels. The analysis also showed that, although this approach describes who is comparatively food insecure, it does not provide a sound basis for concluding why they are food insecure. The relationship between educational attainment of the head of household and food insecurity provides an illustration. The association between educational attainment of the head of household and food insecurity is driven by the fact that Congolese households have both higher levels of head of household educational attainment and Congolese households have lower levels of food insecurity. Controlling for the effect of nationality of origin suggests that there is no association between head of household educational attainment and the CSI measure of food insecurity.

Head of HH Educational Attainment CSI Measure of Food Insecurity

Nationality of Origin Multivariate regression analysis provides a means of controlling for the effect of three or more explanatory variables at one time. As a result it allows for an estimation of the magnitude of the net effect of each variable independently and an assessment of the total explanatory/ predictive power of all variables contained in the regression model. The results, combined with sound theoretical backing, can provide suggestive evidence concerning causal relationships.

Due to the covariance between ‘market access’ and ‘nationality of origin’38 it is not possible to include both in the same regression model. Therefore, two separate models are presented. The statistical output for each model is presented in appendix 5.

38

Because ‘very good access’ predicts Congolese nationality and ‘poor access’ and ‘no access’ predicts Burundian nationality 100% of the time, the regression model considers the inclusion of both variables redundant information.

12

Model 1 – Market Access, Household Size, and Income Sources The first model considers CSI score as a function of market access, household size, and the following sources of income: agricultural labor, selling livestock or livestock products, and working as an incentive earner for an NGO. The net estimated effect of each variable on CSI score39 is presented in figure 9. Figure 9 - Model 1 - Net Effect of Explanatory Variables on CSI Score

15.0

15.0

10.0

11.8 9.0

CSI points

9.0

5.0 0.0 -5.0 -11.1

-10.0

-15.6

-16.7

-15.0 -20.0

M arket A ccess - very go o d, go o d, po o r, no ne

Ho useho ld size categories

Selling livesto ck o r livesto ck pro ducts

Incentive earner fo r NGO

P etty trade (fo od and NFIs)

A g. labo r fo r Tanzanians

A g. labo r fo r refugees

Explanatory Variables (* The net effect of market access reflects a change from 'very good' to 'no' access. The estimated change in CSI per unit change in Market Access is 5.0 CSI points)

The amount of variance in the CSI measure of food insecurity explained by the variables included in model 1 is approximately 16.4%40. This moderate level of explanatory power suggests that one or more variables with a strong predictive relationship to CSI scores were absent in the survey and therefore could not be included in the model. Model 2 – Nationality of Origin, Household Size, and Income Sources The second model considers CSI score as a function of nationality, household size, and the following sources of income: agricultural labor, selling livestock or livestock products, and working as an incentive earner for an NGO. The net estimated effect of each variable on CSI score41 is presented in figure 10. Notably, the substitution of nationality for market access has little effect on other variables in the model (direction and magnitude), suggesting that the two are approximately interchangeable42. 39

All variables included in the model are statistically significant at p<.05 R-square = .164 41 All variables included in the model are statistically significant at p<.05 42 This relationship is expected given the degree of covariance demonstrated in the previous section. 40

13

Furthermore, the magnitude of estimated effect of nationality in model 2 and market access in model 1 are approximately equal. The amount of variance in the CSI measure of food insecurity explained by the variables included in model 2 is approximately 18.3%43. Although slightly higher than model 1, the level of explanatory power remains moderate. Again, this suggests that one or more variables with a strong predictive relationship to CSI scores were not included in the model. Figure 10 - Model 2 - Net Effect of Explanatory Variables on CSI Score

15.0

14.7

10.0

10.3

CSI points

9.4

10.5

5.0 0.0 -5.0

-8.5

-10.0

-15.2 -16.3

-15.0 -20.0

B urundian

Ho useho ld size catego ries

Selling livesto ck o r livesto ck pro ducts

Incentive earner fo r NGO

P etty trade (fo o d and NFIs)

A g. labo r fo r Tanzanians

Ag. labo r fo r refugees

Explanatory Variables

Interpretation of Multivariate Analysis Despite having only moderate explanatory power, some causal propositions can be made concerning the variables included in models 1 and 2 and their relationship to food insecurity measured by the CSI. •

Differentiating the effect of Market Access and Nationality - Due to the covariance between market access and nationality, determining whether or not the difference in food security status between the Congolese and Burundians is caused by market access, by an unspecified factor associated with nationality, or some combination of these factors is not possible.

•

Market Access - The causal association between good access to markets and lower levels of food insecurity is predictable, so much so that WFP program staff used

43

R-square = .183

14

market access as a basis for stratification of the sample.44 Although it is not possible to establish trends with a single data point, the strong association between lack of household access to external markets and mean CSI scores (model 1) suggests that restrictions on movement and closure of markets in recent months have led to an increase in the frequency and severity of consumption coping strategies among refugees. This, in turn, suggests a deteriorating food security situation among households who have been cut-off from external markets. •

Nationality of Origin - A causal explanation for the link between nationality and food insecurity (model 2) is less evident than for market access. As nationality itself has no inherent value related to food security, the challenge for future CSI survey rounds is to attempt to identify the unspecified factor or factors that are associated with both nationality and food security.

•

Household size – Food rations are distributed on the basis of the number of household members as noted on refugee ration cards. Therefore, the relationship between household size and food insecurity (more people, more food insecure), while common in settings where average family size rations are distributed to all households or populations are not receiving food rations, seems counter-intuitive in this setting. One possibility is that larger households have more unproductive household members in relation to productive household members and therefore less income and food production capacity per household member. The proposition that larger households have a higher ‘child under 5 to adult ratio’ appears to be true45, but this same ratio does not exhibit any relationship to the CSI measure of food insecurity. An alternative explanation is that household’s with more members are more likely to have relatives or other ‘hangers-on’ that are not registered on the ration card. As a result, the per person ration is reduced. However, information on actual household size versus registered household size was not collected during the survey46, preventing testing of this proposition.

•

Income: Livestock Production - Only a small percentage of households indicated livestock production as a source of income (5.6%), suggesting that income from this source is limited to better-off households. In addition to contributing to household income and therefore reducing food insecurity, livestock production is also likely to be a proxy indicator of relative wealth.

•

Income: NGO Incentive Earner - Similarly, only 12% of households indicated being an NGO incentive earner as a source of income. Given the vagaries of agricultural and casual labor47, employment as an incentive earner provides households with a

44

The purpose in doing so was to ensure precise and comparable estimates of food security for those with access to external markets and those without. 45 The difference is statistically significant at p<.05 46 It was thought that this would give the appearance that the survey had a direct relationship to registration and would compromise data quality. 47 Demand and labor price is likely to fluctuate much more for private sector labor

15

relatively stable source of income and, as a result, increases household food security. •

Income: Petty Trade - Although the estimated effect of being a petty trader on reducing food insecurity is substantially less than the estimated effect of selling livestock and livestock products or being an NGO incentive earner, the effect is still positive. As with the livestock producers and incentive earners, more income translates into lower levels of food insecurity.

•

Income: Agricultural Labor (for Refugees and Tanzanians) - The lack of capital requirements involved in working as an agricultural laborer provides one possible explanation for the association between households with income from this source and higher levels of food insecurity. Put simply, working as a laborer is the only available income generating option for poorer households that lack capital and inputs and these poorer households are more food insecure48. Furthermore, the fact that many of the agricultural labor opportunities lay outside of the refugee camps also supports this explanation. Poorer households that are more food insecure are likely to be more willing to tolerate the risks associated with illegal exit and entry from the camps. The fact that working for Tanzanian farmers outside of the camps exhibits a larger magnitude of association with the CSI measure of food insecurity than working for refugees provides additional evidence in this regard. However, even working for refugee farmers may require illegal travel outside of the refugee camps.

CONCLUSION AND RECOMMENDATIONS Recommendations for Time Series CSI Comparisons The primary purpose of the CSI baseline study was to establish the initial point of comparison that will be used to gauge changes in the food security status of the refugee population in response to shocks to their livelihoods and access to food. Potential shocks, include, but are not limited to: • • • • • • •

Natural disasters Reductions in the food ration Food pipeline interruption Changes in Government of Tanzania policy Movement restrictions Market restrictions Substantial changes in market demand or terms of trade

1. In addition to assessing overall improvements or declines in food security among refugees, future CSI survey rounds should also examine whether or not the

48

The relationship between agricultural labor as a source of income and food insecurity is likely to be driven by the independent relationship of relative poverty to this income source and relative poverty to food insecurity.

16

comparatively food insecure sub-groups identified in this report are disproportionately affected by a particular shock or combination of shocks. 2. Furthermore, it is likely that the sub-groups affected will change over time depending on the type, duration, and severity of future shocks (e.g. some groups are more vulnerable to particular shocks than others). Therefore, subsequent survey rounds should also be used to identify emergent sub-groups that are disproportionately affected. 3. It is recommended that CSI surveys be scheduled at regular 6 month intervals during the first two years so that seasonal trends can be established. Depending on the results, the frequency of CSI survey rounds may be revised to once per year after the first two years. 4. The timing and frequency of CSI surveys should be adjusted in response to perceived shocks or improvements that are likely to have an effect on refugee livelihoods and food security (see partial list above). Doing so will increase the likelihood of capturing changes in the food security status of the refugee population between two points in time. 5. Detailed guidance for conducting future CSI surveys and making analytic comparisons between survey rounds is provided in a supplementary document49. Additionally, a short guidance note on ‘lessons learned’ from the Tanzania CSI is available as an annex to this report (email [email protected] ) and will be useful for both the Tanzania Country Office and others interested in using the CSI food security monitoring tool. Program Recommendations based on the CSI Baseline The CSI monitoring tool is designed to alert program managers to rapid and significant declines in the food security status of the population, allowing for appropriate programmatic responses. It is also designed to capture improvements in the food security status of the population as a means of assessing the efficacy of WFP programs50. Given that the CSI is designed for these types of time series analyses, the number and extent of programmatic recommendations based on the CSI baseline data is limited. Comparatively Food Insecure Sub-Groups The fact that refugees are currently receiving food rations based on 1857 kcal/person/day suggests that there is no reason to presuppose that the average refugee household51 is critically food insecure. However, almost all households52 are currently enacting consumption coping strategies as a means of managing food shortfalls. While 49

Guidelines for CSI Data Collection and Analysis: WFP Assisted Refugees in Western Tanzania. Improvements in refugee food security status may not be solely attributable to WFP programs. Accordingly, an assessment of extraneous factors that have contributed to improving food security will be an important part of program impact analysis. 51 Mean CSI = 53.02 +/- 2.8 CSI points 52 97.7% +/- 1.1 % points 50

17

some coping is likely to be the norm, this does indicate that there is significant room for improving food security among the refugee population. Government of Tanzania restrictions of movement and closure of external markets due to insecurity also provide an identifiable shock. Although it is not possible to quantify the changes in food security status that have occurred as a result of this shock due to the lack of a pre-shock measure, it is clear that those without access to external markets are less food secure. As highlighted in the section entitled Interpretation of Multivariate Analysis, this suggests that the food security status of household’s no longer able to access external markets has deteriorated. In addition, sub-groups defined by other household characteristics have also been shown to be comparatively food insecure. The value of this differentiation for targeting depends on the type of characteristic used to define these groups. For example geographic comparisons are likely to be more useful for targeting than comparisons by income sources. However, WFP is strongly advised against targeting and making conclusions about the impact of WFP programming on the basis of the CSI baseline results alone. The CSI is designed to track changes in food insecurity, allowing for comparisons within subgroups over time and the identification of sub-groups that are disproportionately affected by various shocks. These time-series analyses will provide a more robust basis upon which to target and gauge the impact of WFP programming. The section entitled CSI Comparisons between Groups provides a basis for identifying sub-groups for time series analyses. Improving Food Security Similarly, a recommendation concerning changes in the general ration size, commodity mix or targeted feeding programs on the basis of the cross-sectional CSI baseline results alone is not defensible53. The upcoming 2004 UNHCR/WFP Joint Assessment Mission (JAM) is better suited to this task and, in fact, is designed to make such recommendations. However, two broad recommendations related to camp closures can be made. Rather than new findings, these recommendations are based on what is more aptly described as confirmation of existing anecdotal evidence. •

Closure of external markets and limited rights of movement constrain the ability of refugees to complement the food aid they receive with other foods and non-food items. Although beyond the scope of this report, it is likely that local traders have also suffered from these closures. To the extent possible54, WFP and its partners should advocate for re-opening/reestablishment of markets that facilitate exchange between refugees and the Tanzanian population.

53 54

Once multiple data points are established, CSI will contribute in this regard. It is recognized that WFP’s influence in this regard is limited.

18

•

Nearly 40% of refugee households are involved in agricultural labor, primarily for Tanzanian farmers55. The evidence provided in this report shows that those who do so are relatively food insecure when compared with those who do not. Combined with the fact that these household’s are also burdened with risks associated with illegal travel outside the camps to access labor opportunities, this suggests that agricultural labor is primarily an income source for poorer households who lack the capital and inputs to engage in other income generating activities. If camp closures and movement restrictions remain intact or intensify, alternative income sources (or other income transfers that allow refugees to meet their food and non-food needs) must be identified for refugees who are currently unable to generate income within closed camp economies.

55

The estimated percentage of refugees involved in agricultural labor for Tanzanians is 38%. The estimated percentage of refugees involved in agricultural labor for other refugees is 10%. The estimated percentage of refugees involved in either or both is 40%.

19

Appendix 1 – Sampling A. Sample Size Parameters and Calculation The required sample size was calculated using the following parameters56: The minimum magnitude of change to be detected between strata and within strata overtime in CSI points. . Power – The level of certainty with which it is desired to capture a change/difference of this magnitude.

=

3

=

80%

Confidence Level - The level of statistical assurance that this change/ difference is not due to chance (corresponds with alpha = .05)

=

95%

Design Effect (none)57

=

1

Estimates58 of a). mean b). standard deviation

= =

13.1 14.2

Sample Size (n) required per strata

=

353

10% additional household allowance for incomplete questionnaires

=

35.3

Sample Size (n) taken from each strata

=

390

Total Sample Size (2 strata)

=

780

B. Total Number of Households per Strata, Sampling Interval, and Random Start

Strata 1 Strata 2 Total # of HH Sample Size required Sampling Interval Random Start

“good”

“poor” market access

48,950 390 125 13,067

55,675 390 142 40,429

104,625 780

56

A web-based sample size calculator was used to perform the calculation and is available at the following website: http://calculators.stat.ucla.edu/powercalc/normal/n-2-eq-var-samp.php 57 The multi-stage cluster sampling approach outlined in the CSI Field Methods Manual (CARE/WFP, 2001) was not necessary given that a complete, accurate sampling frame of households was available. 58 These estimates come from previous CSI studies in Kenya and Ghana. The mean for the CSI baseline was 53.05 and the Standard Deviation was 32.99. Note that the scale and index used in the CSI baseline differs from that used in the Kenya and Ghana studies and that they are not directly comparable.

20

Appendix 2 – Survey Protocols and Special Sampling Instructions 1. Household Selection a. The households to be included in the survey have been pre-selected b. Use the address provided to locate the household to be surveyed (see special note for NYARUGUSU, MTABILA 2 AND MUYOVOZI). 2. Respondent Selection a. Once a household has been selected, ask to speak to the head of the household (registered head of household on ration card) b. If the head of household is unavailable, the following respondents may be substituted (in order of preference): i. Spouse of Head of Household ii. Other adult (16 years and above) c. If none of these respondents are present in the household, follow the household replacement strategy described below. 3.

Household Replacement Strategy a. Replace by choosing the next closest household (in any direction):

Household Pre-selected Household Household Replacement Household

b. Be sure to tick the box on the top of the questionnaire to indicate that this is a replacement household: Tick box if replacement household c. If a suitable respondent is not found in the replacement household, repeat the procedure by the next closest household. Do so until a household with an appropriate respondent is identified.

21

4.

Special Sampling Instructions: NYARUGUSU, MTABILA 1, MTABILA 2, MUYOVOZI

a. Once inside the block, determine the approximate center b. Spin a pencil to choose a random walking direction c. Sample every household in that direction until the end of the block is reached or the number of households needed in the block are interviewed. d. If additional households remain and you’ve reached the end of the block, return to the center of the block and re-spin the pencil to choose another walking direction. Example: Solid Line – First Walking Direction Dashed Lines – Second and Third Walking Directions Selected household No respondent, proceed to next household

Approximate Center of Block

End of Block, return to center and re-spin

End of Block, return to center and re-spin

22

Appendix 3 – CSI Questionnaire World Food Programme: Coping Strategies Index Household Questionnaire (W. Tanzania)

first initial

last name

day

Interviewers Name:

month

year

Date:

Tick box if replacement household

District:

Camp:

Ad 1

Ad 2

Ad 3

Plotnum

RESPONDENT INFORMATION 1. Sex: Tick one only

Male

2. Age (in years):

Female

HOUSEHOLD INFORMATION (a household includes those who share a cooking pot) 3. How many people (including all children and adults) are currently living in this household? Clarify: actual number living in the household, not the number of persons on the registration card > 15 years 5-15 years 4. How many males (and how old is each)

Male

5. How many females (and how old is each)

Female

0-5 years

Check to See if total from q. 4 and q. 5 sum to the answer given in q. 3. If no ask respondent to clarify

6. Is this household registered as an EVI household?

Yes

No

Tick one only

7a. Who is the 'head of this household' (the primary decision maker in the household)?

Male

Tick one only

Female 7b. What level of education does this person (head of household) have?

None

Primary

Secondary >Secondary

Tick one only

Bicycle

8. Does your household currently own any

Radio

of the following items?

TV, Video set

Read each aloud and tick all that apply

Cattle Goats

9. How many household members have earned income (cash or food in-kind) for the household during the last 30 days?

0

1

2

3 or more

Tick one only

10. What are the sources of cash income for this household?

tick all that apply

Agricultural Labor for other refugees

Income Generating Activities

Agricultural Labor for Tanzanian farmers

(brickmaking, hairdresser, baskets, etc.)

Causal and Other Labor

Selling part of the food ration

Selling Crops the HH produces

Incentive earner for NGOs

Selling Livestock or Livestock Products

Gifts

Petty Trade/Shop (food and NFI)

Other (

)

23

11. W hat are the sources of the food this household consumes?

tick all that apply

Purchase in market

Food Aid

Household crop production

Gifts

Household livestock prodcution

Other (

)

Labor is paid in food

12. In the last 30 days:

Yes

No

All the time Pretty often

Once in a while

Hardly at all

Never

13 - 14 days 6 - 12 days

2 - 5 days

1 day

0 days

3.5

1

0

a. Are there markets outside the camp in which you purchase and/or sell food and non-food items? b. Are there markets inside the camp in which you purchase and/or sell food and non-food items?

CONSUMPTION COPING STRATEGIES tick one box only in each row 13. Because you did not have enough food or money to buy food, in the past 2 weeks (14 days), how many days has your household had to: a. Borrow food or money (you have to repay) from neighbors, friends, or relatives? b. Purchase food on credit? c. Send household members to eat elsewhere? d. Send household members to beg? e. Limit portion size at mealtimes? f. Restrict consumption of adults in order for small children to eat? g. Reduce number of meals eaten in a day? h. Skip entire days without eating? i. Sell high value, preferred foods to purchase larger quantity of less expensive foods j. Exchange your labour for food (work for food) k. Sell Household Assets or the NFI's the household owns l. Engage in prostitution or theft of food (illegal activities) m. Have some members of the household migrate elsewhere or repatriate (data entry code)

13.5

9

14. How many days ago did you receive your last ration?

Supervisor Signature (initial after reviewing questionnaire):

24

Appendix 4 – Refugee Camps by District, Nationality, and Market Access

District

Camp

Kigoma

Lugufu 1

Lugufu 2

Kasulu

Mtabila 1

Congolese

Congolese

Burundian

Congolese

Karago

Poor Access (limited supply internal markets)

Burundian

Mtendeli

Poor Access (limited supply internal markets)

Burundian

Kanembwa

Poor Access (limited supply internal markets)

Burundian

Nduta

Poor Access (limited supply internal markets)

Burundian

Lukole A

No Access (no external or internal markets)

Burundian

Lukole B

Poor Access (limited supply internal markets)

Burundian

Nyarugusu Muyovosi

Ngara

Nationality of Origin

Good Access (internal markets) Good Access (internal markets) Good Access (internal markets)

Mtabila 2

Kibondo

Market Access very good, good, poor, none Very Good Access (external and internal markets) Very Good Access (external and internal markets) Poor Access (limited supply internal markets)

Burundian

Burundian

25

Appendix 5 - Statistical Output for Models 1 and 2 Model 1

Model Summary Adjusted R Std. Error of Square the Estimate R R Square .405(a) .164 .156 30.20785 a Predictors: (Constant), Ag. labor for refugees, Household size categories, Petty trade (food and NFIs), Incentive earner for NGO, Selling livestock or livestock products, Market Access - very good, good, poor, none, Ag. labor for Tanzanians Model 1

ANOVA(b) Sum of Squares df Mean Square F Sig. Regressio 129659.01 7 18522.716 20.299 .000(a) n 4 Residual 659747.59 723 912.514 6 Total 789406.61 730 1 a Predictors: (Constant), Ag. labor for refugees, Household size categories, Petty trade (food and NFIs), Incentive earner for NGO, Selling livestock or livestock products, Market Access - very good, good, poor, none, Ag. labor for Tanzanians b Dependent Variable: CSI Score Model 1

Coefficientsa

Model 1

(Constant) Market Access - very good, good, poor, none Household size categories Selling livestock or livestock products Petty trade (food and NFIs) Incentive earner for NGO Ag. labor for Tanzanians Ag. labor for refugees

Unstandardized Coefficients B Std. Error 26.120 4.874

Standardized Coefficients Beta

t 5.359

Sig. .000

5.003

1.343

.145

3.724

.000

9.000

2.250

.137

4.000

.000

-15.556

5.687

-.094

-2.736

.006

-11.108

3.559

-.108

-3.121

.002

-16.668 11.815 9.037

3.537 2.692 3.992

-.164 .174 .080

-4.712 4.388 2.264

.000 .000 .024

a. Dependent Variable: CSI Score

26

Model 2 Model Summary Model 1

R R Square .428a .183

Adjusted R Square .176

Std. Error of the Estimate 29.85944

a. Predictors: (Constant), Nationality of Origin, Selling livestock or livestock products, Ag. labor for refugees, Household size categories, Incentive earner for NGO, Petty trade (food and NFIs), Ag. labor for Tanzanians

ANOVAb Model 1

Regression Residual Total

Sum of Squares 144789.8 644616.8 789406.6

df 7 723 730

Mean Square 20684.263 891.586

F 23.199

Sig. .000a

a. Predictors: (Constant), Nationality of Origin, Selling livestock or livestock products, Ag. labor for refugees, Household size categories, Incentive earner for NGO, Petty trade (food and NFIs), Ag. labor for Tanzanians b. Dependent Variable: CSI Score

Coefficientsa

Model 1

(Constant) Household size categories Selling livestock or livestock products Petty trade (food and NFIs) Incentive earner for NGO Ag. labor for Tanzanians Ag. labor for refugees Nationality of Origin

Unstandardized Coefficients B Std. Error 13.392 5.719

Standardized Coefficients Beta

t 2.342

Sig. .019

9.431

2.224

.143

4.241

.000

-15.162

5.621

-.092

-2.697

.007

-8.490

3.575

-.083

-2.375

.018

-16.323 10.335 10.446 14.655

3.470 2.615 3.957 2.625

-.161 .152 .093 .213

-4.704 3.952 2.640 5.583

.000 .000 .008 .000

a. Dependent Variable: CSI Score

27