These figures show that most of deceased due to HIV/AIDS are 22-45 years old males and 15-50 years old females. This observation and the fact that age 15-50 are main labor for household production are the two main reasons why we set the age range to be from 15 to 50. Another reason why we set the upper bound of the age range at 50 is that KHDS did not ask mortality or illness for below 15 or above 50 when KHDS chose sample households. As we discuss in the following subsection, 33% of prime-age adult mortality in the data is enumerated when KHDS chooses sample households. We need to set the upper bound at 50 or less to include these data into our analysis consistently.Here, we show the characteristics of prime-age adult mortality in the data. There are 6,681 individuals are surveyed in wave 1, 2, 3, or 4 . Out of these 6,681 individuals, 988 died between 1991 and 2004 and their deaths are recorded in the KHDS. Note that since wave 5 in 2003 asks mortality only for individuals who were household members in wave 1-4 , there can be other deaths which are not recorded in the KHDS. While these 6,681 individuals have individual ID for KHDS, KHDS records other 377 individuals who do not have individual ID since some of them died in the 12 months just before wave 1 and others joined a survey household and died between waves. Thus, KHDS records the details of total 1,365 deaths. Among 1,365 deaths, 844 deaths are deaths of individuals whose ages are between 15 and 50 when they died. Out of these 844 prime-age adult deaths, 743 deaths are as the result of illness. Out of these 743 illnesses, 398 illnesses are diagnosed by a health professional and 188 are reported as HIV/AIDS. Thus, 47.2% of diagnosed illnesses are reported as HIV/AIDS. KHDS also asks a respondent in a household what illness the respondent think the died person was suffering from. Out of 743 illnesses, 36.7% illnesses are thought as HIV/AIDS. Out of 844 prime-age adult deaths, 32% deaths are due to HIV/AIDS although respondents may not have enough knowledge about health to understand the cause of death correctly.
As mentioned above, KHDS intended to sample households hit by adult mortality more than other households. KHDS calls the sampling stage before main survey as “enumeration”. The enumeration before wave 1 asks whether any adult with age of 15-50 has died in the past 12 months. Then, if so,strawberry gutter system it asks the ages of each adult and the cause of the death. The cause of the death has only 4 categories: illness, accident, child birth, and other. It does not ask gender of each adult nor any further individual characteristics. The enumeration recorded 499 deaths. We checked the duplication of deaths between one in the enumeration and one in wave 1. The enumeration was implemented between March 15 and June 13, 1991 while wave 1 was implemented between September 30, 1991 and May 10, 19922. We found 83 duplications although we could rely on only household ID and the age of died adult to find duplications. Thus, the enumeration before wave 1 provides information on 416 adult deaths. Figure 9 shows the age distribution of these died adults. Out of these 416 died adults, 413 adults died due to illness. Figure 10 shows the age distribution of these adults died due to illness.We think we should include these mortality in analysis since our focus is effects of adult mortality and there are huge numbers of adult mortality in the enumeration and before wave13. As we mentioned in the previous subsection, one of the reasons why we set upper limit of prime-age adult at 50 is that the enumeration does not record mortality of individuals whose ages are more than 50. The reasons why we do not distinguish adult mortality due to HIV/AIDS and one due to other causes are the sample size is not so large, whether the cause is HIV/AIDS is not clear, and the enumeration does not ask whether the cause is HIV/AIDS. Previous studies mentioned that HIV/AIDS is more harmful than other mortality or illness since a household suffers from the longer period of sick before death and other members’ care for the sick. Since we do not think we have proper data to study the difference in the effects of HIV/AIDS and those of other illness and mortality, we focus on the effects of prime-age adult mortality on long-term agricultural production. Table 1 shows the number of prime-age of adult deaths by cause and by year. Most of deaths recoded in the data are in 1990 and 1991. This characteristic is due to KHDS’s unique sampling strategies. First, KHDS intentionally sample households which suffered from prime-age adult mortality, more precisely, 14 out of 16 households have prime-age adult mortality in the last 12 months, prime-age adult who is too sick to work or both in the enumeration. Second, in wave 5 , KHDS does not ask death of individuals who were not household members in previous waves even if an individual was a household member when he or she deceased.
We should take into account that even we call prime-age adult mortality between 1990 and 2003, most of death occurred in 1990 and 1991. Table 2 shows the number of households by year and by number of prime-age adult death. As we explain in Section 4.3, we use 401 households out of all households in the original data. There are households which suffer multiple deaths. The number of households which has 0, 1, 2, 3, 4, 5, and 6 deaths are 152, 117, 82, 38, 10, 1, and 1, respectively as shown in Table 2. 56% households have prime-age adult mortality between 1990 and 2003. This table also show that most of prime-age adult death in the data occurred in 1990 and 1991, which is due to KHDS’s sample selection scheme as mentioned above. Wave 5 of KHDS asked households whether each of the past ten years was a very bad year or not, if so, why it was, and if so, how did they cope with it. As the answer to for year 2003, 25% of 376 individual singled out death of family member, 22% did poor harvest due to weather and 20% did serious illness. As the answer to , each individual could answer at most two and there are 525 answers for 2003 from 376 individuals. The content and percentage of each answer is as follows: rely on support from family and friends , reduce consumption , take casual employment , introduce other crops , sell livestock , sell other assets , start other business , start selling processed food , and sell land . These results imply that mortality and illness are the most serious negative economic shock for the households and households respond to it in various ways. We do not study short-term responses although Beegle studies short-term labor responses to prime age adult mortality as mentioned in Section 2. Instead, we study the long-term consequences in agricultural production after being hit by prime-age mortality and responding to it.We need homogeneity in households in the sense that households solve the same or at least a similar economic problem. In this subsection, we discuss what sub-sample of households we choose from the original data. In summary, we choose households which engage in agriculture mainly and we exclude households which emigrate from the original location and new households which split from the original households over a decade from our analysis. Wave 5 of KHDS tracks households and their members who emigrated between 94 and 03. However, investigators do not ask those emigrated households about their agriculture less than non-emigrated households in order to reduce work load for tracking phase and thus the data on agriculture are much less complete compared to non-emigrated households. Since the data on agricultural outputs and productive assets for emigrated households are not collected, we simply drop emigrated households from our analysis. Unfortunately, the number of emigrated household are large: there are 1,413 emigrated households out of all 2,774 households in 2003.
However, we should not say 51% households emigrated. First, these 2,774 households in 2003 includes split households from the original 919 households in 1991 and 1992. Second, 540 out of 1,413 emigrated households emigrated to nearby villages. If we take household unit in 1992, total 830 households are resurveyed in 20034. Out of them, 733 households have at least one new household unit which remained in the same village. 46 households do not have any new household units which remained in the same village but have at least one new household unit which emigrated to a nearby village. The remaining 51 households emigrated in the most restricted definition, that is, do not have any new household units which remained in the same village or emigrated to a nearby village. We exclude households in the most urbanized four clusters since the model does not have occupational choice and poverty dynamics in urban area is very different from the one in rural area we study. The ratio of employment income compared to agricultural income increased a lot in these four most urbanized clusters from 1994 to 2003. Although one fourth of households in wave 1 live in urban zone as mentioned above,hydroponic fodder system we include households in urban zone except households in the most urbanized four clusters since urban zone except the most urbanized four clusters seems to be as agriculture-oriented as other zones in 1991-19945. We drop 55, 51, and 41 households in these four clusters in 1991, 1992, and 2003, respectively. In order to focus on agricultural households, we drop households whose non-agricultural income or transfer income is larger than agricultural income.We exclude households which split from the original household between 1992 and 2003 and which do not seem to be continuing households from 1992. More particularly, we exclude the following households: If there is a main household where household head is the same over 1992 and 2003 and there is another household which was split from the main household between 1992 and 2003, for example, a son’s new household, we exclude the split household and focus on the main household.
If a household head passed away between 1992 and 2003 and there are two households in 2003, for example, older brother’s new household and younger brother’s new household, we choose only one household as the continuing household and exclude the other household from our analysis. Table 3 shows the results of this selection of households. See Appendix A.1 for the detail on how to choose a continuing household.In this subsection, we discuss the relevancy of our specification of agricultural production function . We use the sub-sample of households whose income is mainly from agriculture for our analysis. We think household members, land and livestock are the three main productive factors/assets for the agricultural production in Kagera region. We use the number of household members instead of labor hour input into agricultural production. Although main labor input is household member’s labor, some household use hired labor. For example, in the original KHDS data, 26% of and 33.3% of households used hired labor on their shamba in the past 12 month in wave 1 and wave 5 , respectively. Also, 10.9% of households used paid labor for herding in the past 12 month in wave 5 . In order to control this heterogeneity among households, we subtract the cost of hired labor from agricultural output/sale. We exclude a household from analysis if its agricultural income is smaller than non-agricultural income in order to focus on household income generation with subsistence agriculture. Although we do not take into account 1) that household members use some labor hours in non-agricultural activity and 2) the differences in gender and age among household members, we do not think it is a shortcoming for our purpose. Our objective is to understand the effects of prime-age adult mortality on long term income generating power of subsistence agricultural households and production function is a reduced form of household income generation.