Income Diversification Among Farming Households Headed by Women in Rural Kenya

ABSTRACT

This article discusses barriers to women’s economic empowerment and opportunities for households headed by women to diversify incomes in the rural parts of Kenya. The study analyzes the full range of income-generating activities at the household level and also accounts for the different types of female-headed households, each of which face different constraints. The findings show that not only do female-headed households diversify and combine their incomes differently than male-headed households but also that there are differences among different groups of female-headed households.

HIGHLIGHTS

• Increasing women’s economic empowerment requires identifying barriers to and facilitators of women’s opportunities to diversify income.

• In rural Kenya, female-headed households (FHHs) are heterogeneous when it comes to how they diversify their incomes.

• FHHs in rural Kenya are more reliant on income from transfers than male-headed households.

• FHHs receive a smaller share of their earned income from the nonagricultural sector.

• FHHs are more dependent on work on own farm than MHHs.

• It is important for future research to account for all types of FHHs.

KEYWORDS:

• Female-headed household
• income diversification
• livelihood
• Kenya

JEL Codes:

• D1
• I30
• Q1

INTRODUCTION

Although we do not know the full implications of the pandemic, its effects have already shown that African households are vulnerable to external shocks. Lockdown measures and the global economic recession have had dramatic effects on labor markets and particularly on vulnerable groups, such as impoverished households, informal sector workers, women, and youth (World Bank 2021). One way to increase income while decreasing household vulnerability to shocks is to take advantage of different, diversified sources of income (Ellis 2000; Freeman, Ellis, and Allison 2004; Barrett et al. 2005; Christiaensen and Subbarao 2005; Kristjanson et al. 2010; Bigsten and Tengstam 2011; Hoang Trung, Phamb, and Ulubaşoğlu 2014). To take advantage of these more attractive income strategies, households need to overcome barriers to entry, such as insufficient skills, contacts, and capital (Barrett, Reardon, and Webb 2001). These constraints are closely linked to gender.

Addressing inequalities at the microeconomic level is important to achieve high sustained economic growth. Neoclassical growth models that incorporate gender inequality in education show that gender inequality in education directly affects economic growth by lowering the average level of human capital (Klasen 2002). Gender education inequalities also affect economic growth indirectly through their impacts on fertility and labor market opportunities (Seguino 2020).¹ A less-recognized channel is that gender inequalities also affect total factor productivity, the most important factor affecting income differences between countries. When a large part of a labor force is working in low-productivity activities, the overall productivity in an economy is bound to be low. This is not only devastating for women workers but also leads to large welfare losses for countries in aggregate terms.²

Despite certain favorable trends, such as the expansion of women’s education, declines in fertility and higher per capita incomes, women are still more likely to work and remain in the agriculture sector (Fox and Thomas 2016) and/or small-scale household service work, primarily as family workers (care work). Even if women are able to enter into nonfarm employment (whether self-employed or as an employee), social norms enforce occupational and sectoral segregation (Fox, Senbet, and Simbanegavi 2016; Borrowman and Klasen 2020).

In this article, we analyze income diversification among households headed by women in the rural parts of Kenya. Specifically, we examine (1) whether income among female-headed households is less diversified than in male-headed households, (2) whether female-headed households diversify their income differently than male-headed households, and (3) which factors determine how female-headed households diversify their incomes.

However, female-headed households represent a heterogeneous group with different reasons for having household heads who are women, and they face different constraints that determine their access to sources of income generation. Thereby, including only a dummy variable to capture the gender of the head of household, may yield misleading results. This approach has been shown to give the sometimes incorrect conclusion that female-headed households are “the poorest of the poor” (Chant 2008). Furthermore, we find that dividing households into groups with married and unmarried heads is not sufficient. Instead, we control for all types of marital status (monogamously married, polygamously married, divorced, widowed, or never married), and whether there is an adult man present in a household. This is especially important in the Kenyan context, where there is a long tradition of household members migrating to work outside the household. Even though these members do not live in the household, they might still contribute to its income by sending remittances, thereby becoming part of the household’s overall income-diversification strategy. To the best of our knowledge, no previous work on income diversification has used this refined distinction.

LITERATURE REVIEW

Income-diversification strategies

Income diversification can occur out of necessity or by choice (Ellis 2000). Diversification out of necessity refers to situations in which the income from a household’s farm production is not enough to sustain an acceptable standard of living. Diversification by choice refers to voluntary reasons for diversification, which are often linked to a desire for higher returns from off-farm activities. Having different income sources can also be considered a risk-coping strategy because diversified households are less vulnerable to economic shocks than undiversified households (Ellis 2000). In Kenya, increased access to nonfarm employment has been identified as an important approach to reduce household vulnerability (Christiaensen and Subbarao 2005). To capture different forms of diversification, it is important to consider not only whether a household has a diversified income portfolio but also how it diversifies its income.

Some households rely on farming their own land and investing in livestock. This group generally consists of two types of households. First, there are impoverished households that due to different constraints do not have access to the labor market and have limited choice but to rely on their own production for survival. Second, there are households for which this livelihood strategy can yield relatively high returns through their high-quality land endowments and adequate market access. Because this approach requires sufficient access to land, it is not possible for many impoverished households, which instead can diversify their incomes by earning wages through working on other farms, an option that does not require schooling and is accessible to individuals without an education.

If a household is able to diversify income via off-farm wage labor, this generally offers a higher return (Davis, Giuseppe, and Zezza 2017). Off-farm labor can also provide more effective insurance against shocks because it is less correlated with the output of a household’s farm. However, there is also considerable diversity in terms of returns within the nonfarm sector. Because of high entry barriers, we expect relatively wealthy households to dominate higher-return nonfarm activities (Woldenhanna and Oskam 2001). A household can also choose to diversify income through nonfarm self-employment. However, starting a business requires capital.

Female-headed households and livelihood diversification

A female-headed household’s diversification options are typically determined by the specific constraints linked to gender. Because female-headed households are generally smaller, they sometimes lack sufficient labor to diversify their income sources. In addition, Christopher Udry (1995) shows that the productivity of plots farmed by women is lower than that of plots farmed by men because less-productive resources are allocated to the plots controlled by women. This gendered difference in productivity can also be explained by the difference in the returns of these resources (Aguilar et al. 2015, Oseni et al. 2015). However, in Kenya, Mwangi wa Githinji, Charalampos Konstantinidis, and Andrew Barenberg (2014) conclude that the difference in productivity is also driven by crop choice, since Kenyan men have a higher probability of producing market-oriented crops than women.

Furthermore, while women are less likely to own land, those who do often own plots that are smaller and of lower quality than those owned by men. In nine of fourteen countries (in the RIGA database), landholdings were smaller among female-headed households than among male-headed households (FAO 2011). Formal law in Kenya allows women to both inherit and buy land. However, as in most African countries, land rights are also regulated by customary law (OECD 2010; Ossome 2014), and in practice, only 4 percent of the land is owned by women (OECD 2010).

In addition to a lack of access to productive resources, female-headed households sometimes lack access to credit. This is especially important if a household wants to diversify into nonfarm self-employment. In the abovementioned survey, rural, female-headed households were less likely to use credit in seven of nine countries (FAO 2011). In Kenya, this is primarily a problem for impoverished women and can be explained by the fact that they do not often have sufficient assets to use as collateral (OECD 2010).

Furthermore, access to high-return off-farm wage employment requires a certain level of human capital (Escobal 2001; Woldenhanna and Oskam 2001; Bigsten and Tengstam 2011; Senadza 2014). Although the gender gap in education has decreased in most developing countries, among older female heads of household, levels of education are lower than their male counterparts. In Kenya, the gender gap in education has narrowed, and in 2016, there were as many girls as boys enrolled in primary education. In secondary education, the number of girls overtook the number of boys in 2019. However, regarding tertiary education, the ratio of enrolled girls to boys is approximately 0.7 (UNESCO 2021). Women might also lack the connections needed to access high-return off-farm wage employment, and certain social norms regarding women’s workforce participation can hinder them from entering a labor market (Mammen and Paxon 2000; De Giusti and Sarada Kambhampati 2016; International Labour Organization 2017). In Kenya, even though women constitute 48 percent of the labor force (World Bank 2018), they account for only 34 percent of wage-earning employees in the formal sector (Kenya National Bureau of Statistics 2017). Furthermore, even when women participate in high-return off-farm wage employment, they tend to be concentrated in specific sectors, such as education and the service industry (Kenya national Bureau of Statistics 2017).

In general, Kenyan women are more likely to work in informal sectors (Wanjala and Were 2009) and in low-return nonfarm activities (Loison 2015). Giovanna De Giusti and Uma Sarada Kambhampati (2016) find that different representations of gender norms in Kenya impact women’s labor force participation. Furthermore, since women often have the primary responsibility in caring for children and other family members, they are also typically constrained by time (International Labour Organization 2017).

However, in households where there is a man present, he may be able to help his family overcome some of these constraints. Even if a man does not live in a household, he might help his family through his access to productive resources, such as land and credit. Migrating to work in the urban parts of the country is common in Kenya and should be viewed as a part of a household’s income-diversification strategy (Bigsten 1996). For example, Benjamin Davis, Stefania Di Giuseppe, and Alberto Zezza (2017) show that among the nine African countries in their study, Kenya had the highest percentage of income shared through this transfer process (19 percent). Nine percent of rural Kenyan households acquired more than 75 percent of their income from urban transfer, making it their most important income source.

Even though it is possible that an entire family can migrate to an urban area, the most common type of migration is when only the husband migrates while the wife stays in the rural area. This norm can be linked to the colonial era, when men were used as inexpensive labor in urban centers. However, those men were only allowed to stay for a limited time and were not paid enough to allow their family to move with them. The women were supposed to stay in rural areas and take care of the land (Kinuthia 2003). Today, there are still more men than women who move to urban areas to work. One explanation for this could be that the income returns from migration to urban jobs are still higher for men (Agesa and Agesa 1999). Another reason could be that in some parts of the country the husband is still viewed as the household breadwinner, so, when forced by economic hardship, he must migrate to find a way to provide for his family. When a husband migrates, the wife becomes the head of the household. By sending remittances to their rural households, men maintain their ties with their rural families. In this way, rural-based households function as safety nets for husbands to return to in case of unemployment or illness (Kinuthia 2003).

To the best of our knowledge, the study most similar to ours is Agnes Andersson Djurfeldt, Göran Djurfeldt, and Johanna Bergman Lodin (2013), who look at differences in farm and nonfarm income between male- and female-headed households. Although their focus is on discrimination in income levels (in eight African countries), some of their conclusions are relevant for our study. They find that households that diversified out of agricultural work have higher incomes than others and that participation in nonfarm sectors is related to higher incomes. They find some evidence that female-headed households are more likely to obtain incomes from both farm and nonfarm sources. However, looking at the share of households with nonfarm incomes in Kenya, they do not observe any statistically significant difference between households headed by women and men.

Although female-headed households are not the main interest of most previous research, the gender of heads of household is often included as a control variable. The results from these studies are mixed, with some finding that female-headed households are less likely to diversify incomes (Block and Webb 2001) and others finding no impact (Escobal 2001). Arne Bigsten and Sven Tengstam (2011) find that female-headed households in Zambia are less likely to diversify into nonagricultural wage work and are more likely to diversify into a business. Christopher B. Barrett et al. (2005) observe that female-headed households in Rwanda are more dependent on unskilled labor than male-headed households. Bernardin Senadza (2014) finds that female-headed households in Ghana are more likely to combine their farming activities with self-employment. However, these studies are limited by the fact that they treat female-headed households as a homogeneous group, without considering the diverse circumstances that can lead them to be female-headed.

In Kenya, Menale Kassie, Simon Wagura Ndiritu, and Jesper Stage (2014) observe that female-headed households with single, widowed, divorced, or separated heads face higher probabilities of food insecurities than female-headed households with married heads whose husbands live elsewhere. This is in line with the hypothesis that husbands, even if they do not live in the household, sometimes still contribute to households’ overall income diversification by sending money. Agnes Andersson Djurfeldt (2018) shows that female-headed households with husbands living elsewhere were more similar to male-headed households in regard to access to land than to other female-headed households.

DATA

The income statistics used in this study come from the RIGA database, which is a joint project between the Food and Agriculture Organization of the United Nations (FAO) and the World Bank to produce aggregated income data that are comparable across countries. For Kenya, the data are aggregated from the Kenya Integrated Household Survey (KIHBS) 2005/06.³ This survey contains 13,430 randomly selected households from all districts in Kenya. The sample is representative at the national, urban/rural, provincial, and district levels. The RIGA data were merged with the underlying KIHBS data to yield further insights into differences in income diversification.

All aggregates are calculated at the household level. Wage employment is disaggregated by industry according to the United Nations International Standards and Industrial Classification of All Economic Activities (ISIC). These categories are then aggregated into nonagricultural and agricultural wage earnings. Earnings from wage employment are net and include all in-cash and in-kind benefits received from employers. Farm income includes crop and livestock production. Crop production includes sales of crops (and crop byproducts), sharecropping earnings, and consumption of homegrown crops. Livestock income includes the sale and barter of livestock and the sale and household consumption of livestock byproducts. Self-employment is defined as work in an enterprise owned by any member of a household that is not directly connected to its agricultural production. Transfers are reported as gross figures and are divided into private and public transfers. Private transfers are primarily remittances but include transfers from private organizations/associations. Social transfers consist of pensions and social benefits (Davis et al. 2009). Total income for household i is given by: $\begin{array}{rl}Incom{e}_i& =Agriculturalfarmincom{e}_i+Agriculturalwageincom{e}_i\\ & +Non-agriculturalwageincom{e}_i\\ & +Selfemploymentincom{e}_i+Transfe{r}_i+Otherincom{e}_i\end{array}$

Descriptive characteristics of households

Our definition of female-headed households comes from self-reported information in the KIHBS. After we cleaned the data, we obtained a final sample of 7,685 households, of which 2,784 (36 percent) were headed by a woman and 4,901 were headed by a man. Among female-headed households, the largest group was households headed by a widow (35.5 percent), followed by a monogamously married woman (22.4 percent),⁴ a polygamously married woman (8.7 percent),⁵ a divorced woman (4.7 percent), and a woman who never married (2.9 percent; Table 1). The largest group among male-headed households (Table 2) was households headed by a monogamously married man (79.5 percent), followed by a polygamously married man (15.3 percent).

Table 1 Household characteristics and structure of income, female-headed households

	Mono. married	Poly. married	Divorced	Widowed	Never married	No male	All
Observations	625	241	132	989	81	716	2784
Distribution (%)	22.4	8.7	4.7	35.5	2.9	25.7	100.0
Age^a	38.34	46.38	44.30	54.58	39.33	43.02	47.43
	(0.56)	(1.17)	(1.24)	(0.55)	(1.37)	(0.67)	(0.41)
Years of education^a	6.35	3.36	3.92	2.88	6.96	4.76	4.24
	(0.20)	(0.32)	(0.42)	(0.16)	(0.53)	(0.20)	(0.12)
Household size	5.06	5.48	4.44	4.98	3.97	4.00	4.99
	(0.09)	(0.19)	(0.32)	(0.09)	(0.22)	(0.09)	(0.06)
Male labor	0.82	0.99	0.63	1.05	0.57	0.00	0.92
	(0.05)	(0.08)	(0.09)	(0.04)	(0.11)	(0.00)	(0.03)
Female labor	1.55	1.65	1.75	1.51	1.50	1.44	1.55
	(0.04)	(0.07)	(0.13)	(0.04)	(0.11)	(0.03)	(0.03)
Land/labor ratio	1.59	1.91	1.27	1.55	0.99	2.32	1.56
	(0.37)	(0.76)	(0.61)	(0.32)	(0.23)	(0.58)	(0.21)
Credit constraint	0.35	0.48	0.50	0.45	0.41	0.40	0.42
	(0.02)	(0.04)	(0.05)	(0.02)	(0.06)	(0.02)	(0.01)
Expenditures per capita	26,090	18,617	21,776	22,017	31,989	23,893	23,307
	(1523)	(972)	(1542)	(653)	(3847)	(1072)	(665)
Farm income	46.39	54.92	47.31	47.52	36.76	46.28	47.51
	(2.15)	(3.36)	(6.15)	(1.87)	(7.48)	(2.19)	(1.40)
Agricultural wage income	4.93	4.07	16.71	5.31	18.01	6.24	6.29
	(0.84)	(1.23)	(6.43)	(0.68)	(4.98)	(1.34)	(0.68)
Total agricultural income	51.32	58.99	64.02	52.84	54.77	52.52	53.80
	(2.18)	(3.32)	(5.43)	(1.80)	(7.57)	(2.17)	(1.36)
Nonagricultural wage income	8.11	6.16	12.80	9.39	7.71	5.14	8.79
	(1.04)	(1.43)	(3.19)	(1.01)	(2.85)	(0.90)	(0.66)
Non-farm self-employment income	4.96	6.00	6.56	7.48	12.48	7.18	6.70
	(0.97)	(1.68)	(2.26)	(1.02)	(3.85)	(1.89)	(0.67)
Total wage and self-employment	13.08	12.16	19.36	16.87	20.20	12.32	15.49
	(1.39)	(2.17)	(3.79)	(1.36)	(4.43)	(1.54)	(0.92)
Transfers	31.72	26.38	16.32	27.48	20.72	32.22	27.70
	(1.92)	(2.79)	(3.61)	(1.48)	(4.73)	(1.90)	(1.16)
Other income	3.88	2.47	0.29	2.81	4.31	2.94	3.01
	(0.57)	(0.82)	(0.15)	(0.54)	(3.94)	(0.68)	(0.38)
Total nonagricultural income	48.68	41.01	35.97	47.16	45.23	47.48	46.20
	(2.18)	(3.32)	(5.43)	(1.79)	(7.57)	(2.17)	(1.37)

Notes: ^arefers to the female household head. Weighted sample data. Mean and standard error. The distribution of income by source is presented as mean of shares.

Table 2 Household characteristics and structure of income, male-headed households

	Mono. married	Poly. married	Divorced	Widowed	Never married	All
Observations	3895	751	75	95	85	4901
Distribution (%)	79.5	15.3	1.5	1.9	1.7	100.0
Age^a	44.79	53.72	42.97	60.20	28.08	45.86
	(0.28)	(0.65)	(1.04)	(1.48)	(1.18)	(0.27)
Years of education^a	6.93	4.32	5.77	3.92	7.08	6.53
	(0.09)	(0.21)	(0.66)	(0.61)	(0.44)	(0.08)
Household size	5.79	6.79	2.65	4.14	1.73	5.76
	(0.06)	(0.15)	(0.32)	(0.26)	(0.19)	(0.05)
Male labor	1.49	1.54	1.32	1.34	1.28	1.49
	(0.02)	(0.05)	(0.08)	(0.13)	(0.08)	(0.02)
Female labor	1.45	1.67	0.26	0.81	0.18	1.42
	(0.02)	(0.05)	(0.07)	(0.10)	(0.05)	(0.02)
Land/labor ratio	0.92	1.18	1.17	1.33	1.05	0.97
	(0.04)	(0.17)	(0.18)	(0.19)	(0.27)	(0.05)
Credit constraint	0.40	0.40	0.54	0.48	0.36	0.41
	(0.01)	(0.01)	(0.07)	(0.06)	(0.05)	(0.01)
Expenditures per capita	24,057	18,090	35,474	28,356	48,804	24,038
	(606)	(636)	(3255)	(3014)	(5168)	(552)
Income structure (% of total income)
Farm income	50.77	55.21	42.06	58.66	40.57	51.14
	(1.11)	(1.97)	(6.46)	(5.13)	(5.90)	(1.00)
Agricultural wage income	6.56	3.23	13.31	3.69	17.91	6.41
	(0.54)	(0.69)	(5.12)	(1.47)	(5.92)	(0.49)
Total agricultural income	57.33	58.44	55.37	62.35	58.48	57.55
	(1.03)	(1.91)	(5.46)	(5.00)	(5.47)	(0.92)
Nonagricultural wage income	16.19	12.50	15.29	11.06	21.21	15.71
	(0.87)	(1.38)	(4.25)	(3.11)	(4.59)	(0.77)
Self-employment income	10.91	10.14	11.04	9.80	4.13	10.67
	(0.76)	(1.33)	(3.42)	(4.06)	(2.11)	(0.67)
Total wage and self-employment	27.10	22.64	26.33	20.86	25.34	26.37
	(1.08)	(1.69)	(5.09)	(4.83)	(4.85)	(0.94)
Transfers	13.91	16.26	15.63	15.82	13.38	14.26
	(0.63)	(1.23)	(4.27)	(3.04)	(3.30)	(0.57)
Other income	1.66	2.66	2.67	0.97	2.80	1.81
	(0.18)	(0.56)	(1.29)	(0.47)	(1.06)	(0.17)
Total nonagricultural income	42.67	41.56	44.63	37.65	41.52	42.45
	(1.03)	(1.91)	(5.46)	(4.99)	(5.47)	(0.92)

Notes: ^arefers to the male household head. Weighted sample data. Mean and standard error. The distribution of income by source is presented as mean of shares.

Comparing socioeconomic characteristics for the different groups, we find that all groups of female heads of household, except those who never married, have, on average, a lower level of education compared to their male counterparts. Female-headed households have, on average, four years of education, while male-headed households have almost seven years of education. However, the size of the difference varies between the groups. For example, for the group of monogamously married heads the average difference between female- and male-headed households is only 0.6 years, while for the group of divorced heads the difference is almost two years. We also find considerable variation within the group of married female-headed households. The monogamously married group has, on average, six years of education, while the corresponding polygamously married group averages three years of education. The group of never-married heads, which, among both female- and male-headed households, is the youngest group, is also the group with the highest average years of education, about seven years of education.

Male-headed households are larger and have on average close to six members, compared to five for female-headed households. Female-headed households have a higher land/labor ratio, as these households, on average, have less labor. Finally, even though expenditure per capita is on average lower in the sample of female-headed households, the difference is rather small and not statistically significant. However, there is a significant variation between groups in both categories. Households headed by a polygamously married woman have, on average, lower per capita expenditure than other female-headed households, while those headed by a never-married woman have the highest.

Do households diversify their incomes?

Tables 1 and 2 show the breakdown of total income by the income source reported as the mean of shares.⁶ On average, households obtain 45 percent of their income from the nonagricultural sector. This estimate is in line with earlier research from Kenya, where 1997 data show that nonfarm income represents 40 percent of total rural household income (Jayne et al. 2003). Compared to other African countries analyzed by Benjamin Davis et al. (2010), it seems that households in Kenya obtain larger portions of their incomes from nonagricultural sectors (Malawi, 23 percent; Madagascar, 23 percent; Ghana, 39 percent; Nigeria, 20 percent).

On average, male-headed households obtain a larger part of their total income from agriculture than female-headed households. The reason why female-headed households obtain such large portions of their incomes from nonagricultural activities is that a large part of their income is received via transfers. Most reliant on income from transfers are female-headed households with monogamously married heads and female-headed households with no male adult present; both receive close to one-third of their incomes from transfers. This pattern is expected; some of these women have husbands who have moved away to work and send money home as part of their household’s income-diversification strategy. However, it is also the case that all-female-headed households, except divorced women, receive a larger share of transfers than equivalent male-headed households suggesting that transfers have a broader target than just husband and wife. To explore whether other structural factors such as altruistic behaviors, prevalence of shocks and poverty can explain allocation of transfers between household groups is an interesting question for further studies.

Regarding wage labor and self-employment income in nonagricultural sectors, male-headed households receive 26 percent of their income from wages and self-employment, while female-headed households receive only 16 percent of their income from these two income sources. However, again, there are a substantial variation within both groups. In female-headed households, those with a divorced or never-married head of household receive the largest share from these sources.

Within the never-married group, male households notably receive a much larger share of their income from wage labor than their female counterparts, even with similar years of education. However, an important difference between the two groups is the average age. Although the never-married female heads are, on average, younger (39 years) than other female heads, they are older than their typical male counterparts (28). The average size of a household is therefore larger among never-married female heads (four family members) than among never-married male heads (close to two members). This partly explains the vast difference in per capita expenditures between the two groups.

Thus far, we have summarized some descriptive statistics for the different types of households. To obtain a more complete picture of their income diversification and its underlying determinants, we need to elaborate further. From this part on we will only focus on earned income.

EMPIRICAL STRATEGY

There is no consensus in the literature about how a level of income diversification should be measured. Therefore, in this article, we use several measures to address the first empirical question: number of income sources (total and per adult), percentage of total (earned) income obtained from nonagricultural sources and the Herfindahl index.⁷ To analyze the number of income sources, we need a model that can address count data. Because our data have a variance smaller than the mean (under-dispersion), we use a model based on the generalized Poisson distribution, which is more effective than the standard Poisson distribution when analyzing under-dispersed data (Harrison, Yang, and Hardin 2012).

In the second empirical section, we analyze the decision to participate in a specific income-generating activity. We use a probit model, in which a household chooses to participate if the expected net utility is positive. A household’s choice to participate in an income-generating activity is expected to be influenced by several variables, including the head of household’s gender and marital status, and other control variables. Several of these are likely endogenous. Although we use different strategies to address this, we do not claim causality, instead we interpret the coefficients simply as correlations.

We are also interested in the combinations of income sources. We divide households into five groups based on their combination of income-generating activities: own farm only, own farm and agricultural wage, own farm and nonagricultural wage, own farm and self-employment, and mixed. This allows a model with five unordered alternatives, where a household is assumed to maximize its utility by choosing a combination of income-generating activities. Here, we use a multinomial logit model to analyze the choice of income combinations, and as above, factors that is expected to influence the outcome include the gender and marital status of a head of household and a number of control variables.⁸ All variables are defined in the Appendix (Table A1).

Entering the off-farm wage sectors often requires a certain level of education (de Janvry and Sadoulet 2001); therefore, the options available to a household will be dependent on their level of human capital. To control for this, we include the highest level of education in a household and the years of education of the head of a household. Working solely on the farm requires a sufficient amount of land. The larger the household, the more land is required to absorb its labor supply. T. Woldenhanna and A. Oskam (2001) find that surplus labor is an important determinant of diversification into off-farm wage employment, and Bigsten and Tengstam (2011) observe that households with more land per laborer are more likely to have full-time farmers and less likely to diversify into either nonagricultural wage labor or agricultural wage labor. Therefore, we choose to include the amount of land per laborer as an explanatory variable. We also include the number of men and women individuals ages 15–59 and the number of children in different age categories.

If a household wishes to diversify into self-employment, we expect that its access to credit is important. Unfortunately, we do not have a direct measure of this. Instead, we create a dummy variable that indicates whether a head of household applied for a loan and was turned down or whether they did not even try to borrow because they believed that the application would be refused, that it was too expensive or too much trouble, that they did not have enough collateral, or that they were simply not aware of any lenders.

A household’s possibilities for diversifying its income can also be influenced by gender norms in general and opportunities for women to participate in a labor market in particular. Previous work on Kenya finds that gender norms are significantly correlated with women’s labor market participation (De Giusti and Sarada Kambhampati 2016). To measure gender norms, we include the percentage of women participating in a labor market; however, a larger value could mean either that social norms make women’s labor market participation acceptable or that a household is located in a district with a high demand for labor. Therefore, we divide the variable by the percentage of men participating in the labor market, creating a variable that can be viewed as a proxy for norms of women’s labor force participation. This variable is calculated at the district level.

However, gender-specific job opportunities may exist, and these opportunities may differ among districts. Furthermore, women who want to participate in the labor market may move to districts with high women’s labor force participation. Approximately 5 percent of the households in our sample had at least one individual who stated that they had moved to a specific district because of a job transfer or to look for a job. To control for this reason, we reestimate our models and exclude these households. Furthermore, we reestimate our model by using an alternative measure of gender norms via data from the Afrobarometer (data from 2005, round 3). These data measure the percentage of individuals in a district who agreed with the statement that women should have equal rights and receive the same treatment as men (instead of agreeing that women have always been subject to traditional laws and customs and should remain so).

In addition, the possibility of diversifying into different income-generating activities depends on where a household is situated. Different locations have varying degrees of market access, different labor market opportunities and diverse suitability’s for farming. We include several variables to control for these characteristics at the district level: mean land per labor, mean wage (logged), percentage of adults who are employed, and distance to market and mean expenditures (logged). Living close to a market might be important for female-headed households since women are sometimes more constrained in terms of time and space; they may be unable to be away from a household for extended periods due to domestic and parental responsibilities. Moreover, to capture other variations in the possibility of income-generating activities, which might vary by location, we also include province dummies. However, because some individuals might move to a specific district or province to obtain a job, start a business, or buy land to farm, we re-estimate our models and exclude these households (approximately 10 percent of the households in our sample).

RESULTS

Are female-headed households less diversified?

We start by examining whether female-headed households are less diversified than male-headed households. We measure the level of diversification by the number of income sources, total and per adult, and the percentage of earned income that comes from nonagricultural sources. Calculating the number of income sources, we find that female- and male-headed households, on average, obtain earned income from 2.53 and 2.55 different sources, respectively (Table 3).⁹ In terms of income sources per adult, female-headed households in total have a somewhat larger number of income sources than male-headed households. This can be explained by the fact that households with a female head are smaller than male-headed households. Female-headed households without men have, on average, more income sources per adult than other female-headed households. Looking at the Herfindahl index, there is no major difference in the overall level of diversification between female- and male-headed households.

Table 3 Level of diversification

Female-headed households	Mono. married	Poly. married	Divorced	Widowed	Never married	No male	All
Number of income sources	2.64	2.42	2.16	2.57	2.23	2.33	2.53
	(0.05)	(0.08)	(0.12)	(0.04)	(0.10)	(0.04)	(0.03)
Number of income sources per adult	1.54	1.25	1.21	1.38	1.37	1.90	1.40
	(0.05)	(0.07)	(0.08)	(0.04)	(0.10)	(0.04)	(0.03)
Herfindal index	0.65	0.67	0.70	0.67	0.65	0.69	0.67
	(0.01)	(0.02)	(0.03)	(0.01)	(0.04)	(0.01)	(0.01)
% from nonagric.	18.7	14.1	24.7	20.8	22.6	15.9	19.7
	(1.99)	(2.53)	(4.95)	(1.61)	(4.63)	(1.98)	(1.20)
Male-headed households
Number of income sources	2.57	2.58	2.31	2.35	2.05		2.55
	(0.02)	(0.05)	(0.12)	(0.09)	(0.13)		(0.02)
Number of income sources per adult	1.05	1.04	1.81	1.49	1.66		1.08
	(0.01)	(0.04)	(0.13)	(0.12)	(0.12)		(0.01)
Herfindal index	0.65	0.64	0.75	0.71	0.75		0.66
	(0.01)	(0.01)	(0.04)	(0.03)	(0.03)		(0.01)
% from nonagric.	30.6	26.8	28.6	22.3	30.0		30.0
	(1.13)	(2.09)	(5.63)	(4.94)	(5.25)		(1.00)

Notes: Weighted data. All calculations are based only on earned income. % from nonagricultural is calculated as the mean of shares.

An alternative way to measure the level of diversification that is often used in the literature is to use the percentage of earned income that comes from outside the agricultural sector (Table 3). We find that all types of female-headed households receive, on average, a smaller portion of their earned income from nonagricultural sectors than households headed by monogamously married men. Households with a polygamously married female head and households without any adult male obtain approximately 14 and 16 percent of their incomes from this source, respectively (compared with 31 percent for households with a monogamously married male head). Descriptive statistics and bivariate correlations show that a larger share of earned income from nonagricultural sectors is associated with higher household per-capita income, indicating that this can be an important way out of poverty.¹⁰

By estimating a generalized Poisson regression that includes the control variables presented above, we find that having a monogamously married female head of household is correlated with an increase in income sources by a factor of 1.094 versus having a monogamously married male head (Table 4).¹¹ Even though the difference is statistically significant the difference is small.

Table 4 Incident-rate-ratio (Poisson regression on number of income sources)

Constant	2.586***	0.685
	(0.047)	(0.365)
Female head
Monogamously married	1.031	1.094***
	(0.023)	(0.019)
Polygamously married	0.928*	0.984
	(0.037)	(0.031)
Divorced	0.848***	0.956
	(0.033)	(0.029)
Widowed	0.973**	1.019
	(0.015)	(0.016)
Never married	0.828***	0.909***
	(0.029)	(0.026)
Male head
Polygamously married	0.993	1.034*
	(0.028)	(0.021)
Divorced	0.878**	1.052
	(0.046)	(0.084)
Widowed	0.879***	0.918**
	(0.041)	(0.035)
Never married	0.812***	1.031
	(0.048)	(0.059)
Controls	No	Yes
Dispersion	−0.496	−0.605
Log pseudolikelihood	−6052770	−5676048

Notes: *, **, *** denote significance at the 1, 5, and 10 percent levels, respectively. Weighted data, standard errors are in parentheses and clustered at the district level.

Do female-headed households diversify differently?

Now, we move to our second question and analyze whether female-headed households diversify their incomes differently than male-headed households. We begin by examining the probability of participating in various income-generating activities. We find that female-headed households, except when the head is divorced or widowed, are less likely to participate in nonagricultural wage labor than households headed by a man (Table 5). A household headed by a monogamously married female head is approximately 5 percentage points less likely to participate in nonagricultural wage work than a household headed by a monogamously married man.

Table 5 Participation in income generating activities, average marginal effects (probit regressions)

	Nonagricultural wage	Agricultural wage	Self-employment	Nonagricultural wage	Agricultural wage	Self-employment
Female head
Monogamously married	−0.083***	−0.018	−0.048**	−0.054***	−0.003	−0.061***
	(0.017)	(0.017)	(0.022)	(0.020)	(0.019)	(0.019)
Polygamously married	−0.130***	−0.009	−0.064**	−0.094***	−0.012	−0.075***
	(0.023)	(0.029)	(0.029)	(0.030)	(0.030)	(0.024)
Divorced	−0.066	0.136*	−0.083**	−0.008	0.146**	−0.060
	(0.043)	(0.072)	(0.040)	(0.044)	(0.063)	(0.037)
Widowed	−0.070***	−0.011	−0.031*	−0.017	−0.002	−0.018
	(0.017)	(0.012)	(0.018)	(0.020)	(0.012)	(0.018)
Never married	−0.137**	0.122*	−0.035	−0.115***	0.099*	−0.106**
	(0.043)	(0.059)	(0.054)	(0.039)	(0.056)	(0.047)
Log pseudolikelihood	−2,384,830	−1,643,580	−2,250,793			−2,150,276
Female head
No adult male in household	−0.171***	−0.024	−0.031	−0.122***	−0.015	−0.039*
	(0.015)	(0.018)	(0.020)	(0.016)	(0.021)	(0.020)
With adult male in household	−0.033**	0.016	−0.049***	−0.009	0.025	−0.040***
	(0.006)	(0.014)	(0.015)	(0.016)	(0.014)	(0.014)
Log pseudolikelihood	−2,363,716	−650,631	−251,508	−2,237,509	−573,947	−152,926
Controls	No	No	No	Yes	Yes	Yes

Notes: *, **, *** denote significance at the 1, 5, and 10 percent levels, respectively. Weighted data and standard errors are clustered at the district level. Comparison group: Monogamously married male head.

The probability of participating in nonagricultural wage labor is lowest for households without any male adult. These households are approximately 12 percentage points less likely to participate in nonagricultural wage work than a household headed by a monogamously married man. Households with a divorced or never-married female head have a higher probability of participating in agricultural wage work than households headed by monogamously married men. Female-headed households with a married head also have lower probabilities of participating in nonfarm self-employment.¹² This is in contrast to previous research that argues female-headed households are more likely to diversify into self-employment or a business (Bigsten and Tengstam 2011; Senadza 2014).

To better understand why a female-headed household might engage in different income-generating activities, we use a probit model to analyze a participation decision (Table 6). We find that access to labor is important to explain participation in nonagricultural wage work. Additionally, gender norms, measured by the gap between women’s and men’s labor market participation at the district level, seem to be important; a smaller gap increases the probability of a female-headed household participating in nonagricultural wage work. Reestimating the model to exclude households that have moved because of a job does not influence this result. However, this result is not robust to using a more general measure of gender norms. This could indicate either that specific labor market participation norms play a role or that our gender norms measure is endogenous. Households with a polygamously married head have a lower probability of participation in nonagricultural wage work. However, a higher level of education for the head of household increases the probability to participate in nonagricultural wage work but decreases the probability that a household participates in agricultural wage labor. Although the effect is small, it is in line with the view that households prefer to diversify into nonagricultural wage work when they can overcome constraints.

Table 6 Participation in income-generating activities for female-headed households (average marginal effects, probit regression)

	Nonagricultural wage	Agricultural wage	Self-employment
Polygamously married	−0.060**	0.017	−0.028
	(0.030)	(0.029)	(0.035)
Divorced	0.035	0.143**	−0.013
	(0.043)	(0.063)	(0.047)
Widowed	0.010	−0.002	0.020
	(0.025)	(0.019)	(0.024)
Never married	−0.044	0.099	0.071
	(0.049)	(0.066)	(0.071)
Age of head	0.002*	−0.001	−0.000
	(0.001)	(0.001)	(0.001)
Years of education, head	0.008*	−0.005*	0.004
	(0.004)	(0.003)	(0.003)
Highest education in household	0.006	−0.002	0.005
	(0.004)	(0.003)	(0.003)
Male labor	0.045***	0.026***	−0.004
	(0.011)	(0.010)	(0.008)
Female labor	0.038***	−0.010	0.018*
	(0.011)	(0.010)	(0.010)
Children aged 0–1	0.028	−0.044***	−0.002
	(0.019)	(0.013)	(0.017)
Children aged 2–5	0.008	0.018	0.020
	(0.012)	(0.015)	(0.016)
Children aged 6–10	−0.019*	0.014	0.011
	(0.010)	(0.012)	(0.010)
Children aged 11–14	0.021	−0.010	0.004
	(0.013)	(0.012)	(0.013)
Credit constraints	−0.023	0.005	0.007
	(0.020)	(0.024)	(0.018)
Land per laborer	−0.001	0.000	−0.002
	(0.002)	(0.001)	(0.001)
Land in district	0.003	0.009*	0.007
	(0.006)	(0.005)	(0.005)
Gender norms	0.030**	0.013	−0.002
	(0.014)	(0.014)	(0.010)
Rate of wage labor	0.772***	0.660***	−0.003
	(0.063)	(0.242)	(0.221)
Distance to market	0.000	−0.000	0.002***
	(0.001)	(0.001)	(0.001)
Log mean wage in district	0.029	0.037	0.035
	(0.038)	(0.032)	(0.059)
Log mean expenditures in district	−0.008	0.145***	0.108*
	(0.049)	(0.046)	(0.059)
Province dummies	Yes	Yes	Yes
Log Pseudolikelihood	−567,841	−460,173	−571,698

Notes: *, **, *** denote significance at the 1, 5, and 10 percent levels, respectively. Weighted data and standard errors are clustered at the district level. Comparison group for marital status is monogamously married.

Choice of income combination

Finally, we investigate how various households combine different income sources. We start by analyzing how the gender and marital status of a head of household influence a household’s income-diversification strategy (Table 7).¹³

Table 7 Income-diversification strategy (multinomial logit average marginal effects)

	Own farm combined with:
	Own farm only	Agricultural wage	Nonagricultural wage	Self-employment	Two or more other income sources
Female head
Monogamously married	0.130***	−0.012	−0.063***	−0.033*	−0.022*
	(0.025)	(0.013)	(0.016)	(0.017)	(0.012)
Polygamously married	0.173***	0.010	−0.102***	−0.051*	−0.030*
	(0.038)	(0.025)	(0.018)	(0.027)	(0.016)
Divorced	−0.033	0.151**	−0.039	−0.040	−0.039
	(0.058)	(0.064)	(0.037)	(0.035)	(0.027)
Widowed	0.096***	−0.009	−0.057***	−0.015	−0.015
	(0.022)	(0.012)	(0.015)	(0.015)	(0.011)
Never married	0.010	0.116**	−0.069*	−0.020	−0.037
	(0.064)	(0.051)	(0.040)	(0.042)	(0.029)
Female head
No adult male in household	0.161***	0.003	−0.111***	0.011	−0.064***
	(0.023)	(0.016)	(0.012)	(0.017)	0.007
With adult male in household	0.069***	0.009	−0.034***	−0.047***	0.003
	(0.019)	(0.011)	(0.013)	(0.011)	(0.010)

Notes: *, **, *** denote significance at the 1, 5, and 10 percent levels, respectively. Standard errors of the changes in parentheses. Weighted data and standard errors are clustered at the small community level. Comparison group for headship status: monogamously married male head.

Our results show that households with married female heads and widowed female heads have an increased probability to rely only on their own farm production (compared to households headed by a monogamously married male). Interestingly, we also find that households headed by divorced women and never-married women have a 15 and 12 percentage points greater probability of combining work on their own farms with agricultural wage labor.¹⁴ With regard to nonagricultural wage work, all female-headed households except the ones with a divorced head of household show a significantly lower probability than monogamously married male-headed households to combine work on the own farm with these activities. We find that both groups of households with a married female head have a lower probability to use self-employment as additional income sources, than male-headed households.

There is also a difference in female-headed households with and without a male adult. Both groups have a higher probability than the comparison group (monogamously married male head) of getting all their earned income from the own farm. Female-headed households without a male adult are about 11 percentage points less likely to combine work on own farm with nonagricultural wage work and 16 percentage points more likely to rely only on income from the own farm than households with a monogamously married male head. This suggests that female-headed households without male adults are more likely to be more dependent on agriculture and have less-diversified incomes.

To better understand what determines the preferred income-diversification strategy among female-headed households, we run the model only for this group, including all explanatory variables (Table 8). We find that a household with a divorced female head is approximately 19 percentage points less likely to rely only on the production of its own farm than a household whose female head is monogamously married. Having one more female adult (aged 15–59 years) in a household decrease the probability of a female-headed household to rely only on income from own farm by approximately 2.8 percentage points and increase the probability of combining work on own farm with nonagricultural wage work by approximately 1.9 percentage points. The pattern is similar when an additional male worker is present in a female-headed household, but the magnitude is larger. The amount of male labor seems more important for an income-diversification strategy than the amount of female labor.

Table 8 Income-diversification strategy with explanatory variables (multinomial logit average marginal effects)

	Own farm combined with:
	Own farm only	Agricultural wage	Nonagricultural wage	Self-employment	Two or more other income sources
Polygamously married	0.038	0.048	−0.046*	−0.021	−0.019
	(0.045)	(0.034)	(0.024)	(0.033)	(0.019)
Divorced	−0.190***	0.162***	0.039	0.014	−0.026
	(0.059)	(0.060)	(0.043)	(0.043)	(0.028)
Widowed	−0.045	0.006	0.014	0.033	−0.007
	(0.034)	(0.020)	(0.022)	(0.024)	(0.019)
Never married	−0.128**	0.080	0.000	0.057	−0.008
	(0.064)	(0.052)	(0.046)	(0.060)	(0.038)
Age of head of household	0.000	−0.001*	0.001	0.000	0.001
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Education, head of household	−0.007	−0.003	0.007**	0.003	0.000
	(0.004)	(0.003)	(0.003)	(0.003)	(0.002)
Highest education in Household	−0.005	−0.003	0.003	0.003	0.002
	(0.005)	(0.003)	(0.003)	(0.003)	(0.003)
Male labor	−0.035**	0.01	0.021**	−0.024***	0.029***
	(0.014)	(0.008)	(0.009)	(0.008)	(0.008)
Female labor	−0.028**	−0.014	0.019*	0.003	0.019***
	(0.013)	(0.010)	(0.010)	(0.009)	(0.007)
Children aged 0–1	0.012	−0.034***	0.029	−0.002	−0.005
	(0.024)	(0.011)	(0.018)	(0.016)	(0.012)
Children aged 2–5	−0.031**	0.009	0.000	0.014	0.008
	(0.015)	(0.010)	(0.011)	(0.014)	(0.010)
Children aged 6–10	−0.003	0.012	−0.014	0.004	0.000
	(0.014)	(0.010)	(0.009)	(0.009)	(0.009)
Children aged 11–14	−0.016	−0.015	0.023*	0.013	−0.005
	(0.017)	(0.010)	(0.013)	(0.012)	(0.009)
Credit constraint	−0.004	0.009	−0.004	0.013	−0.014
	(0.026)	(0.018)	(0.018)	(0.018)	(0.012)
Land labor ratio	0.002	0.000	0.000	−0.001	0.000
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Mean land labor ratio in District	−0.012	0.003	−0.001	0.004	0.006
	(0.009)	(0.006)	(0.005)	(0.004)	(0.005)
Gender norm	−0.029	−0.001	0.024*	−0.006	0.012
	(0.018)	(0.012)	(0.013)	(0.012)	(0.009)
Rate of wage labor	−0.549***	0.17	0.258	−0.132***	0.252
	(0.019)	(0.403)	(0.453)	(0.010)	(0.473)
Distance to market	−0.001	0.00	−0.001	0.001	0.001**
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Log wage in district	−0.049	−0.007	−0.008	0.007	0.058
	(0.055)	(0.031)	(0.033)	(0.042)	(0.050)
Log mean expenditures	−0.187***	0.100*	−0.044*	0.046	0.085
	(0.056)	(0.059)	(0.027)	(0.058)	(0.069)

Notes: *, **, *** denote significance at the 1, 5, and 10 percent levels, respectively. Discrete changes from 0 to 1 for all variables except Gender norm, which shows the marginal effect of increasing the gap by one standard deviation. Standard error of changes in parentheses. Weighted data and standard errors are clustered at the small community level.

The number of children in a female-headed household has different effects, depending on the age of the children. The number of infants (0–1 year) has a negative impact on the probability to combine work on own farm with agricultural wage employment, while the number of children age 2–5 years reduces the likelihood of working on farms only. The number of children in older age groups (11–14 years) does instead have a positive effect on the probability to combine work on own farm with nonagricultural wage employment.

CONCLUSION

An important element in a gender-inclusive growth agenda is the opportunity for women workers to diversify their incomes and move into activities with higher productivity. Investing in women’s economic empowerment is important not only to reduce households’ vulnerability to shocks and to remove them from poverty, but also to achieve sustained inclusive economic growth. In this article, we analyze income diversification among female-headed households in rural Kenya.

The first question we address is whether the income of female-headed households is less diversified than that of male-headed households. Looking at the household’s total income most female-headed households, on average, gets a larger part of their income from outside the agricultural sector than their male counterparts. However, this can be explained by the fact that female-headed households in general receive a larger portion of their income from transfers than male-headed households. This is to be expected since in Kenya there is a long tradition of men migrating to work in the urban parts of the country and sending money home to the rural household. If we instead focus only on the households’ earned income a somewhat different picture emerges, and there is no clear answer to the question regarding if female-headed households are more or less diversified than male-headed households. Instead, the answer seems to depend on which measure of diversification we use and on which group of households we look. However, we do find that female-headed households receive a smaller share of their earned income from the nonagricultural sector. This is important since previous research has shown that increased access to nonfarm employment is an important approach to reduce household vulnerability (Christiaensen and Subbarao 2005). This is also supported by a positive correlation between percentage of earned income that comes from the nonagricultural sector and total income in our data.

The second question we address is whether female-headed households diversify their earned income differently than male-headed households. We find that female-headed households, when the head is married or never married, are less likely to participate in nonagricultural wage labor compared to households headed by a monogamously married male. Access to the nonagricultural sector is important since this reduces the household’s vulnerability to shock since it is less correlated with income from the own farm. Regarding the barriers that affect women’s participation decisions, gender norms, access to labor, education, and age and number of children (age 6–10) affect the probability of participating in nonagricultural wage work.

Our final question is how female-headed households combine different income sources and what the determinants of such combinations are. Female-headed households are, in general, more dependent on work on own farm than male-headed households. However, we also find that households headed by a divorced or widowed woman have a higher probability of combining work on own farm with wage employment in the agriculture sector than a monogamously married male-headed household. Furthermore, female-headed households without an adult male are less likely to combine working on own farm and with a nonagricultural wage activity than female-headed households with male labor.

In the final section of this article, we analyze what influences the income-diversification choices of female-headed households. Again, we find that there are large differences among female-headed households. More resources within households, in terms of the number of workers, increase the probability of diversified incomes.

What are the policy recommendations informed by this empirical analysis? An agenda for increasing women’s economic empowerment needs to identify barriers to and facilitators of women’s opportunities to enter into sectors with higher productivity as part of their income-diversification strategies. From a policy perspective, a distinction can be made between barriers to and facilitators of women’s opportunities to diversify their incomes (Peters et al. 2019). Some facilitators include inputs that have broader societal benefits, such as economic growth, infrastructure/transport, legal environment, and childcare. Certain barriers include both deeply embedded traditional norms regarding women’s role in society and more actionable problems, such as access to resources and low social capital.

Overall, we find empirical support for the existence of actionable and more rigid barriers. With regard to the latter, we discern some support for the hypothesis that women might be hindered by existing social norms regarding women’s labor force participation. Even though changing norms takes time, it is an important step to give women and female-headed households the same opportunities as their male counterparts. Among the actionable barriers, we find that, to various degrees, access to resources (labor) and education all influence the likelihood of female-headed households diversifying their incomes.

Our results clearly show that female-headed households are heterogeneous when it comes to how they diversify their income and which constraints they face. One important implication from this is that treating female-headed household’s as one homogenous group, which today is standard in the literature about income diversification (and several other fields) could lead to incorrect conclusions and inefficient policy recommendations. We argue that it is important for future research to account for all types of marital status.

ACKNOWLEDGMENTS

We are thankful for comments from three anonymous reviewers, Fredrik Sjöholm, Camilla Andersson, Therese Nilsson, and Daniela Andrén.

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/13545701.2022.2159056.

Additional information

Funding

This work was supported by Vetenskapsrådet and the Nordic Africa Institute.

Notes on contributors

Elin Vimefall

Elin Vimefall is Researcher at Örebro University, Sweden. She earned her PhD in 2015 and her main research interest is in development economics and economic evaluations, with a focus on gender aspects and investment in children.

Jörgen Levin

Jörgen Levin is Researcher at the Nordic Africa institute and at Örebro university. He has many years of experience as both lecturer and researcher on topics including Economic Growth and Distribution, Macroeconomics, Inclusive Growth, Taxation and Public Spending, and Sustainable Development Goals (SDGs).

Notes

1 Stephanie Seguino (2020) provides a comprehensive review of gender inequality and macroeconomic outcomes focusing on both demand and supply effects.

2 Factor market distortions have very large effects on productivity in Sub-Saharan Africa. For example, in Kenya, income per capita could be two and a half times larger if physical and human capital were reallocated to sectors with higher productivity (Vollrath 2009).

3 http://www.fao.org/economic/riga/en/. For more detail about how the aggregate data were constructed, see Davis et al. (2009). Information about KIHBS 2005 database can be find here: http://statistics.knbs.or.ke/nada/index.php/catalog/8.

4 Among the monogamously married female-headed households, 80 percent of wives stated that they did not currently live with their husband. A monogamous household includes individuals who live in the same compound, answer to the same head, and share a common source of food and/or income. An individual who has been away from the household for more than nine months during the last year is not counted as a member. Therefore, a husband who has migrated for work is not counted as a household member. Even if the husband sends money back to the household and thereby is part of its income-diversification strategy, our data does not allow us to identify these cases. Therefore, we expect transfers to be important income sources for households with a monogamously married female head.

5 Polygamy varies by groups and locations; 22 percent of the household heads in the Northeastern province stated that they were polygamously married, while the respective number in the Central province was 2 percent. In this study, we do not address regional patterns by location directly. De Giusti and Kambhampati (2016) analyze how religion and ethnicity influence women’s labor market opportunities in Kenya. When wives live in separate dwellings within the same compound or nearby, share housekeeping arrangements and have a common household budget, they are defined as a single household. However, if this is not the case, they are defined as separate households.

6 This is done by first calculating the income shares for each household and then calculating the means of these shares for each income source. An alternative would be to calculate the share of means, which shows the importance of a specific source to the overall agricultural economy.

7 Herfindahl Index, defined as: H = $\sum _i^N{s}_i^2$ere s_i is the sh. are of income from income source i and N is the total number of income sources. The index ranges from 1/n – 1 and decreases with the level of diversification.

8 We cluster the error terms at the small community level (ten households per community).We use the small community level instead of the district level used in previous models, as using the district level in this multinomial context would yield too few degrees of freedom.

9 There are forty-four potential sources of income. We have information only about whether a household obtains income from a specific source. Therefore, if two individuals in the same household obtain their incomes from the same source, this will count as one income source for the household. Thus, it is possible that we underestimate the total number of income sources.

10 We find a positive and significant correlation between household income per capita (excluding transfers) and different measures of income diversification (see online appendix). Noteworthy, is that the correlation is often stronger for male-headed households compared to female-headed household on all three measures of diversification.

11 The results are presented as incident-rate-ratios (IRR). An IRR of 0.6 means that the expected count is multiplied by a factor of .6 when the independent variable increases by one unit. Therefore, an IRR smaller than 1 indicates a negative effect on the expected count, while an IRR larger than 1 indicate a positive effect. The results for control variables are available upon request.

12 As a robustness check, we re-estimate the model to include rural households that do not participate in farming. This makes the coefficient for a divorced female head statistically significant in the equation, explaining participation in nonagricultural wage work when the controls are not included. Furthermore, for the equation explaining participation in nonfarm self-employment, the coefficient for a divorced female head becomes statistically non-significant when the controls are not included. When the controls are included, we do not find significant differences.

13 We use a Wald test to determine whether any of the groups can be combined. The result leads us to reject the hypothesis that any of the groups are indistinguishable. As a robustness check, we also estimate this model as five separate logit regressions. This only has a small impact on the results and does not change the overall conclusions (results available upon request).

14 As a robustness check, we re-estimate the model to include rural households that do not participate in farming. This makes the coefficient for having a monogamously or polygamously married female head statistically significant in explaining a mixed income diversification. In addition, having a divorced female head has a statistically significant negative effect on the probability of not participating in farming at all.

Appendix

Table A1. Variable definition

Variable	Definition
Dependent variables
Own farm only	Household obtains all income from work on own farm
Own farm and agricultural wage	Household obtains income from work on own farm and agricultural waged work
Own farm and nonagricultural wage	Household obtains income from work on own farm and nonagricultural waged work
Own farm and self-employment	Household obtains income from work on own farm and non-farm self-employment
Mixed	Household has more than two income sources
Characteristics of head
Female head:
Monogamously married	Head of household is female and monogamously married
Polygamously married	Head of household is female and polygamously married
Divorced	Head of household is female and divorced
Widowed	Head of household is female and widowed
Never married	Head of household is female and never married
Male head:
Monogamously married	Head of household is male and monogamously married
Polygamously married	Head of household is male and polygamously married
Divorced	Head of household is male and divorced
Widowed	Head of household is male and widowed
Never married	Head of household is male and never married
De jure female head	Head of household is a female who is separated, divorced or has never married
De facto female head	Head of household is a female who is married or living with a man with whom she is in a relationship
Age of head	Age of head of household
Education of head	Years of education for head of household
Characteristics of household
Highest education	Highest level of education in the household
Male labor	Number of males aged 15–59 in the household
Female labor	Number of females aged 15–59 in the household
Children 0–1	Number of children aged 0–1 in the household
Children 2–5	Number of children aged 2–5 in the household
Children 6–10	Number of children aged 6–10 in the household
Children 11–14	Number of children aged 11–15 in the household
Land labor ratio	Area of plot/individuals aged 15–59
Constrained	A dummy that indicates whether the household applied for a loan and was turned down or whether they did not even try to borrow because they believed they would be refused, that it was too expensive or too much trouble, that they did not have enough collateral, or that they did not know any lenders.
Characteristics of district
Gender norm	Percentage of women aged 15–59 working for a wage/ percentage of men aged 15–59 working for a wage. Excluding the observation itself
Mean plot size, district	Mean area of plot, by district. Excluding the observation itself
Log(mean wage)	Mean wage in district (logarithm)
Expenditures district	Mean per capita expenditures in district (logarithm). Excluding the observation itself
Province
Central	Household lives in the central province
Coast	Household lives in the coastal province
Eastern	Household lives in the eastern province
Northeastern	Household lives in the northeastern province
Nyanza	Household lives in Nyanza province
Rift Valley	Household lives in the Rift Valley province
Western	Household lives in the western province