Data and self-employment definition

This study uses data from the Nigeria General Household Surveys Panel (GHS-Panel), conducted in 2010/2011, 2012/13, 2015/16, and 2018/19. This nationally representative survey of 5,000 households results from collaboration between the Nigeria Bureau of Statistics (NBS), the Bill and Melinda Gates Foundation (BMGF), and the World Bank. The GHS-Panel data has been widely used in various studies of Nigeria (Guven-Lisaniler et al Reference Guven-Lisaniler, Tuna and Nwaka2018; Nwaka et al Reference Nwaka, Guven-Lisaniler and Tuna2016; Nwaka et al Reference Nwaka, Uma and Ike2020). Vijverberg (Reference Vijverberg1992) show that household surveys provide a more accurate representation of small-scale enterprises than dedicated enterprise surveys.

We used data from the four GHS-Panel waves, sourced from the World Bank’s Living Standard Measurement Study (LSMS) microdata library (see Nigeria Bureau of Statistics, 2010 for more details). The survey includes questionnaires on personal, labour and household characteristics, as well as inquiries into non-farm household enterprises and housing details. Typically, household heads or those most knowledgeable about household business operations serve as respondents.

Focusing on the gender gap in family business choices—home-based or non-home-based—we restricted the sample to female and male owners aged 20 to 64 who own at least one family enterprise. This age range includes crucial decision-makers in self-employment and household management. We matched individuals based on their characteristics, households and business attributes. Each wave’s sample comprises an average of about 2,014 unique households, 2,840 non-farm enterprises, and 11,396 observations. We used survey weights to accurately reflect the broader population and adjust for any survey biases. Detailed definitions of the variables used in the estimation, along with descriptive statistics, are presented in Table 2.

Table 2.

Variable’s definition and description statistics by business owner’s gender

Independent variables

Empirical methodology

In this study, self-employment participation is examined based on whether owners run businesses at their residence or elsewhere. The owner’s gender and other relevant factors influence the choice of home-based self-employment (HBS). We use a standard probability model to explore how gender and these factors impact the decision to engage in HBS as follows:

(1) $$HBS_{it}^* = {\beta _0} + {\beta _1}{X_{it}} + {\varepsilon _{it}} + {\mu _i}$$

(2) $$HB{S_{it}} = \left\{ {\matrix{ {1\;if\;HBS_i^* \gt 0\;ie\;homebased\;\;\;\;\;\;\;} \cr {0\;if\;HBS_i^* \le 0\;ie\;nonhomebased\;} \cr } } \right\}$$

where $HBS_{it}^*$ is the unobserved latent variable and $HB{S_{it}}$ represents a binary variable that denotes an individual’s probability of operating home-based self-employment given period (t). ${X_i}$ is a vector of variables determining HBS choices including individual, household, business, regional, housing and institutional characteristics. The parameters to be estimated are represented by ${\beta _0}$ , ${\beta _1}$ and ${\beta _2}.$ ${\varepsilon _{it}}$ is a normally distributed stochastic error term with zero mean and unit standard deviation, ${\mu _i}$ capture individual fixed effects. Since both ${\varepsilon _{it}}$ and ${\mu _i}$ are independent normal random variables, it follows that ${\varepsilon _{it}} + {\mu _i}$ is normally distributed with zero mean and variance of $1 + \sigma _\mu ^2$ . Given the panel structure of the data, Equation 1 estimated using both pooled probit (Probit^all) and random effects probitFootnote ¹ (REP^all) estimators as follows:

(3) $${\rm{Pr}}\left( {HB{S_{it}} = 1|\;{X_{it}}} \right) = \Phi \left( {{{{\beta _0} + {\beta _1}{X_{it}}} \over {\sqrt {1 + \sigma _\mu ^2} }}} \right)$$

where ${\rm{\Phi }}\left( . \right)$ capture the standard normal cumulative function (see Bland and Cook Reference Bland and Cook2019)

Machine learning estimator

To identify factors predicting HBS, we use machine learning estimators. Unlike baseline models like Probit^all and REP^all, these estimators minimise residuals with a tuning parameter (Böheim and Stöllinger, Reference Böheim and Stöllinger2021). We apply the least absolute shrinkage and selection operator (LASSO), which has seldom been used for examining the gender gap in HBS. LASSO selects relevant variables, eliminates correlated ones, and shrinks some coefficients by adding a penalty to the conventional models (Probit and Random Effects Probit), penalising the sum of absolute coefficient values. Given $n\;$ observations, each labelled with $i$ , and p variables, Equation 3 is modified as follows:

(4) $${\widehat \beta _l} = \arg min\sum\nolimits_{i = 1}^n {{{\left( {\Pr \left( {HB{S_{it}} = 1|\;{X_{it}}} \right) - \Phi \left( {{{{\beta _0} + {\beta _1}{X_{it}}} \over {\sqrt {1 + \sigma _\mu ^2} }}} \right)} \right)}^2}} + \lambda \sum\nolimits_{j = 1}^p {\left| {{\beta _j}} \right|} $$

where ${{\rm{\widehat \beta }}_l}$ captures the LASSO coefficients in vector form, while p and $\lambda $ are the number of independent variables and tuning parameter respectively. As $\lambda $ increases, the LASSO estimator diverges from the probit model, increasing the bias (Bonaccolto-Töpfer and Briel Reference Bonaccolto-Töpfer and Briel2022, Böheim and Stöllinger Reference Böheim and Stöllinger2021). A $\lambda $ of zero produces the original probit model. The choice of $\lambda $ is determined by the algorithm. Figure 7 shows the cross-validated $\lambda $ using the one standard error bound, and ${\lambda _{cv}}$ show the chosen $\lambda $ value=0.0096 according to this bound as well as the number of variables included in the model. Refer to the Online Supplementary Material for the post-selection goodness of fit test result (Table A).

Figure 7.

Cross-validation for Lambda.

Following Böheim and Stöllinger Reference Böheim and Stöllinger2021, we apply pooled probit and random effects probit models to all variables listed in Table 2 for both female and male business owners, conducting separate analyses for each group. We then re-run the regression using only the variables selected by LASSO estimator (POSTLASSO models).

To analyse how individual, household, business, regional and housing characteristics contribute to the gender gap in HBS choices, we employed the Blinder–Oaxaca (B–O) decomposition technique for nonlinear binary outcomes. According to Averkamp et al. (Reference Averkamp, Bredemeier and Juessen2024), the extended decomposition method used in gender wage gap analysis attributes a larger portion of the gender gap to differences in observed endowments. In this study, the characteristics of male business owners are key determinants of the gender gap in HBS. This approach examines whether the gender gap in HBS engagement would be greater if male business owners were assumed to possess female characteristics, and vice versa. The B–O method (Blinder Reference Blinder1973; Oaxaca Reference Oaxaca1973) is commonly used to study gender and group-based wage differences and has been applied in health (Fagbamigbe et al., Reference Fagbamigbe, Adeniji and Morakinyo2022), food security (Zingwe et al., Reference Zingwe, Manja and Chirwa2023), and entrepreneurship (Hewa-Wellalage et al., Reference Hewa-Wellalage, Boubaker, Hunjra and Verhoeven2022). Details of the decomposition model are provided in the Online Supplementary Material.

Empirical results and discussions

Table 3 presents estimates of factors influencing home-based business (HBS) engagements. Following Böheim and Stöllinger Reference Böheim and Stöllinger2021, we use three specifications: a probit regression with all explanatory variables (PROBIT^all, Model 1), a random effects probit regression with all explanatory variables (REP^all, Model 2), and the POSTLASSO specification, which re-estimates the probit and random effects probit regressions using only the variables selected by the LASSO model (POSTLASSO^probit and POSTLASSO^rep). The POSTLASSO model is applied to the overall sample (Models 3 and 4) as well as to female and male samples (Models 5 and 6). All models are statistically significantly different from 0 at the 1% level, indicating that observable characteristics collectively explain the probability of HBS operation among surveyed business owners.

Table 3.

Probit (Random Effect Probit) average marginal effects of factors determining homebased business choices, overall sample and business owner’s gender (compared to Non-homebased)

Note: Robust standard errors clustered at the enterprise and household levels are shown in parentheses. PROBIT^all and REP^all are the probit and random effects probit model specification that utilized all explanatory variables. POSTLASSO is the re-estimation of the probit regressions of the homebased business choices that includes only the explanatory variables selected by the LASSO model according to the one standard error rule.

*** p < 0.01, ** p < 0.05, * p < 0.1.

The likelihood of engaging in home-based self-employment (HBS) is significantly higher for female business owners, as shown in models 1 to 4, supporting hypothesis 1a. This finding contrasts with Reuschke (Reference Reuschke2016) for the UK, where gender does not influence HBS decisions. Developmental disparities between countries may explain this difference. In the UK, gender norms are less pronounced due to established institutions and human rights, unlike Nigeria, where gender-biased patriarchy and traditional values confine women to domestic roles (Anderson & Ojediran, Reference Anderson and Ojediran2022). Age affects HBS engagement non-linearly for females, with older women being more inclined due to household responsibilities and intensified networks for entrepreneurial opportunities (Millan et al Reference Millan, Congregado and Roman2012). Although work hours are insignificant overall, they impact genders differently: increased self-employment time raises HBS likelihood for females but decreases it for males. Higher monthly rent also boosts HBS participation for females, indicating how economic burdens affect HBS decisions differently for women.

Regarding institutional variables, the absence of household shocks affects HBS engagement similarly for both genders. Marriage positively influences HBS engagement overall but is insignificant for males, supporting hypothesis 1d. For females, marital circumstances, including the need to balance paid and unpaid work and enhance household income, are factors that drive HBS participation. Economic benefits from marriage also prompt female HBS engagement, as seen in Tanzania where sales among small business owners increase during marriage seasons due to gifts and incentives (Anderson & Ojediran, Reference Anderson and Ojediran2022). Divorced and separated individuals, particularly females, exhibit higher HBS engagement compared to never-married individuals, indicating persistent work balance needs post-separation (Nwaka et al Reference Nwaka, Guven-Lisaniler and Tuna2016).

HBS owners with post-secondary education are generally more likely to engage in home-based self-employment, but this trend does not hold for females. Higher household income slightly reduces the likelihood of HBS for females, suggesting that HBS is neither particularly risky nor profitable and that it is not complementary to higher income, supporting the notion of subsistence self-employment (Nwaka & Uma, Reference Nwaka and Uma2021). The presence of children under 19 increases HBS likelihood in both the full sample and among males, indicating a greater burden on females. These findings contrast with Reuschke (Reference Reuschke2016) for the UK, highlighting how gender norms and social barriers differently affect female HBS motivations.

The choice of HBS among female business owners is linked to flexibility and balancing economic and household responsibilities, reflecting the primary caregiving role of Nigerian females. Higher dependency ratios slightly increase female HBS participation, reinforcing gender norms. Ownership of a bank account reduces HBS likelihood for females, indicating some financial empowerment and linkage to formal sector employment, though the evidence is weak. Membership in business associations significantly decreases HBS likelihood across all specifications, suggesting a preference for formal sector employment among members.

Business characteristics further influence HBS participation. Compared to the wholesale and trade sector, HBS is more prevalent in the manufacturing sector and, to a lesser extent, the services sector, though the latter effect is absent in the male sub-sample. This finding is consistent with Reuschke (Reference Reuschke2016). Manufacturing sector businesses are highly capital intensive and outright ownership of a home can form a significant portion of that capital. Higher sales and capital stock reduce HBS likelihood, particularly for males, revealing the economic disadvantages of HBS. Gender norms constrain female business owners, who remain primary caregivers despite higher capital. Consequently, males view higher capital as an opportunity to maximise profits through non-HBS engagements, while females prioritise balancing paid and unpaid work.

Housing characteristics show that renting a home significantly decreases HBS likelihood across all samples, supporting hypotheses 2a and 2b. This effect is stronger for males, suggesting that homeownership compels both genders to participate in HBS. This finding is consistent with Reuschke (Reference Reuschke2016) and validates the assumption that home ownership reduces the start-up capital that would likely accumulate if the business was non-home-based. Larger dwelling spaces increase HBS participation only for males, implying that females are indifferent to dwelling size and prioritise work–life balance. Electricity connection increases HBS likelihood for females but not males, reflecting economic constraints and a persistent need to balance work-life despite challenges. Energy intensity is not significant across self-employment types, although small-scale business operations generally require some form of energy.

Regional characteristics reveal significant differences across Nigeria’s geopolitical zones. Business owners in the North-East are more likely to engage in HBS than those in the South-South. Female business owners in the North-Central have a higher HBS likelihood compared to males, reflecting varying social and religious norms. In the North-West, a predominantly Muslim region, similar patterns are observed, though males do not differ significantly from South-South counterparts. In the South-East, business owners are less likely to engage in HBS compared to the South-South, a significant effect for both genders. In the South-West, HBS likelihood is lower in the full sample but not for females, indicating no statistical difference between South-West and South-South female business owners. These variations are influenced by religious and cultural differences between Northern and Southern regions.

Geographical factors further affect HBS engagement. The greater the distance between the business owner’s home and the market, the less likely they are to engage in HBS, a significant effect for females only. Over time, the likelihood of female HBS engagement has decreased across different survey waves from 2010/2011 to 2018/2019, suggesting changing dynamics in female self-employment.

Gender gaps in HBS

This section presents the B–O decomposition model results. A set of nonlinear decomposition models is initially estimated using all explanatory variables as well as POSTLASSO selected variables, as reported in Table 4. Further analyses are conducted by waves and homeownership status (Figure 8), educational level and industry types, (Figure 9), geopolitical zones (Figure 10) and owner’s membership to an organisation.

Table 4.

B-O decomposition between male and female business owners: female coefficients–Grouped variables

Note: Robust standard errors clustered at the enterprise and household levels are shown in parentheses. PROBIT^all is a probit model specification that utilized all explanatory variables. POSTLASSO is the re-estimation of the probit regression of the homebased business choices that includes only the explanatory variables selected by the LASSO model according to the one standard error rule.

*** p < 0.01, ** p < 0.05, * p < 0.1.

Figure 8.

POSTLASSO B-O decomposition: gender gap by waves (years) and homeownership status.

Source: Authors’ computation using Nigerian GHS-Panel data (2010–2019).

Figure 9.

POSTLASSO B-O decomposition: gender gap by owner’s educational level and industry types.

Source: Authors’ computation using Nigerian GHS-Panel data (2010–2019).

Figure 10.

POSTLASSO B-O decomposition: Gender gap by geopolitical zones.

Source: Authors’ computation using Nigerian GHS-Panel data (2010–2019).

Figure 11.

POSTLASSO B-O decomposition: Gender gap by business registration status (formality) and owner’s membership to an association.

Source: Authors’ computation using Nigerian GHS-Panel data (2010–2019).

The results from Table 4 and Figures 7 and 8 reveal a significant gender gap in HBS operation across all categories. In Table 4, the average probability of females operating an HBS is 27.9% higher than that of males. The gender gap is more pronounced among homeowners (Figure 8), and businesses in wholesale trade show lower gender gaps than those in manufacturing and services (Figure 9). Primary education or less exhibits the highest gender gaps (Figure 9), while Northern geopolitical zones have higher gender gaps than Southern zones (Figure 10). Also, relative to non-registered businesses, registered businesses revealed lower gender gaps with statistically insignificant endowment effects (Figure 11).

Furthermore, Table 4 confirms that if male business owners had characteristics similar to females, their HBS probability would increase by 4.8% (endowment effect). Conversely, if male business owners had female business owners’ coefficients, their HBS probability would decrease by 25.4% (coefficient effect). In essence, if male business owners had female characteristics, it would slightly close the gap (4.8%), but if they had female returns, the gap would increase significantly (25.4%).

About 91% of the total gender gap in the non-HBS group is explained by coefficient effects, while the remaining gap is due to interaction effects (26.16%) and endowment effects (-17.20%). From Table 4, 3.5% (-0.013/0.048) and 43% (-0.111/-0.254) of the endowment and coefficient gaps, respectively, come from returns to formal institutions (marriage and children). Additionally, about 64% (-0.047/-0.073) of the interaction gap is due to informal institutions (such as business association membership or financial inclusion). Additionally, Table B (see the Online Supplementary Material) reveals a larger gender gap among married business owners (-0.304) compared to non-married owners (-0.207), with partner characteristics contributing marginally. This highlights the impact of institutions, cultural differences and socioeconomic factors on gender disparity in home-based self-employment.