Gender differentials on productivity of rice farmers in south-western Nigeria: An 1 Oaxaca-Blinder decomposition approach

16 One of the critical constraints hindering the transformation of African agriculture in general, 17 and Nigerian agriculture in particular, is gender disparities in productivity. This study, 18 therefore, examines gender inequality in farm productivity and sources of the productivity 19 differentials among rice farmers in Nigeria, using the Oaxaca-Blinder (OB) gender 20 decomposition model. The results revealed an uneven situation between men and women, 21 leading to a gender productivity gap of about 29% in favour of men. Thus, female-managed 22 plots are 29% less productive than male-managed plots. The decomposition of the sources of 23 gender productivity differences shows that marital status, education, farm size and access to 24 market information are the significant determinants of the endowment factor that contribute 25 to about 15% of the productivity gap. The study, therefore, concludes that gender 26 productivity inequalities exist in the Nigerian agricultural sector, hence, paying attention to 27 these productivity gaps and factors contributing to these gaps is crucial in formulating policy 28 interventions oriented towards women empowerment.


32
Agricultural production risks are dominant in developing nations. Most farmers in these Gender difference in farm productivity in subsistence farming had become an issue of discourse in public policy in developing countries (Dossah & Mohammed, 2016). 104 Determining the productivity level of both male and female farmers is significant in 105 enhancing food security in Africa where there is high disparity in cultural and religious 106 beliefs. The widened gender gap in adopting modern crop varieties and other agricultural 107 technologies is detrimental to the empowerment of women in developing countries, and 108 imposes real costs on societies in terms of untapped potential in agricultural output, food 109 security and economic growth (Ragasa, 2012). Empirical studies had indicated that, if women 110 farmers had the same access as men to improved agricultural inputs such as fertilizer and 111 seed, maize yields would increase by as much as 16% in Malawi,17% in Ghana and 19% in 112 western Kenya (Diiro et al., 2018). Addison, Ohene-Yankyera and Fredua-Antoh (2016) 113 reported that female farmers are more productive than male rice farmers in Ghana. This 114 finding is also in consonance with the study of Due and Gladwin (1991) who observed that 115 inefficiency of women rice farmers was due to constraints in accessing productive inputs. To 116 the contrary, in a similar study, Omondi and Shikuku (2013) provided evidence that male 117 farmers are more efficient than female rice farmers in Kenya.  (2010) observed a significant difference in paddy rice yield of men and women farmers in 119 Benin, disclosing a larger yield by male farmers, although it revealed that there was no 120 significant difference in technical efficiency of men and women farmers. They explained that 121 the higher productivity of male farmers as opposed to their female counterparts was due to 122 the possession of larger land holding sizes allocated for rice farming by male farmers.

123
Furthermore, research in developing countries showed that income controlled by women has 124 a greater positive effect than income controlled by men in terms of calorie intake, nutrition, 125 health and educational attainment of household members (Quisumbing et al., 1995). While contrary results as compared to previous studies by Chirwa (2005) and Smale (2011), which 129 revealed that gender was an important determinant of modern maize adoption, even after 130 having controlled for individual, household and community-level characteristics. Their Following the classical approach of the wage gender gap literature of Oaxaca (1973) and136 Blinder (1973), an OB decomposition methodology was employed by dividing the 137 productivity gap between the average male and female farmer into two components: (i) the 138 endowment effect and (ii) the structural effect. The former is accounted for by unequal access 139 to production inputs and farmer characteristics, while the latter is associated to gender 140 differences in the returns to such factors. Since this method builds on basic OLS regression 141 estimates, the problem of endogeneity is inevitable in production inputs as found in most 142 other studies (Peterman et al., 2011;Quisumbing, 1996).

143
Various decomposition methods had been extensively employed in the gender and union 144 wage gap literature. They had also been used to understand which factors explain changes in 145 inequality and account for economic growth (Fortin, Lemieux & Firpo, 2011). In evaluating 146 the decomposition of gender differentials in agricultural productivity in Ethiopia, Aguilar et 147 al. (2015) suggested in their study that differences in the returns to extension services, land 148 certification, land extension and product diversification may contribute to the unexplained 149 fraction of smallholder farmers' productivity in Ethiopia. This article adopts a decomposition 150 method to determine to what extent differences in levels and returns to productivity 151 determinants explain the overall gender differential in agricultural productivity. The main 152 purpose of the decomposition method is to partition the overall difference of a given 153 distribution's statistic of interest between two groups: where . u is the statistic in which ν (·) is the statistic of interest, usually the mean, and is the cumulative distribution of the potential outcome, qi Y , for individuals of group S. In this article, the mutually exclusive groups are male and female smallholder rice farmers.

158
To be able to construct counterfactuals, a structural form relating to the outcome with 159 observed and unobserved individual characteristics ( i X and i , respectively) is specified.

160
Making simple counterfactual treatment, overlapping support, and ignorability assumptions, where u s or the "structural effect" (also called the unexplained effect) represents 165 differences in returns to observable characteristics, and u X or the "endowment effect" 166 reflects differences in the distribution of observable characteristics between both groups.

167
The basic mean decomposition follows the method developed by Oaxaca (1973) where i X is a vector of characteristics and q is a vector of OLS coefficients, estimated 174 separately for each group.

175
Zero conditional mean indicates that 0 Applying assumptions 1 and 2 to the OB framework is obtained as: . To the extent that the additive linearity assumption is 184 satisfied, a detailed decomposition assessing the contribution of each covariate to the 185 structural and endowment effects can be estimated.
It was assumed that agricultural productivity is a function of the manager's socio-187 demographic factors, labour and non-labour inputs, and farm size. Following a yield-based 188 approach, this study's regression analysis expresses these covariates as inputs per unit of land 189 (Quisumbing, 1996). Although this specification may suffer from an endogeneity problem 190 (Peterman et al., 2011;Quisumbing, 1996), the objective of this study is not to infer causality,

211
A multistage sampling technique was used to select the respondents for the study. The first 212 stage involved a typical case purposive selection of three states, Ekiti, Ondo and Osun, 213 located in the same agro-ecological area. In the second stage, four local government areas 214 (LGAs) were selected from each state, based on the predominance of smallholder rice 215 farmers in these areas and using a typical case purposive sampling. In the third stage, five 216 villages were randomly selected from each of the four LGAs. Following Tesfahunegn et al.

217
(2016), the sample size for the study was determined at 95% confidence level and 5% margin 218 of error, using the sample determination formula as described by Cochran (1977) and allowing for six smallholder rice farmers to be selected from each of the five villages earlier 220 selected. This produced 360 respondents interviewed for the study. Data was collected by 221 means of a pre-tested, well-structured questionnaire on respondents' socioeconomic 222 characteristics as well as the productivity of rice. 223 4 Empirical results and discussions 224 The descriptive statistics of the surveyed rice farmers are presented in Table 1. The standard 225 OB decomposition technique was applied here to divide the productivity gap between men 226 and women with respect to farm plot management into a part that is explained by differences 227 in the factors influencing productivity and the parts that cannot be explained by those factors.

228
The results of the group-specific regression for males and females are presented in Table 2.

229
The main determinants of productivity differences include age, marital status, formal 230 education, and farming experience.

Empirical discussions 250
Age has a positive effect on productivity and affects both male and female plot managers.

251
However, age has a marginally higher private rate of return for males than for females. This which potentially make them weak. As a result, their strength to work hard for higher 257 productivity declines as they age. Moreover, some of these old women may be widowed, 258 leading to less financial muscle and support than that of their male counterparts to purchase 259 farm inputs required for higher productivity.

260
Marital status has no influence on male managers but a negative significant effect on female

Oaxaca-Blinder (OB) mean decomposition 283
As mentioned above, the mean difference in farm productivity between male and female is 284 based on the Oaxaca (1973) and Blinder (1973) decomposition approach. In Table 3, the 285 decomposition output reports the mean predictions for male and female and the difference 286 between the two. In the Table, the mean value of productivity reflects 9.36 for men and 9.08 287 for women, leading to a significant gap difference of 0.29. This suggests an uneven situation between men and women regarding farm productivity. Thus, women are about 29% less 289 productive than their male counterparts.  Vargas and Vigneri, 2014) which state that gender inequalities in farm productivity are 312 mostly explained by differences in production characteristics. The interaction term is a 313 measure of the simultaneous effects of differences between endowment and coefficient.
After the decomposition of the farm productivity gap into endowment and structural effects, 315 the study boosts the analysis by identifying factors contributing to these two effects. As stated 316 earlier, if the additive linearity assumption is satisfied, a detailed decomposition of the 317 determinants of endowment and structural effects can be estimated.    Netherlands.