Comparative analysis of rice milling strategies : Evidence from rice millers in Benin

Milling and related activities are central to the improvement of the quality of locally produced rice in many African countries. Fee-based milling and paddy buying-milling-selling services are among the strategies used in the milling business. The Heckman two-stage model and gross profit method were used on survey data, collected in 2011 on rice millers in major rice-producing regions of Benin to identify the factors that influence millers’ choice of strategy and determine the quantity of paddy milled under the buying–milling–selling strategy. Results from the first stage of the Heckman model identified previous salary activity, utilization of modern technology, and trade as main activity as factors positively influencing millers’ decision, while the second stage showed that the milling capacity and length of service in milling activity as factors positively influencing the quantity of paddy milled. The average profit of miller-traders was higher than that of millers-only.


INTRODUCTION
Since the 2008 food-price crisis, agricultural policy measures in many West African countries have focused on boosting domestic rice production.Consequently, West Africa's annual production of 7.7 MT is higher than that of any other region in Africa (AfricaRice, 2012).The consumption of rice is also growing in West Africa but rice processing is very fragmented with no major commercial players (Bill and Melinda Gates Foundation, 2012).Benin's Strategic Plan to Revitalize the Agriculture Sector (PSRSA, 2009(PSRSA, -2015) ) includes the development of 13 promising sectors, including that of rice because of its socio-economic and dietary importance (NRDS, 2011).This has contributed to an increase in the annual growth rate of paddy production from 12% in 2000-2007to 14% in 2008-2013(USDA, 2014)).This significant achievement should increase the volume of local rice available in Benin's major food markets.However, because of the low local rice processing capacity in the country, it is doubtful if the quality of such rice meets consumers' expectations.Locally produced African rice often fails to compete with imported rice because of outdated processing technologies and ill equipped infrastructure (Stryker, 2013;Seck et al., 2010Seck et al., , 2013)).Most operators of rice hullers only provide milling services and this practice contributes to the fragmentation of the market for milled local rice (AfricaRice, 2011).
In Benin, the harvesting, threshing, drying and milling of rice are generally done manually and the end product is therefore often of low quality compared with imported rice.However, small threshing, winnowing and milling equipment have been introduced in some areas.Two processing methods are used -direct rice milling without parboiling and milling after parboiling.Rice milling without parboiling is common in the South while milling after parboiling is practiced in the North and Central zones.Parboiling itself is of two types, namely traditional parboiling and improved parboiling.Private millers in some villages use the Engelberg mill which produces grains of acceptable quality from parboiled rice.Processing is done by both men and women but, while women predominantly purchase, parboil and sell rice (usually through women's associations), milling is mostly done by men.
As mentioned earlier, rice produced in Benin is marginally less valued and attracts a lower price than imported rice due to perceived differences in quality.The low quality of local rice is also due to the status of the rice processing sector in which 76% of millers use traditional sub-standardized milling plants, 21% use small and medium-sized modern mills, and only 3% (in general the state mills) have automated plants that produce relatively good quality rice (AfricaRice and PAPA, 2011).Consequently, most of the locally processed rice contains impurities and mixed colors due to the use of inappropriate post-harvest handling such as post-milling sorting, polishing and grading.Much of this rice is consumed in rural areas (Demont, 2013).However, the product of mills that have key processing tools is clean and suitable for urban consumers who value convenience due to their busy work schedules (Lançon and David-Benz, 2007;Fiamohe et al., 2013).Rice processors have a major role to play in the improvement of the quality of local rice and their efficiency can be considered as key elements in the development of a new promising local rice market.According to Lançon (2003) the issue of the quality of locally produced rice is actually only relevant for miller-traders who have direct control over the type of paddy they process and bear the risks of selling a type of rice that does not respond to market demand; apart from specific paddy parboilers, parboiling is also done by millers who purchase paddy; most millers in Nigeria do not systematically combine milling with trading (that is, purchase paddy and sell milled rice) but rather mill paddy for a fee (that is, millers are service providers).Surveys of the rice-processing sector in Benin in 2011 showed that only 9% of millers combine milling with trading, 74% of millers are services providers, and 9% do both.This situation raises the following questions: (i) why can millers not purchase paddy and process it to a suitable standard for selling?(ii) which of these activities (providing only milling service or purchasing paddy to mill for sale) has positive effects on millers' welfare?(iii) which factors determine the economic attitude of millers and therefore the level of paddy purchased for milling and sale?Astewel (2010) identified education level of the head of household and the quantity of rice produced as important variables that determine the volume of rice sold in the market by rice farmers.In contrast, the objective of Nigerian rice millers/traders is beyond maximizing the volume of rice processed; rather it is to generate additional income by adding marketing value to the product (Lancon, 2003).Through financial analysis Diarra (1994) showed that the buy-mill-sell activity is far more profitable than the provision of milling service.The many studies on rice post-harvest processing in Africa (Diarra, 1994;Lançon, 2003;Shwetha et al., 2010;Astewel, 2010) have ignored the factors that determine whether millers purchase paddy or not.This study assessed the factors that could explain the economic behavior of millers faced with the alternative strategies, and the quantity of paddy milled by those who adopt the buymill-sell strategy.

THEORETICAL FRAMEWORKS
Both descriptive methods and econometrics techniques were used.MCA provided an exploratory view of factors which explain how the rice milling sector in selected regions of Benin function.The Heckman two-stage model was then applied to determine the factors that influence millers' decision to buy-mill-sell and determine the quantity of paddy milled by miller-traders.Finally, profits of both categories of milling activities were calculated and compared.

Multiple correspondence analyses
MCA is a potentially useful exploratory data analysis technique for the identification of stable patterns in the data (Kaciak and Louviere, 1990).This technique was used to identify categories of millers that have similar characteristics in terms of criteria listed in Table 2. Normally, it consists of representing individuals or points (here millers or processing units) in a multidimensional space in such a way as to get an overview of the relative positions of individuals.However, it is difficult to observe points in a space with more than three dimensions.Therefore, MCA was used to reduce multidimensional space in a 3 orthogonal axes also termed as factors which maximize the information content (inertia).These axes define 2×2 factorial planes and are associated with eigenvalues.The eigenvalues are defined as the difference between the total number of modalities (K) of the variables and the total number of variables (Q) themselves divided by the total number of variables, [(K- Each modality of variables that characterize the economic behavior of millers is represented by a point.On the factor plane in the MCA points (projected points) that images of each other represent individuals that have similar or the same characteristics.Some criteria such as contribution of a modality to the total inertia borne by an axis (absolute contribution), good position of a modality on an axis (relative contribution) and appropriate modalities to use for the similarity interpretation of individuals on a given graphical display (test-value) are necessary.For instance, the higher the test-value of a modality on an axis, the more useful this modality is for interpreting that axis.Variables used in the MCA analysis are described in Table 1.

Heckman two-stage model
The micro-data for the quantity of paddy purchased and processed are available for only some millers since not all of them purchase paddy.Thus, we have zero quantity of paddy purchased for some millers.This situation created an econometric problem known in the literature as 'sample selection bias as specification error' (Heckman, 1979).To resolve this problem, we used Heckman's two-stage procedure.This procedure is applied when the data used are censored (Green, 2005).
According to Gujarati (2003, pp. 563-636) a sample in which information on the regressand is available only for some observations is known as a 'censored sample'.This technique is appropriate because in this study we are not only interested in knowing factors that determine the economic attitude of millers, but also the level of paddy purchased and milled.If only the probability of a miller purchasing or not purchasing paddy were to be analyzed, Probit or Logit models would be adequate.Since we are interested in both objectives, there is a need for a model that is a hybrid between the Logit or Probit and ordinary least squares (OLS).Sample selection bias may arise in practice for two reasons: (i) there may be self-selection by an individual or data units being investigated; (ii) sample selection decisions by analysts or data processors operate in much the same fashion as self-selection (Heckman, 1979).This method can also be used to address the issue of endogeneity.In that case, the endogeneity is supported by the presence of sample selection bias (Manuel, 2010).For this study, this model was chosen due to missing data for the regressand (quantity of purchased paddy).The Heckman method consists of a two-step estimating procedure.In the first stage, a 'selection equation' attempts to capture factors affecting millers' decision on the purchase of paddy.The second stage provides 'heckit' analysis that determines the quantity of paddy purchased and milled.The probability of millers purchasing paddy was modeled by the Maximum Likelihood Probit, from which the inverse Mill's ratio was estimated.Its specification is as follows: The selection equation where Y i is a binary variable that assumes 1 if rice- processing unity (i) purchases paddy and 0 otherwise; i  is the vector of unknown parameters to be estimated; i X the vector of explanatory variables that determine whether paddy is purchased or not; i  is the random error terms that are assumed to be independently and

The regression equation (OLS)
where i Q is the purchased paddy milled; i z the vector of explanatory variables which determine the quantity of paddy purchased and milled; i  , which is related to the conditional probability that a rice-processing unit will decide to purchase paddy (given a set of independent variables) is determined by the following formula: where  and  are the probability density function and the cumulative distribution function of the distribution, respectively, and  is a vector of unknown parameters that contains the correlation coefficient   .It is worth noting that rho () is the parameter that helps detect whether there is selection bias.For instance, if rho is statistically different from 0, there is selection bias and the two-stage model is relevant.Where rho is not statistically different from 0, OLS techniques will be appropriate.

Profit calculation of milling activities
We used the gross profit method to compute the profitability of rice milling for both categories of millers.The usual gross profit (GP) formula is specified as follows: where P is the price of the milled rice sold, Q the total quantity of milled rice sold, i p the price of input i and i q the quantity of input i.In the model, cost variables include repair and maintenance and depreciation costs (DC) of material (mills) which is specified as follows 1 : 1 Costs that are common for both types of activities include expenses on electricity, repairs, maintenance and labor.The calculation of the value of production (V) and the cost of production (C) for each category of miller differed.
For the miller-traders, the value of production was the market price per kg of milled rice multiplied by the quantity of milled rice.Also, costs incurred by the millertraders took into account the value of paddy purchased (that is, quantity of paddy multiplied by the price per kg of paddy).On the other hand, the value of production for custom-millers was the fee per kg of milling service multiplied by the quantity of rice milled.

Data collection
The data used in this study came from a survey conducted in 2011 across 16 communes in Benin by the national agricultural research system (Institut National des Recherches Agricoles du Benin, INRAB) and the national agricultural statistics systems (NASS) in collaboration with the Africa Rice Center.This survey covered all actors across the rice value chain -farmers, consumers, traders and processors (millers and parboilers).In this study, we used only data that are relevant to rice millers since our goal was to investigate the factors that could explain the economic behavior of millers faced with alternative types of activities in Benin.
The survey used the same standard questionnaire and the stratified two-stage sampling method adopted by participating member states of AfricaRice.In the first stage, eight villages and peri-urban areas were purposively selected to ensure a good representation of the major rice processing zones with intensive milling activities.In the second stage, 90 processors were selected randomly in each village and Peri-urban area.However, because there are few millers in Benin, almost all of them were surveyed in each village and Peri-urban area and only 34 respondents were found.They fall into the different categories of small, medium and large-scale.Information collected with the structured questionnaire included detailed personal information of millers, sources of paddy procured, milling practices, milling costs, equipment, constraints, labor demand, labor skills, milling capacity, prices of local paddy and milled rice, energy sources, processing losses, and capacity utilization of the processing facilities.The geographical distribution of the surveyed millers is presented in Table 3.

Description of associated characteristics of surveyed population
Table 4 shows the socio-demographic characteristics of the surveyed millers.Of the 34 respondents, 82% were millers-only and 18% miller-traders; 94% were males and 6% females.Millers-only comprised 96% males and 4% females, compared to 83% males and 17% females among the miller-traders.This shows that the milling business in Benin is dominated by men.
Both groups of millers were 'young' with an average age of 28 to 50 years.More than 44% of millers in both categories were educated (4% of the millers-only group attended high school) while more than 40% were not educated.In addition, 33% of miller-traders used modern mills as against 14% in the millers-only group.The old types of mill were largely used by most millers in the two groups.The remaining variables were distributed moreor-less equally.

Map of rice millers
The map of millers with their respective characteristics is shown in Figure 1.The first factor axis explains 11.34% of the total variance, while the second axis explains 10.19%.The two factor axes explain a total of 21.53%.In the factor plane, the first axis is determined by the variable named contract with customers who have two modalities (contract with customers and no contract with customers).In the positive value is the modality contract with customers, (contribution=15.1,test-value=4.3,and distance=1.83)while in the negative value is the modality no contract with customers (contribution=8.2,test-value=-4.8, and distance=0.55).The second axis is determined by the variable funding source with two modalities (self-financing and donation/support).In the  The map reveals two main subgroups -miller-traders led by the modality no contract with customers and millersonly led by the modality contract with customers.Millers in first subgroup are miller-traders, have less than 10 years of service (LS<10), do not belong to a processors' association (PrAssM/no), practice farming as a secondary activity (SA/farming), had salary service as a previous activity (PA/salary service), use more modern machinery, and have at least primary school education.
The second subgroup comprises millers mill paddy as a fee-based service (Millers-only), belong to a processors' association (PrAssM/yes), have more than 10 years of service (LS>10), mill as a secondary activity (SA/milling), have general trading as a previous activity (PA/general trade), and use more old-fashioned machinery.Members of the second group often make short-term credit and inkind payment contracts with their customers.The exploratory analysis identified previous salary activity, the use of modern technology, length of service and trade as main activity as factors that are likely associated with the decision to purchase paddy or not.These results highlighted the similarity between the variables miller-traders (equivalent to purchase paddy) or millers-only (equivalent to not purchase paddy) and other characteristics of millers, such as the use of modern technology, previous salary activity, length of service, and trade as main activity.This apparent relationship between these variables is deepened by the econometrical analysis.

Results of the Heckman two-stage model
Table 5 presents results from the use of the Heckman two-stage model.This model explains factors that influence both the economic behavior of millers and the quantity of purchased paddy milled by them.The results show that the model is globally significant (P<0.01) and a rho figure that indicates whether there is statistically significant selection bias.The significance of rho confirms the appropriateness of the use of the Heckman two-stage model.On one hand, in the selection equation, the variables previous salary activity, modern technology and trade as main activity are statistically and positively significant (P<0.01).The significance of the previous salary activity suggests that millers who were previously in salaried employment had bought their mills using money they had saved.This is confirmed by the fact that on the MCA map, miller-traders bought mills by their own means.The significance of modern technology may be related to the fact that this technology is more used by miller-traders since they are the ones that purchased paddy.As for the variable trade as main activity, its significance can mean that millers that have trading as their main activity are more likely to purchase paddy and mill it for sale.Because paddy production is a seasonal activity, they diversify their activities by combining milling with trading.In Nigeria, 41% of all the millers closed their workshop for an average of three months because they were millers-only (Lançon, 2003).This fact highlights the difference between the two categories of millers.
On the other hand, the regression equation showed that the variables machine capacity and length of service in the milling business significantly and positively influence the quantity of paddy purchased and milled, while modern technology was not statistically significant.The significance of machine capacity means that the greater the capacity of a machine, the greater the quantity it mills when paddy is available.The significance of length of service could mean that over time, millers gain experience and acquire more skill and therefore their activity yields more.However, this is not corroborated by the MCA map, which indicates that length of service is associated with millers-only and is supposed to negatively influence the quantity of paddy purchased and milled since millers-only do not purchase paddy.This disparity can be explained by the fact that miller-traders are mainly those who entered the sector after liberalization, the length of service being more associated with millers-only (cooperative and public milling units which were established a long time ago) than millertraders (more private sector operators who are newcomers to the sector).The non-significance of modern technology may be due to the fact that the difference between modern technology and ordinary technology does not lie in the quantity of paddy a machine can mill, but rather on the quality of the rice it produces.The estimated profit indicates the relative profitability of the two categories of milling activities and could be used by rice sector-oriented policy makers to offer financial-based advice.

Computation of profit
Table 6 shows the profit calculated for both types of millers, that is, those who purchase paddy (miller-traders) and those who do not (millers-only).The table describes the operating accounts of the two types of milling business as a simplified balance sheet providing each major cost item and source of revenue as well as the profit for each type of activity.The average profit for miller-traders for 1 kg was 35.465 FCFA compared with 15.98 FCFA for millers-only.Thus, milling was more profitable for miller-traders than for millers-only, others things being constant.It should be noted, however, that the profit of miller-traders is influenced by fluctuations in the price of paddy and milled rice since these millers are also involved in marketing.

Conclusions
Due to the inappropriate post-harvest techniques (threshing, drying, parboiling, and milling) used by processors, the quality of locally produced rice remains far below that expected by urban consumers.Processors thus have a major role to play in improving the quality of locally produced rice.This study assessed the factors that can explain the economic behavior of millers faced with the alternative types of activities -"providing only milling service for a fee" (millers-only) and "purchasing rice paddy and milling it for sale" (miller-traders).This was done by using multiple corresponding analysis (MCA) and the Heckman two-stage model.Whereas MCA facilitated the categorization of millers with the same characteristics, the Heckman two-stage model highlighted not only factors that influence millers' decision to purchase paddy or not, but also factors that influence the quantity of purchased paddy milled.
The MCA showed two subgroups of millers.Most of the miller-traders procured their mills with their own means; they are often not members of a processors' association; most of them have worked less than 10 years in the milling sector; they were previously in salaried employment, and some of them buy paddy and process it for sale.Most of the millers-only got their mills through support or donation; they are often members of a processors' association; do not purchase paddy to mill but process paddy for others for a fee (farmers, traders or consumers); most of them have spent more than 10 years in the sector and were previously involved in farming and handicrafts.
The Heckman two-stage model showed that previous salary activity, modern technology and trade as the main activity are likely to influence the decision of millers.It also showed that factors such as machine capacity and length of service influenced the quantity of purchased paddy milled.The average profit of miller-traders (35.46 FCFA kg -1 ) is higher than that of millers-only (15.98 FCFA kg -1 ).
These findings indicate that processors gain experience and acquire more skill over time.It is therefore, necessary to organize periodic training of processors, especially the newcomers.Policy should create an enabling environment by ensuring easy access to capital to facilitate investment in improved technologies by processors.Such policy could increase the number of miller-traders which is currently low.Millers also need a financial incentive to encourage them to purchase paddy.The expected outcome of this policy will be an increase in the use of modern mills that will produce good-quality rice as required by consumers.

Figure 1 .
Figure 1.Map of rice millers with their distinctive characteristics.

Table 1 .
Description of variables used for the MCA.

Table 2 .
Description of variables used in the model.
* i y is the latent variable which is observed.

Table 3 .
Regional distribution of surveyed millers.

Table 4 .
Socio-demographic characteristics of surveyed millers in Benin.

Table 5 .
Estimates from the Heckman two-stage model.