34
Determinants of Changing Behaviors of NERICA Adoption: An Analysis of Panel Data from Uganda Yoko Kijima* A , Keijiro Otsuka B , and Dick Sserunkuuma C A University of Tsukuba, B Foundation for Advanced Studies on International Development, C Makerere University, Kampala, Uganda Abstract Using panel data of 347 households collected in 2004 and 2006 in Uganda, we identify four types of NERICA adoption behavior; continuous adoption in the two years, dropout (adoption only in 2004), late adoption (adoption only in 2006), and non-adoption. We found that NERICA yield is significantly higher among continuous adopters than among dropouts and late adopters. Further, results of the estimated NERICA adoption and yield functions indicate that the extension system failed to disseminate appropriate production knowledge, particularly of NERICA seed production, while the development of the market for rice via improved provision of milling services and the use of farmer-produced seed led to wider dissemination of NERICA rice. * Graduate School of Systems & Information Engineering, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki, 305-8573 JAPAN, E-mail: [email protected] Financial support was provided by the 21 st Century Center of Excellency project at National Graduate Institute for Policy Studies for collection of the data used in this paper. We would like to thank Carlos Bozzoli, Kei Kajisa, Takashi Yamano, participants in CSAE Conference 2008 in Oxford University, COE Tokyo workshop 2008 in National Graduate Institute (GRIPS), and GSID seminar in Nagoya University for helpful comments.

Determinants of Changing Behaviors of NERICA …otsuka/Papers/4. Kijima et al...Determinants of Changing Behaviors of NERICA Adoption: An Analysis of Panel Data from Uganda Yoko Kijima*A,

Embed Size (px)

Citation preview

Determinants of Changing Behaviors of NERICA Adoption: An

Analysis of Panel Data from Uganda

Yoko Kijima*A, Keijiro OtsukaB, and Dick SserunkuumaC

A University of Tsukuba, B Foundation for Advanced Studies on International Development, CMakerere University, Kampala, Uganda

Abstract

Using panel data of 347 households collected in 2004 and 2006 in Uganda, we identify

four types of NERICA adoption behavior; continuous adoption in the two years, dropout

(adoption only in 2004), late adoption (adoption only in 2006), and non-adoption. We

found that NERICA yield is significantly higher among continuous adopters than among

dropouts and late adopters. Further, results of the estimated NERICA adoption and

yield functions indicate that the extension system failed to disseminate appropriate

production knowledge, particularly of NERICA seed production, while the development

of the market for rice via improved provision of milling services and the use of

farmer-produced seed led to wider dissemination of NERICA rice.

* Graduate School of Systems & Information Engineering, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki, 305-8573 JAPAN, E-mail: [email protected] Financial support was provided by the 21st Century Center of Excellency project at National

Graduate Institute for Policy Studies for collection of the data used in this paper. We would like to

thank Carlos Bozzoli, Kei Kajisa, Takashi Yamano, participants in CSAE Conference 2008 in

Oxford University, COE Tokyo workshop 2008 in National Graduate Institute (GRIPS), and GSID

seminar in Nagoya University for helpful comments.

1

1. Introduction

The recent sudden and sharp increase in food prices in the international markets has

significantly and adversely affected food security in developing countries, especially

among smallholders in the Sub-Sahara Africa (SSA), who tend to be net food buyers, as

well as the urban poor. There is thus an urgent need to increase food production to

alleviate widespread poverty and food insecurity in this region. Given the increasingly

limited room for further expansion of cultivated area, there is no substitute for a Green

Revolution, which enhances crop yield per unit of land, in an effort to boost the food

production in SSA (Otsuka and Kijima, 2008).

In Uganda, New Rice for Africa (NERICA), a high-yielding upland rice variety

suitable for African environment, was introduced for food security and poverty

reduction in rural areas in 2003. The release of NERICA was received enthusiastically

in Uganda because of its high potential for increasing crop yield and reducing rural

poverty. Indeed, the area under upland rice has increased rapidly in the past six years,

from 1,500 ha in 2002 to 35,000 ha in 2007 (Tsuboi 2008). Furthermore, the

introduction of NERICA in Uganda is estimated to have increased per capita income by

12%, decreased the poverty head count ratio by 5 percentage points, and decreased the

squared poverty gap from 22 to 15 without deteriorating income distribution. Such

effects were even greater if the learning effect, manifested in rice growing experience,

was taken into account (Kijima et al. 2008).

This the poverty-reducing impact of NERICA adoption would be expected to

2

motivate a steady increase in its adoption. However, panel data of 374 rural households

interviewed in both 2004 and 2006 shows that more than 50% of the NERICA adopters

in 2004 abandoned it in 2006, although 20% of the non-adopters in 2004 adopted

NERICA in 2006.1 This adoption pattern raises an important question regarding the

sustainability of the profitability of NERICA adoption. The objective of this study is

to analyze the determinants of the changing behavior of NERICA adoption, using a

two-year (2004 and 2006) panel data set.

In addition to the peculiar pattern of NERICA adoption, several important

observations were made. First, we found that the availability of rice millers has

improved remarkably in recent years. Our informal interviews with rice farmers in

2004 revealed serious problems in marketing the harvested paddy rice because of the

absence of rice millers in nearby towns (Kijima et al. 2006). The number of rice

millers in Uganda, however, has increased rapidly increased in recent years, and farmers

reported a major improvement in access to rice millers in 2006. Second, we also found

that while a limited amount of rice seed was produced by seed companies and 1 A number of studies have been conducted on the dynamic adoption decision of new technologies in developing countries. Suri (2007) observes that many farmers in Kenya using hybrid maize varieties in one year tend to switch to local maize variety in subsequent years. Moser and Barrett (2006) and Moser and Barrett (2003) examine the adoption dynamics of agricultural technologies (yield-increasing and low-external-input technology for rice cultivation) in Madagascar and show that own experience of the technology decreases the probability of the non-adoption. Corletto et al. (1999) analyze the determinants of unsustainable adoption of nontraditional agro-export crops in Guatemala and find the importance of cooperative membership for accessing to credit. Cameron (1999) used panel data from India to show that as the average differential of profits from high yield variety and traditional variety of cotton in the past are greater, household tends to use new variety this year. She calls this learning effect. Neill and Lee (2001) examine the adoption of sustainable but labor-intensive cropping systems for maize production in Northern Honduras and find that higher opportunity costs of land and labor measured by accessibility to road and cultivation of high-valued crops increase the probability of abandoning the new cropping system..

3

distributed mainly by subsidized seed programs in 2004, the use of farmer-produced

seed became common and was even traded among farmers in 2006. Third, the yield of

NERICA was much higher among the continuous adopters than among the dropouts

(i.e., adopters only in 2004) and the late adopters in 2006. We hypothesize in this

study that while the massive abandonment of NERICA adoption, and the lower yields

among dropouts and late adopters are largely explained by failure of extension services,

the continued and increased NERICA adoption is attributed to the development of rice

milling markets, and expansion of seed supply by farmers themselves.

The rest of this paper is structured as follows. Section 2 presents the

descriptive data used in this study and examines their major characteristics. Section 3

explains the empirical models and variables used in this study, while results of the

estimated NERICA adoption and yield functions are examined in Section 4. Section 5

discusses the conclusions and policy implications.

2. Data and Sample

The data used in this paper were collected in 2004 and 2006. Since the dissemination

of NERICA started in 2004, farm households growing NERICA rice were found mainly

in areas with NERICA seed dissemination programs. Thus, we intentionally selected

10 NERICA growing areas covering the Central and Western regions (Kijima et al.

4

2008)2. In each sample area, we drew a random sample of 25 households that grew

NERICA rice in the second cropping season of 2004, and 15 households that did not.

In the second survey (2006), there was attrition due to out-migration from sample areas,

dissolution of households, and absence of household members during the data collection

period. In total, we used a panel sample of 347 households for this analysis.

Based on differences in NERICA adoption behavior, we stratified the sample

into four groups. The first group, non-adopters, has never adopted NERICA; the

second group, dropouts, includes households that grew NERICA in 2004 but not in

2006; the third group, continuous adopters, consists of the households that grew

NERICA in both 2004 and 2006; and the last group, late adopters, refers to the

households adopting NERICA in 2006 but not in 2004. Table 1 presents the data on

the household and community characteristics of the four adoption categories.

The early adopters, who include the dropouts and continuous adopters, are more

experienced with rice cultivation than the late adopters and non-adopters. Among the

early adopters, the rice cultivation experience of continuous adopters is significantly

higher. This may suggest that learning from own experience heightens the profitability

of rice cultivation, which affects the dynamic adoption decision of NERICA, as shown

2 Two to three local council 1 (the lowest administrative unit in Uganda called LC1) constitute each NERICA growing area and our sample covers 29 LC1s.

5

in Moser and Barrett (2006). It is also clear that continuous adopters tend to be

younger than the other groups. The significant differences in the household head’s age

between continuous adopters and dropouts suggest that younger households are likely

not to abandon rice cultivation in a short period of time.

The early adopters are significantly more educated than the non-adopters and

late adopters. These findings are consistent with those in the large body of literature

on the adoption of agricultural technologies (Sunding and Zilberman, 2007). The early

adopters are also endowed with a larger number of adult household members than the

non-adopters and late adopters. Since NERICA is much more labor intensive than

other common field crops in Uganda (Kijima et al. 2008), this observation may be

expected. The early adopters also cultivate lager areas of land than others.3 It is,

however, important to recognize that the early adopters (both the dropouts and

continuous adopters) are almost equally educated and endowed with the family labor

and land, indicating that the reason for abandoning NERICA rice cannot be attributed to

lack of human capital, family labor and land. Regarding endowment of household

assets (such as furniture, bicycle, and electrical appliances) and livestock, a major

difference is found between the late adopters and other categories, suggesting that

3 Since mailo tenants have secure land rights according to the government Land Act of 1998, we include mailo tenanted land in the land area along with owned land. The difference between the sum of owned area and the mailo tenanted area and the cultivated area is fallow area.

6

poverty (or lack of assets) compelled some farmers to adopt NERICA later, even though

they may have been aware that it is potentially more profitable than other crops.

Female headed households are less likely to be the early adopters, which is consistent

with the common belief of women being disadvantaged, particularly in the rural African

setting. The proportion of Bakiga tribe, which is well recognized as a tribe of

hard-working people because they originate from land-scarce hilly areas, is markedly

high among the late adopters.

Being a new crop to many farmers in Uganda, the seed distribution programs for

NERICA rice in Uganda are intended to provide not only seed but also training on rice

production in areas where NERICA was newly introduced. As a consequence, the

availability of seed distribution program is expected to significantly affect NERICA

adoption. In fact, Table 1 clearly indicates that the early adopters in 2004 had better

access to seed distribution programs in 2004. However, availability of such programs

decreased in 2006 among all NERICA adoption categories, except in areas where the

late adoption took place. Therefore, it seems reasonable to hypothesize that the seed

distribution programs influenced NERICA adoption by providing information on the

new technology and seed. Because NERICA rice seed can be produced by farmers,

unlike improved seed for other crops, such as hybrid maize, it is also reasonable to

7

postulate that the effect of seed distribution programs is moderated over time as farmers

produce own seed.

One of the major constraints to NERICA adoption in 2004 was identified to be

the absence of rice millers in nearby towns to mill or buy the paddy rice. As is

indicated in Table 1, in 2004, a typical farmer had to haul 100 kg paddy sack by bicycle

to rice millers located 15 to 30 km away from one’s residence. However, the number

of rice millers, who are sometimes buyers as well, has increased rapidly in Uganda as a

whole (see Figure 1), presumably responding to the increasing demand for the rice

milling services that followed the increase in NERICA rice production. In all the four

sample categories, access to rice millers has improved in the past two years, which is

clearly reflected in the considerably shortened distance to the rice millers from a

distance of 15 to 30 km in 2004 to 6 to 11 km in 2006, particularly among the

continuous adopters and late adopters. These observations indicate that improved

access to the market for paddy is a critical factor promoting NERICA adoption.

The relative profitability of NERICA can be an important determinant of its

adoption. Since NERICA is highly labor intensive, its adoption may be higher in

communities where land is scarce relative to labor. However, data on community

(village) area per household suggests that the availability of land is similar across the

8

four adoption categories. Since maize is a major alternative crop to rice that is less

labor intensive, the relative price of maize to rice can be an important factor affecting

the relative profitability and, thus, adoption of NERICA rice.

Given that NERICA in Uganda is grown without irrigation yet it requires more

water than the other subsistence crops, reliable and sufficient amount of rainfall is likely

to have a critical impact on rice yields. Table 1 shows the rainfall in 2004 and 2006, as

well as the average rainfall for 2001-06 periods and its coefficient of variation between

2001 and 2006, which intends to capture the long-term rainfall patterns. The rainfall

in 2004 and 2006 as well as the six-year average is not significantly different across the

four adoption categories. However, continuous adopters tend to be located in areas

with lower rainfall variation than the other categories, suggesting that rainfall reliability

is critical for sustainable adoption of NERICA.

Table 2 shows NERICA yield on sample plots and sources of NERICA seed in

2004 and 2006 by the different adoption categories. The average yield for continuous

adopters in 2004 is 3.0 tons per hectare, which attests the high yield potential of

NERICA.4 It is also significantly higher among the continuous adopters than among

the dropouts in the 2004 data set. This result is to be expected because the poor

4 The average yield of upland rice in SSA is about 1 ton per hectare (Balasubramanian et al. 2007).

9

performance of NERICA production likely discourages NERICA production in

subsequent periods.5 On average, the yield for late adopters is much lower than that

for continuous adopters in the 2006 data set. This low yield among the late adopters

may be attributed to their lower human and physical capital (as shown earlier in Table 1)

and lower experience in rice cultivation than the continuous early adopters.6

There are four possible ways through which farmers obtain NERICA seed: (1)

participating in the seed distribution program, (2) directly purchasing seed from seed

companies, (3) using own seed saved from the previous harvests, or (4) purchasing seed

from other farmers. Note that the seed distribution programs procured certified seed

from seed companies and distributed it to farmers via NGOs and extension workers, but

the direct purchase of seed from seed companies by individual farmers are observed

mainly in areas close to seed companies where farmers engage in contract farming for

the seed companies which advertise farmers to purchase treated seeds from the company.

In 2004, the proportion of sample plots planted to seed obtained either from seed

companies or from seed distribution program reached to 80-90%. Use of own seed

5 A recent study conducted in Uganda (Sserunkuuma, 2008) shows that the occurrence of severe drought conditions during the cropping season significantly reduced rice yield and negatively affected the number of Japan International Cooperation Agency (JICA)-trained households growing NERICA in the subsequent cropping seasons. 6 Sserunkuuma (2008) shows that farmers trained by JICA in 2007 had significantly lower NERICA yield than those trained in 2005 who started growing NERICA earlier and accumulated experience over the years.

10

and seed purchased from neighbors were rare in 2004, but the proportion of farmers

using farmer-produced seed increased to 54% of the continuous adopters and to 52% of

the late adopters in 2006. While NERICA seed can be produced by farmers

themselves like other rice varieties, farmers have to remove undesirable plants to obtain

“pure” or high-quality rice seed. Once farmers learn how to produce high-quality seed,

the farmer-produced seed can be as good as that sold by seed companies.7 Given that

rice production was only recently started by many of our sample farmers, there is a

possibility that the quality of farmer-produced seed is not as good as that of purchased

seeds, unless the extension service provides the required information for seed

production. If pure seed is not produced, the quality of seed is expected to deteriorate

gradually over time, thereby lowering rice yield. As shown in Table 2, the average

yield of self-produced seeds tends to be lower than the other seed types (except

continuous adopters in 2004). In particular, the yield from self-produced seeds among

continuous adopter in 2006 is far lower than the other seed types, which suggests the

possibility of seed-quality deterioration.

Even controlling for the effects of using different types of seeds, however, the

yields of continuous adopters are higher than the other groups. Therefore, it may be

7 This is why seed suppliers cannot make large profits in Asia, where farmers are adept at producing high-quality seeds. In Uganda, certified seeds are actually produced by farmers under contract with seed companies, which provide detailed instructions for seed production.

11

the case that there are differences in management of rice cultivation other than selection

of seed type. Though field practitioners have observed that farmers growing NERICA

in plots with higher moisture content (such as the lower part of hill) could obtained high

yields, there are no statistical differences in plot choice across adopter types and in

yields between the plot types.

Similarly, it is widely believed that farmers who planted rice late tended to fail

crops since they missed enough rainfall. We constructed the dummy variable whether

farmers planted too late according to their planting date, in order to examine if the

differences in the timing of planting can explain the higher yields of continuous adopter

than the dropout and late adopters. In 2004, the continuous adopters are less likely to

plant late than dropouts, while in 2006, the proportion of late planning is higher among

continuous adopters than late adopter. In 2004, the plots with late planting ended up

with low yield (many are zero yield) among both dropouts and continuous adopters. In

2006, however, there is no such trend. Significantly different between continuous

adopters and late adopters is the proportion of plots with zero yields, which is

responsible for lowering the average yield of late adopters.

3. Model Estimation

In order to rigorously identify the critical factors affecting the changing behavior of

12

NERICA adoption, we conduct regression analysis in this section. The major

questions addressed are: (1) why so many early adopters dropped out; (2) what the

underlying factors are for the superior performance of continuous adopters; and (3) how

to increase NERICA yield in a sustainable manner. In particular, we are interested in

the effects of rainfall, the availability of seed programs, access to rice millers, and the

effect of using farmer-produced seeds on the adoption and yield performance of

NERICA rice.

To answer the first two questions, we use multinomial logit model to identify the

major factors underlying the four types of adoption decisions. Although we estimate

multinomial regressions with cross-sectional data, we use variables in 2004 and 2006

for time-variant characteristics, such as the availability of seed distribution programs

and the distance to rice millers. By comparing the estimated coefficients of these

variables in 2004 and 2006, we can infer the changing impacts of seed distribution

programs and access to rice millers.

To answer the last question, we estimate the yield function, separately for 2004

and 2006. Since the data on rice yield are available only for adopters, a sample

selection problem may arise. To control for this problem, we apply standard method

Heckman two-stage regression model and include a selection correction term calculated

13

from probit model of adoption decision in each year..8

To investigate the determinants of adoption decisions, we use adoption category

(dropouts, continuous adopters, late adopters, or non-adopters) as the dependent

variable, while the amount of rice harvest per hectare is used as a dependent variable in

the yield function. Possible determinants of the NERICA adoption and yield include

(1) a set of variables indicating the suitability of rice production and costs of acquiring

seeds and selling paddy rice such as the average rainfall, rainfall variation, availability

of seed program, and distance to rice miller, and (2) household characteristics such as

rice cultivation experience,9 the household head’s education, the number of adult male

and female household members aged between 15 and 59 years old, and asset holdings.

Note that while we use the average rainfall and its variations over the six-year

period in the estimation of adoption function as proxies for expected rainfall patterns,

we use rainfall in a single year (one cropping season) to explain NERICA yield in each

year. It must be also noted that while the availability of seed distribution program in

the village, which can be considered as exogenous, is used to explain NERICA adoption

and yield, the seed type (i.e., the use of program seeds, farmer-produced seed saved

from the previous harvest, or purchased from neighbors) is also used to explain the

8 The results of the first stage regression are provided from the authors upon request. 9 Rice cultivation experience for 2006 captures the experience in 2004 and 2005.

14

NERICA yield.10

4. Empirical Results

NERICA Adoption

Table 3 shows the marginal effects of the explanatory variables on NERICA adoption

evaluated at their mean values. Unexpectedly, higher average rainfall and lower

rainfall variation, which are favorable for the NERICA adoption, significantly raised the

probability of being late adopters. In contrast, higher rainfall variation increases the

probability of being dropouts. These results suggest that the new adoption in 2006

took place in areas that are more suitable for NERICA production, while some early

adopters in less suitable areas for NERICA production (manifested in higher rainfall

variation) stopped growing NERICA by 2006. In other words, some of the areas that

received NERICA dissemination program support at the start of the NERICA campaign

were likely to be mis-targeted.

The availability of seed distribution programs in 2004 increases the probability

of being early adopters and decreases the probability of being late adopters. In

contrast, the availability of seed programs in 2006 did not have any significant effect on

10 One may wonder if the use of self-produced and purchased seeds is endogenous. It is not necessarily so, however, because farmers are almost forced to use such seeds, if the seed distribution programs are unavailable. .

15

NERICA adoption. These distinct effects of NERICA dissemination program may be

taken to imply that because of the expanded use and trade of farmer-produced seed, the

importance of seed distribution programs as a driver of NERICA adoption significantly

reduced by 2006.

The coefficients of community area per household are positive and significant

for non-adopters and negative and weakly significant for dropouts and continuous

adopters, suggesting that in communities where land is abundant relative to population

(labor), there is a lower incentive to adopt labor-intensive crops like NERICA. The

high price of maize relative to rice may also partly explain the late adoption of NERICA

rice.

The effects of distance to rice millers/buyers are as expected: For the late

adopters the coefficient is positive in 2004 but negative in 2006, whereas for the

non-adopters the coefficient in 2006 is positive and highly significant, indicating that

NERICA is less likely to be adopted where the cost of marketing paddy rice is higher,

(i.e., closer to millers/buyers). Consistently the coefficients of this variable are

negative and significant for dropouts in 2004 and for continuous adopters in 2006.

These results strongly suggest that improvements in market access in the two-year

period partly explain the changing behavior of NERICA adoption. Traveling time to

16

the nearest town affects dropout and continuous (early) adoption positively and the late

adoption negatively. This suggests that proximity to towns was not originally

considered to be important by the efforts to promote NERICA adoption through seed

distribution.

The results clearly show that rice cultivation experience tends to promote the

adoption of NERICA, as reflected in its positive and significant coefficients for the

three categories of adopters and its negative coefficient for non-adopters. This is likely

because experience tends to reduce the risk of crop failure as well as inappropriate

cultivation practice and thus increases the probability of adoption. For farmers without

experience, rice cultivation may be considered riskier than other food crops with

negative consequences for rice adoption (Kijima et al., 2008). It is also important to

note the positive effect of household head’s education on continuous adoption and the

negative effect on non-adoption. It appears that the ability to decode new information

and the rice production knowledge acquired through experience do matter in NERICA

adoption.

The households belonging to the Bakiga tribe tend to adopt NERICA later or not

to adopt at all, judging from the positive and significant coefficient for the late adopters

category and the negative and significant coefficient for the non-adopters category.

17

Being a predominantly migrant tribe in NERICA growing areas, limited access to

suitable land and seed for NERICA production may partly explain this. The

coefficient of the number of male adult members is negative for the late adopter

category, which is unexpected in view of the high labor requirement of NERICA

cultivation.

It is important to note that, apart from livestock for the later adopters, asset

ownership (landholdings, household assets, and livestock) do not significantly affect

NERICA adoption. Presumably this is because purchased inputs, such as chemical

fertilizers, are seldom used in NERICA production in Uganda, and seed was provided

free of charge or as in-kind credit to the early adopters. This finding is also consistent

with the finding of Kijima et al. (2008) that NERICA is a pro-poor crop.

In conclusion, the suitability of agro-ecological conditions for NERICA

production (measured by rainfall patterns and relative land abundance) and transaction

costs (measured by accessibility to rice millers, and availability of seed distribution

program) are critical factors explaining the changing behavior of NERICA adoption.

Our analysis also suggests that the inappropriate targeting of areas for NERICA

promotion by the seed distribution programs is responsible, at least partly, for the

massive dropouts.

18

NERICA Yield

Table 4 reports results of the estimated NERICA yield function. As was explained in

the previous section, the yield function is estimated separately for each year with

self-selection correction terms. The first two columns use availability of seed program

to examine possible effects of differences in seed quality, while the latter two columns

include three seed type dummies for the same purpose.11 As mentioned earlier,

although these seed type dummies can be endogenous in the yield function, we failed to

find suitable instruments for these variables. Thus, we have to interpret the estimated

coefficients with caution. The adopters in 2004 are the dropouts and continuous

adopters, while the adopters in 2006 are the continuous adopters and late adopters.

Interestingly, there are some marked differences in determinants of NERICA

yield between 2004 and 2006. One is the availability of seed distribution program for

the first two columns and the use of seeds obtained from seed distribution program for

the last two columns: These variables significantly affect NERICA yield only in 2006.

The estimated coefficients literally imply that rice yield increases by as much as 0.38

ton per hectare if seed distribution program is available, or 1.14 tons per hectare if seeds

from seed distribution program were used instead of self-produced seeds (see the last

11 A comparison group is those who used self-produced seeds.

19

part of Table 4).12 These findings suggest that the quality of farmer-produced seed has

substantially deteriorated during the past two years, leading to reduced yield. The

possible reason for seed quality deterioration is the failure of extension service to

provide proper information for production of high-quality seed. In order to fully

exploit NERICA’s high-yielding traits, there is an urgent need to provide appropriate

information to farmers on how to produce good seed. Note that the coefficient of

seeds purchased from seed companies is negative. Since the purchased seeds from

seed companies are supposed to be the same quality as the program seeds, it is possible

that the positive coefficient of availability of seed distribution program on NERICA

yield also capture the effect of information and training provided by the seed program.

Higher rainfall seems to have increased yield, which is expected in Uganda

where upland NERICA rice is grown under rainfed conditions. Thus, NERICA rice

production ought to be promoted in areas with sufficient and reliable rainfall to enable

farmers to enjoy its high production potential without exposing them to high production

risk. The coefficient of distance to rice millers changed from being positive in 2004 to

negative in 2006, which means that the reduction in marketing costs contributed not

only to increased adoption but also to increased yield. Another important finding is

12 It may well be that those farmers who sell seeds are specialized and knowledgeable rice farmers.

20

that the rice production experience tends to increase yield in both years, suggesting that

there is potential for farmers to raise yield from learning-by-doing. The knowledge

acquired through experience, however, can be also acquired through training.

The age of household heads has negative and significant coefficients, as may

be expected, while their education has a negative (unexpected sign) and significant

coefficient in 2006. The coefficients of per capita land are positive and significant

only for the latter two columns, which is probably due to the differences in cropping

pattern between small and large farmers. As per capita land increases, the proportion

of households who planted rice after tobacco and leguminous crops in the previous

season increases. Since ample application of chemical fertilizer to tobacco fields and

nitrogen fixation of leguminous crops raise soil nutrients on NERICA fields (Kijima et

al. 2006), positive relationships emerge between the per capita land and yield.

Finally it must be pointed out that judging from the estimation results of yield

function, significantly higher yields of continuous adopters than dropouts in 2004 and

than late adopters in 2006, as reported in Table 2, can be explained largely by

significantly longer experience of rice cultivation and younger age of household heads

among continuous adopters than others (see Table 1).

5. Conclusion

21

Using panel data of 347 households collected in 2004 and 2006 in rural Uganda, we

identified four types of NERICA adoption behaviors; continuous adoption in the two

years, dropout, late adoption, and non-adoption. A major determinant of dropout,

which accounts for 37% of our sample households, is large variation in rainfall,

indicating that some farmers adopted NERICA in 2004 in the areas unsuitable for its

production. We found that the availability of seed distribution programs was a critical

determinant of NERICA adoption in 2004 but not in 2006, most likely because the use

of farmer-produced seed was widespread in 2006. The use of farmer-produced seed,

however, led to substantial reduction in rice yield in 2006, suggesting that farmers do

not have appropriate knowledge on the production of high quality seed. It is also

important to note that seed was traded among rice farmers, suggesting emergence of an

informal market for NERICA seed. Shortened distance to rice millers, whose number

has increased rapidly in the past 2 years in response to increasing demand for milling

services, significantly increased NERICA adoption. These findings suggest that the

extension system failed to disseminate appropriate production knowledge, particularly

for seed, but improved access to rice millers and the use of farmer-produced seed

stimulated wider dissemination of NERICA rice.

The first policy implication of this study is that in order to achieve wider

22

dissemination of NERICA and to realize its yield potential, the extension system must

be strengthened. According to our nation-wide survey of about 900 farmers in Uganda

conducted in 2005, the adoption rate of NERICA is disappointingly low, ranging

between 1% and 2% (Kijima and Sserunkuuma, 2008). This study shows that the

failure of widespread NERICA diffusion is partly due to inappropriate extension

activities to promote NERICA in unsuitable areas, such as those predisposed to

excessive variations in rainfall. Failure to disseminate appropriate methods for

producing high-quality farmer-produced seed is another important factor, which will

likely reduce not just the yield of NERICA but the adoption of NERICA as well.

Capacity building for extension workers and allocation of resources to extension

activities are keys to realization of a “NERICA Revolution” in this country.

The second implication is that rice development policies should be so designed

as to support the development of the market for rice seed and milling services, as they

promote both NERICA adoption and yield. Although we do not have concrete

evidence, in all likelihood, the development of rice milling market for paddy rice

responded to increased demand for these services. The fact that the distance to rice

millers declined more significantly in areas where continuous adopters are located than

those where dropouts and non-adopters are located is consistent with our conjecture.

23

The burgeoning trade of seed from experienced rice farmers to new farmers is indicative

of the emergence of an informal seed market. Thus, there seems to be virtuous circle

of increased production and market improvement. To the extent that these markets

“fail” due to credit constraints and imperfect information about seed production and rice

milling as well as the quality of seed, there is room for the government to support the

further development of these markets for both efficiency of rice production and poverty

reduction among smallholders in Uganda.

24

References

Agribusiness Development Center (ADC) (2001). Upland Rice Production and

Marketing Feasibility Study. Kampala: Independent Consulting Group.

Alphonse, C., Richard, S., Samuel, O., Tobias, O. (2008). “Survey Report on the Status

of Rice Milling Industry in Uganda.” Mimeo, Japan International Cooperation

Agency.

Balasubramanian, V., Sie, M., Hijmans, R.J., and Otsuka, K. (2007). “Increasing Rice

Production in Sub-Saharan Africa: Challenges and Opportunities.” Advances in

Agronomy, 94: 55-133.

Bandiera, O., Rasul, I. (2006). “Social Networks and Technology Adoption in Northern

Mozambique.” Economic Journal 116: 869-992.

Barrett, C., Moser, C., McHugh, O., Barison, J. (2004). “Better Technology, Better Plots,

or Better Farmers? Identifying Changes in Productivity and Risk among

Malagasy Rice Farmers.” American Journal of Agricultural Economics 86:

869-88.

Cameron, L. (1999) “The Importance of Learning in the Adoption of High-Yielding

Variety Seeds.” American Journal of Agricultural Economics 81: 83-94.

Carletto, C., de Janvry, A., Sadoulet, E. (1999) “Sustainability in the Duffusion of

25

Innovations: Smallholder Nontraditional Agro-Exports in Guatemala.”

Economic Development and Cultural Change 47: 345-69.

Conley, T., Udry, C. (2005) “Leaning about a new technology: pineapple in Ghana.”

Mimeo, Yale Univeristy.

Dadi, L., Burton, M. Ozanne, A. (2004) Duration Analysis of Technological Adoption in

Ethiopian Agriculture. Journal of Agricultural Economics 55:3, 613-631

Dercon, S., Christiaensen, L. (2007) “Consumption risk, technology adoption and

poverty traps: evidence from Ethiopia.” World Bank Policy Research Working

Paper 4257.

Duflo, E., Kremer, M., Robinson, J. (2007) “Why Don’t Farmers Use Fertilizer?

Experimental Evidence from Kenya.” Mimeo, Department of Economics,

MIT.

Feder, G,., Just, R., Zilberman, D. (1985) “Adoption of Agricultural Innovations in

Developing Countries: A Survey”, Economic Development and Cultural

Change 33: 255-98.

Kherallah, M., Delgado, C., Gabre-Madhin, E., Minot, N., Johnson, M. (2000) “The

Road Half-Traveled: Agricultural Market Reform in Sub-Saharan Africa,”

Food Policy Report, International Food Policy Research Institute.

26

Kijima, Y., Sserunkuuma, D., Otsuka, K. (2006) “How Revolutionary is the “NERICA

Revolution”? Evidence from Uganda,” Developing Economies 44(2): 252-67.

Kijima, Y., Otsuka, K., Sserunkuuma, D. (2008) “Assessing the Impact of NERICA on

Income and Poverty in central and Western Uganda,” Agricultural Economics

38: 327-37.

Kijima, Y., Sserunkuuma, D. (2008) “The Adotion of NERICA Rice Varieties at the

Initial Stage of the Diffusion Process in Uganda,” East African Journal of

Rural Development (forthcoming).

Moser, C., Barrett, C. (2003) “The disappointing adoption dynamics of a

yield-increasing, low external-input technology: the case of SRI in

Madagascar.” Agricultural Systems 76: 1085-100.

Moser, C., Barrett, C. (2006) “The complex dynamics of smallholder technology

adoption: the case of SRI in Madagascar.” Agricultural Economics 35: 373-88.

Neill, S., Lee, D. (2001) “Explaining the Adoption and Disadoption of Sustainable

Agriculture: The Case of Cover Crops in Northern Honduras.” Economic

Development and Cultural Change 49: 793-820.

Otsuka, K., Kijima, Y. (2008) “Technology Policies for a Green Revolution and

Agricultural Transformation in Africa.” Presented in Plenary Session of

27

Biannual Meeting of African Economic Research Consortium, June 2008.

Spencer, D., Dorward, A., Abalu, G., Philip, D., Ogungbile, D. (2006) “Evaluation of

Adoption of NERICA and Other Improved Upland Rice Varieties following

Varietal Promotion Activities in Nigeria” A Study for the Gatsby and

Rockefeller Foundations Final Report.

Sserunkuuma, D. (2008) “Assessment of NERICA Training Impact.” A Study Report

Prepared for the Japan International Cooperation Agency (JICA).

Sunding, D., Zilberman, D. (2001) “The Agricultural Innovation Process: Research and

Technology Adoption in a Changing Agricultural Sector,” Handbook of

Agricultural Economics, B. Gardner and G. Rausser, eds, vol. 1A: 207-61.

Amsterdam: Elsevier.

Suri, T. (2007) “Selection and Comparative Advantage in Technology Adoption.”

Mimeo, Sloan School, MIT.

Tsuboi, T. (2008) “Towards Rice Green Revolution in Africa: The Case of NERICA

Promotion in Uganda.” Presented in Tokyo Conference on African

Development, Yokohama, May 2008

Uganda Bureau of Statistics (UBOS). (2002) Uganda National Household Survey

1999/2000. Report on the crop Survey Module. Entebbe: UBOS.

28

West Africa Rice Development Association (WARDA). (2001) “NERICA Rice for

Life.” http://www.ward.org/publications/NERICA8.pdf (accessed March 27,

2005).

29

Figure 1 Total Number of Rice Mills in Uganda

0

100

200

300

400

500

600

700

1999 2000 2001 2002 2003 2004 2005 2006 2007

Source: Alphonse et al. (2008), Figure 18.

30

Table 1 Household and Community Characteristics by Type of Adopters a

Non-

adopters Dropouts

Continuous

adopters

Late

adopters

Number of Households 94 129 99 25

Household characteristics

Rice cultivation experience (years) 0.07 1.49 1.88 0.50

Household head’s age 48.7 48.2 43.9 49.4

Household head’s years of schooling 4.9 7.0 7.7 4.5

Number of adult male aged 15-59 1.43 1.80 1.81 0.99

Number of adult female aged 15-59 1.50 1.96 1.68 1.34

% Female headed households 32.8 10.2 8.0 29.7

% Bakiga tribe 0.4 8.3 9.0 20.3

Land area per capita (ha) b 0.24 0.38 0.47 0.38

Land area per household (ha) b 2.56 4.52 4.23 2.53

Land cultivated in 2nd season (ha) 0.92 1.19 1.30 0.78

Household asset (USD) 160 149 172 54

Value of livestock (USD) 307 371 390 80

Community characteristics

Availability of Seed Program in 2004 (%) 18.6 37.4 33.7 17.3

Availability of Seed Program in 2006 (%) 10.7 20.4 28.6 23.9

Distance to rice miller in 2004 (km) 19.0 15.4 26.9 28.9

Distance to rice miller in 2006 (km) 14.1 11.1 6.2 5.5

Traveling time to town (hour) 0.42 0.62 0.77 0.66

Community area per household (mile2) 0.023 0.020 0.022 0.019

Relative price of maize to rice 0.385 0.512 0.465 0.448

Annual rainfall in 2004 (mm) 368.5 424.0 429.6 393.0

Annual rainfall in 2006 (mm) 446.2 435.8 432.0 450.2

Average annual rainfall c 409.4 435.5 441.3 421.6

C.V. of rainfall c 0.20 0.17 0.15 0.17

a The data pertain to 2004 unless stated otherwise.

b Land area refers to owned land and tenanted land under the mailo regime.

c Average of six years from 2001 to 2006.

31

Table 2 NERICA Yields and Sources of Seeds by Type of Adopters 2004 2006

DropoutsContinuous

adopters Continuous

adopters Late

adoptersNumber of plots 129 107 100 23 Yield (ton per ha) 2.01 2.97 2.54 1.49 % Self-produced seeds 5.2 7.7 41.5 5.8 % Purchased seed from neighbors 3.8 11.7 12.0 46.5 % Program seeds (NGO, VP) 53.8 42.9 15.0 10.2 % Other seeds (purchased from traders, contract farming) 37.2 37.7 31.5 37.5 Yield in plots from self-produced seeds 0.99 3.41 2.06 1.15 Yield in plots from purchased seeds from neighbors 1.35 2.72 2.97 1.17 Yield in plots using program seeds 1.76 2.96 2.95 3.46 Yields in plots using seeds from seed companies 2.49 2.99 2.75 1.39 % plots in low lying location 11.8 14.1 21.7 21.8 Yield in plots in low lying location 2.04 2.83 2.09 1.22 Yield in plots not in low lying location 2.01 3.00 2.67 1.56 % Late plantation 8.5 2.8 11.1 6.1 Yield in plots without late plantation 2.20 3.05 2.56 1.43 Yield in plots with late plantation 0.68 0.63 2.44 2.31 % Plots with zero yield 4.4 0.4 6.9 24.9 Size of NERICA plot (ha) 0.423 0.377 0.471 0.241

32

Table 3 Determinants of NERICA Adoption (Multinomial Logit Model, Marginal Effects) a Dropouts Late

adopters Continuous

adopters Non-adopters

Community characteristics Average Annual Rainfall, mm (2001-2006) -0.000 0.001 -0.000 -0.000 (-1.50) (3.02)** (-0.26) (-1.44) C.V. of rainfall (2001-2006) 0.421 -0.655 -0.131 0.365 (1.89)+ (-3.08)** (-0.74) (1.08) Availability of seed program in 2004 dummy 0.096 -0.122 0.082 -0.056 (1.93)+ (-1.80)+ (2.24)* (-0.59) Availability of seed program in 2006 dummy 0.017 0.077 -0.021 -0.073 (0.33) (1.08) (-0.59) (-0.72) Relative price of maize to rice in 2004 0.042 0.143 -0.042 -0.142 (0.65) (1.73)+ (-0.99) (-1.27) Community area per household in 2004 -3.286 -1.561 -2.112 6.959 (squared mile) (-1.93)* (-0.92) (-1.77)+ (2.46)** Distance to rice miller (km) in 2004 -0.002 0.006 -0.000 -0.004 (-1.93)* (3.13)** (-0.09) (-1.56) Distance to rice miller (km) in 2006 -0.001 -0.006 -0.003 0.007 (-0.63) (-2.48)** (-2.30)* (2.58)** Traveling time to town (hours) in 2004 0.165 -0.201 0.145 -0.109 (2.44)* (-2.24)* (2.50)** (-0.90) Household characteristics in 2004 Rice cultivation experience (year) 0.063 0.060 0.039 -0.163 (2.32)* (1.90)+ (2.00)* (-2.50)** Number of male adult members aged 15_59 0.021 -0.069 0.005 0.043

(1.42) (-2.74)** (0.43) (1.38) Number of female adult members aged 15_59 0.023 0.007 0.002 -0.032

(1.49) (0.38) (0.16) (-1.14) Female headed household dummy -0.075 0.022 -0.024 0.078 (-1.65)+ (0.54) (-0.65) (1.15) Bakiga tribe dummy 0.116 0.367 0.034 -0.517 (1.28) (2.62)** (0.60) (-2.35)* Household head’s age 0.001 0.000 -0.001 -0.000 (0.46) (0.27) (-0.63) (-0.14) Household head’s education (years of schooling) 0.005 0.007 0.008 -0.020 (1.32) (1.38) (2.39)* (-2.68)** Per capita land owned (mailo included) (ha) -0.001 0.036 0.002 -0.037 (-0.05) (1.13) (0.09) (-0.69) Household asset (thousand US$) 0.009 -0.034 0.005 0.021 (0.18) (-0.37) (0.14) (0.19) Value of livestock (thousand US$) 0.010 -0.155 0.016 0.130 (0.29) (-2.05)* (0.68) (1.53) Constant -0.256 -0.292 -0.102 0.650 (-2.10)* (-1.90)+ (-1.02) (3.10)**

a The numbers in parentheses are t-statistics. **, *, and + indicates the significance at the 1%, 5%, and 10%, respectively.

33

Table 4 Determinants of NERICA Yield a 2004 2006 2004 2006 (1) (2) (3) (4) Community characteristics Availability of seed program -0.347 0.378 (1.24) (2.64)** Rainfall during the cropping season -0.000 0.003 0.003 0.002 (90 days, mm) (0.06) (4.08)** (2.76)** (3.03)** Distance to rice miller (km) 0.010 -0.023 0.008 -0.024 (1.50) (2.30)* (2.04)* (2.54)* Traveling time to town (hours) 0.160 0.172 0.292 0.450 (0.42) (0.99) (1.45) (2.40)* Household characteristics Rice cultivation experience (year) 0.073 0.124 0.031 0.128 (2.04)* (6.51)** (1.29) (6.86)** Number of male adult members 0.217 0.064 -0.176 0.137

aged 15_59 (1.68)+ (0.86) (2.86)** (1.86)+ Number of female adult members 0.162 -0.114 -0.024 -0.165

aged 15_59 (1.08) (1.54) (0.33) (2.23)* Female headed household dummy -0.106 -0.216 -0.341 -0.131 (0.22) (1.15) (1.26) (0.72) Bakiga tribe dummy -1.529 -0.328 0.159 -0.586 (3.12)** (1.69)+ (0.60) (3.20)** Household head’s age -0.032 -0.021 -0.026 -0.012 (3.19)** (5.00)** (4.26)** (2.92)** Household head’s education 0.045 -0.034 0.024 -0.026 (years of schooling) (1.37) (1.80)+ (1.19) (1.48) Per capita land owned (ha) -0.120 0.243 0.253 0.253 (mailo included) (0.51) (1.56) (2.49)* (1.66)+ Household asset (thousand US$) 0.082 0.495 -0.242 0.577 (0.18) (1.65) (0.89) (1.99)* Value of livestock (thousand US$) -0.265 0.365 -0.291 0.280 (0.92) (2.02)* (2.01)* (1.55) Plot level characteristics Seed is produced by neighbors b -0.294 0.051 (dummy) (0.64) (0.31) Program seed b -0.614 1.140 (1.58) (5.60)** Purchased from seed company b -0.670 -0.219 (1.62) (1.22) Constant 2.843 1.745 2.386 1.095 (3.50)** (4.78)** (4.65)** (2.37)* Self-selection correction term -2.317 1.343 0.237 2.361 (0.88) (0.79) (0.50) (1.43) R-squared 0.22 0.30 0.21 0.35

a The numbers in parentheses are t-statistics. b Comparison group is self-produced seed. **, *, and + indicates the significance at the 1%, 5%, and 10%, respectively