INTRODUCTION
Rice is staple and strategic crop in Malaysia. The consumption for rice in
2012 was about 2.1 million t and the production was only able to meet 68.5%
of that consumption (Department of Statistics Malaysia, 2011).
Despite those strategic function, the increase of rice production triggered
by the land expansion is quite difficult currently due to the decreasing of
land used for the food crop since the country’s rapid economic development
occupies more agricultural area mainly for housing, business and industrial
purposes. In 1960, for example, land used by food crops accounted for 31.5%
of the total agricultural land in Malaysia, then it has decreased to 16.3% in
2005 (Alam et al., 2010). Further, for two last
decades, total number of paddy area is not more than 0.7 million ha with the
average growth only about 0.27% year^{1}.
This fact has prompted the Malaysian authority to consistently increase the rice production by the improvement of the yield through the utilization of the optimal input used, new technology, farm management and provides the insentive for farmers in increasing production such as the paddy price support and the yield increase insentive. For example, government provide various input subsidy schemes which are 240 kg ha^{1} mixed fertilizer and 80 kg ha^{1} for organic fertilizer as well as RM 200/ha/season subsidy for pesticide control. The price support is currently at RM 248.1 t^{1} with the guaranted minimum price of RM 750 t^{1}.
Although, there have been many efforts and policies on rice farming, however,
there was no significant improvement in the yield. The average yield at 3.9
t ha^{1} (Department of Statistics Malaysia, 2011)
was not differ from previous studies by Coelli and Battese
(1996) that mentioned the actual paddy farm yields in Malaysia vary from
35 t ha^{1}. Comparing to the neighboring countries that yield was
lower than Indonesia and Vietnam at 4.9 and 5.5 t ha^{1}, respectively
(FAOSTAT, 2012). Further, rice farms in Malaysia were
characterized as less efficiently managed farms compared to industrial farms
since rice farms mainly enganged by small farmers and are not so wellmanaged.
There were 677,884 ha paddy planted which was managed by 300.000 farmers with
average farm size about 1.45 ha (Man and Sadiya, 2009).
Those condition conceive that the difficulties in improving the yield is potentially
caused by the unintensive use of input due to the inefficient management on
rice farm. Thus, in this context, the measurement of the existing farms efficiency
much more useful since it could provide the information about the gap of efficiency
performance among the farms and the potential to be improved (Kumbhakar
and Lovell, 2000). It also shows the possibility to increase the yield without
increasing the resourse base or developing new technology and it helps determine
the under and over utilization of inputs (PadillaFernandez
and Peter, 2012). Moreover, the analysis of technical efficiency in agricultural
sector has been widely used in developing countries due to the importance of
productivity growth in order to improve the economic development (Ogundari,
2009). Therefore, this study aims to measure the rice farm efficiency and
factors affecting that efficiency.
METHODOLOGY
Data envelopment analysis: Data Envelopment Analysis (DEA) was formally
developed and named by Charnes et al. (1978) where
efficiency was defined as the weighted sum of outputs over the weighted sums
of input in the constant return to scale assumption. Further, Banker
et al. (1984) extended the model to consider the Variable Return
to Scale (VRS) assumption and named as the pure technical efficiency. DEA involves
the use of liner programming methods to construct a non parametric piece wise
surface of frontir over the data. Then, efficiency measure are calculated relative
to this surface and this technique identifies the efficient production unit
which belong to frontier, otherwise the inefficient ones is remain below the
frontier (Coelli et al., 2005). Thus, DEA assumes
there are no random effects in the production and does not require the specification
of the production function. It just uses a set of inputs that farm want to minimize
and a set of output that farms want to maximize. Theoretically, technical efficiency
can be examined from an inputorientation or outputorientation. Input orientated
technical efficiency means that farms minimize the quantity of inputs while
holding output constant and the output orientated technical efficiency means
a farm want to maximizes the output given the fixed current quantity of inputs.
Two measures provide the same technical efficiency scores when Constant Return
to Scale (CRS) technology applied but are unequal when Variable Return to Scale
(VRS) was applied.
This study focused on the measure of output orientated technical efficiency
due to the main concern on Malaysian rice farming is maximizing the output from
a given set of input. Further, we examine the Variable Returns to Scale (VRS)
assumption on paddy farm technology since the CRS assumption is more appropriate
when all farms are operating at an optimal scale and it is not supported by
actual condition on rice farms. The measure on technical efficiency based on
the VRS assumption is also named pure technical efficiency since it is free
of scale effects (PadillaFernandez and Peter, 2012).
By assuming there are n farms which produces a single ouput using i different
inputs and the Variable Returns to Scale (VRS) output oriented DEA model, developed
by Charnes et al. (1978) can be expressed as:
where, 1≤φ<∞ and φ1 is the proportional increase in outputs that could be achieved by the ifarm, with input quantities held constant. Technical efficiency score is defined as 1/φ and varies between zero and one.
It is possible that eventhough the production units technically efficient but they are not equally productive due to the effects of scale. If the underlying production technology is a globally Constant Return to Scale (CRS), then the production unit is automatically scale efficient. However, when the farms might be too small in its production scale or the production unit are too large and it may operate within the decreasing return to scale, efficiency level might be improved by changing their production scale or the size of operation. Thus, the scale efficiency measurement is one of the important view on paddy farms management.
Scale efficiency is measured as the ratio of technical efficiency on CRS to
technical efficiency on VRS. Following Fare et al.
(1994), we can define an output orientated measure of scale efficiency at
a given input, x and the output, q as follows:
Bootstrap DEA: The recently advance on DEA estimate is the using of
bootstrapping technique which is inspired by the drawback of DEA approach. According
to Schmidt (1986), DEA approach did not assume the statistical
noise so that all the error term was attributed to inefficiency. Therefore,
the efficiency scores generated by DEA were not very robust and highly sensitive
to sample selection. Further, Simar and Wilson (2000)
mentioned that DEA as the nonparametric approach has been characterized as the
deterministic as if to suggest that the method lack any statistical properties.
The bootstraped DEA was suggested by Simar and Wilson
(2000) that derived from some unobservable data generating process, could
remove inherent dependency among efficiency scores and to obtain the bias corrected
DEA efficiency scores. The bootstrap is defined as the resampling technique
as a mean of approximating the properties of the sampling distribution of an
estimator when this is difficult to obtain by using alternative means and hence
allowing one to construct the confidence interval (Simar
and Wilson, 2000).
According to Nastis et al. (2012), the bootstrap
method is aimed to analyze the sensitivity of efficiency scores relatives to
the sampling variations of the estimated frontier and provide the statistical
basis for nonparametric efficiency measures. Hence, bootstrapping is the useful
way illustrating the sensitivity of DEA efficiency estimates to the variation
in sample composition. In particular, the width of the confidence interval for
the efficiency of farms located on the fringes of the data set will tend to
be quite wide, indicating that the degree to which these estimates are generally
based upon rather thin data and hence should be interpreted cautiously. Then,
when one has a small sample and a large number of dimension, the confidence
interval tend to be wide (Diler, 2011).
Data: Data were collected from the survey that conducted at Muda Agricultural Development Authority (MADA), the largest granary areas in Malaysia. Research sample of 150 farmers were drawn using simple random sampling. The data collection used a structured questionnaire on farmer’s production activities including input and output on paddy farm as well as socioeconomic characteristics.
In measuring the technical efficiency level of individual farms, one output and five outputs were used whereas the output was defined as the quantity of paddy production for one season (t). Five production inputs included land (ha), seed (kg), fertilizer (kg), pesticide (L), labor (man h). After detecting the outler, from 150 samples, we dropped eight extreme observations as the outliers in order to reduce the possibility of DEA estimates sensitivity to those outliers. Then efficiency scores were recalculated using the final sample of 142 farms.
EMPIRICAL RESULTS
Technical efficiency estimate: The summary of statistics for variables gathered from the survey are reported in Table 1. The average paddy production of the sampled farms was 2.37 t with the minimum production at 1.05 t and maximum production at 5.42 t. Standard deviation of the production was quite high (77.38) which indicated the large variability on paddy production among the sampled farms.
On average, the seed used on paddy farms was about 68 kg and there were some farms used it until 113 kg. Yet, the variability of used seed among the sampled farms was not large since the standard deviation for seed was lower than for other inputs. The large variability on input used among farms was found on the utilization of fertilizer which was 83.79 and indicated that many farms did not comply with a good farm practice especially in the utilization of the fertilizer.
The result of VRS or pure technical efficiency, its bootstrapping methods and
the scale efficiency are presented in Table 2. On the average,
the bias corrected TE scores were significantly lower than the VRS TE scores
and similar result have been reached by Linh (2012)
for rice farming in Vietnam.
Table 1: 
Summary statistics of variables used on the study 

SD: Standard deviation 
Table 2: 
Technical and scale efficiency score estimated by DEA and
bootstrap methods 

The technical efficiency score estimated by DEA was about 0.6375 and means
that with a given amount of inputs, on average, the rice farms could increase
its output by 57.31%. These result was consistent with Thiam
et al. (2001) on his study about technical efficiency in developing
country agriculture, Nargis and Lee (2013) in the study
on efficiency analysis of boro rice production in north central region of Bangladesh
and PadillaFernandez and Peter (2012) with study about
farm size and its effect on productive efficiency of sugar cane farm in Central
Negos, Phillipines.
However, after correcting for the bias, the technical efficiency score was 0.5366 and indicated that rice farms in MADA still could increase its output by 86.35% on the same level of inputs. This bias corrected TE score obtained from the bootstrap methods was more robust than the VRS TE estimated by DEA due to its adjustment to the sample variation. Further, by considering the lower and the higher bounds, on the average, the rice farms could increase its output in the range from 20.1399.12% with 95% confidence interval. This confidence interval was rather wide due to the small sample in this analysis.
Further, the scale efficiency of rice farms was about 0.8576. It implied that
essentially the average farms were very close to optimal scale since only 14.24%
additional productivity gain was feasible to reach the optimal scale by assuming
no other constraining factors. The lower VRS technical efficiency scores compared
to the scale efficiency scores suggested that inefficiencies were mostly due
to inefficient management or inefficient technical practices rather than the
scale of production or the size of operation. This result is consistent with
study by PadillaFernandez and Peter (2012), Yusuf
and Malomo (2007) and Rios and Shively (2005).
Table 3 shows the distribution of the paddy farm and seperated
on two categories. According to the VRS technical efficiency, the best practice
farms with the TE score above 0.90 were about 13.4% and it was higher than those
on the bias corrected TE (5.63%). Conversely, based on the bias corrected TE,
there were 3 farms (2.11%) which were not efficiently managed the rice production
and those farms were not founded on the VRS TE.
Table 3: 
Distribution of initial and corrected technical efficiencies 

These difference potentially caused by the sensitivity of VRS TE to the sample
variation.
The comparison of farms distribution on both categories can be expressed clearly as well by dividing those farms into below or above the average technical efficiency scores (0.63). On the VRS/pure technical efficiency categories, there were 46.5% paddy farms with the efficiency score above the average. Yet, on the bias corrected TE category, there were less farms (35.2%) with the efficiency score above the average value. Hence, according to the corrected bias TE, mostly paddy farms (59.2%) were not efficient due to its scores still below the average value.
Factor affecting the efficiency: Regarding the previous discussion that
inefficiencies on sampled rice farms were mostly due to the inefficient management
rather than the scale of operation, we attempt to examine factors affecting
the efficiency by following the two step approach as suggested by Coelli
and Battese (1996). Factors that supposed to be related to the management
on rice production were inluded on the analysis. As we know, a common practice
in the DEA literature for estimating the factors affecting the efficiency had
employed the Tobit model regression since the efficiency scores as the dependent
variables had the range from 01 (Sharma et al., 1999).
We regressed the estimated efficiency scores obtained from first step as a
function of explanatory variables which were sociodemographics variables, ownership
and access to credit. Those sociodemographics variables were the farmers’
age, household size, the main job (whether the full time farmer), education
and involvement on the extension activities. Two special variables were the
land ownership and access to credit were included in the model in order to estimate
the role of those variables to the rice farm efficiency. The following was the
Tobit model used in this study:
Ø_{i} = β_{0}+ β_{1}
Age_{i}+ β_{2} HHSize_{i}+ β_{3}
Job_{i}+ β_{4
}Ext_{i}+ β_{5} LandOwn_{i}+ β_{6}
Credit_{i}+ β_{7} Noedu_{i}+ β_{8}
Primary_{i}+
β_{9} Secondary_{i}+ β_{10} Highedu_{i}+ε_{i} 
The variable of age was defined as the household head’s age and the household size was the total number of household members. Other variables that might affect the technical efficieny were expressed as the binary/dummy variables. The dummy variables of job was divided into one if the paddy farmer as the main job and zero for others. The involvement on extension services was defined as one and zero for otherwise.
Table 4: 
Factors affecting the rice farm efficiency 

Values in parentheses are tstatistics; **,***significantly
at 5 and 1%, respectively 
Land ownership was another dummy variable which one for the farmers who has their own land and zero for others who rent the land. Further, access to credit that often characterized as the farms capability criteria to improve their production also expressed as dummy whereas one for farms that have the access and zero for otherwise. Household head’s education was divided into four categories: No formal education, primary school (from 16 years), secondary school (from 712 years) and high education ( 12 yeras and up).
The result in Table 4 shows that the model appears to fit data well due to the positive sigma coefficient and statistically significant at 1% level. Out of ten variables included in the model, household size, land ownership and secondary level of education provide the the significant effect to the rice farms efficiency. The positive significant effect of household size implied that farms with more household member was appeared to be more efficiently manage their production since many worker that potentially involved in the production.
Land ownership also had the significant effect to the rice farm efficiency but on the negative direction. It implied farmers who own the land were tend to be more inefficient than those who rent the land. It was related to the motivation on production whereas tenant farmers were more motivated to improve their production and get higher income so that they strived to manage the production in a professional manner and receptive to new technology as well. Further, from four categories of farmer’s education, farmers with secondary education level more efficiently managed the paddy farm than others because of their passion for managing the rice farm. Other factors such as age, job, involvement on extension service, acces to credit were not significant in affecting the rice farm efficiency.
CONCLUSION
The findings of this study show that on average, the bias corrected TE scores
(0.5366) were significantly lower than the variable return to scale TE scores
estimated by DEA (0.6375). Efficiency score estimated by DEA at 0.6375 means
that with a given amount of inputs, rice farms could increase its output by
57.31%. However, after correcting for the bias, the technical efficiency score
was about 0.5366 and indicated that rice farms in MADA could still increase
its output by 86.35%. By considering the lower and the higher bounds of efficiency
scores, on the average, the rice farms could increase its output in the range
from 20.1399.12% with 95% confidence interval.
The scale efficiency of rice farms at 0.8576 implied that essentially the average farms were very close to optimal scale. The interesting condition can be seen on the lower VRS technical efficiency scores compared to the scale efficiency scores because it suggested that inefficiencies were mostly due to inefficient management. Therefore, the analysis on factors affecting the efficiency is supposed to conducted by applying the Tobit regression model.
Three factors that significantly affect the rice farm efficiency were household
size, land ownership and secondary level of education of sampled farmers. The
positive significant effect of household size implied that farms with more household
member was appeared to be more efficiently manage their production. Then, the
negative effect of land ownership implied farmers who had the own land were
tend to be more inefficient than those who rent the land. It was related to
the motivation on production whereas tenant farmers were more motivated to improve
their production and get higher income so that they strived to manage the production
in a professional manner and receptive to new technology as well. Further, farmers
with secondary education level more efficiently managed the paddy farm than
others because of their passion for managing the rice farm.
ACKNOWLEDGMENT
This study was supported by the LRGS Food Security: Enhancing Sustainable Rice
Production through innovative Research, Vot. No. 5525017.