Toward Sustainability or Efficiency: The Case of Smallholder Coffee Farmers in Vietnam

Round 1

Reviewer 1 Report

The manuscript could be published as it is. The authors has incorporated the revised required.

Author Response

Dear Professor,

Thank you very much for your decision. 

Best regards, 

Reviewer 2 Report

Use of BC inefficiency effects with a Cobb-Douglas production function is quite out-dated. I would suggest consulting several of the more recent reviews of SFA, including Greene (2008) and Parmeter and Kumbhakar (2014) for more discussion on the state of the art for applied efficiency analysis. The BC approach only parameterizes the mean of the truncated normal. Why does the distribution have to be truncated normal? Why not exponential or half normal? Why not have the variance parameterized as well? None of these important modeling issues are even discussed.

With all the z variables you have there is no need to even specify the distribution of inefficiency, can be done with semiparametric methods. 

Additionally, no specification tests are undertaken for the production frontier, Cobb-Douglas is extremely narrow in scope. What about translog, generalized quadratic or some other flexible functional form. 

I wonder if selection is an issue between SC and non-SC coffee farmers. No real discussion of this is presented. See papers by Greene (2010) and Kumbakar, Tsionas and Sipilainen (2009) for more background. 

Lots of typos and grammatical issues.

Author Response

Dear Professor,

Thank you very much for reviewing this research paper.  We highly appreciate your concerns in regard of methodological approach. Your detailed remarks are important that help to improve the quality of the empirical research.

To the best of our knowledge, the following are details of our revision.

1. Use of BC inefficiency effects with a Cobb-Douglas production function is quite out-dated. I would suggest consulting several of the more recent reviews of SFA, including Greene (2008) and Parmeter and Kumbhakar (2014) for more discussion on the state of the art for applied efficiency analysis. The BC approach only parameterizes the mean of the truncated normal. Why does the distribution have to be truncated normal? Why not exponential or half normal? Why not have the variance parameterized as well? None of these important modeling issues are even discussed. With all the z variables you have there is no need to even specify the distribution of inefficiency, can be done with semiparametric methods. 

We are totally agreed that BC approach is quite out-dated. However, we have provided the reason of choosing generalized truncated normal distribution assumption of inefficiency effects model due to its computational simplicity as well as its ability to examine the effects of various specific variables of smallholder farmers, as opposed to 2 stage approach. Although, there is no priori basis of choosing one distribution over another. Different specifications lead to different estimates of cost inefficiency. In practice, this choice is made for reason of convenience. Details of our revision at this point are in line 61 to 65 (introduction), line 132 to 139 (theoretical framework), line 772 to 780 (conclusion).  

2. Additionally, no specification tests are undertaken for the production frontier, Cobb-Douglas is extremely narrow in scope. What about translog, generalized quadratic or some other flexible functional form. 

We have employed generalized likelihood ratio test for coefficients of the inefficiency effects model in line 264 to 277 (Data analysis and model specification), for Gamma in line 517 to 524 (Stochastic frontier production model). The test statistics are significant in favor of alternative hypothesis.

Concerning heteroscedasticity of the noise term, Koenker test was also used in line 239 to 242 

I wonder if selection is an issue between SC and non-SC coffee farmers. No real discussion of this is presented. See papers by Greene (2010) and Kumbakar, Tsionas and Sipilainen (2009) for more background. 

We leave this concerns regarding heterogeneity bias in the limitation of the study as the number of possible differences among individual farmers is infinite (line 783 to 786). Also, another limitation of the study concerns partial effects in line 781. This remarks from you is really helpful for our direction to further research. 

Last but not least, we also thank you for the useful references.

Author Response File: Author Response.pdf

Reviewer 3 Report

The is a good empirical paper.  Data source, estimation strategy, and estimation results are clearly explained. 

I only have one substantial and two minor comments. 

1. The result of the Stochastic Frontier Estimation (SFE) depends on a model assumption of the Technical Inefficiency Function (TIF). It will be desirable to show that the results are generally in agreement regardless of different TIFs. A few examples are truncated normal (R.Stevenson,

1980), exponential (Meeusen and van Den Broecke, 1977), gamma (Greene, 1990), and Rayleigh models (Hajargasht, 2014). If the half-normal model (the author’s choice) is only a suitable one, authors might need to explain why that is that case. 

 2. Data section (2.3) needs to be expanded and show which variables are to be analyzed in the regression. Also, the authors need to explain “data filtering” in more detail. 

3. Tables need to be self-contained with all necessary descriptions in the note. 

Typo

maximum livelihood -> maximum likelihood 

Author Response

Dear Professor,

Thank you very much for reviewing this research paper.  We highly appreciate your concerns in regard of methodological approach. Your detailed remarks are important that help to improve the quality of the empirical research.

To the best of our knowledge, the following are details of our revision.

1. The result of the Stochastic Frontier Estimation (SFE) depends on a model assumption of the Technical Inefficiency Function (TIF). It will be desirable to show that the results are generally in agreement regardless of different TIFs. A few examples are truncated normal (R.Stevenson,

1980), exponential (Meeusen and van Den Broecke, 1977), gamma (Greene, 1990), and Rayleigh models (Hajargasht, 2014). If the half-normal model (the author’s choice) is only a suitable one, authors might need to explain why that is that case. 

We have provided the reason of choosing generalized truncated normal distribution assumption of inefficiency effects model due to its computational simplicity as well as its ability to examine the effects of various specific variables of smallholder farmers, as opposed to 2 stage approach. Although, there is no priori basis of choosing one distribution over another. Different specifications lead to different estimates of cost inefficiency. In practice, this choice is made for reason of convenience. Details of our revision at this point are in line 61 to 65 (introduction), line 132 to 139 (theoretical framework), line 772 to 780 (conclusion).  We also thank you for the useful references.

 2. Data section (2.3) needs to be expanded and show which variables are to be analyzed in the regression. Also, the authors need to explain “data filtering” in more detail. 

We added variables used in the estimation in the section 2.3 in 190 to 197. 

8 cases were excluded from data filtration (ANOVA and cross tabulation) due to missing values (line 187 to 189)

3. Tables need to be self-contained with all necessary descriptions in the note. 

We have provided descriptions of SC and non-SC at the note of all tables. Details of variables and measurement unit are described in section 2.4 (line 248 to 252), (line 257 to 261)

Typo

maximum livelihood -> maximum likelihood  (I must have been foolish sometimes)

In addition, we have some other revisions. 

Concerning heteroscedasticity of the noise term, Koenker test was also used in line 239 to 242.

We have employed generalized likelihood ratio test for coefficients of the inefficiency effects model in line 264 to 277 (Data analysis and model specification)

We also have concerns regarding heterogeneity bias in the limitation of the study as the number of possible differences among individual farmers is infinite (line 783 to 786). Also, another limitation of the study concerns partial effects in line 781. 

Last but not least, thank you very much!

Author Response File: Author Response.pdf

Reviewer 4 Report

The paper should be published without major changes. The paper is well written, is clear and gives robust results which support the conclusions. The quality of writing is good, with clearness and good format of presentation. The scientific content is accurate, balanced, and interesting.

This paper deals with an interesting and crucial topic about sustainability and efficiency of agriculture. It tries to examine certified sustainable coffee farming versus conventional coffee farming. It considers an interesting question to answer about the positive impacts of sustainable farming versus the issue of technical efficiency. This paper gives a core contribution to the reader to think about the spectrum of sustainability and also to realize the economic efficiency aspect for small scale coffee farmers.

In this paper, the authors used a stochastic frontier and cost-benefit analysis, to investigate the economic impacts of sustainable and conventional coffee farming on farmers’ welfare. The description of the method is used sufficiently.

Moreover, this study made key findings. It has a fruitful results section.  Therefore, it will be valuable to rural development policymakers.  The significant value of this work and the importance of its results are essential for the development of strategies.

Please, I would like to comment on one point. Please, could you in Tables not to use the abbreviations for SC, Non-SC, TVC, TVC/ha, R, R/ha, it is not clear. Moreover, could you use an estimation of VND to euro or dollar?

Author Response

Dear Professor,

Thank you very much for reviewing this research paper.  We highly appreciate your remarks. They are important that help to improve the quality of the empirical research.

To the best of our knowledge, the following are details of our revision.

The paper should be published without major changes. The paper is well written, is clear and gives robust results which support the conclusions. The quality of writing is good, with clearness and good format of presentation. The scientific content is accurate, balanced, and interesting.

This paper deals with an interesting and crucial topic about sustainability and efficiency of agriculture. It tries to examine certified sustainable coffee farming versus conventional coffee farming. It considers an interesting question to answer about the positive impacts of sustainable farming versus the issue of technical efficiency. This paper gives a core contribution to the reader to think about the spectrum of sustainability and also to realize the economic efficiency aspect for small scale coffee farmers.

In this paper, the authors used a stochastic frontier and cost-benefit analysis, to investigate the economic impacts of sustainable and conventional coffee farming on farmers’ welfare. The description of the method is used sufficiently.

Moreover, this study made key findings. It has a fruitful results section.  Therefore, it will be valuable to rural development policymakers.  The significant value of this work and the importance of its results are essential for the development of strategies.

Please, I would like to comment on one point. Please, could you in Tables not to use the abbreviations for SC, Non-SC, TVC, TVC/ha, R, R/ha, it is not clear. Moreover, could you use an estimation of VND to euro or dollar?

We have provided details of these variables in Table 7. Production cost and rate of return line 577.

We first thought to use estimation in euro or dollar. However, this might lead to different estimation results and bias estimates. Can we provide rate of exchange instead? The average exchange rate in 2017 were 22.650 thousand VNDs/US dollar and 27.110 thousand VNDs/Euro.

Last but not least, thank you very much!

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Many typos and missing references (relative to what is cited in paper). I am fine with deferring selection to future research, but it should be explicitly mentioned in the paper. Note that Parmeter and Kumbhakar (2014) discuss the different versions of selection. Not sure I buy the response on why the normal-truncated normal is kept and only the mean is modeled as a function of z. This still needs more attention in my opinion. And the argument of a two stage approach is wrong regardless of which distribution you select. 

I would like to see much more attention paid to the use of the BC model with inefficiency effects through the mean only. Writing needs to be cleaned up, citations fixed and better caveat regarding potential selection and the likely impacts this will have on the interpretation of the results. 

Author Response

Dear Professor, 

Thank you very much for your time and supporting remarks. I hope for this time the revision satisfies your request.  

Many typos and missing references (relative to what is cited in paper). I am fine with deferring selection to future research, but it should be explicitly mentioned in the paper. Note that Parmeter and Kumbhakar (2014) discuss the different versions of selection. Not sure I buy the response on why the normal-truncated normal is kept and only the mean is modeled as a function of z. This still needs more attention in my opinion. And the argument of a two stage approach is wrong regardless of which distribution you select. 

I have added the selection issues in lines 252 to 256 as below.

However, sample selection might be an issue because there exist unobservable variables that are correlated with both W_i and V_i, which would lead to bias estimates of stochastic production frontier. According to (Greene 2010), the choice of technology (if based on some aspect of inefficiency) is influenced by correlation between W_i and V_i (random error), then the different form of sample selection issues arise.  

The revisions on the argument of a two stage approach are:

According to (Coelli 1995), there is a significant problem in two-step procedure. In this regard, the assumptions of inefficiency effects used in either stage are violated as they are assumed to be independently and identically distributed in the first stage, while in the second stage inefficiency effects are assumed to be a function of a number of farm-specific factors which implies that they are not identically distributed. This inconsistency problem and specification of stochastic frontier models were also in research papers of (Kumbhakar, Ghosh, and McGuckin 1991). Therefore, the application of one-step approach proposed by Battese and Coelli (1995) allows exogenous variables to incorporate directly into the efficiency error component (Huang 1994; Reifschneider and Stevenson 1991).

Also, argument of two stage approach regarding the limitation of the study in the conclusion was removed and replaced by the selection issues. I have already rechecked the typos throughout the paper. Detected typos from Word may be due to many specific Vietnamese words and also some Latin words of pests and diseases of coffee plant.

I would like to see much more attention paid to the use of the BC model with inefficiency effects through the mean only. Writing needs to be cleaned up, citations fixed and better caveat regarding potential selection and the likely impacts this will have on the interpretation of the results. 

More discussion on the use of BC model I have added in line 161 to 179 

One aspect of using truncated normal is the implications it presents regarding inefficiency for the industry as a whole. The truncated normal density has mode at 0 only when μ ≤ 0, but otherwise has model at μ (Parmeter and Kumbhakar 2014). As the truncated normal distribution reduces to the half-normal distribution on U_i when μ = 0, the estimation of  provides the average level of technical inefficiency in the sample (unconditional mean of U_i). However, if interest lies in the level of inefficiency for a given farmer, knowledge of  is not enough because it does not contain any individual farmer-specific information. It is worth noting that the truncated normal distribution depends on two parameters μ and that affords more flexibility in the shape of the density (Parmeter and Kumbhakar 2014). Hence, the identification of under-performing farmer or farmers using best practices in production demands for further investigation. In addition, assuming production and distribution shocks, estimation results do not entirely represent inefficiency. The solution for these concerns is the JLMS efficiency estimator that was first proposed by (Jondrow et al. 1982). The method is to estimate U_i from the expected value of U_i conditional on the composed error of the model ε_i ≡ V_i – U_i. The composed error contains individual farmer-specific information leading to the observation-specific value of the inefficiency from conditional expectation. Since the conditional density function (Jondrow et al. 1982) of U_i given ε_i is known, we can derive observation-specific of the efficiency index (Battese and Coelli 1988). The efficiency index ranges between 0 and 1, with a value of 1 indicating a fully efficient farmer.

Again, thank you very much for your useful remarks

Best regards,  

Author Response File: Author Response.pdf