Agricultural Informatization and Technical Efficiency in Maize Production in Zambia
Round 1
Reviewer 1 Report
The article has improved and added explanations provide useful information. It is also more relevant to talk about association instead of causal effects when it is difficult to control selection bias.
Using GR in the second stage of the estimation is not a valid method in SFA although otherwise it is can be used in the regression type analysis in general.
I would suggest the authors to apply the procedure described e.g.,in
Boris E. Bravo-Ureta · William Greene ·Daniel SolÃs. 2012. Technical efficiency analysis correcting for biases from observed and unobserved variables: an application to a natural resource management project. Empir Econ 43:55–72.
The first stage takes into account the unobserved heterogeneinity. If PSM does not work in the sample, the authors could state that the observable differences in farm size and age may exaggerate the actual TE effect of mobile phones.
The technical inefficiency part of the equation should be written in a bit different form.
The inefficiency term u follows (usually) a positive truncated normal distribution with a constant scale parameter sigma u^2 and a location parameter μ that depends on additional explanatory variables:
u ~ N+(μ, sigma u^2) with μ = δ z, where δ is an additional parameter (vector) to be estimated.
The interpretation of gamma is not exactly correct. As the sigma u^2 is not equal to the variance of the inefficiency term u (this can be shown), the estimated parameter cannot be interpreted as the proportion of the total variance that is due to ineffciency.
Author Response
Dear Reviewer,
We once again thank you specially for taking your time to review our submission. Your comments are very constructive and useful. We have done our very best and we hope our submission will be deemed as satisfactory. Below is our response to your comments;
Comment 1: The article has improved and added explanations provide useful information. It is also more relevant to talk about association instead of causal effects when it is difficult to control selection bias.
Response 1: Thank you so much for your validation of the improvement in our previous submission. It was extremely encouraging to see those kind and candid remarks from you. It helped us to refocus, reflect and refuel during the course of the revision.
Comment 2: Using GR in the second stage of the estimation is not a valid method in SFA although otherwise it is can be used in the regression type analysis in general.
Response 2: Use of GR in the estimation has been dropped. Thank you for the clarification.
Comment 3: I would suggest the authors to apply the procedure described e.g.,in Boris E. Bravo-Ureta · William Greene ·Daniel SolÃs. 2012. Technical efficiency analysis correcting for biases from observed and unobserved variables: an application to a natural resource management project. Empir Econ 43:55–72.
Response 3: We appreciate you substantially for your suggestion. We have used both the conventional approach and the sample-selection method for more robust implications and lessons. The quality of our paper has significantly improved. Thank you profoundly
Comment 4: The first stage takes into account the unobserved heterogeneity. If PSM does not work in the sample, the authors could state that the observable differences in farm size and age may exaggerate the actual TE effect of mobile phones.
Response 4: Since the proposed procedure starts by implementing PSM, this has been solved for one part where sample-selection model was used. However, your comment is very useful for the other part where conventional approach was used. As guided, we have stated (emphasized) this starting from the abstract.
Comment 5: The technical inefficiency part of the equation should be written in a bit different form. The inefficiency term u follows (usually) a positive truncated normal distribution with a constant scale parameter sigma u^2 and a location parameter μ that depends on additional explanatory variables: u ~ N+(μ, sigma u^2) with μ = δ z, where δ is an additional parameter (vector) to be estimated.
Response 5: This has been done as suggested. Thank you so much for your resolve to improve the quality of our paper
Comment 6: The interpretation of gamma is not exactly correct. As the sigma u^2 is not equal to the variance of the inefficiency term u (this can be shown), the estimated parameter cannot be interpreted as the proportion of the total variance that is due to ineffciency.
Response 6: We appreciate your careful observation in an attempt to improve the quality of our paper. We have revised the interpretation of gamma as follows:
The value is widely used as an indicator to measure the influence of the inefficiency component in the overall variance, in which case close to 1 implies deviations from the frontier dominates total variance, and = 0 denotes no technical inefficiency. Bear in mind that gamma is not the contribution of inefficiency in total variance, because the variance of the truncated normal random variable is .
Thank you so dearly for the constructive and highly valuable review.
Author Response File: Author Response.pdf
Reviewer 2 Report
no additional comments
Author Response
Dear Reviewer,
Thank you very much for your positive evaluation on our manuscript. We have employed MDPI english editing services to polish our writing again. Attached please find the certificate for english editing. We hope the edited manuscript meet your expectation.
All the best.
Round 2
Reviewer 1 Report
The authors have taken into account my earlier suggestion very well. The PSM provides a good match eliminating the largest observed differences between users and non-users of mobile phones. The PSM eliminates observed heterogeneity and the sample selection SFA accounts for unobserved heterogeneity. These procedures are technically properly done but there are a couple of problems which still need to be solved.
In production functions (user cases), the coefficient of labor input is very significantly negative which is not theoretically consistent. We expect that elasticities are positive or insignificant but large negative and significant values are questionable. Is there any explanation for that? The elasticities are also otherwise quite different in the two groups. This suggests that their production technologies are different in groups of users and non-users.
It is also unexpected that none of the TE effect model’s coefficients are significant in the group of non-users when they are significant in pooled sample and to some extent also in users’ subsample.
Additionally, the coefficients of gamma are very close to one in the unobserved heterogeneity models indicating that all or almost all deviations from the frontier are inefficiency and there is no noise at all. This is also to some extent questionable result.
Taking into account the points above, my suggestion is the following:
You discuss theoretically the procedure how to solve the problem at hand. The explanation about unobserved heterogeneity could probably be shorter since the second stage seems not to be applicable in the current case. Then you perform the PSM. I think it would be better to present only the pooled results of all observations and pooled results of PSM observations and compare the efficiency distributions of pooled and PSM pooled for all and for users and non-users. In both cases, we assume that the frontier is the same in each analysis for all observations. This procedure could probably avoid the negative elasticities of inputs.
Author Response
Dear Reviewer 1, Special thanks for the extensive and insightful comments. It is our wish that our revision is satisfactory. All revisions are marked using red color in the manuscript. Below is our response: COMMENT 1: The authors have taken into account my earlier suggestion very well. The PSM provides a good match eliminating the largest observed differences between users and non-users of mobile phones. The PSM eliminates observed heterogeneity and the sample selection SFA accounts for unobserved heterogeneity. These procedures are technically properly done but there are a couple of problems which still need to be solved. RESPONSE 1: We are very delighted to note that you found our revision properly done technically. Your validation and acknowledgment is gratifying COMMENT 2: In production functions (user cases), the coefficient of labor input is very significantly negative which is not theoretically consistent. We expect that elasticities are positive or insignificant but large negative and significant values are questionable. Is there any explanation for that? The elasticities are also otherwise quite different in the two groups. This suggests that their production technologies are different in groups of users and non-users. RESPONSE 2: We agree considering large significant negative value violates the positive marginal product. Thank you for your careful observation. Action has been done based on comment 5. COMMENT 3: It is also unexpected that none of the TE effect model’s coefficients are significant in the group of non-users when they are significant in pooled sample and to some extent also in users’ subsample. RESPONSE 3: Noted. Thank you. We have revised our empirical strategy. COMMENT 4: Additionally, the coefficients of gamma are very close to one in the unobserved heterogeneity models indicating that all or almost all deviations from the frontier are inefficiency and there is no noise at all. This is also to some extent questionable result. RESPONSE 4: Noted. Thank you. This has been revised. After pooling data together, the gamma is more reasonable. COMMENT 5: Taking into account the points above, my suggestion is the following: You discuss theoretically the procedure how to solve the problem at hand. The explanation about unobserved heterogeneity could probably be shorter since the second stage seems not to be applicable in the current case. Then you perform the PSM. I think it would be better to present only the pooled results of all observations and pooled results of PSM observations and compare the efficiency distributions of pooled and PSM pooled for all and for users and non-users. In both cases, we assume that the frontier is the same in each analysis for all observations. This procedure could probably avoid the negative elasticities of inputs. RESPONSE 5: Thank you so much for your suggestion. Our revision is based on this extremely useful suggestion. We sincerely hope our submission will be deemed as necessary and sufficient.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
The introduction could be more compact. General discussion about production could be shortened.
The objectives of the study should be clarified. For example, the significance of the output differences is not actually tested. Therefore, it is not possible to conclude if the difference is significant (in statistical sense).
It is not clear why the probit models are introduced when the selectivity bias is no more addressed. The authors could apply e.g., PSM matching for finding a comparable group for MP users and then apply efficiency model as is done in the article.al
Technical efficiency effect model is not presented in the method section. Only standard cross sectional model is discussed.
The data set is not properly described. At least averages and distributions of variables should be presented. The number of users and non-users of MPs are not presented either.
TE effect model is not testing causal effects, only associations. This concerns also other determinants of efficiency in addition to the use of mobile phones.
Reviewer 2 Report
Thank you for taking the comments to your paper.
The paper has been improved in many parts.
One minor comment is as follows.
SSA line 33 <-> Sub-Saharan Africa (SSA) line 50
Reviewer 3 Report
see attached
Comments for author File: Comments.pdf