Factors Influencing the Adoption of Agricultural Practices in Ghana’s Forest-Fringe Communities
Round 1
Reviewer 1 Report
The manuscript is well-written ad addresses specifically the aim the authors want to achieve.
To me, everything is OK.
I take the liberty to suggest that authors improve the discussion considering a newly published paper that could be worth as it tackles a complementary aspect of Ghana's forest-fringed rural communities.
Amadu et al. (2020). Assessing the value of forest resources to rural households: A case of forest-fringe communities in the Northern Region of Ghana. Environmental Development, art. No. 100577. https://doi.org/10.1016/j.envdev.2020.100577
Author Response
Dear reviewer, thank you for your time invested in reviewing this manuscript. We appreciate your suggestion which has added to the quality of the manuscript. Below is our response to your comment. Thank you once again.
Comment 1: I take the liberty to suggest that authors improve the discussion considering a newly published paper that could be worth as it tackles a complementary aspect of Ghana's forest-fringed rural communities.
Amadu et al. (2020). Assessing the value of forest resources to rural households: A case of forest-fringe communities in the Northern Region of Ghana. Environmental Development, art. No. 100577. https://doi.org/10.1016/j.envdev.2020.100577
Response 1: Thank you for your suggestion. I have improved the discussion with the finding from the suggested paper that indicates that compensations/motivations may encourage farmers who adopt forest-friendly inputs not to encroach adjoining forest for more land for production. Lines 353-357 present this as follows,
“……This can be more achievable through motivations given to the farmers by the Forest Services Division of Ghana for not encroaching the forest. Evidence shows that farmers in forest-fringe Ghana are willing to forgo exploitation of forest resources if they are compensated [54]. Our suggestion is not in a form of compensation but motivational packages for practicing forest-friendly agriculture.”
Reviewer 2 Report
In conclusion chapter, specify what complimentary agricultural practices you consider a priority for farmers. Referencing policy aspects in this section would also be useful, especially because Ghana is such a particular case, basically, what options would be available to farmers?
Author Response
Dear reviewer, thank you for your time invested in reviewing this manuscript. We appreciate your suggestion which has added to the quality of the manuscript. Below is our response to your comment. Thank you once again.
Comment 1: In conclusion chapter, specify what complimentary agricultural practices you consider a priority for farmers. Referencing policy aspects in this section would also be useful, especially because Ghana is such a particular case, basically, what options would be available to farmers?
Response 1: Thank you for your relevant comment which has improved the quality of the paper. The complimentary agricultural practices we consider a priority for farmers have been stated. This has been backed by Ghana’s policies which have been stressing on these practices for two decades now. The addition to the paper based on your comment is in lines 579-587 as follows,
“The complimentary agricultural practices under study have been part of Ghana government’s agricultural modernization policies since 1997 [76, 80-83]. The adoption of improved seeds and planting materials for crops survival and the application of inorganic fertilizers for increased outputs in particular have been pursued by the government of Ghana for decades. The promotion of these practices together with other improved agricultural technologies have mainly been through extension services delivery although not highly effective due to resource scarcity and other challenges. Promoting the adoption of the studied complementary agricultural practices is therefore in line with Ghana’s agenda of transforming its agricultural sector in a sustainable manner.”
Reviewer 3 Report
1. Introduction
The introduction reads very well. The research objective, knowledge gap and contribution are clearly indicated.
2. Adoption of agricultural practices: a review
The review of existing studies on Ghana is scanty in this section. I believe there are more recent papers on adoption of agricultural practices. Hence, I want to suggest that the authors provide more empirical evidence on Ghana.
3. Materials and methods
Study area and sampling technique are well described. However, I have some concerns with the choice of model. First, since the dependent variable - number of practices adopted at time is censored, that there are about 103 zero observation. For this reason, using linear multiple regression tends to yield biased estimates. instead the authors can use Tobit regression model. Looking at the nature of the objective research objective, it would be interesting to look at double hurdle model which simultaneously analyses the adoption decision and intensity of adoption. This model considers the general adoption process. Alternatively, the authors can use multivariate probit model to jointly evaluate the farmers adoption of individual farm practices and further analyses the correlation among the practices. The author may agree with me that the adoption of individual farms practices may be influenced by different set of factors. After running the multivariate probit model, you can look at the intensity of participation using Tobit regression model. Based on this recommendation, the entire results related to the econometric model will change entirely. Another suggestion is that the authors to look at their data set and explore more factors. For example, gender, access to credit, farm experience, household size, off-employment or incomes from off-farm activities, membership of farmer association, land tenure system, farming practices, etc. All these suggested factors have been proven to be some of the important determinants of adoption of farm practices.
If the authors have data on farm output or incomes, it would be valuable to explore the effect of number of farm practice adopted on farm output and or income.
4. Results
Based on the suggestion from above, the results require major revision. Hence, I will not comment deeply on the content of the discussion. Another suggestion is that the authors should create a table of descriptive statistics (means, standard deviation) for all the variables included the models for adopters and non-adopters. Use t-test to compare means between the two groups.
Tables A2 - A10 can be deleted. They appear to be redundant.
Author Response
Dear Reviewer, we express our thankfulness to you for your time invested in reviewing this paper, especially taking your time to make those significant comments and suggestions. Below are our responses to your comments and suggestions and we hope that you will be satisfied with our responses. Thank you very much for contributing to the improved quality of the manuscript.
Comment 1: The review of existing studies on Ghana is scanty in this section. I believe there are more recent papers on adoption of agricultural practices. Hence, I want to suggest that the authors provide more empirical evidence on Ghana.
Response 1: More recent empirical evidence on Ghana concerning adoption and intensity of adoption of agricultural practices have been added to section 2. The revised paragraph is in lines 75-96 as follows.
“In Ghana, researchers have documented the extent of farmers’ adoption of various agricultural practices and their effects on soil fertility, agricultural productivity, food security and household incomes [12, 18, 25-31]. Some of these studies have demonstrated that the adoption of agricultural practices is influenced by farm size, the effectiveness and frequencies of agricultural extension services, farmer education, input availability, and distance to sources of inputs [14, 16, 25, 29]. For instance, Kotu et al. [25] observed that most farmers are unwilling to travel long distances to purchase agricultural inputs, e.g., chemical fertilizers; hence, they continue with slash-and-burn farming. Others even do not have access to such inputs and as a result do not adopt innovative technologies [27, 28]. Issahaku and Abdulai [30] and Zakaria et al. [26] observed that education of household head, access to extension and weather information, and membership of farmer-based organizations influence farmers’ likelihood of adopting climate-smart agricultural practices. Zakaria et al [26] further stated that the intensity of adoption of climate-smart practices depends on farmers’ participation in capacity building programs, family labour, and access to agricultural insurance. Ehiakpor et al. [29] identified that the intensity of adoption of sustainable practices is influenced by farmers’ access to agricultural credit, participation in field demonstrations, and farm size. Similarly, [31] found that the intensity of adoption of sustainable soybean production technologies in northern Ghana is determined by age, education, extension visits, mass media, and perception of adoption. Generally, farm characteristics, socio-economic, and institutional factors determine the agricultural practices a farmer adopts, while the intensity of adoption in a given place also reflects a farmer’s motivation and capacity [32]”
Comment 2: Study area and sampling technique are well described. However, I have some concerns with the choice of model. First, since the dependent variable - number of practices adopted at time is censored, that there are about 103 zero observation. For this reason, using linear multiple regression tends to yield biased estimates. instead the authors can use Tobit regression model. Looking at the nature of the objective research objective, it would be interesting to look at double hurdle model which simultaneously analyses the adoption decision and intensity of adoption. This model considers the general adoption process. Alternatively, the authors can use multivariate probit model to jointly evaluate the farmers adoption of individual farm practices and further analyses the correlation among the practices. The author may agree with me that the adoption of individual farms practices may be influenced by different set of factors. After running the multivariate probit model, you can look at the intensity of participation using Tobit regression model. Based on this recommendation, the entire results related to the econometric model will change entirely.
Response 2: Thank you very much for your insightful comments and suggestions for improving the quality of the paper. We initially did not include a model on adoption since we analysed the section on adoption descriptively. However, based on your comments, we have carried out a multinomial logistic regression on the influence of the factors on adoption of the practices. We did not use the Tobit or multivariate probit model because we are familiar with logistic regression model and research shows that logistic regression produces results that are similar to that of multivariate probit model. The data analysis section in the methodology has been revised to include the use of logistic regression. This revision is from line 227-232 as follows,
“We used multinomial logistic regression (MNR) to assess the effects of the socio-economic, institutional and farm factors (Table 1) on farmers’ adoption of any of the aforementioned agricultural practices. The Chi-square likelihood ratio test statistic (c2 = 575.88, df = 102, r<0.0001, n = 288) and the goodness-of-fit test statistic (Pearson: c2 = 511.18, df = 1620, r = 1.000; Deviance: c2 = 333.01, df = 1620, r = 1.000) indicate that the data are appropriate for the multinomial logistic model.”
The logistic regression results is tabulated in Table 3 from line 327-330 and its interpretation is just before the table, from line 318-326 as follows.
“The multinomial logistic regression results show that seven variables have significant effects on the adoption of the complementary practices with four having the most influence (Table 3). We observed that farmers that cultivate more than one plot (c2(6) = 85.46, r<0.0001), have more household members to help on farms (c2(6) = 79.41, r<0.0001), have access to extension services (c2(6) = 89.25, r<0.0001), and perceive agricultural inputs to be more beneficial (c2(6) = 67.87, r<0.0001), are more likely to adopt some practices to enhance their farming business. Gender, land tenure, and farming system also influence a farmer’s likelihood of adopting agricultural-enhancing practice. These factors however have lower significance according to our model.”
The addition of the logistic model to the results did not change the results section of the manuscript. It is rather an improved extension of the adoption section (section 4.1). The last sentence of section 4.1 (lines 342-344) which reads “According to our model, distance to sources of inputs, educational level, ages of farmers, and hired labour availability have no effects on the likelihood of a farmer to adopt any agricultural practice identified in our survey.” and the sentence under section 4.2 (lines 3358-361) which reads “While our survey found that seven factors influence farmers’ adoption of complementary practices, four factors (Table 4) significantly influence the number of practices a farmer adopts at a time (F(7, 280) = 58.968, r < 0.001, R2 = 0.596, N = 288)” indicate the connection between the logistic regression and the multiple linear regression.
We used multiple linear regression for the intensity of adoption because the dependent variable (number of complementary practices adopted) is a continuous variable. All the explanatory variables are also continuous except access to extension services and perception for adoption which are dummy variable and multiple linear regression takes dummy variables as well. Although 103 zero counts (for non-adoption) may influence the results towards adoption, yet we could not have included the dummy variable perception for adoption without the value of non-adoption. Inasmuch as there may be bias in using multiple linear regression, the results significantly support the predictions and realities about factors of adoption and non-adoption of agricultural practices. Secondly, a paper is soon coming out which analyses the factors of adoption of each of the practices as well as the social, economic, and environmental costs and benefits of each of the practices. Adopting another model to evaluate each of the practices will therefore duplicate the results in our new paper. In the conclusion section, lines 588-599 have been added to reflect this limitation of our research and the areas of further research that we are carrying out. This conclusion paragraph reads as follows,
“We acknowledge that our research is limited to the assessment of the factors of adoption treating all the identified practices as a combined component and that we failed to investigate the extent to which the factors affect each of the identified practices. We chose this method to provide a broader overview of factors of adoption and non-adoption in order to make generalized recommendations for agricultural development and forest sustainability. Since the adoption of complementary agricultural practices around forest reserves is a critical and contested issue, further research is required to investigate the factors affecting the adoption of each of the practices, the social, economic, and environmental costs and benefits associated with each of the practices, and the effects of adoption on outputs and income of the farmers. These will provide a more concrete rationale for the adoption of forest-friendly agricultural practices within forest-fringe communities while not compromising the conservation of the forest reserves.”
Comment 3: Another suggestion is that the authors to look at their data set and explore more factors. For example, gender, access to credit, farm experience, household size, off-employment or incomes from off-farm activities, membership of farmer association, land tenure system, farming practices, etc. All these suggested factors have been proven to be some of the important determinants of adoption of farm practices.
Response 3: Land tenure system, farming system, and gender are part of the variables for the multinomial logistic regression. Some other variables were initially explored but were insignificant. For instance, household size duplicated with household labour and was also insignificant and had multi-collinearity issues. Non-farm employment was insignificant and was negatively affecting the significance level of the other variables. Thank you.
Comment 4: If the authors have data on farm output or incomes, it would be valuable to explore the effect of number of farm practice adopted on farm output and or income.
Response 4: We have data on farm output but it is very complex and measurement is inconsistent from one farmer to another and even at the market level. As a result we could not include it in this paper. Data on income was not available. Thank you
Comment 5: Another suggestion is that the authors should create a table of descriptive statistics (means, standard deviation) for all the variables included the models for adopters and non-adopters. Use t-test to compare means between the two groups.
Response 5: The descriptive statistics of all the variables of the study have been presented in Table 1 with their associated tests of normality. In addition to the tests of normality, Q-Q normal probability plots have been generated in Figure A1 on line 637 all of which show the approximate normal distribution of the variables. We have revised the data analysis section of the methodology to include these tests. This is from lines 240-243 as follows,
“……Table 1 displays the normality tests using skewness and kurtosis and the Q-Q normal probability plots (Figure A1) indicate the approximate normality of the quantitative variables which are all in line with the acceptable standards for normality [49, 50].”……
We did not use the t-test to compare the means between adopters and non-adopters with the reason being that our focus was on adopters and all the quantitative variables were directed towards adopters. The variables for non-adopters were qualitative and were used for descriptive analysis. Thank you.
Comment 6: Tables A2 - A10 can be deleted. They appear to be redundant.
Response 6: Tables A2 to A10 have been deleted with the exception of Table A7 which is now Table A2. We did not delete this table because we only wrote a simple summary about it in the text. Therefore interested readers can refer to it for more details. The others appeared to be redundant as you noticed. Thanks for this comment.
Reviewer 4 Report
This paper presents an interesting topic, both from a scholarly perspective and in support of policies aimed at increasing the sustainability of farming practices in agricultural-dependent communities. The scope of the paper is regional, focusing on the performance and outcomes of the farming practices of forest-fringe communities in Ghana, but it may be of interest to the wider public, taking into account the global challenge of tackling the pressure of securing environmentally friendly agricultural production.
Regarding the study area of Ashanti region, it would be useful for the reader to have a perspective in terms of similarities and differences between the communities selected from the 10 reserves that influence the adoption of agricultural practices. For instance, are some of these communities located closer to urban areas or main city markets/local district market which influence the use of agricultural input? Can some socio-economic characteristics of these communities picture a broader context for understanding the predominance of certain agricultural practices?
Concerning the Results section, as data on incomes were difficult to obtain from the farmers, it would be interesting to know how could these data gaps be addressed in the future and how it could improve the analysis. It would be relevant to include some points on the advantages and the shortcomings/uncertainties of this research, considering the methods at hand. The discussions could include a point on the role of farmer-based organizations in advancing the use of certain agricultural practices.
Author Response
Dear Reviewer, we express our thankfulness to you for your time invested in reviewing this paper, especially taking your time to make those significant comments and suggestions. Below are our responses to your comments and suggestions and we hope that you will be satisfied with our responses. Thank you very much for contributing to the improved quality of the manuscript.
Comment 1: Regarding the study area of Ashanti region, it would be useful for the reader to have a perspective in terms of similarities and differences between the communities selected from the 10 reserves that influence the adoption of agricultural practices. For instance, are some of these communities located closer to urban areas or main city markets/local district market which influence the use of agricultural input? Can some socio-economic characteristics of these communities picture a broader context for understanding the predominance of certain agricultural practices?
Response 1: Our study communities are similar in terms of their locations (distances to central markets where they sell outputs and buy inputs) and only different in terms of what others have that others do not have. Of more importance is that each community has a mix of adopters and non-adopters. Even in communities where they have agro-chemical shops, there are still non-adopters. Nevertheless, we have added a paragraph to the results section lines 287-298 to reflect your comment and indicate that our results is based on individual farmers and not from a community’s point of view. The additional paragraph reads as follows,
“The mean distance from the communities to the nearest central markets where the farmers sell their produce and purchase agricultural inputs is 11 kilometers. Twelve of the 20 study communities have agro-chemical shops from which inhabitants can purchase agricultural inputs. Almost all (18) of the communities have information centers that relay various information, including that on agriculture, to their members. Eleven of the 20 communities had extension service visits at least before our survey. It is however worth mentioning that according to our study, adoption or non-adoption of complementary agricultural practices is not based on a community’s nearness to central markets, availability of agro-chemical shops and information centers in community, operation of periodic markets in community, or any other characteristic of a community. This is because each community had a mix of adopters and non-adopters based on the farmers’ perceptions and some other probable factors which are elaborated in the succeeding sections”
Comment 2: Concerning the Results section, as data on incomes were difficult to obtain from the farmers, it would be interesting to know how could these data gaps be addressed in the future and how it could improve the analysis. It would be relevant to include some points on the advantages and the shortcomings/uncertainties of this research, considering the methods at hand.
Response 2: Thank you for this insightful comment. We have added an additional paragraph to the conclusion from line 588-599 to address this comment which stresses on the need for further research. The additional paragraph is as follows,
“We acknowledge that our research is limited to the assessment of the factors of adoption treating all the identified practices as a combined component and that we failed to investigate the extent to which the factors affect each of the identified practices. We chose this method to provide a broader overview of factors of adoption and non-adoption in order to make generalized recommendations for agricultural development and forest sustainability. Since the adoption of complementary agricultural practices around forest reserves is a critical and contested issue, further research is required to investigate the factors affecting the adoption of each of the practices, the social, economic, and environmental costs and benefits associated with each of the practices, and the effects of adoption on outputs and income of the farmers. These will provide a more concrete rationale for the adoption of forest-friendly agricultural practices within forest-fringe communities while not compromising the conservation of the forest reserves.”
Comment 3: The discussions could include a point on the role of farmer-based organizations in advancing the use of certain agricultural practices.
Response 3: We have added a point on the role of FBOs not only in accessing agricultural inputs but also, marketing farm produce and securing financial credits. The revised section of the discussion is from line 487-493 as follows,
“Forming farmer-based organizations will also be an effective means to access agricultural inputs in bulk for distribution among the members. This will reduce the retail and transportation costs incurred by individual farmers to access agricultural inputs. Farmer-based organizations are not only helpful in procuring bulk inputs, they also aid in securing good prices for agricultural produce marketed by members and accessing agricultural financial credits. Farmer-based organizations are known to play important roles in the farming operations of members [64-68]”
Reviewer 5 Report
The manuscript is dealing with very important issue and bring interesting results that are worth to be published. However, I have some major concerns on the writing of the paper.
- Introduction is too long. I must be shortened to make clear the main topic of the manuscript with clear stated aim of present study.
Information regarding Ghana, agriculture, degradation, etc. in Introduction should be moved into section 2. perhaps labelled as Conceptual background
- Adoption of agricultural practices: a review
This part is not prepared well. Please make first subchapter dealing with agriculture – degradation – Ghana and then prepare proper literature review including hypotheses to be tested – it is not clear why the chosen independent variables were measured and others were not.
- Material and Methods and 4. Results
It seems that the sample is OK and is well descripted. I recommend to move “data collection“ from part 3.3. to 3.2 a to make 3.3 only for data analyses.
I think that there should be reported the distribution of number of adopter among number of practices adopted – it seems that normal distribution among six categories (From 1 adopted practice to 6 adopted practises) for 188 observations is not possible. Please report the result of Shapiro-Wilks test or other test used. The authors should present the graphs of the interactions between the dependent variable and the others to tell the findings directly.
The decision tree model of classification and regression trees should be also done. Because its results are relatively easy and straightforward for interpretation. Split the sample of respondents into two groups – training data set and data set for testing. Calculate the decision tree model and then apply on the testing data set with the model based on the results of the training data set.
I think (based on my field experience among farmers in CEE countries) that there is important not only the number of adopted practices, but also the structure of adopted practices, so multivariate analyses should follow after the multiple linear regression – according the data the DCA with following CCA (with forward selection of variables) may be appropriate (or nMDS with permutation test to test the relationship between nMDS axes and independent variables).
- Discussion
The part 5.2 is extremely interesting. However, it seems, that it is (mostly) not discussion for the results. This should be shortened and some theoretical parts moved into Conceptual background.
Author Response
Dear Reviewer, we express our thankfulness to you for your time invested in reviewing this paper, especially taking your time to make those significant comments and suggestions. Below are our responses to your comments and suggestions and we hope that you will be satisfied with our responses. Thank you very much for contributing to the improved quality of the manuscript.
Comment 1: Introduction is too long. I must be shortened to make clear the main topic of the manuscript with clear stated aim of present study. Information regarding Ghana, agriculture, degradation, etc. in Introduction should be moved into section 2. perhaps labelled as Conceptual background
Response 1: The introduction has been shortened. Sections relating to agriculture and forest degradation have been moved to section 2. Section 2 has now been revised in lines 75-125 under the heading “Adoption of agricultural practices: a conceptual review” to reflect this. Thank you.
Comment 2: This part is not prepared well. Please make first subchapter dealing with agriculture – degradation – Ghana and then prepare proper literature review including hypotheses to be tested – it is not clear why the chosen independent variables were measured and others were not.
Response 2: Section 2 has been revised. The first paragraph now deals with issues relating to agricultural practices in Ghana. The second paragraph extends this argument to capture the practices adopted in other African countries. This serves as an extension of the literature review beyond Ghana. The third paragraph builds on the research gap which argues that none of these research was carried out at forest frontiers where the application of some agricultural practices may conflict with forest and biodiversity conservation. This paragraph concludes with a re-iteration of the research objective. This revised section is in lines 75-139 as follows.
“In Ghana, researchers have documented the extent of farmers’ adoption of various agricultural practices and their effects on soil fertility, agricultural productivity, food security and household incomes [12, 18, 25-31]. Some of these studies have demonstrated that the adoption of agricultural practices is influenced by farm size, the effectiveness and frequencies of agricultural extension services, farmer education, input availability, and distance to sources of inputs [14, 16, 25, 29]. For instance, Kotu et al. [25] observed that most farmers are unwilling to travel long distances to purchase agricultural inputs, e.g., chemical fertilizers; hence, they continue with slash-and-burn farming. Others even do not have access to such inputs and as a result do not adopt innovative technologies [27, 28]. Issahaku and Abdulai [30] and Zakaria et al. [26] observed that education of household head, access to extension and weather information, and membership of farmer-based organizations influence farmers’ likelihood of adopting climate-smart agricultural practices. Zakaria et al [26] further stated that the intensity of adoption of climate-smart practices depends on farmers’ participation in capacity building programs, family labour, and access to agricultural insurance. Ehiakpor et al. [29] identified that the intensity of adoption of sustainable practices is influenced by farmers’ access to agricultural credit, participation in field demonstrations, and farm size. Similarly, [31] found that the intensity of adoption of sustainable soybean production technologies in northern Ghana is determined by age, education, extension visits, mass media, and perception of adoption. Generally, farm characteristics, socio-economic, and institutional factors determine the agricultural practices a farmer adopts, while the intensity of adoption in a given place also reflects a farmer’s motivation and capacity [32].
It is not only in Ghana that farmers adopt and intensify their adoption of new or existing practices for various reasons. For instance, smallholder farmers in Zambia and Kenya who own their farmlands tend to practice agroforestry and mixed cropping to sustain production, while conversely farmers with insecure land tenure tend to use chemical fertilizers to sustain production [33, 34]. Faße and Grote [35] observed that experienced farmers in Tanzania employ crop diversification and agroforestry more than inexperienced farmers because the latter have little knowledge about farming techniques. Kassie et al. [36] found that households with short-lease land tenure adopt legume intercropping and chemical fertilization to increase short-term productivity, with the intensity of their adoption correlating with farm size, distance to farms, and availability of household labour. According to Kassie et al. [36], household size positively influences farmers' use of manure since collecting and transporting manure to farms is labour intensive .
However, the issue surrounding the Ghanaian studies reviewed above is that none of them considered the locations of farms to determine whether the practices adopted could conflict with the surrounding landscapes and the possible resolutions that could be offered. Research sites were generally arable lands used for subsistence and commercial farming, not designated as forest reserves, and where sometimes there are almost no forest. Farmers in forest-fringe communities in Ghana however cultivate within and around forest reserves that are officially protected [37-39]. Evidence shows that these farmers often rely on the forests as land banks for agricultural production when their existing farmlands become infertile [37, 40].
According to Acheampong et al. [17], agricultural expansion between 1986 and 2015 caused 78% of the deforestation in the forest reserves of the Ashanti region. The underlying factors were that, first, before the demarcation of the areas as forest reserves, human settlements and farms already existed within the forests [39]. The Forestry Commission of Ghana allowed the settlers and their farms to remain in the reserves with their boundaries delineated to prevent further encroachment into the reserves. Population growth and weak enforcement of forest protection laws led to the expansion of the farms into the reserves. Second, a majority of the inhabitants interplant their food crops with tree crops for cash and depend on natural soil fertility to increase output. According to these farmers, when the tree crops form a canopy, they encroach more of the forest in the search for fertile land to cultivate their food crops [17]. Since the tree crops are the main source of income for the farmers, after about two years of cultivating the newly cleared land, they will interplant their food crops with tree crops.
This continuous conversion of protected forestlands to agriculture reduces forest cover and biodiversity, limits the provision of ecosystem services, and contributes to climate change [17, 41-43]. The adoption of certain high-yielding agricultural practices by farmers in forest-fringe communities may enhance agricultural sustainability and forest conservation. This is possible via a presumed ‘land-sparing’ effect whereby higher yields on existing plots diminish the need to convert surrounding forests [44]. We explore the factors influencing the adoption and intensity of adoption of the agricultural practices listed above at forest frontiers of Ghana and offer possible recommendations for agriculture and forest sustainability.”
Comment 3: I recommend to move “data collection“ from part 3.3. to 3.2 a to make 3.3 only for data analyses.
Response 3: Data collection techniques under 3.3 has been moved to 3.2 to make 3.3 only for data analyses. Section 3.2 has now been titled “Sample size selection and data collection. Thank you.
Comment 4: I think that there should be reported the distribution of number of adopter among number of practices adopted – it seems that normal distribution among six categories (From 1 adopted practice to 6 adopted practises) for 188 observations is not possible.
Response 4: Data on the distribution of the adopters among the number of practices adopted has been reported from line 352-354 as follows,
“The farmers that adopt one, two, three, four, and five practices at a time represent 43.6%, 33.0%, 13.3%, 8.0%, and 2.1%, respectively of the 188 adopters. No farmer adopts all the six practices.”
Comment 5: Please report the result of Shapiro-Wilks test or other test used. The authors should present the graphs of the interactions between the dependent variable and the others to tell the findings directly.
Response 5: We have reported the results of skewness and kurtosis tests of normality in Table 1 and presented the Q-Q probability plots in Figure A1 on line 637 for the variables all of which show the approximate normality of the quantitative variables. We have revised the data analysis section of the methodology to include these tests. This is from lines 240-243 as follows,
“……Table 1 displays the normality tests using skewness and kurtosis and the Q-Q normal probability plots (Figure A1) indicate the approximate normality of the quantitative variables which are all in line with the acceptable standards for normality [49, 50].”……
Comment 6: The decision tree model of classification and regression trees should be also done. Because its results are relatively easy and straightforward for interpretation. Split the sample of respondents into two groups – training data set and data set for testing. Calculate the decision tree model and then apply on the testing data set with the model based on the results of the training data set.
Response 6: Thank you for this suggestion. Unfortunately we have no idea how this is done. Since this suggestion does not mean that our results is limited but rather a suggestion that will be an addition to our results, we humbly request that we decline this suggestion because we do not know how to produce them. Thank you.
Comment 7: I think (based on my field experience among farmers in CEE countries) that there is important not only the number of adopted practices, but also the structure of adopted practices, so multivariate analyses should follow after the multiple linear regression – according the data the DCA with following CCA (with forward selection of variables) may be appropriate (or nMDS with permutation test to test the relationship between nMDS axes and independent variables).
Response 7: Thank you for your suggestion. A paper is soon coming out which analyses the factors of adoption of each of the practices as well as the social, economic, and environmental costs and benefits of each of the practices. Carrying out multivariate analysis following the linear regression will therefore duplicate the results in our new paper. We however value this your insightful suggestion which will improve our forthcoming paper, and we have added a paragraph in the conclusion section from lines 588-599 which reflects on this area of further research. The paragraph is as follows,
“We acknowledge that our research is limited to the assessment of the factors of adoption treating all the identified practices as a combined component and that we failed to investigate the extent to which the factors affect each of the identified practices. We chose this method to provide a broader overview of factors of adoption and non-adoption in order to make generalized recommendations for agricultural development and forest sustainability. Since the adoption of complementary agricultural practices around forest reserves is a critical and contested issue, further research is required to investigate the factors affecting the adoption of each of the practices, the social, economic, and environmental costs and benefits associated with each of the practices, and the effects of adoption on outputs and income of the farmers. These will provide a more concrete rationale for the adoption of forest-friendly agricultural practices within forest-fringe communities while not compromising the conservation of the forest reserves.”
Comment 8: The part 5.2 is extremely interesting. However, it seems, that it is (mostly) not discussion for the results. This should be shortened and some theoretical parts moved into Conceptual background.
Response 8: The first paragraph of section 5.2 which is more of a conceptual background has been moved to section 2 to strengthen the rationale for the study. This is found from lines 118-130 as follows,
“According to Acheampong et al. [17], agricultural expansion between 1986 and 2015 caused 78% of the deforestation in the forest reserves of the Ashanti region. The underlying factors were that, first, before the demarcation of the areas as forest reserves, human settlements and farms already existed within the forests [39]. The Forestry Commission of Ghana allowed the settlers and their farms to remain in the reserves with their boundaries delineated to prevent further encroachment into the reserves. Population growth and weak enforcement of forest protection laws led to the expansion of the farms into the reserves. Second, a majority of the inhabitants interplant their food crops with tree crops for cash and depend on natural soil fertility to increase output. According to these farmers, when the tree crops form a canopy, they encroach more of the forest in the search for fertile land to cultivate their food crops [17]. Since the tree crops are the main source of income for the farmers, after about two years of cultivating the newly cleared land, they will interplant their food crops with tree crops.”
Other sentences that are also not so relevant have been removed to shorten the section 5.2. Thank you for this suggestion.
Round 2
Reviewer 3 Report
Compared to the previous version, the revised version has improved. Im responding based on the comments number provided by the authors. The authors have satisfactorily comments 1, 3, 4, 6. However, I am not convinced with the responses to comments 2 and 5 which are major concerns.
With response 2 on the choice of the model, I think it is not convincing for the authors to indicate that they are not familiar with Tobit or multivariate probit models although they are the appropriate models. Hence, they applied those ones they are familiar but are inappropriate in the context of their study. I think I provided adequate explanation on why the authors should use multivariate probit and tobit model. Multinomial logit model may not be appropriate for the study context. The reason is that farmers adopt multiple agricultural practices, hence, the options are not mutually exclusive. You use multinomial logit when the choices are mutually exclusive. Hence, multivariate probit model is suitable when you farmers use multiple practices. An interesting thing about the multivariate probit model is that it allows the research to explore correlation among the practices. Furthermore, using multiple linear regression is still inappropriate because the data is censored. For this reason, I still hold that the authors should explore Tobit model. The results presented in Table 3 appears to be a binary logistic model and not multinomial logitic as claimed by the authors.
Response 5: The reason provided by the authors are not convincing enough. Readers would want to know the characteristics of the adopters and non-adopters. In many empirical studies on adoption, this is the convention. Hence, I encourage to authors to do that. In the statistics, I also encourage them to report only mean, standard deviations, mean difference and t-values.
General comment: I would encourage the authors to critically consider these comments.
Author Response
Dear Reviewer,
We thank you for your further comments and suggestions to further improve the quality of the manuscript. Please below are the responses we have provided to your comments. We admit that addressing these comments have further sharpened the manuscript. Thank you.
Comment 1: With response 2 on the choice of the model, I think it is not convincing for the authors to indicate that they are not familiar with Tobit or multivariate probit models although they are the appropriate models. Hence, they applied those ones they are familiar but are inappropriate in the context of their study. I think I provided adequate explanation on why the authors should use multivariate probit and tobit model. Multinomial logit model may not be appropriate for the study context. The reason is that farmers adopt multiple agricultural practices, hence, the options are not mutually exclusive. You use multinomial logit when the choices are mutually exclusive. Hence, multivariate probit model is suitable when you farmers use multiple practices. An interesting thing about the multivariate probit model is that it allows the research to explore correlation among the practices. Furthermore, using multiple linear regression is still inappropriate because the data is censored. For this reason, I still hold that the authors should explore Tobit model. The results presented in Table 3 appears to be a binary logistic model and not multinomial logitic as claimed by the authors.
Response 1: We have adopted a multivariate model, specifically, canonical correlation analysis to evaluate the relationships between the two dependent variables (adoption and intensity of adoption) and the factors of adoption. We admit that you were right. Multivariate model limits the probability of committing a type 1 error. Some of the adoption variables that we considered significant in our previous regression model were actually found to be not significant after running the multivariate model on the dependent variable set and independent variable set. We have revised the data analysis section of the methods to reflect the new model we have used. Lines 213-232 present this revision as follows,
“The data on farming practices the farmers adopt and the perceptions for adoption and non-adoption were first descriptively analysed and related to the farming operations of the respondents. We used the multivariate technique, canonical correlation analysis (CCA), to assess the relationship between adoption and intensity of adoption of agricultural practices and a set of adoption factors, namely, age, education, household size, household labour, hired labour, access to extension services, number of farm plots, land tenure, distance to sources of input, and perception for adoption. We adopted this technique to limit the probability of a type 1 error [1] by performing one statistical test on the same predictors for the two dependent variables (adoption and intensity of adoption) instead of running separate univariate models. We relied on Wilks’ test of significance to assess the significance of the full model and the proportion of the variance explained by the variable sets [2]. We then tested the hierarchical arrangements of the canonical covariates for statistical significance through the dimension reduction analysis. This was done to determine whether only the first canonical covariate or both are worthy of interpretation. We adopted a cut-off correlation of .30 to determine the variables that contribute significantly to the relationship between the adoption variables and the factors of adoption [3]. We checked the data for normality, linearity, and absence of multi-collinearity for the purpose of the multivariate analysis, specifically the correlation analysis. The data were generally normally distributed and when they were not they were log transformed to approximate normality [48]. Multi-collinearity was not problematic [51].”
The analysis section has also been reivsed to reflect the new results. Lines 300-328 and 355-410 reflect these changes as follows,
300-328 “We conducted canonical correlation analysis to evaluate the multivariate-shared relationship between social, institutional and farm factors and adoption and intensity of adoption of agricultural practices. The analysis produced two canonical covariates based on the two dependent variables (i.e. adoption and intensity of adoption) with squared canonical correlations (Rc2) of .716 and .094 for canonical variates 1 and 2, respectively (Table 3). The test statistics of the multivariate model adopting Wilks’ Lambda criterion (l = .257, F(20,552.00) = 26.791, r < .001) indicate that the full model is statistically significant and that the model explains 74.3% (1-l) of the variance shared between the two sets of variables. The dimension reduction analysis for canonical variates 1 to 2 (F(20,552.00) = 26.791, r < .001) and 2 to 2 (F(9,277.00) = 3.198, r < .005) indicate that both functions are statistically significant (Table A2). However, given the Rc2 effect of each function (71.6% and 9.4% of shared variance for canonical variates 1 to 2 and 2 to 2, respectively), the first canonical variate is more noteworthy of interpretation although the second function is still significant for interpretation.
The coefficients and proportions of variance explained in the first pair of canonical variates show that both adoption and intensity of adoption of complementary agricultural practices correlate with the canonical variate (Table 3). The first pair of canonical variates indicate that farmers that have lesser number of farm plots (-.31), have to travel long distances to purchase inputs (-.71), and that have negative perceptions about complementary agricultural practices (-.98) do not adopt any complementary agricultural practices (.855). On the other hand, possession of more farm plots (-.31), short distances to sources of inputs (-.71), and positive perception about complementary agricultural practices (-.98) influence adopters to increase the number of practices they adopt (-.899). The second pair of the canonical variates indicates that adopters (.519) increase the number of practices they adopt (.438) when they have access to agricultural extension services (.713) and cultivate more than one plot (.32) but do not adopt or decrease intensity of adoption as they age (-.422) and also with long distance to sources of inputs (-.356). The farmers that adopt one, two, three, four, and five practices at a time represent 43.6%, 33.0%, 13.3%, 8.0%, and 2.1%, respectively of the 188 adopters. No farmer adopts all the six practices.”
355-410 “As aforementioned, the first canonical variate demonstrates that positive perceptions (-.98) emanating from the need to improve yield and control weeds, pests, and diseases (Table 4) are the main motivational factors that increase the number of practices a farmer adopts (-.90). According to the farmers, application of these practices makes their farming operations easier. Some adopters stated that adopting more of the practices results in harvesting more outputs on a relatively smaller plot compared to farming with no complementary inputs. However, a comparative analysis between the adopters and non-adopters revealed that the adopters farm on larger plots. For instance, while 54.8% of the adopters have their total farm sizes larger than 2 ha and the rest with plots sizes smaller than 2 ha, 40.8% of the non-adopters have same. The adopters however harvest almost three times the outputs of the non-adopters. Data collected from maize growers for instance indicated that while adopters of complementary agricultural practices harvest averagely 19 bags of maize per hectare, non-adopters harvest averagely eight bags.
Both canonical variates show that multiple-plot farmers (-.31 and .32 for functions 1 and 2, respectively) tend to intensify their agriculture through adopting more practices at a time (-.90 and .44 for functions 1 and 2, respectively). Cereals (maize, rice) farmers constitute 40.4% of the adopters followed by tree crops growers (28.2%) and tubers cultivators (18.1%). According to some of the multi-plot farmers, they are able to control weeds, pests and diseases and increase crops yields at the same time using complementary agricultural practices. Cross-referencing the number of farm plots cultivated with the main practice adopted indicated fertilizer application dominating all the categories of farmers except those who cultivate three plots where herbicides usage is the main practice followed by legume crop rotation (Table A3).
According to the second canonical variate, access to agricultural extension services has a strong positive correlation (.71) with both adoption (.52) and number of practices adopted (.44). While 64.6% of the farmers adopt at least one of the above practices, only 35.6% had access to extension services at the time of the survey. These adopting farmers could not explain why extension agents generally do not visit their communities. The same farmers stated that they instead rely on their own knowledge and that of other adopters to apply the agricultural practices they have adopted.
While perceptions for adoption, number of farm plots cultivated, and access to agricultural extension services increase the intensity of adoption, two other factors are on the contrary – age of a farmer, and distance to sources of agricultural inputs. The second canonical variate demonstrates that the age of a farmer (-.42) negatively correlate with adoption and number of practices a farmer adopts. Further enquiry into the ages of the adopters revealed that 35.6% are over 50 years old. Out of this, 52.2% adopt only one practice while 28.4% adopt two practices at a time. Ideally, these farmers (>51 years) should be using more agricultural inputs to boost productivity because age might reduce their physical capacity. Yet our results indicate that the more farmers age, the lesser the number of complementary practices they adopt.
Agricultural inputs play vital roles in farming. However, the distance a farmer has to travel to purchase these inputs (-.36) reduces adoption and for the farmers adopting, the intensity of their usage. We identify that distance to sources of inputs (-.71) in the first canonical covariate positively correlates with the number of practices adopted (-.90). This interpretation is however questionable considering the small standardized coefficient of distance to input (.06) perhaps resulting from a multi-collinearity issue. Nevertheless interpretation is still valid for the second canonical covariate. The majority of the adopters travel less than 20 km to purchase agricultural inputs. For instance, all the farmers that adopt five practices purchase their inputs within 5 km of their residence. A third (66.7%) of those that adopt four practices, 72% of those that adopt three practices, 67.7% of adopters of two practices, and 70.7% of one practice adopters travel within 20 km to purchase their inputs. This implies that shorter distance to sources of inputs correlates positively with increased adoption of agricultural practices. Eight of the fringe communities have access to agrochemical shops that supply inputs to the farmers. According to the adopters in these communities, purchasing inputs from shops in their communities is easier and preferable to shops outside of their communities.”
Other sections of the discussion that referred to the results have also been updated to reflect the new results.
Comment 2: Response 5: The reason provided by the authors are not convincing enough. Readers would want to know the characteristics of the adopters and non-adopters. In many empirical studies on adoption, this is the convention. Hence, I encourage to authors to do that. In the statistics, I also encourage them to report only mean, standard deviations, mean difference and t-values.
Response 2: We have calculated the independent sample T test and have recorded the mean, standard deviation, mean difference and t-values of the quantitative variables for both adopters and non-adopters. This has been added to Table 1 on lines 197-199. Thank you.
Comment 3: General comment: I would encourage the authors to critically consider these comments.
Response 3: We have critically considered the above comments and acknowledged that they have made significant improvement to the quality of the paper.
Reviewer 5 Report
Thank you for new version of manuscript and the changes made. As have written in my first review this is interesting topic that deserves to be published.
Here are some minor remarks:
– as the distribution of dependent variable is not normal (as I have expected) and multiple non-linear regression was used instead of multiple linear regression, the Q-Q plots are not needed,
- decisions trees are great and were used in various studies such as e.g. here http://www.preslia.cz/P194Pinke.pdf I like this procedure very much – it is easy to follow and in graphical presentation it is really clear interpretation of results of the regression, that is often confusing for many readers
- it is great that other paper based on multivariate techniques is being prepared,
- CAUTION – you have not renumbered tables after preparing of new table 3
- CAUTION – models reported in section 4.2 are wrong – they are from former multiple linear regression – you must replace them with new results of multiple non-linear regression
- CAUTION – table 4 have to be deleted – new statistics are calculated in new table 3
Author Response
Dear Reviewer,
Thank you for your time for the minor remarks made. We have attended to all of them and below are our responses to your remarks. Thank you so much for contributing to the quality of this paper.
Remark 1: – as the distribution of dependent variable is not normal (as I have expected) and multiple non-linear regression was used instead of multiple linear regression, the Q-Q plots are not needed,
Response 1: We have removed the Q-Q plots from the appendix. Thank you.
Remark 2: - decisions trees are great and were used in various studies such as e.g. here http://www.preslia.cz/P194Pinke.pdf I like this procedure very much – it is easy to follow and in graphical presentation it is really clear interpretation of results of the regression, that is often confusing for many readers
Response 2: Thank you for providing this source to us. We will surely explore its usefulness in our forthcoming paper. Thank you.
Remark 3: - it is great that other paper based on multivariate techniques is being prepared,
Response: Thank you.
Remark 4:- CAUTION – you have not renumbered tables after preparing of new table 3
Response 4: The tables have been renumbered now. Thank you.
Remark 4:- CAUTION – models reported in section 4.2 are wrong – they are from former multiple linear regression – you must replace them with new results of multiple non-linear regression
- CAUTION – table 4 have to be deleted – new statistics are calculated in new table 3
Response 4: We have replaced the regression model with multivariate model utilizing canonical correlation analysis. We have revised the data analysis section of the methods to reflect the new model we have used. Lines 213-232 present this revision. The results section from lines 300-328 and 355-410 have been revised to reflect these changes. Table 4 has been deleted and new table 3 of results has been reported on lines 328-333.
Round 3
Reviewer 3 Report
The current version has improved. All concerns have been addressed.