o :2“ _ u'“ L ‘n h ‘10.. | }. “'11?”ka “I“? A! “1“"? .» "VI. “(3%” i ,'I"S'«'h ”'93:; W 'I l ‘I.‘ :‘ T -. Jr". " ! \'~‘..‘."..f:.:., '.;.-.”" |":'I 2.9"“ fill": ' fi‘xaéfl, ~ a. g ' ' ' . :Lp'”. :IH-U m3 ..." .!.I:h" :H. l w,“ " .. . v 0 "'t' -_:. J‘ ‘_-' .. "7 -;-“.*;."z'u:' 'n 'rwp» Jig?‘vfiiji.'y,"fixwl.'.}..:'§:‘, ‘ ‘ . . l {"1 .‘ZE'K " :' ’1: iv. -!, W . ' ll.:' WI?) h“, '01:“ "'0'“ |}. it; A}: . I" 0 ' ' ID”, ' ‘ " I. o, P '1. ' 1| _:f ' "UHF” V“?!- ‘ ’ | I'..' g! [I 6_ s. ' ‘ n, ' l ' H '_?.' ,1 ’1: . l 1‘! \g'; "I" - lg'E' -- ,1... 2'23"? (,K'VM . 3.0 : . ‘ . . I‘ll" l " '61. ,’|' ‘ H '3 a: | ". fail“ (Ru? "1 I‘EI'! 1‘ full“ 0...! "" ol * I oh“ ' A ll .3“ Mt“ M H..:xl;.ifl':{rtl ' '3‘ ' ”9’33 1|: {Ugh-1 lll'l‘ W) Refit- M” . I“? U '3; v r - ., .zs‘H" " "Fifi? ~ 5'47" 3";"1‘ " ' " .i' we :4“. .. - fl ‘ I“: 2“: U " SI 51.. \l'. i I, :VWV .lr'il' 5“! )' 'IJI‘ ”1... "ml ’25\ “{0}"; 31‘." l!‘ o g l ' ‘ o 0' ‘ V g : ‘0‘ ;.'.'a'- M'Hu’ . NM” '.'.| I' I Wu? 0 M if“: u ’1' . . 1' "" 'K‘"--- " ' ”9' .. .I II' 1 » . ' ‘ ' . Ll; D)”, l.' I .l'h‘l.“-|131\u§wfilk'l.‘| v ' 1.9+!) ‘; ‘(j 3.11M ‘ H 3‘ s: ' . : :" I'Wf' "' 'v‘ it 1‘ III W'.’ .1. I. » '1. "4""?! .I u' ..' I . ‘ E H I n .n I . I lirki . '3'..- fdhl‘ 111,3“) I; Q I .f'b‘l) 'l IQ'I’J l{HI H: 'IPLIL fig]? ‘lf‘flu‘. lll‘f‘f'fl3NJxm ‘ - ‘ .. . .' «.lhwq/A ., av” - it w' ’ r, In; , . 5"?“ LIL Michigan gum; University firmer: This is to certify that the thesis entitled FACTORS AFFECTING ENTERPRISE CHOICE: AN ANALYSIS OF TRADITIONAL FOOD PRODUCTION IN SOUTHEASTERN MINAS GERAIS, BRAZIL presented by Carlos da Silva has been accepted towards fulfillment of the requirements for _Eh._D._degree in WT Economics z.\fl Major professor Date February 10, 1981 0-7639 1 Tliiimilfliliifllmifliiiflifl 3129 OVERDUE FINES: 25¢ per day per item RETUMIM LIBRARY MATERIALS: \ _.. , Place in book return toremo “Ft-'73, ' arge from circulation records |\\ J (III-“fl“ 7k JL/ ciao A211 FACTORS AFFECTING ENTERPRISE CHOICE: AN ANALYSIS OF TRADITIONAL FOOD PRODUCTION IN SOUTHEASTERN MINAS GERAIS, BRAZIL BY Carlos Arthur B. da Silva A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Agricultural Economics 1981 ABSTRACT FACTORS AFFECTING ENTERPRISE CHOICE: AN ANALYSIS OF TRADITIONAL FOOD PRODUCTION IN SOUTHEASTERN MINAS CERAIS, BRAZIL By Carlos Arthur B. da Silva This research contributes to a better understanding of small scale farming in a depressed rural area in the state of Minas Gerais, Brazil. With a stagnant agriculture and widespread socio-economic problems, the Zona da Mata region of Minas Gerais is the current site of a compre- hensive development program--the PRODEMATA project. The study utilizes survey data from the project's monitoring and evaluation component to examine aspects related to farm enterprise Choice among producers in the region. Specific objectives are: (l) to provide a characterization of farming in the survey area; (2) to investigate the relationship between income and enterprise emphasis for farmers with similar levels of resource endowment; (3) to identify and study factors influencing the selection of alternative farm systems among area farmers; and (£0 to discuss the implications of the research for the design of plans and programs aimed at the improvement of farm incomes. A brief cross tabulation of the survey data revealed that farming in Zona da Mata maintains the traditional characteristics reported in earlier studies. Labor is by far the most intensive input used in the diverse number of activities performed, and crop yields were generally lower than state averages. Moreover, a low level of market orientation was observed, particularly for the smaller scale farmers. These producers 'It‘ ‘ I III. 5:! (1.7 Carlos Arthur B. da Silva were also at the lower end of a farm earnings distribution schedule, which depicted a high concentration of earnings among larger scale farms. Given that significant differences in income also exist within farm size strata, the hypothesis that such differences could be a result of dissimilarities in enterprise emphasis was examined. It was shown through potential net margin analysis that the hypothesis is accepted for producers in both the 10- 50 hectare and 50—100 hectare strata, but not for the 0-10 hectare group. The implication of the analysis is that there is potential to improve incomes of farmers with more than ten hectares via reorganization of farm plans such that the enterprise systems of high income producers are approximated. In view Of the severe limitations imposed by the land constraint, this is not true for the 0—10 hectare producers. Polychotomous logit models were estimated for the investigation of factors affecting producer participation in the important farm systems existing in the area. The models specify probabilities of participation into the alternative systems as functions of a number of variables hypothesized to be influential in the decision processes of interest; their parameter estimates would provide a basis for testing the signifi- cance of each individual variable. Further, they would allow the prediction of probabilities of participation in the observed farm types, given alternative scenarios. Results of the analysis indicate that there are differences among the three farm size strata in terms of both the nature and relevance Of factors affecting participation. Credit per hectare, and the farm location in one of the three sub-regions in which the sample area was Carlos Arthur 8. da Silva divided were factors which tended to be more influential among smaller scale producers. Conversely, risk attitudes and perceptions, and the size of the household unit were more Significant for larger scale farms. The level of labor/hectare was relevant in all three strata. Effects of Changes in the average levels of the significant variables suggested that there exists an overall responsiveness to market Oppor- tunities and policy incentives. The examined scenarios indicated that Changes in existing patterns of enterprise emphasis are likely to stem from policies such as the provision of more credit, or the reduction of yield and price variability normally associated with higher return enter- prises. Additionally, the effects of Changes in the average levels of labor/hectare reinforced earlier observations about the importance of off farm enterprises in the utilization of surplus family labor from the smaller scale farms. The research results are encouraging evidence of the potential Of policy actions aiming at adjustments in patterns of enterprise emphasis as components of broader rural development strategies. ACKNOWLEDGEMENTS I wish to express my sincere appreciation to my major professor, Dr. Darrell Fienup, for his skillful guidance throughout my graduate program, and for his constructive suggestions in the development of this study. Dr. Fienup's constant support was instrumental in making my stay at Michigan State University a most rewarding experience. I am also indebted to Dr. Michael Weber for his concern with my professional development, and for his insightful criticisms during all the stages of this research. The participation of Drs. Gerald Schwab, Thomas Pierson, John Allen and Anthony Koo on my Thesis and Guidance Committees is also acknowledged. I want to thank the Brazilian Ministry of Education and the Federal University of Vicosa (UFV), for providing the financial support for my graduate studies. In particular, I am indebted to Dr. Adao Pinheiro and the Faculty of UFV's Departamento de Tecnologia de Alimentos for the Opportunity to come to MSU. I am also grateful to Dr. Antonio Bandeira, Chairman of UFV's Departamento de Economia Rural for allowing me to use the survey data of the PRODEMATA project, and for the financing of my field work in Brazil. The assistance of Chris Wolf and Paul Wolberg with my programming and data management needs is appreciated. Thanks are also due to Dr. Peter Schmidt, who kindly made his logit programs available to this research . Special thanks are due to Mrs. Cathy Cooke and Mrs. Joy Walker who respectively typed the final version and drafts of the dissertation. Appreciation is extended to my friends and relatives in Brazil, particularly to Alexandre Aad Neto, who kept some of my personal affairs in order, and to my mother and brothers, whose contributions were too many to describe here. Finally, I would like to Sincerely thank my wife, Salete, for the critical prodding and loving support which made this long process more endurable . If TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES Chapter I. INTRODUCTION Problem Definition Research Objectives Research Organization ll. GENERAL CHARACTERISTICS OF THE STUDY AREA Introduction Geographic Setting Socio-Demographic Characteristics The Agricultural Economy of Zona da Mata III. THE INTEGRATED RURAL DEVELOPMENT PROGRAM FOR THE ZONA DA MATA REGION OF MINAS GERAIS STATE (PRODEMATA) Introduction Project Background Project Components and Costs Monitoring and Evaluation The Sampling Plan Some Preliminary Evaluation Results Summary IV. OVERALL CHARACTERISTICS OF THE SAMPLED FARMERS Introduction Farm Sizes and Allocation of Land Resources Crop Yields Levels of Input Use Market Orientation Farm Earnings Family Composition and Education of HouseholdMembers Summary iv Page vii xi mam—o momma: 21 21 21 22 25 26 30 31 33 33 34 40 42 45 ll9 53 5'4 Chapter V. ENTERPRISE EMPHASIS AND FARM INCOMES Introduction Related Research Farm Classes Appraised Measurement of Incomes Enterprise Returns Producer Emphasis on Alternative Enterprises The 0-10 Hectare Class The 10-50 Hectare Class The 50-100 Hectare Class Assessment of the Relationship Between Farm Incomes and Enterprise Emphasis Relative Emphasis by Income Classes: The Test of Differences Among Sample Proportions The Test of Independence of Classification The Indexes of Potential and Performance: Upton's Approach Summary Vl. SELECTION OF THE FARM SYSTEM: AN ANALYTICAL FRAMEWORK Introduction Smallholder Decision Making: Theory and Fact A Conceptual Model of Enterprise Choice For the Zona da Mata Farmer Enterprise Selection: A Quantitative Model Summary VII. LOGIT ANALYSIS OF ENTERPRISE CHOICE Introduction Defining the Relevant Enterprise Combinations Explanatory Variables Distance (DIST) Location (R2 and R3) Sensitivity Index (SI) Capital/Hectare (K/Ha) Labor/Hectare (L/Ha) Credit/Hectare (ClHa) Family Size Model Structure 0-10 Hectares and 10-50 Hectares 50-100 Hectares Model Estimation and Results 0-10 Hectares 10-50 Hectares 50-100 Hectares I? 57 57 58 60 61 65 69 73 75 77 80 82 88 91 98 101 101 102 107 116 129 .131 131 132 137 137 137 138 139 1110 llIO 141 “I1 141 1H2 143 1113 153 163 Chapter - Page VII. LOGlT ANALYSIS OF ENTERPRISE CHOICE (continued) Predicted Probabilities Under Alternative Scenarios 170 0-10 Hectares 170 10-50 Hectares 180 50—100 Hectares 186 Summary 190 VIII. SUMMARY AND CONCLUSIONS 192 The Problem and Research Objectives 192 Research Methodology and Empirical Findings 1911 Implications and Policy Issues 2011 Limitations and Suggestions for Further Research 212 APPENDIX A 215 APPENDIX B 220 APPENDIX C 221 BIBLIOGRAPHY 2le vi LIST OF TABLES Urban, Rural and Total Production Zona da Mata—-1950—1975 (1000 inhabitants) Land Use and Number of Farms by Group of Activity; Zona da Mata--1975 Total Output of Major Crops and Average Yields--1975 Land Ownership Zona da Mata--1975 Selected Sample Municipalities by Administrative Region and Corresponding IBGE Micro—Region Sample Size by Class of Farm Purged Files: 1976/1977 and 1977/1978 CrOp Years Average Size of Holding: 1977/1978 Crop Year Average Land Allocation Among Major Uses (Ha) Zona da Mata: 1977/1978 Crop Year Land Use and Farmers Participating in Major Activities Average Yields of Major Crops (Ton/Ha) Average. Levels of Input Use by Farm Class Average Percentages of Consumption and Sales Relative to Farm Production Total Value of Farm Ea‘rnings (Cr$) Average Family Composition and Age Distribution of Family Members Formal Education of Zona da Mata Farmers (Percent) vii Ii 15 16 18 28 29 32 30 35 37 III H3 '47 51 53 56 Table Page 5.1 Average Earnings, Variable Costs 64 and Net Farm (Cr$) 5.2 Average Returns to Labor and Land 66 5.3 Enterprise Emphasis 72 0-—10 Hectares 5.8 Enterprise Emphasis 76 10-50 Hectares 5.5 Enterprise Emphasis 78 50-100 Hectares 5.6 Average Farm Incomes by Size Class 81 and Income Groups (Cr$) 5. 7 Relative Emphasis on Alternative Enterprises 814 by Income Groups 0-10 Hectares 5.8 Relative Emphasis on Alternative Enterprises 85 by Income Groups 10—50 Hectares 5.9 Relative Emphasis on Alternative Enterprises 86 by Income Groups 50-100 Hectares 5.10 Classification of Farmers by Enterprises 89 Emphasized the Most 5.11 Average Indexes of Potential and Performance: 94 Land as a Constraint 7.1 Enterprise Combinations by Farm Size Classes 13¢! 7.2 Most Relevant Farm Types and Respective 137 Numbers of Sample Observations by Size Classes 7.3 Results of the Model Estimation Under the Initial 104 Specification 0-10 Hectares Theil's Normalization 7.11 Results of the Model Estimation Under The Initial 106 Specification 0-10 Hectares ANOVA Normalization viii Table 7.5 7.6 7.7 7.8 7.9 7.13 7.15 7.16 Results of the Model Estimation Under The Reduced Specification 0-10 Hectares Theil Normalization Results of the Model Estimation Under The Reduced Specification 0—10 Hectares ANOVA Normalization Results of the Initial Model Estimation 10-50 Hectares Theil's Normalization Results of the Initial Model Estimation 10-50 Hectares ANOVA Normalization Results of the Model Estimation Under the Reduced Specification 10-50 Hectares Theil Normalization Results of the Model Estimation Under The Reduced Specification 10—50 Hectares ANOVA Normalization Results of the Initial Model Estimation 50-100 Hectares Theil's Normalization Results of the Initial Model Estimation 50-100 Hectares ANOVA Normalization Results of the Model Estimation Under the Reduced Specification 50—100 Hectares Theil's Normalization Results of the Model Estimation Under The Reduced Specification 50-100 Hectares ANOVA Normalization Predicted Probabilities 0-10 Hectares Predicted Probabilities 10-50 Hectares ix Ii 148 199 151i 155 156 157 161i 165 166 167 171 181 Table 7.17 Predicted Probabilities A. 1 50-100 Hectares Average Levels of Input Utilization Sharecroppers Average Levels of Input Utilization 0-10 Hectares Average Levels of Input Utilization 10-50 Hectares Average Levels of Input Utilization 50-100 Hectares Average Levels of Input Utilization 100-200 Hectares Computation of the Indexes of Potential and Performance for a Sample Farm Enterprise Emphasis by Region Juiz de Fora Enterprise Emphasis by Region Muriae Enterprise Emphasis by Region Vicosa 216 217 218 219 220 221 222 223 LIST OF FIGURES Figure 2.1 Brazil, Minas Gerais State and the Zona da Mata Region 2.2 Lorenz Curve for Land Distribution Data, Zona da Mata, 1975 4.1 Distribution of Farm Earnings 6.1 Schematic Presentation of a Smallholder Decision Making Process 7.1 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Zona da Mata Farmers--0—10 Hectares 7.2 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Muriae Farmers-~0—10 Hectares 7.3 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Vicosa Farmers—-0—10 Hectares 7.11 Predicted Probabilities Over the Range of Sample Observations: Credit [Hectare Juiz de Fora Farmers—-0-1O Hectares 7.5 Predicted Probabilities Over the Range of Sample Observations: Credit [Hectare Zona da Mata Farmers--10-50 Hectares xi Page 19 52 111 175 176 177 178 185 31" ‘i 3):”? up... him a IU" "I Gel". CHAPTER I INTRODUCTION Past efforts on the promotion of rural development have often emphasized improvements in farm productivity. By accepting the contention that a more productive agriculture would result in increases in output, employment and incomes for the masses Of poor rural resi- dents, strategies based on technological change have become key components Of overall plans of economic development. Recently, there have been Charges that most of these actions have bypassed the neediest of the rural poor--the small farmer.1 Yet, the number of individuals worldwide who would conform to the usual definitions of "small farmer" is estimated to be as high as to encompass 60 percent of all farmers.2 The constraints faced by this vast number of peasants on their productive processes, as well as their socio-cultural traits are rather unique. As a consequence, attempts to improve their well being require explicit consideration of these constraints and characteristics. In 1David Norman: "The Farming Systems Approach: Relevancy for the Small Farmer;" MSU Rural Development Paper NO. 5, 1980, p. 2. 2Clifton Wharton, cited by John Dillon in "Broad Structural Review of the Small-Farmer Technology Problem;" in Economics 1'1" the Design of Small Farmer Technology; edited by A. Valdes, Grant Scobie and John Dfllon (Ames: Iowa UniverSIty Press, 1979). affine ‘iis re- 1'0” ‘ ‘ w letlg 2 fact, part of the problem with traditional approaches to foster growth in agricultural productivity resides on the inadequate understanding of the workings of the small farm sector. Awareness of the need for the consideration of small farmers as unique entities has fortunately increased in more recent times. In this respect, the role of research in meeting information needs for‘the design of strategies to help the rural poor has become all important. As knowledge increases about the complexities of the small farm sector, the prospects for the success of such strategies are certainly improved. This study represents an effort to enhance existing knowledge on the small farm sector of a lesser developed region of the Brazilian state of Minas Gerais. lts specific concern is the investigation of factors impinging on farm planning processes among small scale pro- ducers, since empirical evidence on this aspect is virtually nonexistent for the area. Problem Definition The Zona da Mata region of Minas Gerais state is typical of areas which were marginalized in the overall process of economic growth. Despite its relative proximity to the large urban centers of Southeastern Brazil, the region is disproportionately poor and backward, with a stagnant agricultural sector and a multitude of Chronic social problems. Adverse historical Circumstances are partially to blame for the current state of affairs, though erratic governmental intervention and inequalities on the control and ownership of productive resources have also played a role in this regard. M vii ‘. .6," 0“. x (I! ”v. Iv -3 3 To address the needs of the vast majority of small farmers who eke out a subsistence in the area, a joint effort of state development planners and World Bank officials is currently under way. The proposed remedial measures are expressed in a comprehensive rural development project--the PRODEMATA project. Its chief component is a large program to promote increased credit use, but also important are the strong socio-infrastruc— tural components which should benefit rural residents in general. An implicit hypothesis in the project's design is the belief that increased access to inputs through more availability of credit should provide important incentives fOr the improvement of small farm incomes. Yet, another relevant and interrelated aspect of farm income determina- tion is the planning of the farm system. Since it is known that there exists a large number of farm and non-farm activities performed in the region, it should be of interest to study the differences between the farm plans of lower and higher income producers to ascertain if such income variations can be attributed to the selected farm enterprise mix. If a relationship between selected farm systems and relative income levels is shown to exist, this becomes a logical area for policy intervention. This research utilizes information generated by the surveys of the PRODEMATA monitoring and evaluation component to investigate specific aspects of farm planning among Zona da Mata farmers. The major concern is the identification of factors influencing producers' decisions on the selection of alternative farm plans. It is argued that knowledge of these factors can be a valuable input for development planning in the region, particularly if Significant differences in incomes can be related to alter- native farm plans . 1h isar sale F gses I: It gak I; Research Objectives The overall objective of the study is to deveIOp a better understanding of planning processes related to the selection of enterprises by small scale producers in the area of concern, as an aid to the design of strate- gies to improve their income levels. The following specific objectives have been set to accomplish this goal: - To provide a Characterization of farming in the region for all classes of producers originally considered in the PRODEMATA sample survey. This Classification should provide initial insights into the nature of the problems being faced by area farmers. Moreover, it should be helpful as a means of validating common beliefs about agriculture in the region. - To investigate the relationship between income levels and enterprise selection for producers with similar levels of resource endowment. Emphasis will be placed on the identification of the major farm enterprises in the area, and on the assessment of their returns to fixed factors. Additionally, the analysis will identify income classes for small farmers in the sample and determine the degree of emphasis placed by these producers on alternative enterprises. - To identify and study factors influencing the selection of alternative farm plans among producers in the area. Econometric models relating the choice of a farm plan with a group of independent variables hypothesized to impinge on the selection process will be estimated and utilized to predict probabilities of participation into alternative farming systems, given selected policy alternatives. 5 Further, the framework will be used to examine the potential role of the PRODEMATA project in the promotion of changes on existing patterns of enterprise emphasis. - To discuss the implications Of the research for the design of plans and programs aiming at the improvement of farm incomes in the study reg ion . Research Organization The study is divided into eight Chapters. The introductory Chapter has presented the research problem and objectives. Chapter II presents a brief characterization of the study region, with emphasis given to a discussion of specific features of its agricultural sector. Chapter III summarizes the scape of the PRODEMATA project and presents the sampling plan which generated the data utilized in the analysis. Chapter IV is devoted to a descriptive presentation of characteristics of the sampled farmers. Aspects related to resource endowment and allocation, market orientation and socio-demographics are discussed. The analysis of the relationship between enterprise Choice and farm incomes is carried out in Chapter V. Chapter VI develops the analytical framework for the analysis of factors affecting enterprise choice, whereas the results of the Choice model estimation and the predicted probabilities of participation on selected farm systems are discussed in Chapter VII. Finally, Chapter VIII summarizes the major findings of the study, discusses its policy implications and evaluates its limitations. Areas for further research are also suggested. CHAPTER II GENERAL CHARACTERISTICS OF THE STUDY AREA Introduction This Chapter presents descriptive aspects pertaining to selected characteristics of the area under consideration in this study. The initial section deals with the geographic setting Of the study. The socio-demographic aspects are examined in the second part of the chapter. To conclude, the major Characteristics of the agricultural economy of the area are appraised, with particular reference to the identification of land use patterns and structural issues facing the farming sector. The characterization which follows draws basically from two sources: the "Economic Diagnostic of Zona da Mata," a comprehensive study of the area undertaken by the Agricultural Economics Department of the Federal University of Vicosa, and the "1975 Agricultural Census of Minas Gerais," published by the Brazilian Institute of Geography and Statistics (IBGE) . Further sources are referenced throughout the Chapter . Geographic Setting The Zona da Mata region of Minas Gerais is located in the southeastern part of the state, bordering the states of Rio de Janeiro and Espirito Santo (Figure 2.1) . The large urban centers of Belo Horizonte, Sao Paulo BRAZIL Mina s Gerais State MINAS GERAIS Iona da Mata Region Figure 2.1 Brazil, Minas Gerais State and the Zona da Mata Region 8 and Rio de Janeiro are situated respectively at about 150, 000 and 200 km from the area's central part. The boundaries of the study region were formerly defined so .that it would constitute one of 15 geographic zones of the state. This division has been modified over the years, but for the purposes of this study, theorlginal configuration will be used. As such, the study area encom— passes the following micro-regions defined by IBGE:1 - Micro-Region 188: Mata de Ponte Nova - Micro-Region 189: Vertente Ocidental do Caparao - Micro-Region 192: Mata de Vicosa - Micro-Region 193: Mata do Muriae - Micro-Region 196: Mata de Uba - Micro-Region 200: Juiz de Fora - Micro-Region 201: Mata de Cataguazes The region defined above has a total area of 36012 km which corresponds to 6.1 percent of the total area of the state. Its climate is generally mild, with average temperatures ranging from 20° to 22° C. Rainfall averages 1000 mm a year, with maximum precipitation occurring from October to March. The drier season takes place during the fall/winter months of April to September. The topography of the study area is characterized predominantly by rolling to hilly terrains. It is estimated that only 16 percent of the land can be regarded as flat, whereas 00 percent is rolling terrain and V 1Fundacao lnstituto Brasileiro de Geografia e Estatistica (IBGE); Censo Agropecuario de 1975--Minas Gerais (Rio de Janeiro, 1979) . 9 00 percent hilly lands. The soils are often poor and eroded due to inadequate management and intensive exploitation. Latosols prevails in the highlands, while alluvial soils are the characteristic of the lower lands, particularly the valley bottoms . Soda-Demographic Characteristics The total population of the study region in 1975 was 1,623, 713 inhabitants of which 01.0 percent lived in the rural areas, as shown in Table 2.1 below. The population density is roughly 05 persons per square kilometer. The growth rate of the total population is low when compared to Brazil as a whole. During the 1950-1975 period, the population of the Zona da Mata region grew at an average annual rate of 1.6 percent compared to 2.8 percent for the whole country. Even though a consistent decline in rural population is observed from the aggregate statistics of Table 2.1, it should be noted that within the study area the urban-rural population mix varies considerably. Particularly those "municipios" (municipalities) located at the northern part of the region still show much lower numbers of 'urban inhabitants, whereas the situation is reversed in the southeast. The diminishing number of rural residents, and the slow growth rate for the study region reflect the outmigration trend which has charac- terized the area in recent decades. Even though precise statistics on this issue are lacking, it is estimated that migration rates from the region are among the highest in the country. The reasons for this trend have been mostly linked to a saturated labor market in the agricultural sector 10 Table 2.1 Urban, Rural and Total Population Zona da Mata--1950—1975 (1000 inhabitants) Urban Rural Year Total Percent _ Total Percent Total 1950 385.1 30.0 898.2 70.0 1,283.3 1960 567.2 37.2 955.8 62.8 1,523.0 1970 795.6 09.7 805.2 50.3 1,600.8 1975 951.0 58.6* 672.7 01.0* 1,623.7 Source: Fundacao Instituto Brasileiro de Geografia e Estatistica (IBGE); Anuario Estatistica do Brasil, 1955, 1965, 1975 and 1978 issues. *Estimated percentage refers to Minas Gerais state as a whole. 11 and relatively more favorable urban wages. Furthermore, the poor social infrastructure has most likely an active role in influencing decisions to migrate. Deficiencies in the social infrastructure are substantial in both health and education. The health care in particular is very limited, and the rural poor are the most seriously affected by the lack of proper services. Mortality rates are high, especially among infant and preschool Children. Malnutrition and a high incidence of infectious and other communicable diseases are believed to be at the root of the problem. Furthermore, schistosomiasis is endemic throughout the region, and some rural areas have almost all their pOpuIationS affected by the disease.2 An examina- tion of the basic health statistics for five of the Six Micro-Regions which constitute the study region shows that about 25 percent of the municipal- ities do not have resident physicians. No hospital beds were available in 58 percent of the "municipios" examined. The usual statistics of physicians and hospital beds per thousand inhabitants are also low by most acceptable standards. The former was estimated at . 77, while the latter was found to be 5.3, Of which 3.2 were concentrated in the indus- trial municipality of Juiz de Fora (280,000 inhabitants in 1975) .3 Education opportunities are limited. The illiteracy rate among the 5—10 age group was about 55 percent in 1975. Although the absolute number of municipally sponsored elementary schools in the area is con- siderable, the quality of the teaching staff is generally poor. A large 2The World Bank; Brazil-Staff Project Report of the Integrated Rural Development Project in the State of Minas Gerais; Report No. 1291 BR Washington, 1976f. 3Fundacao Instituto Brasileiro de Geografia e Estatistica; Informacoes Basicas dos Municipios--Minas Gerais (Rio de Janeiro, 1978) . 12 number of school teachers in the area are not properly qualified to meet the legal requirements pertaining to the profession's regulations. The result of this flawed system is a labor force with poor educational back- ground, and according to the World Bank ". . . well over 60 percent of the agricultural workers have not had any formal education.” In summary, the Zona da Mata region of Minas Gerais state can easily be placed among the many "poverty pockets" which have developed in Brazil in the recent period. Contrasting with rapidly developing areas within the state, Zona da Mata is today a region with a stagnated economy. Some of the reasons and consequences for this depressed economic status are examined next . The Agricultural Economy of Zona da Mata Historically, the settlement of the area can be traced to the early years of the mining cycle in Minas Gerais (XVllIth century). The geographic location of the region, between the important mining centers of Minas Gerais and the regional metropolis of Rio de Janeiro, favored the establishment of small villages which played the role of intermediate supply points. An incipient agriculture began to develop in this period, mostly based on subsistence crOps, sugar cane, tobacco and coffee. Although the growing of sugarcane soon became an important commer- cial activity, it was not until the expansion of the coffee frontier from the southeast that real incentives were present for the rapid development of agriculture. Coffee found in Zona da Mata favorable soil and climate conditions as well as privileged location in relation to major markets. “The World Bank, Op. Cit., p. 10. EYE.” ‘ I 'ué': 'M' iv. "SC nib. Fm- u ‘ v LC: in 13 With the end of the mining cycle, this export oriented agriculture transformed the region into one of the leading economic bases of Minas Gerais. Coffee revenues in particular created a demand for basic food craps and allowed the advent of several light industries, of which tex- tiles and dairy processing had major relevance. Up to the first decade of this century, the region experienced constant growth. However, intensive exploitation of the land, and increased competition of coffee growers from the southern states of Sao Paulo and Parana created a gradual process of economic decline in the area. The southern land areas of Brazil were more suited for mechanization and large scale production. Growers in Zona da Mata could not compete favorably with these areas and the results are still observed today. Later, in the early 19605, a government sponsored coffee eradication program further contributed to the decay of the coffee based economy of the region. Today, Zona da Mata is seen by development planners as a problem region. Per capita income in the rural areas is among the lowest in the country, and 1975 estimates place this figure around U.S. $250, or about 25 percent Of that of Brazil. 5 Agriculture still generates about 80 percent of all job opportunities and 05 percent of the total income. The number of farm properties is estimated at 67,070 in 1975, with a total area of 3,163,118 hectares.6 Additionally, some 20,000 share- crOppers are present in the region, engaging primarily in subsistence crOp production . 5The World Bank, op. cit. GIBGE, op. cit., 1979, p. zuu (Vol. I). 10 Land use data shows Clearly the relevance of the livestock subsector. Considering land allocated to pastures as a proxy, the figures in Table 2.2 Clearly support the above assertion. Dairy, in particular, has gained increased importance over the years, and Zona da Mata has become an important milk supplier for the greater Rio de Janeiro area. In fact, the area's topographic Characteristics and the comparative liquidity Of dairy enterprises have provided strong incentives for these activities to develop. Milk production reached 016.1 million liters in 1975, or approximately 15 percent of the state's output for that year, while the size of the dairy herd was 367,600 cows.7 Ranking second in land use, annual craps also play an important role in the area's agricultural economy. Corn and beans, often inter- cropped, are cultivated by the vast majority of farmers. Other impor- tant annuals are rice, sugar cane and tobacco. Coffee is the leading activity among the perennial craps, whereas garlic and tomatoes are the main activities among a smaller group of vegetable growers. The total 1975 output of the major crOps mentioned above is shown in Table 2.3. The table also depicts the average yields for these crops, and for comparative purposes the yield estimates for both Minas Gerais state and Brazil are presented as well. The relatively low yields of most of the area's main crops can be interpreted as an indication of the traditional nature of its agricultural sector. Although the total output has been experiencing increases over the years, such a process is mainly a consequence of the expansion and intensification of land use. Clearly, this process is naturally 7IBGE, op. cit., 1979, pp. 1237-1235. 15 Table 2.2 Land Use and Number of Farms by Group of Activity; Zona da Mata--1975 Land Use Number of Farms Item Area (Ha) Percentage Total Percentage Perennial Crops 113,379 3.7 25,786 38.2 Annual crops 382, 800 12. 0 58,173 86. 2 Pasture 2,166,650 70.3 59,812 88.6 Forests 303,239 11.1 38,961 57.7 Other 76,028 2.5 11,009 16.0 TOTAL 3, 082, 500* 100. 0 67, 070 ---—** Source: IBGE, op. Cit., 1979. *Total area does not include unproductive lands. “Given the existence of diversified enterprise mixes, percentages do not add to 100. 15’! 9' .16 Table 2.3 Total Output of Major Crops and Average Yields-—1975 Yields (ton/ha) Total Total Zona Minas Production Area da Gerais CrOp (tons) (Ha) Mata State* Brazil“ Corn 206,039 151,319 1.63 1.03 1.00 Beans 25, 802 79, 907 . 32 . 50 . 60 Rice 99,851 83,208 1.20 .78 1.52 Coffee 00, 758 70, 837 . 60 . 83 . 75 Sugar cane 1,013,078 30,301 29.55 30.05 06.23 Tobacco 0,312 7,010 .61 .70 1.0 Source: IBGE, op. cit., 1979. *EPAMIG, Departamento de Economia, Internal Documents. **Ruy, Paiva; Salomao Schattan and Claus Freitas. Setor A ricola do Brasil: Comportamento Economico, Problemas e Possibilidades; Sao Paulo (Secretarfida Agricultura, 1973). 1970 figures. .. ‘4‘. Illlll 2’95 1.»- I" 395i! N “'2‘ x; it is ill 17 limited. Soils are already exhausted, and tOpographiC conditions largely preclude the development of new agricultural lands.8 A further factor often associated with the economic problems of Zona da Mata is the prevailing tenancy structure. According to UFV researchers, ". . . there is a strong contrast between land use and its ownership."9 Recent figures from the 1975 Agricultural Census still demonstrate high levels of concentration of land ownership. These figures are summarized in Table 2.0. The predominance of small sized farms is Clear from the available evidence. About 52 percent of the farms in the region have less than 20 hectares of land, and altogether they account for only 8.7 percent of the total agricultural area. On the other hand, a small number of larger sized farms (10 percent) hold about 55 percent of the area's land resources. The extent of inequality in distribution of land owner- ship is also lllustrated by the Lorenz curve presented in Figure 2.2. The area between the equality line and the curve itself is a measure of the degree of concentration in land distribution, and the associated Gini coefficient was estimated at .68.10 Recalling that total inequality is expressed by a coefficient of 1.0, the estimated ratio further indicates the intensity of the land problem. 8Universidade Federal de Vicosa, DER; Dia nostico Economico da Zona da Mata (Vicosa: lmprensa Universitarla, 1971T, p. xiv. 9 lbid, p. xiv. 10The estimation followed the methodology proposed by N . Kakwani and N . Podder in "Efficient Estimation of the Lorenz Curve and Associated Inequality Measures from Grouped Observations," Econometrica 00 (1) , 1976: pp. 137-108. 0” €.F~:LI 18 53— ..20 .QO .mOm. "occaom I I 8 .2: 8 .2: 2 _ .8. .m a; .2. .ZFOF 8.2: 8.2: i. 8. So...“ _ cocmE oco 88 ~a.ao oa.ea ea._ Ne. mmo.na m. ooomioeom am.ae Ra.aa ma.~ e_. m~m.ma no ooe~-oeo_ .m.ma Ra.eo .a.n em. a~e..a~ .om ace—loom aa.na am.aa __._~ Rm.m omo.eoc ~e~.~ oem.oe~ a~.oo ne.mm oa._~ ea.h oeo.aeo aoa.a ee~.ee_ ~a.=a Am.aa o~.e_ me.~F am_.oeo R~o.a eo_-em e~.m~ mk.mn oa.o_ n~.a~ am~.mmm ohm.c. onioN an.» .m._m ~m.m mo.a_ ac_.aa_ mam.~. e~-e_ am.m oa.~m em.~ mn.a_ _ma.an hmm.e epim .o._ m_.a_ Re. no... nao.n~ ooa.n mIN a_. ha.o __. No.m oam.m ~aa.~ N-— mo. ma.~ me. ma.~ mao.. one._ _.o NOL< w meta". w no.2 meta”. A2,: meta”. no :3: Eco“. 023.:an w m mo..< 59:52 .30... Co 3.20 mum—leans. up aCON QEmcocio new... a.~ oeaoc. Cumulative Percentage of Farm Area 19 100 7'5 - so - 25 — l J 25 so 75 100 Cumulative Percentage of Farms Figure 2.2 Lorenz Curve for Land Distribution Data, Zona da Mata, 1975 20 The predominance of small sized farms has often been associated with 11 This the relative backwardness of Zona da Mata's rural economy. limitation in land resources places a serious constraint on the expansion of the marketable surplus, and generally contributes to an oversupply of family labor in most farm households. Off farm work then becomes a critical source of family income, and under limited job opportunities, the incentive to migration is further reinforced. In the absence of policies aimed at structural change, the avenues being followed by planners addressing the problems of the study region have been mostly oriented toward a better allocation of available resources. Use of new agricultural inputs, reorganization of enterprise mixes, land use management and introduction of new activities are some of the alter- natives being proposed. Furthermore, investments to upgrade the social infrastructure are usually coupled with these efforts. This study will utilize information generated by an important develop— ment program under implementation'in the Zona da Mata. The World Bank sponsored integrated Rural Development Project for the Zona da Mata region (PRODEMATA) , represents a conscious effort towards the minimiza- tion of the region's malaises. The scope of this project, as well as the sampling plan of the study will be the subject of the next Chapter. "R. Singh, et al.; Poor Rural Households, Technical Change and Income DistributToTn infiLDCs, Brazil: A Summary (Department of Agri- cultural Economics, Purdue University, 1979) . CHAPTER III THE INTEGRATED RURAL DEVELOPMENT PROGRAM FOR THE ZONA DA MATA REGION OF MINAS GERAIS STATE (PRODEMATA) Introduction The main objective of this chapter is to present an overview of the World Bank sponsored PRODEMATA project. The several components of the project and the results achieved to date are discussed. Emphasis is placed on the aspects dealing with the monitoring and evaluation surveys carried out by the Agricultural Economics Department of the Federal University of Vicosa. The sampling plan used for these surveys is examined In some detail as the information generated constitutes the data base for the analyses developed in this study. Project Background As part of a continued effort towards the revitalization of the agri- cultural economy of the study region, several agencies of Minas Gerais state have joined development planners of the World Bank in preparing and implementing the Mated Rural Development Program for the Zona da Mata Region (PRODEMATA) . The program, whose Chief plan of action is expressed by the "First Rural DeveIOpment Project in the State of Minas Gerais,"1 has as its target population the large number of small farm owners and sharecroppers of the region. 1The'World Bank, op. cit. 21 22 Essentially, the project aims at an overall upgrading of welfare standards for the target population. This goal is to be achieved through a series of incentives to induce real increases in farm incomes, and enhancement of the area's social infrastructure. Real increases in farm incomes are expected as a consequence of expansion in the area farmed and from better yields, whereas investments in health, education and sanitation should provide better living conditions for the rural pOpula- tion in general. The measures for attaining these objectives as enumerated in the Integrated Rural Development Project include four major elements, namely (1) provision of agricultural credit for crop and livestock produc- tion, as well as reforestation, land reclamation and rural electrification; (2) provision of supportive agricultural services, including extension, research and demonstration facilities and COOperative organization; (3) investments in health, sanitation and education; and (0) monitoring, administration and evaluation. Project Components and Costs The major thrust of the PRODEMATA project is its agricultural credit component, which includes some 60 percent of the total U.S. $139.0 million of project costs. This component provides low interest loans2 for on-farm investments, working capital, reforestation and land reclamation, bene- fitting a sizeable number of peasants who did not previously have access to normal credit sources because of their lack of collateral or other institutional rigidities. The allocation of credit among project beneficiaries was designed to benefit primarily sharecrOppers and farm 2Given Brazil's high inflation rates, interest for these loans are actually negative. 23 owners with holdings of up to 100 hectares. Larger sized farms (100 to 200 hectares) could qualify for credit for special purposes such as reforestation and land reclamation. Other provisions were designed to ensure that most of the available credit would indeed be used by smaller farmers and sharecroppers. Further project components include a program to reclaim 8,000 hec- tares of poorly drained lands (Land Reclamation component) and a pro- gram to improve the quality and outreach of the state's Agricultural Extension Service (Agricultural Extension component). The former component would provide financing for equipping and staffing the state's Rural Development Agency (RURALMINAS) to design and supervise the necessary engineering projects. The latter, coupled with an Agricul- tural Research and Demonstration program, would consist primarily of an effort to deliver extension services to the majority of small farmers so far deprived of such aid. The extension staff would be augmented, trained and equipped to maximize their area of coverage. The State's Research Agency would be responsible for the deveIOpment and demon- stration of improved agronomic practices to be further disseminated among the project beneficiaries. Project costs for these components are about 21 percent of the total funds available, including a provision for the development and strengthening of cooperative groups. The social infrastructure component includes investments in basic aspects of health and education. Existing health services would be improved by upgrading or constructing several health care delivery posts throughout the area. Further investments would be made in a basic sanitation program and in nutritional improvement action plans. Specifically, pregnant and lactating women, as well as children under 20 five years of age, would be the target beneficiaries of the health and nutrition plan. Emphasis would be given to maternal and Child health care and to preventive vaccination against communicable diseases. A food supplement plan would ensure adequate nutritional intakes for the target group, through the distribution of enriched foodstuffs. Other sanitation measures, including the construction of latrines and distri- bution of water filters were expected to reduce the incidence of endemic diseases such as schistosomiasis in the rural areas. With regard to investments in education, the program calls for the betterment of both primary schooling and agricultural training for the rural populations. Actions would include construction and improvement Of primary schools and vocational centers, training of teaching staff, promotion of short term instructional programs for farm families and assorted supply of equipment and materials. Costs for both components are approximately 13 percent of the total. Finally, an administrative component representing three percent of project costs was designed to oversee and continuously appraise its implementation. The administrative and coordinative tasks would be performed by RURALMINAS which, in turn, would report to the State Planning Secretariat (SEPLAN) for overall policy coordination and guidance. Other state agencies would be responsible for the specific components within their jurisdictional boundaries (viz. Health Secretariat, Education Secretariat, etc.) . Monitoring and evaluation were delegated to the staff of the Agricultural Economics Department of the Federal University of Vicosa, a leading agricultural university in Brazil, which is located in the center of the study area. 25 As previously stated, the monitoring and evaluation activities of UFV have provided the data on which the study is based. The pro- cesses whereby these activities were performed, as well as some of the preliminary results attained by the project are reviewed next. Monitoring and Evaluation In consonance with the specifications of the loan terms, the progress of the project would be monitored continuously by an evaluative team which in turn would provide the necessary feedback to administrative officers at RURALMINAS. The essence of the evaluation task would be to compare the with and without project situations as measured by a set of major indicators including net farm income, changes in net worth, productivity indexes for different farming systems and measures of social welfare. Furthermore, other aspects of project implementation would be appraised by interaction with the diverse project units. The institution selected for this evaluation has substantial exper- ience on the agricultural issues facing the study region. For several decades UFV has been a leading institution in agricultural education, research and extension in Brazil, and its contributions to the deveIOp- ment of the country's agriculture are numerous. The Agricultural Economics Department, in particular, has been a pioneer in this field in Brazil, and was responsible for a major diagnosis of the study region in the late sixties.3 AS such, the appraisal team could build upon pre- vious experience to perform the functions associated with this task. 3Universidade Federal de Vicosa, op. cit., 1971. 26 To accompany the progress of the project, comprehensive surveys of the target population would be performed annually, covering primarily crOp production, acreage and prices; livestock production; on farm consumption and marketed surplus; levels of input use, technology and prices; credit use and sources; tenancy arrangements; off farm work; cooperative participation; socio-demographics and other aspects relating to health, education and welfare; reforestation and land reclamation. AS of this writing, surveys have been conducted in 1977 (baseline year), 1978 and 1979, covering respectively the crop years of 1976/1977, 1977/1978 and 1978/1979. The Sampling Plan The sampling process followed a careful selection of units to properly represent the target beneficiaries of this project. Farm owners were chosen from prepared master lists in accordance with the criteria described next. For purposes of definition of the geographical boundaries of the survey area, the regional division utilized for administrative purposes by the State's Extension Agency (EMATER-MG) was adopted. Accordingly, the three administrative units of Muriae, Vicosa and Juiz de Fora had their municipalities and populations identified and listed. It is impor- tant to mention here that the boundaries defined above correspond to the ones described in Chapter II by the IBGE system of Micro-Regions. The next step in the sampling process was to identify those communities which would be assisted by the project in each of the municipalities previously listed. Then, selection of four municipalities by region was undertaken, of which at least two were randomly Chosen. 27 The remaining were purposedly included in the sample whenever their populations were considered large insofar as target population Charac- teristics were concerned, and whenever regionally specific crops assisted by the program (e.g., tobacco and sugar cane) were not present in the formerly selected municipalities. The Chosen municipalities, their WATER-MG administrative regions and the corresponding IBGE Micro-Regions are presented in Table 3.1. The size of the total sample was determined to be 850 farmers, and the allocation of sampling units among the selected municipalities followed a criteria of proportionality to the Size distribution of farm prOperties in the respective sampling areas. Thus, each municipality would be represented in the sample with farm owners in one of four classes of size of holding, namely 0—10 hectares, 10- 50 hectares, 50-100 hectares, and 100-200 hectares. Based on the records of the National Institute of Settlement and Land Reform (INCRA), a random drawing of 700 farm owners in the ranges above was then carried out. Concerning sharecroppers, the lack of an appropriate sampling frame required that sample elements in this class be indicated by the selected farm owners. It was decided that at least 50 sharecroppers should be sampled in each administrative region, Of which a minimum of 12 Should be present in each municipality. The final configuration of the sampling plan is as shown in Table 3.2. The sample as defined above was surveyed for the three consecutive crop years previously indicated. Further surveys are planned for the entire period of project implementation (up to 1981) . Ian 1' Jul: ‘IJrI 28 Table 3.1 Selected Sample Municipalities by Administrative Region and Corresponding IBGE Micro-Region Administrative . . . * IBGE Micro-** Region MunlClpallty Region Juiz de Fora Alto Rio Doce Mata de Vicosa Juiz de Fora Juiz de Fora Sao Joao Nepomuceno. Juiz de Fora Santos Dumont Juiz de Fora Muriae Carangola Mata de Muriae Leopoldina Mata de Cataguazes Manhuacu Vert. Ocid. do Caparao Muriae Mata de Muriae Vicosa Ervalia Mata de Vicosa Ponte Nova Mata de Ponte Nova Raul Soares Mata de Ponte Nova Uba Mata de Uba Source: *Municipality, “IBGE Micro-Region. *Universidade Federal de Vicosa, DER; Programa lnte rado de Desenvolvimento ch Zona da Mata - MG: Primeiro Relatorio AnuaI de Avaliacao (Vicosa, 1979). “IBGE, op. Cit., 1979. 29 Table 3.2 Sample Size by Class of Farm Class Size Sharecroppers 153 Farm Owners 698 0-10 Hectares 232 10-50 Hectares 316 50—100 Hectares 98 100-200 Hectares 52 TOTAL 851 30 Some Preliminary Evaluation Results Analysis of the project's impact has only been performed thus far for the first year of project implementation. Hence, the results outlined here correspond to the immediate effects of the induced measures on some major areas of actuation. A comparison of the baseline data with that of the 1977/1978 crop year shows that the project is exerting substantial impact on the target population. By and large, positive results were observed in different degrees for most components. Cropped area increased by approximately 21 percent, whereas gross farm income grew an impressive 33 percent in real terms. Particularly sharecroppers and smaller farmers with less than 50 hectares of holdings have achieved substantial increases in productivity for both subsistence and commercial crOps. Beneficiaries in the 50 to 100 hectare range (mostly dairy farmers) have shown sizeable increases in pasture area and productivity. With respect to the social components, most of the planned actions were reported in progress, although information was still insufficient to provide meaningful figures on the project's impact. The comparative analysis performed also provided a basis for the elimination and. substitution of farmers in the sample which did not meet previously specified consistency criteria in their responses. Thus, the original data file was purged of the conflicting and incomplete ques- tionnaires, resulting in the creation of files with less respondents, but presumably with higher degrees of accuracy. This study utilizes these purged files in the subsequent analyses. The new sample sizes for the baseline year and for the 1977/1978 crop 31 year are shown in Table 3.3. Data for the third year survey were not yet available for analysis as of this writing. mm This chapter has presented an overview of an important Rural Develop- ment effort currently underway in the Zona da Mata region of Minas Gerais state. The PRODEMATA project aims primarily at the improvement of living conditions of its target pOpulation, which consists of large numbers of small farm owners and sharecroppers. This goal is to be accomplished through a series of incentives to promote increases in farm incomes and through investments to upgrade the region's social infrastructure. The four major elements of the project are the provision of agricultural credit; the provision of supportive agricultural services; investments in health, sanitation and education; and monitoring, administration and evaluation. Monitoring and evaluation are being performed with basis on infor- mation derived from yearly surveys covering several aspects of farm production and socio—demographic characteristics of target beneficiaries. The sample consisted originally of 700 farm owners and 150 Sharecroppers which later (1977/1978 crop year) was reduced to 517 farm owners and 130 sharecroppers. The reduced sample for the 1977/1978 crop year is utilized for the analysis carried out in this research. 32 Table 3. 3 Purged Files: 1976/1977 and 1977/1978 CrOp Years Class 1976/1977 1977/1978 ShareCroppers 129 130 Farm Owners 021 517 0—10 Hectares 112 139 10-50 Hectares 220 266 50—100 Hectares 59 72 100-200 Hectares 19 00 TOTAL 550 651 Source: UFV evaluation team; personal communication. CHAPTER IV OVERALL CHARACTERISTICS OF THE SAMPLED FARMERS Introduction Although primarily intended as an information source for the PRODEMATA monitoring and evaluation team, the sample surveys described in the previous Chapter also provided the basis for a detailed description of the rural economy in the project area. For the purposes of this study, such a Characterization has been developed to give a more thorough understanding of the area's food production system. The information provided by the sample surveys covers a large number of aspects relating to both the farming systems and the quality of life in the Zona da Mata region. In view of the main objective of the present study the latter features were left for further appraisal. The discussion which follows will emphasize only those aspects considered relevant to the analytical purposes of this research. Readers are referred to the UFV reports"2 for details on socio—economic traits of the Zona daOMata's rural population. It Should be pointed out that all the descriptive information pre- sented in this chapter refers to the crop year 1977/1978, which corresponds to the purged file (651 farms) described in Chapter III. 1Universidade Federal de Vicosa, Op. cit., 1971. 2Universidade Federal de Vicosa, op. Cit., 1979. 33 30 Farm Sizes and Allocation Of Land Resources The average Size of holding for the 651 farmers interviewed in the 1978 survey was 30.2 hectares, with Sizes ranging from as low as .1 hec- tare up to 202. 5 hectares. Given the Characteristics of the sample design, most of the sampled farms are grouped in the lower boundary of the range. Thus, the median size of holding is 17.9 hectares which, when compared with the average size, can provide an illustration of the skewness of the sample distribution. Within the sample, the distribution of farm sizes is as depicted in Table 0.1. Table 0.1 Average Size of Holding: 1977/1978 Crop Year Average Size Standard Number of Class of Farm Of Hglading Deviation Observations Sharecroppers 7.1 11 0 130 0-10 ha 7.0 5.3 139 10-50 ha 26.3 12. 9 266 50—100ha 71.0 13 9 72 100—200 ha 138.0 25 6 00 Total Sample 30. 2 35. 9 651 Source: PRODEMATA Evaluation Survey Patterns of land allocation among major uses in the surveyed area show that this resource is employed intensively in a large diversity of crop and livestock related enterprises. On the average, farmers utilize some 20 percent of their land in cropping activities: 59 percent is for pasture and the remaining is either unused, under forest cover or under other uses (Table 0.2). 35 Table 0. 2 Average Land Allocation Among Major Uses (Ha) Zona da Mata: 1977/1978 Crop Year 0-10 10-50 50—100 100-200 Total Use Sharecroppers Ha Ha Ha Ha Ha % Crops 0.6 3.0 6.9 12.7 20.7 7.1 23.5 Pasture 2.0 2.7 15.0 06.9 91.1 17.9 59.3 Forest .3 .S 3.0 8.8 19.0 3.6 11.9 Unused .1 .0 .7 2.0 0.2 .9 3.0 Other Uses .1 .0 .7 1.0 2.6 .7 2.3 TOTAL 7.1 7.0 26.3 71.0 138.0 30.2 100.0 Source: PRODEMATA Evaluation Survey The percentage Of land used for crops diminishes as the size of holding increases. This negative association reflects the emphasis placed on subsistence oriented crops by the smaller farmers. In fact, smallest farm owners (0—10 Ha) and sharecroppers combined have up to 60 percent of their land resources allocated to crOps. Larger sized holdings on the other hand tend to allocate more land to pastures, suggesting more emphasis on livestock activities (Table 0.2) . Although information on land types is not available from the sample survey, it is quite obvious that land use in the general survey area is conditioned by, topographic characteristics to a large extent. Cesal and Bandeira have examined the technical feasibility Of different farm activities under varied land types in the Zona da Mata region.3 Their findings suggest that very few enterprises could be carried out indis- criminately in the region. Pastures and corn growing are among these 3L. Cesal and A. Bandeira; "Uso da Terra na Zona da Mata de Minas Gerais;" in Estudos Sobre uma Regiao Agricola: Zona da Mata de Minas Gerais; MonografTa No. 9 (Rio de Janeiro: IPEA, 1973), p. 126. 36 activities. In consequence, they predominate in Zona da Mata. Corn is grown by 80.5 percent of the sampled farmers, and represents approxi- mately half of the average cropped area (Table 0.3) . The importance Of pastures has already been pointed out. As mentioned in Chapter II, intercropping is a common practice among farmers in the region. Corn is intercropped with beans by more than half of the producers surveyed, and the percentage is much higher for the smaller sized holdings. The extent of this practice is so prevalent that Agricultural Research Agencies in the state have reduced their efforts to promote the cultivation of corn and beans in separate fields to research more extensively the agronomic aspects of the association. Vieira, in pointing out this fact, suggests that the intercrOpping practice is both a resource use Optimization strategy and a means to ensure diversified diets and income sources." He further states that the strategy minimizes the risk of crop failure, since the incidence of some common diseases is reduced with this sort of cultivation.5 Yet, management difficulties inherent to the strategy usually lead to lower productivities than could otherwise be attained, Specially for beans. Despite their unsuitability to cultivation on hilly lands, beans are the second major cropping activity of the sampled farmers. About 75 percent of all growers interviewed produced this crop under one or more of the cropping systems shown in Table 0.3. Being an important item in the daily diet of most Brazilians, dry beans are also appealing to small llClioas Vieira ; "Cultivo Consorciado do Milho com Feijao," lnforme Agropecuario 06 (0), 1978, pp. 02-05. 5ibid., p. 02. 37 .Ecn» 9:3 or: c. mocnum o_Om CCn ooQQOLuLouc. Son CO 95395 o5 3 982.3...he .momnocon “cocoa—to o5 CO omnco>n “.3592? $3.5m 828.96 S N. c on. a 3. NM .N. a S. a. Nm. mdomu 55.0 on. .m. o. RN 2 am. a... 2. MN m... a. 2. uz Anzv mcozn> A2,: mcozn> .nz. mCOZn> AnIv mCOSn> Ant. . mc02n> Ana >z>tu< ILomnO no..< Icomno no..< iLomn—O noL< Lomno no..< lLomnO no..< lLomnO no..< o: .98. a: 8T8. a: 8.3... a: one. o: S... a: cocoocuococm 33323.. .3722 c. mc_una_o_u..nn_ mLoELnn. ocn ow: DEN.— m .3 San... 38 scale growers, to the extent that crops can normally be harvested twice yearly under traditional agricultural practices. Moreover, its traditional Characteristic of being an appendage to corn production further enhances Its importance in terms of acreage and farmer participation. Smaller sized farms tend to allocate relatively more land to beans than larger sized ones. About 36 percent of the crOpped area of the three smallest classes are under this crop, while the figure for the two larger classes is about 28 percent. The former were also observed to have relatively higher percentages of grower participation in the activity. Rice ranks third in cropped area. Even though this cereal is also consumed daily, topography constraints likely preclude larger acreages being planted by Zona da Mata farmers. In fact, the requirement of flat lands for best agronomic results is in itself a major barrier to rice cultivation in the survey area. As a result, the average percentage of crop area under rice cultivation is rather low in comparison to its role in family diets. Twenty-one percent of the crOpped land was devoted to rice by 60 percent of all survey farmers. As expected, the acreages are prOportionately larger for the smaller sized holdings. However, partici- pation within farm Classes tends to increase with farm sizes, ranging from 53.2 percent for owners with 0-10 hectares to 69.0 percent for those with 50-100 hectares. Notwithstanding the eradication efforts of the mid-sixties, coffee is still an Important agricultural activity in the survey area. Perhaps by virtue of long established traditions, crop land allocated to this perennial is second only to the corn lbean cropping systems discussed earlier. This figure ranges from 17 percent for sharecroppers to roughly 35 percent 39 for farm owners in the range of 10- 50 hectares, with an average of 30 percent. Participation by survey farmers is not very pronounced, averaging 00.6 percent for the sample as a whole. The presence of share- croppers engaging in this activity is an interesting feature emerging from the survey information. Since most sharecropping contracts are verbal and Short term, sharecroppers do not usually cultivate perennial crops. The UFV evaluation team hypothesizes that this Characteristic may be explained by an increased number of father—son sharecropping contracts.6 Still, the larger sized holdings predominate in terms of relative partici- pation. Within the survey area, two localized crops were important enough to influence the inclusion of specific municipalities in the sample design. These are sugar cane and tobacco, but for the purposes of this study the latter was grouped in the category "other crops." Sugar cane production is common in the region, although the majority of growers are located in the area's centermost municipalities. As expected, the sugar processing industry is also located primarily in this subregion. The participation of growers in this activity is only 20 percent, with an average acreage of .51 hectare. Farm owners in the range of 100-200 hectares predominate in terms of participation (08 percent), and this largest sized Class also ranks first in relative sugar cane acreage. Finally, the remaining crop land is allocated either to fruits, vegetables or other crOpS. Citrus and bananas are the Chief fruits cultivated, whereas tomatoes and garlic take the lead in vegetable growing. These activities are still very incipient, and commercial cultivation is limited to a few producers, which 6Universidade Federal de Vicosa, op. Cit., 1979, p. 88. 00 are generally favored by proximity to centers of demand. "Other crops" include tobacco, the leading crop of the Uba Micro-Region, and other less important activities such as cassava, potatoes or groundnuts. Cry Yields Yields for the main crops cultivated by the sample farmers were found to be generally lower than the averages for the entire state in the same crop year. As evidenced by the figures in Table 0.0, the only crOp which displayed results comparable to the state average was rice, with an average yield of 1.11 tons/hectare. Performance was especially poor for beans, which yielded only 37 percent of the state's average. As indicated in Chapter III, low yields may be viewed as an indica- tion of the traditionalism of agricultural practices in the survey area. As a matter of fact, an examination of the survey data concerning agro- nomic aspects of crop cultivation has shown that very few of the farmers in the sample use modern inputs or follow technical recommendations in their cultivation practices. The use of improved seed varieties in bean cultiva- tion, for instance, is practiced by only four percent of the surveyed farmers, and only six percent follow technical recommendations for plant spacing. 7 The pattern is similar for most crops grown in the area, par- ticularly for those regarded as subsistence crops. An inspection of Table 0.0 does not suggest any general pattern of association between yields and size of prOperty for the more traditional subsistence crops. However, such a tendency is more apparent for coffee and rice, with outputs tending to increase with the size of property. Sugar cane, a 7PRODEMATA Survey Information processed in this study. 01 . 83. 6:3an amnmv 33.62:: oanam comings. .mu_Eocoow _n..3_:u_..m< CO «coECnQoo SN .02 Loan.”— >Eocoow .nLNE cnofc< ..Nn_..om_z cLoLaLOz .ocn_nm:n: of c. LoELnu :nEm of ocn omcncu .nu_ccuo._... 329.0 .0 ocn Lotta .9 £58.52 .Q ocn um: .Q .mnm. 35.95395 ounum comings. 5033.835 .95. oosnznaaca ...m_m>.nc< octon. oi... < "mLoELnn. cnu..c< mcoE< 358:. .9595... Co cotantuma och... "33.3.0 .0 moans 9539: Lo» oomn .o_nn__n>n ac: m_ cflunELOCc; .ocn; an momoacsa .nuzrncn of L8 acn>o.o.. .233 «c: oLos none—o... connexion... c. >:mcoo «:2... c. moococoti Lo» «.5093 3 moocuos. .noLn ooQQOLOLouc..n Co «E: Lon 9.3 c. 2.5.5ro 95 mco_un>:_:u ooaQOLuLouc. LO» mo_o_>m .co=n>_:ao CO non: ooh: o5 Looc: omnocun ocn comuusoOLQ mocEEOUN 63. £3 ms anaoommocm< 258c— ”8.50m. .>o>..:m co_.n:_n>m (C‘s—woo”?— "8.50m is. 5.... and. 3...: NN.... 2.: So. mz< 29m annm a: 878. a: 873. a: o... -e. o: 3-.. . acoddoco 8.6 £950 mnczz .nuoh Iocncm .nIEoF. mQOLU L922 no mo.o_> oano>< a .2 Bank. 02 more commercial crop, shows an inverse relationship, with relatively lower yields for larger farms. Levels of Input Use Further understanding of the low agricultural productivity of Zona da Mata farmers is shown by the survey data on levels of input use. For crOp activities, the major inputs are seeds (or shoots), organic fertilizer, chemical fertilizer, machinery utilization, use of animal labor and use of family or hired labor. Feeds, vaccines and labor are the basic inputs for livestock activities. Tables A.1 to A.5 in Appendix A summarize input expenditures by class of farmer for the major crop and livestock enterprises. To ensure uniformity in comparisons among classes, the figures are reported on a per hectare of cropped area basis, or per animal unit of livestock for each specific activity.8 Labor figures are presented in hours per hectare (or hours per animal unit), since the information available does not allow a breakdown of the input into family or hired labor by activities.9 The data in Appendix A, which are summarized in Table 0.5 indicate clearly that labor is the most intensively used input for all activities and farm Classes. Assuming the opportunity cost of family labor to be equal to the average market wage of Cr$09.00 per man day,10 labor costs would account for at least 70 percent of input costs for most activities 8Such reporting approach is suggested in Martin Upton; Farm Management in Africa (London: Oxford University Press, 1973) . 9For purposes of cost allocation among alternative enterprises, a propor- tionality method was developed to circumvent this problem. Details are provided in Chapter V. 10Equivalent to U.S. $2.72, at the exchange rate of June 29, 1978 (U.S. $1.0=Cr$18.03). 03 .213 do .23. .O .>..s oom .mounL chLo>cou co mznuoo Lou .zc: .nE_cn:moo «an... CO «.5 c. commocaxo 95 no.5?» No.33 Coon ocn moo: $3.38 eaten—cu: .noLn ooQQoLO no ncRLO c. commocaxo oLn 3..sz .oorzuoan 33.350 32.5... .>o>..:m oEEnm 8.8. 8.25 3.5... $52 9...... 3...: 88:35. 38“. 3...: No... 3... SN. NN.... 2.... 68598 x :95 Loan: 9.50 >Eo._ omnso>< llll and farm classes. Furthermore, it can be inferred that smaller scale producers tend to use relatively more labor than their larger counterparts. Farm owners in the range of 0-10 hectares, in particular, consistently utilized more hours of labor per unit of cropped area than any of the other farm classes. By comparing the observed levels of labor utilization with the tech- nically recommended labor/land ratios for the region, the intensity of the use of this factor again becomes apparent. For the most important crops (corn, beans, and rice), observed levels are generally higher than the technical recommendations, with a few exceptions among the larger farm owner classes. On the other hand, a similar comparison for the use of animal traction and machinery demonstrates that these two inputs are, for the most part, underutilized. Recommended hours of animal traction per hectare under technically defined practices, for instance, are in most cases three to five times higher than actual use. Concerning the levels of expenditure in fertilizer and pesticides, a substantial portion of input costs is accounted for by these inputs. In fact, expenditures on chemical fertilizers are second only to labor costs in almost all cases analyzed. Although information on technical recommendations for fertilizer and pesticide use could not be readily obtained, data from the 1975 Agricultural Census indicate that average expenditures in the sample are below the state's average at 1978 prices.11 Average figures for the state are about no percent higher than the ones for the sample as a whole, with the gap tending to become narrower as 11IBGE, op. cit., 1979, p. Sll, corrected for inflation by the Index of Prices Paid by Minas Gerais Farmers; published by EPAMIG in lnforme AgrOpecuario 53 (5), 1979, p. 79. us the size of holding increases. An analogous pattern was observed for expenditures in seeds and shoots. Livestock enterprises are also labor intensive with respect to input expenditures. For all of the livestock activities considered, labor expenses using the prevailing market wage rank first for all farm classes. With the exception of farm owners in the 100-200 hectare range, labor usage in the sample was more intensive than technically recommended for the dairy activity-~the most important of the livestock group. On the other hand, expenditures in feeds were found to be relatively in line with the state's average at 1978 prices, the same being true for the average costs of veterinarian care. Summing up, this brief appraisal of levels of input use has shown that, on the average, farmers of Zona da Mata are by and large bound to traditional labor intensive practices, and the role played by modern agri- cultural inputs is, with few exceptions, very limited. This observation is in agreement with the findings by the 1969 Economic Diagnostic.12 Apparently the agricultural sector of the study region has experienced little, if any, changes in this respect during the last decade. Market Orientation Additional insights into the nature of agriculture in Zona da Mata can be obtained from an examination of the allocation of farm production between on farm consumption and sales. Conventional wisdom has been that farming in the study region is largely at subsistence levels and is characterized by low levels of market participation. Moreover, a commonly proposed hypothesis by researchers concerned with the general problem 12Universidade Federal de Vicosa, op. cit., 1971. #6 area states that market participation is closely associated with size of holding. 13'” The survey data seems to confirm those beliefs. Table 4.6 depicts the average percentages of on farm consumption and sales relative to farm production for the main activities considered. Since storage, payments in kind, retention of production for seed, losses or other uses are not explicitly considered, percentages will not add to 100 in most cases. Aggregated activities have percentages com- puted from the total monetary value of production, consumption and sales. It is recognized that using declared prices as a weighting factor for grouped activities may bias the estimates toward the high valued crap or livestock enterprises. Yet, the market value was considered the only feasible alternative of a common denominator to allow comparisons at aggregated levels. It is also relevant to mention that the portion of sharecropping production remaining with the land owner was not accounted in the computations. As expected, most of the output of corn and beans did not find a way to markets. Only 16.7 percent of all corn and 37 percent of the beans produced by the sample farmers was sold. Sales tended to be lower than the average for farmers in the 0—10 hectare range, whereas the highest levels were found among producers in the 10—50 hectare class. Plausibly, larger farm owners have relatively low sales levels because of their greater utilization of corn as a livestock feed. Availability of on farm storage could also explain relatively low levels of corn and bean sales for these farm classes. On farm consumption, on the other hand, is generally high, with averages above 50 percent for most cases analyzed. 13Universidade Federal de Vicosa, 0p. cit., 1971. 1"The World Bank, op. cit. '47 .>o>.=..m 035mm 135 N6. .1: .1... .1... ..... .1NN N... .1» .1... .12 a... m6N mks... .6. 6.. .<.z .<.z .13 .. .1... m6 .1... m. 1.... .. mmum m6. 6... .<.z .<.z .1NN .. .1.. ...N .1... .1... .1.... .1. mmomu mmzho ..... .1... m6. .1... 1: .1... .1... 66.. ....m .....N N6. .... mz2>=U< 20.5mm .23 a: 8...... a: 8...... a: 8-... a: S-.. m.aao_o..uo..£m c3523.... Egan. o. ”$323. mofim cc... cgudsamcou uo mommucoULon. omnco>< o... 0.2m... us As to rice, it can be noted that although the level of sales is relatively low, averaging just 33.0 percent for the total sample, consumption levels were likewise low. In fact, a marked characteristic among rice growers was the relative balance between on—farm consumption and sales. Thus, more details on the disposition of the remaining portions of the rice crop would have to be known before a more prOper assessment of the market orientation of this crap could be made. For most of the remaining activities considered, the degree of market participation was considerably more intensive. One exception was sugar cane, which had an average sales level of just ill percent, a figure actually lower than the observed consumption level of “7.0 percent. Sugar cane is often used as a raw material for rudimentary farm based industries producing syrups and "rapadura"--a hard, brick sized bar of raw sugar, among other products. However, such uses accounted for just 13.“ percent of the total production in 1975, according to census figures for the state of Minas Gerais.15 Hence, the high observed con- sumption level is more likely reflecting the importance of the crop as a livestock feed. Commercial activities, in the more strict sense of the word, include coffee, dairy, livestock, vegetables and "other crops." Sales levels were the highest for this group of enterprises with averages ranging from “6.8 percent for "other crops" to 89.8 percent for dairy. So far the market orientation of specific farm activities has been emphasized. In order to evaluate this characteristic for the farm system as a whole, an index of market participation was computed for each of ”IBGE, op. cit., 1979, pp. 98, 194. 149 the surveyed producers. The index is defined as the ratio between the total sales of the farm and the value of total farm production at market prices. By definition it would range from zero for farms with no sales, to 1.0 for farms selling the entirety of their outputs. The average index for the sample was .39, with a standard deviation of .29. Farm owners in the loo-200 hectare class had the highest participation with an index of .57, while sharecroppers had the lowest index of just .le. The figures for the 0-10 hectare, 10—50 hectare and 50—100 hectare classes were respectively .31, J” and .ll9. The index also allowed the testing of the hypothesis that participation is a function of size of holding. By estimating a simple linear regression model relating these two variables, we verify a close association between them. Although just seven percent of the variations in the market index could be accounted by the size of holding, the estimated coefficient had the expected sign and magnitude, and was highly significant. The hypothesis of no association is easily rejected at the .01 confidence level. It becomes clear from the foregoing that agriculture in the study region has not yet achieved a degree of market orientation which could dismiss previous beliefs as to its semi-subsistence characteristics. Pro- duction oriented to meet on farm needs is still very much a reality, and commercially oriented farming seems to be a major occupation only for the larger sized farms. Farm Earnings Given the different characteristics of the sampled farmers with regard to size of holding, enterprise mixes, cultivation practices and resource endowment among others, a large variability was encountered in the estimation of gross value of farm earnings. The gross value of 50 farm earnings, as discussed by Yang,16 is here utilized as an initial measure of farm income. It encompasses the market value of total crop and livestock production plus any additional earnings accruing to the farm family. As such, the measure takes into account income derived from off farm work, which is known to be an important source of farm household earnings in the study region. The average earnings of the Zona da Mata farm family were estimated at 868111 cruzeiros (U.S. $0815.0l, with a standard deviation of 188020 cruzeiros (U.S. $10u28.0) (Table lL7). Considerable skewness was observed in the distribution of farm earnings, with incomes tending to cluster around values lower than the mean (Figure 11.1) . The positive skewness coefficient of 13.611 computed for the distribution indicates that farm earnings are grouped more to the left of the mean, with most extreme cases clustered towards the right. Indeed, close to 70 percent of the sampled farmers had earnings below the estimated mean. The median earnings, estimated at approximately 110,000 cruzeiros (U.S. 52218.5) provides another reference point to characterize the skewness of the sample distribution. Variability both within and across farm classes was very high, plausibly because of the reasons pointed out above. Within farm classes, the coefficients of variation were above 99 percent for all but the 10—50 hectare group. This strongly indicates that size of holding by itself accounts for only a portion of variations in farm earnings. Across farm classes, earnings increased along with the size of holding, but the coefficient of variation was extremely high, being estimated at 16Yang, 0p. cit., p. 511. 51 Table ll. 7 Total Value of Farm Earnings (Cr$) Average Standard Farm Class Farm Earnings Deviation Sha recroppers 29, 350 39, 255 0—10 Hectares 32,090 38,2911 10- 50 Hectares 72, 987 72, 573 50-100 Hectares 175,915 133,702 100-200 Hectares 400,998 609,276 Total Sample 86,811 188,020 Source: PRODEMATA Sample Survey. 52 mmc_c..mw Econ. no co::£.3m_o $26.6. 3553. 35...... E. ..N. 2. .6 .... 6.5m... on as. c..— a; can suoue/uasqo 53 approximately 217 percent. The sources of such high variabilities will be examined in more detail in Chapter V. Coffee, dairy, livestock and off farm work in that order were the major sources of earnings for the sample as a whole. Within farm classes, subsistence oriented crops and off farm work were relatively more impor— tant earning sources for the smaller sized producers while commercial activities took the lead among large scale growers. Family Couposition and Education of Household Members To conclude the characterization of the surveyed farmers, this section examines the basic demographic data of the sample. The only variables discussed are family sizes, age distribution and educational achievement of family members, in view of their importance for the analytical aims of this study. The average family size in the total sample was determined to be 5.5 persons, with a standard deviation of 2.8 (Table 11.8) . Family sizes ranged from two to 111 members. Table 4.8 Average Family Composition and Age Distribution of Family Members 0-10 10-50 50-100 100—200 Total Age Sharecroppers Ha Ha Ha Ha Sample Less than 15 years 2.8 1.8 1.8 1.9 1.2 2.0 15—50 years 2.6 2.2 2.11 2.5 2.7 2.5 Over 50 years .6 1.0 1.0 1.1 1.1 1.0 TOTAL 6.0 5.0 5.2 5.5 5.0 5.5 Source: PRODEMATA Sample Survey 54 The typical family is composed of the farmer, his wife, one to two sons, one to two daughters and up to one relative living under the same roof. Sharecroppers have the largest families (6.0 members) whereas farm owners in the range of 100-200 hectares have the smallest (5.0 members). Family sizes in the survey area are larger than the average for the state as a whole which, in 1970 was determined to be 5.1 members per household.” in terms of age distribution, 46 percent of all family members are in the age bracket which contributes more potentially to the composition of the labor force (15 to 50 years) . About 37 percent are under 15 years of age and 17 percent are above 50 years (Table 4.8). The educational achievement of the sampled farmers is low. Approxi- mately 23 percent of them are illiterate, and another 50 percent had incomplete elementary education (Table 4.9) . On the average, the sampled farmers have attended school for just 2.1 years. Sharecroppers had the least time of formal schooling (1.4 years) whereas farm owners in the 100—200 hectare range had the most (4.4 years). The average figures for the 0-10 hectare, 10-50 hectare and 50—100 hectare classes are respectively 1.7, 2.3 and 2.8 years. These findings are in agreement with the World Bank's discussion of educational opportunities in the survey area.” Summary This chapter presented descriptive information characterizing the farmers in the PRODEMATA Sample Survey. The evidence discussed tends to reinforce most known beliefs about the nature of agriculture in 17IBGE; Censo Demografico de 1970-~Minas Gerais (Rio de Janeiro, 1972) . ”The World Bank, op. cit. 55 the survey area. Generally, farming is traditionally oriented, with high labor intensity being a major trait among most types of farm enterprise. There is a large diversity of farming activities in which producers engage, but subsistence crops (corn and beans), coffee and dairy stand up as the most relevant, both in terms of land use and farmer participation. Crop yields were found to be generally lower than averages for the state as a whole, and use of modern inputs was not a widespread practice. More- over, agriculture in Zona da Mata has a relatively low degree of market orientation, particularly among small scale producers. This group of farmers was also shown to be at the bottom of an earnings distribution schedule which, in view of its high degree of skewness, suggested a large concentration of earnings among large scale producers. Finally, demographic and educational data provided a general picture of relatively large farm families and low levels of formal schooling among area growers. This characterization clearly demonstrates the need for policies to promote rural development in the region. Though the PRODEMATA pro- ject should have an important impact in this respect, it will be argued in the next chapter that an aspect deserving further consideration is the evaluation of the potential for improving farm incomes through increased knowledge on returns of alternative farm enterprise mixes. 56 Table 4.9 Formal Education of Zona da Mata Farmers (Percent) 0-10 10-50 50-100 100-200 Total Level Sharecroppers Ha Ha Ha Ha Sample llliterate 10.0 5.6 6.5 .7 .0 22.9 Elementary incomplete 10.7 12.2 19.8 6.2 1.1 50.0 Elementary complete 1.3 2.0 6.7 2.7 .9 13.6 Secondary incomplete .4 .5 .9 .2 . 7 2. 7 Secondary complete .2 .0 .5 .4 .4 1.5 Higher .0 .0 .0 6 . 0 6 Private Instruction .9 2.0 4.9 .5 .4 8.7 Source: PRODEMATA Sample Survey CHAPTER V ENTERPRISE EMPHASIS AND FARM INCOMES introduction Rural deveIOpment strategies are often focused on the removal of constraints perceived to preclude the full realization of potential benefits from farming activities. This emphasis is clearly demonstrated by the strategies being proposed today for the Zona da Mata region. At the center of the prescribed actions is the implicit belief that access to modern agricultural inputs is a major factor conditioning rural incomes. Therefore, it is reasoned that easier-to-get credit would help circumvent this basic restraint. Preliminary results of the PRODEMATA project seem to validate this hypothesis. However, it is further argued that farm incomes are also a function of the type of farming system followed by producers. It can be hypothesized that differences in incomes within groups of relatively homogeneous farms with similar resource endowments are associated with different farm enterprise mixes. Low income farmers. could conceivably be faring poorer than their higher income counterparts because they are undertaking activities which yield relatively lower returns to fixed factors of production. Assuming this hypothesis to be true, policy makers should develop programs to provide more participation in higher return farm enterprise mixes. 57 58 It is recognized that the question of farm enterprise selection is not as simple as most choices made on a day-to-day basis. The decision is very crucial for a producer with limited resources, for his livelihood and that of his family will be greatly affected by the final outcome. Thus, it is likely that this decision will be influenced by many factors, of which access to inputs may be one of the more important. Yet, the research question which must be addressed before endeavoring into an analysis of choice conditioning variables is the nature of the relationship between enterprise mix and farm incomes. Once this association is better under- stood, the study of factors affecting choice is certainly more meaningful. The relationship discussed above is examined in this chapter. Essen- tially, the analysis will develop a basis for testing the statistical signifi- cance of the association by means of both parametric and non-parametric procedures . Related Research Although the theoretical aspects of the question under appraisal have received fairly adequate treatment in the literature,1 the same could not be said for the empirical testing of the underlying hypothesis. In fact, for the general setting of this study, some evidence can be found indi- cating that the. relationship between income and farm systems is largely accepted as given. Stated another way, researchers have been sub- stantially preoccupied in the reorganization of farm plans by means of mathematical programming techniques or partial budgeting approaches. 1See, for instance, the excellent reviews of Gerald Helleiner, "Smallholder Decision Making: Tropical African Evidence," in A riculture in Develo ment Theor (New Haven: Yale University Press, 19755; and o n Cleave, "Decision Making In the African Farm," contributed paper read at the 16th International Conference of Agricultural Economists (Oxford: Oxford Institute of Agricultural Economics, 1977) . 59 The selection of an enterprise mix is thus implicitly accepted to be a conditionant of farm incomes.2 Being essentially normative in character, this approach fails to provide insights into the differences in outcomes of observable farm plans. Positive approaches to the investigation of the question at hand have been adopted in Africa in studies by Franzel,3 Matlon,ll and Upton and Petu.5 Franzel investigated the relationship between net farm return and choice of enterprise for a sample of smallholders in Sierra Leone. His analysis indicated that although low income farmers tended to emphasize low return enterprises (and vice versa) , variations in income were more associated with variations in resource productivity than with the choice of farm activity itself. These findings were partially in agreement with Upton and Petu's analysis of a Nigerian farm sample survey. By utilizing some interesting measures to represent a farm system, they found that gross margins per hectare in the sample were more significantly associated with the farm plan than with an index of farm production efficiency. When returns were. expressed in terms of gross margin per man day, the farm 2Farm planning studies for the Zona da Mata region and neighboring areas of Minas Gerais state are myriad. See, for instance, the review provided in Mussolini Greco, "Determinacao da Renda das Empresas Rurais, em Relacao a Melhor Combinacao de seus Empreendimentos Basicos;" unpublished M.S. thesis, Universidade Federal de Vicosa, 1972. 3Steven Franzel; "Enterprise Choice, Enterprise Combinations, and Income Distribution Among Farmers in Sierra Leone;" unpublished M.S. thesis, Michigan State University, 1979. “Peter Matlon; "Income Distribution Among Farmers in Northern Nigeria: Empirical Results and Policy Implications;" African Rural Economy Paper No. 18; Department of Agricultural Economics, Michigan State Univer- sity (East Lansing: 1979). 5Martin Upton and D. Petu; "An Economic Analysis of Farming in Two Villages of lllorin Emirate;" Bulletin of Rural Economic and Sociology, Vol. 1, No. 1, 1964; Department of Agricultural Economics, University of lbadan, Nigeria. aim TBS rel He pn ex PF re ta th fa 60 plan had a negligible effect though, and differences in incomes were almost totally accounted by measures of management performance. Matlon's research found no clear relationship between income differences and the types of farm plan followed by smallholders in three Nigerian villages. As it can be inferred from the sparse evidence identified, the hypothesis that differences in farm incomes are related to differences in the farm plan can by no means be outrightly accepted nor rejected. On the contrary, the contradictory empirical findings suggest that the relationship is plausibly specific and therefore non-generalizable. Hence, testing is warranted before any inferences on the matter can be properly made. Farm Classes Appraised Although the original PRODEMATA sample survey is stratified into five classes of producers, the two extreme classes (sharecroppers and larger farm owners in the range of 100—200 hectares) are not considered in this analysis. The characterization of the sample developed in Chapter IV demonstrated that the larger farms (100—200 hectares) differ considerably from the smaller sized farm groups in most of the aspects examined. Furthermore, these farmers are not in the major group of PRODEMATA target beneficiaries, as they are assisted only for land reclamation and reforestation purposes. Sharecroppers, though impor— tant beneficiaries of the project, pose a serious research difficulty in that the decision making constraints they face are rather unique. Contractual arrangements have large variability across farms and farming activities. Generalizations about sharecroppers' behavior under 61 these circumstances are risky and of doubtful value.6 The analysis then emphasizes only the farm owners in the range of 0-100 hectares, and the breakdown into three strata is maintained. Measurement of Incomes Measures of farm income are available in different degrees of sophistication. 7 Because of the characteristics of the sampled farmers, net household income was selected over alternative standards. Net house— hold income is defined as the value of total farm production plus receipts from other income generating activities (including wages from off farm work) minus the costs of purchased inputs.8 The measure is appealing for the Zona da Mata region, since on farm consumption is explicitly accounted for in the computations. The low degree of market participa— tion earlier characterized calls for an income definition which does not attribute an excessive weight to cash sales. Farm output was valued at the prices farmers declared to be receiving for sales of their products at the time of theinterviews. By using this approach instead of a standard market price for each product, variability attributable to differences in marketing conditions is preserved in the measurements. Costs of purchased inputs were also valued at declared 6Recognizing these problems, the UFV evaluation team is supervising research work dealing specifically with the sharecropping question, as indicated by Evonir Oliveiraf(personal communication). 7Yang;.op. cit., presents a good discussion of these measures. 8Robert King and Derek Byerlee; "Income Distribution, Consumption Patterns and Consumption Linkages in Rural Sierra Leone;" African Rural Economy Paper No. 16; Department of Agricultural Economics, Michigan State University (East Lansing: 1977). prices. of anim difficu'. Ti activit tad where manp farm the a diffe fernal c0m). 62 prices. The exceptions are hired labor, service of machinery and service of animals, which were valued at standard rates for the region in view of difficulties with the data base. The allocation of hired and family supplied labor hours between farm activities was performed by a proportionality method because of the same data difficulties. The formula adopted is: n—s n LH..= L../ X L.. - Z L.. - FL. , for farms with ll {( ll i=1 ll) (i=1 I] J) n X L.. 3» Fl. i=1 II I where: LHii = man days equivalent of hired labor allocated to the ith activity at the jth farm9 Lii = man days equivalent of total labor allocated to the ith activity at the jth farm Fli = man days equivalent of family labor available at the jth farm 5 = activities involving selling of family labor n = total number of activities It is assumed that labor is hired whenever the total stock of family manpower is lower than the declared total labor usage in farm and non- farm activities. The difference between usage and availability is thus the amount of hired labor employed by farm owners. To allocate this difference among activities, the observed proportions of labor usage in 9One hour of male labor is considered equivalent to 1.5 hours of female labor or two hours of child labor. Such a scale uses the standard conversion rates adopted in the PRODEMATA Evaluation Report (op. cit.) . usual iikeli size alloc- mac run 63 each farm are applied accordingly. For obvious reasons, the hours of family labor Spent in off farm work are excluded from the computation of proportions. Another assumption implied by the allocation criteria is that hired labor is only a complement to family supplied manpower. No allowance is made for the engagement of agricultural workers in activity-specific tasks not performed by family members. Given the low levels of specialization usually encountered among farm workers, the assumption provides in all likelihood a fair representation of practices followed by producers in the size classes under appraisal. Consistency checks with the results of the allocation showed no major incoherences. Capital costs were regarded as fixed and therefore are not included in the adopted income measure. Although depreciation of tree crops and livestock can be considered as a variable cost, the inclusion of such items in the analysis would impose specific data requirements beyond the scope of the sample survey. To avoid unwarranted arbitrariness in the estima— tion of the number of trees per farm, life span of items, interest rates and other information needs for depreciation accounts, a decision was made to regard these depreciation costs as relatively fixed in the short run.10 The summary results of the estimates discussed above are presented in Table 5.1. Differences between farm classes are significant at the .05 level.11 10For a discussion of the nature of tree and livestock depreciation costs, see Upton (op. cit., Chapter 12) . 11Differences among means are tested in this analysis through use of standard ANOVA procedures. For details, see Norman Nie, et al.; Statistical Package for the Social Sciences (New York, McGraw Hill, 1"“)975 , p. "259. 64 Table 5.1 Average Earnings, Variable Costs and Net Farm Income (Cr$) Average Average Average Farm Variable Net Farm Class Earnings Costs Income 0—10 Hectares 32, 090 5, 962 26,128 (38, 294) (9, 544) (36, 503) 10- 50 Hectares 72, 987 13, 999 58, 988 (72, 572) (21,175) (63, 028) 50-100 Hectares 175,916 42,519 133,397 (133,702) (61,674) (114,892) NOTE: 1. Figures in parentheses are standard deviations. 2. U.S. $1.0 = Cr$18.03 65 Enterprise Returns Based on the net farm income figures computed for each producer in the sample, returns to labor and land were estimated for the different enterprises identified (Table 5.2) . The results indicate that differences among farm classes are more pronounced in terms of returns per man day than when the measurement is done in relation to cropped area. Indeed, differences in return to labor among alternative farm sizes are statistically significant for all but five of the activities considered. Smaller sized operations, known to be more labor intensive, obtained lower returns for most enterprises than the larger sized farms. Conversely, returns per hectare do not differ significantly among farm classes in a statistical sense. The only exception is the dairy enterprise, which yields higher returns to the larger sized farm classes. Since a single measure of enterprise profitability is needed for the proposed analysis, an evaluation of the alternative standards initially considered resulted in the indication of returns to labor as a preferred choice. On the one hand, the measure differentiated returns among farm classes in a better fashion, as seen above. Moreover, the provision of a common denominator for crop, livestock and nonfarm enterprises alike is a convenient attribute for comparative purposes. Considering that comparisons among enterprise returns are a prime concern of this study, returns expressed in cruzeiros per man day of labor equivalent are certainly the more adequate measure. For the stratum of smaller sized farms, returns for the alternative enterprises ranged from as low as Cr$35.0 per man day for corn cultivated under both regimes (intercropped and sole), to Cr$216.6 per man day for fruit growing. Generally, the more commercial enterprises tended to yield higher returns than the subsistence oriented ones, as evidenced by ._0>O— m6. 05H um «Cmuzmcmmm 0L0 mflmmfl-U ELM..— Cwflgufln— MOUCGLflhbmcflfi .32: _GE_C<\WLU C— UNmmOLQXO 0..” m0L3m_..— XUOum0>_I_ USN >meot "MNFOZ 66 3.0. $9.38: Log 3...:qu 3.0. >3. cw... Led mccaaom ll ll ll .13. 6...... .6... v3.0... .52.. to 66...... N...3.N .636 362.. .663 «.1... xoo._.mm>_.. ..........N 36.... .66.... .66... .N... .66.. $.33 .63.. 6...... 6. .63.. .1NN.. ........ 66.N 3.5.... .1..N..6 . 63.6. 66.... ...N .13 6...... N.....N mmum .63... 6...... 663... 6.... ....N N.3 macmo mmzho 6.....66 ..........N .636 ...N.. .66 .13 mz< ~.m 0.00... 67 the ranking of enterprises in descending order of profitability. The highest returns group encompasses fruits and vegetables, coffee, beans cultivated under both regimes and off farm work. The average returns per man day from off farm work were roughly twice as high as the average daily wage paid to agricultural laborers in the region. Since the off farm enterprise combines labor use in both the agricultural and non-agricul- tural sectors, such results were expected. For producers engaging in off farm activities, non-agricultural occupations accounted for the highest portion of the hours allocated to this enterprise category. Because average returns to labor in the non-agricultural sector are usually higher than agricultural wages, its heavier price weight is responsible for the high returns observed in the off farm work activity. The more traditional cropping activities--oorn and beans-—tended to cluster in the bottom portion of the distribution of enterprise returns. Middle return enterprises on the other hand include corn and beans under single cnop cultivation, other crops, livestock, dairy, rice and sugar cane. Finally, it should be pointed out that differences in returns among alternative enterprises within the 0-10 hectare class are significant at the five percent level. Concerning the 10- 50 hectare stratum, the grouping of enterprises into high, medium and low return classes produced somewhat different results from the previous group of farm owners. Traditional craps were also found in the lower portion of the distribution, but surprisingly the dairy enterprise was classified in the lowest returns group as well. An examination of the return components in the dairy enterprise for this group of farmers showed that the comparatively high level of labor inputs in the dairy activity was most likely the reason for the observed low return. 68 Compounding the problem, feed costs per animal unit were also found to be relatively higher for that size group of dairy farmers. In the middle returns class the activities were off farm work, sugar cane, rice, corn ‘ in sole stands, and beans in sole stands. Corn cultivated under both regimes, as well as the combined figure for this grain are also classified in the group. Similar to the previous farm size class discussed, off farm work had returns well above regional agricultural wages. These results are consistent with expectations for the same reasons cited earlier. The group of higher return activities includes basically those which are more commercially oriented. Fruits, livestock, other crops, coffee and vegetables are the enterprises classified in this group. Differences in enterprise returns for this class of farmers were also found to be sig- nificant at the five percent level. For the third farm size stratum, enterprise returns ranged from a low of Cr$57.9 per man day for beans (sole), to as high as Cr$1279.5 per man day in the case of livestock. Livestock, fruits, dairy, rice and corn under both cultivation regimes were grouped in the highest return class. The you: of medium return enterprises embraces off farm work, beans under both regimes, vegetables, other crops, coffee, and both intercropped and combined corn, Finally, sugar cane, combined, intercropped and sole beans, and corn in sole stands comprise the group of low return activities. As in the two former farm classes, differences among returns were found significant at the .05 level. The differences in enterprise returns among farm classes suggest that essential dissimilarities may exist in the way producers in these groups conduct their activities. Resource endowment is certainly one of the major factors influencing returns, but productive efficiency, price ' 69 relationships or even agronomic conditions could be playing a key role in the relative performance of these farm system components. One of the aspects related with the efficiency question--the variability in factor intensities--has been observed in the characterization of the surveyed farmers. Yet, to infer from observed factor intensities without concern for the specific constraints impinging upon producers would be erroneous. By the same token, careful judgement would also be warranted if inferences about the role of input loutput price relations (or agronomic conditions for that matter) in the determination of enterprise returns are to be made. Although a detailed examination of factors affecting returns of smcific enterprises is an important research question, such a pursuit is beyond the purposes of this work. Keeping with the stated objectives, we will concentrate on the role of alternative enterprise returns in the determination of total and relative farm incomes. Producer Emphasis on Alternative Enterprises In order to determine the degree in which producers emphasize alternative enterprises, it becomes necessary to establish a set of dimen- sions characterizing enterprise emphasis. Conceptually, these dimensions should be sufficiently representative to provide a clear differentiation of the extent to which farmers concentrate their resources in alternative activities. Franzel12 measured the degree of enterprise emphasis in his analysis by the percentage of labor each activity absorbs, and by its relative contribution to household income. Additional dimensions, which may 12Op. cit., p. 23. 70 prove useful under specific conditions include the contribution of an enter- prise to household cash receipts, the percentage of crop land it absorbs, the amount of expenditures in purchased inputs and even a farmer's subjective ranking of the importance of each of his enterprises. The measures adopted by Franzel are rather appealing, as they allow ready comparison among farm and nonfarm enterprises. It is recognized, nonetheless, that the stress on the labor dimension may overshadow the importance of other constraints (e.g., land or capital), which could be just as relevant in defining the relative emphasis of each enterprise. This would be especially problematic if the group of farmers being analyzed was not relatively homogeneous in terms of resource endowment. Ideally, enterprise emphasis would be better measured by some combination of the important dimensions. But as Franzel warns, this would cause conceptual difficulties, as measures of economic performance would be combined with measures of resource allocation. ‘3 Considering the need for uniform characterization, and the fact that the farm classes analyzed were stratified by farm size, the measures proposed by Franzel seemed appropriate. Moreover, an empirical comparison between the labor and income dimen- sions with the plausible alternative of using cash receipts, shows that the latter is so. much associated with income that for most practical pur- poses the results would not differ. Therefore, income and labor are adopted as the dimensions characterizing enterprise emphasis in the analysis which follows. According to the selected criteria, an enterprise is considered impor- tant if it contributes ten percent or more to the composition of net farm 13lbid., p. 23. 71 income or if it absorbs ten percent or more of the total labor used at the farm (family and non-family) . Specialization is characterized if more than 30 percent of labor or income is accounted for by one specific activity. The rationale for the cut-off points selected is based on the notion that diversification is often used in small scale agriculture as a risk minimiza- tion strategy. Hence, farmers may well emphasize more than one activity, and they may simultaneously engage in several other important enterprises. By using larger reference points to define importance or specialization, one could easily overlook this important aspect. Theoretically, the specified reference points allow for specialization in three crops per farm in the extreme case. Also, up to ten important enterprises per farm could be identified by the criteria. Following the discussed procedures, the determination of the number of farms specializing or attributing importance to one or more of seventeen alternative enterprises was carried out. The results are depicted in Tables 5.3, S.“ and 5.5, respectively for the 0—10 hectare, 10—50 hectare and 50—100 hectare farm classes. Two of the features presented in Tables 5.3, 5.4 and 5.5 have been discussed in Chapter IV, namely the proportion of farmers engaging in alternative enterprises and the proportion of land absorbed by cropping activities. It is relevant to recall that the clearest features emerging from the discussion were the diversity of enterprises farmers undertake, and the importance of traditional crops in the sample area. These characteris- tics are again apparent in the analysis of enterprise emphasis. 72 ll «.3 9mm 93 93 93 V35; .52“. "to ll a... m... 5.... can mam qumeS: ll 5m .3 ~.~_ 92 .12 >545 a. as N. ~.~ ea ea mtg... _._ m. m; a; .3 9m mmfim o; m. m... m... m... a...” $9.0 51.5 a...” .2 a; ma «.2 3. mz53 .. .... m. ... a... a... who”... u. 5. ~.. .2 e... e... mm4m ..N e. e. e.~ e... ...~. $9.0 $1.5 e... .... m.~ a... ...~. e..~ mz54... e. m. N. ... a. N... 9.5.... N. e. m. ... ... ...N mm...m<._.mom.> ma m.N m... e... N... N... meomu «5.5 m... m. ...N .... m.N. ...N mzo_ ma. of am 33.2.53 3.8.5.35 9.0 momma? oEouc. «meow mcmoE :33qu mmucocota "ouoz NN ”N NN mcozmtomno 8.2:: .83 5.2a .2; $— 68.2. 5:235 2855 2.9352 3.38.8. 3535.2 :82 $528: 873 S 2: 3 3253890 END}: Gas—MN: Eat}: 5:325 2855 NN.N.N.NN_ £53.; 2.2.; :82 3.5.8: 8-2 N: mm N: 3253890 39.2.2: 5.25.3 Snags. 8.333 P5955 2.68.8 8.ch .2 2.2; .N :82 3.58: 2.... 3.20 02m 0500:. 9:02... 058:. £9: 5:522 30.. mam—U oEouc. 3.5. mango 0:32... can mun—U 05m .3 358:. Egan omuco>< w .m 0.0m... size gro the als: the CFC cl; le 82 size class defined the "high returns" group, whereas the five lowest were grouped as "low returns." The remaining activities were then placed in the medium returns group. It should be noted that the combined figures for corn and beans were also ranked along with the different activities. Since most research in the area does not distinguish the alternative cultivation methods for these crops, the combined figure is maintained as a potential reference for comparative purposes. In accordance with the classification above, alternative testing pro- cedures were followed. The results are summarized next. Relative Emphasis by Income Classes: The Test of Differences Among Sample Proportions The rationale for this first testing approach is that an association between farm incomes and stress on alternative activities would be denoted by the existence of differences in enterprise emphasis among the income classes under appraisal. In other words, the statistical hypothesis tested by the approach is whether the proportions of farmers emphasizing each possible enterprise are significantly different among income groups. To illustrate, if the proportions of high income farmers found to emphasize higher return enterprises are larger than the proportions of low income farmers emphasizing these same activities (or vice versa), then the association between enterprise emphasis and income classes would be accepted to prevail. This is the basic approach adopted by Franzel in his analysis. 15 Nevertheless, differences in measurements and result interpretations are here introduced to reflect specific characteristics of the area analyzed. 15Op. cit., pp. 31-34. reh acc size spe par will 5.9 das lowe This exce beloi gene lowe. in tu lbni ”fleet The f Mfka: ClBSS’ 15F See C. [New \I Shtht efien 83 To perform the testing of differences among sample proportions, the relative numbers of farms where alternative enterprises are important in accordance with the labor and income criteria were determined for each - size class and respective income groups. Importance, rather than specialization, was adOpted to ensure adequate degrees of freedom, a paramount consideration in this approach. 16 The results of the classification with an indication of significant differences is shown in Tables 5.7, 5.8 and 5.9 respectively for the 0-10 hectare, lO—SO hectare and 50-100 hectare class. The evidence points to a tendency for lower income farmers to emphasize lower return activities more frequently than higher or medium income farmers. This observation applies for the three farm classes alike, where with few exceptions the larger sample prOportions stressing low return activities belong to the low income categories. Analogously, higher income farmers generally stressed higher return enterprises more intensively than their lower and medium income counterparts. Medium return enterprises were in turn emphasized by the three income groups, without any clear associa- tion being suggested by the data. But despite these observed patterns, there is no statistical basis to reject the hypothesis that such was merely a product of sampling errors. The fact is that very few of the differences among proportions were sig- nificantly different at the .05 level. One exception is the 10-50 hectare class, where six out of seventeen enterprises differed significantly in 16For a discussion of the test of differences among sample prOportions, see C.A. Moser and C. Kalton, Survey Methods in Social Investigation (New York: Basic Books, 1972) . See also C. Clark and L. fihkade Statistical Analysis for Administrative Decisions (Cincinnati: South Western Co., 19710. mvaI—LQLQeCW mv>~uNCLmvu-( :0 NEWQCQEW 0>~un=0m K. in eunuch: 84 , .mcflucaocd 0.9.5... 95:5 30:282.... .0 «no. a 3 953.08.. ..e>o_ ma. 9.: 5 «cost? >_ucmu_:cm_m 0cm xmtoumm cm 5.3 beta... 3.59“. .thz G 3.5,. «9.2. 3.3 mcmom 39.3500 2: «can «~65 .l..: 500 359500 : a... .2: ..N 58 32.9.. 53 ...N. ...N ...8 3.... 38m 3&8835 mm min mém 9...... Foo DoQQcLchc. 2:30”. 26.. mN m .m ...3 m.m~ 550 20m Nm N..N... Q... ...N :8 m. a... “.2 ..h 9.00 ..mmam we «5 .mm 3.3 .3 .~.. mu;— o iii a.— m;— mcmom 0.0m em m .3 min m .3. x0332.— o ...ii m .m a... macho cameo mccauom 53:55. 2. 7mm Tam ...pN oetou _ iii. a; ll mcmom magma”. 50m mp K .3 «6.3 «m .2. x33 Egan. to .. ..h 9. ii: 3.3.". m ..n m.m ..iii 3 930mm 2:33. c .1 9.03.; 9.50:. 9.50:. 0:50:— >22 uu< -330 .o gm... 53.3... 3.. cons-.2 «:2..an— m. 329.0ch 30...... c. «Egan. no cflacodok. 8.28: 3.... 2.39.0 2.50:. >9 www.cdcoucw 02.2232 co 2359.5 02.23. N. .m 2an 85 6:03.500... 0.9.50. 95:... 30.5.0.2... 5 .00. 0 0. 9.55000 ._0>0_ m... 0... .0 50.0.0... 2:50:59... 0.0 225.00 :2... 003.0... 00.59“. "0.02 s m .N a .. a .m ..50 “.0530”. ...om cm «0.». «n ...m «a .9. mcmom 009555.... N 2 «m ...N «m . 2. «m ...m 0:00m 00:29.50 0: «m .2. «o .00 «m .2. >200 a: «m .2 «m .3 ..m . N... ...00 00.05.05...— mc.5.0m 30.. mm. «m ...m 1.5 «a .3 ..50 0059.50 N... 0.... a... m.- :50 05m $2 1...... .16.. 5.8 005. mm «m.~— E... «m...— 0:00 .095 N. c.m m.. m8 mcm0m 0.0m mm m... m.~— mg ..50 0.0.5001 50m we m..m m...~ m.—~ «...53 5.0“. to 0.53.01 5300.2 0. «m .0 «a . ..o .m m0_n0.0m0> m I «0 .cm i .2. «0 .mm 00.50 m. m... 0.. m. 2.9.0 .050 mm. m .3 N .2. ...8 200.002.. o. «m... .3.— ..m... 0.5.“. 0.53.0”. :9: mco_.0> 0:50... 0.50... 0:50... 3.2.04. 3.35 .o :9: 523.2 .2... 59.5.2 50.59... m. 0m...n.0...m 20...... c. 05.0". 5 2820.5...— 00550: on no. 095.0 0.50... .3 50:052.”... 02.2.0.2 .5 0.00.3.5. 02.0.01 m.m 030... 86 0:03.000... 0.0800 .5050 50:05.30 5 .00. 0 0. m:.0.0000 ..0>0. m... 0;. .0 50.0.20 >..:00....:m.0 0.0 20.5.00 :0 2...... 002.0... 00.30.". "3.02 N :1 .... i..- 0...... 000 E N... ...N 0... :30 060 . 0... .5. ...N 0:00 .0030 .N .0... . ...N. 0.... 0:80 0.2.0.80 ..N .0... R... .0... 0:80 02.8.9.2... mar-SHON— 30:— c. ...0. ...m 0.... 550 00009.22... 0. ...0 a... ...0. 0.80 09.3.80 0. ...0. ...0 o... 8.30 0 ..... .... ..... 8.9.0 .050 u- 1!... 1-..: .iu: 00.00.0m0> . :1... 1!. 0... 0:00m 00:..m0m ...0m . ... ..S 0... .8015... ..0 mCLZHQK 3.3.302 .. 0.... m... N... 550 002.03. ...om .. 0... 0.... .1... on... .0 0.0... m... .... 2.00 .. ~... ..-i s.-- 0.3... e. .0 .0. .0 ...0 .0 ...0 0.0200... mcgauwm— S 2..— 0:0. .0> 0500:. 0:50:. 0.50: . >.. 2 .0< L800 .6 .0... 52022 30.. .00.:32 50.50:: 0. 00..0.0.:m 20...... :. 0:20“. 5 :0...000..n. 00.0.00... 00.10... 0030.0 0.50:. >0 000..0.0.:m 0>..0:..0..< :0 0.00....Em 02.0.0”. m .m 0.00... 87 support of the postulated association, though four significant differences did not support it. The other two farm classes had respectively four and three significant differences suggesting a relationship between the group of the enterprise emphasized and the corresponding income class. The main results of this analysis can be summarized as follows: 1) For the 0—10 hectare farm class, emphasis on corn and beans (combined) was more intensive among low income farmers. This suggests an association between stress on subsistence-oriented activities and lower farm incomes. Conversely, off farm work--a higher return enterprise-was associated with higher farm incomes. 2) The 10-50 hectare farm class showed more signs of association between enterprise emphasis and farm income classes. Higher income farmers, for instance, tended to emphasize fruits, coffee and vegetables more intensively than the other farmers. On the other hand, lower income producers were shown again to emphasize intercropped corn and beans, as well as combined beans, which are also low return for this group. Dairying showed a stronger association with higher income farmers, despite its classification as a low return enterprise. Medium return activities were in turn more associated with low income farmers. 3) For the larger farms (SO—100 hectares), livestock was the only high return activity emphasized more strongly by high income farmers. Beans (intercmpped and combined) were emphasized by low income producers. Summing up, this analysis shows that there is little statistical basis to state that the proportions of farms emphasizing each enterprise are significantly different among farm income group in the study region. Yet, the relationships observed for some of the enterprises warrant the perfor- mance of alternative testing approaches to seek for more conclusive results. 88 The Test of Independence of Classification One of the major drawbacks of the test of sample pr0portions resides in its consideration of enterprises individually, without concern for the relative stress placed in more than one enterprise. Given the farm income classification criteria, the same producer could be represented in more than one group of enterprise returns even if one of its performed activities is considerably more relevant than the others. Clearly, a more desirable approach would overcome this problem by classifying each farm in that group of enterprise returns to which the mostly emphasized activity belongs. The basic idea of the analysis performed in this section is the classifica- tion of the survey farms in accordance with the criteria above and the sub- sequent testing of its meaningfulness. If the principles of classification are independent, the implied conclusion is that there is no association between farm incomes and the emphasis placed by the producer on his most relevant activity. Otherwise, the association would be evidenced.17 Since the proposed classification required a unique dimension to define enterprise emphasis, an examination of the two factors utilized throughout this study was carried out to determine which one had influenced the groupings of importance and specialization the most. The inspection showed that labor was the overriding factor in defining the adopted emphasis measures, and for this reason the factor defines the classification of this section. Income is, in turn, used to classify farms in which labor usage was the same for more than one activity. The classification as dis- cussed is presented for all farm classes in Table 5.10. 17A discussion of the principles of this testing approach is provided by Clark and Schkad, op. cit., pp. 376-378. 89 .0538 5:03.03. 0.35000 .0500 0:00 .302 2. - 0.. mm .05... lwm ................. ...m .................. mm ................. will---llll-w......:..w.m...m....._ mm a m. ~. 0:.3.0~. 0.3.005. m o m N 09.3.0”. 30...? 00.0.00... 2.7.0 emu c. 0... o. .05... {mm ................. mm}---llllllJml--l----l-llmmllllllll...............w.fi.fi..n 00 0N 3. 0.. 09.3.0”. 0.3.00.2 mm mm ..m mm 055.00. 2.0.. 00.0.00... 00:... mm. 2. 00 ~.. .05... ...N ................. mm .................. mm ................. mall--l---ll..-w......:..w.m.....m..m me o. 2. ~— 0:.3.0m. 53.00.... em m m N. 0:.3.0¢ 30H 00.0.00: 0.10 _0:o..0> 0.50:. 0.50:. 0050:. 0:.3.0~. 00.... 1.0090 cm... 0.3.00.2 3o. ...0.:w no 00320 .05... 0:0 00000.0 02m .022 0... 02.02%... 000.55.:m .5 0.05.00. 5 :o..00..=000.0 c. .m 0.0.0... 90 The null hypothesis under scrutiny states that the number of farms classified in each of the three income groups considered is independent of the grouping of returns to which the most emphasized activities belong. If this is correct, then one would expect a relatively uniform distribution of farms with respect to enterprise returns in Table 5. 10. At a confidence level of five percent, this hypothesis is rejected by a chi-squared test for both the 0—10 hectare class and for the 50-100 hectare class, but not for the farms in the 10—50 hectare range. Consequently, the results indicate that there is an association between farm incomes and the return characteristic of the enterprise emphasized the most by farmers in the smallest and largest size classes. The more a farmer in these classes emphasizes a high (low) return enterprise, the more likely it is that he would be classified as a high (low) income farmer. On the other hand, such a rule would not hold for farmers in the 10-50 hectare class. The implication is that the activity stressed the most has no impact on the classification of these farmers into one of the alternative income groupings. The outcome of this testing approach allows the inference of an impor- tant farming characteristic of the study region. Single enterprises receiving substantial emphasis can be a consequence of either resource limitations, as in the case of smaller farms, or extreme specialization, in the case of the larger, market oriented farms. This being the case, it can be argued that there is a greater likelihood that extreme emphasis in individual activities would be more associated with income groups for the two classes where this was indeed observed. On the other hand, the middle sized farms are less likely to substantially stress individual enterprises and, 91 therefore, one would not observe a significant association among individual emphasis and farm income groups. In fact, the results of the former testing approach tend to support this inference. Another characteristic shared by the two testing approaches considered is nevertheless a shortcoming. Both fail to take into account the plausible off-setting effects of alternative farm plans, whereby different enterprises belonging individually to different groups of returns could be jointly associated with an income group when considered as a combination. Because many of these enterprises are performed simultaneously, it can be argued that a more appropriate testing methodology would attempt to regard farm and non-farm enterprises as components of an integrated system. Therefore, the hypothesis tested would be the existence of an association between the farm system and farm income levels. The approach discussed next provides the means for this testing. The Indexes of Potential and Performance: Upton‘s Approach Whereas the two testing approaches followed in the foregoing were concerned with the association between enterprise emphasis and farm incomes, the analysis of potential returns here performed will provide a basis for evaluating the relative roles of the farm plan and the individual farmer's management performance in accounting for variations among incomes. This analysis, which was developed by Upton, is often referred in the literature as potential net margin analysis.18 Conceptually, the approach attempts to evaluate the extent to which farmers could achieve potentially higher returns to given factors, were 18Martin Upton, "A Development of Gross Margin Analysis," Journal of Agricultural Economics 16 (1), 1961!. 92 they to work under the standard operating conditions of their respective groups. The potential returns of each activity in each farm are determined by taking the product of actual factor levels times the standard return (average return for the relevant sample) to the factor for that particular activity. By adding up potential returns for each enterprise present in the farm plan, the ratio of potential return to'the input considered is computed and compared to the standard ratio for the group, thereby defining the index of potential. On the other hand, the percentage of a farmer's actual returns in relation to his potential returns defines the index of performance. ”Mathematically, the indexes can be expressed as follows: n m-Si 2 {Li. . ( z Yi.)} - i“ J H l - 100 1 ”30' — 1/m m n n ( ) {Z ( Z Y../£ L..)} i=1 i=1 " i=1 '1 1/m-si n n m--si lPEj = 100 £1 Yii/if1{l'ii - 1/m-s #1; viii} (2) where: IPOj = index of potential of the jth farm lPEj . = index of performance of the jth farm t.ij = units of input utilized on the ith activity at the it“ farm ii = returns of the ith activity at the jth farm n = number of activities m = number of farms in the group 5. = number of farms where the ith activity is not performed 93 The index of potential expressed in (l) is thus simply defined as the ratio between the potential returns of any given farm and the standard--or average returns over the sample--for the input considered. Analogously, the index of performance is given by the ratio between actual and potential farm returns. In his original application of this concept, Upton utilizes returns to land as the relevant measure of farm income. He argues that whenever land is the limiting factor of production, farmers tend to be primarily concerned with the achievement of high returns per hectare. Therefore, potential net margin analysis becomes a more helpful means of understanding farm systems when these particular decision maker characteristics are properly accounted for. Throughout this study it has been noted that farm sizes tend to be closely associated with most of the farming characteristics appraised. There seems to remain little doubt about land being the limiting factor in the majority of farms in the study area. Bearing in mind this basic fact, the indexes of farm potential and performance discussed next were computed under the assumption that land is indeed the constraining pro- duction input. Nonetheless, returns of dairy and livestock activities are considered in terms of animal units, whereas off farm work is main- tained as an activity where return to man day is the relevant consideration. Since the figures for all enterprises are combined in an index, these different measurements will not pose a conceptual difficulty for comparative purposes. The average indexes of performance and potential computed for the sample farms in accordance with expressions (1) and (2) are shown in Table 5.11. An illustration of the computations for one of the farms is presented in Appendix B. 914 00.05300 0... 5 0:0..0.>00 0.00:0.0 0... 0.0 0000...:050 :. 00.39”. .m .00m0.0>0 000.0 00.0 c. .00000. 5...... 00..0.00:0.0 0.0 08.00:. .m . 300. n: 00.300080 <>Oz< .0:o..:0>:o0 o. 0:.05000 .0>0. 500.00 0>C 0:. .0 0:00... 038.... 0.50:. 0:95 0050.0...0 .:00....:m.0 05:00 0x0..0.0< .. "mm—0.02 $.05 .N.00. 2...... «N . .0. «m . .0. i ..m 000.05.00.00 3...... ...00. 8...... .10.. ...N .0. .0 .0. .0530. 00.0.00: 0..-... 8.... ...N: ...N... .0 ..0. ..N .0. .0 .0. 0805.00.00 8.... 8.... 0...... «o . :2 «m.mm «m ...N 33.3.00 . 00.5001 amic— ......... 2.0.. ...000. .. . um. .. .mm 0 . .N. 00:0E.o...00 0...... :00... ...N: ...0. 0.... ..N. 30:80.. 00.0.00: o—ic 0.50:. ..m... 0.50:. ...3.00.2 0.50:. 25.. 00x00:_ 0:0 00000.0 00...... .:.0..0:oU 0 00 0:0... ”00:0...5050 0:0 .0..:0.o0 5 09.00:. 000.03.. ...m 0.0.0... 95 The role of the farm plan in accounting for variations in farm incomes is expressed by the standardized indexes of potential. Farms with indexes below 100 are engaging in activities which yield a lower overall return than the standard enterprise mix of the group. Conversely, indexes of more than 100 denote more efficient farm plans than the average with respect to overall returns. Significant differences in average indexes among income groups would suggest a relationship between the farm plan and the respective income group. On the other hand, the index of performance reflects the relative success of a given farmer with respect to his management methods and experienced input/output price ratios. Farms with indexes above (below) 100 had a better (worse) performance than the average for his respective group. Differences in these average indexes among groups would suggest an association between farmer performance and respective income groups. An examination of Table 5.11 reveals that there are no significant differences in either the average index of performance or in the average index of potential for farmers in the 0-10 hectare size range. This implies the existence of a relative similarity of management methods and farm systems among this group of producers. Observed variations in total farm income would then be primarily accounted for by variations in farm size within the group. Recalling that the classification of farmers in income deciles was done in terms of absolute, rather than relative farm incomes, the observed relationship in the 0-10 hectare class is conceivable. The Opportunity set faced by these farmers does not offer enough leeway to allow the observation of substantial differences in pursued enterprises. Although 96 some differences in relative enterprise emphasis among income groups could be observed in the preceding testing approaches, the consideration of the farm as system averages out such dissimilarities to reflect the relative uniformity which one could expect. It should be noted that the absence of significant differences in the indexes of potential and performance among income groups does not necessarily imply that these factors are not associated with incomes for the size group as a whole. The inference which can be properly made is that the factors do not exert enough impact to influence the classifica— tion of a farm into one of the income groups. If we want to evaluate the relationship between the factors and returns to farm size, correlation analysis would be the appropriate statistical tool. Such an analysis for this group reveals that there is a highly sig— nificant association between the system potential and returns to farm size, though this is not the case for farmer performance. The coefficient of correlation for the former index is .87 and its significance level is .001. On the other hand, the same coefficient for the index of performance is .02, and it is not significantly different from zero at the .05 level. Thus, although the system potential is relatively uniform in the 0-10 hectare class, it is highly associated with farmers' returns per hectare. More insights into the important types of farm systems prevailing in this size range will be obtained in the definition of enterprise combinations discussed in Chapter VII. With respect to the two other size classes, the evidence indicates a clear association between performance, potential and alternative income groups. The average indexes for these two farm classes differ signifi- cantly at the .05 level among their respective income groups. It can be 97 seen that farmers in the high income groups fared better than their lower income counterparts both in terms of performance and potential. The indexes assume higher values as we change from the lower to the higher income groups, suggesting that there are relevant divergences both in management procedures and farm systems within the two size classes. The extent to which the two factors considered in this analysis are associated with returns to farm size can be determined by simple correla— tion analysis. For the 10—50 hectare class, the coefficients of correlation between returns to farm size and indexes of potential and performance are respectively .52 and .65, both being significantly different from zero at the .05 level. Returns are associated with both variables, but farmer performance seems to be playing a somewhat stronger role in their deter— mination. Previously observed differences in emphasis for some enter- prises were reinforced when farms are regarded as a system, but although they: can be associated with farm returns, the relationship is more impor- tant in terms of farmer performance. Differences in enterprise productivity, prices received and costs experienced are some of the factors embodied in the index of performance which could be accounting for part of the observed income variability. In an analogous way, farmers in the 50-100 hectare size range appear to differ more in terms of performance than system potential, though the two factors are significantly associated with their returns to farm size. The correlation coefficients are respectively .74 and .65 for the indexes of performance and potential, and both are significantly different from zero at the .05 confidence level. To sum up, the analysis has shown that the roles of farmer performance and farm systems in accounting for differences among farm income groups 98 is more pronounced for farms where the land basis is relatively larger. For these larger farms, both factors show a relationship with returns to farm size, though farmer performance appears to be more important than the system potential in this respect. The smallest sized farms were found to be more uniform in terms of the factors considered. Summa ry The basic purpose of this chapter was to examine the relationship between farm incomes and specific enterprises pursued by individuals in the sample. The underlying hypothesis is that differences in income levels among producers with similarity of resource endowment may be partially accounted for by differences in enterprise emphasis. If this is the case, greater participation in higher return activities (or combina- tions of activities), would be a logical strategy towards the improvement of income levels. To test this hypothesis, producers were classified into income ter- ciles and their emphasis on alternative enterprises under different groupings of return was analyzed. As a rule, we observed more tradi— tional and lower return activities to be stressed the most by low income farmers within each size class. However, the opposite is true only for a few of the higher return activities and the evidence in this regard is stronger for the 10-50 hectare farmers. In view of the mixed indications from the test of differences among sample proportions, a complementary approach was further adopted. This method investigates the relationship between the returns nature of that enterprise emphasized the most by each producer, and his classification as a low, medium or high income farmer. The results showed that the classification is not independent for the smallest or the largest farms, 99 though it is for the 10-50 hectare producers. Hence, the enterprise emphasized the most by the 0-10 hectare and 50—100 hectare producers has a significant relation to their income classes, and it has not for the 10— 50 hectare farmers. Despite the useful information conveyed by these two testing approaches, the main question asked is still not clearly answered. Although there are indications that an association between enterprise emphasis and income levels exists, there is also the possibility that individual enterprises when analyzed as a group could yield different results. Therefore, the examina- tion of the farm unit as an integrated system should provide a more realistic notion of the extent to which the farm plan accounts for differ- ences in income levels. Potential Net Margin analysis was the method followed in this investi- gation. The results showed that both the farm plan and the relative per- formance of the producer differ among income classes, and that both are significantly correlated with farm incomes at the .05 level for farmers with more than ten hectares of land. The smaller producers in turn are found to be more similar in terms of performance and farm plans, such that observed differences across income levels are not statistically significant. For the larger farms, in both classes the correlation of the index of performance with incomes was higher than that of the index of system potential (the farm plan). The suggestion is that even though both factors are associated with incomes, performance is somewhat more pro- nounced. The results of these analyses provide enough evidence to support the contention that changes in the farm plan towards higher return mixes already emphasized by some producers can bring about increases in income 100 for farmers with more than ten hectares of land. For the smaller sized farms, it appears that variations in income are more attributable to differences in size within the group. Therefore, a study of factors affecting enterprise choice will be con- ducted as a means to improve the understanding of farm planning pro- cesses in the Zona da Mata region. Elements from such an analysis should be helpful in assessing income enhancement strategies based on reorganization of farm enterprise mixes. Additionally, the analysis should provide information on the likely impact of the PRODEMATA pro- ject on existing patterns of enterprise emphasis. This in turn justifies the inclusion of all three farm classes in the study. lts analytical frame- work is presented next. CHAPTER VI SELECTION OF THE FARM SYSTEM: AN ANALYTICAL FRAMEWORK Introduction It has been shown in the foregoing chapter that differences in farm systems are associated with income classes for the majority of farmers in the sample area. Though income variability is likewise linked to farming performance, empirical evaluation of farm management methods, allocative efficiency and related managerial processes are not uncommon for the Zona da Mata regiOn or for other developing areas in Brazil and elsewhere.1 On the other hand, the analysis of factors affecting the selection of farm systems is generally limited to speculation or hypothetical exercises. Factual inquiries upon underlying decision making processes in the choice of enterprise mixes which could be identified in the conventional literature were few and far between. 2 1See for instance, Joao Carlos Garcia; "Analise da Alocacao de Recursos por Proprietarios e Parceiros em Areas de Agricultura de Subsistencia;" unpublished M.S. thesis, Universidade Federal de Vicosa 1975. See also Jacksonwilliam Nagornni; "Analise da Eficiencia Tecnica e da Substituicao de Fatores na Agricultura de Subsistencia da Zona da Mata de Minas Gerais; unpublished M.S. thesis, Universidade Federal de Vicosa, 1980, and references thereof. Further references are provided by Robert Stevens; "Transformation of Traditional Agriculture," in Tradition and Dynamics of Small Farm A riculture, edited by Robert StevensTAmes: Iowa—UnWersity Fress, 1977). 2Studies which went as far as identifying some of the factors in- fluencing choice of farm systems were: Franzel (op. cit.); Edinaldo Bastos; "Farming in the Brazilian Sertao: Social Organization and Economic Behavior;" unpublished Ph.D. dissertation, Cornell University, 1980; and David Norman; "Rationalizing Mixed Crapping Under Indigenous Conditions: The Example of Northern Nigeria," Journal of Development Studies 11 (I), 1970, pp. 3-21. 101 102 Yet, knowledge about the constraints which influence producers towards the pursuit of specific farm systems may be an important element of an integrated rural development strategy. Inasmuch as it is known that farmers with similar levels of resource endowment in this relatively imageneous geographical area behave differently insofar as enterprise selection is concerned, the identification of the afore- mentioned constraints is warranted. This chapter develops an analytical approach whereby the proposed research question can be addressed. On the basis of a broad review of smallholder decision making theories, the polychotomous logit model is presented as a framework to estimate probability functions expressing the relationship between the selection of alternative farm systems and a group of factors hypothesized to influence the relevant choices. The model estimation and results are then discussed in Chapter VII. Smallholder Decision Making: TheorLand Fact Agricultural development planners must base their recommended actions on a set of notions about the behavioral characteristics of target decision makers. Embodied in any conceivable development policy mechanism, is some sort of assumption as to the type of response which should be expected from alternative stimula. Needless to say, prOper understanding of the rationale of farmers' decisions is a §_i_r_I§_ 90:1 291 condition for the accomplishment of rural deveIOpment policy goals. It is not surprising then that the debate about smallholder decision motives in less developed countries (LDCs) has been exhaustively engaged in over the years. Earlier studies tended to present the "ignorant and 103 lazy . . . farmers" (sic), and a backward bending supply curve as major retardants of agricultural growth in these areas.3 This colonial mentality and the associated idea of target levels of income beyond which higher utility is attributed to leisure is certainly a convenient way to justify the status quo. But the extent to which it can explain behavior for the aggregate of smallholders in given areas is of limited, if any, value. Nonetheless, much of the deveIOpment work carried out in southwest Asia in the I950s was inspired by similar views." Professor Schultz's "poor but efficient theory" which appeared in the early 19605 set forth the basis towards dispelling some of the earlier notions about allocative behavior in traditional agriculture.5 The Nobel laureate utilized the works of Tax6 and Hopper7 to prove the hypothesis that "There are comparatively few significant inefficiencies in the alloca- tion of the factors of production in traditional agriculture."8 The empirical basis of this landmark book tended to show that despite the 3The quotation is attributed to Rene Dumont and is cited by Carl Eicher in Research on Agricultural Development in Five English Speaking Countries in West Africa (New York: Agricultural DeveIOpment Council, 1970, p. Ill). “Stevens, op. cit., p. 5. 5Theodore Schultz; Transforming Traditional Agriculture (New York: Arno Press, 1976) . Reprint of the original 1960 edition. 6Sol Tax; PennyCflDitalism (Chicago: University of Chicago Press, 1963) . 7David Hopper; "The Economic Organization of a Village in North Central India;" unpublished Ph.D. dissertation, Cornell University, 1957. 8Schultz, op. cit., p. 37. iou absolute levels of poverty observed in both the Guatemalan (Tax's) and Indian (Hopper's) cases, resource allocation followed closely the neo- classical principles of equalization of marginal costs and returns. What follows as a major implication of the "poor but efficient" theory is that little gain would be derived from the mere reallocation of resources in traditional agriculture. Since resources are already allocated efficiently-—given the terms of trade and technological possibilities faced by producers-~agricultural output and associated income levels would not be substantially affected by recombination of resources. Agricultural development should then be fostered by technological change, bringing about conditions whereby the farmer ". . . has access to and has the skill and knowledge to use what science knows about soils, plants, animals and machines."9 Although Professor Schultz's theory has been criticized under alter- native grounds, its major element-~the rationality of small farmer's behavior-~has endured the criticisms. Disagreements with the theory range from the supposed impropriety of generalization from isolated case studies, 10 to the failure to explicitly consider the crucial dimensions of risk and uncertainty.11 Lipton's argument in particular, though not refuting the postulates of rationality, has provided a theoretical basis which inSpired some of 9|bid., p. 205. 10George Beckford; "Transforming Traditional Agriculture: Comment ;" Journal of Farm Economics 08 (l), 1966, pp. 1013-1015. "Michael Lipton; "The Theory of the Optimizing Peasant;" Journal of Develgpment Studies ll (3), 1968, pp. 327-351. 105 the most recent work in smallholder decision analysis. Stated simply, the argument accounts for the fact that farmers may not always be profit maximizers in the orthodox sense because of both the unsuitability of some theoretical aspects to LDCs and the specific socio-environmental traits of these areas. In fact, production variability (arising mainly from highly variable year-to-year rainfall distribution), impedes the equalization of marginal value products in alternative uses because with these circumstances ". . . there is no unique marginal physical product (MPP) associated with any factor (for given inputs of all 12 More- other factors), but only a probability distribution of MPPs." over, equalization of marginal value products in alternative uses cannot account for trade-offs between variance and expected profits. Thus, even if the problem with the MPPs is assumed away, and if knowledge of previous conditions can reduce uncertainty to measurable risk, the neoclassical rule would not be optimal for the small farmer. The individual cultivator, he argues ". . . requires risk premium, and the risk is abnormally large owing to high rainfall variance, and of an abnormally 13 lmperfection in factor markets, social severe outcome, starvation . " traditions and a relatively unchanging environment are further reasons to defend the argument that the optimizing peasant searches for a survival algorithm, rather than a maximizing one. Hence, small farmers will more likely strive to attain some expected mean income in any given year, thereby maximizing expected mean income over the long run. 12lbid., p. 330. 13ibid., p. 330. 106 By bringing in the risk and uncertainty dimensions, this conceptual- ization constructs a framework whereby observed sub-optimal or "pro- ficient" patterns of resource allocation in traditional agriculture can be reconciled with rationality and utility maximization. Other works which present strong cases in support of the need to understand smallholder behavior in this context include Wharton” and Dillon and Anderson. ‘5 For the area of interest of this study, it has been shown that con- ventional productivity measures indicate a relatively stagnant agricultural sector. Farmers can safely be assumed to have reached the technical and economic equilibrium trap implied by Professor Schultz's theory. Yet, we observe sufficient inter-farm variations in patterns of resource allocation which may be seen to denote inefficiency in the orthodox sense. Why, one could argue, do we observe different enterprise mixes in relatively equal sized farms in the same general area, when some of them generate higher returns to fixed factors? If the conditions are homogenous, should we not expect a more similar pattern of resource allocation? Essentially, the implication of the observed patterns is that there are potential benefits to be gained from shifts into high return farm systems, provided we can relax the constraints which impede decision makers to do so in the first place. Thus, there is a need to conceptualize 1“Clifton Wharton; "Risk, Uncertainty and the Subsistence Farmer;" paper presented at the Joint Session of the American Economic Associa- tion and the Association for Comparative Economics; Chicago, lllinois, December 28, 1968, mimeo. 15John Dillon and J.R. Anderson; "Allocative Efficiency, Traditional Agriculture and Risk," American Journal of Aggicultural Economics 53 (I), 1971, pp. 26—32. 107 the enterprise mix selection decision process in a way which can allow an empirical evaluation of the relevant constraints. The approach developed next utilizes elements of the major theories discussed above to derive a suitable model of enterprise selection for the area of concern. A Conceptual Model of Enterprise Choice For the Zona da Mata Farmer Modeling small farmer behavior is without a doubt a difficult task. Whether for normative or descriptive purposes, it is likely that the exercise will require abstractions and generalizations which will limit inferences to a certain extent. Yet, if we are to understand decision making in the context of smallholder agriculture, we must accept the inherent limitations. Small farms have been characterized by Anderson and Hardaker as "medium number" systems, subject to Weinberg's Law of Medium Numbers. 16 Therefore, the authors warn, their analysis can be ". . . difficult, may 17 These remarks be impossible and will probably lead to many mistakes." notwithstanding, it is here argued that enough gains can be obtained from the modeling exercise to justify its development. On the one hand, the difficulties faced in modeling small farmer behavior are not insurmountable, as we know from the large number of studies which have evolved in this 18 field. On the other hand, the ultimate goal of the exercise is not the 16Jock Anderson and J. Hardaker; "Economic Analysis in Design of New Technologies for Small Farmers," in Economics and the Design of Small-Farmer Technology, edited by Alberto Valdes, Grant Scobie and John Dillon (Ames: Iowa State University Press, 1979), p. 12. The law is "For medium number systems . . . large fluctuations, irregularities and discrepancy with any theory will occur more or less regularly." 17lloid., p. 12. 18A comprehensive reference list in this regard (87 citations) is provided in John Hardaker's insightful review article entitled "A Review of Some Farm Management Research Methods in LDC's," Journal of Agricultural Economics 30(1), 1979, pp. 315-331. 108 perfect representation of the decision making process of smallholders from a behavioral standpoint, as such a model would be devoid of prescriptive value.19 Rather, the purpose which should be kept in mind is the establishment of a framework which enables the understanding of the nature and relevance of constraints facing small farmers' decisions. The translation of such a conceptual framework into a quantitative model further allows the measurement of impacts of alternative actions on the processes analyzed. From an analytical standpoint, there are no essential differences between decision making in small farm agriculture and decision making in any other setting. The process consists of the allocation of resources among a range of choices, such that one or more objectives can be attained. As shown in the preceding section, decisions are made rationally, given specific constraints, objectives and range of alternatives. Hence, under these provisions, decision making can be understood in light of conventional economic theory. The proposed investigation of decision processes related to the choice of enterprise mixes will require that a priori hypotheses be made on the nature of factors influencing these processes. Statistical testing should then allow theascertainment of the most relevant factors. However, the initial hypotheses must be founded in- an appropriate conceptual frame- work. In this connection, the economizing principles of resource allocation among competing alternatives are helpful. Moreover, Cleave's synthesis of extensive empirical evidence on decision making in African farming ‘9lbid., p. 319. 109 provides a concise classification of the variables involved in the processes.20 Therefore, there exists sufficient basis from both theory and past research to formulate initial hypotheses. The conceptual model developed in the forthcoming paragraphs is essentially an approach towards the identification of the constraints which preclude small farmers from fully realizing the potential of their resources.21 As such, it is basically a non-normative model, though with prescriptive value. It is non-normative, for the results of the derived quantitative model will not imply any optimal course of action for groups of decision—makers to follow. The prescriptive value, it is hoped, will follow from the applicability of the model to "what if" type questions which, at the very least, provide the directions of change in the outcome of the decision process given selected policy interventions . That rationality will be taken as axiomatic should have been implied in the foregoing. It is further assumed that the overall process of selection of an enterprise mix is the same for farmers in any of the three classes analyzed. Clearly, the constraints faced, as well as their relative impact may differ. Moreover, the existence of a decision function relating the choice of a given farm system with attributes of choice alternatives and individual characteristics of decision—makers must also be assumed. As Bastos points out, choice in this respect can be inter- preted in the context of a random decision (utility) model, whose 200p. cit. 21The model presented here has no claims of originality. In fact. it follows closely Cleave's conceptualization for the African farm, though with some modifications. 110 framework is well known in the literature.22 Details on the derivation of choice probability models from this framework are provided by Domencich and McFadden.23 The classes of variables which are hypothesized to bear on the decision process analyzed, as well as a schematic presentation of the process are depicted in Figure 6.1. It should be mentioned that the relevant decision is defined in the context of a given time frame: a crop year in this par- ticular case. Motivated by his basic goals, a producer makes an evaluation (myopically or otherwise) of the allocation alternatives available to him in view of the set of constraints faced. Such an evaluation, which is permeated by his risk preferences, will in turn determine the pattern of farm operations in that crop year. Though the decisions on enterprise mix, technology and disposition of output are interdependent, the latter two aspects will not be appraised in this study, as the major concern is with the selection of the farm system. However, it is noteworthy that the general framework applies to the three aspects alike. The general groupings of constraints in the diagram present the guidelines for the inclusion of explanatory variables in the enterprise selection model. It should be noted that not all of the aspects in the diagram lend themselves to straightforward quantitative measurement. Further, even some of the readily measurable factors cannot be brought into the analysis, for the exercise is constrained by the data base. 220p. cit., p. 138. 23Thomas Domencich and Daniel McFadden; Urban Travel Demand: A Behavioral Analysis (New York: North Holland, 1975). 111 Y’"-_--——‘—°‘—----'T'-"'-—""--""-_--W 0.0.055 00.0w .5. ...EJZSU #30:... 0O ZC..._WC..m_D 002:0... ......0>.::U 0:0...0. .000 .c ESE... 0730. 2.0:. >33.DZ:Um—.—. fa... .5... :o 0.....0 0:0. , 01...... £00U x2... um.~.n.~.m:2w .mm. ... ....0 ...o .0>00.U Es... 00.~.00< ”0......cmc .0005... 3:...03. ..30.u.0.u ..00.c.._.c=.m 0 .: .5..0.:_...0....n. 0....E0..Jm ..w 0.33.0 ZC wO.UmQ /\ nlllllll ll'l L.T.---_----- Till m.Z.<~. ..mz..U ..l. m HZ.<~....MZCU .5359: 0.. :0... :20. ..O .o .0I:U 35.3.0003 U2.b..0—<3U l-l: lll.|ll—lllll.ll m. .>_h<2~.m.... _< 0:520... .........c..U .00.. 0......0...0c~. 5.52.0.5. .0......U 0.0.003.— 0xdo.m 0.......:.::00< mhz.<:bm200 ~_w.>O>~.~.0:0.0...:m..0m 05... cc..0~......0.m 9.50:. .0>.>.:m . 00...>.0m 5.0.... ....0 ...0..c..:..m. 2.. 0:050: 0.0.5.. :02 5.. 0:05.; 0:02.09 0.50.... 00.00? ...§.00..c..l ... :2....:0...< 5:0. lCmU..n—Z. A. .3: ..m..C—. 0.00.7.5“. 00.5.01... .5553. 0.5.0:..m 5.5... . :c..0~.:0$.5 3:200:02 00.0.30 $50.... 2.3.52.5... 520...... 0.5—Eh 5.5.0.; 0.05:0 ;.<.—Zw=Zl. :00.U 0.2.... .20....2 c. 00..00< 0:35. 52.31.52 .93.. .00.... .3 >......==s>< . .27.. 3.50.. .0 >.....:m m,Pz.<~:mZCU uUZ~.me~. l l l l l l I 112 Resource constraints in the diagram refer to the available stock of basic factors of production as well as to the decision maker's ability to increase these stocks by market transactions or similar means. It is clear that the levels of resource endowment have a direct influence on the particular enterprises a farmer decides to pursue. This characteristic is well documented in Cleave's work, where he shows how the scarcity of land may lead farmers to engage in high yielding crops.“ Likewise, capital intensive enterprises will be avoided where this factor is in short supply, and labor intensive activities will be followed where labor is not binding. Labor demands during crucial periods in the production process (e.g., weeding, harvest, etc.) may in turn reduce the Opportunities for steady off farm employment. interacting with the resource constraints, a group of factors which is often completely beyond the control of the decision maker can also be identified. This group is labeled "environmental constraints." Though it is likely that not all of these aspects are explicitly evaluated by the farmer, their influence in the decision process is certainly present. The frequency and intensity of rainfall, for instance, should constitute a strong influence on perceived ranges of crop choice. Evidence in this regard is particularly marked for drought prone areas.25 Another aspect which can easily misdirect research on small farms is the effect of cultural values and beliefs on decision making. A whole series of traditions ranging from household division of labor to the social values zqu. cit., p. 162. ZSAUQUStO Scares; "Resource Allocation and Choice of Enterprise under Risk on Cotton Farms in Northeastern Brazil;" unpublished Ph.D. dissertation, Ohio State University, 1977. 113 attributed to specific enterprises is well documented as influencing 26 The value attached by some African ethnic groups farming activities. to livestock possession is a classical example in this respect. Though it is often impractical to account for these factors in formal models, their importance should not be overlooked. Also important are the interactions of decision makers with the broader institutional setting, reflected in the decision framework by their evaluation of market conditions and terms of trade. The selection of the farm enterprise mix, and in par- ticular the balance between cash and food crops, will be closely linked to perceptions of relative input/output price ratios. Studies in the general area of supply response in traditional agriculture present exten- sive evidence on this aSpect.27 Household related activities and the induced demands for foods and nonfoods alike constitute the third group of constraints considered in the model. The tasks related to the maintenance of the household in a day to day basis compete with farm activities for the members' labor, particularly that of the women. Again, in periods of peak labor demand, the problem will be felt more intensively, and it is likely that considera- tions on this aspect should be built in the decision function. Also, the family needs for clothes, processed foods, services and other nonfarm 26The works of economic anthropologists are usually the best references on such aspects. See for instance Stephen Gudeman; The Demise of a Rural Econom : From Subsistencgto Capitalism in a Latin American Village (London: fiztledge and Kegan Paul, 1978), Chapter 3. See also Albert Ravenholt; "The Gods Must Go," in Selected Readings to Accompany Gettin Agriculture Movi_ng; edited by Raymond Barton (New York: The Agricultural Development Council, 1966). 27Gerald Helleiner (op. cit.), presents an interesting discussion on this respect. His references are a valuable source for readers with a particular interest in the subject. iill produced goods will influence the balance between cash, food crops and off farm work in the enterprise selection process. it is known that farmers often do not have complete flexibility to change farm plans from one year to the other. There are rigidities introduced by past farming patterns, rotational restrictions, accumulated stocks and invested capital, to name a few. Hence, such initial conditions can also be interpreted as constraints impinging on the mix selection decision. in the case of tree crops such as coffee for instance, it is rather clear that the flexibility to pursue alternative enterprises is greatly reduced by the initial conditions. The last group of constraints in the diagram has to do with the off farm work opportunities which may constitute viable alternatives to on farm activities. If off farm work is available in either the agricultural or non-agricultural sectors, and if the benefits of these activities are perceived to outweigh those from farming one's own land, then we should expect a lower emphasis on farm vis a vis nonfarm work. Likewise, smallholders are not completely unaware of the opportunities available in the urban sector, as the ranks of unskilled workers in the booming civil construction industry of southeastern Brazil well document. in fact, the outmigration trend characterized for the study region as well as the great importance of off farm work are proof that such considera- tions are present in the decision framework of the Zona da Mata smallholder. The groups of constraints outlined above illustrate the complex environment in which decision makers must operate. Yet, the difficulties of the process are compounded by the need to evaluate both the con- straints and the outcomes of alternative farm plans under conditions of 115 imperfect knowledge. Risk and uncertainty, in the traditional meaning set forth by Frank Knight,28 are then present on several facets of the farm system organization plan. The types of risk involved in traditional agriculture arise primarily from yield and price variability. Both play important roles in the enter— prise selection process, though it has been argued that price risks are relatively less important.29 ideally, a model which attempts to concep— tualize and quantify smallholder decision making should account for individual attitudes toward risk. Knowledge of those attitudes coupled with information on the riskiness of alternative decision choices should increase the power of a model to a considerable extent. Though the methodologies for the measurement of risk attitudes are well developed, none would conform to the available data of this analysis. Elicitation methods following Von Newmann and Morgenstern;"0 (the "direct" approach), 31 as well as "indirect methods" are the usual approaches to estimate 28Frank Knight; Risk, Uncertainty and Profit (Chicago: University of Chicago Press, 1921), p. 11. 29James Roumasset; "Risk and Choice of Technique for Peasant Agri— culture: The Case of Philippine Rice Farmers," unpublished Ph.D. dissertation, University of Wisconsin, 1973, p. ‘13. See also John Mellor; "Relating Research Allocation to Multiple Goals," in Resource Allocation and Productivity in National and international A ricuTturafl Research: edited byfiThomas Krndt, Uana Dalrymple and Vernon uttan (Minneapolis: University of Minnesota Press, 1977), p. 1191. 30J. Von Newmann and O. Morgenstern; Theory of Games and Economic Behavior (Princeton: Princeton University ress, £17) . For applications in the Brazilian context see J. Dillon and P. Scandizzo, "Risk Attitudes of Subsistence Farmers in Northeast Brazil: A Sampling Approach," American Journal of Agricultural Economics 60 (3), 1978, pp. 1125-1135. Sééalso Celso Crocomo; Risk ETficient Fertilizer Rates: An Application to Corn Production in the Cerrado Region of Brazil," unpublished Ph.D. dissertation, Michigan State University, 1979. 31Eduardo Moscardi and Alain de Janvry; "Attitudes Toward Risk Among Peasants: An Econometric Approach," American Journal of Agricultural Economics 5901), 1977, pp. 710-716. 116 degrees of risk aversion for individual decision makers. The direct method is obviously out of consideration. The indirect methods, despite their intuitive appeal, also required Specific data which could not be obtained in a timely basis for the present study. To circumvent the lack of measurement on risk attitudes, an approach was adopted which allows inferences on the role of safety considerations in enterprise selection to be derived from the results of the model. This approach is utilized by Roumasset in a study of Filipino farmers.32 Details are provided in the next chapter. The outcome from the decision process of direct interest to this analysis is the choice of the farm system. A farm system will be defined as a pairwise combination of any of five major groups of enterprises, ' namely food crops, cash crops, dairy, livestock and off farm work. Therefore, decision makers in each size class will be assumed to select one of the possible choice alternatives, given the observed con- straint levels or attributes of the alternatives. Details on the definition of the relevant farm systems, as well as on the presumed conditioning factors will be presented in Chapter VII. We turn now to the presentation of the quantitative model proposed for the analysis . Enterprise Selection: A Quantitative Model Hardaker has classified approaches to planning for small farm development into two major classes: the statistical and the modeling 32James Roumasset; Rice and Risk: Decision Making Among Low income Farmers (Amstefilam: North Haland, 1976i. 117 approach. Statistical studies are the ones in which production functions are fitted to cross sectional data for subsequent analysis, whereas modeling usually involves the use of mathematical programming or simu- lation techniques. Both approaches, he argues, fit into the category of behavioral or positive farm management studies.33 Choice of alternative enterprises in a farming context has generally been analyzed by means of the modeling approach characterized by Hardaker. Based on specific assumptions as to the nature of the decision makers' objectives, such studies provide allocation solutions consistent with optimization behavior. The degrees of sophistication vary from the simple profit maximization Linear Programming Models” to the recent emphasis on constrained optimization of objective functions, where risk is explicitly accounted for.35 Though the programming approach is appealing (particularly the most recent methods), it has not been free of criticism. As Hardaker indicates: 33‘0p. cit., p. 319. 3“For instance: Judith Heyer; "A Linear Programming Analysis of Constraints on Peasant Farms in Kenya;" Food Research institute Studies 10 (1), 1971, pp. 55-67. Also, Stahis Panagides and Leo Ferreira; "Absorcao da Mao de Obra na Agricultura da Zona da Mata de Minas‘Geraisfl' in Estudos Sobre uma Regiao Agricola: Zona da Mata de Minas Gerais; Monografia No. 9 (Rio de Janeiro: IPEA, 1973) . See also footnote 2 in Chapter V. 35Michael Schluter and Timothy Mount; "Management Objectives of the Peasant Farmer: An Analysis of Risk Aversion in the Choice of Cropping Pattern, Surat District, lndia;" Occasional Paper 78, Depart- ment of Agricultural Economics, Cornell University, 19711; Bryan Schurle and Bernard Erven; "The Tradeoff Between Return and Risk in Farm Enterprise Choice," North Central Journal of Agricultural Economics 1 (1) , 1977, pp. 15-21. See an Hardaker‘s reffi'ences (op. cit.) . 118 Most of (the programming approaches) can be criticized on the grounds that, unlike analyses based on utility maximiza- tion, they lack sound axiomatic foundation (Anderson, Dillon and Hardaker, 1977, Chapter 7). In other words, computational tractability is won by representing the farmer's actual preferences in a particular, arbitrary way.36 Evidently, the selection of a "best" approach to go about analyzing farm enterprise choice should be a function of the type of research question one is attempting to address. if the interest is with the deter- mination of optimal enterprise mixes, given assumptions on constraint levels or decision functions, then the programming approach should suffice. in the case of this study, the concern is with the identifica- tion of factors which influence producers to engage in different kinds of farm systems. A model which enables the association of these factors with the alternative farm system choices should thereby provide an adequate analytical framework. Conventional econometric models are the standard approach to the study of relationships among variables. Continuous dependent variables can be related to any number of continuous or categorical explanatory factors and parameters expressing the relationship between the variables can be easily estimated by ordinary least squares or similar methods. The difficulty in considering enterprise choice in a conventional econometric framework arises from the fact that the dependent variable (choice) is assumed not continuous, but categorical. Other than that, the problem can be treated by the standard approach. The analysis of discrete dependent variables by econometric methods can nonetheless be carried out through proper specification of functional 36Hardaker, op. cit., p. 320. 119 forms and use of special estimation procedures. Therefore, an econometric model in which the alternative enterprise mix choices will be treated as dependent variables, while constraints or choice attributes will constitute the "right hand side" variable is proposed. its mathematical structure and estimation properties are discussed next. For the sake of clarity, we begin with the binary choice case, and then proceed with the generalization to the n—chotomous models. Assume that the ith farmer in a sample of size N has the option to pursue one of two alternative farm systems. Let Yi represent the selection such that Yi=1 if the first system is dwosen and Yi=0 otherwise. Assume further that the M factors influencing the decision can be represented by an array X of size M x N . The relationship between the selection and the explanatory factors in the traditional regression approach would be expressed as - I Yi - Xi B + ei (1) -where B = a M x 1 vector of unknown parameters, and e. = a disturbance term. 1 Under this approach the parameters could be estimated by ordinary least squares (OLS), and the predicted values of Yi may be interpreted as the probabilities that the first choice will be made given the levels of the explanatory variables. This linear model, which is simply a straightforward extension of conventional regression analysis to the discrete dependent variable case, is nonetheless flawed by two major difficulties. 120 On the one hand, a basic assumption of OLS estimation-~that of homoscedasticity--is violated. Since the disturbance term of the model is expressed by ei=Yi—Xi'B (2) and since Yi can assume only the values 0 or 1, we have a distribution of the error terms of the form: -X.'B if Y. = O, with probability 1-P. l I I (3) 1-Xi'8 if Yi=1, with probability Pi it can be shown then that: E(ei2) = (Xi'B) - (I-xi'e) (4) implying that the disturbance terms are heteroscedastic, i.e., the variances are not independent of the values of the X5. Even though OLS parameter estimates under this problem are unbiased, they are no longer efficient. Further, the variance estimates will be themselves biased, and conventional statistical tests will not be valid. On the other hand, even if generalized least squares (GLS) methods are used to deal with the heteroscedasticity problem, we would face another difficulty, namely that the predicted values of Yi would not necessarily be bounded between 0 and 1. Since probabilities have no meaning outside this range, such is a major drawback of the linear model. Moreover, GLS as suggested by Goldberger could imply negative variances if the predicted values of Yi happen to be outside the 0-1 range.37 37A. Goldberger; Econometric Theory (New York: John Wiley and Sons, 19611), Chapter 5. 121 A simple solution to the problems above has been proposed by the imposition of an arbitrary definition of the probabilities estimated by the model such that they forcibly assume the value zero if the pre- dicted value is smaller than 0, or one, if the predicted value is greater than one. The predictions which lie between zero and one are maintained.38 The limitations of this approach are discussed by Domencich and McFadden.39 To avoid the problems with the linear model, econometricians have used alternative specification forms which constrain the probability values within the relevant range. Such specifications include the logit and probit transformations, as well as some trigonometric forms. Logits and probits, in particular, have been used more frequently. The logit model is a simple monotonic transformation of the logistic function Pi=1l1+exP(-Xi'8) (5) whose predicted values (pi) range from 0 to 1 as Xi'B varies from ~00 to +00. Of course, 1-2Pi= 1/1 + exp(Xi'8) (6) Expressions (5) and (6) can be rewritten as Pi/1-Pi = exp Xi'B (7) it follows then that Loge (Pill-Pi) = Xi'B (8) The logit transformation of Pi gives origin to expression (8), which has the necessary features of a consistent probability model. 38Neil Wrigley; "Introduction to the Use of Logit Models in Geography;" Concepts and Techniques in Modern Geography 10, 1976, p. 9. 390p. cit., Chapter II. 122 Alternatively, a probit model can also be specified such that the desirable characteristics are maintained. This model assumes the existence of an unobservable index Ui which is associated with a response (choice) situation. The response is expressed as a linear function of a set of independent variables such that Ui = Xi'B (9) The observed response (Yi=1 or 0) is further assumed to be conditional to an unobserved threshold Ui* which is particular to each individual. The threshold establishes the criteria for expressing the individual behavior: lf Xi'B > Ui* Y.=1 ' (10) If xi's 5 ui* Yi=0 Assuming that Ui* is a standard normal variable (i.e., Ui*'\:N(0,1)), the probit model is eXpressed as X.'B *2 1 f I exp—U2 dU* (11) 4.1T ‘w Pi = Prob(Yi=1) = Hence, Pi is given by the area under the standard normal distribution between -co and Xi'B. The two specifications are virtually equivalent for values of Pi not lll) close to 0 or 1'. As a consequence, the selection between the alterna- tive models for the analysis of dichotomous dependent variables has become somewhat arbitrary. As Wrigley points out: As a result (of the similarity) the choice between logit and probit models as alternative regression models more suited to the utegorized response variable situation is essentially a matter of computational convenience.“ ”This is shown both graphically and through empirical estimations in Eric Hanushek and John Jackson; StatisticaltMethods for Social Scientists (New York: Academic Press, 1977K pp. 188, 206. “Op. cit., p. 11. 123 Although the above remarks can be corroborated by some evidence “2 the in the literature insofar as the dichotomous models are concerned, generalization to the multiple response categories will introduce some a priori grounds to the selection of a particular specification form. In fact, in situations where the dependent variable may assume more than two unordered values the logit model becomes a more adequate alternative. In the case of ordered response categories, a n-chotomous probit specifi- cation lessens the computational burdens by reducing the number of parameters to be estimated. Moreover, the unordered probit model of Hausman and Wise1m has been shown elsewhere to be ". . . very difficult and expensive to estimate, especially for (the number of categories) greater than three.”5 in our case, the relevant choices are unordered. A farmer is assumed to select any of several alternative farm systems, which can only be represented at the nominal level of measurement. Therefore, the logit model is adopted in the discussion which follows. Details on the I”See for instance the applications in Philip Garcia; "Market Linkages of Small Farms: A Study of the Maize Market in Northern Vera Cruz, Mexico;" unpublished Ph.D. dissertation, Cornell University, 1978, Chapter VII. See also Stanley Thompson and Doyle Eiler; "A Multivariate Probit Analysis of Advertising Awareness on Milk Use;" Canadian Journal of Agricultural Economics 23 (1), 1975, pp. 65—73. 113 Hanushek and Jackson, op. cit., p. 211. “Jerry Hausman and David Wise; "A Conditional Probit Model for Qualitative Choice: Discrete Decisions Recognizing Interdependence and Heterogemus Preferences;" Econometrica 146 (2), 1978, pp. 1103-826. usPeter Schmidt and Ann Witte; The Economics of Crime (provisory title) (New York: Academic Press, forthcoming), Chapter 2. ll3 1211 polychotomous probit are provided in McKelvey and Zavoina,“6 as well as in Amemiya.“7 The generalization of the binary choice logit to the multiple response cases is attributed to Theil.“8 The underlying idea of the generalization is that the estimated functions will represent the logarithm of the odds of one selection versus an alternative one. The log odds ratio is then expressed as a linear function of the choice influencing factors. To illustrate, let us consider a situation where three selections are available to a decision maker. The selections Yi=1, 2, or 3 have associated probabilities P“, P2i' P3i' We define P1i as the reference alternative and express the model as “gown/PH) .-. x{821 (‘2) l'°9I=:(P3i"’1i) = xsi'831 “3) I'°9¢==“”3i"’2i) = x{832 (1“) where the B coefficients in (12) measure the effect of the independent variables X on the logarithm of the ratio between the probabilities of selection of alternative 2 and the probability of selection of alternative 1. Expressions (13) and (ill) represent the log odds ratio (P2/P1) and (PB/P2) respectively as functions of X. in fact, the unknown parameters of expression (111) can be derived from those of (12) and (13) . This is so because “Richard McKelvey and William Zavoina; "A Statistical Model for the Analysis of Ordinal Level Dependent Variables;" Journal of Mathematical Sociology (II), 1975, pp. 103-120. “T. Amemiya; "Qualitative Response Models;" Annals of Economic and Social Measurement (ll), 1975, pp. 363—372. “Henry Theil; "A Multinomial Extension of the Linear Logit Model;" International Economic Review 10 (3), 1969, pp. 251-259. 125 Loge(P3i/P2i) =L°ge(P3i/P1i) - Loge(P2i/P1i) and .. I - I _. _ "9 L°9e(P3i/P2i) ‘ xi 831 xi B21 ’ X{”331 821’ Therefore, in a model with "n" choices, the estimation of n—1 equations should provide the basis for all possible comparisons between Pis. Given the n-1 equations, and the requirement that the probabilities add up to unity, the Pis can be uniquely expressed. Thus, _. I . P21- exp X1821 P1 (15) .. I . P3i ' 2"" x1831 P1 (16) ._ . I I 1-P1i (1 +exp xi821 +exp Xi831) (17) - l l PIi-1/(1+expxi821+epriB31) (18) The n-chotomous generalization of the above model as presented by Schmidt and Straussfio is given by P.. l .. I ~ _ . - _ Logepfl—i—Xisj j—2,3,ll,...n,I-1,2,3,...N (19) sUch that the solution for the n probabilities is n P .=1/1+ 2 exp X.'8. 1| .__2 I j l' (20) n - I l _ Pei--eprBe/1+jfzepriBj 9—2,3,...n Unless the independent variables considered in the polychotomous logit model are all categorical, the estimation of the unknown parameters must be done by maximum likelihood methods. Generalized least squares can be applied in the categorical x categorical case, and a discussion on ”9R. Pindyck and D. Rubinfeld; Econometric Models and Economic Forecasts (New York: McGraw-Hill, 1976), p. 258. 50Peter Schmidt and Robert Strauss; "The Prediction of Occupation Using Multiple Logit Models;" international Economic Review 16, 1975, pp. 1171-1186. 126 this specific procedure is provided in Wrigley. 51 For the purposes of this analysis, a maximum likelihood estimation procedure will be adopted, as the choice influencing factors can be both continuous or categorical. The likelihood function is the expression denoting the joint prob- ability distribution of the selected alternatives in the sample. Con- sidering that the observations are assumed to be statistically independent, the likelihood function is the product of the probabilities of selecting each particular alternative. in the three choices case, where N1 individuals have selected alternative 1, N2 selected alternative 2 and the remaining selected the third option, we may write the function as N N+N 1 1 2 N L = Tl P . 11 P . 11 P . (21) ._ 11 ._ 21 . 31 .-—1 I-N1+1 l=N1+N2+1 The maximum likelihood estimates of the unknown parameters (85) will be those values of B which are the "most likely" to have generated the observed data.52 L is maximum at these values of 8. But since maximization of (21) can be very complex, it is usual to maximize the logarithm of L. As such .we have: N1 N1+N2 N L*=LogL=£LogP.+ Z LogP.+ Z LogP. (22) 6 i=1 9 1' i=N +1 9 2' i=N +N +1 9 3' 1 1 2 which is further reduced to 51Neil Wrigley; "Analyzing Multiple Alternative Dependent Variables;" Geo raphic Analyses 7 (ll) , 1975. A maximum likelihood approach for t e same casfisfiescribed in Shelby Haberman.Analysis of Qualitative Data, Vol. 2 (New York: Academic Press, 1979?, Chapter 6. 52Hanushek and Jackson, op. cit., p. 202. The formulations which follow are also drawn from this source. 127 N N * = __ 1 I I L i:1Loge(1+ exp Xi821+exp Xi 831) + if] D1Xi 821+ N I + 321 szi831 (23) where D1 assumes the value one if the second alternative is selected and zero otherwise, while Dz equals 1 if the third alternative is observed, and zero if not. The first order conditions for the maximization of LogeL are I I 8L* "g" xi exPX1321 +§ox~o "' ‘1 I - " 38721 i=1 1 + ”‘9 xi B21+ ex" xi B31 i=1 1 ' (24) I I an __;‘xie"px1831 +§ox'—o — l I ° — 3 B31 i=1 1 + EXP X1821 + exp x1831 i=1 2 ' Generalizing to the n-chotomous case, we will have a set of n-l vector equations in which the (n-1)th will be of the form &* N n N ——=-£ X.‘ exp X.'B /1+ 2 exp X.'B. + 2 D X.‘ = 0 (25) 88m i=1 I I n1 1:2 I jl i=1 n I it can be seen that the number of parameters to be estimated in a logit model with 11 alternative categories and M explanatory variables is given by M(n-l) . If the system above is written explicitly for the individual elements in each of the 85, we would have as many equations as parameters. Hence, the estimation of the unknown parameters is extremely complex, and it is generally accomplished iteratively by means of nonlinear optimization methods. Since it is known that the log like- lihood function of this model is globally concave, the maximum will be 53 reached " . . . from any set of starting values by any reasonable algorithm." 53Schmidt and Witte, 0p. cit., Chapter 2. 128 The estimated 85 will provide a measure of the relation between the corresponding explanatory variables and the probability that individual choices would be observed, rather than the reference alternatives. Statistical inferences on the estimated parameters can be performed through the usual significance tests (t tests) .5" The polychotomous logit model as presented above has followed the usual "normalization" procedure for the estimation of the unknown para- meters. Since in a model with n alternatives only n—l 85 can be uniquely defined, the convention was to make 81:0 and proceed with the estimation. Schmidt and Witte call this the "Theil normalization."55 An alternative n way is to make 2 B. = 0, following Nerlove and Press.56 i=1 Witte call this latter approach the ANOVA normalization, since it is Schmidt and similar to the type adopted in analysis of variance. They show that there is no difference of substance between the two normalizations, and that the parameters of either one can be easily derived with knowledge of the maximum likelihood estimates of the other. Also, the predicted probabilities of occurrence are the same regardless of the selected normalization. There- fore, these authors contend that the choice of which convention to adopt is basically a matter of the perceived ease of interpretation of coefficients. While the 85 of the Theil normalization express the effects of X on the log 5“For a discussion of the statistical properties of the parameters of the multinomial logit model, see Daniel McFadden; "Conditional Logit Analysis of Qualitative Choice Behavior;" in Frontiers inéconometrics, edited by Paul Zarembka (New York: Acadefic Press, 19713 . 55Op. cit., Chapter 2. 56Marc Nerlove and 5. James Press; "Univariate and Multivariate Log-Linear and Logistic Models;" R—1306-EDA/NIH (Santa Monica: The Rand Corporation, 1973) . 129 odds ratios, those of Nerlove and Press simply denote the effects of X on each P]. It is then often simpler to interpret the parameters of the latter specification, particularly when the number of alternatives is large. The estimation program used in this analysis was written in accordance with Theil's normalization. Nonetheless, the inclusion of an additional routine allowed the derivation of the alternative parameters, such that both results will be presented. Details on the relationship between the parameters of both normalizations can be found in Schmidt and Witte.57 Elma This chapter has presented the analytical framework for the proposed study of farm enterprise selection. Based on the major theories of decision making in traditional agriculture, a conceptualization of the processes whereby important choices in the farming environment are arrived at was developed. This in turn should provide a basis for the inclusion of explanatory variables in a quantitative model of enterprise choice. The polychotomous logit model is pr0posed as an adequate approach to identify the nature and relevance of factors affecting the selection of farm systems. Based on the postulates of rationality and utility maximization, this model draws from observed behavioral patterns to estimate parameters expressing the relationship between the probabilities of selection of alternative systems and a number of choice influencing factors. Its mathematical structure was examined in some detail. 57Op. cit., Chapter 2. 130 With knowledge of the relevant parameters, the model can be used to test the effects of different policy interventions on the outcome of the process. Thus, it is believed that important insights on specific characteristics of decision making in the study region can be gained from its estimation. Chapter VII will present a discussion on the relevant choices, explanatory variables, model estimation and results. CHAPTER VII LOGIT ANALYSIS OF ENTERPRISE CHOICE Introduction The analysis performed in this chapter is motivated by a need to gain increased knowledge about the nature and relevance of variables affecting the selection of farm enterprise combinations in the Zona da Mata region. its results can be a valuable component of future develop- ment efforts for the area, particularly in view of the virtual nonexistence of empirical evidence on farmer decision making processes. The analysis follows the framework developed in Chapter VI, whereby a polychotomous logit Specification was proposed as a suitable model to examine the relationship between a group of explanatory factors and the probabilities of selection of alternative farm systems. It begins with a discussion of the methodological aspects of farm classification, followed by a description of the independent variables considered. The model structure is then formally presented and the results of the estimations for each farm class are examined. In keeping with the objec- tives of the study, several scenarios are evaluated in light of the model results. The predicted probabilities of participation into the alternative farm systems, given the different scenarios, are used as a basis for the assessment of the impacts of alternative strategies on farmer partici- pation into target farm types. 131 132 Defining the Relevant Enterprise Combinations As indicated in Chapter VI, the dependent variable of the proposed logit model is defined only at the nominal level of measurement. It will simply indicate whether or not a given farm belongs to a particular category, in many ways paralleling the definition of a "dummy" variable in conventional econometric models. The categories are defined as the alternative farm types (farm systems) observed to exist in the study region. Classification of farms in accordance with the nature of the activities performed is an old methodological challenge to agricultural economists. The criteria adopted in farm type classification systems in several North Central states of the U.S., for instance, are so varied that a set of farms classified under each criteria would produce widely divergent groupings.1 This is a good illustration of the difficulties inherent to the establishment of a meaningful classification system. These problems notwithstanding, elements from the Missouri system2 and from the rules adopted by Franzel3 in his classification of enterprise combinations are utilized to derive the criteria for this analysis. The methodology selected to classify the sample farms consists of an initial pairwise combination of the fifteen activities being considered, in terms of labor inputs absorbed. By eliminating those pairs which are 1Don Pretzer and Robert Finley; "Farm Type Classification Systems: Another Look at an Old Problem;" American Journal of igicultural Economics 56 (1), 1979, pp. 1115-1119. 2ibid., p. 1215. 3Op. cit. 133 known a priori not to be present in the sample (e.g., sole beans and beans in both pure and intercropped stands), a total of 85 possible combinations resulted. Since it is possible that more than one pair of activities be present on a given farm, farms were characterized by that pair which absorbs the most labor from the total used. If present, ties would be broken by the net income contribution criteria utilized earlier in this study. Frequency distributions of the sample farms by the possible combinations were then constructed for each farm class. The next step in the classification process was to group each of the fifteen original activities into five major classes, namely food crops, cash crops, dairy, livestock and off farm work. Food crops are defined as those which are known to be grown primarily for the fulfillment of on-farm consumption needs. From the evidence of this research, and based on reported evidence in the literature, corn, beans and rice are considered as the relevant food crops.” The remaining are considered as cash crops. Results of the frequency distribution constructed in the initial step were then reclassified into pairs of these five major classes. The resulting frequency distributions are shown in Table 7.1. Table 7.1 shows that some farm types are clearly more frequent in each of the size classes. The five most frequent types in the 0-10 class encompass 77 percent of farms in this class, whereas the same figure for the 10—50 class is 75 percent. For the 50-100 class, four farm types are responsible for 86 percent of the observations. Hence, the categories considered in the ensuing analysis will be defined by these most frequent ”Works referring to these enterprises as food crops in Zona da Mata include Singh, et al. (Op. cit.) and the evaluation reports of the PRODEMATA project. Enterprise Combinations by Farm Size Classes 134 Table 7.1 O- 1 O Hectares 10—50 Hectares 50-100 Hectares Farm Type Observations % Observations % Observations % Food only 17 12.2 21 7.9 3 11.2 Food and cash 18 12. 9 66 211. 8 13 18.1 Food and dairy 6 ll. 3 37 13. 9 19 26. ll Food and livestock 6 11.3 12 l1.5 3 9.2 Food and off farm work 37 26. 6 26 9. 8 -- -- Cash only 19 13.7 35 13.2 19 26.11 Cash and dairy 16 11.5 23 8.6 2 2.8 Cash and livestock 1 . 7 1 . Ll -- -— Cash and off farm work 5 3.6 35 13.2 11 15.3 Dairy and off farm work 10 7. 2 6 2. 3 -- -— Livestock and off farm work ii 2.9 1 .II 1 1.11 Dairy and livestock -- -- 3 1.1 1 1. ll TOTAL 139 100.0 266 100.0 '72 100. 0 Source: PRODEMATA sample survey 135 farm types. The relevant categories, as well as the respective number of observations in the sample are summarized in Table 7.2. As with any classification norm, the one selected here may be viewed as excessively reliant upon just a few of many characteristics of a farm type. Yet, the selected standard is a realistic alternative in view of the information available and research objectives. As concisely expressed by Wharton, reduction is a necessity, if insights into the nature of any complex system are to be gained. His comments are noteworthy. While I readily applaud any stricture against univariate classification which masks the fundamental heterogeneity of any universe, one should be equally prepared to eschew the counsel of no classification at all.5 Besides encompassing the majority of observations in the sample, the selected number of farm types is very convenient, particularly in view of the complexity involved in the estimation of logit models with a large number of alternative values for the dependent variable. Since the number of parameters to be estimated by the maximum likelihood procedure is given by the number of independent variables multiplied by the difference between the number of alternatives and one, it is easy to see how important it is to keep the number of alternatives at a manageable level. Moreover, a large number of alternatives would probably result in too few observations for at least some of the categories, unless of course the distribution is relatively uniform, the categories being evenly represented in the sample. By defining these groups such that the majority of the farms in the sample can be classified into a small number of categories, an adequate number of observations per alternative is ensured. A further reason for keeping the number of alternatives at a 5Clifton Wharton, op. cit., p. 6. 136 low level resides in the difficulty of interpreting the results of the model when the dependent variable assumes several values. Under Theil's normalization the number of equations representing a pairwise comparison in the model is given by (n) , where n is the number of alternative values of the dependent variable. With five alternatives, we have ten equations; with six, the number of equations grows to fifteen and so on. Hence, for the sake of tractability in the estimation and interpretation of the model, the criteria established seems appmpriate. Explanatory Variables The diagram presented in Chapter VI provides the basis for the selection of variables hypothesized to affect the farm system selection process. These are defined below. Distance (DIST) This variable measures the distance from the farm site to the central area of the respective municipalities. It is included as a proxy for the relative ease of input/output marketing. It is hypothesized that higher transaction costs are associated with larger distances to the municipalities. Given the generally low quality of the transportation infrastructure, it is likely that farmers would take such aSpects into consideration on making decisions about enterprise mixes. Moreover, the variable may be indicative of one's relative ability to secure work in the non—agricul- tural sector. The unit of measurement is kilometers. Location (R2 and R3) Dummy variables are included in the model to represent the sample regions to which the farms belong. They are intended to account for possible variations in farm types attributable to regional specialization. 137 Table 7.2 Most Relevant Farm Types and Respective Numbers of Sample Observations by Size Classes Observations Farm Type 0-10 Ha 10-50 Ha 50—100 Ha Food Only 17 -- -- Food and Cash 18 66 13 Food and Dairy -- 37 19 Food and Off Farm Work ’37 26 -- Cash Only 19 35 19 Cash and Dairy 16 -- -- Cash and Off Farm Work -- 35 11 TOTAL 107 199 62 138 Since the sample area is divided into three regions, two dummy variables are defined. R2 assumes the value 1 if the farm is located in the region of MurIae and 0 otherwise, while R3 equals 1 if the farm is in Vicosa and 0 if not. The region of reference is Juiz de Fora. Although the study area is relatively homogenous in terms of agronomic characteristics, there are some differences in other aspects among the three sample regions which may influence the processes under analysis. The region of Juiz de Fora, for instance, has a privileged location in relation to the major metropolitan area of Rio de Janeiro. ' In fact, most of the dairy products consumed in the greater Rio area originate in Juiz de Fora. Moreover, its main municipality—~Juiz de Fora--is in itself an important center for light industries. The region of Muriae, on the other hand, is located more to the northeast of Zona da Mata, and its economy is primarily based on a diversified agriculture, where coffee and dairy are the most relevant activities. The region is crossed by the major highway linking northeastern and southeastern Brazil, and it has good access to important demand centers. Finally, the region of Vicosa--Iocated towards the center part of Zona da Mata-~is also characterized by a diversified agriculture. Traditional crops, such as corn and beans, as well as sugar cane and tobacco are among the main agricultural activities. Apart from sugar processing, there are no major industries, and access to demand centers is perhaps the most limited of the three sample regions. Sensitivity Index (SI) The sensitivity index (a variation of Roumasset's risk sensitivity indexe) is a measure of the farmer's ability to withstand financial losses 6Op. cit., 1973, p. 1113. 139 before he needs to dispose of now-liquid assets to provide for minimum household needs. It is defined as the per hectare difference between a farm's amount of liquid assets and the minimum level of income necessary to meet household needs in a given crop year. Specifically, we have: SI = (LA - M)/PL where SI = sensitivity index LA = amount of liquid assets (assets convertible into cash within the accounting period, such as stocks or livestock) M = minimum level required to meet household expenditures, defined as the per capita regional minimum wage (in a family of four) times the number of household members in the respective farm unit PL = hectares of productive farm land Although the index should not be regarded as a measure of risk preferences, it will provide indications of the role played by risk con— siderations on enterprise choice decisions. Details on the theoretical aspects on this approach are indicated by Roumasset. 7 Capital/Hectare (K/Ha) In theory, we should expect a more relevant role for this variable in farm systems which require larger amounts of capital investments, such as dairy and some types of livestock production. Though agriculture in Zona da Mata is generally not intensive in terms of capital inputs, the inclusion of the variable in the model will allow the ascertainment of the 7lbid. 1110 extent to which the selection of enterprises as such is influenced by this constraint. The farm stock of capital items includes buildings, machinery, implements and assorted farm equipment. The unit of measurement is thousands of cruzeiros per hectare of productive land. Labor/Hectare (L/Ha) The rationale for including this variable is straightforward. In conditions where the labor input is abundant, it should be expected that labor intensive farm enterprises-—or off farm work--will be preferred. The opposite should be true if labor is a limiting resource. in view of the observations in previous chapters, whereby an excess labor supply was identified for the smaller farms in the sample, it should be of particular interest to study the impact of these conditions in enterprise selection decisions. The variable encompasses the total amount of labor available to the farm unit, including both family and hired manpower. It is expressed in hundreds of man days equivalent per hectare of productive land. Credit/Hectare (C/Ha) Investment and working capital credit obtained in the crop year of reference are aggregated into the credit variable. As a proxy for access to inputs in general, the variable can be indicative of the potential impact of the PRODEMATA project in altering patterns of enterprise emphasis in Zona da Mata. Moreover, its inclusion should allow inferences on the potential of credit as a policy tool for the promotion of emphasis in target farm systems. The variable is measured in thousands of cruzeiros per hectare of productive farm land. 1111 FamilLS ize Size of the household unit is included to account for the plausible effects of family demands in the selection decisions considered. These demands are known to differ in accordance with the age composition of the family unit, its size in relation to availability of land resources, and other factors which are unique to each farm family. Decisions on the balance between food and cash activities are particularly likely to be influenced by such factors. The variable will include all persons living in the same household unit. Model Structure Once both the dependent and independent variables considered in the analysis have been defined, the proposed model can be formally presented. its structural form in accordance with Theil's normalization is as follows: 0-10 Hectares and 10—50 Hectares LogelP2/P1) = 811+512R2 +313R3 + B1uSl+ BISKIHa + 316L/Ha +817C/HA + BIBF + B1QDIST Loge(P3/P.1) = 32] +322R2 +323R3 + 8245' + 825K lHa +326L/Ha +827C/Ha +528F + BZSDIST Loge(Pan1) = 831 + B32R2 + 333R3 + 83115] + B35K/Ha + B36L/Ha + B37C IHa + 838': + B3QDIST Loge(P5/P1) = Bl.” + BMRZ + Bu3R3 + 8M5] + 8115K /Ha + anL/Ha + BwC/Ha + 8118': + BugDIST 1112 where the codes of the dependent variables denote the respective farm types, as specified below: Type 1 = Food Only (0-10 Hectares), or Food and Cash (IO-50 Hectares) Type 2 = Food and Cash (0-10 Hectares), or Food and Dairy (10—50 Hectares) Type 3 = Food and Off Farm Work, for both classes Type II = Cash Only, for both classes Type 5 = Cash and Dairy (0-10 Hectares), or Cash and Off Farm Work (IO-50 Hectares) Six other equations are derivable from the above for the remaining comparisons among the probabilities. As shown in the previous chapter, we have: Log(P3/P2) Log(P3/P1) — Log(P2/P1). Thus: LOQIPBIPZ) = (B 21- B“) + (822- B12)R2 + (823- 813lR3 + (8211- 61,151 + ( 825-8151K/Ha + (826- BmiL/Ha + (827- SWIG/Ha Similar derivations are made for all the other comparisons. 50-100 Hectares The model structure is basically the same as above, the only difference being the number of alternatives (categories), which is four for this group. Thus, we will have the same equations for (P2/P1) to (Pu/P1) as formerly presented, plus two other comparisons derived from the estimated coefficients (Pu/P3 and P3/P2). The codes 1 to £1 will denote respectively the farm types "food and cash," "food and dairy," "cash only," and "food and off farm work." 1113 Model Estimation and Results The maximum likelihood estimates of the unknown parameters of the logit models above were computed following a numerical maximization algorithm developed by Marquardt.8 This procedure combines the method of Gauss with the method of steepest descent. A FORTRAN program written by Professor Peter Schmidt of the MSU Economics Department was used for the estimation of the parameters of the model under Theil's normalization. The program provides the parameter estimates and their asymptotic t ratios, which are then used as inputs by another routine to compute the parameters of the ANOVA normalization, their t ratios and the chi-squared statistics for each independent variable considered. The estimated parameters and associated statistics are discussed next. 0-10 Hectares The initial estimation of the model for the 0-10 hectare class included the eight explanatory variables presented earlier. Table 7.3 gives the results for the Theil normalization. To illustrate the interpretation of the coefficients, the numbers under the first column (P2/P1) are the parameters of the equation expressing the logarithm of the ratio of the probability that a farmer would belong to the second farm group (food and cash) to the probability of participation in the first group (food only). The positive sign of R2, for instance, denotes the fact that P2 increases relative to P1 given that the farm is in Muriae. The figures 8Donald Marquardt; "An Algorithm for Least Squares Estimation of Nonlinear Parameters;" Journal of the Society for Industrial and Applied Mathematics 2 (2), 1963. illl-l .CEQ Ucm :mmU n m “3:0 :30 u 3 32oz Egan. to ccm coo... u m EmmU Dcm coon. .I. N “Eco coop. n — 09»... Econ. a... $5. 2..-. .3... 8m. 5:. E... E... 5... 8e... 5.. 25 3:. 2:..- Sflr £3. 22..- .Re. RS. 2...... 3.... mm... Sean. NE. a”. 8.... 5.. E... 8.... Sr. 5.: 5.3 5.... 5.. $2.. ES. 22. .58. 3:..- memf 2m... 22... 8:. men... oonocu ~e.~_ .m~.. Ame.-. ANN.-. .em.. .em.. Amm.. .me.m. .me.m. ._~.N. A_e.. 2:... 2.2..- e:~.- £2... 28. Se... is... $3.. 8:... $5. extoofi t.— .2. 8m. .8.-. 2:... a... 5... an. 8.... :~.-. 5... 82.. $8. 28.- 38. 88. 83. £8. £2..- oeeef 22..- o::o=oou :3 5.. 2... 8m.-. .3... $5.. 2:... 8m... 8.... SN... .2... xouc. 22. 32.. 28.- :2. :3. 8:. 22. M.S.. $2. 3.2.. 3.322% m=.~ .he._-. Ame.-. Amo.. .ma.-. .Ne.-. ..e.-. .em._-. Ame.-. Ae._-. A_s.-. 52.. SS.- 22.. m3..- meeer 2:..- Emef 2:..- mmsr 3:..- 8535 3.2 an... E... :5... 5.7. 5.7. 2....-. E... 5r. 5.. .22. am. 9.23 82. 82.7 25....- meee.m- £3:- fimm; ~25- 82. 226 $85 3.... 8:... 3e... .3... 6.... 5.7. .8... ca... .8... E... 5.3 5.. See. 824 Rid min..- RS..- $.35- ~e-.~ 2:3 93. 8%.... oozes. Ace.-. .mm.-. ..e.-. Ame.. .e~.~. ..m... .mo._-. AMN..-. a~_.-. .mm.-. 3.- 28.7 32..- ..eme; .33.. 22..“ 22.? £3..- 22.- Sema- acoamcou Nx some men... men... None 3:... Nwe}... {me (a... {.mc .29. spot; 532.3582 9:9: mOLmuUOI G — 10 538.28% ...Ec. 9: ..ooc: cocoéaw .ooos. of co 3.38m N..N. 03c... 1115 in parentheses are the t ratios, which are asymptotically distributed as N(0,1) under the null hypothesis that the respective coefficients are zero.9 The chi-squared column shows the values of this statistic for eaCh of the independent variables of the model. These are used to test the joint significance of the coefficients, under the null hypothesis that the variable tested has no effect on any of the probabilities. ‘0 The results of the ANOVA normalization are presented in Table 7.11. The parameters in the first column, for example, express the effect of each independent variable on the probability of belonging to the "food only" category. The negative coefficient of credit/hectare indicates that this probability decreases with increases in the level of credit/hectare. Figures in parentheses are also asymptotic t ratios, and the chi-squared statistics are presented in the far right column. Unlike the t statistics, the values of X2 are the same for both normalizations. A From an inspection of the computed X2 values, it can be seen that not all variables initially considered have an impact on the farm system selection decision. The critical values for X2 with four degrees of freedom are 9.119 at the five percent confidence level and 7.78 at the ten percent level. Thus, at these usual levels, the hypothesis of no effect on the probabilities is rejected only for the location variables, 9Ann Witte and Peter Schmidt; "An Analysis of the Type of Criminal Activity Using the Logit Model;" Journal of Research in Crime and Delinquency (1), 1979, p. 169. 10This is known as the "Wald test" or "W-test." For details on the procedure see G.S. Maddala, Econometrics (New York: McGraw-Hill, 1977), p. 180. See also 5.0. Silvey, Statistical Inference (London: Chapman and Hall, 1973), pp. 115-118. Results of the Model Estimation Under 1'46 Table 7.11 The Initial Specification 0-10 Hectares ANOVA Normalization Variable Food Food 6 Food 8 Cash Cash 8 )(2 Only Cash Off Farm Only Dairy Constant 1.11898 -1.8932 1.2189 -.0128 —.8027 (1.118) (-2.52) (.77) [-.02) (-.96) Muriae -1.6077 1.9131 -1.S333 .6090 .6186 (—1.70) (2.110) [—.97) (.96) (.79) 8.1114 Vicosa -.8507 2.11116 -.0209 -1.5769 .0870 (-.8‘l) (3.17) (-.30) (-2.ll9) (.1111) 23.50 Distance .0159 .0053 -.0080 .00119 -.0181 (.98) (.35) (-.60) (.111) (-1.19) 2.48 Si -.0979 -.0519 .0583 .0198 .0718 (-1.13) (-.88) (.92) (.30) (.93) 5.01 K/Ha .0088 -.0228 .0002 -.0026 .0163 (.52) (-.73) (.01) (-.1l-l) (.87) 1.17 Labor/Ha -1. 30111 -. 11253 . 67119 . 11572 . 5973 (-2.70) (-.55) (1.10) (1.211) (1.36) 12.62 Credit/Ha -.3209 .1136 .0191 .0711 .1170 (-2.12) (.26) (.11) (.40) (.77) 8.72 Family - .1035 .1006 . 0822 —. 0525 . 0637 Size (-.90) (.09) (.86) (-.62) (.57) 1.811 147 for credit/hectare, and for labor per hectare. Distance, the sensitivity index, capital/hectare and the size of the family have no significant influence on the farm system selection decision for this particular farm class. These variables are therefore eliminated from the analysis and the model is re-estimated with the remaining ones. Results are shown in Tables 7.5 and 7.6 for the two normalizations. The results of the reduced model indicate that the farm location is an important factor affecting the probability of participation in the farm systems considered. Farmers in the Muriae area are more likely to be characterized by farm types involving cash crOps than their counterparts in the reference area of Juiz de Fora. The odds of "cash only" or "cash and dairy" relative to "food only" farms increase, given the fact that the location is in the Muriae region. Also, indicative are the coefficients of R2 in the ANOVA normalization, showing a positive and highly signifi- cant association with the probability of "food and cash," and a negative association with P (Food Only), though at a lower significance level. It can further be seen from Table 7.5 that relative to "food and cash" enterprise combinations, both the odds of "cash only" and "cash and dairy" farms are inversely associated with R2. Thus, compared to Juiz de Fora, farmers in this region are most likely to be in the second group (Food and Cash),md least likely to belong to the "Food Only" category. These findings are consistent with expectations, inasmuch as they largely reflect the concentration of coffee farms in the sample region of Muriae. Some 55 percent of the farmers sampled in that area were found to attribute importance to coffee growing, in accordance with the criteria of Chapter V (Appendix C). Since it is not entirely unusual 1118 Table 7.5 Results of the Model Estimation Under The Reduced Specification 0— 1 0 Hectares Theil Normalization Log Odds Constant Muriae Vicosa Labor/Ha Credit/ Ha Ratio (R2) (R3) P2/P1 —6.7467 6.7089 6.4715 1.1555 .7397 (-5.62) (5.47) (5.20) (1.17) (.79) P3/P1 -.6078 .2959 .5292 1.8540 .6866 (-.07) (.03) (.06) (2.67) (3.78) Pu/P1 -2.0823 2.3313 -.6442 1.6876 .7141 (-1.94) (2.16) (-.56) (4.29) (4.11) Ps/P1 -2.8165 -2.4939 1.3889 1.7632 .8145 [-2.27) (2.02) (.86) (4.04) (5.30) P3/P2 6.1389 46.9129 -5.9423 .6985 -.0531 (.71) (-.74) (-.69) (.65) (-.06) Pu/PZ 4.6644 -4.3775 -7.1157 .5321 -.0256 (6.32) (-5.16) (-9.59) (.56) (-.03) P5/P2 3.9302 -4.2149 -5.0825 .6077 .0748 (3.95) (-4.10) (-3.74) (.62) (.08) Pu/P3 -1.4744 2.0354 -1.1734 -1.6640 .0275 (-.17) (.23) (-.14) (-.25) (.14) P5/P3 -2.2087 2.1980 .8597 -.0908 .1279 (-.25) (.25) (.10) (-.13) (.70) PSIPu . -.7342 .1626 2.033 .0756 .1037 (-.87) (.19) (1.59) (.21) (.58) x2 38.35 93.63 22.46 33.99 Farm Type 1 = Food Only Type 2 = Food/Cash Type 3 = Food/Off Farm Work Type II = Cash Only Type 5 = Cash/Dairy 189 Table 7.6 Results of the Model Estimation Under The Reduced Specification 0—10 Hectares ANOVA Normalization Variable Food Food 6 Food 5 Cash Cash 6 )(2 Only Cash Off Farm Only Dairy Constant 2 . 11506 -II. 2960 1. 81129 . 3683 - . 3659 (1.28) (-2.35) (.27) (.21) (-.20) Muriae -2.3660 11.31128 -2.0700 -.0346 .1279 (R2) (-1.28) (2.36) (-.30) (-.02) (.07) 38.35 Vicosa -1.5491 11.92211 -1.0199 -2.1933 -.1602 (R3) [—.79) (2.69) (-.15) (-1.29) [-.08) 93.63 Labor/Ha -1.2921 -.1366 .5619 .3955 .4711 (-3.63) (-.19) (1.08) (1.34) (1.38) 22.116 Credit/Ha -.5910 .11187 .0956 .1231 .2235 (-2.79) (.20) (.43) (.56) (1.06) 33.99 150 to intercrOp coffee with any of the food crops considered,11 "food and cash" farm types are likely to be found in this particular region, especially among small scale growers such as the ones in the 0—10 hectare size class. This same farm type (Food and Cash) also has its probability of selection positively associated with the dummy variable for the region of Vicosa. But unlike the former region, we do not identify a higher propensity to observe emphasis in cash crops in general, relative to the standard of comparison (Juiz de Fora). Though the signs of both "cash only" and "cash and dairy" compared to "food and cash" in Table 7.5 are negative as in the former case, the comparisons with "food only" have all insignificant coefficients. Thus, the only inference that can be properly made for farms located in Vicosa is that mixes involving food and cash crops are more likely to be observed there than in the sample region of Juiz de Fora. Such results are also plausible, since the Vicosa region is characterized in the sample by a lower emphasis on commercially oriented enterprises than for any of the other regions, at least for this size class. Yet, we still find some degree of emphasis in a few of the activities here characterized as cash crops, such as coffee, tobacco or sugar cane. Higher levels of availability of labor per hectare are found to increase the probabilities of participation in both "cash only" and "cash and dairy" farm types, though the significance of the respective coefficients is low (20 percent). Conversely, the likelihood of "food only" farms being observed has a highly significant inverse relation to labor/hectare. In fact, the coefficients of Table 7.5 show that the odds of participation in 11Carlos Mells and Cicero da Silva; "Culturas lntercalares;" Informe Agropecuario ii (111)), 1978, pp. 70-71. 151 all of the farm types relative to the "food only" group increase with more availability of labor per hectare. With the exception of the comparison "food and cash" to "food only," the coefficients for the comparisons are highly significant. This indicates that lower levels of labor input per hectare tend to be associated with farms which are characterized by an emphasis on food crops. Thus, greater availability of labor/hectare makes it more likely for a farmer to belong to the alternative farm groups. However, it is not clear from the model results which specific farm type would have a greater probability of observation with increases in the amount of labor. These findings suggest that most producers in this size range are not bound by labor constraints to meet food requirements. As observed earlier (Table 5.3) , food crops absorb no more than 38 percent of the total labor availability for the group. Hence, as food demands are met, the surplus labor is employed into alternative activities, of which off farm work appeared to be the most important. Moreover, the higher levels of labor/hectare observed for the more market oriented activities reflect the utilization of hired manpower to supplement family labor during periods of peak demand. It is unlikely that farmers would hire any additionallabor on farms with stronger subsistence orientation. Given these observations, the estimated relationships between availability of labor/hectare and the probabilities of participation are justifiable. Credit per hectare has shown effects similar in essence to those of labor/hectare. Specifically, a negative association of this variable with the probability of "food only" farms was observed, at a high level of significance (Table 7.6) . Also, relative to this farm type the odds of 152 "food and off farm work," "cash only" and "cash and dairy" mixes increase, given increases in credit/hectare. These results are hardly surprising, given that credit should provide a farmer with flexibility to pursue more lucrative enterprise mixes. It is clear then that we should observe a movement away from the more traditional farm mixes as credit/hectare becomes more available. But one intriguing aspect related to this variable is the increase in the odds of "food and off farm work" farms relative to the "food only" group. Unless one hypothesizes that increases in credit are associated with improvements in the produc- tivity of labor in growing food crops (with an associated reduction in labor requirements for these activities), more emphasis in off farm work should not follow increases in credit/hectare. Considering that off farm work is a high income activity for this group of growers, the above hypothesis may well be a correct one. To summarize the results of the model, it can be said that mixes in which food crops are relatively more important are more likely to be emphasized by a farmer who is not located in Muriae or Juiz de Fora, and who has lower levels of both labor/hectare or credit/hectare. Con- versely, mixes where cash crops, dairy and off farm work gain prominence have a higherprobability of being observed among farmers who are not in Vicosa and who have higher levels of both labor and credit/hectare. The average farmer in the sample, that is, one for which the values of the explanatory variables equal the sample averages, will have probabilities of 11 percent, ten percent, 111) percent, 416 percent and 19 percent of belonging respectively to each of the five farm classes under consideration. Further results on these predicted probabilities of participation are discussed later in this chapter. 153 10-50 Hectares The initial estimation of the model for this size class also included all of the eight independent variables presented earlier. The results are shown in Tables 7.7 and 7.8 respectively for the normalization of Theil, and Nerlove and Press (ANOVA) . As in the former analysis, not all of the variables initially taken into account were found to influence the probabilities of participation in the farm types evaluated. Recalling that the critical values of the chi-squared statistic with four degrees of freedom are 9.119 at the five percent level of confidence and 7.78 at the ten percent level, it can be seen that the variables distance, capital/hectare, and the dummy variable for Vicosa do not exert any significant influence on the probabilities of participation. The former two variables were dropped from subsequent estimations. R3 is kept in the model to allow the breakdown of predicted probabilities by all of the three regions con- sidered. The results of the new model estimation are presented in Tables 7.9 and 7.10. In considering the effects of the location variables, we observe that these are less pronounced than for the former farm size group. In fact, the hypothesis of no effect is rejected only at the 20 percent level for both R2 and R3. Relative to Juiz de Fora, the farms located in Muriae have a higher probability of emphasizing "cash only" systems, and a lower probability of being in either "food and dairy" or "food and off farm work" groups. But the association is only significant at the 20 percent level for the latter two farm types. The coefficients of Table 7.9 reinforce this pattern. Only the odds of either "food and dairy" or "food and off farm work" relative to "cash only" show a significant association with R2, and the indication is that the latter group has its odds of 1511 x33 Egon. to DE... :30 n m “Eco ammo n a 3.3.5 Econ. to new coon. n m “EEG .ocm coon. -I. N .meU can noon. 0 p out. 8.5.... :3... 8.... ...N. 8.... 8.3-. 5..-. 5.7. 8.... 3.... ..~.. 3...: 62m 3:..- om.e. m8... Ea.- mme..- 82.- 2.3. 3:. :N... 8.3. 3.56”. mm.- ANN..-. Am..-. ..m.. .mo.~. Aem.~. Ame... .mo.-. .mn.-. Ae:.-. Am..s-. mamaf mace.) Ncmm. 93... came. man... NMMh.I mmm~.1 hmwmf gum—.7 NIEBGLU e=.e. .em..-. .m~.-. .e...-. .mm.. ....~. .se.~. .am..-. .h~.-. .aw.. .ah..-. :3..- EN...- 8:1..- SS. 323 M.S.; .38....- 2....- 22; .33...- 2.236.. m~.~ .me.-. .me... .sm.. .em.. ..~.. .mm.-. ..m.~. .om.. .m~.-. .5... £2..- 22.. so... is. 3.... 2......- mmme. Ea. £8.- 38. 6:3. mm.mm ..~.~. ....m. .Ne.. .me..-. ..s.m-. .ee.=-. .e~.e. .ee... .em.. .sm.m. eemm. mNee. .eea. see..- ~.e..- meem.- peep. eeNN. e.ee. ~.me.. _m mad EMT—I. 8.va 33.1. So.~1. :N.I. 3N; Ammf. 3...... 3a.: 73.. hmmcf as...) Nmocf awn...) Each 32.. 2.2.... oN—c. 5:... .32.. oucmufla s... .....-. G»... .2... 8... .mo... .3... ..m... .8. 37.-.73..-. .2: .52..- emmmr Sch. 8:. 32.. 8mm. 29....- oer 2.3..- ~$R7 88; .m... 8e.-. an... 3.5 :3... 3...: 8... .3.-. 8.... .2... 5..-. .3. ~85.- BZ. 8:... 3m... 8m... 32: 2:..- e.$. 2.3..- ..S....- 68.5.2 ..o... .m~.. .ea.-. .m.... .5..-. .mm.. .e~.. ..m..-. .mm.-. .me.-. 38.. 8.5. 8......- emeN. mefir 8mm. «3.. 2m...- Sam.- 32..- 2.36:5 N E o E o E o E e E o E o E c E c E e .ENe 656...; 5.32352 9:65 8.28: em .3 836533 aces. .oEc. 6.: .6 3.36m h .5 03m... Results of the Initial Model Estimation 10- 50 Hectares 155 Table 7.8 ANOVA Normalization Variable Food 8 Food 6 Food 8 Cash Cash 8 x2 Cash Dairy Off Farm Only Off Farm Constant . 5573 - . 3505 . 0091 -1. 0709 . 8551 (.84) (-.60) (.01) (-1.57) (1.13) Muriae . 3664 -. 6956 - . 9225 . 9574 . 2942 (.68) (-1.34) [-1.91) (1.76) (.45) 8.31 Vicosa 1.0151 -.7541 -.2235 .5415 -.5791 (1.67) (-1.42) (-.42) (.98) (-.84) 5.81 Distance .0062 .0105 .0156 .0064 -.0263 (-.44) (.94) (1.15) (4144) (-2.08) 5.98 51 - . 4376 . 6136 - . 3756 -. I676 . 3672 (~3.68) (5.21) (-2.98) (-.97) (3.02) 39.93 K/Ha -.0287 .0093 -.0362 .0289 .0267 (-1.29) (.27) (-1.34) (.56) (1.19) 7.29 Labor/Ha 1.2378 -3..2727 3.9106 .4908 -2.3665 ~ (.66) (-2.75) (1.95) (.33) (—1.77) 10.48 Credit/Ha .5747 -.5883 -.1090 .2811 -.1585 (3.04) (-3071) .(“.‘i8) (1.02) (-095) 22089 Family -.0929 .1951 -.0718 .0218 -.0522 Size (-1.36) (3.54) (-1.11) (.29) (-.89) 13.41 156 Table 7. 9 Results of the Model Estimation Under the Reduced Specification 10—50 Hectares Theil Normalization Lo? Constant Muriae Vicosa SI LiHa C lHa Family 0 95 (82) (R3) Size PZIP1 -.5099 -1.0419 -1.8418 1.0741 -4.1373 -1.1320 .2907 (-.54) (-l.30) (-2.09) (5.76) (-1.77) (-4.19) (2.98) P3/P‘ -.2972 -1.0817 -1.1353 .0695 2.4672 -.6747 .0351 (-.33) (-1.53) (-1.37) (.44) (.37) (-2.47) (.35) P4/P1 -1.2750 .5154 -.5997 .3245 -.C332 -.2636 .1271 (-1.27) (.63) (-.66) (1.36) (-.01) (-.66) (1.09) PSIPI .2143 -.3131 -1.7149 .8423 -3.0093 -.6648 .0414 (.20) (-.34) (-1.67) (4.57) {-1.24) (-2.44) (.41) P3/P2 .2127 -.0399 .7065 -1.0046 6.6045 .4573 -.2557 (.23) [-.05) (.87) [-5.121 (2.58) (1.44) (-2.67) P“/P2 -.7651 1.5573 1.2420 -.7497 4.1041 .8684 -.1637 (-.86) (2.01) (1.59) (-3.40) (2.48) (2.51) {—1.67) F’s/P2 .7242 .7288 .1269 -.2319 1.1280 .4672 -.2493 (.73) (.81) (.13) (-1.36) (.70) (2.31) {-3.031 l’u/P3 -.9778 1.5972 .5355 .2550 -2.5004 .4111 .0920 (-.98) (2.04) (.64) (1.05) (-.93) (.95) (.81) PSI’P3 .5115 .7686 -.5796 .7728 -5.4765 .0099 .0063 (.48) (.88) (-.60) (4.00) (-2.07) (.03) (.06) P51?“ 1.4893 -.8285 -1.1152 .5178 ~2.9761 -.4012 —.0857 (1.40) (-.91) (-1.16) (2.24) (-l.56) (-1.10) (-.30) x2 7.00 6.03 42.92 11.06 23.26 13.92 Farm Type 1 a Food and Cash Type 2 = Food and Dairy Type 3 a Food and Off Farm Work Type 4 = Cash Only Type 5 8 Cash and Off Farm Work 157 Table 7.10 Results of the Model Estimation Under The Reduced Specification 10—50 Hectares ANOVA Normalization Variable Food 8 Food 8 Food 8 Cash Cash 8 )(2 Cash Dairy Off Farm Only Off Farm Constant .3735 -.1363 .0765 -.9014 .5879 (.60) (-.24) (.12) (-1.44) (.84) Muriae . 3843 - . 6576 - . 6975 . 8997 . 0712 (R2) (.76) (-1.31) (-1.49) (1.73) (1.16) 7.00 Vicosa 1. 0583 - . 7834 - . 0769 . 4586 -. 6566 (R3) (1.79) (-1.52) (-.15) (.85) (-.99) 6.03 SI -.4621 .6121 -.3926 -.1376 .3802 (-4.00) (5.25) (-3.18) (-.84) (3.20) 42.92 Labor/Ha . 9425 -3. 1948 3. 4097 . 9093 —2. 0668 (.55) (-2.85) (1.82) (..68) (-1.62) 11.06 Credit/Ha . 5470 - . 5850 -— .1277 . 2834 -. 1178 (2.97) (-3.76) (-.57) (1.02) (-.71) 23.26 Family - . 0989 .1979 -. 0638 .0282 - . 0575 Size (-1.47) (3.55) (-.98) (.38) (-.98) 13.92 158 observation increased, provided R2 assumes the value one (e.g. , the farm is in Muriae) . These results come as no surprise, as both dairy and off farm work activities are more frequent in Juiz de Fora than in Muriae (Appendix C) . By the same token, the higher likelihood of observing "cash only" systems in Muriae could be expected, particularly for this group of farmers. Since the land constraint is relaxed to some extent, it is entirely rational for a farmer to allocate more resources to enterprise mixes which tend to generate higher returns to fixed factors. This fact, coupled with regional specific incentives, plausibly led Muriae farmers towards a comparatively greater emphasis in "cash only" systems, other things held constant. The dummy variable for Vicosa shows a positive and significant association with the probability of belonging to the "food and cash" group, and a negative association with P (food and dairy). though the coefficient of the latter is significant only at the 20 percent level. Once again, the results of Table 7.9 are indicative of this pattern. The odds of either "food and dairy" or "cash and off farm work" systems relative to "food and cash" are reduced, given that the farm is in Vicosa. These are very consistent findings, since the sample data has shown farmers in the area of Vicosa to be less oriented towards cash crops than those ' in the other two regions (Appendix C) . Lesser binding land constraints (vis a vis the former size class) are then the possible cause for the increased relevance of "food and cash" mixes among Vicosa farmers. The sensitivity index appeared as a highly significant variable influencing participation in alternative systems. Specifically, we infer that those farmers who need larger returns per hectare to avoid the necessity of selling non-liquid assets to provide for household needs are 159 more likely to emphasize "food and cash" or "food and off farm work" systems. As the sensitivity index increases, the probabilities of these two systems decrease, while those of "food and dairy" and "cash and off farm work" increase. Hence, the greater his ability to withstand loss before sale of fixed assets, the more likely is a farmer to move away from mixes where food crops tend to be relevant. In this context, it should be noted that the basic feed grain in dairy enterprises--corn--is also classified as a food crop. Therefore, the "food and dairy" mixes for the most part reflect this characterization. The odds ratios of Table 7.9 are helpful in supporting the observations above. Compared to the first system (food and cash), we see that the odds of both "food and dairy" (P2) and "cash and off farm work" {(PS) increase with higher $15. On the other hand, the comparisons of P3 and P4 ("food and off farm work" and "cash only") to P2 show that their odds of observation are inversely related to SI. Finally, both odds ratios PS to P3 and P5/P4 are directly associated with SI. Farm types 2 and 5 are then the two systems more likely to be observed, as $1 increases in magnitude. The association of higher Sis with two systems which are primarily market oriented suggests that "safety first" considerations are a very likely underlying motivation in the selection of enterprise mixes for farmers in this size range. The emphasis on systems involving food crops characterized for farms with low 515 can be interpreted as indica- tive of the importance attributed to fulfillment of basic household needs via traditional, less risky farm systems (e.g., systems with less sus- ceptibility to price and yield fluctuations). It follows then that risk preferences and perceptions are in all likelihood strong elements in the shaping of farm enterprise choice decisions. 160 Concerning the availability of labor/ hectare, the coefficients of Table 7.10 show that the probability of "food and dairy" systems being observed decreases as more units of this variable are present on the farm. In fact, comparisons with the alternative farm types reveal that all but "cash and off farm work" have their odds of occurrence increased relative to "food and dairy," given more availability of labor/hectare. The implication is that this particular farm type is characterized by requirements of labor/ hectare which are lower than for any of the alter- natives. Hence, higher levels of the explanatory variable make it more likely for a farmer to engage in enterprise mixes other than "food and dairy." The farm type which is more likely to be observed, given greater levels of labor/ hectare is "food and off farm work," as inferred from the positive and significant coefficient for this variable on the third column of table 7.10. This is not at all surprising, for it was suggested in Chapter V that off farm work for this size class is an activity mostly pursued by farm units whose labor/land ratios are large. The patterns emerging from this estimation thereby lend support to that observation. Credit per hectare is a further variable whose effect on the prob- abilities of participation is significant. What emerges from the model estimation is a_ tendency for increases in this variable to decrease the probability of "food and dairy" farm types, while increasing that of "food and cash" systems. All farm types have their probabilities of participation decreased when compared to "food and cash" systems, given increases in credit/hectare, though one comparison (Pu/P1) does not have a significant coefficient. Moreover, the comparisons of alter- native systems to "food and dairy" show that the odds of each of them 161 increase with higher levels of credit/hectare. Again, only one comparison (P3/P2) showed a low level of significance for the respective coefficient (20 percent). The results for this particular variable may be interpreted as a first indication of the role of the PRODEMATA deveIOpment project on altering credit use patterns in the sample area. An earlier study has shown that in the past banks tended to concentrate loan concessions on areas offering lower risk, such as investments in the expansion of herds or acquisition of fixed assets.12 Thus, the distribution of credit among uses was relatively biased towards capital intensive activities, particularly in those loan operations carried out by banks other than the federal controlled one. This pattern has changed over the years. From the survey data we observe a higher proportional utilization of credit by farm systems involving cash and food craps, while "food and dairy" farm types were the second lowest users. This evidence is reflected by the model results. As more credit becomes available, the average farmer in the 10-50 hectare class will have a greater likelihood of emphasizing combinations involving cash crops, especially the "food and cash" system. Turning now to the last independent variable considered, we note that contrary to expectations the size of the household unit does not appear to have a direct effect on the probabilities of observing farm types where traditional food crops are more prominent. Instead, the model results reveal an association of this variable with "food and dairy" systems. Al other effects are not significant. In accounting for the finding, it is important to note that one reason often mentioned for the 12T. White, Jr. and Dilson Rocha; "Credito Agricola na Zona da Mata de Minas Gerais;" in Estudos sobre uma Regiao Agricola: Zona da Mata gefMinas Gerais, Monografia, No. 9 (Rio de JaneIro: IPEA, 1973), p. 226. 162 importance of dairy systems in the study region is the liquidity charac- teristic of this activity. 13 Dairy sales provide a relatively stable source of cash income year round, and this is particularly attractive for a household unit which must meet high demands for foods and non-food alike. Although this hypothesis is not untenable, it cannot be tested without additional data. In any case, it will be seen later that the same association is encountered for the 50—100 hectare class. To summarize, it is noted that the clear dichotomy among more and less commercially oriented farms observed in the former farm class appraised is not as apparent here. The effects of the independent variables on the alternative probabilities are in most cases specific, making generalizations improper, to say the least. Looking at the farm types on an individual basis then, we note that the farmer more likely to be in the "food and cash" group is that who is in Vicosa, has a low sensitivity index, has a relatively high amount of credit/hectare, and whose household unit is small, relative to the others. "Food and dairy" farm types are more likely to be selected by Juiz de Fora farmers, with higher 515, low levels of both labor and credit/hectare and rela— tively large families. Juiz de Fora farmers are also the more likely to belong to the "food and off farm work" group, given low 515 and high levels of labor per hectare. The fourth group--cash only--was charac- terized only by a larger presence of Muriae farmers, whereas farmers who are more likely to belong to the last group are characterized only by high 515. Finally, the probabilities of participation in the alternative groups for the average farmer are respectively .35, .14, .12, .22 and .17. 13Universidade Federal deVicosa, OP- Cit" 1971' 163 50—100 Hectares Following the procedure of the two former cases, an initial model with all of the explanatory variables was estimated. Tables 7.11 and 7.12 present the coefficients, asymptotic t ratios and the chi-squared statistics for the two normalizations. At the ten percent confidence level (x: = 6.25) we fail to reject the hypothesis of no effect on the probabilities for the variables R3, DIST, K/Ha, F and R2. The latter two have coefficients which are jointly significant at the 20 percent level though, and are therefore maintained in the model. R3 is also kept to ensure the breakdown of predicted probabilities by all of the three regions. Results of the model estimation under this reduced specification are presented in Tables 7.13 and 7.14. For this size class, the impact of location variables is much smaller than for the earlier cases. Only R2 shows some degree of association with the probabilities of participation in the alternative groups. The results reveal that relative to Juiz de Fora, farmers located in Muriae have a higher probability of belonging to the "cash only" group, and that at a lower significance level (20 percent), these farmers are less likely to belong to the "food and dairy" group. Vicosa farmers, on the other hand, show a slight tendency towards emphasis on the first farm group (food and cash), but the hypothesis of no effect on the probabilities cannot be rejected at the usual confidence levels. One is therefore led to conclude that regional effects on the probabilities of participation in the alternative systems are very minor for larger sized farms. The sensitivity index in turn appears again as an important variable affecting probabilities of participation. Very much like the effects on the previous size class, increases in this variable tend to reduce the 164 Table 7.11 Results of the Initial Model Estimation 50—100 Hectares Theil's Normalization Variable Pz‘lP1 P3/P1 Pu/P1 3 2 4 2 4 3 X Constant 3.8735 -1.2528 -.3779 -5.1263 -4.2514 .8749 (.93) (-.41) (-.07) [-1.56) (-.71) (.16) Muriae -6.0570 1. 5305 -2. 7908 7. 5875 3.2663 -4.3213 (R2) (-1.61) (.59) (-.51) (2.36) (.55) (-.81) 5.89 Vicosa -5.6014 -1.7292 -4.5339 3.8722 1.0675 -2.8047 (R3) (-1.46) (-.69) (-.82) (1.23) (.19) (-.53) 2.49 Distance .0044 .0055 -.0073 -.0039 -.0117 -.0078 (.07) (.01) (-'.20) (-.06) (-.20) (-.20) .08 51 3.1146 1.2469 2.5329 -1.8677 -.S817 1.2860 (3.70) (2.06) (4.76) (-2.00) (-.83) (1.88) 24.78 K/Ha .1120 .1791 .3414 .0671 .2294 .1623 (.30) (.67) (2.07) (.17) (.66) (.62) 4.51 Labor/Ha -58.5639 -3.1457 -20.0011 55.4183 38.5628 ~16.8555 (-4.52) (-.19) (-2.21) (2.72) (3.58) (-.92) 20.80 Credit/Ha -.4998 .4710 .1985 .9708 .6983 -.2725 (-.46) (.60) (.24) (.90) (.76) (-.31) 1.06 Family .4462 -. 0979 .1563 -.5442 -.2899 .2543 Size (1.69) (-.41) (.65) (-2.23) (-1.38) (1.06) 5.33 Farm Type 1 = Food and Cash Type 2 - Food'and Dairy Type 3 = Cash Only Type 4 = Cash and Off Farm Work 165 Table 7.12 Results of the Initial Model Estimation 50-100 Hectares ANOVA Normalization Variable Food 8 Food 8 Cash Cash 8 x2 Cash" Dairy Only Off Farm Constant -.5606 3.3128 -1.8135 -.9386 (-.22) (1.20) (-.93) (-.23) Muriae 1.83 —4.2277 3.3598 -.9615 (R2) (.83) (-1.59) (1.76) (—.25) 5.89- Vicosa 2.9661 -2.6353 1.2369 -1.5678 (R3) (1.33) (-.98) (.68) (-.39) 2.49 Distance .0006 .0049 .0011 -.0067 (.02) (.12) (.04) (-.29) .08 SI -1.7236 1.391 -.4767 .8093 (-4. 6) (2.46) (-1.0) (2.41) 24.78 K/Ha -.1582 -.4613 .0209 .1833 (-1.06) (-.18) (.11) (1.37) 4.51 Labor/Ha 20. 4276 —38. 1362 17. 2820 . 4265 (2.85) (-4. 13) (1.30) (.06) 20.80 Credit lHa - . 0424 —. 5422 . 4286 .1561 (-.08) [-.79) (.77) (.31) 1.06 Family - .1262 . 3201 -. 2241 . 0302 Size (-.80) (2.17) (-1.51) (.22) 5.33 166 Table 7.13 Results of the Model Estimation Under the Reduced Specification 50-100 Hectares Theil's Normalization Variable P2/P1 P3/P1 Pit/P1 P3/P2 P4/P2 Pit/P3 x2 Constant . 9181 -1. 0106 - . 4497 -4. 7664 -4. 2055 . 5609 (.92) (-.34) (-.08) (-1.47) (-.72) (.11) Muriae -5.6997 1.8210 -1.8242 7.5206 3.8755 -3.6451 (R2) (-1.S1) (.69) (-.34) (2.38) (.67) (-.70) 5.94 Vicosa -5. 6874 -1. 5982 -3. 8781 4. 0893 1. 8094 -2. 2799 (R3) (-1.46) (-.63) (-.71) (1.29) (.31) (-.44) 2.42 Si 2.9448 1.2180 2.5095 -1.7269 -.4353 1.2916 (3.86) (2.16) (5.15) (-2.04) (-.67) (2.03) 28.70 Labor/Ha -55.5709 -.0615 -15. 7839 55.5094 39.7870 ~15.7224 (-5.13) (-.01) (-1.92) (2.89) (4.58) (-.86) 29.16 Family .4465 -.1019 .1323 -.5484 -.3142 .2342 Size (1.79) (-.44) (.58) (-2.35) (—1.S7) (1.01) 6.06 Farm Type 1 = Food and Cash Type 2 = Food and Dairy Type 3 = Cash Only Type a = Cash and Off Farm Work 167 Table 7.14 Results of the Model Estimation Under The Reduced Specification 50-100 Hectares ANOVA Normalization Variable Food 8 Food 8 Cash Cash 8 x2 Cash Dairy Only Off Farm Constant -.5738 3.1819 -1.5845 —1.0236 (-.23) (1.17) (-.83) (-.26) Muriae 1. 4257 -4. 2740 3. 2467 -. 3984 (R2) (.64) (-1.63) (1.75) (-.10) 5.94 Vicosa 2.7909 -2.8965 1.1927 -1.0871 (R3) (1.28) (-1.08) (.67) (-.28) 2.42 Sl -1.6681 1.2768 -.4501 .8414 (-4.88) (2.49) (—1.03) (2.67) 28.70 Labor/Ha 17.8541 -37.7268 17.7926 2.0702 (2.69) (~4.83) (1.35) (.33) 29.16 Family -.1192 .3273 —.2211 .01309 Size (-.80) (2.34) (—1.54) (.11) 6.06 168 likelihood of observing farmers in the "food and cash" group while increasing that of the farm types "food and dairy" and "cash and off farm work." Actually, compared to "food and cash" all groups have higher probabilities of being observed with increases in SI, as evidenced by the highly significant coefficients in the first three columns of Table 7.13. The two other significant comparisons in the SI row in this table show that the third farm system (cash only) is less likely than both the second (food and dairy) and the fourth (cash and off farm work), given increases in the respective independent variable. However, the corresponding coefficient of the "cash only" farm type in the ANOVA normalization is not significantly different from zero at usual confidence levels. The same comments made earlier on the effects of the sensitivity index are applicable here. An interesting feature is that the increases in farm size we observe when moving from the 10-50 hectare to the 50—100 hectare class have apparently reinforced the effects of the variable. This suggests that concern with safety aspects in farm decisions does not vary with the size of the farm, at least in this par- ticular range of observations. Similarly to the previous size class discussed, we encounter a negative association among the probability of belonging to the "food and dairy" group and the levels of labor/hectare. More specifically, 8"V 0f the alternative groups has a higher likelihood of being observed than "food and dairy," as evidenced by the significant coefficients of all comparisons involving P2 in Table 7.13 (and their respective signs). But unlike the 10-50 hectare farmers, we find here that increases in Iabor'lhectare will also increase the probability of "food and cash" farm 169 systems. However, these results are not worrisome, as it is well known that this particular crop system is generally more labor intensive than dairy enterprises. The model results are thus consistent. Finally, we again encounter an association between higher family sizes and the probabilities of observing the "food and dairy" farm type. The chi-squared statistic is slightly below the ten percent critical point of 6.26, an indication that the significance of the effects of this variable on the alternative probabilities was somewhat improved in the reduced estimation. Moreover, both the comparisons of "food and dairy" systems to "food and cash" and "cash only" have significant t values, suggesting that larger household family units increase the odds of "food and dairy" relative to these two groups. Thus, the hypothesis proposed on the former size class is also appropriate in this particular case. To sum up, we can characterize a farmer likely to belong to the "food and cash" group by his comparatively lower sensitivity index and higher level of labor per hectare, and to a lesser extent by his location in Vicosa. "Food and dairy" groups are more likely to be observed among Juiz de Fora farmers, with relatively higher Sls, lower levels of labor/hectare and larger families. The last two groups are more likely only for,Muriae farmers (cash only) and farmers with higher Sls (cash and off farm work). Probabilities of participation in the four groups are respectively 13 percent, 20 percent, 40 percent and 27 per- cent, for a farmer whose levels of the independent variables equal the sample averages. 170 Predicted Probabilities Under Alternative Scenarios Once the consistency of the model results are established, the estimated coefficients are used to predict the effects of changes on the levels of the independent variables. insofar as these changes may reflect a number of policy alternatives, the model becomes a useful instrument in the evaluation of the impacts of these actions on partici- pation in the different farm types. The predicted probabilities can be derived from the coefficients of either one of the normalizations considered. If the Theil normalization is adopted, the equations are: n P1=1/1+ 2 exp XB. , and -- l 1-2 P.=ex X.fP ,'=2,3,...n l P 81 1 1 Conversely, if the parameters of the ANOVA normalization are employed, we have n P. =exp XBH 22 exp X8? , i=1, 2, . . . n l l 1:1 l where the is denote the vectors of coefficients for each respective alter- native. Of course, the predictions are the same regardless of the normalization . Average values for the independent variables are used to compute the probabilities to be used as a standard of reference. Further, these are disaggregated so that predicted reference values are available for each of the three regions of the sample area. 0:10 Hectares The predicted probabilities of participation for the average farmer in this size class, as well as the disaggregated values by region, are shown in Table 7.15. 171 Table 7.15 Predicted Probabilities 0-10 Hectares Scenario Food Food 8 Food 8 Cash Cash 8 Only Cash Off Farm Only Dairy Average Values Zona da Mata .11 .10 .44 .16 .19 Muriae .06 .18 .21 .34 .21 Vicosa .10 .25 .49 .03 .13 Juiz de Fora .20 .00 .60 . 13 .07 25% Reduction on Average L/ Ha Zona da Mata .14 .10 .42 .16 .18 Muriae . 07 . 19 20 . 33 . 21 Vicosa .13 .27 .45 .03 .12 Juiz de Fora .26 .00 .56 .12 .06 increase on Average C/Ha by 1 Standard Deviation (+ Cr$1750lHa) Zona da Mata .03 Muriae .02 Vicosa .03 Juiz de Fora .07 .11 .19 .29 .00 .45 .20 .49 .68 .17 .34 .03 .15 .24 .25 .16 .10 172 At the average levels of the independent variables, the model estab- lishes "food and off farm work" as the most likely group to be observed, with a probability of 44 percent. Considering the characteristics of farmers in this size range, emphasis on enterprise mixes where crops grown for subsistence requirements are coupled with sale of labor for cash income is indeed a rational strategy. It is very clear from all available evidence that there is an excess labor supply on farms with less than ten hectares in the sample region. If conditions are such that growing cash crops is not feasible, the viable alternative is the employ- ment of surplus labor in off farm activities. That is exactly the strategy being reflected by the model. It is also interesting to note that the effects of regional specialization on enterprise mix decisions earlier pointed out are clearly evidenced by the predicted probabilities. In particular, a propensity for emphasis on cash crops is apparent for Muriae farmers, largely reflecting the concen— tration of coffee farms in that area. This shows that small farmers are not unaware of conditions favoring specialization, nor of the benefits arising from it. Further evidence on such rational decision making is found from the high probability predicted for a Juiz de Fora farmer to be in the "food and off farm work" group (60 percent). Since the Juiz de Fora region has the greatest potential for the absorption of excess agricultural labor, we find a larger share of small farmers benefiting from off farm work opportunities in that area. In sum, the predicted probabilities show very clearly that an average farmer would be likely to emphasize different farm enterprise mixes, depending on the location of his farm. 173 Of interest for this research are the likely effects of modifying average constraint levels on the predicted probabilities. For this particular farm size class, we are enabled to test the impact of changes in both the availability of labor/hectare and credit/hectare. Let us consider first the effects of changes in the availability of labor/hectare. Since labor is not a limiting factor, the impacts of a 25 percent reduction on the average level of this variable are determined. Compared to the average farmer, such a change would cause a movement away from the third and fifth groups towards the first one (Table 7.15) . Specifically, an increase of three percent on the probability of belonging to the "food only" farm type was observed, stemming from a two percent reduction on the probability associated with "food and off farm work," and from a one percent reduction on P (cash and dairy). The pattern is pretty much the same across regions: farmers will be likely to shift towards the "food only" farm type, in the process diminishing the emphasis on those more frequent groups in the respective regions. Given that a 25 percent decrease on the average level of labor/hectare translates into a reduction of only 170 man days equivalent/year in a ten hectare farm, the noticeable changes in the probability breakdown arising from this relatively small change are illustrative of the significant impact of this variable. An important finding of this exercise is the marked reduction in emphasis on the off farm work group, obtained with the decrease in labor availability. This confirms that this particular activity is exer- cising a strong role in the absorption of surplus agricultural labor. From a policy perspective, it seems clear that expansion of the labor market could then be a realistic avenue towards improvement in income levels for this group of farmers. In view of the severe limitation imposed by 174 the smallness of the land base, it is unlikely that substantial increases in income would be achieved via reorganization of existing farm mixes. In fact, this was observed in Chapter V, where differences in income could not be linked to the index of system potential. Therefore, from the evidence of this analysis, it appears that potential gains can be derived from the introduction of incentives towards the inducement of upward shifts in labor demands from both agricultural and non-agricul- tural sectors of Zona da Mata. The second independent variable susceptible to policy intervention is the level of credit/hectare. Credit has for long been a major instrument of agricultural policy in Brazil. Particularly in an area such as Zona da Mata, where conditions do not favor capital accumulation via increased savings, the role of financing institutions in providing for investment and working capital needs in agriculture is essential. Yet, in spite of negative effective interest rates and despite a major governmental commitment to increase credit use among small farmers, we still observe a very low average availability per hectare (Cr$641) among the 0—10 hec- tare farmers. In fact, 75 percent of these producers reported no use at all. Such is exactly the major problem being addressed by the PRODEMATA project. Hence, as the project reaches more mature stages in its implementation process, we should expect the present levels of credit use to grow. To test the effects of increases in credit availability, predicted probabilities were computed over the range of observed levels of this variable. The results are shown in Figure 7.1 for the average farmer in the area, and in Figures 7.2, 7.3 and 7.4 respectively for those pro- ducers located in Muriae, Vicosa and Juiz de Fora. The impact of Predicted Probabil lty 175 Food and Off Farm Work Cash and Dairy Cash Only / \ Food and Cash Food Only 1. 7 3. 4 5.1 6.8 8. 5 Cr5103/Ha Figure 7.1 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Zona da Mata Farmers--0-10 Hectares Predicted Probability 176 Cash and Dairy Cash Only Food and Ca sh Food and Off Farm Work Food Only 1.7 3.4 5.1 6.8 8.5 Crsio3/Ha Figure 7.2 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Muriae Farmers-0-10 Hectares Predicted Probability 177 Food and Off Farm Work Food and Cash Cash and Dairy Food Only Cash Only 1 g M 1.7 3.4 5.1 6.8 8.5 Cr$103lHa Figure 7.3 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Vicosa Farmers--0—10 Hectares Predicted Probability a '9 178 Food and Off Farm Work \ 7 l Cash Only Cash and Dairy Food Only Food and Cash A L L ‘fi ‘1 1.7 3.4 5.1 6.8 8.5 Cr$103/Ha Figure 7.4 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Juiz de Fora Farmers--0~10 Hectares 179 increases in the average level of credit is illustrated by the probabilities in Table 7.15, computed by augmenting use by one standard deviation (Cr$1750) . In considering the results of these tests, we first note a steady increase in the probability of observing "cash and dairy" farms, that is invariant across regions. The increase is mostly offset by the sharp decline in the probability associated with the "food only" group, which approaches zero, as the value of credit/hectare increases. Little availability of credit thus appear to be a factor constraining higher par- ticipation in enterprise mixes where less traditional activities gain relevance. As availability increases, the marked increase in the prob- ability of participation in a more market oriented farm type presents decisive evidence in this regard. This shift is therefore an encouraging sign of the potential of the project in enabling small farmers to make a more effective utilization of available resources from an income generation standpoint. Also interesting are the effects on the probabilities of participation on the farm type "food and off farm work" in Vicosa and Juiz de Fora (Figures 7.5 and 7.4) . The initial increases in credit/hectare are apparently contributing towards the displacement of labor from the more subsistence oriented farms, as suggested in the increases in the off farm work activities observed at first. However, as additional increases are introduced, reduction on participation in this group starts to be noted. This is further evidence of the earlier noted role of off farm work activities for the 0-10 hectare farmer. 180 Finally, the alternative groups tended to remain relatively stable over the observed range of credit/hectare, with slight variations occurring from region to region. Among the variations, the decline observed in the probability of "cash only" farms in Muriae is a bit surprising (Figure 7.2) . Nonetheless, the offsetting increase is also in a less traditional enterprise combination involving cash crops and the predictions are not to be seen as inconsistent with the observed characteristics of Muriae farmers. In brief, the changes introduced in both exercises above have shown that decisions on enterprise combinations are responsive to levels of resource availability and perceptions of market opportunities. This is an important finding of the analysis, for it suggests that emphasis on target enterprise combinations can be promoted by appropriate policy incentives. Although income levels are not likely to increase substan— tially unless the severe land limitation is relaxed, farmers in this size group will certainly benefit from policies such as expressed by the PRODEMATA project, or implied by the above discussion on absorption of surplus labor . 10- 50 Hectares An average farmer in this size class would be more likely to be in the "food and cash" group, with a predicted probability of 35 percent (Table 7.16). Yet, we observe that the regional breakdown is not as skewed towards one particular farm type as it was in the former case. With the exception of Vicosa, the predictions are very close for two or more farm types. This is illustrative of the smaller impact of regional factors on these producers. As land constraints are relaxed, incentives for regional specialization are apparently being reduced, although not 181 Table 7.16 Predicted Probabilities 10— 50 Hectares Scenario Food 8 Food 8 Food 8 Cash Cash 8 Cash Dairy Off Farm Only Off Farm Average Values Zona da Mata .35 .14 .12 .22 .17 Muriae .29 .13 .09 .28 .21 Vicosa .50 .10 .15 .16 .09 Juiz de Fora .21 .26 .20 l .12 .21 increase Average Si by 25% Zona da Mata .33 .15 .12 .22 .18 Muriae .27 .14 .09 .28 .22 Vicosa .48 .11 .15 .16 .10 Juiz de Fora .19 .29 .18 ' .12 .22 Zona da Mata .34 .16 .11 .21 .18 Muriae .28 .15 .08 .26 .23 Vicosa .50 .12 .13 .15 .10 Juiz de Fora .20 .30 .17 .11 .22 Increase Average Credit/Ha by 25% Zona da Mata .37 .12 .12 .23 .16 Muriae .31 . 11 .09 .29 .20 Vicosa .53 .09 .14 .16 .08 Juiz de Fora .23 .2u .19 .13 .21 182 completely eliminated. in particular, there are still some indications of a preference for mixes involving cash craps in Muriae, and of a stronger emphasis in "food and cash" systems in Vicosa. More evidence on the diminished relevance of regional differences are provided by the pre- dicted effects of changes in average values of the independent variables. Changes in the sensitivity index provide a general idea of the extent to which safety concerns affect resource allocation decisions. An increase in the average value of Si, for instance, is akin to a reduction of the probability of disaster for a particular farmer, disaster being defined as a loss which would result in indebtedness, sale of fixed assets or worse, starvation. The higher the Si, the more a farmer is able to withstand loss, and consequently the lower is his probability of disaster for a given distribution of returns. For an average farmer, an increase of 25 percent in Si results in a two percent reduction on the probability of being in the "food and cash" group, followed by an increase of one percent in the probabilities associated with both the farm types "food and dairy" and "cash and off farm work." The pattern is very similar across regions. Only a slight difference in the intensity of the effects is observed in Juiz de Fora, where the third group (food and off farm work) has its probability of participation reduced by the same proportion as that of the "food and cash" farm type. The implications of the predicted changes are clear. Since the con- cern with the relative safety of alternative strategies is so apparent, the design of policies aiming at greater emphasis in less traditional farm mixes must be based on a better understanding of farmers' perceptions for risk. We have noted that there is potential for income gains from shifts into already existing and higher return enterprise mixes in this 183 class of producers. In particular, mixes involving cash crops such as fruits, vegetables or coffee were shown in Chapter V to be emphasized by high income farmers. Based on the experiment with Si, a possible reason for a low participation into such less traditional enterprise com— binations is the high degree of risk aversion which, from all indications, characterizes most producers in this size range. if such a hypothesis can be shown true, a reduction on both production and market risks through crop insurance schemes, price supports or similar policies should provide the necessary incentives for increased emphasis in these target activities. The effects of a 25 percent reduction on the availability of labor/ hectare only reinforce the comments made on the discussion of the model results. Specifically, we observe decreases in the probabilities of participation in the knowingly more labor intensive farm types, with corresponding increases in the "food and dairy" category. However, an unexpected effect is the increase in the probability of emphasis on enter- prise combinations where cash craps and off farm work are the relevant activities. Generally, decreases in the average level of labor [hectare were found to reduce the probabilities of both these activities when components of_alternative mixes, or even when highly stressed, as in the case of "cash only" farms. Yet, if a situation arises such that the availability of labor is not sufficient for placing large emphasis in more than one cash crop, the allocation of any surplus labor from the first crop to the off farm activities would be a conceivable strategy. it should be noted nonetheless that the increases in P (food and dairy) will tend to outweigh that of P (cash and off farm work), as successive 184 reductions on the average level of labor/hectare are introduced. But, apart from this, no essentially new information is obtained from this particular experiment. Finally, increases in credit/hectare over its range of observed values are evaluated. Use of credit among the 10-50 hectare farmers is somewhat more frequent than in the former case, although in absolute terms a large proportion of farmers (40 percent) still reported no use at all. The average level is also low (Cr$741/hectare), and here too we can expect the PRODEMATA project to exert an important impact. The extent of its likely effects at later stages of implementation may be approximated by the predicted probabilities over the full range of observations on credit availability/hectare (Figure 7.5) . Further, a numerical illustration of a 25 percent increase in the average level of credit/hectare is shown in Table 7.16. Since the variation of effects across regions is very minor, a dis- cussion of the average results should suffice. Figure 7.5 shows clearly that higher levels of credit use will cause a sharp and steady increase in the probability of participation in the "food and cash" farm type. This is offset by reductions on all alternative predictions, although over a small range ofobservations (0-Cr$12500/hectare) there is an increase in P (cash only). These predictions denote that traditional patterns of enterprise com- binations are likely to be strengthened with greater availability of credit per hectare, other things being held constant. Essentially, they are mirroring the existing levels of credit use across farm types, which show comparatively higher values associated with "food and cash" farms. Predicted Probability l. .6 0 0| 185 Food and Cash Food and Dairy Off Farm Work \ caSh only 1 ._;_______________. 2.5 5.0 7.5 10.0 12.5 15.0 Cr$103/Ha Food and Off Farm Work Figure 7.5 Predicted Probabilities Over the Range of Sample Observations: Credit/Hectare Zona da Mata Farmers—-10-50 Hectares 186 This particular system is a traditional one among the 10—50 hectare producers and it is interesting that more credit appears to reinforce stress on it. But one aspect which cannot be assessed from the pre- dictions is the effect of credit on marketable surplus of food crops. Since on farm consumption cannot grow indefinitely, increased emphasis on combinations involving food craps through more credit use may generate opportunities for larger market participation. in fact, this has been shown to happen in a similar analysis for small farmers in northeastern Brazil.” Again, the model results do not allow a hypothesis as such to be tested, and further analyses are needed on the aspect. In brief, the information yielded by the predictions suggest that credit by itself is not likely to induce changes in existing patterns of farm enterprise combinations. Unless the measures expressed by the project are coupled with efforts such as reducing risks generally associated with higher return activities, observed patterns of enterprise planning are likely to persist. in any case, this is not to say that gains in income will not be forthcoming from the project. Of course, this discussion is limited to one of many dimensions which the measures will affect . 50-100 Hectares Producers in this size range have been characterized in Chapter V as more oriented towards less traditional activities than those in farms with less than 50 hectares. As such, the predicted probabilities for the average farmer in Table 7.17 denote a greater likelihood of parti- cipation on the "cash only" group. 1“Edinaldo Bastos, op. cit., pp. 218-220. 187 Table 7.17 Predicted Probabilities 50—100 Hectares Scenario Food 8 Food 8 Cash Cash 8 Cash Dairy Only Off Farm Average Values Zona da Mata . 13 .20 .40 .27 Muriae .06 .06 .67 .21 Juiz de Fora .00 .92 .01 .07 increase SI by 25% Zona da Mata .05 .31 .29 .35 Muriae .03 .10 .56 .31 Juiz de Fora .00 .93 .00 .07 Decrease Labor by 25% Zona da Mata .07 . 48 . 22 .23 Muriae .05 . 19 .51 .25 Juiz de Fora .00 .97 .00 .03 Note: The location variable for Vicosa does not significantly affect the probabilities of participation. See Tables 7.13 and 7.14 for the statistical proof. 188 The pattern varies dramatically among Muriae and Juiz de Fora farmers, with the latter having a larger probability of belonging to the "food and dairy" group. Although this is also a consistent prediction, we must recall that the significance of the location variables for this region was shown to be low (20 percent). As to Vicosa, no predictions were computed for that area, since the significance of its location variable was lower than 20 percent. Unlike the former cases appraised, only two variables are susceptible to short run changes here and none of them is directly affected by the PRODEMATA project. The sensitivity index has its average level incremented by 25 percent and average labor/hectare is reduced by 25 percent. Results are presented in Table 7.17. The index shows once again a direct association with farm systems involving dairy activities. The increase in the mean level of Si causes a gain of 11 percent on the probability of participation in "food and dairy" systems for the average farmer in Zona da Mata. Increases are also observed both in Muriae and Juiz de Fora, though with different magnitudes. "Cash and off farm work" farms have their probability of participation increased with'a 25 percent increment in SI as well, but successive changes in the same direction would eventually reverse the initial effect. Reductions in P (food and cash) and P (cash only) offset the observed effects on the other two farm groups. This is further evidence of the role of safety concerns in decision making, even among farmers which are generally better off than those in the former classes. There is little doubt that "food and cash" systems are more likely to be selected by farmers who would be affected the most in the event of loss or low returns per hectare. in fact, if we increment 189 SI successively starting from the lowest observed level, all alternatives to "food and cash" have their probabilities of participation increased, up to values close to the average index. From there on, "food and dairy" systems increase more sharply, with offsets from the remaining groups. Therefore, it is once more evident that there is a need to account for attitudes and perceptions of risk on the design of policies aiming at reorganization of farm plans for this class of producers as well. With respect to the labor variable, the effects of changes on average levels do not convey any new information. Essentially, we observe that less labor causes decreases in the probabilities of participation in more labor intensive farm systems, and the opposite is true for less intensive ones. One logical inference from these results is the potential role that these larger farmers can play in absorbing excess labor from the smaller sized properties. The more frequent farm type on the sample for this class is among the lowest users of labor of the alternatives considered. Therefore, increased emphasis on cash crops by these producers--especially coffee, fruits and vegetables--could increase the demand for agricultural labor to a considerable extent. This aspect has been examined in detail elsewhere, and the conclusions are analogous.15 Changes in the independent variables have shown again that the probabilities of participation in alternative farm types are responsive to characteristics of the decision making unit. In particular, it can be inferred that the promotion of changes in existing patterns of enterprise combinations as an income enhancement strategy will depend primarily 15Lon Cesal and Antonio Bandeira, op. cit. 190 on increased knowledge of risk attitudes and perceptions. Moreover, such induced changes may also be instrumental in providing increasing Opportunities for absorption of agricultural labor. Summa 51 An analysis of factors affecting participation in selected farm systems was performed in this chapter through the specification and estimation of polychotomous logit models. These models allowed the assessment of the relevance of each of a number of explanatory factors in affecting the probabilities of participation into the alternative farm systems considered. Labor absorption was the criteria adopted to define the pairs of activities characterizing a farm system. Further, the pairs were grouped into combinations of the categories "food crops," "cash craps," "dairy," "livestock," and "off farm work." For the 0-10 hectare class, the most important systems were "food only," "food and cash," "food and off farm work," "cash only," and "cash and dairy," which altogether accounted for 77 percent of the observations in the subsample. Seventy-five per- cent of the farms in the 10-50 hectare class were classified into the systems "food and cash," "food and dairy," "food and off farm work," "cash only," and "cash and off farm work." Finally, four farm types--"food and cash," "food and dairy," "cash only," and "cash and off farm work"--accounted for 86 percent of the observations in the 50-100 hectare class. These more frequent farm types were selected for the analysis. Selected explanatory factors included the distance from the farm to the reSpective municipality, indicator variables for location within the sample region, a measure of a farmer's ability to withstand loss without need for disposing of non-liquid assets (the sensitivity index), the levels of capital, labor and credit per hectare, and the size of the farm family. 191 The maximum likelihood estimates of the model coefficients were obtained through an iterative algorithm for each of the farm classes. The estimates for the 0-10 hectare class revealed that farm location, and the levels of labor and credit per hectare are the most significant variables affecting participation for these farmers. At the average levels of the independent variables, the system "food and off farm work" is the most likely to be observed, with a predicted probability of 44 percent. For the 10-50 hectare farmers, the important variables are the farm location, the sensitivity index, the levels of labor and credit per hectare, and the size of the family. An average farmer in this class is more likely to engage in "food and cash" farms, with a predicted probability of 35 percent. The average 50-100 hectare farmer has a higher probability of participation (40 percent) in "cash only" farms. important influencing variables were the sensitivity index and the level of labor per hectare. The model was finally used to predict probabilities of participation given changes in the levels of the independent variables. Since these changes may reflect alternative policy options, the results represent an indication of their likely consequences on patterns of enterprise emphasis in the study region. Policy implications of the findings of the analysis are discussed in the next chapter. CHAPTER Vlll SUMMARY AND CONCLUSIONS This chapter has the purpose of presenting an overview of the study as well as its conclusions and recommendations. The first section briefly states the problem and research objectives. Secondly, the research methodology and empirical findings are presented, followed by a dis- cussion of the implied policy issues. The last section points out some of the limitations of the study and suggests areas where further analyses are needed. The Problem and Research Objectives Despite its proximity to rapidly developing areas of southeastern Brazil, the Zona da Mata region of Minas Gerais state has failed to over- come the consequences of adverse historical circumstances and it is today poor and backward, with an estimated annual rural per capita income of U.S. $250.00.1 its economy relies heavily on a stagnant agricultural sector. In consequence, social problems are widespread. Awareness of these problems has led the state's Rural Development Agency to join the World Bank in the design and promotion of a series of remedial measures. They are contained in an integrated rural develop- ment project--the PRODEMATA project. By addressing the needs of small farmers via increased credit availability, the project hopes to provide 1The World Bank, op. cit., 1975 Figure. 192 193 enough incentives to foster growth of agricultural output and associated incomes. Moreover, its strong social infrastructure component should be instrumental in ameliorating the substandard living conditions of rural residents. implicit in the project measures is the belief that increased access to inputs should play an important role in improving farm incomes. Yet, another relevant and interrelated aspect of farm income determination is the planning of the farm system. in particular, it should be of interest to study the differences in the structure of farm enterprise combinations among lower and higher income farmers to ascertain if such income variations can be related to the farm organization. if this is the case, a logical area for policy intervention exists. This is the essential motivation of this study. it draws from the data base of the project's monitoring and evaluation component to examine specific aspects of farm planning among area farmers. The general objective is the development of a better understanding of planning pro— cesses related to the selection of enterprise combinations, as an aid to the design of income enhancement strategies. Specific objectives include: a characterization of farming by classes of producers, an assessment of returns for the major enterprises, an investigation of the relationship between enterprise emphasis and farm income levels, the development of a statistical model relating selected farm plans to charac— teristics of specific decision makers, and the utilization of the model to ascertain the impacts of alternative actions on the probabilities of par- ticipation in target farm plans. 194 Research Methodology and Empirical Findings The initial steps of the research were to characterize the farms in the sample. A brief cross tabulation of the survey data mostly confirmed previous beliefs about the nature of agriculture in Zona da Mata. In general, the sector was depicted as traditional, both in terms of resource utilization and productivity. Corn, beans, coffee and dairy appeared as the more frequent of a large array of enterprises pursued. Other important activities are rice, sugar cane, livestock, fruits, vegetables, and off farm work. Yields of the relevant crops were by and large below the state's average for the same crop year. The low agricultural productivity is consistent with observed patterns of input use. Labor is by far the most intensive resource employed, whereas use of modern inputs was found to be lower than the technically recommended levels, particularly for the smaller sized farms. In terms of market orientation, the evidence reveals that farming in the study region still has characteristics of semi-subsistence. With the exception of some clearly defined cash enterprises (dairy, fruits, vegetables and coffee), output sales were not substantial. In fact, the ratio of cash sales to value of farm output for the entire sample was only . 39. On the average, sales tended to increase with the sizes of farm units; the larger the farm, the higher was the sales/output ratio (and vice versa) . Farm earnings were highly variable both across and within farm classes, mostly reflecting variations in size, enterprises pursued, prices or other related factors. The concentration of earnings among larger farmers was very evident from the sample data. About 70 percent of the farmers had gross earnings below the average of Cr$86811 195 (U.S.$4815) , and the class averages were much higher for the larger farmers. Coffee, dairy, livestock and off farm work in this order were the major sources of earnings for the entire sample. Within classes, there was a tendency for more traditional activities (e.g., corn, beans, rice) and off farm work to be more important sources of earnings for producers with less than 50 hectares, whereas commercial enterprises took the lead among large scale producers. Finally, demographic and educational data provided an overall picture of relatively large farm families and low levels of formal schooling. Average family sizes were slightly larger than the average figures for the state. Forty-six percent of family members are in the 15-50 age bracket, while 37 percent are under 15 and 17 percent are over 50 years. Moreover, 73 percent of the interviewed producers were either illiterate or had incomplete elementary education. This brief characterization has highlighted some of the difficulties that development planners are attempting to surmount. in this context, one area which has not been addressed explicitly in past efforts concerns itself with the feasibility of increasing farm incomes by means of larger participation into already existing enterprise types. Since the answer to this research question requires a more thorough understanding of the nature of enterprise returns among different sized farms, a descriptive analysis of enterprise emphasis and returns was performed for farm owners with farms up to 100 hectares in size. Sharecroppers and the largest farm owners were not considered in view of the more specific nature of their farming methods and related constraints. Net farm incomes were determined for all farm owners with less than 100 hectares and disaggregated by specific enterprises and main production 196 factors. Returns to labor were found to differentiate producers in the three farm classes in a better fashion than returns per hectare. Although variations within classes existed, the general pattern observed was a tendency for less traditional enterprises (i.e., those with a stronger commercial orientation) to generate higher returns to labor. The five highest return enterprises included one traditional activity in the 0—10 hectare class, two in the 50-100 hectare class, and none among the 10-50 hectare farmers. Conversely, the five lowest return enterprises included only one non-traditional activity in both 10-50 hectare and 50—100 hectare classes, and none in the 0-10 hectare class. Therefore, there are indications that incomes could be raised by enhanced emphasis on high return activities, particularly the less traditional which are followed by small numbers of sample farmers. Insights into the relationship between enterprise emphasis and farm incomes were provided by an examination of the importance attributed to alternative activities by producers with similar farm sizes but belonging to different income groups. Importance was defined as the allocation of ten percent or more of labor to a specific activity, or by the fact that more than ten percent of the net farm income was derived from a given enterprise. Although a tendency for low income farmers to give more emphasis to low income activities (and vice versa) was apparent in all three farm classes, the observed differences were statistically significant at the five percent level for only a few of the enterprises considered. In the 0-10 hectare class, off farm work was the only high return activity associated with high income levels, whereas corn and beans (the combined figures) were low return crops associated with low income levels. Fruits, coffee and vegetables were the high return activities emphasized more by 197 high income farmers, while combined corn, and beans (intercropped and combined) had more importance among low income farmers in the 10-50 hectare group. Finally, the largest sized farms in the analysis had livestock as the only high return activity associated with high income levels, and beans (combined and intercropped) as a low return activity associated with low incomes. The results of this analysis show that in terms of individual emphasis few enterprises differentiate farm plans of high income farmers from those of low income producers. Therefore, the evidence for rejecting the hypothesis that differences in farm incomes are primarily a function of differences in emphasis on individual activities in a farm plan appears to be stronger than the evidence for its acceptance. However, an entirely conclusive rejection is prevented by the significant differences observed for some of the enterprises. For this reason, a complementary approach was pursued to seek for more definite results. The approach classifies farmers in the group of returns to which their mostly emphasized enterprise belongs. This ensures that no farmer will be in more than one group, as it could happen in the former testing if more than one important enterprise were present in the farm plan. The test performed then examined the relation between the classification of low, medium and high income farmers into the low, high and medium return enterprise classes. If the principles of classification are indepen— dent, no relationship exists. At the five percent confidence level, the hypothesis of independence is rejected for the 0—10 hectare and 50-100 hectare classes, but not for farmers in the 10-50 hectare range. Hence, the returns characteristic of the enterprise emphasized the most by the smallest and the largest farmers is associated with their income classification 198 levels. These findings are further evidence on the role of individual enterprise emphasis on the determination of farm income. But since results based on emphasis in individual enterprises may mask the joint effects of enterprise combinations on farm incomes, it seems more appro- priate to examine the farm system as an integrated unit. Potential net margin analysis allowed the investigation of effects of both the farm plan and management performance on the determination of net farm income. This analysis revealed that the two factors differ significantly among income groups for farmers with more than ten hectares of land. The smaller farms (less than ten hectares) are more homogeneously characterized in terms of farm plans and management performance, so that observed differences across income levels were not statistically significant at the .05 level. Though both the farm plan and farmer performance were associated with incomes for the larger farms, corre- lation analysis showed that the latter factor is somewhat more pronounced in both groups. The analysis of enterprise emphasis and returns sufficiently evidenced that farm systems differ among income classes for the majority of farm owners in the study area. High income farmers pursue systems which generate more. returns per hectare than those followed by their lower income counterparts in two of three size classes appraised, which together embrace 65 percent of the farm owners in the entire sample. There is also evidence of significant differences in management performance, but that is an area believed to have received adequate research attention in the past. As an alternative course of action, an analysis of factors associated with observed farm types was performed. The motivation was the need 199 for more information on farm planning processes as an input for develop- ment policy formulation. If the nature and relevance of the constraints related to selection of farm plans are better understood, then the promo- tion of participation into target farm types becomes a strategy with a higher probability of success. Moreover, this analysis should enable the assessment of the impact of the PRODEMATA project on existing patterns of enterprise combinations. Therefore, the 0—10 hectare class was also included in the analysis of factors affecting enterprise choice. ‘Usual approaches for studying farm planning involve modeling farmer behavior in alternative mathematical programming formulations. Based on assumptions about the nature of decision makers' motivations, these approaches provide allocation solutions consistent with constrained optimization behavior. These methods are essentially normative, and although appealing under several circumstances, they were not the most apprOpriate for the proposed analysis. Since the concern was the identi- fication of factors affecting participation in existing farm plans, a non- normative approach seemed more adequate. In particular, a model which enables the association of hypothesized factors with observed farm plan choices under the usual assumptions of rationality and utility maximiza- tion would provide an acceptable analytical framework. Therefore, an econometric model was adopted whereby the hypothesized factors are considered as independent variables and the observed farm type choices are included as dependent variables measured only at the nominal level. From alternative specification forms, the polychotomous logit model was selected. This form specifies probabilities of participation into the alternative farm types as a function of the independent variables con- sidered. The maximum likelihood estimation of its parameters would 200 provide the basis for testing the relevance of each of the individual explanatory factors. Furthermore, the estimated model would allow the prediction of probabilities of participation in the observed farm types under alternative scenarios. Farms were classified by the pair of activities accounting for the maximum absorption of labor in each unit. By classifying each activity into five categories-~cash crops, food craps, dairy, livestock and off farm work--twelve farm types were characterized. A frequency distri— bution of these types further revealed that a number of them was clearly more important in each farm class. At least 75 percent of the observations would be accounted for by five of the types in the 0-10 hectare and 10-50 hectare classes, whereas four of them encompassed 86 percent of the sampled farmers in the 50-100 hectare class. For the sake of tractability in the model estimation and interpretation of results, these most frequent types were selected for the analysis. Thus, for the smaller farmers, the types are "food only," "food and cash," "food and off farm work," "cash only," and "cash and dairy;" in the 10-50 hectare class we have "food and cash," "food and dairy," "food and off farm work," "cash only," and "cash and off farm work;" finally, for the largest farms the important types are "food and cash," "food and dairy," "cash only," and "cash and off farm work." Independent variables hypothesized to impinge on participation in alternative farm types were selected with basis in a conceptualization of small farmer decision making adapted from the work of John Cleave,2 under the constraints imposed by the data base. They include location 2Op . cit . 201 dummies, the distance from the farm to the municipality, a measure of a farmer's ability to withstand loss without need for disposing of non-liquid assets (the sensitivity index), the levels of capital, labor and credit/hec- tare, and the size of the farm family. The general procedure followed in the determination of the important variables affecting participation consisted initially in the estimation of models with all explanatory variables. Subsequently, variables deter- mined not to affect the probabilities of participation at usual significant levels were dropped and the models for each farm class were re-estimated. Estimated parameters for the 0-10 hectare class revealed that the location dummies, the level of labor/hectare and the level of credit per hectare were the most influential variables affecting participation in different farm systems. The individual farmer who is more likely to emphasize mixes involving food crops is characterized by his location in Vicosa, and by his relatively lower levels of labor and credit/hectare. Conversely, mixes where cash crops, dairy and off farm work gain prominence have a higher probability of being pursued by a farmer who is not in Vicosa, and whose levels of labor and credit/hectare are com- paratively higher. At the average level of the independent variables, "food and off farm work" appears as the more likely farm type to be selected, with a predicted probability of 44 percent. The effects of change in these average levels reinforced the importance of the location variables, inasmuch as the predicted probabilities varied in magnitude across regions to a large extent. Less labor/hectare increases the prob— abilities associated with "food only" farms, in the process diminishing those of the more frequent groups within the respective regions. More 202 credit/hectare, on the other hand, should promote participation in "cash and dairy" farms, and reduce sharply the likelihood of a farmer to engage in "food only" farm types. In the 10-50 hectare class, the relevant variables affecting the prob— abilities of participation are the dummy variable for Muriae, the sensitivity index, the levels of labor and credit per hectare and the size of the family. The indicator variable for Vicosa was kept in the reduced model specification to allow the breakdown of predicted probabilities into all of the three regions. Results of the estimation showed that the more likely farmer to belong to the "food and cash" group is that one who is not in Muriae nor Juiz de Fora, has a low sensitivity index, has a comparatively higher amount of credit per hectare, and whose household unit is small relative to the others. This farm type is also the one with the highest probability of participation, which is predicted at 35 percent. "Food and dairy" farm types are more likely to be pursued by Juiz de Fora farmers, with higher sensitivity indexes, low levels of both labor and credit/ hectare and relatively larger families. Farmers located in Juiz de Fora are also more likely to engage in "food and off farm work" farm types, given low sensitivity indexes and high levels of labor per hectare. "Cash only" farms were characterized only by a larger presence of Muriae farmers, while the only characteristic associated with "cash and off farm work" farms was their higher sensitiVity indexes. Changes in the independent variables illustrated the reduced importance of location factors in this case, since most effects were similar across regions. Increases in the sensitivity index showed that concerns with safety are a plausible factor associated with producer emphasis in more traditional farm types. As the ability to withstand loss increases, so does the probability of 203 participation in "cash and off farm work" and "food and dairy" farm types. At the same time, the probability associated with "food and cash" farms is reduced. This suggests the relevance of risk attitudes and perceptions on planning processes among these farmers. Concerning labor/hectare, changes in the average level of this variable generally resulted in con- sistent effects. The probabilities of participation in labor intensive farms are increased with more labor and vice versa. The likely effects of the PRODEMATA project were assessed by changes in credit/hectare. More credit should strengthen existing patterns of enterprise combinations. The project is not likely to encourage changes in this respect unless its measures are coupled with alternative actions such as the promotion of policies aiming at reduced marketing and production risks normally associated with high return enterprises. Finally, the initial estimated model for the 50-100 hectare farmers had only two significant variables at the ten percent level, namely the sensi- tivity index and the level of labor per hectare. Two other variables--the indicator for Muriae and the size of the family--were significant at the 20 percent level. These four were maintained in the re-estimation, along with the dummy variable for Vicosa, which is included to allow the break- down of predictions by all three regions. The results indicate that the more likely farmers to be in the "food and cash" group have comparatively lower sensitivity indexes and higher levels of labor/hectare. Juiz de Fora farmers are the more likely to participate in "food and dairy" farm types, given high sensitivity indexes, lower levels of labor/hectare and larger families. "Cash only" farms are characterized only by a larger participation of Muriae farmers, whereas higher sensitivity indexes increase the probability of participation in "cash and off farm work" farms. 204 The predicted probabilities of participation point out "cash only" as the most probable farm type to be pursued by an average farmer in this size range, with a probability of 40 percent. The predictions are responsive to changes in the levels of the independent variables. Changes in the sensitivity index show that "food and cash" systems are the more likely farm types to be selected by farmers who would be affected the most in an adverse year. Conversely, higher indexes increase the probability of "food and dairy" farms, while decreasing that of "food and cash." This is further evidence of the role played by risk attitudes and percep- tions on farm planning decisions. With respect to the labor variable, no new information is derived from the effects of changes in its levels. Probabilities of participation in more labor intensive enterprises increase consistently with higher levels of the variable and vice versa. In sum, the analysis of factors affecting participation in the impor- tant farm types of Zona da Mata suggests that changes in these factors through policy actions can be instrumental in the promotion of emphasis in target types. The overall responsiveness of the predicted probabilities to the introduced changes is encouraging evidence of the potential of such actions as components of broader development strategies. Implications and Policy issues The findings of the analyses performed in this study are suggestive of a number of important policy areas for consideration by development planners. Some of the relevant issues have already been pointed out in the presentation of results. These and other implications are brought into perspective in this section. 205 Perhaps the most striking evidence derived from the study is the marked difference in farming characteristics noted when observations are made across size classes. Although in essence the nature of the problems facing producers in Zona da Mata is the same, their extent and the mag- nitude of some tested policy impacts does vary among farm sizes. More specifically, differences were shown to exist across sizes in terms of farm incomes, enterprise emphasis, market orientation and input usage, to name a few. The main implication of such evidence is straightforward: there is a need for disaggregation of rural development policies in the study region such that fairly specific and well defined groups of decision makers can be targeted. Farm size appears as the most important differentiating factor, although dissimilarities were also shown to exist across income classes and to a lesser extent across regions. All these factors must be taken into account in the definition of target beneficiaries of specific policies. Based on the definition of producers by their sizes of property, let us examine first the policy options available for increasing incomes of the 0-10 hectare farmers. It was noted that there was little scope for increasing incomes of these producers via reorganization of their farm plans in such a way that the plans of high income farmers in the same size group would be reproduced. In fact, no significant differences in farm plans or management performance were observed among these pro- ducers. The only plausible factor differentiating incomes among them are the variations in size within their range. The logical sequel to this ob- servation is the sensitive issue of promoting changes in existing patterns of land distribution. Need for actions in this area is recognized by the National Institute for Settlement and Land Reform (INCRA), inasmuch as 206 the "modulo rural"--a measure of the minimum farm size needed to provide full employment and an adequate income for a family with four workers-~has been determined by lNCRA to be close to 30 hectares for the area of the study.3 Given the difficulties involved in the implementation of policies directed to such major structural changes, realistic measures should aim at areas where transformation can be brought about more gradually. From the results of this research, the most promising area for the short run increase of incomes of these smallest producers is the promotion of off farm job opportunities. In this respect, two major areas for policy evaluation arise. On the one hand, there is the option of promoting increases in labor demands from the non-agricultural sector of Zona da Mata. Tax incentives for the hire of unskilled workers by private con- cerns could increase employment opportunities for part of the producers in the area, particularly those located in regions where industry is gaining importance. Incentives could also be provided for the develop- ment of small scale industries covering sectors such as usual crafts, small commerce, basic food processing and the like. Additionally, public expenditure programs (say, road construction and maintenance) at times of low agricultural demand may be another relevant policy with both short and long run benefits. Still concerning the non—agricultural sector, proposals have been made in the past for more investments in training for provision of skills which would enable potential migrants to qualify for better paying positions in the urban areas.ll A program as such is 3Singh, op. cit., p. 14. ulbid., p. 63. 207 appealing, though its implications in terms of regional development policies deserve better appraisal. With respect to labor demands within the agricultural sector, one policy which could reduce the pressures on labor from the smallness of the land base is the development and promotion of equitable contractual arrangements for sharecropping or similar forms of lease agreements with larger land owners. Under proper supervision and enforcement such arrangements are likely to bring more land into production, at the same time contributing towards a better resource allocation by the smallest farm owners and the landless alike. A related area is the promo- tion of c00perative farming, and the economic and political feasibility of such an institutional arrangement deserves evaluation. in a more direct manner, the promotion of emphasis on labor intensive enterprises for the larger sized farms should result in increased demands for agri- cultural labor. Clearly, absorption of surplus labor from the smaller sized properties could be achieved by such a policy. Apart from the issues presented above, there are a few other impor- tant implications related to the 0-10 hectare farmer. Despite the severe resource limitations of these producers, they were found to be responsive to economic incentives and market opportunities. It was observed that their patterns of enterprise combinations vary from region to region within the sample area, and that changes in the levels of both credit and labor/hectare resulted in consistent changes in these patterns. As a policy instrument, credit should promote their participation in less tradi- tional farm types. But since usage of this input by the sample farmers is impressively low, there is a need to investigate the reasons for such a fact. Although credit use is expected to increase at later stages in the 208 implementation of the PRODEMATA project, the analysis of factors affecting credit use is recommended. The more that is known about the nature of these factors, the better are the prospects for a successful utilization of the project's incentives as means towards the change of traditional patterns of resource allocation among the smaller farm owners in Zona da Mata. Still concerning the role of the PRODEMATA project, the provisions for evaluation and dissemination of technological improvements contained in the research and extension component indicate another area for policy consideration. We have noted that existing patterns of managerial per- formance did not differ among income classes for producers in the 0-10 hectare stratum. Accordingly, there is little scope to diffuse improved methods based on existing farming practices, as it can be proposed for farms with more than ten hectares. This observation suggests that research on improved technologies should have a stronger orientation towards the needs of this group of small scale producers. However, gains in income from such improved technologies are likely to be limited, given the low resource base of these producers. Turning to the 10-50 hectare farmer, most of the observations made above would apply for producers at the lower side of the size distribution within this range. In fact, it has been noted that similar problems of surplus labor are being faced by some producers in this group. Though the magnitude is much smaller than in the 0-10 hectare class, the need for promotion of employment alternatives as discussed above is also present. But, unlike the former size class, there is potential for increasing incomes of farmers in this range via reorganization of farm plans. Since farm types which could be looked upon as targets for 209 increased participation by these producers already exist in the area, the prospects for a successful reorganization strategy are promising. If there is interest in promoting emphasis in less traditional farm types, a crucial issue to be dealt with will be the explicit consideration of the risk and uncertainty factor. From the indications of the analysis, it can be argued that reduction of risks normally associated with farm plans which attribute heavier emphasis to cash crops and other non-tradi- tional enterprises should result in more participation in these plans. in this regard, usual measures such as price supports or crop insurance schemes appear appropriate. However, it should be mentioned here that very little is known empirically about risk preferences and attitudes of small farmers, or even about the riskiness of alternative enterprises in the Zona da Mata region. The benefits of more investigation on these aspects should be substantial. Indeed, increased knowledge about the risk question is a necessary prerequisite for the promotion of emphasis on target farm types. The results of the analysis of factors affecting selection of enter— prise combinations also evidenced a limited role for credit as an instru- ment to promote changes in existing patterns of enterprise emphasis. The most frequent farm type in the class (food and cash) had its prob- ability of participation sharply increased as more credit was made avail- able. The implications of this observation are twofold. First, it suggests that credit availability per se is not a relevant constraint in precluding higher participation in less traditional farm types. More likely, risk preferences are the overriding consideration in this respect. Second, it may be that an increased emphasis on this traditional farm type will result in higher levels of marketable surplus, such that income gains 210 coUld be attained. Therefore, the potential of credit as a mechanism for increasing incomes of these producers will depend on the extent to which its provision is coupled with policies such as risk reduction. By itself, more credit may increase incomes of these producers on the average only if the marketable surplus of the most traditional farms increases as well. The observed differences on farmer performance in the analysis of potential net margin points to another important policy area. High income farmers in the 10-50 hectare class were shown to have signifi- cantly higher performance indexes than their medium and low income counterparts. Since there have been a number of studies on management efficiency in the study area, a basis for the design of actions to improve farmer performance can be derived from a review of these works. Addi- tionally, the data base of the PRODEMATA monitoring and evaluation component can be utilized to provide further insights on the factors affecting the observed performance dissimilarities within farm size classes. The extension component of the project can then utilize the derived infor- mation to diffuse improved farming procedures among low income producers. in order to improve incomes of the third size class considered in the research, some of the observations made above will apply as well. Specifically, the role of safety concerns in the selection of farm types was also shown to be relevant among the 50—100 hectare farmers. Since reorganization of farm mixes to reflect those of high income producers in the class can lead to higher incomes, the discussed policies aiming at reduction of risks should produce the desirable results. Moreover, the observed differences in farmer performance also suggest the need for consideration of this important aspect in an income enhancement strategy. 21 1 Although farmers in this group were generally better off than those with less than 50 hectares, there are still a number of them with absolute income levels‘which are low enough to justify their inclusion in such a strategy. Additionally, it was noted that these producers have a favorable potential as absorbers of surplus agricultural labor from the smaller farm properties. If emphasis on labor intensive cash crops is to be promoted as suggested in the discussion of policy options for the 0-10 hectare farmers, then these producers should receive special consideration. An important difference among the 50—100 hectare farmer and those with less than 50 hectares of land is the effect of credit on farm enter- prise selection. lt was evidenced from the study that no significant effects on their participation in alternative systems can be attributed to the levels of credit/hectare. This implies that this variable cannot be regarded as an instrument to promote changes in enterprise emphasis among these producers, given current credit policies. The implementation of strategies such as strict supervision of credit use, or the provision of preferential interest rates for target enterprises, is needed to strengthen the role of credit as a policy tool in this respect. The policy issues outlined in this section suggest that there is potential to improve income levels in the Zona da Mata region by means of appropriate strategies which do not require substantial structural changes. Of course, their cost/benefit characteristics need further evaluation. Furthermore, the implications of such proposed actions at more aggregated levels must be considered before implementation. In particular, the effects of increased emphasis on specific enterprises on supply and prices should be taken into account on policy evaluation. 212 Since supply shifts likely to be caused by increased emphasis may lead to lower prices, there will be a need to evaluate demand conditions to ascertain the extent to which these effects may offset potential income gains. Limitations and Suggestions for Further Research This study has adopted an analytical framework which is flexible enough to allow the inclusion of several improving modifications. Since the framework was found adequate, improvements would be essentially achieved by refinements in the data base. Availability of timely information was indeed the single most important constraint on the definition of both dependent and independent variables in the model. Accordingly, a possible extension of the study would be the devising of a less aggregative definition of the relevant farm types. This would require a thorough review of current farm typification methodologies followed by Brazilian researchers, and in all likelihood it would entail a need for more detailed data on input utilization and dis- position of output. Lack of data on several of the aspects depicted on the schematic presentation of a farmer's decision process (Figure 6.1) also reduced the explanatory power of the model to some extent. To exemplify, we have been unable to account for variations in agronomic conditions (such as land quality) in the analysis of enterprise choice. Additionally, it was impossible to include variations in labor use during different periods of the year. Thus, possible complementarities in labor use which could account for preferences for some crop mixes could not be identified. Perhaps most importantly, it was not feasible to account for risk preferences 213 in a more explicit way. These and other important aSpects should be considered in further applications of the model. There are also a number of areas where additional research may generate benefits. The most important ones have already been pointed out in the discussion of policy issues. Recalling, they included an analysis of risk attitudes and preferences among area producers, a study of factors affecting use of credit and an investigation of the effects of increased use of this input on expansion of marketable surplus. More- over, a review of existing studies in the area of management efficiency was suggested as an initial step towards the understanding of observed dissimilarities among income groups within two of the size classes. In addition to these aspects, there are grounds to suggest a few alternative research issues. First, because of the availability of survey data on a yearly basis, it seems feasible to bring into a model of enter- prise choice the all important time element. The comparative static framework utilized in this study can be considerably improved if dynamic elements are introduced. It should be especially interesting to evaluate the effects of the PRODEMATA project on changes in farming patterns over a longer time period. Secondly,, it should also be relevant to investigate the effects of the proposed changes in farm plans on patterns of income distribution. The underlying hypothesis of the suggestions was that income levels of the low income producers would raise with their adoption of high income enterprise mixes. Given the large concentration of income among larger scale producers, it seems appropriate to investigate the extent to which the existing situation would be changed. In this respect, a related area for investigation is the analysis of the effects of the project on 214 income distribution. There have been charges that credit projects such as the one currently under implementation in Zona da Mata tend to aggra- vate disparities in income distribution.5 Thus, such an analysis assumes substantial importance. Since this study did not have the necessary elements for a precise evaluation of the role of marketing conditions on the decision processes appraised, a third area for further investigation emerges. One variable included in the model as a surrogate for possible differences in marketing costs--the distance from the farm to the municipalities--was not signifi- cant in any of the farm classes. On the other hand, there are indications from the significant impact of the location variables (R2 and R3) in smaller farmer decisions that differences in marketing conditions may in fact exert an influence on farm enterprise choice. However, the information available from the adopted proxies is limited and there is a need for better measures if a more adequate analysis of this important aspect is to be made. In particular, there is a need for more detailed data on such key aspects as timing and location of sales. Finally, in view of the strong correlation between farm plans and income levels in all three farm classes,6 it should be relevant to examine the economics of the introduction of new enterprises in the area. In particular, enterprises which would potentially contribute to a better utilization of labor inputs should receive priority in this regard. 5Cheryl Payer; "The World Bank and the Small Farmers ;" Journal of Peace Research 16 (4), 1979, pp. 293-312. 6Although farm plans did not differentiate the 0-10 hectare farmers by income classes, they were correlated with absolute income levels. See Chapter V for details. APPENDICES APPENDIX A AVERAGE LEVELS OF INPUT UTILIZATION 215 m5 N.E. m.— ll ll Iiii ll.- Iiii Ill. ¥UOme>_-. 5.: as: ea -- -- -- -- -- -- >~: -- -- ...2 ...em 93m .1: e52. -- 8.2» $05 mmzho -- II «.2 Re mg -- rim. -- ~52... mz coco; scion—2 Lamon— uc oaaoeooc ncm c3325: «sac. Co 205-. ommeo>< — . < Boa-P 216 m .2 e .2... m .e -- -- -- -- -- -- 8.08.25... N .m .. .2... a .3 -- -- -- -- -- -- .225 ii i..- c .c— m . Km Ii- .i-i ..i-i ili l-i 9:31“. -- -- 3.... ~12 -- 2...... 3.3 22 .1: mmam -- -- 22 ...m. .12 ex: 22.. -- RB meomo mmzho -- -- ...m... 22 ..2 -- «.8 -- 2m. mz «coon. cmE m_mE_c< to 320 313.5. .3095 mac—00.; Loam-d . -chm_2 Aenea— eodztou— e mooom 3:38: :7: cot—2:3: Sec. .3 302.1. 00983.. N . < 03:. 217 ..... .22 c ... -- -- -- -- -- -- 0.00.5020 ...m. 3.2 22 -- -- -- -- -- -- >m.c2c:.80. 5:80. 3:80. .3520 0.595 3:80. 3.25.. 0.50 .0> mooou cm... m_mE_c< .Co 3.0.0 Aer—3.5. .muoocm moc_000> cone-— -c_.._0m.2 utmom .8255“. w mtoom no... 300: 2.. la— coznzzu... Sac. ..o 202.... ommeo>< m .< Each 218 .. .2 m .52 .... ..-- --- -- -- -- -- 50502.. 2... 233 o... -- --- -- --- -- -- 5.3.5 --- -- ..m --- -- ---- .. ...mm -- --- 953.“. -- -- 8:. -- ...m: -- 88 ...2. ....... mmqm<500> -- -- 3...... 2.... 22. 28 22.2. 22 0...... 9.9.0 .315 --- -- ...N ....... ...am. -- 2:... 2... ...m. mz<0 «<03 -- --- 2....” «z... e. -- 282. 2.2 2.2 mmumoo --- ---- ..Z 22. .4... --- 2.... 2.... 22 m0... -.... -...... m...~ a... mém ...i- --.. -- 23. 50m -..- -..- 2: 2.. -- -- -- 2.... 23. 0.9.5- -- -- a... ...2 ..... mg 23 .... ...2. 02.9.23... --- -- ...N. ...Z a... 2N. a... ...... 22. mz..>..u< 0.80 .0> £000“. :0... 0.0.5.5. >3 00.0.0 313.0. .3095. m0£00~> .30.. union-z nzmmn. .02—2.5". w mnumm 09.300: 2.7:... £02355 «an... ..o m_0>o-_ am~..v>< a . < 0.00... 219 N . .. .. ..... .. .~ ---- -- ---- -- ---- -- «00502.. 0... 0......” 0K -..- -- -- -- -- ..-- >~: -- -- ...m 82. ........ -- 8...... 82.. .32 8.0«0 «01.5 ---------- 2... 2m... 8...: -- 88.. .8... --- mz<0 «<03m --- -- 8.... 82 .4... -- 83.... 8.3 8m... mmuuou --- --- ..2 ....... 8.... ...N ...e... 8.. 8.... m0_« --- --- 8.. I: 8.... -- --- 2.... ...N... San ---- -- 8.: --- 2..... 82m 8... -- 822 29.3.. --- --- 8.. 8.... 2N it 8.... ...2 8...: 023222... ---.. --- 2.. 2.... 22 --- 2.... 8.... ........ mz woman. :0... £2.52 >3 320 2.13.0. .3093. m0£00m> .30.. unions. :33... .0253“. w mvmom 3.300... ecu-2.. cot—2:3: «an... be m_0>0-_ wow-.03: m .< 0.9m... APPENDIX B COMPUTATION OF THE INDEXES 0F POTENTIAL AND PERFORMANCE FOR A SAMPLE FARM 220 $2.. .... u 2.. x 3.3 23.5.9. .o .80... I 3.: R I 2.— x ~mo.~ “wucmsgazm .8 x20... EBmNJ H95.0 9.. .8 9.2301 0.5255 m:......8~ 3533. 3.5.9. aimed .2355 2.33. .39.. a: 8... 3.2m 5...". 8...... m... 2. 9:... cm... 8. {oz 5...“. to N..N... S8~ 2.. .:.< ...m x833... 58m 2. .. 28. .:.< 8 i=5 2..... ....... a... a: 8. 23m Sow 2.8.. ..R .. 2:3 2. 8.. :30 29.... 3.0. 3.5.3.0. 3.53.0. 3...: 3.5.35 mCLDHOM— mcgauwm mCL—auom 335.9. 0.853.... .86.. mum-U 0.200... cm no. a... c. 5.5.... 9.50:. 82005. ...N .3532 «2.0 5.0... 0.953 a .8 00:05.0...om 0:... 33:33. .0 3.80:. 05 mo CESSQEOU —.m 030... APPENDIX C ENTERPRISE EMPHASIS BY REGION 221 II ..2 ..2. II ...mm ..2. v3.0... .52“. ..o ...3 ma... 92 ...S m... ...N... 209.32.. ...8 «.3 a... ...8 mac n.3, >m. ..m ...N ...N ..m ..m m.~ mmomu mmzho II II II ms: ...N a... mzz>20< 0.9.3 “Sta“. .0 w “amazon...— m. >z>$u< 9.033 2.3... ..o w 9.0.. 0.. 2..... 20501 >9 m_mm..QEm mm..a.3:m —.U 0.0m... 222 ..N ..2 «.«m «..: ...«~ «.3 v5.02. 2...... ..o ...m «...N «.2 T... ««.. «.8 505.3... «.8 «.2 «... «..... ..S ...«« >53 II .. «.. II «... «... «:3... II m .« « .. ..« ... « .. mm4m II .... «.. ...~ «.« «.. «.96 5...... II II ..«. ...m «.m .K «25 «(03 ..«« «... ..2. «..... «.S ...«m mutou «..: «.«. ..Nm ....« «.9... ..8 mo... II II II II ..~ II 52. II .. «.. ..N «... «.. «.8 II ..N « .. m ...N .. .N « .«m uoaaEu.m.c. II «.N «.« «.2 «..... 5...... «2.3... ...m N... «.. «.«. «... ... 53. II «.N. «.« «...N «..: «..: 22m ..« N .« .. .m « . .~ ...~« « .~.. .83888... «.m. ..«N ..«. «..... ...2 «..... 2:8 2. ««..«m a: «m I... a: «I. a: 8...... a: «... -8 a: «..... >..>..u< anzflooam m. >z>=u< 32.3 2:...”— uo w «£3.09... a. >2>=u< 0.9.3 main... ..o m 03.32 .533. >9 m_m9._QEm 93.9.3.5 N . U 032. 223 .... II ....N ...: N..N ««.. v3.0: .52.. ..0 ....«N ...«N N.« «.«m «.8 «...... .0055... ....« .... N.« N.«. ...... ..N. >m.<0 II II .... II N .. .... «.50... N.. II ...N N... «.m .... «0.5.308; .... «.N. N.« ...: «..: N.« «.05 «01.0 «... . «... .... ...: «.«. «.... mz<0 «<03 ...... «.«m ...«. ...N« ...... «.«N HES ...N. ...N. «.... ««... «..m ....« 00... II II II N... ...N ...N .08 N .. II II N .. m .« II «.8 « .« II .... . . .. ...... «.«« 08.8.82... ... N... .... «..... ...... «..... mz..>..u< 002.500.; m. 5.3.0.. 0.0:; £50”. .0 m “05.00.... m. >z>=0< 0.053 05.0“. .0 w 0m00.> 00301 .5 «.mEQEw 0m..0.0.:m m .U 039—. BIBLIOGRAPHY BIBLIOGRAPHY Amemiya, T. "Qualitative Response Models;" Annals of Economic and Social Measurement (ll), 1975, pp. 363-372. Anderson, Jock; Dillon, John; and Hardaker, J. Agricultural Decision Analysis. Ames: Iowa State University Press, 1977. Anderson, Jock, and Hardaker, J. "Economic Analysis in Design of New Technologies for Small Farmers." Economics and the Desi n of Small Faring Technology. Edited by AlEerto VaTdes, Grant Scobie ant—{John Dillon. Ames: Iowa State University Press, 1979. Bastos, Edinaldo. "Farming in the Brazilian Sertao: Social Organization and Economic Behavior." Unpublished Ph.D. dissertation. Cornell University, 1980. Beckman, George. "Transforming Traditional Agriculture: Comment." Journal of Farm Economics 48 (ll), 1966, pp. 1013-1015. Cesal, L. and Bandeira, A. "Uso da Terra na Zona da Mata de Minas Gerais." Estudos Sobre uma R iao A ricola: Zona da Mata_d_g Minas Gerais. Monografia—No. 9. IO déTaneiro: lPEA, 1973. Clark, C. and Schkade, L. Statistical Analyis for Administrative Decisions. Cincinnati: S'outTa Western Company, 1974. Cleave, John. "Decision Making in the African Farm." Contributed paper read at the 16th International Conference of Agricultural Economists. Oxford: Oxford Institute of Agricultural Economics, 1977. Crocomo, Celso. "Risk Efficient Fertilizer Rates: An Application to Corn Production in the Cerrado Region of Brazil." Unpublished Ph.D. dissertation. Michigan State University, 1979. Dillon, John. "Broad Structural Review of the Small Farmer Technology Problem." Economics and the Design of Small Farmer Technology. Edited by Alberto Valdes, Grant Scobie and John Dimm. Ames: Iowa State University Press, 1979. Dillon, John and Anderson, J. "Allocative Efficiency, Traditional Agriculture, and Risk." American Journal of Agricultural Economics 53 (1), 1971, pp. 26-32. Dillon, John and Scandizzo, P. "Risk Attitudes of Subsistence Farmers in Northeast Brazil: A Sampling Approach." _AI_nerican Journal of Agricultural Economics 60 (3), 1978, pp. ll25-335. 224 225 Domencich, Thomas and McFadden, Daniel. Urban Trafl Demand: A Behavioral Analysis. New York: North Holland, 1975. Eicher, Carl. Research on Agricultural Development in Five En lish Speaking Countries in West Africa. New York': AgricuTtura Development Council, 1970. Franzel, Steve. "Enterprise Choice, Enterprise Combinations and Income Distribution Among Farmers in Sierra Leone." Unpublished M.S. thesis. Michigan State University, 1979. Fundacao Instituto Brasileiro de Geografia e Estatistica. Anuario Estatistico do Brasil. Rio de Janeiro: IBGE; 1955, 1965, 1975 and 1978. . Censo Agropecuario de 1975 - Minas Gerais. Rio de Janeiro: IBGE, 1979. . Censo Demografico de 1970 - Minas Gerais. Rio de Janeiro: IBGE, 1972. . Informacoes Basicas dos Municipios - Minas Gerais. Rio' de Janeiro: IBGE,1978. Garcia, Joao Carlos. "Analise da Alocacao de Recursos por Proprietarios e Parceiros en Areas de Agricultura de Subsistencia." Unpublished M.S. thesis. Universidade Federal de Vicosa, 1975. Garcia, Philip. "Market Linkages of Small Farms: A Study of the Maize Market in Northern Vera Cruz - Mexico." Unpublished Ph.D. dissertation. Cornell University, 1978. Goldberger, A. Econometric Theoq. New York: John Wiley and Sons, 196“. Greco, Mussolini. "Determinacao da Renda das Empresas Rurais em Relacao a Melhor Combinacao de Seus Empreendimentos Basicos." Unpublished M.S. thesis. Universidade Federal de Vicosa, 1972. Gudeman, Stephen. The Demise of a Rural Economy: From Subsistence to Capitalism in a—Eatin American Villa; London: Routledge and Kegan Paul, 1978. Haberman, Shelby. Analysis of Qualitative Data, Vol. II. New York: Academic Press, 1979. Hanushek, Eric and Jackson, John. Statistical Methods for Social Scientists. New York: Academic Press, 1977. Hardaker, John. "A Review of Some Farm Management Research Methods for Small Farm DeveIOpment in LDC's." Journal of Agricultural Economics 30 (1), 1979, pp. 315-331. Hausman, Jerry and Wise, David. "A Conditional Probit Model for Qualitative Choice: Discrete Decisions Recognizing Interdependence and Heterogeneous Preferences." Econometrica 146 (2), 1978, pp. .103-1126. ' 226 Helleiner, Gerald. "Smallholder Decision Making: Tropical African Evidence." %griculture in Development Theory. Edited by Lloyd Reynol s. New Haven: Yale University Press, 1975. Heyer, Judith. "A Linear Programming Analysis of Constraints on Peasant Farms in Kenya." Food Research Institute Studies 10 (1), 1971, pp. 55-67. Hopper, David. "The Economic Organization of a Village in North Central India." Unpublished Ph.D. dissertation. Cornell University, 1957. Informe AgrOpecuarlo 53 (5), 1979, p. 79. Kakwani, N. and Podder N. "Efficient Estimation of the Lorenz Curve and Associated Inequality Measures from Grouped Observations." Econometrica MI (1), 1976,.pp. 137-1&8. King, Robert and Byerlee, Derek. Income Distribution, Consumption Patterns and Cfiqnsumption Linkggfs in Rural Sierra Leone. African Tural Economy Paper No. 16. East Lansing: Micl'figan State Univer- sity, 1977. Knight, Frank. Risk, Uncertainty and Profit. Chicago: University of Chicago Press, 1921. Lipton, Michael. "The Theory of the Optimizing Peasant." Journal of Development Studies II (3), 1968, pp. 327-351. Maddala, G.S. Econometrics. New York: McGraw Hill, 1977. Marquardt, Donald. "An Algorithm for Least Squares Estimation of Non—Linear Parameters." Journal of the Society for Industrial and Applied Mathematics 2 (2), 1963. Matlon, Peter. Income Distribution Amogg Farmers in Northern Ni eria: Empirical ResuLtLandPolicy Implications. African Rural Economy Paper No. 18. East Lansing: Michigan State University, 1979. McFadden, Dam... "Conditional Logit Analysis of Qualitative Choice Behavior." Frontiers of Econometrics. Edited by Paul Zarembka. New York: Academic Press, 19711. McKelvey, Richard and Zavoina, William. "A Statistical.Model.for the Analysis of Ordinal Level Dependent Variables." Journal of Mathematical Sociology ll (1), 1975, pp. 103-120. Mellor, John. "Relating Research Allocation to Multiple Goals." Resource Allocation 32d Producthrity in National and International Agricultural Research. Editedby Thomas ArndT, Dana Dalrymple, and Vernon Ruttan. Minneapolis: University of Minnesota Press, 1977. Mels, Carlos and da Silva, Cicero. "Culturas Intercalares." Informe Agropecuario an (11), 1978, pp. 70—71. 227 Moscardi, Eduardo and de Janvry, Alain. "Attitudes Towards Risk Among Peasants: An Econometric Approach." American Journal of Agricultural Economics 59 (II), 1977, pp. 710-716. Moser, C. and Kalton, G. Survey Methods in Social Investigation. New York: Basic Books, 1972. Nagornni, Jacksonwilliam. "Analise da Eficiencia Tecnica e da Substituicao da Fatores na Agricultura de Baixa Renda da Zona da Mata de Minas Gerais." Unpublished M.S. thesis. Universidade Federal de Vicosa, 1980. Nerlove, Marc and Press, S. James. Univariate and Multivariate Lo -Linear and L istic Models. R-13063EDA7Nil-I. Sinta Monica: Tlfie Rana Corporatio,I n, 1973. Nie, N.; Hull, C.; Jenkins, J.; Steinbrenner, K.; and Bent, D. Statistical Packa e for the Social Sciences. New York: McGraw Hill, 1975. Norman, David. The Farmin Systems Approach: Relivancy for the Small Farmer. U RuraFDevelopment Paper No. 5. East Lansing: Michigan State University, 1980. . "Rationalizing Mixed Cropping Under Indigenous Conditions: The Example of Northern Nigeria." Journal of Development Studies 11 (1), 197“, pp. 3-21. Norman, David; Pryor, David; and Gibbs, ChristOpher. Technical Change and the Small Farmer in HausalandL Northern Ni eria. African Rural Economy Paper No. 21. fist Lansing: Michigan State University, 1979. Oliveira, Evonir. Personal communication. Vicosa, May 1980. Olukosi, James. "The Distribution of Personal Incomes Among African Farmers--A Two Period Analysis." Unpublished Ph.D. dissertation. Michigan State University, 1979. Paiva, Ruy; Schattan, Salomao; and Freitas, Claus. Setor A ricola do Brasil: Comportamento Economioo, Problemas e Possibilidaaes. Sao Paulo: Secretaria da Agricultura, 1973. Panagides, Stahis and Ferreira, Leo. "Absorcao de Mao de Obra na Agricultura da Zona da Mata de Minas Gerais." Estudos Sobre uma Regiao Agricola: Zona da Mata de Minas Gerais. Monografia No. 9. IO de Janeiro: lPEA, 1973. Payer, Cheryl. "The World Bank and the Small Farmers." Journal of Peace Research 16 (II), 1979, pp. 293-312. Pindyck, R. and Rubinfeld, D. Econometric Models and Economic Forecasts. New York: McGraw Hill, 1976. 228 Pretzer, Don and Finley, Robert. "Farm Type Classification Systems: Another Look at an Old Problem." American Journal of Agricultural Economics 56 (1), 19711, pp. 1115-189. Ravenholt, Albert. "The Gods Must Go! " Selected Readin s to Accompany Getting Agriculture Moving, Vol. I. Edited By Raymond Borton. New York: The Agricultural Development Council, 1966. Roumasset, James. Rice and Risk: Decision Making Among Low Income Farmers. Amsterdam: North Holland, 1976. . "Risk and Choice of Technique for Peasant Agriculture: The Case of Phillipine Rice Farmers." Unpublished Ph.D. dissertation. University of Wisconsin, 1973. Scandizzo, P. and Dillon, J. "Peasant Agriculture and Risk Preference In N. E. Brazil: A Statistical Sampling Approach." Paper presented at the CIMMYT Risk Conference. EI Batam, Mexico, 9/15 March 1976. Schluter, Michael and Mount, Timothy. Management Objectives of the Peasant Farmer: An Analysis of Risk’Aversion in the Choice of Cro in Pattern, Surat District, India. Occasional Paper No. 78, [repartment ofIgricuItural Economics. Ithaca: Cornell University, 197II. Silvey, S.D. Statistical Inference. London: Chapman and Hall, 1978. Schmidt, Peter and Witte, Ann. The Economics of Crime (provisory title). New York: Academic Press, forthcoming. Schmidt, Peter and Strauss, Robert. "The Prediction of OCCUpation Using Multiple Logit Models." International Economic Review 16 (2), 1975, pp. 471486. Schultz, Theodore W. Transforming Traditional Agriculture. New York: Arno Press, 1976 (Reprint of original 196“ editionf Schurle, Bryan and Erven, Bernard. "The Trade-Off Between Return and Risk in Farm Enterprise Choice." North Central Journal of Agricultural Economics 1 (I), 1977, pp. 15-21. Singh, R.; Patrick, G.; Schuh, E.; and Kehrberg, Earl. Poor Rural Households, Technical Change and Income Distribution in LDC's, Brazil: A Summary. Department of Agricultural Economics. "Test Lafayette: Purdue University, 1979. Soares, Augusto. "Resource Allocation and Choice of Enterprise under Risk on Cotton Farms in Northeastern Brazil." Unpublished Ph.D. dissertation. Ohio State University, 1977. Stevens, Robert. "Transformation of Traditional Agriculture." Tradition and Dynamics of Small Farm Agriculture. Edited by Robert Stevens. Ames: Iowa State University Press, 1977. 229 Tax, Sol. Penny Capitalism. Chicago: University of Chicago Press, 1963. Theil, Henry. "A Multinomial Extension of the Linear Logit Model." International Economic Review 10 (3), 1979, pp. 251-259. Thompson, Stanley and Eiler, Doyle. "A Multivariate Probit Analysis of Advertising Awareness on Milk Use." Canadian Journal of Agricultural Economics 23 (1), 1975, pp. 65-73. Universidade Federal de Vicosa, Departamento de Economia Rural. Dia nostico Econo_mico da Zona da Mata. Vicosa: lmprensa Universitaria, 1971. . Programa Integrado de Desenvolvimento da Zona da Mata - MG: Primeiro Rgiatorio Anual de Avaliacao. Vicosa: lmprensa Univer- siteria, 1979. Upton, Martin. "A Development of Gross Margin Analysis." Journal of Agricultural Economics 16 (1), 1964, pp. 111-116. Farm Management in Africa. London: Oxford University Press, 1973. Upton, Martin and Petu, D. "An Economic Study of Farming in Two Villages in lllorin Emirate." Bulletin of Rural Economics and Sociolo 1 (1), 1964. Department of Agricultural Economic , University of Ibadan. Vieira, Clibas. "Cultivo Consorciado do Milho com Feijao." Informe Agropecuario 46 (4), 1978. pp. 42-44. Von Neumann, J. and Morgenstern, 0. Theory of Games and Economic Behavior. Princeton: Princeton Unfi/ersity Press, 1947. Wharton, Clifton. "Risk, Uncertainty and the Subsistence Farmer." Paper presented at the joint session of the American Economic Association and the Association for Comparative Economics. Chicago, December 28, 1968. White, Jr., T. and Seabra, D. "Credito Agricola na Zona da Mata de Minas Gerais." Estudos Sobre uma R iao A ricola: Zona da Mata de Minas Gerais. Monografia No. 9. Rio 3e Janeiro: IPEA, 1973. Witte, Ann and Schmidt, Peter. "An Analysis of the Type of Criminal Activity Using the Logit Model." Journal of Research in Crime and Delinguengy ('1), 1979, pp. 164—179. World Bank. BRAZIL - Staff chject Report of the lnte rated Rural Development Project in the State of Minas Gerais. Report No. 1291 Br. VTashfigton , 1976 imimeographed) . 230 Wrigley, Neal. "Analyzing Multiple Alternative Dependent Variables." Geographical Analysis 7 (4), 1975, pp. 187-195. . "Introduction to the Use of Logit Models in Geography." Concepts and TechnicLues in Modern Gem (10), 1976, pp. 1-32. Yang, W. Methods of Farm Mana ement Investigations for Improving Farm Productivity. Rome: Fifi, 1971.