Ecological Economics 93 (2013) 330–341

Contents lists available at SciVerse ScienceDirect

Ecological Economics

j ourna l homepage: www.e lsev ie r .com/ locate /eco lecon

Analysis

Organic certification, agro-ecological practices and return on investment:Evidence from pineapple producers in Ghana

Linda Kleemann a,⁎, Awudu Abdulai b

a Kiel Institute for the World Economy, Kiel, Germanyb Depratment of Food Economics and Consumption Studies, University of Kiel, Germany

⁎ Corresponding author. Tel.: +49 4318814249.E-mail address: linda.kleemann@ifw-kiel.de (L. Klee

0921-8009/$ – see front matter © 2013 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.ecolecon.2013.06.017

a b s t r a c t

a r t i c l e i n f o

Article history:Received 18 December 2012Received in revised form 18 June 2013Accepted 18 June 2013Available online 8 July 2013

JEL code:O13Q13Q17Q56

Keywords:Organic agricultureCertificationAgro-ecological practicesReturn on investmentImpact assessment

The recent empirical literature on economic sustainability of certified export crops shows that certificationstandards that enhance yields are important for improving farm revenues and household welfare. However,limited evidence exists on the impact of organic certification on the adoption of agro-ecological practices. Inthis study, we use unique farm-level data from pineapple producers in Ghana to examine the impact of organiccertification on the use of agro-ecological practices such as organic fertilizers, organic pest andweed control, croprotation, and soil and water conservation, as well as how using these measures affect farm outcomes such asreturn on investment. Our empirical results reveal that organic certification increases agro-ecological practiceuse, although from a very low starting point. Using a generalized propensity score approach, we show thatthere is a positive, but nonlinear relationship between the intensity of agro-ecological practice use and returnon investment.

© 2013 Elsevier B.V. All rights reserved.

1

1. Introduction

Concerns over climate change and increasing pressure on land haveresulted in increasedpromotionof sustainable productionmethods thatincrease yields, while protecting the environment as well as increasingthe resilience of crops to climatic change (Branca et al., 2011; Erenstein,2002; FAO, 2011; Kassam et al., 2012; Knowler and Bradshaw, 2007).Such sustainable production methods form part of organic agricultureprinciples, but in practice, low-input productionwith none, or very littlesustainable soil and water management practices are frequently certifiedas organic in many developing countries (see e.g. Blackman and Naranjo,2012; Bolwig et al., 2009).

To encourage the adoption of sustainable production methods,national governments, NGOs and international donors have promotedthe marketing of export crops through certified marketing channels,mostly through farmer-based groups, as an attractive business modelfor smallholders in developing countries (Beuchelt and Zeller, 2011;Hattam et al., 2012). These sustainable certification schemes have be-come increasingly popular in many countries because they combinevalued traits that are related to the environment, poverty alleviation,and health outcomes into a single commodity (Barham and Weber,

mann).

rights reserved.

2012; Hattam et al., 2012). Although organic certification is currently ex-port oriented in most African countries, this is expected to change with arising middle class, as domestic demand for such products increases(Probst et al., 2012).1 Consumers generally show their preferences forsuch products by paying higher prices to support an environmentallyhealthy world. However, the success of these schemes depends to alarge extent on prices received and incomes earned by the farmer.2 Forexample, Bolwig et al. (2009) and Valkila (2009) find in their studiesthat higher incomes from organic farming are entirely due to higherprices received, but not lower costs of production.

The significance of these schemes in promoting sustainable farmpractices and improving the incomes of smallholders in developingcountries has attracted the attention of many policy analysts over thelast few years. In particular, several studies have examined the impactsof certification schemes on farm outcomes such as farm revenues,profits, and household poverty (Barham and Weber, 2012; Beucheltand Zeller, 2011; Bolwig et al., 2009; Ninan and Sathyapalan, 2005;Pretty et al., 2006; Valkila, 2009). Most researchers findmodest positiveimpacts of organic certification on farm revenues and household

Certification in this paper refers to EU organic regulation. Certifying agencies in-clude IMO, Ecocert and Soil Association.

2 As noted by an anonymous reviewer, organic production could also contribute toincreased income through higher output.

331L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

income, using various measures and econometric approaches. Theyattribute the positive impacts of certification to price premiums thatare paid at least for part of the crop sales (e.g. Bolwig et al., 2009; ITC,2011; Valkila, 2009), although it is usually not clear whether the effectcomes from certification, contract farming, or export market access. Itis important to note that some researchers have been rather skepticalon the ability of certification to lift farmers out of poverty, giventhe usually low revenue increases. Themain reasons for this skepticismare the high certification and investment costs involved in the process(Beuchelt and Zeller, 2011; Calo and Wise, 2005; Valkila, 2009).

Despite this increasing number of impact assessment studies, veryfew studies have considered the environmental outcomes of differentcertification programs or the adoption of sustainable farming techniquesfollowing certification (Barham andWeber, 2012). Philpott et al.'s (2007)study on Mexico examines environmental outcomes by analyzing theimpact of certification on vegetation and ant, as well as bird diversityin coffee farms and forests. Rather surprisingly, their findings show nodifferences between in vegetation characteristics, ant or bird speciesrichness, or fraction of forest fauna in farms based on certification.Pretty et al. (2006) conduct a review of 286 interventions to showthat the use of resource conservation practices increases productivityon developing country farms, albeit using best practices. Bolwig et al.(2009) studied the effect of organic contract farming and adoption oforganic practices on 112 coffee producing smallholders, where organicfarmers apparently use more organic practices, and conclude that thereare somewhat higher revenues for farmers that adopt organic farmingtechniques, findings confirmed by Blackman and Naranjo (2012) whofind that organic certified farmers in Costa Rica use less chemicals andadopt some environmental friendly management practices, basing theiranalysis on only 35 certified coffee farmers in Costa Rica.

With the notable exception of the study by Pretty et al. (2006), whichincludes some countries from sub-Saharan Africa, we find no empiricalevidence on the impacts of certified organic farming on environmentaloutcomes in sub-Saharan Africa. In particular, the dependence of theyield impact of organic certification on the intensity of agro-ecologicalpractice use has been hardly studied in the existing literature. Someauthors have argued that organic farming in Africa mostly implies thenon-application of chemical inputs, without necessarily adopting alterna-tive soil fertility management practices (e.g. Ndugire, 2010; UNCTAD,2008a). This is particularly so for the many small-scale farmers inAfrica, who are assumed to produce traditionally, or “organically bydefault”, using virtually no external inputs (UNCTAD, 2008b; van derVossen, 2005). For example, farmers and extension agents in Ghanastated in our field survey that organic certification is usually only asso-ciated with abstaining from the use of chemical substances, but not theactive use of alternative inputs.3

The smallholders who use no chemical inputs, or very low levelsof external inputs normally face relatively lower economic entrybarriers into organic certification programs, since they require small ad-justments to meet certification requirements (UNCTAD, 2008b;Weber,2011). This does not imply that entry barriers for export-orientedcertification programs are not high, but that it is easier for farmerswho do not use chemical inputs to gain access to such programs.

While access to higher-priced organic markets may provide incen-tives to farmers to adoptmore agro-ecological practices, there are severalother factors that serve as constraints to the adoption of organic farmingand certification (Läpple and Kelly, 2013; Läpple and Rensburg, 2011;Mzoughi, 2011; Wollni et al., 2010). In particular, Wollni et al. (2010)point out that demand-side factors such as farmers' access to marketsand transport costs, which are normally influenced by infrastructural

3 There are various definitions of smallholder. In developing countries, smallholderusually refers to farms supporting one family with subsistence farming and limitedproduction of cash crops. We rely on the Ministry of Agriculture extension offices lists,and thereby definition, of smallholders. In addition, in our sample all farmers are cer-tified under a group certification option, which is directed at smallholders only.

development and remoteness from population centers tend to influencetheir decisionmaking in variousways. Distance of the farm to the nearestmarket has therefore been included in some studies to account for therole of market access in the adoption of farm technologies (Amare et al.,2012; Wollni et al, 2010).

A number of authors have also argued that non-economic factorssuch moral and social concerns can be significant in farmers' adoptionof organic farming (e.g., Carlsson et al., 2007; Mzoughi, 2011). Inparticular, Mzoughi (2011) suggests that even innovations thatare both profitable and ecologically-friendly may suffer from a low dif-fusion rate because farmers ignore their capacity to confer moral andsocial benefits. We argue in this paper that farm and household levelcharacteristics, as well as demand-side factors affect farmers' participa-tion decisions in organic farming, and show that organic farming influ-ences the use of agro-ecological practices.

The primary goal of the paper is to examine the effect of organiccertification on the extent to which agro-ecological practices areused, as well as the impact of the intensity of use on the return on invest-ment (ROI). We employ data from a recent farm-level survey of 386small-scale pineapple farmers in the Greater-Accra, Eastern, and CentralRegions of Ghana. These farmers are either organic or non-organic certi-fied and produce mainly for the export market.4 The study accounts forselection bias due to unobservable factors by using the framework ofendogenous switching regression approach (Lee, 1978). The approachallows us to analyze the determinants and effects of the adoption decisionof organic farming on the use of agro-ecological practices, separately foradopters and non-adopters among the sample of 386 pineapple farmers.In investigating the impact of agro-ecological practice on ROI, we use thegeneralized propensity score approach developed by Hirano and Imbens(2004) to control for selection bias.

The agro-ecological practices we consider in this study include ap-plication of organic fertilizers, organic pest and weed control, crop ro-tation, as well as soil and water conservation measures. As noted byRigby et al. (2001), agro-ecological practices are mostly employedby farmers to ensure farm sustainability through increased yieldsand reduced losses, minimizing inputs from non-renewable sources,maximizing use of natural biological processes, as well as promotingenvironmental quality. Knowler and Bradshaw (2007) argue thatthe net financial impact of conservation agriculture, which involvesagro-ecological practices, at the individual farm scale appears to bepositive.

In Ghana as in many other developing countries, crops that are pro-duced for export are usually intensively treated with pesticides to assurethe required quality and uniformity. This is also the case for pineapple,the third most important agricultural export product of the country,after cocoa and palm oil. On the environmental side, climate change isexpected to have negative effects on agricultural production, while popu-lation pressure will contribute to increased soil degradation and conse-quently lower crop yields (Diao and Sarpong, 2007). The Ghanaiangovernment has attempted to address these problems through environ-mental protection (Government of Ghana, 2010) and has established anorganic agriculture desk in the Ministry of Food and Agriculture (MOFA).

The remainder of the study is structured as follows: The next sectiongives an overview of the pineapple sector in Ghana and the data used inthe analysis. It is followed by the presentation of the correspondingdescriptive statistics. Subsequently, Section 3 presents the conceptualframework and empirical strategy employed in the analysis. The em-pirical results are presented in Section 4. The final section providesconcluding remarks and implications.

4 Non-organic certified refers to Global GAP certification, which is a precondition forproducing pineapple for export in Ghana. The Global GAP certification process sharessome practicalities with organic certification, among them regular inspections andupfront training. We may therefore expect that the adoption of agro ecological prac-tices among Global GAP farmers is likely to be higher than among non-certified pine-apple farmers producing for the local market.

6 Ajumako Enyan Esiam, Akuapem South, Ewutu-Efutu-Senya, Ga, Kwahu South andMfantseman.

7 The variable RISK is one factor from a factor analysis of several subjective state-ments on risk, chemical use and input availability. This factor loaded high on the fol-lowing statements: “I always want to try new farming techniques.”, “I need to take

332 L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

2. Background and Data

2.1. Pineapple Production in Ghana

The agricultural sector in Ghana accounts for about 30% of gross do-mestic product (GDP) and employs over 50% of the Ghanaian workingpopulation (WDI, 2011). In recent years, non-traditional exports ofhorticultural products experienced significant growth. Exports of freshfruits and vegetables, especially to Europe, are now themost importantgrowth sector of Ghana's agriculture. The country began exportingpineapples in small quantities in the 1980s. Ghanaian pineapple farmersproduce the varieties MD2, Smooth Cayenne, Sugar Loaf and QueenVictoria, where Sugar Loaf is mainly produced for the local market, orfor processing, while Queen Victoria plays a minor role as a high-priced specialty product. Pineapple exports increased rapidly until 2004,after which it declined towards 2010, partly because of a change invarieties demanded on world markets, to which the Ghanaian pineapplesector reacted slowly.Many farmers gradually switched to the newworldmarket variety (MD2), while some produced organic pineapples for ex-port. According to estimates from theGhanaian export promotion council(GEPC) and the Sea Freight Pineapple Exporters Association of Ghana(SPEG) about 31,000 tons of the estimated 75,000 tons produced in thecountry were exported in 2009 (GEPC, 2010).

2.1.1. Actors in the Ghanaian Pineapple SectorPineapple farming in Ghana is largely located within a radius of

100 km north-west of the capital Accra, in the regions of GreaterAccra, and the Central and Eastern Region. The pineapple industry isdriven by two dominant groups of producers that include few large ormedium-sized producers, and a large number of small-scale farmers.The small-scale farmers mostly sell their fruits on the local market orto exporters. Among these smallholders, there are two clusters, tradi-tional low-input “organic-by-default” producers and another group offarmers that strive to imitate large-scale high-input production Avail-able evidence indicates that about thirty percent of Ghana's total freshexport pineapples is supplied by smallholders (Suzuki et al., 2011).Pineapple export in Ghana is predominantly organized by exportcompanies that are also engaged in pineapple production.

2.1.2. Organic ProductionAs indicated previously, certified organic products normally attract a

price premium. Organic certified pineapples from Africa receive a pricepremium on the European market, which is normally partly passed onto the producers (Kleemann, 2011). About 0.19% of the agriculturalland in Ghana is organic certified, with a presumably higher part inpineapples.5 Organic certification refers to the standards in the EUregulations (EC) 834/2007 and (EC) 889/2008. It entails, among others,refraining from the use of synthetic inputs and the use of stipulatedinputs for flower induction. Organic farming therefore requires higheramounts of labor for manual weeding, pest control and possibly ownproduction of fertilizers (Bolwig et al., 2009; ITC, 2011; Valkila, 2009).The resulting use of on-farm/local inputs in organic farming may bean advantage when input markets are missing, or do not functionwell. However, a major disadvantage of organic farming is potentiallylower yields, in particular, when synthetic inputs are not replaced byorganic inputs and when knowledge about soil nutrient and plant pestand disease management is not sufficient (Beuchelt and Zeller, 2011;Valkila, 2009). Moreover, since organic production involves a long-terminvestment in soil fertility and sustainability, time lags between invest-ment and returns may prove to be an entry barrier for small resource-constrained farmers in insecure environments (Beuchelt and Zeller,2011). The Ministry of Food and Agriculture (MOFA), Agro Eco/LouisBolk, West African Fair Fruit (WAFF) and the German International

5 Source: http://www.organic-world.net/statistics-data-tables-dynamic.html.

Cooperation are involved in the promotion of organic farming inGhana.

2.2. Data

The data used in the study come from a farm household survey thatwas conducted from January to March 2010 in six different districts6

of the Central, Eastern and Greater Accra regions in southern Ghana,where pineapple cultivation is mostly located. A stratified randomsampling approachwas used. First, districts with significant amountsof smallholder pineapple production for export were selected, usinginformation from development agencies and the Pineapple ExportersAssociation of Ghana. Next, lists of all pineapple farmer groups in theselected districts that are certified and producing for the export marketwere obtained. Finally, farmers were proportionately selected fromeach group randomly. The respondents answered a detailed ques-tionnaire on the household's management of the pineapple farm,inputs into pineapple production, harvesting andmarketing, the certifica-tion process, and relationswith exporters. Besides, information on house-hold characteristics, social capital and land dispositionwere requested, aswell as data concerning non-income wealth indicators and perceptionsof different statements about environmental values, organic farmingtechniques and the use of fertilizers and pesticides.

The dataset includes 386 households from 75 villages and 9 (organic)and 14 (conventional) different farmer associations. In total, 185 organicfarmers and 201 conventional farmers were interviewed. All organicfarmers sold part of their produce as organic certified to exporters orprocessors and part of it on the local market, without reference to thecertification. Respectively, all conventional farmers sold their producepreferably to exporters or exporting processors, but also on the localmarket. In theory, there could be one-directional overlaps. This meansthat organic certified farmers could sell as organic certified (which hasthe highest price per kg) as first preference, as conventional exportproduce as second preference and on the local market as last option.However, this is not the case in our sample. The opposite, i.e. conven-tional farmers trying to sell on the organic export market is notpossible.

2.2.1. Descriptive Statistics of Sociodemographic VariablesThe typical household in our sample has a similar income compared

to the average in Ghana (country average 88.83 GHS per month, surveyaverage: highest density in income groups 51–150 GHS per month),and a higher income share from agriculture (47.8% versus 67%; datafrom Ghana Living Standards Survey 5). All sociodemographic variablesthat are included in the estimations are presented in Table 1.

Organic farm household heads are older and less educated thanconventional farm households. They have smaller farms, but are morespecialized in pineapple farming. On average, about 39% of the totalfarm land owned by organic producers is used for pineapple cultivation,while the corresponding figure for conventional farmers is 16%. Organicfarmers also appear to have fewer assets, relative to conventionalfarmers. However, the average organic household head received creditmore often during the last five years, and stated a higher willingnessto take risks in order to achieve success, as well as a greater opennessto innovation.7 Surprisingly, farmers producing non-organic pineapplesappear to have a greater concern for preserving the environment (ENV).This is probably due to the fact that the term environment is normallynot mentioned in organic certification training material, whereas it isspecifically mentioned in non-organic certification training material.

risks to achieve success”, and “Using new agricultural techniques significantly in-creases agricultural income.”

Table 1Descriptive statistics of variables included in the estimations.

Definition Variable Organic farmers(N = 185)

Convent. farmers(N = 201)

t-Stat.

Gender of household head (HHH)0 = female, 1 = male

GENDER 0.891 0. 982 −3.51***

Age of HHH AGE 46.313 42.970 2.82***Household size (persons living in household) HHSIZE 5.23 5.917 −2.35**Fraction of adults in household

(older than 15) (%)ADULT 0.684 0.665 0.75

Being native in community0 = no, 1 = yes

NATIVE 0.738 0.738 −0.01

Maximal educational level in household1 = none, 2 = primary school,3 = junior secondary, 4 = senior secondary,5 = technical/vocational training, 6 = tertiary/university

EDUC 9.470 10.195 −3.19***

Farm size (acre) FSIZE 10.35 18.720 −5.02***Share of land owned OWNLAND 0.549 0.204 7.628***Pineapple land (acre) PINLAND 4.014 3.066 2.07**Access to credit during the last 5 years

0 = no, 1 = yesCREDIT 0.317 0.232 1.78*

Bank account with more than 200 GHS0 = no, 1 = yes

BANK 0.339 0.512 −3.21***

Number of durable goods owned DURABLE 4.765 8.481 −10.875***Relation to the local government

1 = none,2 = HHH knows someone in the local government,3 = HHH has friends in the local government,4 = strong relation/politically active

GOVERN 2.257 1.774 4.27***

Self-stated openness to innovation and risk (factor analysis: thestronger the agreement, the higher the factor)

RISK 0.152 −0.166 3.01***

Years of experience in pineapple farming EXPER 11.557 11.595 −0.05How pineapple farming was learned

from family members and friends0 = no, 1 = yes

LEARN 1 0.863 0.501 7.97***

as a laborer on a farm or from0 = no, 1 = yes

LEARN 2 0.071 0.286 −5.51***

Importance of preserving the environment1 = very important, …, 4 = not important

ENV 1.775 1.281 6.91***

Number of years being certified CERTIFYEARS 3.165 2.032 3.875***Distance to the closest local market (hours) DIST 0.698 0.804 −1.59Soil characteristics

1 = red or black sandy, 2 = white sandy, 3 = white rocky, 4 = rockyred or black, 5 = sandy or rocky clay, 6 = clay, 7 = other

SOIL 2.781 2.304 2.13**

Variety Smooth Cayenne (0 = no, 1 = yes) SC 0.098 0.351 −5.99***Variety MD2 (0 = no, 1 = yes) VARIETYMD2 0.051 0.216 −7.12***Share of production cost for (of total labor costs) hired workers HIRED 0.484 0.607 −3.13***Assistance or training for farming received during last 5 years (0 = no, 1 = yes) ASSIST 0.732 0.708 0.50Number of farm inspection during the last 5 years INSPECT 1.913 2.619 −0.94Written contract with exporter (0 = no, 1 = yes) CONTR 0.410 0.417 −0.13Organizer of the certification process

1 = farmer organization,0 = else than farmer organization

CERTPROC 0.508 0.143 7.84***

Significance levels: *: 10% **: 5% ***: 1%.We use a conversion factor of 1 GHS = 0.46 Euros (calculated on the basis of the exchange rate on January 12, 2012).INSPECT includes both inspections by the certifier and by the buyer. Under group certification schemes not every farm is automatically visited by the certifier every year.

333L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

Even with more labor needed for production, organic farmers moreoften recruit their workers from the family than hiring farm workers,which is reflected in the lower proportion of the production costthey spend on hired labor. Concerning location specific variables, organicfarmers own a larger share of their land and grow pineapple on differentsoil types compared to conventional farmers. There is also a differenceconcerning the variety of pineapples planted: Organic farmers tend toprefer Sugar Loaf, whereas conventional farmers appear to favor SmoothCayenne or MD2.

Of relevance to the adoption mechanism is the fact that organicfarmers seem to have a stronger link to the local government andvisit the capital more frequently. They are also more likely to havelearned pineapple farming from friends or family members comparedto in training courses or as laborers on large farms. The majority offarmers of both groups have been certified within the last two years

and about 40% have a written contract with an exporter, all othershave oral contracts. The number of years that the farmers have beencertified is slightly longer for organic farmers.

2.2.2. Descriptive Statistics of Economic VariablesDifferences in economic characteristics of the farmers are presented

in Table 2. Columns (2) and (3) of Table 2 show the average costs foreach category per kilogram of pineapples. Kilogram is taken as a basefactor instead of pieces to control for the fact that organic fruits are onaverage smaller than conventional fruits, they are on average 0.18 kglighter than conventional fruit. As expected, there are large differencesin labor, equipment and input costs per kg between organic and con-ventional pineapple production, and costs for land are similar for bothgroups. While organic farmers spend much more on labor—hiredworkers as well as household labor—conventional pineapple producers

Table 2Descriptive statistics of economic variables.

Variable Organicfarmers

Conventionalfarmers

t-Stat.

Agricultural equipment (GHS per kg) 0.002 0.009 −2.77 ***Agricultural inputs (GHS per kg) 0.011 0.077 −5.97 ***Renewal of certification (GHS per kg) 0.000 0.006 −4.27 ***Land used for pineapple (GHS per kg) 0.004 0.004 −0.004Hired workers (GHS per kg) 0.037 0.019 3.77 ***Household labor (GHS per kg) 0.034 0.009 5.68 ***Yield (pineapple per acre) 15,780 18,259 −4.11 ***Quantity sold (in kg) 23,486 36,235 −2.81 ***Average local price (GHS per kg) 0.210 0.131 8.50 ***Average export price (GHS per kg) 0.251 0.196 5.40 ***Share sold on local market 0.495 0.354 3.00 ***Revenue (GHS per kg) 0.219 0.170 5.80 ***Production costs (GHS per kg) 0.105 0.118 −0.94Profits (GHS per kg) 0.114 0.052 4.01 ***ROI (ratio) 2.760 1.800 3.11 ***

We use a conversion factor of 1 Ghana Cedi (GHS) = 0.46 Euros. The t-statistic belongs tothemean difference test between column (2) and (3). Significance levels: *:10% **:5% ***:1%.

Table 3Descriptive statistics of agro-ecological practices.

Variable Organic farmers(N = 176)

Conventionalfarmers(N = 168)

t-Statistics Weightsused%

Min Max Mean Min Max Mean

Processed organicfertilizer

0 1 2.164 0 1 0.030 8.288*** 16

Organic pesticides 0 1.01 0.083 0 0.38 0.082 −0.032 14Mulch 0 5 1.590 0 3.85 1.328 5.294*** 17Manure 0 4.78 1.998 0 1.87 0.912 3.543*** 16Weeding 0 5 2.410 0 5 2.327 0.566 10Cover crops 0 1 0.175 0 1 0.161 0.353 7Crop rotation 0 2.23 0.980 0 0.84 0.132 6.343*** 13Trash lines 0 4.99 2.932 0 0.90 1.043 9.451*** 1Infiltration ditches 0 2.90 1.066 0 0.51 0.721 1.979** 2Leguminousresidues

0 1 0.066 0 1 0.018 2.217** 4

Variables run from0 (technique not used) to 5 (highest frequency or intensity this techniquewas used, taking into account the type of material used), except variables for organicpesticide use, cover crops, and leguminous residues use are zero when not used, and onewhen used.Significance levels for the t-statistics of the mean difference test: *: 10% **: 5% ***: 1%.

334 L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

usemore inputs and equipment.8We also observed that (not presentedin the Table), organic farmers do not use any chemicals, and utilize verylittle organic fertilizers, spend a lot of time with manual removal ofweeds andmore often produce their own planting andmulchingmate-rial, or exchange it with other farmers.

The production and certification costs are higher for conventionalfarmers, compared to organic farmers. On average, conventionalfarmers and organic farmers incurred about 444 GHS and 70 GHS, re-spectively, for initial certification, and then 93 GHS and 1 GHS, re-spectively, for yearly renewal. These costs are much less than thetotal costs, since they represent only the part paid by the farmersthemselves. A large share of the total costs incurred is often paid forby the exporters or NGOs. On the revenue side, export prices weregenerally higher than local prices for both groups, with organic pine-apples recording a price premium on both local and export markets.9

The average ROI presented in the last row of Table 2 indicatesthat the average ROI of certified organic farming is higher thanthat of conventional farming. Also shown in the Table are the slightlylower production costs and higher prices obtained for organic pine-apples. Although the production cycle of organic pineapple, which isabout 18.72 months, is on average longer than that of the conventionalpineapple, about 15.46 months, these differences do not affect the valuesof the return on investment (ROI). It is also evident from Table 2 that con-ventional farmers sold 1.5 times as many pineapples as organic farmers.This is mainly because of the larger areas under conventional farming,as well as the higher yields obtained from this farming method.

2.2.3. Descriptive Statistics of Intensity of Agro-ecological Practice UseTo examine the impact of organic agricultural practices on the

ROI we employ the framework proposed by Rigby et al. (2001) toconstruct a variable that consists of the different agro-ecologicalcropping practices most relevant for pineapple production. Theinformation on relevant practices was given by an agronomist andincluded in the questionnaire. The practices considered include organicfertilizer, non-chemical weeding, mulching, manure, trash lines, infiltra-tion ditches and crop rotation. They are assessed using a scoring systemthat ranges from 0 (technique not used) to 5 (highest frequency andintensity this technique was used, taking into account on what fractionof plots it was used and the type of material used). A score of 5 therefore

8 We are aware that measurement errors are frequent in measuring agricultural in-puts and outputs in developing countries. However, when farmers in both groups aresufficiently similar in their sociodemographic characteristics we can assume that mea-surement errors do not significantly differ between farmers.

9 One hint is that the Sugar Loaf variety yielded the highest prices on the local (andexport) market and was produced more frequently by organic than by conventionalfarmers.

indicates that this techniquewas usedwith the highest frequency presentin the sample on all pineapple plots of the respective farmer.Weeding is acommon practice for organic and non-organic farmers, one that is veryimportant in pineapple production, because pineapple grow relativelysmall. The weeding technique is relevant for soil water managementand erosion control.

Variables for organic pesticide use, cover crops, and leguminousresidues use are zero when not used, and one when used and werethen multiplied by five before weighting.10 All the variables wereweighted according to the average importance of each practice forsustainability given by 13 Ghanaian agronomists. The weights usedare presented in the last column of Table 3. The variable used in theanalysis (AGRECPRAC) was then constructed by adding up the weightedvalues of the different practices. Table 3 shows the descriptive statisticsfor each method. Robustness checks were made by a) repeating theanalysis without any weights and b) using an alternative weightingscheme which consisted in giving similar practice groups (fertilizersand fertilizing material, soil cover, and weeding and pesticides) thesame overall weights, and c) by excluding weeding, since we cannot dis-tinguish between different types of weeding (weeding by hand or weedprevention using e.g. beneficial organisms) which may have differentsustainability implications. For a), b) and c) all regressions werereplicated.

Fig. 1 presents kernel density estimates of the intensity of agro-ecological practices by the two categories of farmers. The estimates revealthat although conventional farmers also use sustainable farmingmethods, their intensity of use is generally lower than that of theircounterparts practicing organic farming. Moreover, it is clear from theresults that there are hardly any organic farmers that do not employthese farming practices, whereas some conventional farmers neveremployed agro-ecological practices.

3. Conceptual Framework

The conceptual framework employed here is based on the assump-tion that farmers choose between adopting organic farming and prac-ticing conventional farming. For analytical purposes, we assume herethat farmers are risk neutral, and take into account the potential benefitderived from adopting organic farming or non-adoption in the decision

10 In these cases either it was logical to pose the question with a yes/no option only,as in the case of cover crops or the quality of the data retrieved from the survey did notallow the division into frequency of use, as in the case of organic pesticides and legu-minous residues.

11 The unbiased treatment effect is hard to measure because, when treatment is non-random as in our case, untreated individuals may differ systematically because of self-selection into treatment. A popular approach to avoid biased results is to randomizetreatment. In our case randomization over which farmers use which agro-ecologicalmethods is impossible to realize because all the methods in question are already com-mon or widely known by the farmers. The underlying treatment is not a developmentintervention, but the outcome of various interventions in a longer time horizon.

Source: own estimation.

Fig. 1. Intensity of agro-ecological practice use and certification.Source: own estimation.

335L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

making process. Farmers are therefore assumed to choose the technolo-gy that provides maximum benefits. Under these assumptions, let usrepresent the net benefits farmer i derives from adopting the technologyas DiA and the net benefits from non-adoption represented as DiN. Thesetwo regimes can be can be specified as

DiA ¼ ZiβA þ uiA ð1Þ

DiN ¼ ZiβN þ uiN ð2Þ

Where Zi is a vector of variable factor prices, fixed factors, as well asfarm and household characteristics; βA and βN are vectors of parameters;uiA and uiN are iids. The farmer will normally choose the organic technol-ogy if the net benefits obtained by doing so are higher than that obtainedby not choosing the technology, that is DiA N DiN.

The individual preferences of the farmers are normally unknown tothe analysts, but the characteristics of the farmer and the attributes ofthe technology under consideration are observed during the surveyperiod. Given the available information, net benefits can be representedby a latent variable Di

⁎, which is not observed, but can be expressed as afunction of the observed characteristics and attributes, denoted as Z, in alatent variable model as follows:

D�i ¼ βZi þ μ i; Di ¼ 1 D�

i N0� � ð3Þ

whereDi is a binary indicator variable that equals 1 for household i, in caseof adoption of the technology and 0 otherwise, β is a vector of para-meters to be estimated, Zi is a vector of household and plot-level charac-teristics as defined earlier, and μi is an error term assumed to be normallydistributed. The probability of adoption can then be expressed as

Pr Di ¼ 1ð Þ ¼ Pr D�i N0

� � ¼ Pr μ iN−βZið Þ ¼ 1−F −βZið Þ ð4Þ

where F is the cumulative distribution function for μi.

3.1. Impact of organic farming on agro-ecological practices

As indicated earlier, the intensity of use of agro-ecological prac-tices vary between organic and conventional farmers. To capturethe effects that arise from the difference between organic andnon-organic farms on the use of agro-ecological farm methods,we employ a specification from the impact assessment literatureon outcomes to participation choice. Specifically, we hypothesizethat adoption or non-adoption of organic technology positively

influences the use of agro-ecological farm practices. This may beexpressed as

Yi ¼ Xiβ þ δDi þ εi ð5Þ

where Yi represents the intensity of agro-ecological practices and Di isthe adoption dummy; Xi is a vector of farm-level and household-levelcharacteristics, such as age and education of farmer, access to credit,social network variables, farm size, and soil quality variables. Thecoefficient δ in the specification captures the impact of adoptionon the use of agro-ecological practices. The issue of self-selectionis crucial here because the decision of households to adopt or notto adopt organic farming may be associated with the net benefitsof adoption. Selection bias arises if unobservable factors influenceboth the error term of the technology choice, μi, in Eq. (3) and theerror term of the outcome specification (εi), in Eq. (5), resultingin correlation of both error terms. When the correlation betweenthe two error terms is greater than zero, OLS regression techniquestend to yield biased estimates. To address these issues, we employ anendogenous switching regression model (ESR) to jointly examine thedeterminants of adoption and the impact of adoption on the intensityof agro-ecological practice use.11

The parametric approach of the endogenous switching regression(ESR) model goes back to Lee (1978) and Maddala (1983), and accountsfor self-selection and systematic differences across groups. Outcomeequations are specified differently for each regime, conditional on theadoption decision, which is estimated by a probit model. Thus, if we de-fine YiA and YiN as the intensity of agro-ecological practices for organicand non-organic farmers, respectively, we can specify the outcome equa-tions as:

YiA ¼ X′iβA þ ξiA if Di ¼ 1 ð6Þ

YiN ¼ X′iβN þ ξiN if Di ¼ 0 ð7Þ

Although self-selection based on observables is taken into accountin the above specification, unobservable factors could still create a

336 L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

correlation between μi and ξiA, ξiN. The endogenous switching regressionmodel treats the sample selectivity problem as a missing variable prob-lem, which can be estimated and plugged into the Eqs. (6) and (7).Thus, after estimating a probit model in the first stage, theMills ratiosλN and λA and the covariances σμA = Cov(μξA) and σμN = Cov(μξN)can be computed and employed in the following second stage spec-ification:

YiA ¼ X′iβA þ σAλiA þ uiA if Di ¼ 0 ð8Þ

YiN ¼ X′iβN þ σNλiN þ uiN if Di ¼ 0 ð9Þ

In these equations, the error terms uiA and uiN have conditional zeromeans. Following Lokshin and Sajaia (2004) we use the full informationmaximum likelihood method (FIML) to estimate this model, i.e.the selection equation and the outcome equations are estimatedsimultaneously.

When the correlation coefficients of μ and ξA(ρiA = σμA/σμσA) and ofμ and ξN(ρiN = σμN/σμσN) are significant, the model has an endogenousswitch, i.e. selection on unobservable factors is substantial. The coeffi-cients obtained from the endogenous switching regression model canbe employed to derive the average treatment effect (ATT). This involvesfirst specifying the expected values of the outcome (Lokshin and Sajaia,2004). For an organic farmer with characteristics X and Z, the expectedvalue of the outcome, YiA, is given as:

E YiA D ¼ 1j Þ ¼ XβiA−σAελA:ð ð10Þ

The last term in Eq. (10) accounts for sample selection, showing thatorganic farmers may behave differently from an average farmer withidentical characteristics due to unobserved factors (Maddala, 1983). Theexpected value of the same farmer had he chosen to produce conven-tionally is given as:

E YiN D ¼ 1j Þ ¼ XβiN−σNελA:ð ð11Þ

The change in the outcome variable due to organic is then specifiedas the difference between organic and non-organic production. Thus,the expected outcomes from Eqs. (10) and (11) are used to obtain unbi-ased estimates of the effects of organic production.

τESRATT ¼ E YiA D ¼ 1j Þ−E YiN D ¼ 1j Þ ¼ X′ βiA−βiNð Þ þ σμA−σμN

� �λA

��

ð12Þ

where σ represents the covariance of the error terms and λ the inversemills ratios. These estimates are termed average treatment effect on thetreated (ATT) in the impact assessment literature (Lokshin and Sajaia,2004).12

3.2. Impact of Intensity of Agro-ecological Practice Use on ROI

Given that the intensity of agro-ecological practices is a continuousvariable, we employ the generalized propensity score (GPS) approachdeveloped by Hirano and Imbens (2004) in order to ascertain whetherusing these practices tend to affect the economic viability of the farm.Thus, the analysis in this section considers the treatment variable as a

12 Since λA is included in both Eqs. (10) and (11), unobserved vectors are taken intoaccount. The approach simply assumes that unobserved factors have differential effectson organic and non-organic farmers. Thus, taking the differences in effects, σμA − σμN,while holding λA constant ensures that the effects of unobserved factors are cancelledout. In this case, the estimated agro-ecological use differences would be solely due toorganic impacts, without any unobserved effects.

continuous variable, and not a dichotomous decision variable as wasassumed in the previous analyses.

In line with the GPS approach, Eq. (5) can be re-specified as Yi =f(XiTi), where Yi refers to the return on investment and Ti is the actuallevel of agro-ecological practice of the farm. Of significance is the aver-age dose response function (DRF), which relates to each possible treat-ment level ti, the unbiased potential outcome Yi(t) of the farmer i:

θ tð Þ ¼ E Yi tð Þ½ �∀t inT ð13Þ

where θ represents the DRF. In line with Hirano and Imbens (2004), wepresume that the assignment to the treatment is weakly unconfoundedgiven the controls, i.e.

Yi tð Þ⊥Ti Yi tð Þ∀t inTj ð14Þ

Thus, the treatment assignment process is supposed to be condi-tionally independent of each potential outcome, given the controlvariables. Hence, there is no systematic selection into specific levels ofagro-ecological practice intensity caused by unobservable characteristics(Flores et al., 2009). Weak unconfoundedness implies that this indepen-dence only has to hold for each level of treatment t but not jointly forall potential outcomes. The generalized propensity score (GPS) suggestedby Hirano and Imbens (2004) is defined as the conditional probabilityof a particular treatment given the observed covariates. Whenr(Ti,Xi) = fT ⊥ X(t ⊥ x) is the conditional density of potential treatmentlevels given specific covariates, then the GPS of a household i is given asRi = r(Ti, Xi). The GPS is a balancing score, i.e. within strata with thesame value of r(t, X) the probability that T = t does not depend on thecovariates Xi. Given this balancing property andweak unconfoundedness,Hirano and Imbens (2004) show that using the GPS to remove theselection bias allows the estimation of the average DRF of Eq. (13).

In the first step the conditional expectation of the outcome as afunction of treatment T and GPS R is estimated as

β t; rð Þ ¼ E Y Ti ¼ t;Ri ¼ rj Þð ð15Þ

Then, the DRF at each level of treatment can be estimated by averag-ing the conditional expectation over the GPS at that treatment level:

θ tð Þ ¼ E β t; r t;Xið Þð Þ½ � ð16Þ

In our application, the GPS is estimated using a normal distributionof the logarithmic treatment given covariates Xi. The validity of the as-sumed normal distribution is assessed using theKolomogorov–Smirnovtest for normality.We followedHirano and Imbens (2004) and took thelogarithm of the treatment variable, because the distribution of theagro-ecological practices was skewed. This procedure yielded low skew-ness (0.090) and kurtosis (1.698) values and a positive Kolmogorov–Smirnov test for normality at the 5% level of significance. The balancingproperty of the estimated GPS is tested by employing the method pro-posed by Hirano and Imbens (2004). The common support condition,i.e. that households in one group have to match with comparable house-holds in other treatment groups, is imposed by employing the methodsuggested by Flores et al. (2009). After estimating the GPS, the DRF isestimated using a flexible polynomial function as in Bia and Mattei(2008). The average potential outcome at each treatment level is estimat-ed using a quadratic approximation of the treatment variable and a linearone for the GPS. The specification is estimated using OLS regressionfor the ROI. Confidence bounds at 95% level are estimated using thebootstrapping procedure.

Table 4Estimation results of ESR for impact of organic certification on agro-ecological practice use.

Variable Selection Eq. Organic farmers Convent. farmers

Coefficient Std.Err.

Coefficient Std.Err.

Coefficient Std.Err.

(1) (2) (3) (4) (5) (6)

GENDER −0.122 0.513 0.173 0.101 1.769 0.597AGE 0.004 0.018 −0.006 0.023 −0.056** 0.024NATIVE −0.151 0.275 0.443 0.513 0.401 0.399RISK 0.306*** 0.095 −0.133 0.207 0.135 0.196HHSIZE −0.058 0.052 −0.146 0.099 −0.068 0.071EDUC −0.094* 0.059 0.323** 0.096 0.135 0.286DURABLE −0.296*** 0.090 0.397*** 0.142 0.138 0.089FSIZE −0.012 0.010 −0.006 0.012 0.0001 0.011OWNLAND 0.586** 0.236 1.051** 0.558 0.927 0.827EXPER 0.034 0.032 0.020 0.036 0.024 0.033LEARN1 0.829** 0.597 1.053 1.146 −0.774 0.440LEARN2 −0.537** 0.217 0.357 1.387 −0.223 0.530DIST −0.341** 0.164 −0.130 0.405 −0.947** 0.373SOIL 0.008 0.058 −0.316*** 0.1281 −0.268*** 0.085CERTPROC 1.403*** 0.218 −1.232* 0.665 0.828 0.747ENV 1.431*** 0.325 −1.423*** 0.412 0.038 0.388GOVERN 0.445*** 0.164BANK −0.468 0.558 0.403 0.497CREDIT −0.526 0.466 −0.239 0.487VARIETYMD2 2.550*** 0.989 −0.128 0.396HIRED −0.569 0.821 0.452 0.578INSPECT −0.113 0.026 0.074** 0.029CONTR 1.154** 0.543 −0.083 0.380CERTIFYEARSNO 0.299** 0.141 −0.203 0.457INTERCEPT 1.038 0.985 1.027 1.347 3.455** 1.754ρ1D −0.320 0.466lnσ1 0.916*** 0.070ρ0D 0.260 1.048lnσ0 0.744*** 0.104Log-Likelihood: −993.143Wald test of indep. eqns.: χ2(2) = 3.69***

Significance levels for the t-statistics of the mean difference test: *: 10% **: 5% ***: 1%.

13 Higher values of the soil variable correspond to lower soil quality soil. Thus, rockyand clayey soils are of the lowest quality.14 We are thankful to an anonymous reviewer for drawing our attention to this issue.

337L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

4. Empirical Results

4.1. Empirical Results for Adoption

The full information maximum likelihood estimates of the determi-nants of adoption of organic farming, as well as the impact of adoptionon the intensity of use of agro-ecological practices are presented inTable 4. As mentioned earlier, identification of the model requires thatthere is at least one variable in the selection equation that does not appearin the outcome equation. The variable representing relation to the localgovernment is used as identifying instrument, and as such droppedfrom the outcome equations. Quite interesting is the insignificance ofthe correlation coefficients presented in the Table. This finding indicatesthe absence of any endogenous switch, suggesting that there is no sub-stantial selection on unobservable factors.

The selection equation, which can be interpreted as probit estimatesof determinants of adoption generally indicate that farm-level andhousehold characteristics do influence adoption decisions of farmers.The estimates of the impact of adoption on the intensity of use of agro-ecological practices show that the farm-level and household characteris-tics influence the behavior of adopters and non-adopters differently. Inparticular, education and durable goods owned appear to have positiveand significant effects on organic farmers using more agro-ecologicalpractices, while no significant effect is observed for conventional farmers.Organic farmers who plant the variety MD2 use much more organicpractices. This is expected, since this variety is reported to requiremore fertilizer and be very sensitive towards environmental conditions.Land ownership also appears to influence the intensity of use by organicfarmers, but not by conventional farmers. Similarly, the number of years

being certified positively and significantly influences the intensity ofagro-ecological practices by organic farmers, but exerts a negative,albeit insignificant effect on conventional farmers. The negative andsignificant impact of the soil quality variable indicates that farmerswho cultivated land with better soil quality used agro-ecologicalpractices more intensively.13 The positive impact of soil quality onthe use of agro-ecological practices is consistent with the resultsreported by Abdulai et al. (2011), who found a positive relationshipbetween soil quality and investment in soil conservation measuresin Ghana. The estimates for the average treatments effect (ATT), whichshows the impact of organic certification on the use of agro-ecologicalpractice was computed with Eq. (12). The results are presented in thefirst row in Table 5. Unlike the mean differences in the use of agro-ecological practices shown in Table 3, theATT estimate accounts for selec-tion bias arising from the fact that adopters and non-adopters may besystematically different. The estimated ATT is positive and highly signif-icant, suggesting that organic certification does indeed act as a catalystfor the increased use of agro-ecological practices. Specifically, organiccertification moves the farmer up 15–20% on the full range of possibleintensities, or by about 80%, taking the overall mean use as a referencepoint. Organic certified farmers use significantly more, or a higher in-tensity of agro-ecological practices. The robustness of the ESR is checkedby estimating the same model, but using the three other specificationsdescribed in Section 3. The estimates, which are also reported in theTable 5, also confirm the positive and highly significant impact of organ-ic certification on the intensity of agro-ecological practices.

It is significant to mention that when the farmers were asked directlyfor changes in production methods after certification, 67% of organic andonly 35% of conventional households claimed to have changed their useof agro-ecological practices. Only farmers that answered “no change”for all types of agro-ecological practices were included in the 35%. Thisimplies that about one-third of organic farmers claimed that theydid not change their use of agro-ecological practices after organiccertification. Interestingly, all those who did not change their use ofagro-ecological practices following organic certification were non-usersof AGREPRAC both before and after certification, which is consistentwith the “organic by default” narrative mentioned in the introduction. Itis also in line with our estimation results that certification contributesto the adoption of agro-ecological practices.14

Given the absence of any endogenous switch, we also employedpropensity score matching (PSM) approach to compute the ATT andcompare with those from the ESR. PSM is basically a technique thatmimics an experiment ex post. The results, which are presented in thelower part of Table 5, show that the ATT ranges between 4.07 and4.23, depending on the matching algorithm used. Overall, the resultsconfirm the positive and significant impact of organic certification onthe intensity of agro-ecological practices, indicating that farmers thatproduce organic pineapples also tend to usemore agro-ecological prac-tices. The matching quality test conducted with the Rosenbaum andRubin (1985) test shows that differences in themeans of the covariatesbetween the two groups vanish after matching. The sensitivity of theestimates to unobservable factors was also tested with the Rosenbaum(2002) bounds. Based on kernel matching, the critical value of Γ(Γ*) =1.35 indicates that the ATT would still be significant even if matchedpairs differ in their odds of certification by the factor 1.35.

4.2. Results for the Impact of Intensity of Agro-ecological Practice Use on ROI

In this section, we examine the impact of agro-ecological prac-tices on the return on investment (ROI). The treatment variable isAGRECPRAC, as indicated previously. The results of the maximumlikelihood estimation of the GPS, which are presented in Table A.1

Source: own estimation.

Fig. 2. Impact of intensity of agro-ecological practice use on ROI.Source: own estimation.

Table 5Results of Impact of Organic Certification on Agro-ecological Practice Use.

Method Predicted use of certified Predicted use of non-certified ATT t-Statistic

ESROrganic certified farmers 5.921 2.518 3.403 13.314***Conventional farmers 8.135 3.788

Alternative SpecificationsESR using different weights

Organic certified farmers 6.102 3.046 3.056 11.465***Conventional farmers 7.979 3.594

ESR using no weightsOrganic certified farmers 5.986 2.136 3.851 12.258***Conventional farmers 8.115 3.266

ESR (weeding excluded)Organic certified farmers 5.728 2.667 3.061 10.894***Conventional farmers 7.934 3.363

PSMKernel (bandwidth = 0.4) 6.751 2.680 4.071 7.98***Radius (caliper = 0.05) 6.751 2.523 4.228 7.34***Nearest-neighbor 6.751 2.351 4.400 6.98***

Significance levels for the t-statistics of the mean difference test: *: 10% **: 5% ***: 1%.For PSM, standard errors are calculated with bootstrapping using 1000 replications. Bootstrapping of standards errors is necessary because the estimated variance does not includethe variance that may appear due to the estimation of the propensity score and the imputation of the common support assumption (Caliendo and Kopeinig (2008)). Even thoughAbadie and Imbens (2008) criticize the use of bootstrapping for the nearest-neighbor algorithm, its application is still common practice.

338 L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

in the Appendix A, are not discussed here, since the estimates only serveto balance the observed distribution of covariates across the treated anduntreated groups (Hirano and Imbens, 2004). It is however interestingto note that as in the regressions in the previous section, the organiccertification dummy is again highly significant in the probit regression.The balancing tests indicate that there is no systematic difference inthe distribution of covariates between both groups after matching,suggesting that the GPS eliminates bias in the estimates of the dose–response function.15 Regarding the common support condition, 278

15 For testing the balancing property of the GPS, the treatment variable was dividedinto 4 intervals with cut-off points at 25%, 50%, etc. Without adjusting for the GPS, t-tests of mean difference between the intervals revealed that 14 t-tests were significantat the 5% level, after dividing into 4 intervals and conducting block-wise t-tests thisnumber was reduced to 2. We repeated the analysis with more intervals, namely 7,which did not affect our conclusions, but the number of observations in each intervalbecomes quite small, so the results are weaker.

farmers were on support, which represents 87% of the initial 311farmers for which we have sufficient data to calculate the ROI.

Fig. 2 shows the dose response function of the impact of the use ofagro-ecological practices on the return on investment in pineapple farm-ing.16 There is a non-linear hook shaped relationship, whereby the effecton the ROI is positive, but in different ways at different levels. The impactis high at very low levels of agro-ecological practice use, but declines athigher levels of use, before rising again, with increasing intensity. It is sig-nificant to note that in our analysis, a low level of the index implies verylittle use of agro-ecological practices and even ahigh level is still lowcom-pared to developed country agriculture. At the lowest point, the estimat-ed ROI is just below the mean of the sample (2.265). While the impact ofusing agro-ecological practices is overall positive, relatively low and rela-tively high levels appear to performbetter in termsof rate of return than amedium level of agro-ecological practice intensity. This implies that themotivation to increase the use may be low when farmers are unawareof the shape of this impact curve or have a high discount rate into thefuture.

A look at the kernel density estimates in Fig. 1 shows that mostfarmers are exactly in this impact dip.17 To gain further insights intothe differential behavior of the farmers, we also examined the com-position of agro-ecological practices used at different intensities.Specifically, we divided the sample into several equally sized groups,according to the AGRECPRAC variable, involving those below andabove the low impact dip. It was observed that at low levels of inten-sity, the average farmer restricts the use to very few practices, andfirst starts to use them more intensively, i.e. on more plots or moreoften/in greater amounts, before adding different practices. Noteworthy

16 The estimated quadratic dose response function regression is shown in Table A.2 inthe appendix. All GPSM regressions were also repeated with net farm income as impactvariable. Due to the low investment level of the farmers, the results did not change sig-nificantly. Therefore the results are omitted here, but are available upon request fromthe authors.17 As rightly pointed out by an anonymous reviewer, site-specific variation in agricul-tural conditions could potentially explain the shape of the curve. However, given thatthe climatic conditions are similar in the surveyed area, and we do control for soil char-acteristics in the estimations, this influence should be minimal.

Table A.1Estimation Results of Generalized Propensity Score.

Variable Coefficient Std. Err.

Equation 1ORGANIC 0.165*** 0.026GENDER 0.050 0.052AGE −0.005** 0.002RISK 0.019 0.048HHSIZE −0.041*** 0.018EDUC 0.101** 0.079FSIZE 0.003** 0.001OWNLAND 0.283*** 0.096EXPER 0.042* 0.021LEARN1 −0.101* 0.060LEARN2 0.108* 0.074DIST −0.152 0.135CERTPROC −0.126 0.114SOIL −0.112*** 0.022DURABLE 0.165*** 0.071ENV 0.266** 0.170INTERCEPT 0.935*** 0.149

Equation 2INTERCEPT 0.31*** 0.014

Significance levels for the t-statistics of the mean difference test:*: 10% **: 5% ***: 1%.

339L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

is the fact that it is not the potentially costly organic fertilizers andpesticides that are used significantly less in the low-use groups,but rather manure, animal mulch, and cover crops. Since thefarmers stated that they know what each practice is, the problemcannot be attributed to lack of information or knowledge, but rathereconomies of scale in transport cost. When the farm is strongly spe-cialized in pineapple, mulching material and manure cannot be pro-duced on the farm (see also e.g. Branca et al., 2011). The requiredmaterial is often available at no or low cost, but needs to betransported to the farm. Since this material is relatively bulky,transport costs can impede their use in case of lack of cash, or iftheir perceived benefits are lower than the effort of organizing andpayingfor their transport.

4.2.1. Robustness ChecksThe large confidence bands at the ends of the distribution in Fig. 2

suggest that the impacts are less clear among the non-users and thevery intensive users. We therefore conducted a robustness check inwhich we excluded values of AGRECPRAC of over 13. The resultobtained is shown in Fig. A.1 in the Appendix A. It is slightly differentat high values, with no flattening out, with the predicted impacthigher at the right end. However, the shape of the curve, which is ofprimary interest, remains the same. As a further robustness check,we use different specifications of the agro-ecological practice index.The results, which are presented in Figs. A.2 (different weights de-scribed in Section 3.2), A.3 (no weights), and A.4 (weeding excluded)in the Appendix A appear to be similar to the findings presented inFig. 2.

5. Conclusions

Some concerns have been raised that organic certification and sus-tainable farming practices are insufficiently linked on farms in developingcountries. Most farmers certified as organic producers have thereforebeen considered to be producing organic-by-default, with very little orno use of productivity-enhancing inputs and soil-improving measures,such as organic fertilizers, organic pest and weed control, crop rotation,and soil and water conservation, often resulting in low yields and unsus-tainable production.

In this paper, we employ an endogenous switching regressionapproach to examine the determinants of adoption of organic pro-duction methods, as well as the impact of organic certification onagro-ecological practice use. We also use a generalized propensityscore matching approach to investigate the effect of the intensity ofagro-ecological practices on the return on investment. A recentfarm-level data from organic and non-organic certified small-scalepineapple farms in the Greater Accra, Eastern, and Central regions inGhana are used in the analysis. Our empirical results show that organiccertification increases agro-ecological practice use, suggesting that or-ganic certification serves as a catalyst for the use of agro-ecologicalpractices. A number of factors could be driving this observation. First,farmers may want to replace the banned non-organic inputs with or-ganic inputs. Second, higher incomes from certified organic productscould increase on-farm investments in agro-ecological practices, partic-ularly if farmers face liquidity constraints. Both are consistentwith the lit-erature on investment in land improving agro-ecological practices inGhana, where ownership in land, which serves as collateral for loans,eases liquidity constraints (Abdulai et al., 2011). It is also consistentwith Duflo et al.'s (2011) finding that farmers underuse fertilizer andother inputs due to liquidity or capital constraints. Third, since mostbuyers put financial and time resources into organic certification, theytend to have a longer-term interest in working with those farmers,thereby investing in training to improve on farm soil fertility. Thosethree explanations may co-exist at the same time. The estimates ofthe economic impacts of agro-ecological practices generally reveala positive and nonlinear relationship between the rate of return and

the intensity of agro-ecological practice use, indicating that more inten-sive use of agro-ecological practices is economically beneficial for farmers.This finding suggests that from an environmental policy perspective thelink needs to be strengthened considerably, given the low intensity ofagro-ecological practice use. The low level of use is probably because ofthe nonlinear relationship, which suggests economic benefits at lowlevels and high levels. However, farmers need to surmount a low impactgap to attain high levels, including availability of organic material andhigh transport costs for organic material. Given that external inputsfrom cocoa production and juice factories are normally available for use,but at prohibitive transport costs for individual farmers, governmentagencies or certification agencies could organize intermediates to fillthis gap by purchasing these organic materials from juice factories andcocoa producers for sales to farmers. Certification may therefore helpease the problem through high prices on the produce and the supportby buyers. Moreover, certification systems could also require the activeuse of organic soil fertility management methods to increase their inten-sity of use. However, thiswould set barriers to certification for small-scalefarmers even higher than they already are and may thereby excludemany potential beneficiaries.

If successfully managed, organic certification for the dominantlysmall farmers in Africa may provide two types of economic benefits.It may reduce rural poverty by providing market access and higherprofits through a combination of higher prices and more resilientyields. It could also provide environmental benefits for the local econ-omy in the long term.

Acknowledgments

The authors would like to thank without implicating; the jour-nal editor, and two anonymous reviewers for valuable commentsand suggestions that have substantially improved the paper. Theywould also like to thank the Poverty Reduction, Equity and GrowthNetwork (PEGNet) and the German Research Foundation for finan-cial support.

Appendix A

Table A.2Estimation results of the coefficients of the dose–response function.

Variable Coefficient Std. Err.

T −0.305** 0.121T2 0.019*** 0.004GP S −3.252** 1.401T ∗ GP S 0.385 0.259INTERCEPT 4.638*** 1.251

Significance levels for the t-statistics of the mean difference test:*: 10% **: 5% ***: 1%.

340 L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

Fig. A.3. Impact of intensity of agro-ecological practice use on ROI (no weights).Source: own estimation.

Fig. A.4. Impact of intensity of agro-ecological practice use on ROI (weeding excluded).Source: own estimation.

Fig. A.1. Impact of intensity of agro-ecological practice use on ROI (restricted to valueslower than 13).Source: own estimation.

Fig. A.2. Impact of intensity of agro-ecological practice use on ROI (different weightsfor agro-ecological practices).Source: own estimation.

References

Abadie, A., Imbens, G.W., 2008. On the failure of the bootstrap for matching estimators.Econometrica 76 (6), 1537–1557.

Abdulai, A., Owusu, V., Goetz, R., 2011. Land tenure arrangements and investment inland improvementmeasures: theoretical and empirical analysis. Journal of DevelopmentEconomics 96 (1), 66–78.

Amare, M., Asfaw, S., Bekele, S., 2012. Welfare impacts ofmaize–pigeon pea intensificationin Tanzania. Agricultural Economics 43 (1), 27–43.

Barham, B.L., Weber, J.G., 2012. The economic sustainability of certified coffee: recentevidence from Mexico and Peru. World Development 40 (6), 1269–1279.

Beuchelt, T., Zeller, M., 2011. Profits and poverty: certification's troubled link forNicaragua's organic and fairtrade coffee producers. Ecological Economics 70 (7),1316–1324.

Bia, M., Mattei, A., 2008. A Stata package for the estimation of the dose response functionthrough adjustment for the generalized propensity score. The Stata Journal 8 (3),354–373.

Blackman, A., Naranjo, M.A., 2012. Does eco-certification have environmental benefits?Organic coffee in Costa Rica. Ecological Economics 83, 58–66.

Bolwig, S., Gibbon, P., Jones, S., 2009. The economics of smallholder organic contractfarming in tropical Africa. World Development 37 (6), 1094–1104.

Branca, G., McCarthy, N., Lipper, L., Jolejole, M.C., 2011. Climate smart agriculture: asynthesis of empirical evidence of food security and mitigation benefits fromimproved cropland management. FAO Working paper.

Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation ofpropensity score matching. Journal of Economic Surveys 22 (1), 31–72.

Calo, M., Wise, T.A., 2005. Revaluing Peasant Coffee Production: Organic and Fair TradeMarkets in Mexico. Global Development and Environment Institute Tufts University.

Carlsson, F., Khanh Nam, P., Linde-Rahr, M., Martinsson, P., 2007. Are Vietnamesefarmers concerned with their relative position in society? Journal of DevelopmentStudies 43 (7), 1177–1188.

341L. Kleemann, A. Abdulai / Ecological Economics 93 (2013) 330–341

Diao, X., Sarpong, D.B., 2007. Cost Implications of Agricultural Land Degradation inGhana. IFPRI Discussion Paper 00698.

Duflo, E., Kremer, M., Robinson, J., 2011. Nudging farmers to use fertilizer: theory andexperimental evidence from Kenya. American Economic Review 101, 2350–2390.

Erenstein, O., 2002. Crop residue mulching in tropical and semi-tropical countries: anevaluation of residue availability and other technological implications. Soil andTillage Research 6 (2), 115–133.

Flores, C.A., Flores-Lagunes, A., Gonzalez, A., Neumann, T.C., 2009. Estimating theeffects of length of exposure to training program: the case of job corps. WorkingPapers 2010-3. University of Miami, Department of Economics.

Food and Agricultural Organization of the United Nations (FAO), 2011. Socio-economicanalysis of conservation agriculture in southern Africa. Network Paper 2. FAO,Rome.

Ghana Export Promotion Council (GEPC), 2010. Ghana pineapple export declines.Available at: http://www.gepcghana.com/news.php?news=196&pnm=Ghana(Pineapple Export Declines, 11/29/2011).

Government of Ghana, 2010. Medium-term national development policy framework:Ghana shared growth and development agenda, 2010–2013. Policy Framework,volume 1.

Hattam, C.E., Lacombe, D.J., Halloway, G.J., 2012. Organic certification, export market ac-cess and the impacts of policy: bayesian estimation of avocado smallholder “times-to-organic certification” in Michoacan Mexico. Agricultural Economics 43, 441–457.

Hirano, K., Imbens, G.W., 2004. The propensity score with continuous treatments. In:Gelman, A., Meng, X.-L. (Eds.), Applied Bayesian Modeling and Causal Inferencefrom Incomplete-Data Perspectives. Wiley InterScience, West Sussex, England.

International Trade Center (ITC), 2011. The impacts of private standards on producersin developing countries. Literature Review Series on the Impacts of Private Standards,Part II, Geneva.

Kassam, A., Friedrich, T., Derpsch, R., Lahmar, R., Mrabet, R., Basch, G., Gonzalez-Sanchez, E., Serraj, R., 2012. Conservation agriculture in the dry Mediterraneanclimate. Field Crops Research 132, 7–17.

Kleemann, L., 2011. Organic pineapple farming inGhana—a good choice for smallholders?Kiel Working Paper 1671. Kiel Institute for the World Economy, Germany.

Knowler, D., Bradshaw, B., 2007. Farmers' adoption of conservation agriculture: areview and synthesis of recent research. Food Policy 32 (1), 25–48.

Läpple, D., Kelly, H., 2013. Understanding the uptake of organic farming: accounting forheterogeneities among Irish farmers. Ecological Economics 88, 11–19.

Läpple, D., Rensburg, T.V., 2011. Adoption of organic farming: are there differencesbetween early and late adoption? Ecological Economics 70 (7), 1406–1414.

Lee, L.-F., 1978. Unionism andwage rates: a simultaneous equationsmodelwith qualitativeand limited dependent variables. International Economic Review 19 (2), 415–433.

Lokshin, M., Sajaia, Z., 2004. Maximum likelihood estimation of endogenous switchingregression models. The Stata Journal 4 (3), 282–289.

Maddala, G.S., 1983. Limited-dependent and Qualitative Variables in Econometrics.Cambridge University Press, Cambridge.

Mzoughi, N., 2011. Farmers adoption of integrated crop protection and organic farming:do moral and social concerns matter? Ecological Economics 70 (8), 1536–1545.

Ndugire, N., 2010. Scaling up Organic Agriculture and Enhancing its Foreign MarketAccess: Lessons Learned from Eastern Africa. UN Economic Commission for Africa,ATPC work in progress No. 80.

Ninan, K.N., Sathyapalan, J., 2005. The economics of biodiversity conservation: a studyof a coffee growing region in the Western Ghats of India. Ecological Economics 55(1), 61–72.

Philpott, S.M., Bichier, P., Rice, R., Greenberg, R., 2007. Field-testing ecological and economicbenefits of coffee certification programs. Conservation Biology 21 (4), 975–985.

Pretty, J.N., Noble, A.D., Bossio, D., Dixon, J., Hine, R.E., Penning de Vries, F.W.T.,Morison, J.I.L., 2006. Resource-conserving agriculture increases yields in developingcountries. Environmental Science and Technology 40 (4), 1114–1119.

Probst, L., Houedjofonon, E., Ayerakwa, H.M., Haas, R., 2012. Will they buy it? Thepotential for marketing organic vegetables in the food vending sector to strengthenvegetable safety: a choice experiment study in three West African cities. Food Policy37, 296–308.

Rigby, D., Woodhouse, P., Young, T., Burton, M., 2001. Constructing a farm level indicatorof sustainable agricultural practice. Ecological Economics 39, 463–478.

Rosenbaum, P.R., 2002. Observational Studies. Springer, New York.Rosenbaum, P.R., Rubin, D.B., 1985. Constructing a control group using multivariate

matched sampling methods that incorporate the propensity score. The AmericanStatistician 39 (1), 33–38.

Suzuki, A., Jarvis, L.S., Sexton, R.J., 2011. Partial vertical integration, risk shifting, andproduct rejection in the high-value export supply chain: the Ghana pineapple sector.World Development 39 (9), 1611–1623.

UNCTAD, 2008a. Organic Agriculture and Food Security in Africa. Study prepared byRachel Hine and Jules Pretty, University of Essex and Sophia Twarog, UnitedNations New York and Geneva (UNCTAD/DITC/TED/2007/15).

UNCTAD, 2008b. Certified organic export production Implications for economic welfareand gender equality among smallholder farmers in tropical Africa. United NationsConference on Trade and Development UNCTAD/DITC/TED/2007/7.

Valkila, J., 2009. Fair Trade organic coffee production in Nicaragua—sustainable devel-opment or a poverty trap? Ecological Economics 68, 3018–3025.

van der Vossen, H.A.M., 2005. A critical analysis of the agronomic and economicsustainability of organic coffee production. Experimental Agriculture 41, 449–473.

Weber, J., 2011. How much more do growers receive for Fair Trade-organic coffee?Food Policy 36, 677–684.

Wollni, M., Lee, D.R., Thies, J.E., 2010. Conservation agriculture, organic marketing, andcollective action in theHonduran hillsides. Agricultural Economics 41 (3–4), 373–384.

World Development Indicators (WDI), 2011. The World Bank.