Doris Läpple is a PhD student at the Department of Economics, National University of Ireland, Galway and Rural Economy Research Centre, Teagasc, Galway, Ireland. E-mail: firstname.lastname@example.org for correspondence. The author acknowledges Teagasc Walsh Fellowship Scheme funding for this research, helpful comments from Stephen Hynes, anonymous reviewers and the editor David Harvey, as well as the provision of GIS and time-series data from Stuart Green and Trevor Donnellan, respectively.
There is a considerable literature about the adoption of organic farming. However, possible abandonment of organic farming has received scant attention. Thus, relatively little is known about the exit decisions of farmers. In addition, most studies are based on a static framework where it is not possible to account for changes in farmer decisions over time. This article attempts to fill this gap in the literature by investigating the determinants that affect both adoption and abandonment of organic drystock farming over time. The use of duration analysis allows for the consideration of cross-sectional and time-varying factors over the study period from 1981 to 2008. Using this dynamic econometric framework revealed a significant time effect on entry and exit decisions. Overall, the results highlight that where no attempt is made to account for exit decisions and time effects, important information about sustainable farmer decisions may not be taken into consideration.
Beginning with the MacSharry reform in 1992, several changes in the Common Agricultural Policy (CAP) have taken place which involve an increasing focus on environmental protection. As a result, several European governments, including Ireland, have actively encouraged farmers to adopt organic farming, mainly via subsidy-driven policies. Despite the levels of financial support available to organic farmers, the sector still represents only a small portion of the total utilisable agricultural area (UAA) in most European countries, averaging 4% at the end of 2007 (Willer and Kilcher, 2009). However, these incentives do not seem sufficient to secure the economic viability of organic farms in the long run, as suggested by an average of a 7.3% withdrawal rate of producers from organic farming in 2005 in the European Union (LLorens Abando and Rohner-Thielen, 2007). These figures underline that the design of effective policies to promote organic farming requires an understanding not only of the factors that lead to the adoption of organic farming, but also of the factors that induce the subsequent abandonment of this farming technique.
The Irish government has supported organic farming since June 1994 with the introduction of the Rural Environment Protection Scheme in response to Regulation (EEC) 2078/92. The support payments are based on a five-year contract. As a result, leaving the support scheme early requires the pay back of the subsidies received. The Irish government set a target of 5% of the UAA dedicated to organic farming by 2012. This resulted in changes in the policy design with higher payment rates available to farmers and increased provision of information about organic farming. Yet the uptake of organic farming in Ireland has been small relative to other EU countries, with currently 1,300 organic producers operating just over 1% of the UAA. In addition, more than 200 farmers left organic farming between 2003 and 2006. However, the small scale of the Irish organic sector is somewhat surprising when one considers that Irish conventional agricultural systems have a low level of intensity in comparison with elsewhere in Europe, where more intensive production systems tend to predominate. In fact, the typical conventional systems of dairy, beef and sheep production in Ireland are most often extensive and mainly grass-based. In particular many drystock farmers could easily switch to organic production with relatively little change in farm management practices and very little entry costs.2
A number of studies have examined the adoption of organic farming considering different influences and using various modelling techniques (e.g. Pietola and Oude Lansink, 2001; Burton et al., 2003; Genius et al., 2006). However, few studies account for the possible abandonment of organic farming. One of the few examples is a study by Rigby et al.(2001) which explores reasons for reversion from organic horticultural production in the United Kingdom. In addition to socio-demographic influences, their results reveal a decreasing probability of reversion if the farmer uses various information sources. Similarly, a study by Genius et al. (2006) underlines the importance of information provision on the adoption of organic farming in Greece. Focusing on farmers’ responses to economic incentives to adopt organic farming, Pietola and Oude Lansink (2001), utilising Finish data, reveal that decreasing output prices and increasing subsidies induce a switch to organic production. Although these studies provide valuable insight into farmer decisions, they are either based on a static framework (Rigby et al., 2001; Genius et al., 2006) or concentrate on economic incentives (Pietola and Oude Lansink, 2001) and exclude influences such as time or personal characteristics of the decision maker.
Time plays an important role in explaining farming decisions, as it captures a number of influences which shift the costs of adoption or abandonment, such as changes in the economic environment or learning from others. In comparison with static models, using duration analysis has the advantage of being able to incorporate time in the decision process. For example, it can incorporate cross-sectional and time-varying data and thereby account for effects outside of farmers’ control. In this article, two different models for each decision are estimated and compared. One is a semi-parametric model that allows for a flexible hazard rate, whilst the other one is a parametric model that requires assumptions about the shape of the hazard function prior to the estimation.
Recently, duration analysis has become more popular in the agricultural economics literature in explaining the time it takes a farmer to adopt a certain technique (e.g. DeSouza Filho et al., 1999; Dadi et al., 2004; D’Emden et al., 2006). However, few studies deal with organic farming in particular. One example is Burton et al. (2003). In this article, the authors explain the time to adoption of organic horticultural farming in the United Kingdom using duration analysis. Aside from environmental attitude and information networks, their results highlight the effect of systematic changes over the study period on the adoption decision. However, the study does not account for the abandonment of organic farming.
Clearly, there is a dearth of research considering the adoption and subsequent abandonment of organic farming over time. This article aims to fill this gap by identifying the factors that influence farmers to adopt organic farming, whilst also investigating the factors that influence exit decisions. Furthermore, the use of duration analysis gives important insight into the timing of adoption/abandonment decisions and it is possible to link these to the impact of a changing economic environment. By considering a wide range of economic and non-economic factors, in particular market effects and farmer attitudes, this research aims to provide a detailed understanding of farmer decisions. This information may help to improve policy interventions by targeting policies to farmers who are most likely to remain in the sector.
This article is structured as follows: in the next section the applied methodology is explained and the different approaches for the entry and exit models are outlined, whereas in section 3 the data is described. In section 4 results are presented and discussed, whereas some final conclusions are drawn in section 5.
Duration analysis is used in this article to model the entry and exit decision as a process of choice of when to adopt and when to abandon. In this approach the variable of interest is the length of time until a certain event occurs or until the measurement is taken (Greene, 2008). The main interest lies in the probability that a farmer will adopt/abandon organic farming at time t, given the farmer has not adopted/abandoned at that time.
Let T be a non-negative continuous3 random variable representing the length of a spell with a cumulative distribution function F(t) and a probability density function f(t). However, in duration analysis the hazard function and the survivor function are the key concepts. As they are based on the cumulative distribution function F(t) and the probability density function f(t), there is a one-to-one relationship between all functions. To begin with, the cumulative distribution function F(t) is given by:
The survivor function S(t) gives the probability that the spell is at least of length t, which means the probability of surviving beyond time t. The survivor function is expressed by:
The survivor function equals 1 at t =0 and strictly decreases towards 0 as t goes to infinity. The density function f(t), which is the slope of the cumulative distribution function F(t), can also be obtained from S(t):
In this context, the hazard function h(t) is the instantaneous rate of failure. Usually, it is this function that is of main interest. It provides the probability that the event will end in the next short interval Δt, conditional on survival up to that time:
The hazard function can vary from 0, which implies no risk of failure, to infinity, which expresses the certainty of failure at that instant (Cleves et al., 2008). The shape is determined by the underlying process. Parametric models require the specification of a functional form prior to the estimation process. Several parametric distributions are possible and Table 1 reports functional forms of the exponential, Weibull and log-logistic models, which are used in this article. The choice of a specific model is usually based on theoretical and empirical evidence (Allison, 1984). However, this approach can sometimes be inadequate to discriminate between different models that allow for a similar shape in the hazard function. One possible alternative to compare models is to estimate different models and to evaluate the fit of the models using an Akaike information criterion (AIC).
Table 1. Functional forms for parametric models
Hazard function h(t)
Survivor function S(t)
Note: λ = exp(β′X) and γ = exp(−β′X).
In duration analysis, the exponential model is often considered as the baseline, and unlike the other parametric models in Table 1, there is only one parameter to be estimated. In the exponential model, the hazard is independent of time and for this reason it is sometimes called ‘memoryless’ (Kalbfleisch and Prentice, 2002). However, owing to its constant hazard rate, it rarely fits the data well. The Weibull model allows the hazard to vary monotonically as duration proceeds. The hazard rises monotonically with time when p > 1 or falls monotonically with time when p < 1. The exponential model is a special case of the Weibull distribution when the shape parameter p = 1. A more flexible approach is to specify a piece-wise constant exponential model by allowing for the possibility that the hazard varies across specified time periods, but is constant within each of these intervals, as expressed in the following:
This model is sometimes called semi-parametric as it does not require assumptions about the shape of the hazard prior to the estimation.
Another alternative is the log-logistic model, which allows for a non-monotonic hazard rate that first increases and then decreases if p > 1 and a monotonically declining hazard for p ≤ 1. In the log-logistic model the distribution is scaled by the parameter σ, which is equivalent to p = σ−1 (Box-Steffensmeier and Jones, 2004).
Estimation of parameters in duration models follows maximum likelihood procedures, although the estimation is complicated as a result of right censoring. This occurs because at the time of measurement it is usual that not all spells are completed, that is, not all farmers have adopted/abandoned. Because the spell end dates for these observations are unknown, right censoring at time t of data collection is necessary. The only thing that is known about a censored observation is that the completed spell is of length T > t. In this sense, let ci be a censoring indicator with ci = 0 if censored and ci = 1 otherwise. Assuming independence over i, the following log-likelihood function is obtained:
with θ representing the parameters to be estimated (Greene, 2008). The contribution to the likelihood of completed spells is given by the density function f (ti) evaluated at the ending time ti and with equivalent covariate values. The second part of the term represents the right-censored cases. The contribution to the likelihood is given by the survivor function S(ti) calculated at the censored ending time ti and with appropriate covariate values.
Depending on the specification, covariates are introduced to alter the hazard. The exponential and the Weibull model are specified in a proportional hazard (PH) metric.4 This means that the covariates act multiplicatively on the baseline hazard and are independent of time. The scale parameter λ is parameterised as exp(β′X). This is a widely used specification as it takes only positive values without making restrictions on β′ (Kalbfleisch and Prentice, 2002). The parameters can be interpreted using the partial derivative of the Weibull hazard function (Table 1):
where the sign of β implies the direction of the effect and each parameter summarises the proportional effect on the hazard of absolute changes in the corresponding covariate (Jenkins, 2004).
The log-logistic model exists only in an accelerated failure time (AFT) specification, which follows the parameterisation:
with a logistic distribution for ε. The effect of the covariates is to accelerate/decelerate time by a factor of exp(−β′X) The parameters are interpreted as follows:
Therefore in this specification the parameters relate proportionate change in survival time to a unit change in a given regressor, with all other characteristics fixed (Jenkins, 2004). The vector X includes covariates, which in the simplest case are constant or assumed to be constant. Covariates can also vary over time. In this case, time is split following the change in the variables, but variables are assumed to be constant within these time intervals.
2.1. The entry model
Because this article includes only certified organic farmers, the start date (t = 0) is either 1981 when the first organic certification body began operating in Ireland, or the date when the farmer started farming as the main farm holder, whichever is latest. The end of the spell is either the date when the farmer adopted organic farming,5 whereas for the conventional farmers, spells are right-censored at the time of data collection. A piece-wise constant exponential model and a Weibull model are estimated for the entry decision.
2.2. The exit model
In the exit model, the beginning of a spell is the date when the farmer adopted organic farming (t = 0). The end of a spell is either the date when the farmer dropped out of organic farming, whereas for the organic farmers, spells are right-censored at the time of data collection. As the hazard is expected to be non-monotonic, a piece-wise constant exponential model and a log-logistic model are estimated for the exit decision.
The main part of the data used in this study is from a nationwide survey for Ireland, conducted between July and November 2008. To be classed as an organic farmer, the farm had to be registered as organic with the Department of Agriculture, Fisheries and Food. To be considered as an ex-organic farmer, the farm had to be registered as organic at some point in the past. For the organic and the ex-organic farmers complete address lists were available from the organic certification bodies. A survey was sent to each farmer in these groups. A response rate of 40% for the organic farmers and 22% for the ex-organic farmers was achieved following an announcement of the survey in the Irish Farmers’ Journal newspaper and one reminder letter. The data for the conventional farmers were collected through farms in the Teagasc National Farm Survey (NFS; Connolly et al., 2008).6 After restriction of the analysis to drystock farms and elimination of questionnaires with missing data, the final sample consisted of 341 organic, 41 ex-organic and 164 conventional farmers.
The very small number of organic farms and in particular ex-organic farms in Ireland implies that complete random sampling would not have generated a large enough number of organic and ex-organic farmers for an empirical analysis. Overall then, taking into account the small amount of organic and ex-organic farms in Ireland, these farms are well represented in the analysis, whereas the number of conventional farms in the sample is small considering the proportion of organic farms to the total number of farms in Ireland. Although not representative of the general farming population, the sample provides a good representation of the types of farm operators who participate in organic farming in Ireland as well as the conventional drystock farm operators.
Following the literature, the adoption of a new technology in agriculture depends on a variety of different factors. For example, economic and structural characteristics (e.g. DeSouza Filho et al., 1999), farmer and household characteristics, such as age, education and household size (e.g. Feder et al., 1985), as well as information provision (e.g. Feder and Slade, 1984; Genius et al., 2006). However, a growing body of literature, especially in the uptake of environmental measurements, stresses the importance of farmer attitudes in that decision (Wynn et al., 2001; Defrancesco et al., 2008). In addition, following the literature on disadoption of agricultural technologies (Carletto et al., 1999; Walton et al., 2008), adoption and disadoption are expected to be affected by broadly similar factors, mainly with opposite effects. Based on these findings and on the availability of data, several economic and non-economic factors are considered in this study. Table 2 provides summary statistics for the variables included in the analysis.
Table 2. Descriptive statistics of explanatory variables for the sample
Organic (n = 341)
Ex-organic (n = 41)
Conventional (n = 164)
Notes: *Time-varying variable increasing by 1 with each year of farming.
UAA, utilisable agricultural area.
= 1 if the farmer has an off-farm job
UAA in hectares
Livestock units per hectare
= 1 if soil has a limited use range
Distance to nearest organic mart in km
Farming experience in years*
= 1 if farmer has higher education
Number of household members
Higher value = higher environmental concern
Higher value = higher profit motivation
Higher value = more risk-averse
Higher value = higher information seeking attitude
Distance to nearest advisory office in km
Distance to the nearest organic demonstration farm in km
Index for: farm advisor, information event, training course
Index for: magazines, TV/radio, internet
Knows organic farmer
= 1 if farmer knows another organic farmer
Four attitudinal variables are included in the analysis to reflect the role of farmers’ attitudes in the decisions to adopt/abandon organic farming. The variables included in the analysis are based on a set of 35 attitudinal statements. The initial statements were measured on a seven-point Likert Scale. Principal component analysis with orthogonal (varimax) rotation was used to extract the main components. Consequently, the variables included in the analysis are the calculated component scores.
The variables comprising distances are a surrogate to account for access to organic markets (distance mart) and technical information about (organic) farming (distance advisory, distance demo). The distances are measured as straight lines using ArcGIS based on the X–Y coordinates of the farms which are received from the database GeoDirectory Q1 2008 (Fahey and Finch, 2009).
The variables info advisory and info media are an attempt to measure information use of the farmer. Info advisory measures the frequency of consulting a farm advisor, attending an information event and agricultural training course divided by 3, whereas info media measures how frequently the farmer uses magazines/press, TV/radio or the internet as a source of farming information, also divided by 3.
Two variables are included to incorporate the effect of a changing economic environment. For this purpose, annual values for output and input prices were obtained from Eurostat, and, for a given year, they are the same for all farms in the sample. The series indicates the calculated development of conventional cattle output prices in relation to input costs and serves as a proxy for the profitability of drystock farming. Figure 1 shows the time path of the price series included in the entry7 model (cattlet). In addition, a time-varying dummy variable switching from 0 to 1 in January 2005 is included. The variable single farm payment (SFPt) is an attempt to capture the effect of the decoupling of payments from production, introduced by the 2003 CAP reform.
4. Results and Discussion
4.1. The entry model
The results from the estimated entry models are presented in Table 3, whereas the respective hazard functions are depicted in Figure 2. Two alternative specifications are considered. Model 1 is a piece-wise constant exponential model, with variation in the hazard in the first eight years, but constant thereafter.8 Model 2 is a Weibull specification. Overall, the models show very similar results. However, using an AIC statistic, the preferred model is Model 2 with an AIC statistic of 1,180.39 in comparison with Model 1 with an AIC statistic of 1,318.20. In addition, Model 2 also shows the larger log-likelihood. Because both models are specified as PH models, the signs of the coefficients indicate the direction of the effect that explanatory variables have on the conditional probability of adoption. Hence, the estimated coefficients of the models are directly comparable.
Table 3. Estimates of duration models for the adoption of organic farming
Piece-wise constant exponential model (Model 1)
Weibull model (Model 2)
Notes: Number of observations = 546; ***P < 0.01; **P < 0.05; *P < 0.1.
Looking at the plots of the estimated hazard functions in Figure 2,9 the values of the time dummies in Model 1 and the shape parameter (p) in Model 2 (Table 3), both models show a rapid decline in the hazard after the first periods of duration, although the exact time path is different. In Model 1, the negative signs of the time dummies indicate a lower propensity to adopt in later years of farming in comparison with the first year of farming, which is taken as the reference category. This is also evident in Figure 2, which shows a higher hazard in the first year of farming in comparison with the subsequent years. The value of the shape parameter p in Model 2 is 0.51. This implies a declining hazard rate, which is also depicted in Figure 2. The difference between the two hazard functions can be explained by the different forms specified; one is quite flexible whereas the other one allows for the hazard to vary monotonically only, as well as a decreasing amount of data as analysis time increases.
The variables capturing the economic development over the study period emerge as important to the adoption decision. It is also notable that the magnitudes of the coefficients (cattlet, SFPt) are very similar in both models, confirming their effect on adoption. In line with DeSouza Filho et al.(1999), the development of conventional output prices relative to input prices (cattlet) had a negative effect on the hazard to adopt, suggesting a decreasing probability to adopt organic farming with an increasing profitability of conventional farming. The time-varying dummy variable (SFPt) captures the effect of the decoupling of payments from production, switching from 0 to 1 at the date of introduction in January 2005. Because this incentive favours more extensive farming systems, the variable shows the expected positive sign.
Some structural variables are also found to affect the adoption decision, with farm size and livestock density showing a statistically significant relation to the hazard to adopt. Again, the direction and magnitude of the coefficients (UAA, LU/ha) in both models largely match. Size of the farm measured in UAA has a slightly negative effect on the hazard to adopt. It should be pointed out that the marginal payment level of organic farming decreases with an increasing farm size. The results from this article are consistent with the findings of DeSouza Filho et al.(1999) of a negative association between farm size and uptake of sustainable agricultural technologies, but inconsistent with Burton et al. (2003) and Hattam and Holloway (2007) who found farm size to be not significant on time to adoption of organic farming. A higher livestock density, having a negative correlation to the hazard, appears to act as a constraint on adoption, also found by Wynn et al. (2001). Uptake of environmental schemes is associated with more extensive farming systems, as a result of lower entry costs. In contrast to Hynes and Garvey (2009) who report that farms with poorer soil are more likely than other farms to enter an agri-environmental scheme, in this article no significant influence of soil quality (soil) could be detected.
Increasing farming experience provides better knowledge about the environment in which decisions are made, and therefore, unlike age, an effect in both directions is possible. Farming experience, increasing by 1 for each year, has a positive effect on the hazard to adopt in Model 2, whereas no significant relation could be detected in Model 1, which puts its effect on adoption into question. Similarly, Burton et al. (2003) report no significant effect of a time-dependent age variable on the adoption of organic farming using duration models.
Amongst the variables describing farmer attitudes, environmental and risk attitude are significantly correlated with the probability of adoption. The magnitudes of the coefficients are very similar in the two specifications, confirming their effect on adoption. Farmers who express a higher level of environmental concern face a higher probability to adopt organic farming, also found by Burton et al. (2003). Similarly, Vanslembrouck et al. (2002) and Defrancesco et al. (2008) stress that environmental awareness increases the probability of entering environmental schemes. Considering the fact that farmers adopting organic farming have to leave established markets and familiarise themselves with a different farming technique, the coefficient of risk attitude also shows the expected effect, with risk-averse farmers being less likely to adopt. This outcome is supported by Serra et al. (2007) who show that organic farmers are less risk-averse than their conventional counterparts.
Unlike previous findings (Genius et al., 2006) that report a positive relation between number of different information sources consulted and the adoption of organic farming, no such effect could be detected in this study. The variables included in this study are an attempt to measure the impact of information, by capturing access and frequency of information use. The distance to the nearest advisory office is the only information variable that shows a significant influence on the probability to adopt organic farming, although it is not significant in Model 1. However, according to the expectations, the hazard decreases with an increasing distance.
In terms of social influences, the variable denoting whether the farmer knows another organic farmer or not has a strong positive correlation with the hazard to adopt. Again, the coefficients in both models show similar magnitudes, which is reassuring. This supports extant findings that existing organic farmers are an important source of information and expertise for converting farmers (Lampkin and Padel, 1994).
5.1. The exit model
Two alternative specifications for the exit model are reported in Table 4, whereas the estimated hazard functions are displayed in Figure 3. Model 3 is a piece-wise constant exponential model, with variation in the hazard within the first 11 years, but constant thereafter. In this model time is held constant within years 1–4, 5 and 6, 7–11 and 12–18. This is because of few failures in the data and therefore it was not possible to split the hazard into shorter time periods. Model 4 is a log-logistic specification, which allows for a non-monotonic hazard rate. The two models show similar results and comparing an AIC statistic the preferred model is the piece-wise constant exponential model, which also has the larger log-likelihood value. The estimated AIC statistic for Model 3 is 249.50, whereas Model 4 shows a higher statistic with 257.24. As explained previously, the signs of the coefficients in Model 3 show the direction of the effect on the conditional probability of abandonment, whereas in Model 4 a positively/negatively signed coefficient implies that the expected time in organic farming increases/decreases. This implies that the magnitudes of the coefficients of the two models are not directly comparable, but one expects opposite signs in both models for the same coefficients.
Table 4. Estimates of duration models for the abandonment of organic farming
Piece-wise constant exponential model (Model 3)
Log-logistic model (Model 4)
Notes: Number of observations = 382; ***P < 0.01; **P < 0.05; *P < 0.1.
Both estimates show a non-monotonic hazard rate (timedummy, σ), which is also depicted in Figure 3.10 The positive values of the time dummies in Model 3 imply a higher hazard to exit organic farming in comparison with the first four years of organic farming (reference category), with farmers most likely to drop out in years 5 and 6 (timedummy2). This is also evident in Figure 3 and was expected by the data, considering the five-year commitment to receive organic subsidy payments. In addition, the scale parameter σ of Model 4 implies a first increasing and then decreasing hazard, as depicted in Figure 3. The difference between the graphs is explained by the different functional forms specified for the hazard, a small failure rate, as well as a decreasing amount of data with increasing time. Furthermore, few failures in the exit models explain the small hazard rate, which should be taken into account when comparing the graphs of the hazard functions.
The variable denoting if the farm holder has an off-farm job is positively related to the probability to exit organic farming. This is based on a positive sign of the respective coefficient in Model 3 and a negative sign of this coefficient in Model 4, which implies that time to abandonment decreases. Farmers who are involved in off-farm work face higher opportunity costs of labour, especially if the new method requires higher labour input (Moser and Barrett, 2003). Farmers involved in off-farm work also have limited time to spend on their farm which complicates the ease of adjustment to the new system and meeting the organic regulations.
This effect is confirmed by the effect of some structural variables. Farmers with a higher stocking density (LU/ha) are found to be less likely to exit from organic farming and both models show the expected opposite signs. This is not surprising considering that organic farming is in general assumed to be more labour-intensive (Offermann and Nieberg, 2000), especially at the beginning when the farmer has to adjust to the new farming technique and establish new markets. The distance to the nearest organic mart serves as a proxy to account for access to organic markets, but the variable does not show a significant effect. This was expected to be an important factor given that lack of organic market outlets was stated as one of the main reasons to exit organic farming (Läpple, 2008).
Considering the attitudinal variables, environmental attitude has a negative significant correlation with the hazard to exit, suggesting that farmers who show high environmental concern stay longer in organic farming. Similarly, Rigby et al. (2001) describe an increasing likelihood of reversion from organic farming of farmers who converted because of mainly economic reasons.
Finally, similar to Carletto et al. (1999) none of the farmer characteristics are significantly related to the probability to exit. Furthermore, looking at the variables that represent the impact of information and social interaction, none of these factors were found to be significantly correlated to the hazard to exit, although the literature underlines an influence. According to Bravo-Ureta et al. (2006), farmers who have access to information or use more information sources have a lower hazard to abandon soil conservation technologies. Furthermore, difficulties in interpreting information and converting into a useful management plan are cited as a reason to abandon a technology (Griffin and Lambert, 2005) and using more information sources helps the ease of adjustment to the new system.
This article used a duration analysis approach to model both adoption and possible abandonment of organic farming. By explicitly considering subsequent abandonment in a dynamic framework, this article is the first such study in this field. A variety of farm and farmer characteristics, as well as market effects over the study period from 1981 to 2008 have been considered. In addition to thoroughly investigating the impact of time, two models for each decision have been estimated and compared; one semi-parametric model which allows for a flexible shape of the hazard and one parametric model which makes strong assumptions about the hazard function. Overall, the models highlight three key factors affecting the adoption and abandonment of organic farming.
First, the results underline the importance of farmer attitudes and social interaction, similar to previous studies by Wynn et al. (2001) and Defrancesco et al. (2008), although these mainly focus on environmental attitudes. By using a comprehensive set of attitudinal statements, this study revealed that risk-averse farmers are less likely to adopt, whereas farmers who express environmental concern are more likely to adopt. Moreover, farmers who express a higher level of environmental concern are also less likely to leave the sector. This latter result has not received much attention in the literature yet but further underlines the importance of environmental attitude. However, a note of caution is required when interpreting these outcomes. According to Ajzen (2005), attitudes are assumed to be relatively stable although they may change with the receipt of new information. In contrast, to take advantage of a possible change of attitudes, raising farmers’ environmental awareness may help to increase the number of organic producers in the long-term. As the results of this article suggest, once these farmers convert to organic farming they are also more likely to continue organic farming in the future.
Second, the results of the entry model highlight the impact of market effects. The implementation of the decoupling of payments introduced by the 2003 CAP reform had a positive effect on adoption, whereas increasing profitability of conventional farming slows the adoption of organic farming. In this context, it would have been desirable to have information on the profitability of organic farms over the study period to allow one to consider its effect on exit decisions. Unfortunately, this information is not available for Irish organic farms. Although a limitation, the dataset still provides valuable insight into exit decisions.
Third, in both models a clear time effect emerged, as suggested by the significance of the timedummies of the semi-parametric models and the shape parameters of the parametric models. The estimated signs and magnitudes of the timedummies imply that farmers are most likely to adopt in their first year of farming. In addition, according to the policy design, farmers are most likely to exit after the first five-year contract expires, suggesting that farmers encounter problems with organic farming.
In conclusion, considering both the impact of market effects and the contract expiry effect suggest that a more sophisticated approach to mainly subsidy-driven policies is required. Because farmers are sensitive to price changes, offering fixed organic price premiums and better market outlets may encourage farmers not only to convert but also may secure the long-term economic viability of organic farms. In addition, targeting and providing existing organic farmers with information, especially in their first years after conversion, appears to be important in reducing the numbers of farmers exiting organic farming. The latter has not been an aspect much investigated in the literature, whereas the importance of information provision to increase the uptake of organic farming is underlined by previous studies (Burton et al., 2003; Genius et al., 2006). Although in this study, no clear effect of information provision on farmer decisions could be evaluated, it would be unwise to rule out its importance. It may be too early to identify any discernable effect that is related to the recent increase in the provision of organic information in Ireland or indeed to establish if the available information is still limited. This remains an area for further research in the future.
This study focuses on the drystock sector, because significant numbers, necessary for an empirical analysis, can be found in this sector.
A continuous time specification is used, as spells are measured in months (Jenkins, 2004).
It is also possible to specify the exponential and Weibull models in an AFT metric, which shifts the focus towards actual survival time. In PH models, the main focus is on the hazard function and how the hazard changes with the values of the covariates, which is of greater interest in this study. With constant covariates, the PH and AFT models of the Weibull distribution coincide and coefficients are related as (Jenkins, 2004).
The date when the farmer started the conversion period is used.
The NFS is based on approximately 1,100 farms representing 104,800 farms nationally. The NFS data are EU-Farm Accountancy Data Network (FADN) for Ireland. Data collection for this survey was restricted to a certain time period; therefore, the present sub-sample cannot be regarded as fully representative.
It was impossible to identify appropriate data for organic cattle output prices.
The inclusion of more timedummies for subsequent years did not alter results significantly.
The comparison is based on mean (median) values of significant variables.