The Case of Organic Dairy Conversion in Norway: Assessment of Multivariate Neighbourhood Effects

This study examines the impact of neighbourhood effects and individual farm characteristics on the decision process of organic dairy conversion in Norway, using a unique, spatially explicit farm-level panel set comprising information at the population level from 2003 to 2015. Our results reveal a positive spatial spillover of neighbouring conversion, confirming previous findings. Additionally, we demonstrate that neighbouring organic dairy reversion (i.e. switching back to conventional dairy farming) and organic dairy exits (ceasing to farm altogether) exert notable negative spatial spillovers on organic conversion decisions which have not yet been shown in the literature. If organic dairy production is an important policy goal, such negative spatial spillover requires consideration within policy design and extension.


Introduction
In several countries, policymakers are taking action to promote the adoption of organic farming practices as one way to reduce the environmental impacts of farming in addition to addressing growing consumer concerns regarding food safety and health. The example considered in this study is Norway, where policymakers set ambitious goals for the promotion of organic farming, offering a range of subsidies to support conversion. Although the area of organic land and organic farm management has increased in Norway since the beginning of the millennium (Flaten and Lien, 2006;Røsnes, 2010; Norwegian Ministry of Agriculture and Food, 2010), the goal of the government to increase organic production and consumption from 0.5% in 2005 to 15% by 2020 (Norwegian Ministry of Agriculture and Food, 2009) has fallen far short of this goal. In 2018, only 2.2% of the entire food market was organic, and only 4.4% of the total agricultural area was directly under organic cultivation (Landbruksdirektoratet, 2018). Achieving these ambitious goals requires a more effective policy design to promote the adoption of organic farming practices. We should enhance our understanding of farmers' motives, reservations and goals in terms of organic production to develop more informed policy approaches.
Existing research studies describe organic conversions from different perspectives, encapsulating it as a problem of optimal timing with an uncertain investment that incurs sunk costs (Musshoff and Hirschauer, 2008), analysing the role of moral and social considerations (Mzoughi, 2011). Further studies assume that potential adopters make a discrete choice based on the possibility of maximisation of expected utility subject to prices, profits and inputs, devoting elevated importance to economic and financial concerns (Koesling et al., 2008;Lewis et al., 2011). Some studies categorise farms according to their characteristics (e.g. farmers' attitudes, age, herd and land sizes, natural resource assets or risk deliberation) Knowledge transfer regarding organic technology development and joint input purchase may further reduce barriers to conversion. Hence, co-operation in general leads to a smoother gateway to the supply chain.
One limitation of existing theory and empirical studies on the neighbourhood effect is that only positive external spatial effects have been covered (e.g. Lewis et al., 2011). In this study, we argue that negative external effects can be similarly influential on farmers' conversion decisions. Neighbouring farmers' reversion from organic to conventional farming 1 or neighbouring organic farmers' exiting farming altogether could serve as a negative example that potentially reduces the likelihood of conversion. Here similar mechanisms of the neighbourhood effect can be expected to act correspondingly to positive external effect.
Those negative occurrences could provide knowledge about the disadvantages and challenges of organic production or lead to a weakening of supply chain access in the local neighbourhood.
To our knowledge, this potential deterrence to organic conversion has not yet been studied.
For policymakers and extension services that seek to promote the adoption of organic farming, the role of negative examples must be understood to design the right policy strategies and set achievable goals. With the establishment of only positive externality policies, extension services solely focus on promoting conversion by as many farms as possible, irrespective of how many revert at a later stage. However, the existence of potentially negative externalities warrants a more selective approach to the promotion of organic conversion by only farms that have a low likelihood of reversion. Additionally, this approach would justify additional effort and support concentrating on keeping farms within organic production, rather than only focusing on the promotion of adoption as the primary goal.
Although the question of the role of negative examples is relevant more broadly, we restrict our study to Norway and specifically to dairy production as one of the most prominent agricultural activities in Norway. In this case, we can use a unique yearly panel dataset The remainder of the paper is structured into five sections: Section 2 introduces the natural and economic environment for dairy production in Norway; Section 3 details the adopted and extended theory; and Section 4 presents the data, the applied method and the potential endogeneity issues. Finally, Section 5 discusses the results, and Section 6 concludes and highlights the policy-relevant findings.

Norwegian Case
Norwegian agriculture is characterised by unfavourable natural conditions 3 but enjoys great social and governmental appreciation and support. In 2010, the utilised agricultural area (UAA) was approximately 3.1% of the territory of the country (Eurostat, 2019), measuring about 1 million hectares; one of the smallest ratios within the European Economic Area.
Weather conditions, coupled with the peculiar morphology of the territory, limit agricultural production. In 2010, the most common holdings specialise in sheep, goats and other grazing livestock (27%) and dairy (18%), and the rest are specialist cereals, oilseed and protein crops, general field cropping, specialist cattle-rearing and fattening of the total population of farms (Eurostat, 2012 Agricultural policy endeavours to promote a diversified agricultural sector delivering a 'multifunctional agriculture' (Potter and Tilzey, 2007;Bjørkhaug and Richards, 2008;Forbord et al., 2014). Another goal is to maintain family farming identities (Forbord et al.,2 Excluding only very small farms that do not qualify for any farm subsidies. 3 Norway is one of the northernmost countries of the globe (58-71° latitude) covering oceanic and continental type of climates. The grazing season varies from three to nearly 6 months. In the southern regions, the grazing season runs from May to October, whereas further north, it is shorter, from mid-June to mid-September. 2014) and ensure that the current UAA is sustained. To achieve these objectives, government support for agriculture is significantly higher in Norway than in the EU (OECD, 2017).
Policies targeting organic agriculture came into effect in 2002 and have been renewed several times with an increase in payment rates 4 and organic conversion grants to encourage farmers to convert to organic production. In 2005, the long-term strategic plan adopted by the government sought to increase organic production and consumption from 0.5% to 15% of overall production and consumption by 2015. That ambition was somewhat reduced in 2009, projecting this desired level by 2020 (Norwegian Ministry of Agriculture and Food, 2009).
Organic farming incurs more costs than conventional farming attributable to increased labour intensity, the prohibition of mineral fertiliser and non-organic plant protection and the use of specific technological settings (Flaten et al., 2005). To compensate for these additional costs, organic farms receive significantly higher financial support than conventional producers.
Norwegian agricultural policy divides the country into seven agricultural subsidy zones because of the major differences in longitudinal production conditions. For the organic livestock subsidy, including dairy cows, suckling cows and other cattle, the organic bonus rates vary between regions, with the northern zones receiving more financial support than the southern zones until 2015. For instance, the organic premium for dairy cows in 2003 was 880 NOK 5 in the north and 630 NOK in the south on top of the average 2500 NOK per animal for conventional dairy cows. 6 The geographic differences in such organic bonuses adjusted from almost 30% in 2003 to no difference in 2015. The same trend can be observed for the subsidy adjustments of suckling cows and other cattle. Moreover, organic dairy farmers receive not only livestock-based subsidies but also regional payments. Payments are provided to arable land, including cultivated pastures, corn, potatoes, green manure production and other ecological driven land use, the rates of which are not differentiated by region. motives for organic conversion: food quality; professional challenges; soil fertility and pollution problems; ideology and philosophy; health risks (pesticides, etc); animal welfare; profitability; organic farming payments; natural conditions (soil, climate, etc) and income stability.
reasons for all groups (early, mid and late converters). Furthermore, the motivational force of income stability was cited as the least decisive motive. In Norway, since the 1990s, off-farm income generation has rapidly increased (Fleming and Lien, 2009), and dairy farming is generally a secondary engagement for the majority of Norwegian dairy farmers. This explains the numerus cases of 'hobby' farms featuring rather small herds. These findings indicate that any analysis from a strictly economic perspective-e.g. coupling potential positive net value to organic conversion-should be expected to have only limited explanatory power in Norway. This dichotomy coincides with Lund and Algers' (2003) description of Norwegian organic farmers' motivations consistently being viewed as idealistic and reaching beyond financial concerns.
According to the Norwegian Agriculture Agency (Landbruksdirektoratet, 2018), in 2018, 2.2% of the entire food market was organic, and only 4.4% of the total agricultural area was registered organic. In the same year, the share of organic milk was up to 3.4% of total milk production, whereas Norwegian organic meat production was almost 1%. The government's ambition of increasing organic production up to 15% by 2020 has remained unsuccessful, despite a massive increase in organic livestock and subsidy payments. In June 2018, Norway announced a new national strategy to promote organic farming until 2030. Support for organic farming has remained constant since 2014, but the organic conversion grants were phased out beginning in 2010 and eliminated altogether in 2016. The government's justification for terminating organic conversion grants was to advance the simplification of the support scheme for organic farmers (Pekala, 2019).
In our observation period (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), the number of all dairy farms decreased by almost 60%, but the number of operating organic dairy farms increased by 45% (Table 1). The general annual statistics demonstrate that the Norwegian dairy sector declined, losing almost 60% of its milk producers between 2003 and 2015. This decrease was a result of the reversion of previous organic dairy farms to conventional dairy farms, which is beyond the scope of this paper. Our focus remains on the aspects of neighbouring conformity in operating organic dairy farms (ΣODF in Table 1 Table 1) is approximately 20 farms annually.
Considering the subsequent organic policy developments, in 2010, 2012 and 2013, the number of newly established organic dairy farms is even lower at the post-policy time period than prior. Therefore, the first organic governmental intervention seems to have been successful, whereas additional increases to the organic payment rates do not appear to motivate organic conversion.

Theory
One theoretical explanation of emerging business trends is often tied to the patterns of the physical localisation of decision-makers; therefore, analysing the influence of regional proximity within the course of micro-behavioural decision factors and the possible inferences between spatially corresponding units is of core interest in agriculture and resource economics literature (Bockstael, 1996;Weiss, 1996;Nelson and Hallerstein, 1997;Durlauf, 2001, 2007;Anselin, 2001Anselin, , 2002. Explicitly including the neighbourhood effect in quantifying the social determinants of conversion decisions incorporates the concepts of social learning (Conley and Udry, 2010) and spatial spillover and landscape development (Irwin and Bockstael, 2002). Consequently, spatial processes can be assumed to have a strong influence on farmers' organic conversion decision. We adopt the theoretical concept of developing neighbourhood relations, relying on the assumption that formerly converted neighbours may be instrumental in swaying the decision of potential converters by reducing the fixed cost of learning and risk, contributing to advanced supply level and in enhancing the competitiveness of the sector. Technology adoption literature explores the causes and effects of wide technological distribution in terms of economic and physical environments, individual characteristics, production uncertainty and spatial spillover (e.g. Lindner, 1980;Feder and O'Mara, 1982;Tsur et al., 1990;Adesina and Baidu-Forson, 1995;Purvis et al., 1995;Koundouri et al., 2006;Oliveira and Martins, 2011). For example, major road networks or locally favourable production conditions may both contribute to spatially clustered behaviour. Economic units may be spatially dependent, as locational attributes influence each other and thus tend to share similar attributes (Goodchild, 1992;Moffitt, 2001). Lewis et al. Our key hypothesis in this context is that such diffusion processes are influential in two directions. In previous literature, it is primarily considered that new conversions lead to positive externalities (Lamine and Bellon, 2009;Geniaux et al., 2011;Bouttles et al., 2019).
In this study, we hypothesise that reversions (or exits of organic farms) can exert similar negative externalities; an aspect that has yet to be considered in the literature to our knowledge. As shown in Table 1 iii. the number of neighbouring organic dairy exits at t−1 (which represent bad examples and assimilate a meaning of redundant milk production).
The actual numbers of multivariate neighbour vectors depend on the definition of the radius surrounding each individual farm. Defining this radius is relatively arbitrary and can hardly be based on precise information or theoretical knowledge. We intend to keep the focus on face-to-face information transfer, which assumes close proximity and personal discussions about the technology. For this reason, we calculate two small radii within 3 and 5 km. There are collateral arguments for determining two seemingly similar ranges: (i) the coefficients under different radius settings are comparable; (ii) and opens an opportunity for a potential robustness check on the model outcomes.
The applied econometric model is a reduced form discrete choice panel data model with binary probit link to explain the probability of organic dairy conversion. The 13-year dataset creates unbalanced panels of repeated observations of farm-level decisions.

Data
As noted, our analysis is based on a Norwegian dataset that is unique for two reasons: first, direct payments (DP) are available for virtually any type of crop and animal; second, the Information Act requires full transparency of all beneficiaries of government spending.
Hence, the amount of agricultural land used, herd size and farmers' age are made publicly available for each farm that receives DP. Additionally, geographic information, which is administered by the Norwegian Agriculture Authority, is available for research purposes. The data used in this application is therefore a unique georeferenced farm-level population set of Norwegian dairy holdings that received DP. The 14-year unbalanced panel construction includes more than 179 thousand observations including the annual contributions of 55 thousand dairy farms. We record approximately 4 thousand organic dairy farms in the observed period.
On the basis of the geographic information of each farmstead, we calculate the kilometre distances between the dairies, labelling each dairy farms as either conventional, organic, reverting or exiting in each year, to allow the determination of the dynamic neighbourhood synthesis of individual farms. In this manner, we derive the number of organic, reverting and exiting farms around each conventional dairy (i.e. potential organic convertor) within 3 and 5 km radii. 9 Organic certification is awarded by a Norwegian certifying agency after fulfilling 3-year certification standards for land and 1-year certification standards for a dairy herd. In the estimation, we consider the year that the conversion decision is made and not the certification date. Reversion from organic to conventional is easily identified in the dataset as farms that no longer receive the additional payments for organic production. Table 2 presents the summary statistics of the neighbour variables within 3 and 5 km radii.
Additionally, farm characteristics such as herd size, cultivated area and farmers' age are also available. Table 3 presents the summary statistics of those variables for organic dairy farms.
The mean of organic herd size and land is continuously increasing. Both measures nearly doubled over the examination period. However, the slowly increasing but later constant number of organic cows illustrates the lack of organic breakthrough in the Norwegian dairy industry.

Method
Organic dairy conversion is assumed to be conditional on farm characteristics and on quantified neighbourhood relations. We encode the latter as categorical variables with five levels (0, 1, 2, 3, 4 or more than 4 neighbours) for the number of neighbouring organic farms and neighbouring organic exits and three levels (0, 1, 2 or more than 2) for the number of neigh- 10 The number of individuals surrounded by more than four neighbours is insufficiently small (i.e. lack of frequency); therefore, we assign a joint group for limited individuals collectively. The same applies to creating three categorical levels for reversions instead of five.
where the conditional expectation is a set of vectors of individual characteristics (herd and land sizes and farmers' age 11 ); 12 is the set of factor variables of organic neighbours, organic reversions and organic exits surrounding farm n at time t. is a vector of period tspecific dummy variables, which is assumed to absorb all period-specific shocks, e.g. policy change at the major organic milk processor company that are the same for all farms; is a vector of region r-specific dummy variables that capture time-invariant regional differences in the conditions for organic production. The regional dummies are identical to the agricultural subsidy zones. Moreover, the interaction terms between regional dummies and year-specific dummy variables are included to control for time-varying effects that are the same for all farms in one region. Finally, denotes the unobserved effect, which is restricted in the applied correlated random effect framework and appears additively inside the standard normal cdf (Papke and Wooldridge, 2008). The inclusion of regional and time fixed effect as well as interaction effects is intended to control for potential endogeneity. The specific identification approach is discussed in the next section. The conditions for organic production might differ across space through spatial differences in organic pay price and isomorphic biophysical aptitudes of micro-regions (e.g. valleys or highlands). To control such effects, we include regional fixed effects . Conditions for conversion might also vary across time (equally for all farms); for example, due to a change in the subsidy scheme or a change in market prices or regulations. We control for those effects by including time fixed effects, . There might also be changes that vary across time and regions. Those effects are controlled for by adding interaction terms between the regional and yearly fixed effects. Although those sets of fixed effects should capture a substantial part of the unobserved confounders, there might still be unobserved effects that vary across space at a smaller spatial scale than those variation captured by the regional fixed effects. To control for those effects, we follow Lewis et al. (2011) considering a Mundlak-Chamberlain (MC) device in a correlated random effects estimation strategy (Mundlak, 1978;Chamberlain, 1982;Papke and Wooldridge, 2008). In this setting, we decompose the unobserved individual characteristics into a mean zero normally distributed random variable and the average of the time-varying explanatory variables: Here, is commonly referred to as an MC device defined as the average of all timevarying explanatory variables. In our specific case, it is the average of all spatial neighbourhood variables capturing average neighbouring farm convergence, reversion and exits of organic farmers. In principle, the MC device is identical to the usual fixed effect estimate in linear cases, but the algebraic equivalence does not hold in nonlinear cases (Papke and Wooldridge, 2008). With this approach, we control for the time-invariant unobserved spatial effects on a smaller spatial scale as captured by the regional fixed effects. For example, we can capture the fact that farms in a certain location are more likely to convert to organic. A specific example is the connection to major road networks as a crucial aspect in optimising the collection routes of dairies. The main dairy cooperative in Norway that collects organic milk, TINE, employs business agents to urge those farms that are favourably located to the circuit of the collection trucks to convert. Those types of effects are covered with the MC device as long as they are not time-varying. In fact, in terms of the specific example, TINE decided to cease personal marketing in 2011 since the market for organic dairy was saturated; instead, the focus turns toward preserving and strengthening those who had already converted (Skjelvik et al., 2017). This could potentially lead to confounders that are both spatial and time-varying and hence not fully covered by the MC device or the fixed effects potentially leading to some bias and overestimation of the spatial interaction effects that cannot be controlled by the identification strategy. However, we consider those specific types of confounders, which are both time and spatial varying, as a rather specific case and an acceptable limitation of the identification approach. Table 4 presents the parameter estimates for the predictor variables. The set of indicator variables, the annual and zonal dummies, are obtained in the Appendix. The probit model outputs, for both 3 and 5 km radii, show the coefficients, their standard errors and the zscores, as well as the associated p-values.

Results
Generally, for both radius settings, the results of the factor variables for describing neighbourhood effects and the period-specific variables support the hypothesised tendencies in the decision-making process of organic conversion. Specifically, we find an increasing probability of organic conversion with a higher number of organic neighbours. As the reference level is 0, which implies no organic neighbour, the first contrast predictor vector Organic Neighbour 1 increases the z-score by 0.275 at 3 km and 0.249 at 5 km radius.
Organic Neighbours 2, 3 and 4 demonstrate increased positive effects on conversion, confirming the hypothesis that more neighbours increase the likelihood of organic conversion.
In case of considering the 3 km radius, the parameter vector of 4 or more organic neighbours is insignificant because the occurrence of 4 or more than 4 organic neighbours is very small.
However, when we increase the observation window from a 3 to 5 km radius, the effect becomes statistically significant. Individual reasons may vary between moving, merging, family issues, age, etc, but none of these relates to any system error in the organic dairy industry. However, in case of more than one organic exit at a time, we are entitled to surmise the presence of some unobserved negative external effect that directly makes them exit simultaneously.
Furthermore, as of the farm-specific attributes, our assumption that youth-induced risk-taking ability and potentially higher level of interest in environmental farming promote organic conversion-proxied by farmer's age-seems to be accurate because of the significant and negative coefficients in both models. It indicates that younger farmers are indeed more likely to convert to organic. We find the coefficients of herd and land sizes very small and consider them marginal in the decision process of organic conversion.
The period-specific dummy predictors show a negative tendency. This confirms that the 2005 and 2009 organic farm policy goal is still far from reach, and the Norwegian organic dairy industry is slowly declining. After 2010, this decrease accelerates in both models, which corresponds to the significant decrease in the summary statistics of organic dairy entries in Table 1. The negative and significant parameter coefficients verify that more organic dairy farms exit organic production than conventional farms decide to convert to organic.
The identical model setting for 3 and 5 km radii allows us to analyse the likely differences

Conclusion
This article explores the factors of organic dairy technology adoption in Norway during the years 2003-2015. We find that farmers' age, herd size and agricultural land used are less important elements of potential organic conversion than the neighbouring dairy relations. We demonstrate that spatial spillovers across economic agents are influential in the patterns of technology adoption. Although our model supports the findings of previous studies that the presence of organic dairies in close proximity positively affects organic dairy conversion, more remarkably, we find that neighbouring organic reversions and exits introduce potential negative interaction effects that had not been considered in previous literature.