Revenue efficiency, profitability, and profitability potential on organic versus conventional dairy farms—results from comparable groups of farms

Abstract

The purpose of this paper is to explore whether differences in profitability and revenue efficiency exist between Norwegian organic and conventional dairy farms. With access to accountancy data from more than 1000 conventional farms, it was possible to compare the 59 organic farms with a matched group of 177 conventional farms. The two groups did not differ significantly with respect to share of turnover from milk, forage area, number of cows, milk quota, location, and share of robotic milking. Data spanned over the fiscal years 2014 to 2016. Stochastic frontier analysis was used to calculate revenue efficiency. The results confirmed that organic farms and conventional farms use different production technology, and therefore, we calculated efficiency on two different production frontiers. On average, organic dairy farms were 6.5 percentage points more revenue efficient than conventional farms. Although conventional farms appeared more profitable than organic farms, differences were not significant; however, conventional farms had a higher profitability potential than organic farms.

Appendices

Appendix 1. Matching variables

Table 7 Differences in matching variables between organic farms, the frequency-weighted control group of conventional farms, and the unweighted control group of conventional farms

In Table 7, we could see that none of the matching variables differed statistically significantly between the three groups. However, weighting yielded somewhat larger forage area, milk quotas, more cows, and somewhat more farms that use AMS and farms in zone A + B, compared to the unweighted control group. Taken together, compared to the unweighted group, the farms in the weighted control group were more similar to the organic farms. Therefore, we conclude that frequency weighting was used to create a group of conventional farms which was more similar to the organic farms than an unweighted group.

Appendix 2. PSM

In our setting, Propensity Score Matching (PSM) (Rosenbaum and Rubin 1983) gives the conditional probability that a farm was run organic, given the set of observed variables (Austin 2010). That is, how likely it is that a farm in our dataset was run organic, given the selected matching variables we controlled for. Following Rosenbaum and Rubin (1983), we performed a probability regression based on the selected matching variables. According to ibid., the propensity score p(X) can be written as follows:

$$ p(X)\equiv \mathrm{Prob}\left(\mathrm{D}=1|X\right)=E\left(D|X\right) $$

(5)

where D = {0, 1} is a binary indicator variable which tells whether a farm is run organic (1) or not (0). X represent the independent variables district, farmland, quota, dairy cows, turnover from milk, joint operation, and AMS, which represent the propensity score for each farm. Thus, E(D| X) is the expected outcome of D given X.

In Table 8, we could see that the variable district zone A + B, forage area, number of dairy cows, and share of turnover from milk have a significant positive sign. This means that farms in district zone A + B had a positive probability of being run organic compared to farms in zone E + F + G. This was as expected, as zone A + B has the best production conditions. Furthermore, farms with a large forage area, a high number of dairy cows, and a high share of turnover from milk increased the probability that the farm was run organic. The finding that a large forage area increased the probability of the farm being run organic is in line with the finding of Mayen et al. (2010). Similarly, the finding that a high share of turnover from milk had a similar effect was also observed by Flubacher (2015). Finally, the negative effect of milk quota was in line with the finding that the organic farms in the dataset had a lower median milk quota compared to the conventional farms.

Table 8 Results from the PSM of all 913 farms with probability estimates for whether the farms were run organic

Appendix 3. Testing for technology differences

Following the recommendations of Madau (2007) and Flubacher (2015), we chose a log-likelihood ratio test. To perform the test, we constructed two stochastic frontiers from the matched dataset, where one model included a dummy for organic farming, D_org, in the production function. We tested goodness of fit for the two models by the equation:

$$ \uplambda =-2\left(\ln \frac{L\left({H}_0\right)}{L\left({H}_A\right)}\ \right) $$

(6)

where L(H₀) was the log likelihood value of the model with null hypothesis D_org = 0, and L(H_A) was the log likelihood value of the alternative hypothesis. The parameter λ is chi square distributed, with degrees of freedom equal to the difference between the parameters in the two models. If the null hypothesis was to be rejected, we had to construct two fronts, one for organic farms and one for conventional farms. To perform the test, we applied a likelihood ratio test in STATA (Stata 2020).

The chi square test of homogeneous production technology rejected the null hypothesis at the 5% level (λ = 52.8, χ²= 3.84). Therefore, we needed to construct two different production frontiers.