Enhancing employment conditions for migrant seasonal workers—Assessing consumer preferences for ‘fair-labour’ apples

Abstract

Criticisms over labour practices in agricultural production often target the living and working conditions of migrant seasonal workers. This article assesses consumer preferences for apples produced under social conditions exceeding current legal standards. The analysis is based on a discrete choice experiment (DCE) with a sample of 204 German consumers, who were asked to choose among domestic ‘fair-labour’ apples and standard apples. The former differed in six attributes describing the social conditions facing migrant seasonal workers on German farms. A mixed logit and latent class model were used to analyse the data. The predicted probability of choosing a ‘fair-labour’ apple was 85 per cent. Consumers valued higher minimum wages, the inclusion of migrant workers in Germany's social security system, and bonus payments for work on Sundays and public holidays. Improved accommodation and limits on the maximum permissible weekly working hours were considered less important. The low-price elasticity suggests that farmers could recoup a large share of the extra costs involved in providing enhanced living and working conditions for their workers.

1. Introduction

The EU's agreement on the Free Movement of Persons allows EU citizens to move from one Member State to another to take up work. Free mobility can bring multiple benefits. It facilitates allocation of workforce to countries with labour shortages and provides income opportunities to workers from countries in economic crisis (Barslund and Busse 2016). For this reason, almost a million migrant seasonal workers, mainly from Poland and Romania, travel to northern and central Europe each year. Roughly 300,000 of them work in Germany's agricultural sector (BMEL 2020; Augère-Granier 2021). Although migrant seasonal workers are indispensable for a successful harvest, the media often negatively reports on the prevailing working conditions. Recent reports, for instance, focused on cases where migrant seasonal workers received wages below the legal minimum wage or where accommodations were in uninhabitable conditions and overpriced (e.g. Seiler 2023; Eckinger 2024). Other articles pointed out that the physically demanding harvesting work is unattractive even when current minimum legal standards are met. One reason for this criticism is the lack of comprehensive social security for migrant harvest workers in Germany (Faire Mobilität 2022).

Media coverage and research on migrant seasonal workers received a boost during the COVID-19 pandemic. Large outbreaks on farms (Charlton 2021) and in abattoirs (Pokora et al. 2021) raised public awareness of the topic. Recent research articles thus focused on infection processes (Charlton 2021; Pokora et al. 2021), others analysed the regulations in force during the pandemic (e.g. Neef 2020; Bogoeski 2022; Szelewa and Polakowski 2022), and a further set of articles described migrant seasonal workers’ harsh living and working conditions in detail before and during the pandemic. They found that these workers are not only more vulnerable to infections but also likely to be exploited (e.g. Birkenstein 2015; Fialkowska and Matuszczyk 2021; Bogoeski 2022; Collins et al. 2022; Piracci et al. 2022; Oxfam 2023). Furthermore, some recent research from Europe has started focusing on social aspects of sustainability labelling and whether consumers value such labels (e.g. Piracci et al. 2022; Gallais et al. 2023). Besides legal changes, these sustainability labels might help achieve improvements. However, one shortcoming these articles have in common is that the labels define social aspects rather loosely, lacking clarity of the specific requirements involved (Gallais et al. 2023).¹ The same criticism applies to articles on consumer preferences for groceries produced abroad under enhanced social conditions (so-called fair-trade products).² Two articles analysed the potential of domestic ‘fair-labour’ labels to improve working conditions for migrant seasonal workers. Howard and Allen (2008) explored the potential of domestic fair-labour labels for the US, while Drichoutis et al. (2017) elicited Greek consumer preferences for fair-labour food products.³ Both articles found a positive willingness to pay of their respondents for a ‘fairer’ product, suggesting a potential for these labels in the future. These results are also in line with the fair trade and sustainability label literature that also describes these labels as being popular (Andorfer and Liebe 2012; Bürgin and Wilken 2021; Piracci et al. 2022; Gallais et al. 2023).⁴ However, since all these studies do not consider the working conditions in detail, they cannot provide information, which particular aspects would need to be improved.

The present article addresses the question of how fairer living and working conditions might look like from the perspective of German consumers. The analysis is based on a discrete choice experiment (DCE) with a sample of 204 German consumers. Respondents had the choice between apples harvested by migrant seasonal workers under different living and working conditions. Our study extends previous work by explicitly including detailed legal aspects (such as minimum wage rates, maximum permissible weekly working hours, participation in the social security system, accommodation requirements, and payment for work on Sundays and public holidays) as attributes. The respondents were asked to choose among two apples produced under higher-than-required standards and an apple produced under current minimum legal requirements. We chose apples as the object of consideration for two reasons: First, because (nearly) all consumers are familiar with the product and, second, because migrant seasonal workers play a major role in the supply chain for apples.

Besides filling a gap in the literature, our work may also have some practical use. The European Commission (EC) strives for a more ambitious legal framework to ensure better protection of migrant workers (European Commission 2020; Augère-Granier 2021). The EC states that fair-labour conditions are characterised by minimum wages that allow for a decent living, that health and safety are protected, for instance through social security standards, and that a dialogue between stakeholders exists (European Commission 2024). Given the gaps in current knowledge, it is difficult to derive from these demands concrete action plans for migrant seasonal worker policies. The great depth of detail of our study allows us to identify starting points for improving the social situation of seasonal workers.

The remainder of the article is structured as follows: Section 2 outlines the relevant legislative framework in Germany. Section 3 describes the methodology (both the DCE and the econometric estimation model) and spells out some key hypotheses. Section 4 presents the results that are critically discussed in Section 5. Section 6 concludes.

2. Legislative framework

Seasonal workers are entitled to equal treatment with nationals of the countries in which they are working (Article 45 TFEU, Directive 2014/54/EU) (Augère-Granier 2021). The legal requirements in Germany that were included as attributes in the DCE are described below.⁵

In 2015, Germany introduced a minimum wage of €8.50 per h. Exceptions applied to agriculture in the first few years. The minimum wage in agriculture was set at €7.20 per h and increased incrementally before it became the same for all sectors. At the time of the survey in June 2022, the minimum wage was €9.83 per h, and in October 2022, the level was raised to €12.00 per h in all sectors (Mindestlohngesetz) (Spiess 2021, 2022).⁶

The minimum wage is usually a gross minimum wage, and employees contribute a share of their wages to the social security system. The social security system includes health care, unemployment insurance, long-term care, and a pension scheme. In total, workers contribute 20.5 per cent of their gross wage to the social security system. The same amount is contributed by the employer.⁷ In addition to the social security contributions, an income tax is levied that is based on individual characteristics, such as the level of earnings, marital status, and number of children. The marginal income tax rate ranges from 14 per cent to 42 per cent (Einkommensteuergesetz) (Spiess 2021, 2022). For seasonal work, the income tax is a flat rate of 5 per cent (§40 Einkommensteuergesetz).

Exceptions from the social security contributions apply for migrant seasonal workers. If they are employed on a short-term basis, there is no obligation to make contributions to the social security system. This is the case when migrant seasonal workers are employed for fewer than 3 months or no more than 70 days per calendar year (Spiess 2021, 2022). This non-contributory period was longer during the pandemic (102 days in 2021 and 115 days in 2020) (Deutsche Rentenversicherung 2023).⁸

In addition, the exemption from contributing to the social security system only applies, if the seasonal work is not carried out professionally. This means that only migrant seasonal workers, who have a job subject to social security contributions in their home country and work for fewer than 70 days do not have to pay into the social security system. Consequently, social security is compulsory for the unemployed, but not for migrant seasonal workers who are housewives, students, or employed in their home country but working in Germany during their holidays (Spiess 2021, 2022). The alternative main occupation is proven by what is known as A1 confirmation. If migrant seasonal workers have to pay into the social security system, for instance, because they work for more than 70 days, A1 confirmation is also used to transfer social security payments to the respective entities in their home country (Deutsche Rentenversicherung 2023).

As a consequence of the pandemic, farmers now have to prove that all migrant seasonal workers, regardless of their social security status, have health insurance when they work in Germany (BMEL 2023). Health insurance coverage in Germany is usually offered by private providers (personal communication).

The maximum number of working hours is regulated by the German Working Hours Act (Arbeitszeitgesetz), which also applies to agriculture. It states that an average weekly working time of 48 h per year must not be exceeded. Consequently, workers can usually work six days per week for 8 h per day. However, this period can be extended to 10 h and, during the harvesting seasons, to 12 h per day. Farmers need to apply for permission to extend working hours to prevent crop failure. If granted, up to 60 h per week are permissible. Permanently employed workers are entitled to additional rest periods to ensure that an average of 48 h per week is not exceeded over the whole year. For short-term, temporary contracts, the extra hours worked do not need to be offset during the term of the contract (§ 3Section 2Arbeitszeitgesetz) (Spiess 2021, 2022).

In general, work on Sundays and public holidays is forbidden in Germany (Article 139 Grundgesetz), and therefore the maximum of 48 h is not usually exceeded. Exceptions are also made for migrant seasonal workers, so that work on these days can be approved by the authorities. In the case of employees working on Sundays and public holidays, there is no legal entitlement to higher pay. As migrant seasonal workers are employed for a fixed term, a written employment contract covering the described aspects is mandatory (§ 14 Abs. 3 Teilzeit- und Befristungsgesetz) (Spiess 2021, 2022).

Farmers usually also provide accommodation, and migrant seasonal workers often live on farms in shared rooms (Oxfam 2023). When sharing rooms with up to six people, each person must have at least 6 m² of space. When more than six people share a bedroom, at least 6.75 m² must be provided per person. During the pandemic, rooms could not be shared by more than four people, regardless of the size of the room (Spiess 2021, 2022). Conclusion of rental contracts between farmers as landlords and migrant seasonal workers as tenants is a legal requirement. These contracts must be concluded independently of the labour contracts. The rent must be aligned with prices for comparable accommodation in the region (German Employment Agency 2021).

During the COVID pandemic, accommodation provided needed to be large enough to allow social distancing and reduced contact (Spiess 2021, 2022). Farms had to ensure that hygiene rules were being followed, they further had to offer disinfectants, masks and regular testing (SARS-CoV-2-Arbeitsschutzverordnung). Furthermore, migrant seasonal workers were obliged to be in a so-called working quarantine, meaning that for the first 10 days after their arrival, they could only travel between their accommodation and their workplace (Einreise-Quarantäneverordnung).

The quarantine requirement and the described requirements for accommodation only applied during the pandemic, along with the exception that migrant seasonal workers with an alternative main occupation could be employed for 115 (or 102) days. Overall, the introduction of compulsory health insurance since 2022 seems to be the only legislative change still in force after the end of the pandemic.

3. Methodology

3.1 Research hypotheses—factors affecting the decision to purchase ‘fairer’ apples

Our case study elicited consumer preferences for apples produced under fairer social conditions for migrant seasonal workers in Germany. As indicated in the introduction, apples were chosen for multiple reasons. First, they are the most popular fruit in Germany, with 79 per cent of consumers buying apples (Janson 2017). Second, apples in supermarkets show low price volatility and almost no seasonal pattern. Year round, a kilogram of apples usually sells for €1.99 (Jalsovec 2012). Third, considering the producer side, apples are the most important branch of fruit production: In 2022, they were grown on 33,000 hectares in Germany. This accounts for 67 per cent of German tree fruit cultivation (German Federal Statistical Office 2022). Finally, apples are labour intensive and their harvesting usually hinges upon labour input from migrant workers.

There are only limited options for replacing labour by capital (machines), e.g. bulk harvesting, harvesting robots, and platforms. Bulk harvesting means that mechanical force is applied to harvest multiple apples (e.g. by shaking the tree). Although the technique is effective, it causes extensive bruising, and the harvested apples can only be used for processing. Harvesting robots have not reached market maturity, the selection process of sufficiently ripe fruit is not yet satisfactory, and bruising remains a problem. Harvesting platforms are partially utilised and can speed up the harvesting process, among other things because no ladders need to be moved. However, they are expensive and can only relieve migrant seasonal workers but not replace them (Hu et al. 2022). It is therefore unsurprising that approximately 50 per cent of an apple farm's turnover is used to pay wages (Luer 2022).

The attributes for the DCE were chosen in accordance with the current German legislative framework. These were discussed with farmers to ensure that they address the aspects relevant at farm level. The levels of the attributes in the choice sets represent a tightening of current legislation with the aim of improving living and working conditions. Tightened social standards will lead to increased costs of production and thus higher prices. For this reason, possible increases in apple prices were also considered.

In the choice sets, the prices of apples in supermarkets produced under improved conditions were increased by between 0 and 100 per cent of the price of standard apples. The upper limit accounts for the fact that fair-trade products are sometimes twice as expensive as standard products. Consumers are expected to react sensitively to price increases. Howard and Allen (2008) showed that consumers were willing to accept slight price increases of 3 per cent, with 87 per cent being prepared to buy ‘fairer’ strawberries, but for price increases of 100 per cent, this share fell to 35 per cent.

The second attribute was an increase in the minimum hourly wage rate. The current minimum wage is €12 per h, and was roughly €10 per h in 2022. In the choice sets, the wage rate was raised by between €2 and €6 per h. The highest wage rate was still below the average German wage level as seasonal work in agriculture does not require specific skills or work experience. In 2021, the average wage rate in Germany was €19 per h (Statista 2021). It is assumed that consumers would be in favour of higher wages, as discussed, for instance, in Howard and Allen's (2008) article.

As described, exemptions from the requirement to contribute to the social security system are usually granted to migrant seasonal workers. However, labour unions such as ‘Faire Mobilität’ argue that seasonal work is often the main source of income for migrant seasonal workers, even though they may have other employment in their home countries. They demand that the contribution requirement be extended to all migrant seasonal workers to ensure they have better security in old age (Faire Mobilität 2022). For this reason, participation in the social security system was included as the third attribute.⁹ We expect that this would have a significantly positive effect on consumer choices in favour of fair-labour apples.

The maximum permissible working hours per week were reduced in some choice sets. The levels ranged from 40 h to the current maximum of 60 h. Additionally, some choice sets included bonus payments for work on public holidays and Sundays. These changes were included to account for the fact that seasonal work is associated with long working hours almost every day of the week (Augère-Granier 2021, Collins et al. 2022). Lack of time to recover from stress and fatigue can increase susceptibility to infection. In addition, recovery time is described as a necessary factor for human wellbeing in general (Caldwell 2005). We thus expect fairer payment conditions to increase our respondents’ willingness to choose fair-labour apples.

The experiment sought improvements in accommodation by restricting the maximum number of people per room (single and bedrooms to be shared by two persons). This attribute refers to the criticism over crowded living and working environments that could boost infections by making it more difficult to comply with minimum distancing. This was identified as one of the drivers of COVID-19 infections (Pokora et al. 2021). In addition, it is more comfortable to sleep and recover in rooms with fewer people. Restrictions on the number of people to share a room rather than varying the square meters per person were not included because of COVID-19 only. Restricting the number of people per room was also described to be more influential by apple producers. They primarily use existing buildings or rented residential containers to house their migrant seasonal workers and there is little choice in terms of the size of the rooms. Limiting the number of people per room automatically results in an increase of available space (m²) per person.¹⁰

For these last three attributes (limit on maximum working hours, bonus payments and improved accommodation), positive effects on the predicted probability of choosing a fair-labour apple are expected. The selected attributes also seem appropriate from the perspective of NGOs and other authors as they address aspects that are commonly highlighted as being important (e.g. Neef 2020; Bogoeski 2022; Oxfam 2023).

3.2 Experimental design

Table 1 displays the attributes of the choice experiment and their levels. The design was created using dcreate in Stata (Hole 2016). Designs for DCEs are statistically determined, as typically the number of possible combinations arising from the attributes and their levels would generate so many choice sets that they cannot be presented to respondents. The design process ensures that enough combinations are selected to enable model identification while minimizing collinearity among attributes and attribute levels. To control for model identification, one needs to specify how many parameters will be estimated. The number of parameters depends on whether linear or categorical effects are assumed for the attributes (Reed Johnson et al. 2013).

Dcreate uses the utility function underlying the conditional logit model. All attributes and their levels were treated as categorical variables to be able to account for non-linear effects. The dcreate software first generates a matrix holding all possible combinations of the attributes and their levels (full factorial design). Next, a starting design is drawn. The programme systematically changes levels within the starting design with the help of a Fedorov search algorithm. This continues until D-efficiency does not increase any further (Hole 2016).

The final design had a D-efficiency of 93 per cent and included eighteen choice cards. According to the literature, designs with a D-efficiency above 90 per cent are sufficiently good (Auspurg and Hinz 2015). The eighteen choice sets were distributed across three blocks of questionnaires, and all respondents were asked to answer six choice sets. When deciding the number of choice sets per questionnaire block statistical and response efficiency must be considered. Statistical efficiency increases when a larger number of choice sets is evaluated. However, response efficiency can decrease due to fatigue when too many choice sets are presented (Reed Johnson et al. 2013). Since the survey included an extensive description of the attributes and was already relatively long, it was decided to include only six choice sets in each questionnaire block.

For each choice set, two hypothetical ‘fairer’ apples and a status quo apple were presented. None of the ‘fairer’ apples displayed the same levels as the status quo apple. The latter was an apple produced under current legal minimum standards and offered on all choice cards. Therefore, the participants were not forced to choose a ‘fairer’ apple. Instead, they could always also opt for the status quo apple. Participants were first asked whether they buy apples at all. Only participants stating that they do could continue the survey. Figure 1 displays one of the choice cards used in the survey.

The survey also elicited the respondents’ personal characteristics in terms of income, age, education, purchasing behaviour, and perceived product quality. These questions were included because research articles, for instance, Howard and Allen (2008), Mahè (2010), Drichoutis et al. (2017), and Piracci et al. (2022), show that these characteristics influence consumers’ decisions to purchase fair-trade products or products labelled as more sustainable.

The survey was distributed online by a market research company in June 2022. In the survey invitation, the participants were informed that ‘fairer’ apples would be labelled as such, and that the attributes shown provide information as to the specific aspects of enhanced social conditions for migrant seasonal workers in apple production. The term ‘fair’ was used, following the approach of Howard and Allen (2008) and Drichoutis et al. (2017). It seemed necessary to describe to the respondents that ‘fairer’ apples are labelled because working conditions cannot be determined by simply looking at the product. Fair-labour products are known as credence goods and require external information, such as expert opinions, advertising, or labelling (e.g. Grolleau and BenAbid 2001; Gallais et al. 2023).

The survey also included so-called catch questions and intentionally a dominant choice set to screen out participants who just clicked through the questions without reading them. In addition, participants who only spent a short time on the platform (< 3.5 minutes) were excluded from the dataset. In total, 46 participants who completed the whole questionnaire were excluded prior to the estimations, leading to 204 remaining observations.¹¹ The average time taken to answer the survey was 10 minutes. Quotas and a ‘cheap talk’ script in the invitation were used in the survey to further improve the quality of the sample and the data.

3.3 Econometric estimation

The description of the mixed logit model is based on Train (2009) and Hole (2007). It is assumed that a respondent (n) chooses the apple that is valued the most from a set of J alternatives on each of the T hypothetical choice occasions. The utility of buying an apple (U_njt) depends on |${{x}_{njt}}$|⁠, which is a vector of the attributes, and on an independently and identically distributed random term (⁠|${{\varepsilon }_{njt}}$|⁠). The vector |${{\beta }_n}$| represents individual-specific coefficients (Equation 1):

The original conditional logit model assumes a homogenous distribution of preferences, implying that the estimators are the same for all consumers. In this setting, the predicted probability of the logit (⁠|${{L}_{njt}}$|⁠) of choosing a particular product (apple) is determined as follows (Equation 2):

The estimated mixed logit model assumes that the estimators are continuously distributed with density |$f( {\beta {\rm{|}}\theta } )$|⁠, with θ being the parameters of the distribution. The probability of choosing a fair-labour apple becomes an integral over the values of β and is determined by simulation (Equation 3):

Even though the mixed logit accounts for preference heterogeneity using individual-specific coefficients, it does not provide much information on which attributes are perceived differently. The heterogeneity can be better gauged by using latent class models that allow for a discrete distribution of coefficients. In that model, individuals with the same preference structure are sorted into the same class of Q classes. The predicted probabilities (⁠|${{P}_{njt|q}}$|⁠) of each class depends on the class-specific coefficients and is determined using the formula of the logit model (Equation 4):

In addition to a class-specific predicted probability of choosing a fair-labour apple, the class-membership probability (⁠|${{H}_{nq}}$|⁠) can be calculated. It depends on the characteristics (⁠|${{z}_n}$|⁠) of the individuals n, their influence is determined by |$\gamma $|⁠. To enable identification of the model, the Q^th vector is normalised to 0 (Equation 5).

The predicted probability to observe an individual choosing a fair-labour apple depends on the class-specific predicted probability and the class-membership probability (Hensher et al. 2018). In latent class models, the appropriate number of classes is unknown. It can be determined using the Bayesian Information Criterion (BIC). The value allows for balancing between model complexity and explanatory power. In practice, BIC values are calculated for models with different numbers of classes, and the model with the smallest BIC value is preferred (Greene and Hensher 2003).

Willingness-to-pay (WTP) and marginal effects were computed for both the mixed logit and the latent class models. Marginal effects indicate how the predicted probability of choosing a fair-labour apple changes when the expression of an attribute changes by one unit. The WTP is the rate of substitution between a specific attribute and the monetary variable. A positive WTP reflects the additional amount of money a consumer is willing to pay for a positively evaluated attribute. Accordingly, a negative WTP estimate shows the amount of money consumers are willing to pay less when a negatively evaluated attribute is expressed. In models without individual-specific coefficients, the WTP for marginal changes in an attribute is calculated as the ratio of the coefficient on that attribute divided by the coefficient for price. In models with individual-specific coefficients, like the mixed logit model, this approach is not recommended. WTP calculated directly from two randomly distributed coefficients would lead to unreasonably high values (Train and Weeks 2005).

4. Results

4.1 Descriptive statistics of the sample

Table 2 presents the descriptive statistics of the consumer sample. The average respondent in the sample was 52 years old, slightly older than the German average of 45 years (Statista 2020). The largest share of participants had a household income of up to €3,000/month, while the German average is €3,600 (after income tax) (Statista 2022). In addition, 31 per cent of the participants held a university degree, which roughly fits the German average of 35 per cent (BPB 2019).

Other questions in the survey addressed consumer buying behaviour. Most of the respondents buy their groceries in supermarkets (72 per cent) and discounters (54 per cent). Their purchasing decisions are mostly based on the perceived quality of the product (69 per cent), personal taste (55 per cent), and product price (46 per cent). Nearly all respondents (91 per cent) stated that they are familiar with the fair-trade label used for coffee and chocolate. However, only 50 per cent said that they buy fair-trade products. This is less than the estimate by Fairtrade Germany; they state that 66 per cent of the population buy fair-trade products (Fairtrade Germany 2018).

4.2 Factors affecting consumers’ decision to purchase fair-labour apples

Table 3 shows the results of the mixed logit model to explain the respondents’ decisions to buy an apple produced under enhanced living and working conditions. The model was chosen using the likelihood ratio test. Where the likelihood ratio test was undefined, a robust Wald test was used. Some of the socio-economic characteristics had no explanatory power and were excluded.¹² For the attributes of product price, working hours and minimum wage paid per hour, the linearity assumption was checked using dummy coding (Mariel et al. 2021).

The pseudo-R² of the chosen model is 0.333. According to Hensher et al. (2018), this represents a sufficiently good model fit. The predicted probability of choosing one of the ‘fairer’ apples was 85 per cent. All of the estimates showed the expected signs. The predicted probability increased for products produced with migrant seasonal workers, who receive higher wages and bonus payments for work on Sundays and public holidays, and with the obligation to participate in the social security system. As expected, the predicted probability of choosing one of the ‘fairer’ apples fell as their price increased. In addition, longer working hours per week were evaluated negatively. The attributes referring to accommodation on farms seem to be less important. We observed a positive effect for the obligation to offer a bedroom to be shared by two-persons (Accommodation (two-person room)), but this effect was only weakly significant.¹³

The alternative-specific constant (ASC) assumed a value of one for the alternatives that represented a fair-labour apple, while it took a zero for the standard apple.¹⁴ According to Train (2009), a significant ASC indicates that the positive perception cannot be explained by the attributes alone. It may thus represent the respondents’ feeling of doing something good. The interaction effects for the consumer characteristics further revealed that female consumers are more likely to choose a fair-labour apple. Consumers who were more concerned about production conditions were more likely to choose a ‘fairer’ apple. This is shown by the significantly positive estimates for the dummies ‘Buys based on regionality’ and ‘Buys based on production (organic)’. These dummies took a value of one for respondents who stated that those aspects were among the most important ones for them when buying groceries. In addition, consumers who have been purchasing products with the fair-trade label were more likely to choose one of the apples produced under improved social welfare conditions. Furthermore, higher education levels and more household members decreased the probability of opting for a ‘fairer’ apple. This counterintuitive result may be explained by the fact that the sample's average household income lies below the German average. Especially those with a lower educational attainment are likely to have lower incomes. They may have assumed that ‘fairer’ working conditions and particularly higher wages would also affect their line of work. This could explain why less highly educated participants opt for ‘fairer’ apples.

The significant standard deviations of the estimated coefficients in Table 3 point to preference heterogeneity within the sample. We thus estimated a latent class model to reveal that heterogeneity (Table 4). Based on the BIC values, a model with two preference classes was chosen.¹⁵ The pseudo-R² for the latent class estimation was 0.313. The predicted probability of a respondent choosing a fair-labour apple was 74 per cent. The class-specific predicted probabilities showed that the first preference class chose a ‘fairer’ apple with nearly full certainty (98 per cent), while the second preference class was less inclined to purchase such apples (45 per cent). The first preference class accounted for 75 per cent of the sample and could be dubbed ‘willing buyers’. The second class, the ‘reluctant buyers’, accounted for 25 per cent of the respondents.

The estimated coefficients for the attributes indicated that members belonging to the second preference class were price orientated. Their choice is only determined by price. None of the other attributes had a statistically significant effect on these respondents’ choices. By contrast, the first preference class valued the attributes that represent improved working conditions. As in the mixed logit model, increases in minimum wages, bonus payments, and the obligation to participate in the social security system were all evaluated positively, whereas higher product prices and longer working hours led to a decrease in the probability of choosing a ‘fairer’ apple. Again, the accommodation-related attributes seemed less important to the respondents in the first preference class.

Overall, the class membership variables (shown in the bottom panel of Table 4) mostly confirmed the findings from the mixed logit model: Respondents with higher levels of educational attainment were less interested in ‘fairer’ apples and thus tended to belong to the second preference class. Furthermore, the results confirmed that those who buy products with a fair-trade label and those who buy regional or organic food were more strongly associated with the first preference class, and thus more likely to choose a ‘fairer’ apple. An interesting result is that those who buy from organic supermarkets would rather not choose an apple produced under enhanced social welfare conditions.

4.3 Marginal effects and willingness-to-pay

In linear regressions, the model coefficients allow for direct inferences about the strength of the effects. This is not the case in logistic regressions. Therefore, marginal effects and WTP values were calculated. These are shown in Table 5 for the mixed logit model and in Table 6 for the latent class model. They revealed that the price of the product was the most important attribute. An increase in price of one Euro reduced the predicted probability of choosing a fair-labour apple by 8 percentage points (pp). An increase in working hours by one hour per week reduced the predicted probability by just 0.1 pp. Thus, an increase in the maximum number of working hours from 40 to 50 hours per week would reduce the predicted probability by just 1 pp. This finding suggests that consumers may tolerate slightly longer working hours. An obligation to participate in the social security system was important to our respondents, raising the predicted probability by 5 pp. An increase in the minimum wage was also of great importance: An increase of €1 raised the predicted probability of choosing a ‘fairer’ apple by 1 pp. Consumers valued a bonus payment for work on Sundays and holidays of €25 per day as much as a bonus of €50 per day. Thus, €25 per day seems to have been perceived as sufficient. The small marginal effect for the bedroom to be shared by two persons (Accommodation (two-person room)) indicates that respondents were less concerned about accommodation for migrant workers. A price change of one Euro (e.g. from €1.99 to €2.99 per kg = 50 per cent) results in a reduction in the quantity demanded by only 8 per cent. This means that the demand for ‘fairer’ apples is price inelastic.

The WTP estimates in Table 5 suggest that our respondents would be willing to pay an additional 74 cents per kg for the obligation to enrol migrant seasonal workers in the social security system. High WTP estimates were also found for bonus payments for work on Sundays and public holidays. By contrast, our respondents were less concerned about a potential increase in weekly working hours. In order to be able to better assess the absolute willingness-to-accept (WTA) values, these are also shown as relative values in Tables 5 and 6. For instance, our respondents would be willing to pay 6 per cent on top of the price of a standard apple when the wage of migrant workers is raised by €1. The WTP estimates displayed in Table 6 are higher than those in Table 5. They reflect the WTP of the more ‘willing consumers’ (class 1) rather than the averages for the entire sample.

5. Discussion

This article elicited consumer preferences for apples produced under enhanced living and working conditions for migrant seasonal workers. The mixed logit analysis revealed a high predicted probability of purchasing fair-labour apples (85 per cent). The latent class model divided the sample into two classes with different preferences: the ‘willing buyers’ who would nearly always purchase fair-labour apples and the ‘reluctant buyers’ who focused on price and would not value any of the social welfare attributes. These results are largely in line with the literature. Howard and Allen (2008) also found a high-predicted probability of US consumers buying domestic ‘fairer’ strawberries and associated positive WTP estimates. Drichoutis et al. (2017) confirmed this finding for Greek consumers. Furthermore, numerous articles have established a positive appreciation for imported fair-trade products. Comprehensive review articles on fair trade were provided by Andorfer and Liebe (2012) and Bürgin and Wilken (2021). Mahè (2010) analysed consumer preferences for fair-trade bananas in Switzerland and highlighted that fair trade was particularly successful in Europe, which may also explain the high-predicted probability of buying fair-labour apples in the present study. These preferences may also hold for sustainability labels that include social aspects, as shown, for instance, by Gallais et al. (2023) and Piracci et al. (2022).

5.1 Impact of living and working conditions

While the existing literature on consumer preferences for fair groceries and sustainability focuses on the effect of labels (e.g. Basu and Hicks 2008; Howard and Allen 2008; Mahè 2010; Rousseau 2015; Rombach et al. 2021; Gallais et al. 2023), the present article analysed living and working conditions of migrant seasonal workers in detail by including these as attributes in a choice experiment. Our article thus extends the literature by identifying the particular aspects of working and living conditions that are considered crucial and in need of improvement. Previous work tested labels that refer to production under fair-trade standards or similar expressions. They found positive WTP estimates for fair trade, fair labour, or sustainability labels that also include social aspects. However, these studies do not systematically vary these aspects and thus cannot derive specific conclusions as to what needs to be improved.

The in-depth analysis of migrant workers’ working and living conditions allowed minimum wages, weekly working hours, access to the social security insurance, improvements in accommodation, and bonus payments for work on Sundays and public holidays to be identified as influential factors. Overall, these results confirm the main hypotheses outlined in Section 3.1. However, some of the attributes were found to be less influential than expected. For example, the WTP estimates for the two levels of bonus payments for work on Sundays and public holidays (€25.00 and €50.00 per day) were quite similar. This finding is in line with Basu and Hicks (2008), who studied consumers’ WTP for fair-trade coffee in Germany and the US. They also found that WTP for increases in growers’ revenue fell beyond a particular threshold. This means that once a certain level is reached, WTP does not seem to increase further. Such a threshold also seems to exist in our study: The bonus payment of €25.00 per day seemed to be perceived as sufficiently high.

Surprisingly, an increase in the maximum permissible working hours per week only had small effects on choices and may thus be tolerated to a certain extent. The respondents may be aware that that working hours in agriculture are longer due to seasonal factors and may therefore not consider extended working hours to be unfair. Furthermore, accommodation seemed to be less important. It may be that the respondents assumed that higher wages would enable migrant seasonal workers to find alternative accommodation if the on-farm accommodation is unsatisfactory. Alternatively, clean and mould free accommodation (which is legally required) may have been perceived as sufficiently good, even though media coverage has pointed to gaps in the enforcement of the regulations (Seiler 2023; Eckinger 2024).

As described earlier, the COVID-19 pandemic has led to mandatory health insurance for the seasonal workforce (BMEL 2023). Our estimation results suggest that this change may not be perceived as sufficient. Access to all parts of the social security system was found to be a decisive factor for purchasing decisions. This finding supports the demands presented on the website of the labour union Faire Mobilität (Faire Mobilität 2022) and in its recent joint article with Oxfam on living and working conditions on German farms (Oxfam 2023). One reason for not allowing migrant seasonal workers to contribute to the social security system is that retirement and unemployment may indeed be a marginal issue in terms of occasional seasonal work. However, some countries have a long history of seasonal work and workers come to work in fields for long periods of their working lives and are thus not properly safeguarded in their home countries, where they have probably made limited social security contributions (Birkenstein 2015; Bogoeski 2022).

However, if the different wage levels across countries and possible effects of abroad payments on the mobility of workers are included in the discussion, it becomes clear that changes must be implemented with caution. Regular payments from foreign social systems can disrupt the flow of available labour in Europe and limit the benefits of free mobility for the economy. This may be the case when foreign support to the unemployed is significantly higher than average wages in the home countries. In this case, workers from abroad have little incentive to contribute their labour (Barslund and Busse 2016). The payment entitlements that short-term employment abroad should lead to must therefore be carefully weighed up against the potential drawbacks. Currently, at least 6 months of continuous employment are required for temporary employees to enter the German social security system, and migrant seasonal workers are not entitled to unemployment benefits. If this was changed to, say, a month, workers could expect unemployment benefits of around €900 per month when they worked 173 hours per month at the current minimum wage of €12 per hour (German Employment Agency 2024). This is above the average Romanian wage of €686 per month (Statista 2023). Whether changes to social security law should be aimed primarily at pension insurance or at all parts of the social security system (including unemployment insurance) must therefore be carefully considered.

The results of the latent class model raise the question of whether changes are necessary from everyone's perspective. On one hand, the estimates could be seen to imply that respondents in class 2 are price sensitive and therefore do not buy fair-labour apples. On the other hand, the findings could be interpreted as suggesting that 25 per cent of the respondents (those in class 2) consider the working and living conditions of migrant seasonal workers to be sufficiently good so that no further improvements are needed. When interpreting the overall results in the German context, it should also be noted that compliance with the current legislative framework (and possible changes) was assumed. Both Bogoeski (2022) and Bruzelius and Seeleib-Kaiser (2023) raise doubts of whether enforcement strategies work for multiple countries in the EU, including Germany, where the responsible authorities appear to be understaffed. Only if the legal requirements are adhered to will the above interpretations for class 2 respondents be appropriate.

Looking beyond Germany, whether our findings can be transferred to other European countries remain questionable. One reason is that only German consumers were surveyed, and the changes related to the German legal situation. Another is that Drichoutis et al. (2017) and Piracci et al. (2022) describe farm work in southern Europe as largely relying on workers classified as illegal immigrants. If working conditions were improved using labels or legislative adjustments, these workers are unlikely to be reached. Thus, more research from a southern European perspective would appear necessary.

5.2 Impact of consumer characteristics and willingness-to-pay estimates

Inclusion of the respondents’ characteristics in the estimations allowed us to identify potential target groups for ‘fairer’ fruits. From a policy perspective, they show which parts of society tend to act less socially responsibly or perceive the conditions as sufficiently good. As in Howard and Allen (2008), the present study showed that consumers who buy organic food are more likely to opt for domestic fair labour. Rousseau (2015) found that people who are members of nature protection groups are more interested in fair-trade chocolate, and Rombach et al. (2021) found that consumers familiar with fair trade are more likely to value hypothetical fair-trade flowers as well. This is in line with the present findings that those who buy organic, regional, or standard fair-trade products are more likely also to buy fair-labour apples.

Our findings also suggest that some consumers may not be easily reached. Just as in our study, Howard and Allen (2008) found a lower willingness to buy ‘fairer’ products among consumers with higher education levels. This goes against the conventional wisdom that highly educated people are more interested in (social aspects of) sustainability. Howard and Allen (2008) assumed that this could be explained by the greater social distance between highly paid, educated consumers, and migrant seasonal workers. An alternative explanation is that those with lower levels of educational attainment earn lower wages and may have opted for fair-labour apples hoping that improved conditions would also affect their incomes. In addition, Vermeir and Verbeke (2006), Howard and Allen (2008) and the present article suggest that men are less likely to buy ‘fairer’ products than women.

Product price estimates were significant in all estimations and all latent classes. This means that the probability of purchasing fair-labour apples decreases with price increases for all consumers. The same was found in the study of Howard and Allen (2008), and also Bürgin and Wilken (2021) conclude in their review of articles on consumer WTP for fair-trade coffee that consumers are only willing to pay ten pp more on average. This is not particularly much. Only two studies found high WTP estimates for fruit. Howard and Allen (2008) found that consumers would be willing to pay a premium of 68 per cent, while Drichoutis et al. (2017) estimated this figure at 70 per cent. These values are above the relative WTP values estimated in the present article for the single attributes but may be reached in combination.

Piracci et al. (2022) also found positive WTP estimates for fairer labour conditions but showed that these were less important than organic production. Since consumers appeared price sensitive, it would seem necessary to analyse whether consumers would be willing to pay sufficiently high premiums when fruit and vegetables are also organic and thus already more expensive. An indication that this might not be the case comes from the latent class estimation, where frequent organic supermarket buyers were less likely to buy fair-labour apples.

Concerning price sensitivity, it may also be necessary to consider other more expensive products in future studies. This is necessary because changes in legislation would affect prices of all domestically grown fruit, vegetables, and meat products. The choice experiments of Howard and Allen (2008) and Drichoutis et al. (2017) for fruit, and also the DCE underlying the present study consider voluntary one-off purchases for relatively cheap products. Results should not be extrapolated to more expensive products. The relatively small sample size is another limitation of the present study. If the experiment was repeated with a larger sample, it should be designed to capture the repercussions on other food product markets, ideally for the whole food basket. Furthermore, it needs to be considered that consumers were forced to choose an apple, because a standard apple instead of a non-option was included. Forced choices are reported to lead to higher WTP estimates (Veldwijk et al. 2014).

WTP estimates should generally be interpreted with caution. For instance, what is known as social desirability bias can lead consumers to overstate their true WTP. This has been shown, for instance, by Barber et al. (2016) for sustainably produced wine. Furthermore, it is well documented that stated preferences may not translate into actual purchasing behaviour. The so-called attitude-behaviour gap is described for fair-trade chocolate in Belgium (Vlaeminck et al. 2016) or sustainable food more broadly (e.g. Padel and Foster 2005; Vermeir and Verbeke 2006).

5.3 Research perspective

The present study has focused on consumer concerns about migrant workers’ living and working conditions. Future work should gauge the perspectives of the other two stakeholder groups involved: farmers and migrant workers themselves. Whereas the WTP estimates presented in the present study could be seen to reflect consumers’ demand for better social conditions for migrant workers in agriculture, a survey of farmers (in our case apple growers) could shed light on the potential supply of these improved social conditions. Ideally, such a survey would use the same elicitation method and the same attributes as in the present study. This approach would enable estimation of farmers’ WTA compensation (in the form of higher prices for their products) for providing their migrant workers with improved social welfare conditions. Comparing and contrasting WTP and WTA estimates would allow one to assess the market potential for fair-labour foods (apples). Latacz-Lohmann and Schreiner (2019) used this common-elicitation-format approach to gauge the preferences of both German consumers and pig farmers for enhanced animal welfare standards in pig production. By determining both WTP and WTA, they were able to simulate a market for pig meat produced under animal-friendly husbandry conditions. Besides assessing the market potential, a survey of farmers could also elicit their views on the potential drawbacks of opening the social welfare systems for migrant seasonal workers.

Future work could focus on what migrant workers want in terms of living and working conditions in Germany and what they perceive to be a fair remuneration system. It is not clear, for instance, whether they would want to be included in the German social security system because they would have to pay the employee's share of social security contributions. The remuneration system for seasonal workers is commonly based on piecework wages. At the end of the working day, the harvest is weighed, and the worker is paid accordingly. The daily earning is then topped up if there is a shortfall of the minimum wage on that day. This remuneration system requires extensive documentation of both harvest and working hours—a burdensome task for both the farmer and the workers. Again, it is not clear whether seasonal workers would want to abandon this system in favour of, say, a fixed hourly wage rate. On the one hand, it would release them of the stress that comes with piecework wages. On the other hand, highly productive workers may regard the current remuneration system as an opportunity to earn more than the average worker.

Finally, research could discuss how fair-labour labels could best be implemented. One option is to introduce mandatory labelling. The other is to allow voluntary labelling. Both types can help to address market failure by presenting product-specific information that consumers value. Which policy would be socially desirable depends on the costs of labelling and the benefits of improved purchasing decisions (Roe et al. 2014). As described above for minimum wage levels, gathering, verification, and monitoring processes might be complex when working conditions are considered. If these costs are higher than the expected added value from selling ‘fairer’ fruits and vegetables, labelling will not occur on a voluntary basis (Roe et al. 2014). Besides consumer preferences for ‘fairer’ fruits and vegetables, costs for labelling require further analysis to be able to determine the best implementation strategy.

6. Conclusions

This article analysed consumer preferences for apples produced under living and working conditions for migrant seasonal workers that exceed current legal standards. The high-predicted probabilities indicate that German consumers would be interested in apples produced under such enhanced conditions, pointing to a large market potential for fair-labour apples. Among the most important factors affecting consumer choices was the inclusion of migrant workers in the German social security system. Fair pay, especially for work on Sundays and public holidays, was also found to be an important driver of consumer choices, whereas improved accommodation and stricter limits on the maximum permissible weekly working hours were considered less important. Three quarters of the respondents were classified as willing buyers of fair-labour apples; their predicted choice probability for such apples was close to 100 per cent. One quarter of the respondents was more reluctant to buy such apples; their choices were shown to be price-sensitive, with little attention being given to the social conditions under which the apples would have been produced. The overall price elasticity across the sample was low: A 50 per cent increase in the price of fair-labour apples (e.g. from €1.99 to €2.99 per kg) would lower demand by only 8 per cent. This suggests that the extra costs of providing living and working conditions exceeding current legal requirements could be largely passed on to consumers. However, this conclusion comes with a word of caution due to challenges over external validity of the stated-preference method used in this study.

Acknowledgements

The research received funding from the EU project agroBRIDGES (Grant Agreement N° 101000788). The authors thank the anonymous reviewers for their comments.

Data Availability

The raw data, processed data, and the Stata code used for the estimation will be provided by the corresponding author upon request.

Footnotes

Appendix

Appendix material 1–Attribute explanation and socio-economic questions used in the survey

Mail—Survey invitation + cheap talk

Page one—Welcome site of the survey

Page two—Socio-economic questions for the quota/filter

How old are you?

______years

What is your gender?

• Male

• Female

• Other

Do you purchase apples from supermarkets?

• Yes

• No

Page 3—Attributes

Attribute explanation

In the experiment, you will be offered apples of the Elstar variety (commercial grade 1), produced under improved living and working conditions. You can also decide against a hypothetical product purchase and for an apple produced under the legal minimum standard. The products are distinguished by the following characteristics (current minimum levels set by law are in bold):

Characteristics

• (1) Product price: €1.99 (+0 per cent mark-up), €2.49 (+25 per cent mark-up), €2.99 (+50 per cent mark-up), €3.49 (+75 per cent mark-up), €3.99 (+100 per cent mark-up).

We assume that an improvement in working conditions leads to a different increase in costs and therefore to an increase in sale price (between 0 and 100 per cent).

• (2) Hourly wage (net): €10, €12, €14, and €16/h

The net wage paid to the migrant seasonal worker varies. With a target minimum hourly wage of €12/h gross (from autumn 2023), the employer pays a seasonal worker who is subject to social insurance about €10/h (net wage per hour). In the experiment the net wage is presented (all deductions are included) and raised above the current minimum levels.

• (3) Sunday & holiday bonus: not paid, €25 per working day, €50 per working day

When working on Sundays and holidays, in some alternatives the migrant seasonal workers receive an additional payment as a bonus, in others they do not.

• (4) Maximum weekly working hours: 40 h/week, 50 h/week, 60 h/week

On average, workers in Germany should not work more than 48 hours per week. However, this period can be exceeded temporarily to up to 60 hours per week. The maximum number of working hours varies over the levels above.

• (5) Social security insurance: not mandatory, mandatory

If migrant seasonal workers are not obliged to pay into the social security system, they receive the minimum wages without deductions for care, pension or unemployment insurance. When workers are obliged to pay into the system, it is assumed in the experiment that they immediately qualify for payments e.g. childcare.

Important note: Since 2022 migrant seasonal workers must have health insurance to cover illness in Germany.

• (6) Accommodation: 6 m²/single room, 12 m²/2-person room, shared bedroom (Min. 24m²/4-person)

Accommodation on farms is usually organised in shared rooms. In the experiment, single (German Einzelzimmer) and bedrooms to be shared by two-persons (German 2-Personenzimmer) are introduced for some alternatives.

Migrant seasonal workers pay rent for their rooms. The maximum amount is determined by legislation and depends on local prices. Under the current legislative framework, the obligation is to offer 6m² per person, and rooms can be shared by multiple people.

Pages 4–10 Choice sets and filter question

(1/7) Which of these apples would you purchase?

Which option would you select?

• Product 1

• Product 2

• Product 3

(Six choice sets, including an extra dominant set—this was not included in the analysis).

Filter question

Please select option B to continue (in between the sets):

• E

• B

• A

• C

• D

Page 11—Socio-economic questions

What is your highest academic qualification?

• Secondary school certificate (intermediary level)

• Secondary school certificate (higher level—Abitur)

• University studies (without a degree)

• University studies (Bachelor level)

• University studies (Master level)

• University studies (PhD)

How high is your household income after taxes?

• Up to €1,000 per month

• Up to €2,000 per month

• Up to €3,000 per month

• Up to €4,000 per month

• More than €4,000 per month

Which district do you live in (please indicate your postcode)

___________postcode

How would you classify the location of your home?

• Rural

• Fairly rural

• Fairly urban

• Urban

Please select the location where you purchase groceries (multiple options possible):

• Supermarkets

• Discounters

• Farmers’ markets

• Organic supermarkets (e.g. Reformhaus)

• Others ____________

How often do you purchase groceries per week?

• Once per week

• 2–3 times per week

• More frequently than that

Please select D to continue:

• B

• C

• D

• E

• F

How do you evaluate the price increases in the last few months (on a scale from 1 to 5)?

• Likert scale: 1 (I am still okay with it)–5 (worries me)

Based on which aspects do you buy groceries? Please order from the most important to the least important aspect:

• Price

• Taste

• Season

• Quality

• Appearance (colour and smell)

• Regionality

• Production standard (organic)

• Fairer production standards

• Nutrition

• Preparation time

How price sensitive do you consider yourself to be?

• Likert scale 1(less price sensitive)–5 (highly price sensitive)

Did you follow the media on COVID-19 outbreaks on farms and in slaughterhouses?

• Yes

• No

Do you know the fair-trade label on coffee and chocolate?

• Yes

• No

Do you purchase fair trade products?

• Yes

• No

If no, what is the main reason why you don't?

• I don't know enough about it

• I purchase other brands

• The variety is too limited

• The price is too high

• I think the product quality is poor

Would you like to leave any comments?

References