Data description.

Data for this study were from a 2019 web-based survey of specialty crops growers who were part of e-mail lists of growers’ associations and the Food Industry MarketMaker database. We compiled 3557 e-mail addresses of growers located in 32 states (i.e., Alabama, Arizona, Arkansas, California, Colorado, Florida, Georgia, Idaho, Illinois, Indiana, Iowa, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Minnesota, Mississippi, Montana, New Mexico, New York, North Carolina, North Dakota, Oklahoma, Oregon, Rhode Island, Tennessee, Virginia, West Virginia, Wisconsin, and Wyoming). The list of growers was screened to eliminate duplicate entries and operations. The e-mail databases facilitated access to a wide range of producers growing fruits, vegetables, and herbs (i.e., specialty crops).

Similar to Velandia et al. (2020), we evaluated sample representativeness by comparing the distribution of our survey sample based on acres of specialty crops production to the distribution of specialty crops growers based on the 2017 Census of Agriculture Data. We considered operations from the Census data that reported growing vegetables, fruits, nuts, berries, and other edible specialty crops. Although the survey sample tended to slightly underrepresent specialty crops operations between 1.0 and 99.9 acres and overrepresent farms with 100 acres or more, the farm size distribution from the surveyed sample closely followed the size distribution of specialty crops operations from the United States Census data (Fig. 1).

View Full Size

Percentage of farms in each farm size category based on acres of specialty crops production according to the 2017 Census of Agriculture data and the survey sample from a 2019 survey of specialty crops growers registered in the Food Industry MarketMaker and grower associations databases.

Citation: HortScience 58, 1; 10.21273/HORTSCI16909-22

• Download Figure
• Download figure as PowerPoint slide

The Web-based survey was conducted using a mixed-mode design using Qualtrics software. To increase the participation rate, we included an incentive (a $10 gift card) to the first 1000 farmers who completed the survey. Dillman et al. (2014) noted that including token incentives tends to increase online survey participation. We sent three e-mail reminders at 2-week intervals between March and April 2019. A total of 766 farmers completed the survey, resulting in a response rate of 21.5%, which is considered an acceptable rate for this type of survey (Dillman et al. 2014; Torres et al. 2017). The questionnaire included questions related to the farmer’s demographics (i.e., educational attainment, sex, farming experience), farm characteristics (i.e., crops, markets, and growing technologies), and the farmer’s network perceptions of their farm and adopting VA technologies. The respective Institutional Review Board approved the questionnaire for compliance with ethical standards for human subjects.

The subsample for this study included 558 operations growing fruits, vegetables, and culinary herbs. Farmers who did not respond to the questions regarding VA technologies were excluded from the study. We followed the USDA Rural Business Development definition of VA technologies to categorize farmers into two groups according to their adoption of VA technologies used in 2018. The first farmer category, VA, is for farmers who produced VA products such as dried or dehydrated produce, produce cut into customer-ready portions, or prewashed specialty crops. The second category, no VA, included specialty crops operations that answered that they were not using any VA technology in 2018. From the sample of 558 farmers, 293 farmers (52.5%) produced and sold VA agricultural products and 265 (47.5%) were categorized as no VA. We compared farmer categories using multiple comparisons of means using analysis of variance models and Tukey’s honestly significant difference test at a 10% significance level. Analyses were conducted using Stata (release 15; StataCorp, College Station, TX).

Empirical model specification.

Econometric models were used in this study. Using a standard probit, our primary goal was to investigate the impact of market access on the adoption of VA technologies among specialty crops farmers. We also addressed the potential simultaneous causality that may arise from the relationship between the adoption of VA technologies (vertical diversification) and the number of crops grown by a farmer (horizontal diversification) because these two types of agricultural diversification can be common among specialty crops growers (Barbieri and Mahoney 2009). Similar to Ahmadzai (2017), we used an IV approach to control the possible endogeneity from unobserved characteristics that may lead a farmer to adopt VA technologies. Although we proposed that crop diversity may influence farmers to adopt processing technologies that add value to their agricultural products, lower risk, and increase revenue, it may be argued that the adoption of VA technologies can increase the likelihood to produce more crops to take advantage of the technology investment.

A probit model of the decision to adopt VA technologies.

The first bracket in Eq. [2] accounts for the number of sales methods (market diversification index) used by the farmer, including farm stands, farmers’ associations, community-supported agriculture, farmers markets, food hubs, grocery stores, internet orders, processors, restaurants, school districts, wholesalers, and other markets. According to Eq. [2], the market diversification index increases as the number of methods of sales used increases. For example, a grower using five market outlets would have a value of 0.8 for the first bracket, whereas a farmer selling only through a farmers’ market (one market outlet) would have a value of zero. In other words, a larger number of selling methods used by the farmer illustrates a larger degree of market diversification.

The second bracket in Eq. [2] illustrates the distribution of sales methods (market distribution index), representing the proportion of sales through market outlets. For example, a farmer who reported selling crops in the same proportion (50/50) using farmers’ market and wholesale would have a distribution component of - 0.25 or$\frac{1}{2^2}\hspace{0.17em}-\hspace{0.17em}\left({0.5}^2\hspace{0.17em}+\hspace{0.17em}{0.5}^2\right)=-0.25$. However, a producer who sells 90% of products through farmers’ markets and only 10% via wholesale would have a diversification component of $\frac{1}{2^2}\hspace{0.17em}-\hspace{0.17em}\left({0.9}^2\hspace{0.17em}+\hspace{0.17em}{0.1}^2\right)=-0.57$. In other words, a larger negative number in the diversification component would indicate an unequal distribution of sales.

Table 1 describes the explanatory variables in the set of covariates X_2i in Eq. [1]. The parameter vector $\beta \hspace{0.17em}=\hspace{0.17em}{({\beta }_0,{\beta }_1^{\text{'}},{\beta }_2^{\text{'}}}^{})\text{'}$ was estimated, and Φ(⋅) is the standard normal probability distribution function. The set of covariates X_2i contains important factors influencing the adoption of VA technologies, such as farmer demographics, farm characteristics, farmer attitudes, and network and information variables. The demographics of farmers included educational attainment, sex, race, part-time farming or full-time farming, and years of farming experience. Farm characteristics included the number of crops produced, number of family members working at the farm, number of employees hired (permanent and temporary), total owned and rented land, the legal structure of the farm, and percentage of the farm income from specialty crops. Farm size was based on the USDA Economic Research Service categorization of family farms. Small operations were those reporting gross sales from $5000 to $99,999. Medium farms reported gross sales of $100,000 to $250,000. Large operations reported gross sales more than $250,000 in 2018. Farmers were also asked if they perceived sales. The number of employees increased from 2018 to 2019. We used the United States Census regions to group farmers from the Midwest and compared the results to those of their counterparts in other regions. The rationale for this grouping was based on the fact that grouping growers into more categories would result in some groups with few observations because of small sample sizes. The Midwest region grouped farms located in Illinois, Indiana, Iowa, Michigan, Minnesota, North Dakota, and Wisconsin.

Table 1.

Description of farmer demographics, farm characteristics, and behavioral variables for the sample of specialty crops farmers participating in an online survey about the adoption of value-added (VA) technologies (N = 558).

View Table

Because improved technologies are not adopted immediately, Maertens and Barrett (2013) highlighted the role of networks in mediating the adoption of agricultural innovations. The farmer’s network variable was defined as 1 if the farmer reported that family, friends, or farmers in the community adopted VA technologies. Valuable sources of information included information obtained from industry, other farmers, and extension personnel. Attitudinal questions were used to examine the perceptions of agriculture and VA technologies of farmers. Likert-type scales were chosen to capture respondents’ perceptions because they are easy for individuals to answer in a survey and produce good results (Lusk and Coble 2005). The producer’s perceptions of agriculture and their business profitability were rated using a 5-point Likert-like scale (1, strongly disagree; 5, strongly agree). Perception variables included whether farmers agreed that they should receive government support to add value to produce, and whether they should receive financial assistance to accommodate the changing regulatory landscape. We also asked farmers if they were satisfied with their farm business, if they had a positive experience trying new farming technologies to increase profits, and if it was difficult to find reliable customers for VA produce.

Addressing endogeneity.

The regressor used in Eq. [1] raised the concern of endogeneity from a possible bias caused by simultaneous causality between the adoption of VA technologies (vertical diversification) and the number of crops produced (horizontal diversification). Simultaneous causality occurs when the causality can happen in both directions: from the regressors (i.e., independent variables) to the dependent variable, and from the dependent variable to the regressors (Bascle 2008). Ignoring potential endogeneity can lead to endogeneity bias and, consequently, result in the inconsistent effects of crop diversification on the adoption of VA technologies (Bascle 2008).

We consider that endogeneity may be caused by the relationship between adopting VA technologies and diversifying the crop mix. For example, one can assume the farmer’s decision to increase the crop mix might be driven by an investment in a dryer (i.e., VA technology). A farmer may want to increase the return on investment of the dryer by increasing the number of crops grown, dried, and sold. In the other direction of simultaneous causality, farmers growing a diverse number of crops may want to take advantage of their wide variety of products; therefore, they may be more likely to adopt VA as a strategy to increase their farm income. We proposed that the number of crops produced on the farm influences the decision to adopt VA technologies, and not the other way around. We developed a two-stage probit approach with endogenous regressors to address the possibility of endogeneity. Using an IV approach, we aimed to find a variable that influences the potential endogenous regressor (number of crops) but is not related to the farmer’s decision to adopt VA technologies (VA).

We proposed the use of farming technologies (e.g., artificial lighting, aeroponics, aquaponics, hoop houses, hydroponics, greenhouses, plasticulture, and irrigation) as the instrument for the variable number of crops. According to the diversification literature, farmers using farming technologies tend to report, on average, six additional crops in their crop mix (Lancaster and Torres 2019). The use of these growing technologies can help farmers extend the farming season, increase yield, and improve pest management, which tend to influence the adoption of a greater number of crops in their operations (Carey et al. 2009). The IV captured the effect of the potential endogenous regressor (number of crops) on the dependent variable (VA) by influencing only the potential endogenous variable and not influencing the dependent variable.

Descriptive statistics.

Table 2 describes the explanatory variables with mean differences for all the variables used in the models by the type of farmer. The first bracket of the market variable (i.e., market diversification index) is significantly higher for VA farmers (0.52) when compared with the index for farmers that do not adopt VA technologies (0.33) (P < 0.01). In other words, VA farmers tend to sell their produce through more markets compared with farmers who do not use VA technologies. Female farmers represented one-third of the total sample data, which is consistent with national statistics indicating that 36% of farmers in the 2017 Agriculture Census were women (US Department of Agriculture National Agricultural Statistics Service 2020). Interestingly, more women are VA farmers (38%) than no VA producers (26%) (P < 0.05). This concurs with findings of Dias et al. (2019), which indicated that a more significant proportion of female farmers adopt VA technologies to increase farm revenue, whereas men tend to focus on increasing yield.

Table 2.

Descriptive statistics of the full sample of farmers and mean comparisons of farmer categories from the sample of specialty crops farmers participating in an online survey about the adoption of value-added (VA) technologies among specialty crops growers in 32 states of the United States (full sample, N = 558; VA, N = 293; no VA, N = 265).

View Table

On average, farmers adopting VA technologies increased their crop numbers by more than twice (20) compared to those not adding value to their crops (8) (P < 0.01). These results are consistent with those of De Benedictis et al. (2009), who found that specialty crops operations tend to increase the crop portfolio as they move out of early development stages; in this case, this occurred when operations started to adopt diversification strategies. Expectedly, fewer VA farmers (29%) reported farming part-time, which was significantly lower than the 37% of no VA farmers (P < 0.05). According to Jablonski et al. (2020), agricultural operations that add value to their products tend to be more labor-intensive than their counterparts and may require farmers to farm full-time.

Although there is no significant difference across farm types regarding the number of employees or family working at the farm, farmers adopting VA technologies, on average, reported significantly higher employment growth in their operations (17%) in 2018 than non-VA farmers (12%) (P < 0.1). Similarly, farmers adopting VA technologies reported, on average, more growth in sales (37%) in 2018 than their counterparts (30%) (P < 0.1). These results suggest that farmers who adopt VA technologies experience more economic growth than their counterparts.

Maertens and Barrett (2013) highlighted the importance of networks to agricultural technology adoption. Our findings indicated that 37% of farmers who adopted VA technologies knew at least one member of their community (e.g., family, friends, or farmers) who has adopted VA technologies. This value was significantly lower for the farmers who did not adopt VA (P < 0.01). Interestingly, among all sources of information (i.e., industry, farmers, and university cooperative extension services), only information from industry associations was significantly useful for VA farmers. Approximately 73% of non-VA farmers considered industry associations to be a useful source of information; conversely, 65% of their counterparts reported that they considered this source of information useful (P < 0.05).

We found that all the perceptions of VA technologies evaluated in our model were statistically significant across farmer categories. A higher proportion of VA farmers perceived they should receive government support to add value to their crops (P < 0.1) and financial assistance to be able to accommodate the changing regulatory landscape compared to non-VA operations (P < 0.05). More VA farmers reported a positive experience with VA technologies to increase profits (P < 0.01) and declared that it was more difficult to find reliable customers for VA produce (P < 0.01) than non-VA farmers. In contrast, VA farmers were significantly less satisfied with their farm business than their counterparts (P < 0.05).

Regressions results.

This study provides empirical evidence of the effect of market access and other key drivers on the farmer’s decision-making process regarding VA technologies. Table 3 reports the coefficients, standard errors, and marginal effects from the standard probit and the IV probit regressions. Table 3 shows that results from the IV approach were consistent with the standard probit. To illustrate, we found that standard errors and robust standard errors were similar, suggesting the lack of heteroskedasticity in our data (King and Roberts 2015).

Table 3.

Standard probit and instrumental variable (IV) probit estimation results of the factors influencing the adoption of value-added (VA) technologies among specialty crops growers in 32 states of the United States.

View Table

Results of the first stage (reduced form) of the IV probit regression provided the validity of our IV approach (available on request). The statistically significant results and strong correlation between farming technologies (IV) and the number of crops (potential endogenous variable) illustrate the strength and power of the IV used in this model. Consistent with Lancaster and Torres (2019), we found that farming technologies were a major factor influencing crop diversification among specialty crops operations, demonstrating the relevance of the IV approach.

A key finding was the fact that in the IV probit estimations, the parameter p from the Wald test of exogeneity was not statistically significant (P = 0.36) (Table 3). A p that is not statistically significant is equivalent to the number of crops being unlikely to be endogenous relative to the adoption of VA technologies. Therefore, there was insufficient evidence for declining the null hypothesis that the model is exogenous. In other words, endogeneity is not likely to be an issue in the analysis; therefore, we used the marginal effect results from the standard probit (Table 3) to explain how market access and other factors influence the decision to adopt VA technologies among specialty crops operations. A reason why endogeneity was not an issue may be the extensive list of covariates included in the model. The findings also provided evidence that the number of crops produced by farmers is a crucial factor influencing the adoption of VA technologies, and not the other way around.

Probit results from Table 3 provided robust empirical evidence that the first component (market diversification index) of our key explanatory variable equation (market) significantly influenced the farmer’s decision to adopt VA technologies. On average, the probability of adopting VA technologies significantly increased by 28.7% as the market diversification index increased (P < 0.01). This result suggests that the number of markets accessed is a significant factor determining the adoption of VA technologies. One explanation may be that farmers accessing more market channels have more exposure to a diverse range of consumer trends and demands; therefore, they are more likely to adopt VA technologies to differentiate their products. We can infer that accessing various markets can impact the diversity and differentiation strategies in the specialty crops industry. Therefore, we expect that farmers located in remote areas with less market access may not be driven to add value to their produce through VA technologies, and policies encouraging the adoption of VA technologies among those farmers may not be as efficient as those targeting farmers near or in urban areas. Policymakers can use these findings to consider whether market access policies and incentives are associated with the adoption of VA technologies.

Results from the standard probit regression illustrated that increasing the number of crops grown can significantly drive the adoption of VA technologies. For example, increasing the crop mix by one crop can increase the likelihood to adopt VA technologies by 1.1% (P < 0.01). It seems that horizontal diversification has a significantly positive effect on vertical diversification (VA technologies) for specialty crops growers. Our results were similar to those of Morris et al. (2017), who found that crop diversification tends to influence technology adoption.

Farmers selling only to DTC markets were 9.9% more likely to adopt VA technologies than their counterparts (P < 0.1). We expect that local markets may be providing farmers with social interactions with customers, which, in turn, can be a source of information to tailor products and access price premiums for differentiated VA products. However, Low et al. (2020) found that employment was not significant when adding value to agricultural products. Deogharia (2018) reported that farmers adopting VA technologies hired more labor. Our results suggest that labor was a significant driver for VA agriculture in the specialty crops industry. Results from Table 3 suggest that having one more family member involved in the agricultural operations increased the likelihood of the farmer adopting VA technologies by 3.2% (P < 0.01). For our sample of farmers, it seemed that the family’s participation in the farm positively influenced the adoption of VA technologies. Additionally, Table 3 illustrates that experiencing employment growth was positively correlated with the adoption of VA technologies (P < 0.05). According to Jablonski et al. (2020), agricultural operations adding value to their products tended to be more labor-intensive than their counterparts.

The effects of networks have been reported as a significant factor in the technology adoption process (Maertens and Barrett 2013). Results from the standard probit indicated that having a family member, friends, or fellow farmers in the community who have adopted VA technologies increased the probability of VA technology adoption by 13.7% (P < 0.01). Our result is consistent with that of studies reporting the importance of networks to the technology adoption process (Ward and Pede 2015). Farmers that agreed that it is difficult to find reliable customers for VA produce were 12.5% more likely to adopt VA technologies (P < 0.01). Because farmers adopting VA are the ones looking for markets for their VA products, it is expected that they would tend to perceive access to customers as a barrier.