Survey

The consumer food purchase data used in this study came from a public mail survey of 7500 randomly selected households in Florida implemented by the University of Florida in 2012, as originally reported (Hodges and Stevens, Reference Hodges and Stevens2013). This survey collected information on Florida consumers’ shopping frequencies and values of foods purchased through multiple local food distribution channels. For the specific purpose of this paper, we focused only on purchases at farmers' markets, roadside stands, and U-pick operations. A series of questions included consumers’ perceptions of local food attributes (freshness, nutrition, food safety, food security, organic certification, etc.) and limiting factors (unavailability, seasonality, not knowing if truly local, high price, etc.). The survey was addressed to the primary food shopper in the household. Questionnaires were assigned a unique code number corresponding to the mailing address for geocoding and spatial analysis.

After eliminating duplicates and questionnaires returned as undeliverable, 1599 usable responses were obtained, representing a response rate of 21.4%. The distribution of responses is mapped in Figure 1. The North Central region, Tampa Bay and Miami-Ft. Lauderdale regions were over-sampled or had higher response rates. Nearly three-quarters of respondents were female (72.4%), middle aged between 45 and 64 years (42.5%) and obtained a college degree (32.7%) than the general Florida population (25.4%). Although the survey sample was broadly representative of the Florida population, an overall sample weighting factor was assigned for each respondent in the empirical model to correct for differences in age, income, education and location.

Figure 1.

Survey respondents’ locations by Florida economic regions. (Note: functional economic regions were defined by US Commerce Department, Bureau of Economic Analysis; Johnson and Kort, Reference Johnson and Kort2004).

Among 1599 respondents, 61.7% reported that they had purchased locally grown foods at a farmers' market, roadside stand or U-pick operation in 2012. The most commonly reported food types purchased were vegetables and fruits (73.0%). Regarding consumers’ perceptions of local products, attributes rated as ‘very important’ were ‘freshness’ (89.7%), ‘food safety’ (78.0%) and ‘nutrition’ (67.3%) (Fig. 2). On the other hand, ‘having a relationship with producers’ and ‘organic certification’ were reported as ‘not important’ by a large percentage of respondents. The most limiting barriers found were ‘high price’ (rated as ‘very limiting’ by 34.4%), ‘unavailability’ (26.3%) and ‘not knowing if truly local as labeled’ (24.6%) (Fig. 3).

Figure 2.

Summary of consumers’ perceived attributes of local foods.

Figure 3.

Summary of consumers’ perceived barriers for local food systems.

Theoretical framework and hypotheses

The conceptual framework employed in this study was the random utility model of consumer choice developed by McFadden (Reference McFadden1974). The basic assumption is that an individual maximizes his or her utility given a certain budget constraint, utility is ‘a function of attributes of products or services and characteristics of this person’ (Lancaster, Reference Lancaster1966), as revealed by their purchasing choices. In this study, if we assign the observed choice as the dependent variable Y, then Y = 1 means that this individual chooses a direct food outlet and obtains a higher utility from this food channel. On the other hand, Y = 0 implies this individual prefers other traditional food vendor sources. The predicted value of Y can be interpreted as the expected probability that an individual shops at farmers' markets, roadside stands and U-pick operations.

Now the key question is what explanatory variables should be included in the conceptual framework? Firstly, spatial factors are the central variables of interest in this study. As previous literature suggested, lack of access to markets is a crucial barrier for farmers' market patrons. Moreover, since food products in direct-to-consumer channels share many similar attributes to those in grocery stores or other food outlets, it is possible that having a traditional food outlet nearby makes people less likely to visit farmers' markets, roadside stands and U-pick operations. Therefore, we hypothesize that:

• H1: Ease of access to farmers' markets increases consumers’ direct food purchases.

• H2: Availability of a nearby alternative food outlet negatively affects a consumer's likelihood of shopping at farmers' markets, roadside stands and U-pick operations.

Next, we include consumer perceptions of local foods as control variables. These variables have been widely used in consumer demand analysis because they affect an individual's utility of consuming a good (Brown, Reference Brown2003; Bond et al., Reference Bond, Thilmany and Bond2006; Thilmany et al., Reference Thilmany, Bond and Bond2008; Conner et al., Reference Conner, Colasanti, Ross and Smalley2010). As Lancaster stated, ‘the good itself does not contribute to an individual's utility, but the attributes associated with that good do’ (Lancaster, Reference Lancaster1966). Many studies showed that interest in local foods is a primary motive for many direct food market patrons. For example, a national online survey found that respondents who used direct-to-consumer markets as a primary food source tended to rank ‘variety available’ and ‘support for local producers’ higher than other attributes (Bond et al., Reference Bond, Thilmany and Bond2006). Another survey of New Jersey farmers' markets shoppers indicated that 87% of respondents thought ‘availability’ and ‘freshness’ affected their decisions on where to purchase (Govindasamy et al., Reference Govindasamy, Zurbriggen, Italia, Adelaja, Nitzsche and Van Vranken1998). Other studies have suggested that negative perceptions of farmers' markets include seasonal availability, high price, inconvenient operation times, etc. (Eastwood, Reference Eastwood1996; Andreatta and Wickliffe, Reference Andreatta and Wickliffe2002; Schmidt et al., Reference Schmidt, Kolodinsky, Desisto and Conte2011). Although some of these perceptions may not be justified, they may influence consumer's purchasing decisions. So we hypothesize that:

• H3: Consumers’ perceptions of favorable local food attributes positively affect their participation in direct food markets, and their perceptions of the limiting factors have negative effects.

Some exogenous variables such as demographic characteristics were found to be significant indicators of consumers’ purchasing patterns. For example, consumers who were female, middle aged and with higher income were more willing to purchase directly from local farmers (Adrian and Vitelli, Reference Adrian and Vitelli1982; Smallwood and Blaylock, Reference Smallwood and Blaylock1984; Eastwood, Reference Eastwood1996; Wolf, Reference Wolf1997). In this regard, we test the following hypothesis:

• H4: Consumers with certain demographic characteristics (such as female, middle aged, higher education attainment, high income, etc.) are more likely to purchase at farmers' markets, roadside stands, and U-pick operations.

Model specification

A probit analysis was selected to examine the hypothesized relationships above based on some statistical concerns. If the respondent purchased food from these direct market outlets in 2012, then the dependent variable was 1 or 0, otherwise. Since binary response variable violates the assumption of homoscedasticity for linear models, several nonlinear functional forms have been developed to address this issue. The probit technique is one of the most widely used nonlinear frameworks (McFadden, Reference McFadden1974; Mabiso et al., Reference Mabiso, Sterns, House and Wysocki2005; Conner et al., Reference Conner, Colasanti, Ross and Smalley2010; Ross et al., Reference Ross, Shanoyan, Conner and Colasanti2010). Another popular alternative is the logit model; however, the probit model has a slightly better fit than that of logit model for our dataset. We also chose it because the interpretation of marginal effects tends to be more straightforward than the odds ratio in a logit model.

The rationale to apply the probit model to choice behavior can be explained by the random utility theory (McFadden, Reference McFadden1974). If a respondent purchased any local foods from farmers' markets, roadside stands and U-pick operations in 2012, this implies that he obtained a higher utility than from traditional food channels. Let Z* be the difference of utilities between these two alternatives. Consumers visited direct food markets if Z* > 0 (i.e., Y = 1) and not if Z* < 0 (i.e., Y = 0). Then the latent variable Z* for individual i can be estimated as:

(1) $$Z_i^{\ast} = \mathop \sum \limits_{k = 1}^N {\rm \beta} _{ik} x_{ik} - u_i, \quad Z^{\ast} = U_{\_Direct} - U_{\_Traditional} $$

where the set of x _ik terms are explanatory variables and u _i is the error term. Next, the nonlinear transformation links choice indicator Y with Z* to predict the probability of Y = 1,

(2) $$\eqalign{P(Y_i = 1) &= P(Z_i^{\ast} \gt 0) = P\left( {u_i \lt \mathop \sum \limits_{k = 1}^N {\rm \beta} _{ik} x_{ik}} \right) \cr & =\; \int_{ - \infty} ^{\sum\nolimits_{k = 1}^N {{\rm \beta} _{ik} x_{ik}}} {\,f\,(u)du}}$$

where f(.) is the probability density function of error term u _i , which is assumed to be normally distributed under the probit framework. As $\sum\nolimits_{k = 1}^N {{\rm \beta} _{ik} x_{ik}} $ increases from negative infinity to positive infinity, the predicted probability goes up from 0 to 1.

Data and variables

Spatial factors in the model included ease of access to farmers' markets and existence of other food stores. For the former, we developed two measurements: distance to nearest farmers' market and number of farmers' markets within a 5 km radius for each household. The 5 km distance was considered as a ‘reasonable’ driving distance for food shopping. The addresses of households used in the survey were geocoded, and the geocoded locations of Florida farmers' markets in 2012 were obtained from the National Farmers Market Directory maintained by the USDA Agricultural Marketing Service. We used ArcMap 10.2.2 software (ESRI R, 2011) to put these two data layers into a coherent dataset, and applied the Proximity toolset to find the Euclidean (straight line) distance to the nearest farmers' market. Then a 5 km buffer was created for each household, and the number of farmers' markets within this zone was counted.

Data on locations of other food outlet came from the InfoUSA database and ESRI^® Business Analyst. We grouped these food stores into three types: grocery stores, food specialty stores and convenience stores based on the North American Industry Classification System (NAICS). Again, we used ArcMap 10.2.2 software to determine the numbers of these food outlets within 5 km for each household. In sum, the spatial variables included in the analysis were:

• • Distance to nearest farmers' market,

• • Number of farmers' markets within 5 km,

• • Number of convenience stores within 5 km,

• • Number of grocery stores within 5 km,

• • Number of food specialty stores within 5 km.

Next, we included consumers’ perceptions of local foods as control variables. The respondents were asked to evaluate the relative importance of local food attributes and limiting factors. The attributes such as freshness, nutrition, food safety, etc. were rated on a three point Likert-type scale of ‘very important’, ‘moderately important’, ‘not important’ and ‘don't know’. Dummy variables were constructed for each category and ‘not important’ was treated as the reference group in the probit model. For the limiting factors (unavailability, seasonality, high price, etc.), dummy variables were constructed for ‘very limiting’, ‘moderately limiting’, ‘not limiting’ and ‘don't know’ with ‘very limiting’ as reference group. Specifically, these variables included the following:

• • Local food attributes: freshness, nutrition, food safety, food security, organic certification, pesticide free, price, shelf life, reduced transportation, and relationship to producer.

• • Limiting factors: unavailability, seasonality, not knowing if truly local, high price, inconvenient operation hour, market congestion, time required for preparation of raw foods, lacking knowledge to prepare, lacking transportation and lacking storage capacity.

In addition, the ‘area considered local’, was included as a variable, i.e., what is considered to be the area in which local foods are produced (city, county, 100 mile radius, state of Florida, Southeast U.S. region). Finally, respondents’ demographic characteristics were added to the model, including age, household income, number of persons in household and type of residential area. Again, we constructed dummy variables for each category of these variables. See Appendix 1 for more information.