The Role of Community and Farm Characteristics in Farm Input Purchasing Patterns

Walter Goldschmidt's widely cited but contentious finding that farm size was the most important cause of socioeconomic differences in two rural communities in California, has provoked more than a half a century of research analyzing the relationship between farm structure and rural economic development (e.g., Gilles and Dalecki; Heady and Sonka; Lasley et al.; Strange; Zeuli and Deller). Testing Goldschmidt's premise that smaller farms purchase a greater share of their inputs locally, and thus provide an essential foundation for vibrant rural economies, is a common line of inquiry within this body of literature (e.g., Chism and Levins; Foltz, Jackson-Smith, and Chen; Henderson, Tweeten, and Schriener; Korsching; Lawrence, Otto, and Meyer; Roe and Stockberger; Marousek). Although each of these studies has added insights into the issue, as a whole they offer no definitive support for the hypothesis that farm size is a key indicator of where farm inputs are purchased. Instead, they collectively suggest a more complex situation. A myriad of farm and farmer characteristics can drive input purchasing patterns and the relative importance of each characteristic depends on the good or service being purchased.

To further complicate the issue, a farmer's decision to purchase a good or service locally is also clearly a function of his or her ability to buy locally. Some farmers will simply not be able to find all the products and services they need in their community. Yet, with a few exceptions (most notably Henderson, Tweeten, and Schriener), none of the farm purchasing pattern studies has addressed local supply constraints.¹ More generally, community characteristics that could influence purchasing patterns have largely been ignored in the research. As Carlin and Saupe note, these studies fail to address the “two-way” relationship between communities and farms. This shortcoming may help explain some of their contradictory findings regarding the importance of farm size to purchasing patterns.

The purpose of this paper is to ascertain the driving factors behind a farm's propensity to purchase locally when accounting for differences in both demand (farm and farmer characteristics) and supply (community characteristics). For instance, is farm size an important determinant of local purchasing regardless of local supply constraints? To address this issue, we access a unique data set (from 2003) covering 141 dairy farms in three distinct dairy-dependent Wisconsin communities. These data have panel characteristics that allow us to isolate various community effects on farm purchasing patterns. They also allow us to differentiate purchasing patterns for eleven separate farm inputs.

This paper makes an additional contribution to the literature by using economic theory to explain the logic behind local purchasing decisions. A theoretical factor-demand model indicates three testable hypotheses regarding the influence of farm scale and other farm and farmer characteristics on local demand. Two additional hypotheses related to the supply of goods and services as well as community attachment are also empirically tested in two sets of regressions using double-bounded Tobit models. In the first set of regressions, community differences are represented by a set of dummy variables. In the second set of regressions, we replace the dummy variables with more specific business supply data from each community.

Local Purchasing Theory and Hypotheses

Farm and farmer characteristics as well as community characteristics influence farm purchasing patterns. Community development theory suggests that local supply is indeed an important consideration, since rural communities often differ in terms of the goods and services they offer, as is community attachment. The literature, with one or two exceptions, largely ignore these considerations (i.e., the ability and willingness to purchase locally) and focus instead on the farm-level determinants, using empirical models to explore the significance of various characteristics (e.g., farm size). We expand on this work by developing a theoretical model that explains the connection between farm-level characteristics and local demand. Three additional hypotheses are derived from this model.

A Model of Farm Purchasing Decisions

In order to demonstrate how a farm's local purchasing pattern might vary with farm characteristics, we derive the determinants of a farm's purchasing pattern from a factor-demand model. Following Foltz, Jackson-Smith, and Chen, let the farm purchase two inputs (good A and good B); good A is sold both locally and in a distant location while good B is sold only locally. Good A is assumed to have uniform quality regardless of its source. This makes local good A and distant good A perfect substitutes in production, so the farm will buy whichever has the lowest price. For simplicity, we also assume that the farm's production requires a fixed ratio of inputs, with α being the proportion going to good A and (1 − α) to good B.

Since the farm is able to purchase inputs locally and in a distant location, it faces a nonlinear price structure. We assume that local inputs have higher prices than those purchased in a distant town when a larger quantity is purchased (i.e., the distant supplier provides a quantity discount).² We also assume that local purchases have no transaction costs since the farmer is likely to have long-standing relationships with local businesses, especially if they are co-ops; distant purchases require search costs associated with finding and negotiating with new suppliers. Distant purchases also incur higher transportation costs, greater opportunity costs associated with being away from the farm, and perhaps membership costs in a new co-op.

These assumptions define a cost function for the farm with the following attributes: it is linear in good A and good B while having fixed factor proportions between the local and distant good A (Chambers). The differential transaction costs associated with local and distant sources of inputs create distinct cost-share functions for the two types of good A purchased. Let the input prices be represented as follows: w₁ is the price of locally purchased good A; w₂ the price of good A purchased in a distant location; and w₃ the price of good B.

Assume that the actual price our farmer will face for good A purchased in a distant town will be some function of the posted price w₂, the “transaction costs” she incurs, τ, and the amount she wishes to buy, which in this case can be described by the scale of production, y. For simplicity, τ represents the total transaction costs plus any additional costs associated with purchasing from a distant location (transportation, etc.). The farm's cost function can be written as:

(1)

where g(w₂, y, τ) is a function describing the relationship between the base price, w₂, the amount bought, y, and transaction costs, τ. Let the function g(w₂, y, τ) be described by the following equation:

(2)

The first term is the standard cost of good A, the second term is a markup price that decays by a positive, fixed exogenous parameter, γ, as the quantity purchased increases, and the last term is a fixed transaction cost incurred in the purchase. The resulting cost function,

(3)

has constant returns to scale for w₁y < (w₂y+w₂y^−γ+τ) and decreasing returns to scale otherwise. Thus, the average and marginal cost functions will be constant until a threshold point at which there is equality between the prices of local and distant good A, with the average cost function declining and the marginal cost function increasing from that point.

The local cost-share equation can be derived from the farms' cost minimization problem: {min C(w, y, γ | τ): s.t. y≥y_o}. Let C_L represent local purchases and C_D represent distant purchases. Then the associated local cost share function will be:

(4)

This implies that the local cost share function will be a nonlinear function of the scale of production and the transaction costs of distant purchases. This nonlinearity will show up in an estimation of cost share functions being censored at 1 or 100% of the inputs purchased locally. The local cost share function will be increasing in the transaction costs, τ, since higher transaction costs make distant purchases more expensive. In contrast, the local cost-share function will be decreasing in the scale of production, y, since the scale of production will reduce the costs of the distant good.

Our theoretical model leads to three testable hypotheses. If there are in fact scale effects as suggested by the theoretical model, larger farms should buy a lower proportion of their inputs locally. Another major implication of the theoretical model is that the share of distant purchases will depend on the transaction costs involved. The nature of transaction costs vary and thus can be measured in several different ways. For our purposes, we are primarily interested in coordination costs, which we assume are a function of a farmer's personal and household characteristics. It seems logical to suppose that younger, better-educated farmers will have lower transaction costs when making distant purchases. They are more likely to have an easier time accessing nonlocal suppliers because of greater facility in traveling, reading magazines with supplier advertisements or finding suppliers through the Internet, and entering into long-distance contracts. For older farmers (assuming they also have more farming experience), the local-distant transaction cost differential may be much greater since they are more likely to have established ties in the local community. For instance, it is likely that they would have built up substantial equity in their local co-ops.

In addition, we also expect that as family members spend more time in their local town, the local-distant transaction cost differential and transportation/opportunity costs increase. The more time the farmer or spouse spends in the local community, the higher the probability that they will develop stronger ties with local businesses (and the greater the relative search costs in a distant location). Clearly, transportation costs for local purchases would also decrease, since an additional trip for many supplies would be unnecessary. The more time spent off farm also suggests less time available for searching and traveling to alternative suppliers (the opportunity costs increase).³ Clearly, there are other factors that may determine transaction costs. However, age, education, farming experience, and time spent in town seem to be particularly relevant to purchasing transactions.

Finally, the farther a farm is located from the center of the local town, the transportation cost differential between local and distant purchases decreases. Some farms may be technically located in a certain town (i.e., share the same zip code), but lie fairly close to other towns as well. Therefore, we expect distance from town to be a factor in determining local purchases. The three hypotheses stated more formally follow:

• H3: The proportion of inputs purchased locally declines as farm size increases.
• H4: The proportion of inputs purchased locally will depend on farmer and household characteristics related to transaction costs: age, education, farming experience, and time spent in town.
• H5: The proportion of inputs purchased locally will decrease as distance from farm to town increases.

Survey Methodology and Community Background Information

Farm-level data were collected from 141 dairy farmers who completed a comprehensive mail survey in three dairy-dependent Wisconsin towns: Athens, Chilton, and Richland Center. The three communities were selected based on the following criteria: (a) they comprised a large number of dairy farms; (b) the town center was geographically located near the middle of the community (as defined by a single zip code); and (c) farming was a significant part of the local economy, both in terms of employment and income. They were also chosen to represent contrasting regions of the state, as well as different facets of the evolving dairy industry.

Athens (population 1,095 in 2000) is a traditional small dairy town located in the northwest corner of Marathon County, surrounded by several mid-size cities. Chilton is a moderate-sized (population 3,708 in 2000) commercial dairy town in Calumet County, an area witnessing considerable nonfarm growth pressures. It is in the center of the county, crossed by two major thoroughfares, and is surrounded by several other small and medium-sized towns. Richland Center (population 5,114 in 2000) is an area of more marginal land with many dairy farms undergoing rapid expansion and/or moving toward intensive rotational grazing. Richland Center is also situated in the center of the county, at the nexus of two major highways. Richland County is the only strictly nonmetropolitan county of the three, following the definitions established by the U.S. Department of Agriculture's Economic Research Service (Cook and Mizer).⁴

The survey, sent to all 356 dairy farms in the three towns, was conducted in February 2003; it elicited responses on general farm and farmer characteristics (age, education, farming experience, and off-farm work), community attachment, and local spending patterns. Farmers were asked where they purchased eleven different farm-related products and services (corn grain, corn silage, soybeans and protein feeds, alfalfa and other forages, feed supplements, replacement heifers, veterinary services, milking equipment, farm machinery, farm supplies, and custom harvesting services) over the previous year. They were asked to allocate (by percentage) their annual quantity of expenditures per input among the following five options: (a) not used on their dairy farm; (b) grown or raised (not purchased); (c) purchased in their hometown; (d) purchased in a neighboring town; and (e) purchased elsewhere. The proportion purchased across all five options summed to 100%.

In order to capture information on the local supply of farm inputs, relevant (i.e., farm related) businesses listed in the ReferenceUSA™ database, chamber of commerce records, and local yellow pages in each community were asked to complete a short questionnaire over the phone. They were asked to confirm whether they sold any of the eleven products or services we asked about in the dairy farm survey. If they sold any of the products or services, they were also asked whether they were cooperative businesses.

Empirical Model Estimation

The theoretical factor-demand model established a nonlinear relationship between local cost shares, C* = C_L/(C_L+C_D), and farm/farmer characteristics, X. Local cost shares are censored from above at 100% and below at 0%. Rescaling the local purchases in our data to 0–100% captures a large portion of the observations for all goods and services. Thus, a double-censored Tobit model (Maddala), which takes this censoring into account, is appropriate. For an individual data point with a vector of independent variables x_i, and a vector of parameters to be estimated β, a double-censored Tobit is estimated as follows:

(5)

where C_i = 100 if C^*_i ≥ 100, C_i = C^*_i if 0 < C^*_i < 100, C_i = 0 if C^*_i < 0.

The estimation procedure for this model maximizes a standard Tobit likelihood function with the changes for upper censoring rather than the more common lower censoring at zero. With 100% as C^u, the upper bound of our estimation, and C⁰ denoting the lower bound, 0%, the likelihood function is as follows:

(6)

where ɸ is the normal cumulative distribution function and C_i is the cost share.

Since our analysis focuses on local business effects, we only consider purchases made off the farm, not on-farm production. Two sets of regressions are estimated. In each set, the dependent variable is the share of expenditures on a good or service that is purchased in the farmer's hometown, i.e., Athens, Chilton, or Richland Center. Each of the eleven different goods and services asked about in the survey are treated as independent regressions. Farms that did not purchase a good or service anywhere (i.e., they either did not use it or grew it on their own farm) are dropped from the data for that particular regression.

The independent variables (table 1) include: cows in 2002 (the number of cows on the farm as a measure of farm size); percentage of off-farm work (the percentage of hours per week the farm manager and spouse spend working off farm); age (the age of the farm manager in years); education (an index variable measuring the level of education attained by the farm manager); attachment (an index representing how attached the farm manager is to the community); distance (the distance in miles from the farm to the town center); and local groceries (a dummy variable equaling 1 when the respondent reports buying their groceries locally, 0 otherwise). The hours the farm operator and spouse spend working off the farm and local grocery shopping serve as proxies for a farm household's time spent in town.

In the first set of regressions, two town dummy variables, Athens and Chilton, are included as independent variables. In the second set, a count of the number of businesses that sell the particular product or service represented by the dependent variable (number of businesses) replaces the dummy variables.⁵ For each data observation, the number of businesses variable represents the number of businesses selling the input in question in the appropriate town (e.g., the number of livestock veterinarians in Athens for all farm-level responses from Athens). These two methods provide two types of controls. The first controls for town-specific but unmeasured characteristics such as transportation and marketing infrastructure, their competitive advantage over neighboring towns, and the spatial distribution of business locations. The second controls specifically for the number of businesses and thus offers a cleaner supply-side metric.

Descriptive Statistics of Data

On average, the smallest farms (67 cows) are in Athens and the largest in Richland Center (115 cows) (table 1). They are relatively close to the average herd size (83 cows) for Wisconsin (Wisconsin Agricultural Statistics Service). The range of size for the entire sample was fairly large: from 12 to 800 cows. The histograms (figure 1), however, show that most farms in each community have fewer than 100 cows and the distribution of herd size is fairly consistent across the three communities. The percentage of total hours per week worked off farm for both the farmer and spouse was fairly consistent across communities. The smaller farms in Athens were slightly less dependent on off-farm work (10%) than in either Chilton (13%) or Richland Center (17%). This is surprising given that small farms generally earn less net income and therefore, it is typical for spouses to work off farm, especially for health-care coverage (WASS).⁶ These statistics are also lower than the state average. According to the 2002 Census of Agriculture (USDA), 72.2% of the farmers in Wisconsin that reported any off-farm work (54.6% of the farms) worked 200 or more days off farm.

The age and education of farmers in Athens and Chilton are similar; the average farmer in each community is forty-eight years old with a high school diploma or some college experience but no degree. The average farmer in Richland Center is older (fifty-three) and slightly better educated (some college, but no degree). For comparison, the average age of all farmers in Wisconsin is fifty-three (USDA). As expected, a farmer's age and farming experience were highly and positively correlated (0.799); therefore, only age is included in our analysis.

Community attachment is measured by asking the farmers to rank how attached they feel to their community on a scale of 1–10, with 1 = “no attachment” and 10 = “very strongly attached.”⁷ On average, farmers in all three communities said they were fairly attached (i.e., they ranked their attachment as 5–6), although the range of responses in each community was complete (1–10). Despite being the smallest of the towns, community attachment in Athens was the lowest on average (5.6), but the average distance from farm to town was also highest (7 miles). The communities vary significantly in terms of the percentage of farm households that normally do most of their grocery shopping in their town. In Athens, only 29% of the households shop there for groceries, compared to 90% in Chilton and 97% in Richland Center. This finding is a function of supply; Athens has fewer grocery stores and businesses in general than the other two towns.

Table 2 shows that local spending as a percentage of all farm-related purchases of products and services also vary across towns. Overall, farmers tend to purchase more locally in Richland Center and Athens than in Chilton. In Richland Center and Athens, there are six products and services where, on average, farmers purchased more than 60% (the percentage of their total annual expenditures for each product) locally (highlighted in bold face), while in Chilton there are only four. However, in Athens, for those products and services primarily purchased locally (>60%), the average purchased is 78%, which is higher than Richland Center (69%) and Chilton (69%). Note also that there are differences in terms of what is purchased locally across towns. On average, farms purchase 50% or more of their corn silage and custom harvesting services locally in all three towns. Also, on average they purchase 40% or less of their replacement heifers locally. In Athens, farmers are more likely to purchase feed products locally than any other products and services, with the exception of harvesting services. The reverse is true in Richland Center, where except for corn grain, farmers purchase significantly lower amounts of feed products locally than other products and services. In Chilton, there is no clear purchasing trend.

Analyzing the data by farm size (table 2) provides a different picture. It seems to support Goldschmidt's premise that large farms purchase less locally than small farms. The average small farm (<75 cows) purchases substantially more of eight of the products and services locally than the average large farm (≥150 cows). The exceptions are milk equipment, farm machinery, and farm supplies. The story becomes more complex when medium-size farms are considered (75–149 cows). The typical medium-sized farm buys (in some cases substantially) higher portions locally than small farms for seven products and services and somewhat higher portions locally than large farms for all but one product (alfalfa).

Table 3 shows the number of businesses selling the products and services that farmers were asked about in each of the three towns.⁸ All three towns are relatively well supplied in farm inputs, although the results vary in terms of products and services. For example, Chilton has the greatest selection of firms selling milk equipment and farm machinery, although it offers the least selection of corn grain businesses and has the fewest veterinary clinics. Some of the similarities between towns are particularly important for our analyses. None offers local sources of corn silage or custom harvesting and alfalfa is only sold locally in Chilton.

Estimation Results and Discussion

Tables 4 and 5 show the results of the two sets of regressions. Note that three regressions were dropped from the analysis due to insufficient variation in the data: corn silage, replacement heifers, and alfalfa. In general, the results presented in table 4 show few significant parameters. It becomes immediately apparent that there is very little evidence to support the hypothesis that small farms are more likely to buy locally than large farms. The farm size variable (cows in 2002) is only significant and negative in the case of feed supplements. Therefore, this is the only input for which there is support for the theoretical model and where we can accept our hypothesis that farm size influences purchasing patterns (H3).

The regression results also generally fail to support our fourth hypothesis that certain farmer and household characteristics (used as proxies for transaction costs) influence local purchasing patterns (H4). Age is significant (at a 10% level) for only two inputs: feed supplements and custom harvesting; in the former it has a positive sign (as expected), while the latter is counterintuitively negative. Education is only significant for feed supplements where, counter to our theory about education's relationship to transaction costs, it is positive. The proxies for time spent in town (percentage of off-farm work and local groceries) are also only significant for one input. Off-farm work is significant for corn grain (although it is negative)⁹ and the purchase of groceries locally is significant for milk equipment (although it is also negative).

The results suggest that in general transaction costs do not seem to significantly influence local purchasing decisions (as compared to other explicit cost factors). Since these four variables are only proxies for transaction costs, however, we cannot decisively reject the idea that transaction costs influence purchasing decisions; the variables we chose may not fully capture the transaction cost effects. In another cost-related finding, distance is not significant for any input and thus there is little evidence to support our fifth hypothesis (H5) that farms located closer to town purchase more locally. Clearly, there are other more explicit cost factors that influence the purchasing decision. Unfortunately, the survey instrument did not elicit explicit cost and price information beyond what is included in the empirical estimation. Therefore, we cannot determine definitively whether the transaction costs outweigh other more direct costs.

In terms of the community variables, community attachment is positive and significant for feed supplements and farm supplies. Thus, for those two inputs we can accept our hypothesis (H1) that farmers more attached to their communities will be more likely to shop locally. The finding with respect to farm supplies is perhaps the most important in that this is an item available uniformly in all the study towns as well surrounding towns and, is often available at multiple outlets. The results therefore suggest that community attachment matters if farmers have a lot of choice regarding where they purchase their inputs.

The community dummy variables are significant in all of the regressions except corn grain and custom harvesting. Significantly more is purchased locally in Athens for soybeans, and significantly less for veterinarian services, milk equipment, farm machinery, and farm supplies. Significantly less feed supplements and farm supplies are purchased locally in Chilton. Given that so few of the other variables are significant, these results suggest that community characteristics may be more important than farm characteristics in determining farm purchasing patterns.

Table 5 shows the results from the second set of regressions, where the number of businesses variable has replaced the Athens and Chilton dummy variables. The custom harvesting regression was dropped from the analysis due to insufficient variation in the data (it was not sold by any local businesses in any of the three towns). The first noticeable result is that the change in community variable measures does not significantly change the inference about the importance of any of the other parameters, with the exception of local groceries. That variable is no longer significant in the milk equipment regression (and the sign changes). For all other variables and regressions, our results seem to be robust, suggesting that only a few nonsupply side parameters significantly impact local purchasing decisions.

The second key finding is that for soybeans, feed supplements, milk equipment, and farm machinery, having access to more outlets locally leads to higher local purchases, which supports our second hypothesis (H2). However, this finding is contradicted for farm supplies, where number of businesses is negative and significant, and for corn grain and veterinarian services, where number of businesses is not significant.

For corn grain, the community dummy variables in the first set of regressions (table 4) were also not significant, but in the case of veterinarian services, Athens was significant and negative. This suggests that a combination of community attributes in Athens, including perhaps transportation and marketing infrastructure, Athens' competitive advantage over neighboring towns, and the spatial distribution of business locations, creates more significant disincentives for purchasing veterinary services locally than the number of businesses alone. Another factor that the dummy variable Athens may account for is that no veterinarian businesses in Athens are structured as a co-op while in Richland Center, two-thirds of the veterinarian services are cooperatives. Thus, we would expect farmers in Athens to be less likely to use the local veterinarians.

However, the cooperative factor does not seem to explain why the significant community dummy variables are negative for all of the remaining inputs (the only significant and positive community dummy variable is for soybean and protein feeds). None of the towns have any cooperatives that sell milk equipment or farm machinery and both Athens and Richland Center have a single cooperative that sells farm supplies while Chilton has none. In addition, for soybean and protein feed, half of all the firms that sell the feed are co-ops in each of the three towns. Thus, this factor does not explain the difference in the significance of the Athens versus Chilton variable. Therefore, as with our other results, the cooperative structure in a town seems to only play a role in certain inputs and is not an overall determinant of purchasing patterns.

Conclusion

By investigating the more typical farm characteristics, such as farm size, alongside community supply and community attachment, our analysis shows the complexity of the economic relationship between farms and rural communities. Our findings show that purchasing patterns are commodity specific and are not universally driven by farm size or other farm-level characteristics. Therefore, our findings suggest that the premise that small farms purchase more locally is not the general case.

In addition, our work demonstrates the importance of adequately accounting for the supply side in analyzing purchasing patterns in small towns. We find that the number of local marketing outlets offered positively influences the decision by farms to purchase inputs locally and that this effect is more important than farm-level and other community characteristics (attachment and whether the local firms are cooperatives) for most farm inputs. Attachment to a community seems to most affect spending patterns when there is a large choice available to consumers (e.g., in the case of farm supplies). It should be noted, however, that local markets might be endogenously determined by the farm structure itself. Analyzing the long-term relationship between farm sizes and the types of businesses in a community using cross-section and time-series data would be a productive avenue for future research.

Taken together, the results of this paper suggest that the policy debate on the relationship between farms and main street businesses has focused too much on farm-level characteristics and not enough on the local business climate. Rather than concentrating solely on the economic impact of small versus large farms, today's policy makers need to understand the complex interaction between farms and communities that have already lost essential businesses. Policy makers interested in preserving the rural economy may want to direct their attention to policies that will maintain farm production, irrespective of the size of the operation, and promote the creation or retention of businesses that support farming.

A number of other important questions related to the issues addressed in this paper remain to be answered. Although we have measured some of the input purchase choices of farmers, we have ignored explicit cost and price factors as well as the labor market effects of different scales of farm. The economic effects of differences in labor usage across farm structures are an important issue for future research. In addition, this work has shown how community attachment influences some spending patterns, but is not able to disentangle the various elements of attachment and how they might relate to spending patterns. Future research could help describe the determinants of community attachment as well as how attachment affects economic activities in small towns.