What drives voluntary adoption of farming practices that can abate nutrient pollution?

Materials and Methods

Survey Methods. To study corn (Zea mays L.) and soybean (Glycine max [L.] Merr.) farmers' adoption of conservation and precision agriculture practices, our research focused on four states in the eastern Corn Belt: (1) Illinois, (2) Indiana, (3) Ohio, and (4) Michigan. The data for our study came from a mail survey of farmers who were the main decision makers on their farm. This questionnaire, the 2017 Crop Management and Stewardship Practices survey, was developed at Michigan State University and cosponsored by researchers at Purdue University and The Ohio State University. The survey was mailed in February of 2017, and farmer responses were accepted through April of 2017.

Crop farmers were included in the sampling frame if they had planted at least 100 ac (40 ha) of corn or soybean in 2016 and resided in a county where at least 15% of total land area was planted in corn or soybean. Figure 1 illustrates the counties sampled in these four states. Addresses for 10,582 farmers across the region were purchased from Farm Market iD, an agricultural data firm. Using county-level operation counts from the USDA National Agricultural Statistics Service 2012 Census of Agriculture as the assumed population, we created one stratum at the state level, and two substrata within each state, covering farms above and below 500 planted ac (202 ha). We intentionally oversampled large-acreage farmers in order to capture how most land is managed in these four states and in anticipation of lower response rates among this group. Of the 10,582 addresses contacted, 3,263 responded, for a 30.8% response rate. Of these, 1,129 were complete records, suitable for our analysis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

States and counties sampled.

Characterizing Examined Practices. This research explores what factors drive the adoption of eight conservation and precision agriculture technologies, selected to represent two broad types of tools that farmers can use to reduce nutrient loss. First, conservation technologies keep nutrients in place. Second, precision agriculture technologies use information and variable rate application to reduce the application of nutrients where they are not needed (and augment them where they are). We further distinguish within precision agriculture technologies by distinguishing between applicative and diagnostic technologies (Finger et al. 2019). Applicative technologies adjust management actions dynamically based on field or subfield level information. Diagnostic technologies collect field or subfield level information that farmers can use to inform future management decisions.

This study examines a single conservation technology, cover crops, which protects the soil when a main season crop is not holding it in place. Cover crops offer clear environmental benefits by reducing soil erosion and retaining soluble nutrients in the organic matter of the cover crop. While cover crops often lack product markets, they can be grazed or killed in place to build soil organic matter.

We examine the adoption of seven precision agriculture technologies, including three applicative technologies and four diagnostic ones. The three variable rate (VR) application technologies manage fertilizer and seed inputs. Variable rate phosphorus and potassium (VR-P/K) and variable rate nitrogen (VR-N) allow spatial adjustment of nutrient application to meet crop needs. Variable rate seeding (VR-Seeding) allows crop seed planting according to the expected crop yield potential. Variable rate application of nutrients helps to prevent excess application of nutrients where they are not needed, thereby reducing nutrient pollution of surface and ground water. Variable rate seeding can affect nutrient pollution more indirectly by enabling denser planting (especially of corn) where field conditions support higher yields and sparser planting where yield potential is lower.

The four diagnostic technologies examined are more varied in how they function. Pre-sidedress NO₃⁻ testing (PSNT) entails the just-in-time evaluation of N needs for corn. Unlike the spatial precision agriculture technologies, this diagnostic technology is temporally specific. By providing the farmer with updated soil NO₃⁻ information, it can deter the overapplication of N fertilizer. Soil nutrient mapping builds maps from soil trait and test information. Those maps can then guide VR fertilizer applications.

Aerial scouting is a diagnostic practice involving the use of imagery gathered from above. Such imagery can capture the vegetative vigor across a farm field during the growing season, thereby informing water, nutrient, pest, and disease management decisions with spatial and temporal precision. In recent years, aerial scouting has begun spreading rapidly, as unmanned aerial vehicles (drones) (1) lower the cost of imagery acquisition, relative to piloted airplanes, and (2) improve spatial precision relative to satellite images.

The last diagnostic technology, yield mapping, uses yield monitors to make a spatial record of crop yields. Yield maps can help inform farmers' planting and field management decisions in the next growing season, potentially reducing nutrient loss via spatial management of fertilizers and seeding rates.

Relevant Drivers and Hypothesized Effects. Drawing upon the categories of explanatory variables identified above, we present a set of testable hypotheses about how the following four categories of major drivers impact farmer decisions on the adoption of practices that abate nutrient losses from crop fields: (1) farmer objectives, (2) farm household traits, (3) farm resources, and (4) information access. For clarity, we state the hypotheses in affirmative form rather than the rejectable, null form preferred by statisticians. We separate our hypotheses by category of the technology, including (1) conservation practices that keep nutrients in place (e.g., cover crops), and (2) precision agriculture practices that target nutrients for resource-efficient, optimal crop growth. As noted above, we subdivide the precision agriculture practices into two groups: (1) applicative precision technologies (VR-P/K, VR-N, and VR-Seeding), and (2) diagnostic precision technologies (e.g., PSNT, soil nutrient mapping, aerial scouting, and yield mapping) (Finger et al. 2019).

Farmer objectives, as identified through stated attitudes, are expected to be strong drivers of adoption choices (table 1). We hypothesize that income-oriented farmers and those motivated by social status will adopt precision agriculture technologies. We expect environmentally oriented farmers to adopt conservation technologies.

Among farmer and farm household traits (table 1), literature leads us to hypothesize that older farmers will be less inclined to adopt either type of technology. By contrast, we expect that more educated farmers will adopt both conservation and precision agriculture practices. Because cover cropping entails additional tasks and hence demands more labor than the precision technologies, we expect that larger families with more available labor will be more likely to adopt conservation technologies. As crop insurance enrollment mitigates risk of financial loss for farmers, we expect that farmers who have insured a larger share of their cropped acreage will be more inclined to adopt precision technologies because information can increase a farmer's awareness of his or her operation's vulnerability to financial loss (and these are information-intensive technologies). Finally, because conservation subsidy programs reduce the cost of adopting conservation practices, we expect that participation in a working lands program will favor adoption of conservation practices, but that participation in a set-aside conservation program will have no effect.

Farm resources (table 1) start with an indicator of farm size, which serves as our proxy for farm wealth. We measure farm size as the sum of owned and rented acres, since larger farms tend to own big equipment and rent more land (the Pearson correlation coefficient between owned and rented land in our survey was +0.15). We hypothesize that farms with more cropped acreage will be more inclined to adopt an indivisible technology like those embodied in precision agriculture equipment (notably yield monitors and variable rate applicators). We expect that farmers with a long time horizon, as indicated by owning a large share of the cropland they operate or by the intent to bequeath the farm to heirs, will be more likely to adopt conservation practices that enhance soil quality. Finally, regarding compatibility of agricultural systems with new technologies, we expect that cover cropping will be attractive to livestock farmers as a source of forage. By contrast, we expect that livestock farmers will be disinclined to invest money and time in precision cropping practices (though they would likely take an interest in precision livestock practices, had we measured any of these). We define livestock farmers as farmers for whom (1) milk and dairy products, and/or (2) livestock and animal products other than dairy, account for more than 10% of farm revenues.

Our literature review indicates that an additional category of drivers, information, can play an important role in practice and technology adoption decisions. Information availability can have hard-to-predict effects for at least two reasons. First, the effect of information on adoption decisions depends on prior information available to the decision maker—a prior state that we do not observe with our cross-sectional survey data. Second, decision makers choose their information sources, so the information effect is endogenous to the decision to seek it out. In the data set explored here, we lack information on stated awareness or familiarity with examined practices beyond the level of adoption. Because the potential role of information as a driver of adoption is complex and dynamic, we include a series of variables measuring the frequency with which farmers consult each of nine distinct information sources as controls for information access without hypothesizing about their expected effects on the adoption of specific practices (table 1). The nine information sources are (1) university extension, (2) university faculty, (3) chemical input dealers, (4) seed input dealers, (5) independent crop consultants, (6) other farmers, (7) grower associations, (8) the Web, and (9) print sources. Frequency of access is measured on a five-point scale ranging from never (1) to once a day (5).

Key Explanatory Variables. Table 2 presents summary statistics for the key independent variables. The survey sample was stratified by state and farm size, which entailed oversampling important groups with historically low response rates, such as large farms. In order to adjust the statistical results to account for different probabilities of subpopulations being sampled, we use probability weights in order to permit extrapolation from the stratified sample to the broader population of corn and soybean farms with over 100 ac (40 ha) in the states of Illinois, Indiana, Michigan, and Ohio. Table 3 details how the inverse sampling probability weights were calculated based on the stratum-level response sample and the associated stratum population. All statistical results presented here are computed using these probability weights. For interested readers, we make unweighted results from the ordered probit analysis available in the supplemental material for this article, which can be found online at http://dx.doi.org/10.22004/ag.econ.302418 (Solon et al. 2015).

Factor Analysis of Farmer Attitudes. We measured farmer objectives indirectly via latent attitude factors that were constructed from levels of agreement with Likert-scaled statements. We used a confirmatory factor analysis to characterize the latent attitudes that might influence a farmer's decision to adopt a certain practice. There were 11 statements regarding respondent attitudes toward social status, income, and environmental stewardship. As these variables were measured on a 5-point Likert scale, many were correlated with one another. Through factor analysis, we reduced the 11 Likert variables to the following three attitudinal factors: (1) preference for income, (2) preference for high social status, and (3) preference for environmental amenities, respectively.

Table 4 presents the factors and analogous factor loadings to these 11 statements. These loadings are used to generate attitudinal scores for each farmer that, by construction, have mean zero with unit variance. The 3 attitudinal factors are described as follows:

• The “Income” factor experiences high loadings on statements about optimizing income;

• The “Social status” factor sees high loadings on statements regarding preference for a higher social status in their community;

• The “Environmental” factor has high loadings on preferences for environmental amenities and stewardship.

In the following analyses of technology adoption, these factors are included to represent the farmer objectives inherent in their latent attitudes.

Ordered Probit Analysis. In order to test hypotheses statistically, we employed an ordered probit analysis to assess the factors driving adoption of the eight conservation and precision agriculture practices. The state of adoption of each practice, the dependent variable in each ordered probit model, was self-reported by the farmer at one of four levels: (1) never used, don't want to; (2) never used, might; (3) use sometimes; or (4) use regularly. We control for state-level fixed effects by including dummy variables for Illinois, Indiana, and Ohio, using Michigan farmers as the base group. We also weight by stratum by using the probability weights described in table 3. The ordered probit models were estimated via maximum likelihood. The estimated coefficients are not easily converted to marginal effects, so we draw inferences from their sign and significance, rather than from their magnitudes.