2. Literature Review

The large majority of previous studies of no-till adoption focused on the use of no-till within a single year (see Bergtold and Molnar, Reference Bergtold and Molnar2010; Kurkalova, Kling, and Zhao, 2006; Lambert et al., Reference Lambert, Sullivan, Claassen and Foreman2007; Mezzatesta, Newburn, and Woodward, Reference Mezzatesta, Newburn and Woodard2013; Pautsch et al., Reference Pautsch, Kurkalova, Babcock and Kling2001; Soule, Reference Soule2001; Soule, Tegene, and Wiebe, Reference Soule, Tegene and Wiebe2000; Wade, Kurkalova, and Secchi, Reference Wade, Kurkalova and Secchi2016; Yang, Sheng, and Voroney, Reference Yang, Sheng and Voroney2005). As noted by Prokopy et al. (Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008), very few studies have focused on sustained adoption of conservation practices (not just tillage practices) over time. Although these studies have provided a great deal of information on the determinants of no-till adoption, they do not provide insight on the extent of continuous adoption or how these determinants may affect producer decisions to sustain adoption over time.

A smaller group of studies used panel data to assess no-till adoption over time. Many of these studies, however, used data that were aggregated to the county or state level. Ding, Schoengold, and Tadesse (Reference Ding, Schoengold and Tadesse2009) estimated the impact of droughts and floods on no-till adoption using county-level data for Iowa, Nebraska, and South Dakota. Schoengold, Ding, and Headlee (Reference Schoengold, Ding and Headlee2015) used a similar data set to estimate the joint impact of weather and government programs on conservation tillage use. Fernandez-Cornejo et al. (Reference Fernandez-Cornejo, Hallahan, Nehring, Wechsler and Grube2012) also used county-level data to examine the effect of herbicide-tolerant soybean adoption on conservation tillage use in major soybean-producing states. In aggregated data, information on sustained adoption for individual fields is lost.

To our knowledge, only one study addresses no-till decisions at the field or farm level, over time, using data on revealed preferences. Recently, Perry, Moschini, and Hennessy (Reference Perry, Moschini and Hennessy2016) used farm-level panel data to examine complementarity between conservation tillage and herbicide-tolerant crops but did not address the question of continuous adoption.

Agricultural economists have learned that to fully analyze environmental issues they need to understand and incorporate biophysical characteristics (Lichtenberg et al., Reference Lichtenberg, Shortle, Wilen and Zilberman2010). In a national study of the effect of land tenure on conservation tillage adoption, Soule, Tegene, and Wiebe (Reference Soule, Tegene and Wiebe2000) found that highly erodible land (HEL) designation had a positive and significant effect on conservation tillage adoption, and inherently wet soil had a negative and significant effect. Schoengold, Ding, and Headlee (Reference Schoengold, Ding and Headlee2015) estimated the effect of climate and soil properties on no-till adoption in Iowa, Nebraska, and South Dakota and also found that HEL designation had a positive effect on adoption. In estimating the cost of carbon sequestration in Iowa soils, Pautsch et al. (Reference Pautsch, Kurkalova, Babcock and Kling2001) found land slope (a field characteristic that is highly correlated with HEL) to be positively correlated with conservation tillage adoption. Wade, Kurkalova, and Secchi (Reference Wade, Kurkalova and Secchi2016) and Kurkalova, Kling, and Zhao (2006) both estimated the cost of conservation tillage adoption in Iowa and found that subsidies required for farmers to use conservation tillage were lower on HEL than on non-HEL. Many studies use proxies for soil productivity: Yang, Shen, and Voroney (Reference Yang, Sheng and Voroney2005) used soil capability class (and aggregate yield data) to interpolate missing field-level yield data in the Fairchild Creek watershed in Ontario, Canada; Secchi et al. (Reference Secchi, Gassman, Williams and Babcock2009) used Iowa's corn suitability rating to estimate field-level yields; and Kurkalova, Kling, and Zhao (2006) used the corn suitability rating to estimate net returns to conventional tillage.

No-till generally is less profitable on cold wet soil (Soule, Tegene, and Wiebe, Reference Soule, Tegene and Wiebe2000); therefore, climate regressors are often used in no-till analysis. For example, in studies of conservation tillage adoption in the Upper Mississippi River basin and the Canadian prairies, Wu et al. (Reference Wu, Adams, Kling and Tanaka2004) and Davey and Furtan (Reference Davey and Furtan2008), respectively, found negative and significant effects for precipitation and positive and significant effects for temperature, whereas others like Pautsch et al. (Reference Pautsch, Kurkalova, Babcock and Kling2001) examined conservation tillage adoption in Iowa and found the opposite effects for both variables. Wu et al. (Reference Wu, Adams, Kling and Tanaka2004) also found variation in precipitation to be positively correlated with conservation tillage adoption, whereas Pautsch et al. (Reference Pautsch, Kurkalova, Babcock and Kling2001) had the opposite findings. Kurkalova, Kling, and Zhao (2006) and Wade, Kurkalova, and Secchi (Reference Wade, Kurkalova and Secchi2016) found that temperature had positive and significant effects on conservation tillage adoption. Precipitation is widely used in the literature but is often found to have an insignificant effect on conservation tillage adoption.

Many studies attribute conservation tillage use to farm household and farm characteristics. In a national study of the farm households that adopt best management practices, Lambert et al. (Reference Lambert, Sullivan, Claassen and Foreman2007) found farm size and education to be positively correlated with adoption of a series of management practices including conservation tillage. Soule, Tegene, and Wiebe (Reference Soule, Tegene and Wiebe2000) also found positive and significant effects for farm size and college-educated operators and a negative and significant effect for operator's age for conservation tillage adoption. Soule (Reference Soule2001) examined the effect of education on conservation tillage adoption and also found a positive effect. Baumgart-Getz, Prokopy, and Floress (Reference Baumgart-Getz, Prokopy and Floress2012) conducted a meta-analysis assessing the attributes affecting best management practices (including no-till) and also found that age had a negative effect whereas farm size had a positive effect on adoption. Davey and Furtan (Reference Davey and Furtan2008) found a similar effect for operator age on conservation tillage. Land tenure was also examined widely in the literature: Baumgart-Getz, Prokopy, and Floress (Reference Baumgart-Getz, Prokopy and Floress2012) include several studies that incorporate conservation tillage and land tenure, and Soule (Reference Soule2001) and Davey and Furtan (Reference Davey and Furtan2008) each found that renters were less likely to adopt conservation tillage. Studies that use ARMS data in conservation analysis also often include the Economic Research Service's (ERS) farm typology. Farm typology describes farms in terms of size, sales, and operator lifestyle, all of which can influence a farmer's adoption decision. Soule (Reference Soule2001) examined conservation tillage adoption by farm typology and found that high-sales farms are more likely to adopt conservation tillage.

Tillage is typically used for weed control (Triplett and Dick, Reference Triplett and Dick2008). Operators may choose to alternate between tillage and no-till if cropland becomes resistant to weeds making periodic tilling necessary on some fields (Wilman, Reference Wilman2011). Use of no-till in conjunction with other best management practices, for example, rotating crops or cover crop use, can also help manage weeds. Farmers’ tillage choice could also be a function of the crop grown. Many farmers think no-till is less compatible with corn than other crops and associate using it with a yield penalty on corn but have no such perception for soybean or wheat (crops typically included in corn rotations) (Reimer, Weinkauf, and Prokopy, Reference Reimer, Weinkauf and Prokopy2012). Such perceptions are not unreasonable because no-till yields vary depending on crop rotation, region, climate, and soil type (Ogle, Swan, and Paustian, Reference Ogle, Swan and Paustian2012).

This study is the first to use data on tillage history and current tillage practices to distinguish farms that use no-till continuously from those that alternate no-till with other tillage practices. Revealed preference data are used in an ordered logit regression analysis to determine the effect of land characteristics, climate, farm characteristics, and producer demographics on producer choice among tillage regimes. Our primary objective is to develop a richer understanding of producer no-till use decisions. Although the model cannot be used to estimate the difference in profit across the three tillage regimes, it adds to the literature by providing insight into the characteristics and conditions that are conducive to each tillage regime.

3. The Model and Estimation Procedure

Farmer i is assumed to make tillage choices that maximize utility. We express utility as an unobserved difference between the utility for no-till and other tillage practices; that is,

(1) $$\begin{equation} y_{is}^* = U_{is}^{nt} - U_{is}^t{\rm{\ for\ }}s{\rm{\ }} = {\rm{\ }}0, \ldots ,{\rm{\ }}S. \end{equation}$$

Here, s is the period in which the tillage choice is observed (s = 0 indicates the initial period); y* _is is the unobserved net utility of choosing no-till over tillage; U is a farmer's utility; nt indicates no-till use; t indicates tillage use; and i = 1,. . ., n indexes individual farmers. Equation (1) can be written as

(2) $$\begin{equation} y_{is}^* = {{\bm \beta '}}{\bm{X}_i} + {\varepsilon _i}, \end{equation}$$

where the product β ^ʹ X _i represents the portion of net utility observed by researchers; X _i represents a vector of coregressors thought to affect sustained no-till adoption such as soil, farm, climate, regional, and operator characteristics; β represents a vector of unknown parameters; and ε_i is the random variable representing the producers’ unobserved preference for no-till. However, we observe only the tillage decision in period s:

$$\begin{equation*} y_{is} = 1\,\, \hbox{when}\,\, y^*_{is} > 0; y_{is} = 0\,\, \hbox{otherwise}. \end{equation*}$$

This implies that the farmer will choose either to till or to use no-till in period s. In a multiyear setting, we argue that there is unobserved random variation in profitability of fields and this variation affects whether farmers use the same tillage practice each year or alternate practices. Random shocks may affect yields negatively (and include drought in the previous year, excessive precipitation in the previous year [there may be combine ruts], heavy residue from the previous crop, emergence of herbicide-resistant weeds, machine failures that delay planting, etc.), or shocks could be positive and result from particularly favorable weather, like warm springs and adequate and well-timed precipitation. Over time, these positive and negative shocks likely balance to zero, and experienced farmers can closely predict yields and choose tillage sequences.

For farmers who till continuously, we theorize that the underlying overall relative profitability of no-till must be below a no-till threshold value, τ ₁, to ensure that it is always more profitable than no-till:

$$\begin{equation*} {{\bm \beta '}}{\bm{X}}_i + {\varepsilon _i} < {\tau _1} < 0. \end{equation*}$$

Likewise, farmers who use no-till every year must exceed a no-till threshold value, τ ₂, to ensure that no-till is at least as profitable as tillage:

$$\begin{equation*} 0 < {\tau _2} \le {{{\bm \beta }}^{\bm{'}}}{\bm{X}_i} + {\varepsilon _i}. \end{equation*}$$

Farmers who alternate practices will fall between the thresholds:

$$\begin{equation*} {\tau _1} \le {\bm \beta }^{\bm{'}}{\bm{X}_i} + {\varepsilon _i} < {\tau _2}. \end{equation*}$$

We denote farmer i’s multiyear tillage choice as T_i = j, where j = 1, 2, 3 corresponds to CT, ANT, and CNT, respectively, and assume that the farmer's utility (or overall measure of no-till preference), y* _i , is a random function,

(3) $$\begin{equation} y_i^* = {{\bm \beta '}}{\bm{X}_i} + {\varepsilon _i}, \end{equation}$$

where y* _i is the random utility over S periods.

This discrete and ordered dependent variable (T_i ) and random utility model lend themselves to an ordered regression model. If the random portion of utility follows a logistic distribution, model parameters can be estimated using an ordered logit model. We define

T_i = j if τ _{j − 1} < y* _i ⩽ τ _j , for j = 1, 2, 3, where τ is a vector of unknown thresholds to be estimated by the model, τ ₀ = −∞, and τ ₃ = ∞. Then the probability that a farmer's utility falls within category j is

(4) $$\begin{equation} \begin{array}{@{}*{2}{l}@{}} {\Pr \left[ {{T_i} = j} \right]}&{ = \Pr \left[ {{\tau _{j - 1}} < y_i^* \le {\tau _j}} \right]}\\ \,&{{\rm{ = }}\Pr \left[ {{\tau _{j - 1}} < {\bm \beta '}{\bm{X}_i} + {\varepsilon _i} \le {\tau _j}} \right]}\\ \,&{{\rm{ = }}\Pr \left[ {{\tau _{j - 1}} - {\bm \beta '}{\bm{X}_i} < {\varepsilon _i} \le {\tau _j} - {\bm \beta '}{\bm{X}_i}} \right]}\\ \,&{{\rm{ = }}\Lambda \left( {{\tau _j} - {\bm \beta '}{\bm{X}_i}} \right) - \Lambda \left( {{\tau _{j - 1}} - {\bm \beta '}{\bm{X}_i}} \right),} \end{array} \end{equation}$$

where Λ(·) is the cumulative distribution function of the logistic distribution (Cameron and Trivedi, Reference Cameron and Trivedi2005). The ordered logit model was used to evaluate the factors that influence sustained no-till adoption. Parameters τ and β are estimated using the maximum likelihood procedure with weighted standard errors.

For this initial analysis of CNT, we estimate a reduced form model. There is evidence suggesting that tillage decisions are made in conjunction with crop choice; for example, some farmers use no-till in soybeans (no-till is used on more than 45% of soybean acres) but till when producing corn (no-till is used on less than 25% of corn acres) (Horowitz, Ebel, and Udea, Reference Horowitz, Ebel and Ueda2010). So the estimated effects of soil, topography, climate, and other factors may affect no-till adoption directly or indirectly through crop choice. Of course, the availability of no-till may also affect the crop. For example, no-till may significantly reduce soil erosion when growing low-residue crops (such as soybeans) on highly erodible cropland. Nonetheless, we focus on the effects of soils, climate, and farm characteristics.

4. Data Description and Variable Construction

Data on tillage, field, farm, and farmer characteristics are from the ARMS. The Production Practices and Costs Report (PPCR) is a field-level, commodity-specific survey of production practices. Each survey covers a subset of states accounting for 90%–95% of the production for the surveyed crop. The Costs and Returns Report (CRR) is a farm-level, follow-on survey that provides data on farm and operator characteristics. All data are weighted to ensure that acres sum to state totals. The 2010 corn and 2012 soybean surveys provide usable data on 1,620 and 1,860 fields, respectively (USDA-ERS, 2017).

The dependent variable has three categories identifying no-till intensity over a 4-year period: CT (0 years of no-till), ANT (1–3 years of no-till), and CNT. For fields in ANT, 28%, 35%, and 37% used no-till once, twice, and three times, respectively. No-till in the survey year is indicated by an absence of tillage operations in the spring (as listed by the respondent) and the previous fall as recalled by the respondent. For prior years, we use the same definition of no-till, although no-till during both the spring and the previous fall are based on producer recall. We believe this definition (Uri, Reference Uri2000) is aligned with how farmers interpret no-till adoption (i.e., they are not tilling) and therefore is more consistent with the tillage history data, which are based on the respondent's recollection of the previous 3 years of tillage practices. Fallow fields (1.6% of acres) and fields in perennial crops (3.4%) are excluded from the analysis because farmers do not have to make tillage decisions on these fields. The dependent variable is a quantitative measure of farmers’ preferences/attitudes toward CNT making the ordered logit regression model appropriate.

4.4. Operator Characteristics

Proxies for farmer experience are used widely with mixed results (see Knowler and Bradshaw, Reference Knowler and Bradshaw2007; Prokopy et al., Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008; Wu and Babcock, Reference Wu and Babcock1998). Here, age proxies for farmer experience. Studies have also postulated that younger farmers are more likely to adopt conservation practices because their longer farming horizons afford them time to recapture the cost of new equipment and training (Baumgart-Getz, Prokopy, and Floress, Reference Baumgart-Getz, Prokopy and Floress2012; Lambert et al., Reference Lambert, Sullivan, Claassen and Foreman2007). Wu and Babcock (Reference Wu and Babcock1998) theorize that this negative correlation with conservation tillage adoption is because of long farming tradition and cultural practices.

Like Lambert et al. (Reference Lambert, Sullivan, Claassen and Foreman2007) we hypothesize that full owners, who have a long-term interest in soil health and profitability, are more likely to adopt practices that help improve long-term soil health, including CNT. We note, however, that tenure is used frequently to explain conservation practice adoption with varying results (Prokopy et al., Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008).

College education is also often associated with conservation practices adoption (e.g., D'Emden, Llewellyn, and Burton, Reference D'Emden, Llewellyn and Burton2008; Gould, Saupe, and Klemme, Reference Gould, Saupe and Kleeme1989; Prokopy et al., Reference Prokopy, Floress, Klotthor-Weinkauf and Baumgart-Getz2008; Rahm and Huffman, Reference Rahm and Huffman1984; Soule, Tegene, and Weibe, Reference Soule, Tegene and Wiebe2000; Wu and Babcock, Reference Wu and Babcock1998).

Though not examined directly in this analysis, cropping patterns may play an important role in the tillage decision. Some authors find that farmers alternate tillage with specific crops (Robertson et al., Reference Robertson, Gross, Hamilton, Landis, Schmidt, Snapp and Swinton2014). During the 3-year cropping history, 5% of sampled corn acres and 7% of sampled soybean acres were planted in crops other than corn or soybeans (with farmers in the soybean sample planting soybeans on 86% of cropland and farmers in the corn sample planting soybeans on 69% of cropland). Because cropping history is endogenous, we opt to estimate a reduced-form model with only exogenous variables that change slowly over time (e.g., farm size, age, and education).

The final samples have 1,155 observations for corn and 1,324 for soybeans representing 85 million acres. The average field size is 52 acres. Table 1 summarizes the descriptions and provides descriptive statistics of the common observations. All statistics are weighted by state crop acres.

Table 1. Descriptions of Variables and Summary Statistics