Potato soil health economics

Literature exploring the relationship between soil health indicators or practices and economic outcomes in potato production systems is very limited. Hills et al. (Reference Hills, Collins, Yorgey, McGuire and Kruger2020) provide a comprehensive review of existing literature on soil health management practices in potato systems in the Pacific Northwest region of the United States. Specifically, they review the literature on reducing tillage, adjusting crop rotation length and rotation crops, reducing fumigation, and using cover crops, green manures, and organic amendments. The majority of the articles included in their review focus on addressing soilborne disease risk through crop rotation adjustments and green manures. Few of these studies contain economic analysis and those that do suggest results are inconclusive. Hills et al. (Reference Hills, Collins, Yorgey, McGuire and Kruger2020, p. 17) write: ‘Though the literature underscores a general understanding of soil management practices, the specific circumstances under which implementing these practices in potato production imparts better yield and quality, and economic gain, often is unclear’.

In one of the few potato-specific soil health economics studies, Khakbazan et al. (Reference Khakbazan, Larney, Huang, Dilay, Mohr, Pearson and Blackshaw2016) found that potato yields were higher in potato systems that included reduced tillage, cover crops, and compost but that the increased yield did not make up for higher costs—net income was lower in the systems with conservation practices. A related group of coauthors has made several assessments of potato rotation duration. Khakbazan et al. (Reference Khakbazan, Mohr, Volkmar, Tomasiewicz, Moulin, Derksen, Irvine, McLaren and Monreal2010) found that while 2-year potato rotations in experimental plots in Manitoba were financially preferable over the time period examined, 3-year rotations fell closely behind, given the higher revenues in potato years in the longer rotation. The article suggests that 3-year rotations could be competitive under the right circumstances or potentially, over time as soil health benefits accrue. In a related article. Khakbazan et al. (Reference Khakbazan, Mohr, Huang, Campbell, Volkmar, Tomasiewicz, Moulin, Derksen, Irvine, McLaren and Nelson2018) recommend 3- or 4-year potato rotations over 2-year rotations because shorter rotations can increase plant disease and therefore decrease in economic returns over the long run.

Larkin and Halloran (Reference Larkin and Halloran2014) compared growing barley in rotation with potatoes to disease-suppressive crops (mustard, rapeseed, and sudangrass), either used as a cover crop, green manure, or harvested with residue incorporated. They found that green manure management provided the largest increases in yield. However, when evaluating economic outcomes, the use of mustard or rapeseed as a harvested crop provided the best overall economic returns, surpassing barley, the comparison crop.

Soil health practices and yields

Yields are a critical component of farm economics, although costs, prices, and crop quality also determine profitability. While we focus on evaluating a set of soil health practice economics in potato systems, we are informed by literature on soil health practices and yields in other field crops.

Findings from literature linking soil health practices and crop yields are generally mixed. An article by Bowles et al. (Reference Bowles, Mooshammer, Socolar, Caldero´n, Cavigelli, Culman, Deen, Drury, Garcia, Gaudin, Harkcom, Lehman, Osborne, Robertson, Salerno, Schmer, Strock and Grandy2020) analyzed long-term experiment data from across the United States and Canada. They found that more diverse corn rotations were associated with higher corn yields, yield gains over time, and higher yields during adverse weather such as drought. Gaudin et al. (Reference Gaudin, Tolhurst, Ker, Janovicek, Tortora, Martin and Deen2015) evaluated long-term rotations in Ontario, finding that greater diversity improves yield stability over time. However, Deines et al. (Reference Deines, Guan, Lopez, Zhou, White, Wang and Lobell2023) found that cover crop adoption was associated with small yield decreases in corn and soybean farms. Those yield losses were also correlated with several additional factors including growing season temperatures and rainfall, as well as underlying soil health indicators. (Note that findings regarding rainfall effects from other crops are less applicable to potatoes in the United States because most production is irrigated.)

Other studies examining the relationship between soil health practices and yield outcomes offer inconclusive, widely variable, or heavily qualified results. Bowman, Poley, and McFarland (Reference Bowman, Poley and McFarland2022)) found wide variation in yields when comparing Midwest farms using cover crops to those that did not, but differences between the two groups were not statistically significant. Munkholm, Heck, and Deen (Reference Munkholm, Heck and Deen2013) found that soil structure and yield were correlated, but that unless a diverse crop rotation was employed, no-till was associated with poor soil structure. In a meta-analysis, Marcillo and Miguez (Reference Marcillo and Miguez2017) found a range of yield responses in corn following winter cover crops, ranging from no change to positive change in the United States and Canada. In another meta-analysis, Tonitto, David, and Drinkwater (Reference Tonitto, David and Drinkwater2006) compared yields when conventional nitrogen fertilizers were replaced with cover crops and calculated a roughly 10% yield reduction when comparing legume cover crop fertilization with conventional nitrogen fertilizer. They did not find a difference, however, when green manures were used, and the nitrogen content of their biomass exceeded a certain threshold (110 kg N/ha).

Fewer studies focus specifically on soil health practices and potato yields. However, Larkin and Tavantzis (Reference Larkin and Tavantzis2013) found that compost amendments substantially increased yield in trials in Maine. They also found that compost reduced Rhizoctonia stem canker but increased common scab in some years. Molina et al. (Reference Molina, Tenuta, El Hadrami, Buckley, Cavers and Daayf2014) also assessed compost application in potato systems and determined that compost amendments increased yield, as well as reduced disease. Moore et al. (Reference Moore, Olsen, Carey and Leytem2011) support these results, finding a large (20%) yield increase attributed to composted dairy manure in Idaho, after 3–4 years of applications. Wilson et al. (Reference Wilson, Zebarth, Burton, Goyer, Moreau and Dixon2019) examined compost application, nitrogen availability, and yields in potatoes. While they found differences between ‘mature’ and ‘immature’ compost products in resulting nitrogen availability, they concluded that compost amendments did not significantly influence potato yield.

Soil health practices and economics

Similar to the literature exploring the links between soil health practices and yield, existing economic findings in the literature are mixed and depend on many factors. Many studies with economic implications have focused on the use of cover crops. Bergtold et al. (Reference Bergtold, Ramsey, Maddy and Williams2019) provide a literature review of the economics of cover crop adoption by US farmers, focused on Kansas. This is also one of the few studies providing an overview of indirect costs associated with cover crop adoption, such as decreased water availability for the cash crop, nuisance, learning curves, and delayed emergence of the cash crop because of increased residue. Their overall findings were mixed. Additionally, they noted that the profitability of incorporating cover crops into crop rotations is affected by a farmer’s unique set of management choices and growing conditions.

Using a partial budgeting approach, Plastina et al. (Reference Plastina, Liu, Sawadgo, Miguez, Carlson and Marcillo2018) calculated lower profits resulting from cover crop adoption, unless the farmer participated in cost-share programs or livestock grazed the cover crops, providing an additional source of revenue. Using partial budgets, focus groups, and survey responses, Plastina et al. (Reference Plastina, Liu, Miguez and Carlson2020) estimated negative economic returns to cover crop adoption for most respondents but noted that participants had generally positive associations with cover crops. Thompson et al. (Reference Thompson, Armstrong, Roth, Ruffatti and Reeling2020) also found negative net returns to cover crops in corn and soybean rotations in experimental plots in Illinois. They used experimental plot data coupled with simulations but mentioned that either cost-share programs or feeding the crop to animals could increase returns. Similar to their findings regarding cover crop adoption and yields, Bowman, Poley, and McFarland (Reference Bowman, Poley and McFarland2022) found highly variable differences in profit when comparing corn and soybeans grown on fields that use cover crops with those that do not. They added that the variability in cover crop outcomes makes it difficult for farmers to develop confident expectations surrounding cover crop adoption.

Rejesus et al. (Reference Rejesus, Aglasan, Knight, Cavigelli, Dell, Lane and Hollinger2021) reviewed existing literature on the economics of both cover crop adoption and no- or reduced-tillage production systems. They found that most studies calculating the short-term return to planting cover crops did not find returns that surpass costs but stated that longer-term net returns are more likely positive. Their review of the literature on the economics of no-till production systems concluded that net returns are variable and depend on region, soil characteristics, and other factors.

Much remains uncertain when assessing the economic returns to practices meant to improve soil health, even with the more extensive body of work that has been created around soil health practice adoption in corn and soybean crops. Our research addresses some of the gaps in this literature by incorporating a comparative analysis of several soil health management practices in a tuber crop across a range of scenarios and locations.

Potato soil health project trial data

We incorporate data from several sources for our analysis. Our potato and rotation crop yield data are from the Potato Soil Health Project, a USDA Specialty Crop Research Initiative (SCRI) effort to assess soil health indicators and methods to improve soil health in potato cropping systems. To learn more about this project, visit https://potatosoilhealth.cfans.umn.edu.) As a part of that project, researchers in seven states conducted coordinated trials that incorporated soil health practices into potato rotations over 4 years, 2019–2022. The seven field sites were located in Minnesota (two sites),Colorado, Wisconsin, Oregon, Idaho, and Maine. Because agronomic conditions and predominant potato production practices vary regionally, researchers in each location designed a set of treatments specific to their area. Researchers in each state applied 12 treatments, made up of alternative combinations of soil health practices, potato varieties, rotation crops, and rotation lengths. In each location, researchers either completed four or five replications of each treatment.

Potato rotations included in the study were either two (potato-other crop-potato) or three (potato-other crop-other crop-potato) years in duration. The rotations were aligned so that both 2- and 3-year rotations would have two potato production years, and all locations and treatments produced potatoes in 2022. For each location, six of the treatments were in 3-year rotations and six were in 2-year rotations. In each location, researchers applied either manure or composted manure in some treatments. Minnesota and North Dakota applied turkey manure, while Idaho and Oregon used composted dairy cattle manure. All states except Idaho varied mustard biofumigation across treatments; Idaho had no treatments with mustard biofumigants. Some treatments also involved other practices, such as cover cropping.

Researchers in North Dakota administered field trials in northcentral Minnesota because a suitable field trial location in North Dakota was unavailable. They selected the location to best approximate North Dakota growing conditions. Researchers in Minnesota administered trials in central Minnesota. Aside from differences in climate and weather stemming from differing latitudes, the central Minnesota location (referred to hereafter as Minnesota or MN), had higher soilborne disease pressure than the northcentral Minnesota location (referred to hereafter as North Dakota or ND). Appendix Tables A1.1 to A1.4 provide a summary of treatments for each location in our analysis and Appendix Table A2 summarizes field trial locations.

We limit our analysis to four locations because of data availability. Some locations were unable to collect yield data in some of the years. Additionally, for certain types of potatoes predominant in some states, pricing data, which we require to estimate revenues, are not published. We are left with datasets adequate to support our analysis from researchers in Idaho, North Dakota, Minnesota, and Oregon.

Our economic analysis focuses on four soil health management practices: rotation duration, fumigation, biofumigation using mustard green manure, and application animal manures. In the case of fumigation and mustard biofumigation, the study sought to test rather than assume their substitutibility versus complementarity. Therefore, in states with both practices, some treatments contained one or the other practice, others included both (see full treatment descriptions in Appendix Tables A1.1–A1.4). All of these practices, with the exception of mustard biofumigation in Idaho, were varied within locations. The specifics of each type of treatment (e.g., type and rate of organic amendment applied) varied across locations to reflect realistic practices in the area. For example, manure rates depend largely on climate and soil type such that similar treatments are not identical in their application rates.

We further segment our dataset to address treatments that included Bannock Russet potatoes, which are more resistant to the soilborne disease Verticillium, than the Russet Norkotah and Russet Burbank potatoes that make up the remainder of our dataset. Bannock Russets were planted in two treatments in North Dakota and two treatments in Minnesota. Because of Bannock’s soilborne disease resistance, it likely responds differently to the soil health management practices being explored in this article. Additionally, because Bannocks were included in a few treatments, we have a very small sample size with which to estimate those responses. Further, no treatments that included Bannock fumigated before the 2022 potato growing season, nor did they plant mustard as a biofumigant in any trial. Treatments with Bannock also did not include any manure applications, so even if we had sufficient observations for statistical power, our study design does not allow us to estimate correlations with three out of four of the soil health practices we examine. (Note, however, that as robustness check, we provide a set of our main regression results that include Bannock potatoes in our main regressions in Appendix Table A7. Our main findings are unchanged.) We therefore include the observations with Bannocks in treatment-by-treatment summaries but remove them for analysis that pools observations over soil health practice type. After removing Bannock observations, we are left with 136 location × treatment × replication observations with data collected for all 4 years of the study and 196 observations for 2022 alone. The number of observations differs because some locations did not collect crop yields in the first year of the study. (Total observations including Bannock = 151 for all four study years and 214 for 2022 alone.)

Revenue data

Out of the seven locations where trials were conducted, four met our criteria for consistent economic analysis. Our main limitation was in obtaining appropriate potato pricing models, which we use to estimate revenues. Relatively detailed crop prices are readily available from the USDA National Agricultural Statistics Service for many crops. (For examples, see USDA National Agricultural Statistics Service (2023a).) However, potato price data collection efforts face several challenges. First, publicly available price data for some types of potatoes are entirely unavailable. Pricing and other details of supply contracts for chipping and several other types of potatoes are often kept confidential by those who hold them. Accordingly, we were unable to obtain price data for chipping potatoes as they are not published, as well as several other potato varieties. Second, potatoes are a heterogeneous commodity in that markets, prices, and production vary greatly across potato varieties, grades, size categories, and regions. Within each market segment, potatoes are assigned prices according to their size and other quality categories such as skin appearance, specific gravity, and sugar content. Revenue can be computed by summing across categories only if pricing information specific to those size and quality categories is available.

The potato price data we used were obtained from USDA Agricultural Marketing Service (AMS) Market News Reports for Russet Norkotah and Russet Burbank potatoes. (See USDA AMS (2023).) We use Russet Burbank as an approximation for Bannock prices. We use averages rather than current prices to model the expected price a farmer would use to make a multi-year decision. While AMS provides price data on sizes and grades, which measure quality, we lack pricing by specific quality characteristics. (Analyzing correlations between these specific quality traits collected in the study--hollow heart, scab, Verticilium vascular ring -- is the subject of ongoing research.) However, the study collected yields by grade, for which AMS also provides pricing data. Potato grading specifics are available from USDA AMS (2011). Generally, US1s are considered higher quality and are more valuable than US2s. We also provide a version of our main regressions incorporating information from an anonymously provided Russet Burbank contract that specifies specific gravity premiums and discounts in Table A6. (Our findings are unchanged)

Price data for the rotation crops were collected from USDA National Agricultural Statistics Service Quickstats (USDA National Agricultural Statistics Service, 2023a). For all price averages, we use a 5-year average of prices (2017–2021) adjusted to 2021 dollars using the US Federal Reserve GDP deflator. (The GDP deflator and documentation can be found at Federal Reserve Bank of St. Louis (2023).) Our price data are summarized in Appendix Table A3.

Cost data

We derive estimates for costs of production from a range of sources. We begin with enterprise budgets for each of the respective crops, selected to represent the locations in which the field trial took place. Enterprise budgets are generally collected and published by land grant universities, often by Extension or Agricultural Experiment Station offices. However, not all states publish enterprise budgets, and no state publishes these budgets for every possible crop. Since not every crop-state combination for the treatments in this study has a published enterprise budget, we substitute from other states and adapt as necessary. Data on the costs of the treatments were gathered from existing enterprise budgets and communication with the researchers who conducted the field trials. Costs include all inputs applied and practices conducted through harvest, and do not include storage costs. While storage costs, as well as disease losses that a grower may experience during storage, are a critical component of potato production profits, neither was collected as a part of this study. Storage costs were also only included in a subset of the enterprise budgets we draw from for our cost estimates.

The costs we compare are variable costs–generally, the costs that would vary according to short-term farm management practices including the soil health practices examined here. Since all plots were irrigated, we include irrigation-related expenses. We use Russet Burbank budgets from Idaho as a baseline for all potato production costs. Russet Burbank potatoes are sold both for fresh/table markets as well as for the frozen processed market which includes French fries, hash browns, and other specialty potato products. While field trial data we analyze includes varieties other than Russet Burbank and most trials took place outside of Idaho, the Idaho Russet Burbank budgets are the most consistently and recently reported and documented. Russet Norkotah cultivars were grown in roughly half of the treatments across the four locations in the study. Russet Norkotah potatoes are generally sold for ‘fresh’ or ‘table’ markets and are commonly found in the produce section of grocery stores. (Both Burbank and Russet Norkotah are also used in the dehydration supply chain that produces potato flakes and products like boxed instant mashed potatoes.) However, although their end markets differ substantially, pre-harvest Russet Burbank production practices sufficiently approximate Russet Norkotah production. (In addition to anecdotal production information provided by co-authors of this study who led the field trials, we also consulted an enterprise budget specialist.)

Appendix Table A4 provides a summary of the cost data used for each crop or practice within each location, and Appendix Table A5 summarizes the total variable cost estimates for each treatment and location. Where practices are employed other than those evaluated, such as cover cropping, we include the costs of those practices as well. The sources for those estimates are also detailed in Table A4.

Data limitations

We have identified several limitations in our approach that impede our ability to make inferential conclusions. The underlying field trials on which the economic work is based lack a treatment-control design that is applicable across all locations. (However, given that different regions use varying potato production systems, a control that is applicable across all locations would be difficult to establish.) Further, confounding factors that are not explicitly included as part of the study complicate the analysis. For example, the crop species planted in non-potato years of the rotations differ by treatments within study locations as well as across locations. Lastly, the timing of each crop necessarily varies across treatments. For example, planting dates differ across locations, crops, and years. Additionally, potatoes in the 3-year rotation are planted in different years than those in the 2-year rotation. These external factors and their interactions with soil health management practices vary within and among treatments, locations, and years. As such, it is not possible to claim that any one practice causes economic outcomes to change.

Additionally, researchers in some states did not vary or employ some of the practices we examine so it is not possible to evaluate them across or within all locations. Specifically, Idaho did not trial mustard biofumigant, and North Dakota chemically fumigated in all treatments before the beginning of the study. (North Dakota did vary the number of fumigations prior to the final potato year, 2022. In analysis for 2022 alone, we use fumigation in the year prior as the measure of fumigation rather than any fumigation during the study period.)

We also only observe 4 years of production. The existing literature supports the idea that the effects of many soil health practices build over time, likely over more than 4 years. (See, for example Wooliver and Jagadamma (Reference Wooliver and Jagadamma2023); Larney et al. (Reference Larney, Pearson, Blackshaw, Lupwayi and Lynch2016); Mohr et al. (Reference Mohr, Volkmar, Derksen, Irvine, Khakbazan, McLaren, Mon-real, Moulin and Tomasiewicz2011).) However, to our knowledge, our dataset is the first to allow an evaluation of correlations between economic outcomes and a range of soil health practices in regionally diverse potato systems in the short term.

Comparisons

Comparing potato production across locations reveals substantial variation in regional productivity. Because potato profits are dependent not only on overall yield but on size category, Figure 1 provides a summary of yields by size category over locations and practices in 2022. On average, trials conducted by researchers in Minnesota produced lower yields with a more uniform distribution than the other three locations. We will address differences in revenues summing across size categories in the following section, but the size profiles provide several insights. First, for most locations, potatoes grown in 3-year rotations had higher yields, led by higher production in the 10- to 14-ounce size category, into which the largest proportion of potatoes fall. Three-year rotations were generally associated with higher yields in the higher-valued, largest-size categories of more than 10 ounces. A similar phenomenon exists for manure application; for most locations overall yield, especially production of 6- to 10-ounce potatoes, was higher in treatments that had manure applied as a part of the treatment protocol. Results from both chemical and mustard fumigation are more ambiguous, although some explanation for a lack of difference is because of treatment design. (As mentioned above, Idaho did not employ any mustard fumigation.)

Figure 1. Comparisons of 2022 potato yield size profiles. Note: See Appendix Tables A1.1 to A1.4 for further details on practices by location. 1 indicates practice was employed; 0 indicates it was not. Chemical fumigation refers to fumigation in Fall of 2021. N=196. Bannock treatments are not included. Size categories are summed across US1 and US2 grades.

We also observe variation within locations over treatments and practices. Tables 1 and 2 provide results of a series of pairwise mean comparison tests for 2022 revenues and ROVCs, respectively, by treatment. We use Stata’s pwmean command with a groups option for these tests, to test differences between treatments within each location. After the tests are performed, treatments are assigned a letter. Treatments sharing a letter in the Group columns are not statistically significantly different from each other; treatments sharing a letter form a ‘group’. While the number of observations in each treatment is very small (either four or five replications), we note patterns both in how means are sorted, and groups are defined. For all locations except Oregon, 3-year rotation treatments have the highest returns and ROVCs and have at least one group that is statistically distinct from the lowest-ranking treatments. Treatments with Bannock potatoes appear at the top of the rankings for both ROVCs and returns in both locations where they were planted. We do not observe the same types of clear patterns for the remaining practices.

Table 1. 2022 Potato revenue by location and treatment

Note: Returns are estimated revenues, calculated as yield multiplied by price for each size category and variety. Prices are 5-year averages detailed in Table A3. Units for mean values are in dollars per acre. Groups sharing a letter are not statistically significantly different at the 5% level. In the Practice column, 3 refers to 3-year potato rotation and 2 years is the default rotation. F refers to fumigation in the previous year, and B refers to mustard biofumigation, done before potato planting. M refers to manure, applied in the Fall prior. In the Variety column, N refers to Russet Norkotah, B refers to Russet Burbank, and Ban refers to Bannock Russet. More detailed information on each treatment is presented in Appendix Tables A1.1, A1.2, A1.3, and A1.4. N= 4 replications per treatment for Idaho and Oregon, five replications per treatment for North Dakota and Minnesota. Table 2 provides an analogous analysis of returns over variable costs.

Table 2. 2022 ROVC by location and treatment

Note: Returns are estimated revenues, calculated as yield multiplied by price for each size category and variety. Prices are 5-year averages detailed in Table A3. Units for mean values are in dollars per acre. Groups sharing a letter are not statistically significantly different at the 5% level. In the Practice column, 3 refers to 3-year potato rotation and 2 years is the default rotation. F refers to fumigation in the previous year, and B refers to mustard biofumigation, done before potato planting. M refers to manure, applied in the Fall prior. In the Variety column, N refers to Russet Norkotah, B refers to Russet Burbank, and Ban refers to Bannock Russet. More detailed information on each treatment is presented in Appendix Tables A1.1, A1.2, A1.3, and A1.4. N= 4 replications per treatment for Idaho and Oregon, five replications per treatment for North Dakota and Minnesota. Table 1 provides an analogous analysis for returns.

To provide an overview of aggregated means by individual practices, we pool across states and varieties. While these comparisons do not allow for isolation of the different practices and treatments since each treatment represents a bundle of practices, they do allow us to compare our results from Tables 1 and 2 across a larger sample size. We use Stata’s t-test command to compute these summary statistics and t-tests comparing means. Table 3 presents means of economic variables by each of the four practices, pooled across locations. In the leftmost column, for each practice, the first three entries show the summary statistics for summed totals across all 4 years of the study, and the latter three show averages for the last year of the study (2022) only.

Table 3. Summary statistics by practice

Note: ROVC is returns over variable costs (revenue minus variable cost). Units for mean values are in dollars per acre, with the exception of the 2022 yield, which is in hundredweight per acre. In the ‘Longer rotation’ category, 3-year rotations correspond to the ‘treated’ column; ‘untreated’ are 2-year rotations. Chemical fumigation refers to any chemical fumigation in the 4-year comparisons and fumigation in the fall of 2021 for the 2022-specific comparisons. N differs because Oregon and North Dakota did not report yield for rotation crops in 2019, so we are unable to use 4-year sums for those treatments. Treatments that planted Bannock potatoes are not included. Asterisks denote statistical significance at the 10% (*), 5% (**), and 1% (***) levels.

Treatments that include manure application have statistically significantly higher direct costs. However, none of the other economic variables exhibit statistically significant differences over manure application. Treatments that included mustard biofumigation had statistically significantly higher variable costs, lower returns, and lower ROVCs than treatments that did not, both across the 4-year period and for 2022 alone. It is important to note that some treatments including mustard biofumigation did not replace chemical fumigation with mustard but rather both chemically fumigated and planted mustard as a biofumigant. Treatments with both types of fumigation have higher costs but do not appear to increase yields or revenues since they remain on average lower than those without mustard or chemical fumigation.

Treatments that were chemically fumigated had higher variable costs and lower returns and ROVCs over the 4-year rotation. All differences are statistically significant at the 5% or lower level. Somewhat surprisingly, when isolating the 2022 potato year, findings are similar. Chemically fumigated treatments had lower yields, lower returns, and lower ROVCs compared with treatments that did not include chemical fumigation. Each is statistically significant at the 10% or lower level. While this finding appears counter to common knowledge as well as several existing studies (e.g., Yellareddygari and Gudmestad, Reference Yellareddygari and Gudmestad2018), these simple comparisons do not allow us to isolate the role of chemical fumigation from other practices taking place as part of the treatment. (See Appendix Tables A1.1 to A1.4 for a list of practices for each treatment and location.) For example, locations with yields and revenues that are on average lower (ND and MN) had more fumigated trials than those with generally higher production values (ID and OR). The seeming lower performance of the fumigated treatments is likely at least partially explained by differing underlying production conditions between locations. These seemingly counter-intuitive findings motivate our regression analysis presented in the following section of the article.

Similar to previous tables, on average, 3-year rotations realized statistically significantly better economic outcomes when compared with 2-year rotations. While total costs over the 4 years of the study were higher in the 3-year rotations, overall returns as well as yield and returns in the last year of the study were significantly higher as well. Each comparison is statistically different at the 1% level, with the exception of 2022 yields, which were significantly different at the 5% level. Differences in crops selected, growing conditions, and management decisions may explain some of the differences in profit-related variables when comparing 2- versus 3-year rotations, as well as the lack of difference or unexpected signs when comparing across other soil health practices. For example, note that higher costs in the 3-year rotations are largely due to more corn being grown in the 3-year rotations. (Corn is a high-residue crop, and researchers reported that a ‘buffer year’ between planting corn and potatoes was common practice for this reason.) The costs for potatoes, which were planted twice in both the 2- and 3-year rotations, are on average essentially equal across rotation duration. These concerns further motivate our regression modeling in the following section.

Regression models

We include a series of fixed-effect regression models to control for differing climate, weather, and management practices in the four study locations and provide further insight into how our set of economic variables correlates with soil health practices. Specifically, we estimate:

(1) $$ {\pi}_j={\displaystyle \begin{array}{l}{\beta}_{0,i,j}+{\beta}_{1,j}{\delta}_{Manure,\ j}+{\beta}_{2,j}{\delta}_{Rotation,\ j}+{\beta}_{3,j}{\delta}_{Mustard,\ j}\\ {}+\hskip2px {\beta}_{4,j}{\delta}_{Fumigation,\ j}+{\beta}_{5,j}{\delta}_{Variety,\ j}+{\beta}_{6,j}{\delta}_{Location,\ k}+{\varepsilon}_j\end{array}} $$

where π_i are the economic variables of interest (i suppressed in equation in the interest of readability) listed across the top row of Table 4) and j is the replication block. The soil health practices all enter as dummy variables such that δ = 1 if the practice is employed and zero otherwise. In the case of rotation, δ_Rotation,j = 1 if the block is in a 3-year potato rotation, and 0 if it is in a 2-year rotation. For each of the four locations, δ_Location,k = 1 if the block is in location k and zero otherwise. The error term is ε _j.

Table 4. Regressions of economic variables on soil health practices

Note: All regression specifications are Ordinary Least Squares (OLS). All covariates enter as categorical or dummy variables. For each practice, default (omitted) category is ‘without treatment’. Idaho is the default location; estimated coefficients for other locations measure the difference from Idaho values. Rotation compares 3-year with 2-year rotations (the default). Mustard refers to mustard biofumigation. Fumigation refers to chemical fumigation, measured as any fumigation during 2021 for the 2022 columns, and any fumigation during the study period for the remaining columns. Burbank refers to Russet Burbank; Russet Norkotah is the default variety. Units on all dependent variables are dollars/acre, with the exception of 2022 (potato) yield, which is expressed in hundredweight per acre. The latter is a sum of all marketable yield. N differs across regressions because of missing yield data for earlier years of the study. All regressions include a constant term.

Table 4 shows our first set of regression results. Manure and mustard biofumigation results are largely inconclusive, especially when looking at the 2022 comparisons. Costs for treatments including both of these practices were significantly higher than treatments that did not include them. While in the comparison tests, treatments that included chemical fumigation had worse economic outcomes than those that did not, the regression results, which take into account other practices and the location of the trials, tell a more nuanced story. Here, we find that treatments including fumigation had higher costs and returns over the 4-year time period (statistically significant at the 1% and 5% levels, respectively), and higher, but not statistically significant, ROVCs over the same period. None of the fumigation results for the 2022 year alone were significantly different from zero.

In comparison to Idaho (the ‘default’ location), Minnesota and North Dakota trials generally realized worse economic performance as measured by both 4-year and 2022 variables. Oregon had higher variable costs and lower returns and ROVCs in the 4-year metrics but had higher yield and statistically indistinguishable 2022 returns and ROVCs. In comparison to Russet Norkotah (the ‘default’ variety), Burbank was generally statistically indistinguishable, with the exception of 4-year variable costs, which were marginally significantly lower.

As in Tables 1 to 3, we find that longer rotations had better economic performance than shorter rotations. All coefficients are significant at the 1% level. These results show that 3-year rotations had higher costs over the 4 years in the study but also received higher returns and ROVCs during that full study period. They also brought in greater 2022 yields, returns, and ROVCs.

Table 5 provides results from a similar exercise, utilizing both years in which potatoes were grown to form a panel over fixed locations and incorporating random effects for years. This linear mixed model can be expressed as:

(2) $$ {\pi}_{j,t}={\displaystyle \begin{array}{l}{\beta}_{0,j,t}+{\beta}_{1,j,t}{\delta}_{Manure,\ j}+{\beta}_{2,j,t}{\delta}_{Rotation,\ j}+{\beta}_{3,j,t}{\delta}_{Mustard,\ j}\\ {}+\hskip2px {\beta}_{4,j,t}{\delta}_{Fumigation,\ j}+{\beta}_{5,j,t}{\delta}_{Variety,\ j}+{\beta}_{6,j,t}{\delta}_{Location,\ k}+{u}_t+\hskip2px {\varepsilon}_{j,t}\end{array}} $$

Table 5. Mixed model regressions of economic variables on soil health practices, using first and second potato harvest data

Note: All regression specifications are Ordinary Least Squares (OLS). All covariates enter as categorical or dummy variables. For each practice, default (omitted) category is ‘without treatment’. Idaho is the default location; estimated coefficients for other locations measure the difference from Idaho values. Rotation compares 3-year with 2-year rotations (the default). Mustard refers to mustard biofumigation. Fumigation refers to chemical fumigation, measured as any fumigation during 2021 for the 2022 columns, and any fumigation during the study period for the remaining columns. Burbank refers to Russet Burbank; Russet Norkotah is the default variety. Units on all dependent variables are dollars/acre, apart from (potato) yield, which is expressed in hundredweight per acre. The latter is a sum of all marketable yield. Each observation in these regressions represents one replication for each year of potato harvest. Locations are treated as fixed effects, and harvest year as a random effect. All regressions include separate constant terms for random (year) fixed (all other covariates) terms. R-squared for mixed models is not directly comparable to that of OLS models; we do not include here it for that reason.

where t represents the year and u_,t is the random effect (intercept), for year. This version of our model has the benefit of using both years of potato data, while taking into account underlying differences between the years of production in the study. While there are some small changes in the levels of significance of covariates, the messages are the same. Our main result, that potatoes grown in longer rotations saw higher returns and ROVCs, still holds.