Measuring Farmer Response to the Rate of Agricultural Innovation: Experimental Evidence from Michigan

Surveys were sent to the first wave of Michigan Wheat Program contacts (160 farmers) during Winter 2019-2020. We also worked with MSU faculty to identify another email list of 600 Michigan farmers who can also serve to increase the sample size. We also adjusted the payment schedule to increase incentives to participate. Farmers now earn a guaranteed $25 for participating and up to an additional $50 based on their performance in the experiment (up to a total of $75). After sending out approximately 760 emailed surveys to Michigan producers, we obtained 19 completed responses for a response rate of 2.5%. While this response rate was lower than we had hoped, it remains in a normal range for email surveys with farmers. Additionally, our response rate may have been affected by both adverse weather events and the COVID-19 pandemic which occurred during the study period. 

Table 3 presents summary results from the survey portion of the experiment. Our respondents were, on average, 50 years old, overwhelmingly male, and had some college education. Approximately 58% of our sample had farming as their primary occupation, 47% had off-farm employment, and our respondents had an average of 24 years of farming experience. The average farmer participating in the survey owned 366 acres and rented in an additional 527 acres. Conditional on growing a specific crop, in 2019 producers planted an average of 241, 374 188, and 461 acres of wheat, corn, soybeans, and dry beans respectively. 

Table 3 Summary statistics

Notes: ^a1=High school/GED, 2=Some college, 3=Associates, 4=Bachelors, 5=Graduate/Professional degree. ^b1=Only when new technology needs replaced, 2=Rarely, 3=Annually, 4=Monthly, 5=Weekly, 6=Daily. ^c1=Extremely unlikely, 2=Unlikely, 3=Neither likely nor unlikely, 4=Likely, 5=Extremely likely. Source: Authors calculations. 

Our main focus in this study is understanding behaviors surrounding the adoption and use of new agricultural technologies among Michigan farmers. We first focused on understand what precision agriculture technologies were being used across our sample. In our sample, respondents reported using an average of approximately 3 precision agriculture technologies, with individual farmers using up to 6 technologies at the same time. The most commonly used technologies in our sample were GPS guidance with autosteer (52.6%), variable rate application services (52.6%), satellite/aerial imagery (47.4%), and GIS field mapping (42.1%). This data suggests that farmers in our sample were receptive to new agricultural technologies and the majority used precision agriculture technology in their operation. 

We next focused on where farmers learned about new agricultural technologies. Two information sources stood out among our sample. First, 78.9% of farmers indicated that they relied on other farmers as a source of information about new agricultural technologies. This highlights the importance of early adopters in a farming community to raise awareness of a potential technology. Second, 68.4% of our sample receive information on new technologies from agricultural input suppliers. Given that suppliers often stock new technologies (e.g. seed varieties) and can provide information on complementary technologies, this is an important information source for producers. Only 15.8% of farmers indicated that they learn about new technologies from extension agents. This may indicate that commodity extension programs are more focused on practices and markets rather than promoting the uptake of new agricultural technologies.

We also asked producers to rank the most important features of new agricultural technologies that they may consider adopting. The majority of our sample indicated that increased yields (94.7%) and decreased costs (84.2%) were among their top three most important features. Additionally contributing to on-farm sustainability (26.3%) and decreased yield variability (21.1%) were also important features. These findings emphasize the yield and cost concerns are at the forefront of the technology adoption decision. 

We find that farmers tend to discuss new agricultural technologies with other farmers (78.9%), input suppliers (47.4%), and company sales representatives (47.4%) with the vast majority sharing this information in person (84.2%) or at farmer/organization meetings (36.8%). These findings emphasize the importance of face-to-face interaction for the dissemination of information about new agricultural technologies. 

We next turn to analyzing the results from our economic experiment with the Michigan farmers. Each subject makes technology adoption decisions in every round of the experiment. We consider the four types of adoption scenarios that can occur within a round: No Change, Adopt Now, Adopt Delay, and Adopt Used. No Change occurs when a subject chooses to continue using a current technology in a given round – there is no technology switch and no fee incurred. Adopt Now is when a subject adopts a new technology in the round when it is first made available. Adopt Delay is when a subject adopts a new technology in any round after the round in which it is first made available. Finally, Adopt Used is when a subject re-adopts a technology that was previously used (and abandoned) in earlier rounds. Together, these four measures are mutually exclusive and cover all of the possible adoption decisions of a participant in a given round. 

Table 4 presents summary statistics of our key outcome variables for the full sample, the low innovation treatment, and the high innovation treatment. Our key outcome variables are experimental earnings (number of points) as well as the number of different technology adoption decisions made by farmers during the game. In the the low innovation treatment, we find that producers adopt an average of 7.18 technologies during the experiment compared to 4.88 technologies among producers in the high innovation treatment. However, while we predicted that the number of technology adoptions would be higher in the low innovation treatment, these two values are statistically indistinguishable (p=0.373) from one another. Similarly across all of our key experimental outcome variables, we find no statistically significant differences between the high innovation and low innovation treatments. One explanation for this finding may be that due to our low sample size, we lack the statistical power to detect an effect.

Table 4: Differences in means of experimental outcomes

Notes: P-values from two-sided t-test. Significance denoted by *** p<0.01, ** p<0.05, * p<0.1, respectively.

We also use multinomial logistic regression regression analysis to investigate the probability that a farmer makes a specific adoption decision (No Change, Adopt Used, Adopt Delay, Adopt Now) within a given round of the experiment. These regression models include several key variables of interest: an indicator for treatment status (1=High innovation treatment, 0=Low innovation treatment), farmer risk aversion levels (Relative risk aversion coefficient), whether or not a farmer observed a higher than average technology return in the previous round (Good tech return), whether a new innovation became available in the previous round (Innovation arrival), and whether a farmer switched their technology in the previous round (Technology switch). We restrict the analysis to only the technologies farmers had in common across the two treatments.

Table 5 presents the marginal effects of each of these variables on the probability of a farmer's decision. We find that being in the high innovation treatment significantly increases the probability that a farmer chooses not to change the technology that they're using within a given round by 10.5 percentage points and decreases the probability that a farmer adopts a technology they've used before by 6.4 percentage points. In other words, when new technologies are arriving at a faster rate we find that farmers are more likely to stick with the technology they're already using. This is somewhat consistent with our expectations, as the faster arrival of new technologies may make a "wait and see" approach the optimal strategy in the presence of fixed and sunk costs to adoption. 

Table 5: Multinomial Logit Marginal Effects for the Effect of Treatment on the Probability of Technology Adoption Behavior

Focusing on other variables of interest, we find that more risk averse farmers are more likely to adopt a technology that they've used before. A farmer who earned an above average return in a previous round with a given technology is more likely to stick with that technology in the current round. If a new technology arrived in the previous round, we find that farmers are significantly more likely to stick with their current choice or adopt something they've used before and significantly less likely to adopt something new immediately or with delay. Finally, if a farmer switched technologies in the previous round they are less likely to stick with their current choice and more likely to adopt something that they've used before.