Think outside the plots: Perimeter measurements and spatial modeling mitigate confounding in a 145-year experiment

Long-term experiments (LTEs) offer unique insights into the effects of agricultural practices on soil organic carbon (SOC). However, early LTEs commonly lack treatment randomization, replication, and initial measurements of SOC. This creates a potential problem of unmeasured confounding. We address this problem using the Morrow Plots (established 1876) as a case study. We start with a standard mixed effects model of SOC and add (i) a spatial kriging component and (ii) SOC measurements in the sod perimeter of the experiment as an additional treatment level. We find that much of the observed SOC variation between treatments after 145 years is not due to treatments but other factors (e.g. initial SOC), attenuating treatment effects by about 50%. Our study demonstrates that creative measurement and innovative modeling can mitigate some deficiencies in early LTEs. However, our improved estimates still have limited precision, suggesting the importance of careful design and measurement in the first place.