Social Comparisons and Groundwater Use

We provide a social comparison mailer to a randomly selected group of agricultural producers in an area of eastern Colorado and western Kansas overlying the High Plains Aquifer. The mailer graphically compares their previous groundwater use to other groundwater wells within their groundwater management district (in Colorado) or county (in Kansas).

2019-05-01

2025-05-01

well-level groundwater use during the 2019, 2020, 2021, 2022, 2023, 2024, 2025 growing seasons

Planting patterns (e.g., acres of land planted in corn) and intensive margin water use (acre inches applied per acre irrigated).

We randomly assigned eligible agricultural producers in the study area to either a treatment or control group. Treatment and control groups remain the same throughout the trial. Eligibility was based on ownership of one or more wells in the study area. The treatment group of producers received a mailer during the beginning of the 2019, 2020, and 2021 growing seasons which compared 2018 groundwater use of up to three wells that they own to groundwater wells within their groundwater management district (GWMD) in Colorado or county in Kansas. Note that the trail was not conducted in Kansas in 2019.

The mailer graphically provided treated producers with information comparing the groundwater use of up to three of the wells that they own to well-level mean groundwater use and the groundwater use of the 20th percentile well in the GWMD/county. The mailer also indicated the percentile of the GWMD/county groundwater use distribution for each of their wells, using the following language “Comparing your YEAR water use to other wells in the YY GWMD/county, your well(s) recorded use higher than X% of wells.”

Many agricultural producers in our study area operate multiple wells. The assignment to treatment or control groups occurred at the producer level, not the well level. Producers receiving the treatment mailers were provided with well-level comparison information, but due to size constraints on the mailer, recipients were provided comparisons for a maximum of three wells. For producers that own or operate more than three wells, we limit the well-level comparison information to the three wells that utilized the most water in the previous year. We also exclude wells that rank less than the 5th percentile among wells within their comparison group. An exception to this rule is if a well owner's wells used ranked above the 5th percentile in the initial mailing and then less than the 5th percentile in subsequent mailings. These well owners continue to receive mailers to maintain the same treatment and control groups throughout the trial. Finally, some producers operate and own wells within multiple GWMDs/counties. For these producers we determine their comparison group based on the GWMD/county the majority of their wells are located within.

To assign agricultural producers to treatment or control we follow a block randomization methodology wherein we block at the GWMD/county level and on past water use compared to their comparison group’s average. Specifically, we generate a variable that indicates whether a producer’s 2018/2019 water use averaged across all their wells exceeded the median well-level water use within their comparison group (2018 for Colorado and 2019 for Kansas). This blocking protocol resulted in 14 mutually-exclusive blocking groups. Assignment of producers to treatment or control was conducted using R’s randomizr package, which in Colorado assigned 487 producers into the treatment group and 489 into the control group based on 2018 pumping data. The same blocking protocol was used in Kansas based on 2019 data and resulted in the assignment of 1414 producers into the treatment group and 1414 into the control group

Not available

We randomize assignment using the R package randomizr

Individual (well owner / agricultural producer)

No

We did not utilize clustering in our methods

976 eligible groundwater well owners / producers in the Colorado study area, 487 of these eligible well owners / producers were randomly placed within the treatment group and received information comparing their groundwater use to that of other wells in their GWMD. 2828 eligible groundwater well owners / producers in the Kansas study area, 1414 of these eligible well owners / producers were randomly placed within the treatment group and received information comparing their groundwater use to that of other wells in their GWMD. Some well owners in our study area rent their land and associated wells to other producers while others farm their own land. Our research design elected to send mailers only to well owners (both producers and non-producers) given a lack of reliable data regarding which producers rent specific agricultural land parcels. The eligibility criteria for well owners / producers is defined based on the uniqueness of the mailing address. The contact database provided by the State of Colorado provides a contact address for 2,558 wells in the study area. However, many agricultural operations utilize multiple wells in their operation. As such, we limit our analysis to 976 unique agricultural operations in one of seven GWMDs, which we define by unique mailing addresses.

In Colorado, 487 individuals (groundwater well owners / agricultural producers) were placed in the treatment group, 489 were placed in the control group.
In Kansas, 1414 individuals (groundwater well owners / agricultural producers) were placed in the treatment group, 1414 were placed in the control group.

Given our sample design and stratification of assignment, our power analysis without covariate control reveals a minimum detectable effect size of 5%, with power = 0.8 and alpha = 0.05, to distinguish groundwater use differences between treatment and control wells. However, with the inclusion of covariate control in the power analysis, specifically groundwater management district specific effects lagged well-level pumping, we find a minimum detectable effects size of 0.03, with power = 0.8 and alpha = 0.05. We use data on past well-level pumping to conduct the power analysis.

Pre-Analysis Plan