This repository contains codes, slides, and other documents that I have used as a guest lecturer for ARE231 Agricultural Demand and Supply at UC Davis in 2021 Fall, a 2nd-year PhD course. ARE231 is taught jointly by Timothy Beatty and Aaron Smith. I held two sessions, each of which took around one and half hours.

In the first session, I covered coding mindset, useful features of Rstudio, short-cuts, snippets and project setup. Before the session, I have asked students to do a mini empirical research project using R (please see at the end of this document). I walked through my workflow and codes for the mini project.

In the second session, I taught how to use R for spatial data analysis. I spent about an hour about how to use R as a GIS tool. In the last half hour, I went through how to construct cropland-weighted county-level temperature data using gridded weather data from the PRISM Climate Group and landcover data from the National Land Cover Database (NLCD).

We are going to run the following panel regression of crop yields for corn and soybeans in Iowa, Illinois and Indiana over the period 1981-2019:

where denotes the yield for crop in county in year . denotes precipitation, and represents beneficial heat exposure (Growing Degree Days: degree days between 10C and 30C) and represents harmful heat exposure (Heating Degree Days: degree days above 30C). We include county fixed effects and county-specific time trend . The growing season for corn and soybeans in the three I states is approximately from April to September.

This crop yield model has become popular in the literature since the seminar work by Schlenker and Roberts (2009). You will learn more about the methodology and other related ones later in the course in more detail.

I do encourage you to spend some time trying to implement the following steps, although I will walk through them in the session.

It is totally ok even if you do not make much progress. But, I would like you to at least try each step so that you can make most out of my session.

1. Download annual county-level yield and acreage data for corn (for grain) and soybeans for all counties in the I states from the USDA NASS quickstats for the period 1981-2019. (Note: There are multiple ways of doing this. I would like your workflow to be reproducible and automated, meaning that executing your script downloads all data you need at once. Please feel free to refer to Accessing Ag Data Using R)

2. Download weather data by clicking Download All Monthly Data from US County Weather once the app is fully loaded. (In a few weeks later, we are going to learn how to construct this data using the gridded daily temperature data from the PRISM Climate Group, county boundaries, and crop frequency map)

3. Using the data you downloaded, create a crop-county-year level panel dataset that contains columns of yield and weather variables necessary for your regressions. This will involve some data cleaning. (Tip: You can construct annual or by summing monthly degree days over the growing season (April to September))

4. Before running regressions, do some data exploration to check your data. You can compare your data with US Crops.

5. Run regressions for the following combinations.

   • crop : 1) corn, 2) soybeans

   • period : 1) full (1981-2019), 2) pre 2000 (1981-2000), 3) post 2000 (2001-2019),

   • regression weights : 1) no weight, 2) acreage

   • cluster standard errors by: 1) year, 2) state

   • time trend : 1) county-specific linear, 2) state-specific linear, 3) county-specific quadratic, 4) state-specific quadratic

   • weather variables : 1) (with precipitation) , , and , 2) (without precipitation) and

That is, you have regression results.

6. Let’s focus on .

   • Obtain s and calculate their confidence intervals.
   • Compare s and their confidence intervals by changing one dimension at a time. (i.e., compare s for different crops 1) corn and 2) soybeans while keeping 1) for all the other dimensions).
   • Do you find any systematic differences between models in any dimensions? If so, what are they? Are they consistent with your intuition?