Gardner-Webb Doctoral Workshop - November 7th, 2026

These materials accompany the 2026 Fall Gardner-Webb Doctoral Workshop on using AI for research. You can follow along during the session or use the exercises on your own. For the live workshop, open Microsoft 365 Copilot Chat and sign in with your Gardner-Webb account.

Suggested readings / review on AI-usage in research:

• Korinek, A. (2023). Generative AI for economic research: Use cases and implications for economists. Journal of Economic Literature, 61(4), 1281-1317. https://doi.org/10.1257/jel.20231736 

• Hosseini, M., Murad, M., & Resnik, D. (2026). Benefits and risks of using AI agents in research. The Hastings Center Report, 56(1), 13-17. https://doi.org/10.1002/hast.70025 

• Modern AI for Economics Research: An Overview Tools w/ Benjamin Golub

  • I really appreciate Ben's explanation that strong prompting and content knowledge are fundamental to being surgical on research. 

• Scott Cunningham (Baylor University) has a Mixtape blog that I encourage checking out.

Practice #1

Papers:

• How different are different Differences-in-Differences? A Large-scale meta-analysis

• Tracking the credibility revolution across fields 

Using Copilot, practice with this prompt:

Use only the two attached working papers. Help me synthesize them in plain language.

1. Make a two-row table showing each paper’s research question, what evidence the author used, one main finding, and one limitation that affects how I should interpret it.

2. Write a three-sentence synthesis. Connect Goldsmith-Pinkham’s findings about where difference-in-differences is mentioned to Hull’s findings about how different ways of estimating it compare. Explain what the two papers help us understand together and identify one question they leave open.*

3. Give the author and printed page or section for each finding, limitation, and claim in the synthesis. Label your own interpretations. End by pointing me to one passage I should open to check a claim.

*For non-economics concentrations, students can swap these suggested titles for the appropriate research. 

Practice #2

Use this prompt to stress-test a statement that is written. I am going to use a snippet of a paper I wrote back in 2023. 

Side Note: Also, I am giving a special shout-out to someone I absolutely enjoy hanging out with at Economics Education conferences: Wayne Geerling. He is an absolute rockstar in teaching economics.

Prompt: Read this as a skeptical dissertation committee member. Identify one concern about how the evidence supports the conclusion. Point to the detail that prompted your concern, ask one question you would put to the researcher, and tell me what additional evidence would help answer it.

Paragraph: Economics classrooms are not foreign the introduction of using music to enhance learning (Hall et al., 2008; Raehsler, 2013; Geerling et al., 2022). The inclusion of diversity can be harvested from instructors who utilize appropriate resources that do not marginalize stereotypes in cultures (Geerling et al., 2021; Wooten et al., 2021). Previous uses include higher education classroom music projects (Al-Bahrani et al., 2017) and curating music for selected courses (Ben Abdesslem, 2022). Economics education supports the usage of music within a teaching environment that fosters diversity, equity, and inclusion (Geerling et al., 2022). However, many of the resources that are available to economics educators lack low-cost curriculum development that benefits access for the busy teacher. The Music4Econ project from Geerling et al. (2022) does have a dedicated section with ready-made questions that one could ask their course. However, the designs of the questions lack specific tailoring towards a target audience. At the time of this writing, only one lesson has specific warm-up questions that are commonly used in tertiary education (Geerling et al., 2021).

Practice #3

Vibe-coding -- the word that I presume makes coders cringe. Copilot can be useful in supporting students in reviewing their coding for errors or supporting a work around.

Download this Excel workbook: https://github.com/BradSc0tt1/bradscott1.github.io/issues/1 

Then pull up RStudio and copy & paste this code:

did_test_data <- readxl::read_excel(

  path = file.choose(),

  sheet = "DiD_Test_Data"

)

View(did_test_data)

did_model <- lm(outcome ~ treat * post, data = did_test_data)

print(summary(did_model))

After the results are posted, let's ask Copilot to code robustness check:

I loaded `did_test_data` into RStudio and fitted `did_model <- lm(outcome ~ treat * post, data = did_test_data)`. Each `id` identifies a study unit with a before and an after observation. Write base R code I can run after my existing script to check whether any one unit drives the `treat:post` estimate.

1. Extract the `treat:post` estimate from `did_model` as the baseline.

2. For each unique `id`, make a temporary subset that excludes both rows for that unit and refit the same model.

3. Print a table with the omitted ID, the number of observations used, the new `treat:post` estimate, and its difference from the baseline.

4. Print the minimum and maximum estimates across all refits. Round the displayed results to three decimal places.