Insurance-Provided Grocery Assistance and Health Care Outcomes Among Patients With Diabetes

ABSTRACT

Objective: To examine the association between a health insurance plan offering a grocery assistance benefit to members with diabetes and their health care use and outcomes.

Study Design: This retrospective, observational study used administrative claims to examine the associations of receiving grocery assistance provided by a health plan and health care use among patients with diabetes. A new health plan with low premiums and a grocery benefit (the intervention) was offered to a select group of members in 2021 and 2022. The intervention group examined in this study was made up of members with diabetes who selected the new health plan with the grocery benefit; the control group consisted of members with diabetes who were eligible for but chose not to select the new health plan with the grocery benefit.

Methods: We used a doubly robust inverse probability weighted difference-in-differences design to analyze the impact on health care use, including inpatient and outpatient utilization and medication adherence, and on health outcomes, including those related to diabetes complications.

Results: The intervention group saw an associated decrease of 3.1 percentage points in the prevalence of diabetes with complications. We also found significant associated increases in adherence to diabetes medications. No significant associated changes in health care service utilization or other health outcomes were observed.

Conclusions: Providing low-touch grocery assistance to members with diabetes improved management of diabetes complications. The findings suggest that integrating nutritional benefits into health plans may enhance health outcomes without increasing overall health care utilization. Further research is needed to explore the long-term impact and scalability of such interventions.

Am J Manag Care. 2026;32(8):In Press

Takeaway Points

• Members with type 2 diabetes who received an insurer-provided grocery benefit saw an associated decrease in the number of visits with diagnoses for diabetes with complications.
• Among the intervention population, overall health care utilization did not decrease and place of service–specific utilization did not change.
• Enrollment in the intervention was associated with an increase in adherence to diabetes-related medications.

Poor nutrition and dietary habits may lead to diet-related diseases, such as type 2 diabetes, heart disease, obesity, and certain cancers.¹ Furthermore, recent research has shown many complex relationships between diet and psychiatric symptoms.^2-4 As such, US policy makers and health care systems have recognized that providing access to nutritious food may enhance medical practice, improve whole health, and reduce existing health inequities.⁵ Despite this increased interest from the public and private sectors, there are still challenges associated with implementing “food as medicine” approaches, with cost being chief among them.⁶ Integration of nutrition into the health care system has been hypothesized to be a cost-effective or cost-saving method of enhancing diet and improving health; therefore, health plans may be uniquely suited to offer nutritional assistance as part of health plan benefits given its potential to reduce health risks while offsetting health care spending.^7-21

Studies on the impact of non–medically tailored meal/food delivery interventions are limited but show promising results, particularly among Medicaid members.^14-18 However, the evidence on the effectiveness of such programs provided by commercial health plans for their members is more scarce.¹⁹ Recent research examined a 6-month pilot program developed by Blue Cross and Blue Shield of North Carolina that implemented a food delivery and health coaching program for plan members with type 2 diabetes living on a low income.¹⁹ Completion of the 6-month program was associated with improvements in health security, obesity rates, self-reported health, diabetes management, and medical expenses. Geisinger Health demonstrated a similar intervention through its Fresh Food Farmacy program, in which participants with diabetes received a prescription from their primary care providers for fresh food for themselves and their families and were provided enough food to prepare 10 meals per week.^20,21 This intervention was associated with improvements in engagement with the health care system and self-reported diet but not hemoglobin A_1c levels in the 6 months following the intervention.

Because the evidence on commercial health plan–provided food/grocery benefits is still quite limited, the purpose of this study was to examine the association between a medium-term health plan–provided grocery assistance program and health outcomes and health care service use among people with diabetes. The intervention coupled a narrow, high-performing network health plan with provision of funds for covered grocery expenses. Unlike prior studies, we were able to follow members using the grocery benefit for up to 2 years.^20,21

Intervention

Starting on January 1, 2021, Elevance Health offered health plans to members that provided financial assistance for nutritious groceries. These plans were offered to members meeting the income eligibility requirement regardless of any history of diagnosis or health risks and had the lowest premium and cost-sharing requirements with respect to co-pays, deductibles, and out-of-pocket limits (eAppendix 1 [eAppendices available at ajmc.com]). The new insurance plan was designed to determine the feasibility of a low-cost health insurance plan that offered grocery assistance in exclusive provider network (EPO) plans or health maintenance organization (HMO) plans that did not cover out-of-network services and encouraged members to use high-quality, lower-cost care.

The grocery coverage option was offered to members who met household income eligibility criteria due to existing evidence on associations between income and low-quality diet, nutritional deficiencies, and diet-related disease prevalence.^22-25 Specifically, the plan offered a grocery benefits card that covered flexible grocery purchases including fresh fruits, vegetables, whole grains, legumes, cooking oils, meat, fish, dairy products, bread, and cereals but excluded alcohol, tobacco, and prepared foods.²⁶ The amount of coverage ranged from $1200 to $3000 annually depending on the household size and coverage (eAppendix 1), and the benefit card could be used at a wide range of network stores (eAppendix 2).

METHODS

Study Design and Population

This retrospective observational study used administrative claims data to evaluate the association between a grocery assistance program and health outcomes. The study period lasted from January 1, 2020, to June 30, 2023. To be included in this study, health plan members were required to have diagnosis codes for type 2 diabetes with or without complications in their claims history (see eAppendix 3 for diagnosis codes) before they were enrolled in the new plan. Enrollment could start in January 2021 or January 2022, with the first 6 months of 2023 included to allow for increased follow-up time.

We limited this study of the intervention to only those with prior diagnoses of diabetes because diabetes is diet related and we sought to examine how more general grocery assistance may benefit a target population with diet-related needs. Members who enrolled in the new health plan were included in the intervention group, and those who were eligible (ie, satisfying the income criteria as outlined in eAppendix 1) but did not select one of the eligible health insurance plans were included in the control group. This work was part of ongoing health plan operations and quality improvement initiatives to enhance benefits offered to members; therefore, an institutional review board review was not needed.

Data and Statistical Analysis

We conducted a retrospective cohort study of commercially insured patients who were eligible to enroll in the intervention between January 1, 2021, and December 31, 2022. We used member-level data, which included intervention eligibility and enrollment status for members on a quarterly basis from January 1, 2021, to June 30, 2023, merged with administrative claims data from the HealthCare Integrated Research Database. The unit of analysis was at the member-quarter level.

Demographic information such as age, sex, urbanicity, and household size was obtained from administrative claims data. Area-level variables such as socioeconomic index quartile, median family income, and indicators for living in a food desert were obtained by linking member data to the American Community Survey and US Department of Agriculture Economic Research Service Food Access Research Atlas by using their census block in their claims.^27,28

We used a doubly robust, staggered adoption difference-in-differences (DRDID) design to examine the impact of receiving the grocery benefits. Regressions included member and quarter fixed effects to account for all time-invariant observed and unobserved characteristics of members (ie, age, sex, ethnicity) as well as common secular time trends. A DRDID methodology is robust to possible biases caused by comparing not-yet-treated members with already-treated members.²⁹ The doubly robust design also uniquely allows for the inclusion of covariates to account for observed selection bias.³⁰ These covariates were used to weight treated and control members by their propensity to enroll in a grocery assistance–eligible insurance plan. We included age, sex, region of residence, household size, rurality, indicators of being in a low-access/low-income food desert, census tract socioeconomic index quartile, count of comorbidities, and race as matching factors. Our sample was limited to those with a preexisting diabetes diagnosis in the year before the intervention. We also limited our sample to those with incomes that would make them eligible for enrollment. For these reasons, we excluded additional clinical and income-related covariates from our weighting process. DRDID designs also rely on less restrictive model assumptions because only the parallel trends assumption of the DID or the propensity score for the likelihood of being treated need to be valid.²⁹

To additionally account for selection bias inherent to insurance plan selection, we conducted subsample analyses using those who had not yet selected into the plan with the grocery benefit but eventually did so as the control variables. This reduced our sample size and limited the years of our sample to only the first year of the intervention, as all members were enrolled in the second year of the intervention. Further sensitivity analysis included matching grocery benefit recipients to nonrecipients prior to conducting the DID estimation. Because our study period overlapped with the COVID-19 pandemic, we conducted an additional analysis starting in 2021 rather than 2020. Our DID design also accounted for secular trends that were common to both enrolled and control members.

We performed several heterogeneity and stratification analyses to assess whether demographic characteristics modified the association between enrollment in a Life Essential Kit (LEK)–eligible plan and the prevalence of diabetes with complications. We estimated this association separately for Black, Hispanic, and White members, by sex, by socioeconomic quartile of member’s residence, and by age group.

All analyses were conducted in Stata/SE 16 (StataCorp LLC), and the csdid module by Callaway and Sant’Anna was used.³¹ All hypothesis tests were 2 sided, with an α of .05 indicating statistical significance.

Outcomes

Outcomes included both diagnoses and measures of health care utilization. Diagnosis outcomes were binary variables presenting whether a member was diagnosed with hypertension with complications (ie, hypertensive heart or chronic kidney disease), hypertension without complications, diabetes with complications (ie, diabetes with renal, ophthalmic, or neurological manifestations), diabetes without complications, obesity, anxiety, and depression in each quarter (eAppendix 3).

Utilization was defined as the total number of visits to a variety of health care settings, including emergency department visits, hospitalizations, and office visits. Medication adherence was defined as mean rolling proportion of days covered (medication codes are included in eAppendix 4). These adherence measures were medication class specific and did not account for medication adherence across classes that a member might have been prescribed. These adherence measures were averaged over 4 quarters to result in the yearly mean percentage of days with supply on hand. This accounted for members filling multiple prescriptions in the quarter and not in the next quarter, as many prescriptions were filled with 90 days’ supply.

RESULTS

The examined intervention group consisted of 411 members who were diagnosed with type 2 diabetes prior to being enrolled in the new health plan and receiving grocery benefits. Among those, 306 started participating in 2021 and 105 started participating in 2022. The control group consisted of 961 members who had diabetes in 2020 and who were eligible to enroll into the new health plan but did not participate.

On average, members enrolled in the intervention group vs those in the control group were slightly younger (47 vs 50 years), were more likely to be Black (35% vs 29%), had a slightly larger household size (2.6 vs 2.4), and lived in areas with lower median family income ($67,546 vs $72,327). Additionally, the intervention group had relatively less health care services utilization and costs at baseline (Table 1).

Table 2 shows changes in outcomes related to common comorbidities associated with diabetes and related medication adherence. We observed a significant 3.1–percentage point (PP) decrease in the prevalence of diabetes with complications. An event study for this outcome showed that the trend in visits with a diagnosis for diabetes with complications did not differ between the control and intervention groups prior to enrollment in an LEK-eligible plan (eAppendix 5). We did not find any association between the grocery benefit and the prevalence of diabetes without complications, hypertension with or without complications, or obesity. We did, however, see that the intervention was associated with improved adherence to diabetes medications and antidepressants for those who ever filled a related prescription by 7.5 PP (95% CI, 1.6-13.4; P = .013) for diabetes medications and 11.6 PP (95% CI, 6.0-17.2; P < .001) for antidepressants. We also found that the proportion of days covered for statins and antihypertensive medications was not significantly impacted by enrollment in an LEK-eligible plan.

Table 3 shows results for use of health care services. We found that enrollment in a grocery assistance–eligible plan was not associated with changes in the number of quarterly health care visits across any place of service.

Sensitivity and Subsample Analyses

We conducted multiple sensitivity analyses and subsample analyses including matching prior to the DID regression as well as including only those who eventually selected an insurance plan with the grocery benefit.

For diagnoses related to diabetes, related medication adherence, and overall utilization, we found results when matching prior to the DID estimation similar to those of our main specification where we simultaneously used inverse probability weighting and DID. The association between diabetes with complications and the intervention was not statistically significant for the model in the sensitivity analyses but was similar in magnitude to our preferred DRDID specification (eAppendix 6). Additionally, enrollment in a health care plan with the grocery benefit was not associated with a change in utilization aside from an associated increase of 0.114 (95% CI, 0.004-0.225; P = .042) in the number of quarterly telehealth visits in the sensitivity analyses (eAppendix 7).

In addition to the matched analysis, we performed a subsample analysis in which we included only those who enrolled or eventually enrolled in the grocery assistance–eligible plan. We found that members who were enrolled in the grocery assistance–eligible plan saw an 8.2-PP (95% CI, –14.9 to –1.4; P = .017) decrease in the likelihood of a diagnosis of diabetes with complications compared with members who had not yet enrolled in the plan (eAppendices 8 and 9).

Our final sensitivity analyses excluded 2020 and looked at those who enrolled in 2022 with 2021 as their preenrollment pretreatment year. As shown in eAppendices 10 and 11, those enrolling in 2022 saw a marginally significant associated 5.1-PP decrease in the prevalence of diabetes with complications (95% CI, –11.1 to 0.7; P = .086).

Regarding the stratification analysis, we found the largest associated decreases in diabetes with complications among Hispanic members, with Black and White members having no statistically significant associated change in diabetes with complications following enrollment in an LEK-eligible plan (eAppendix Table 12). Both men and women saw marginally significant decreases in diabetes with complications associated with enrollment in an LEK-eligible plan, with men having larger estimated associated decreases. Finally, we found statistically significant decreases in diabetes with complications for LEK-enrolled members living in census tracts in the lowest 2 socioeconomic quartiles, with no statistically significant impact on those residing in the highest-quartile census tracts.

DISCUSSION

We found that enhancing access to groceries among a privately insured commercial population with diabetes may yield significant positive changes. Among patients with diabetes, providing grocery benefits was associated with an improvement in medication adherence for some prescriptions and a reduction in the prevalence of diabetes with complications. We conducted several supporting analyses including sensitivity analyses including only those who eventually enrolled, matching before conducting the 2-way fixed effects analyses, and testing for differential pretrends between enrolled and nonenrolled members. In all these analyses, we found little change in the results. The analysis comparing early enrollees with not-yet-but-eventual enrollees was particularly informative, as we found no differences in their preperiod trends but found significant reductions in associated prevalence of diabetes with complications for early enrollees.

The LEK grocery benefit was also a more passive intervention than a grocery delivery or medically tailored meal intervention as this less-involved intervention lacked education on nutrition and required greater effort on the part of the members receiving the benefit. However, this and similar benefits may be more widely feasible due to being simpler and more universally applicable. Even with this less-involved intervention, we found that members with diabetes with complications receiving the insurance-provided grocery benefit showed improvements in medication adherence and reductions in claims with complications.

Because our study overlapped with the COVID-19 pandemic, one of our sensitivity analyses excluded 2020 from our sample. Although this analysis includes 2021 as a preintervention year—during which both intervention and control members were untreated—we believe our results are robust to COVID-19 pandemic–related effects as our included fixed effects control for secular trends common to all members regardless of enrollment status. Regarding the event study, we found no evidence that enrolled members had differential trends in the rate of diabetes with complications. The lack of pretrends supports the assumptions of parallel trends and no anticipation required for a 2-way fixed effects model.

Although estimating the long-term cost-effectiveness of a food-as-medicine program is beyond the scope of this study, insights from this study can shed light on the trade-offs between nutrition and health insurance benefits. Overall, participants spent $428 less on premiums by opting for the newly designed narrow network plans and received $1869 toward nutrition annually. The program’s cost neutrality is suggested by the absence of evidence indicating alterations in health service utilization, including outpatient, inpatient, or emergency department visits. Moreover, findings from a survey study indicate that the new health plan was correlated with enhancements in perceived food security, dietary quality, and overall health.²⁷ The current study’s results, when combined with the results from the survey study, may indicate that the new health plan led to improvements in both self-reported and claims-based outcomes, even though it did not impact the utilization of health care services.

Limitations

We recognize several limitations of this study. Because our intervention was not randomly assigned but based on voluntary enrollment, members who chose to opt in for the new health plan with grocery assistance might have already been on a different path that would have eventually led to improvements in their dietary habits and/or management of diabetes. The voluntary enrollment of members into an LEK-eligible plan may affect the generalizability of these effects to only those with similar motivation and other characteristics to those who enrolled. Similar to other food-as-medicine pilots, the grocery benefit was not randomly assigned in this study; therefore, we cannot say with certainty that the estimated associations would not have been realized in the absence of enrollment in a grocery assistance–eligible health plan. Because the new health plan has only been rolled out since 2021, the long-term impact of such programs, particularly in preventing diabetes complications among those who might have developed them without improved nutritional access, remains uncertain. Also, we did not report changes in cost associated with enrollment into the grocery assistance–eligible health plan because eligible plans had different payment schedules and included a narrower network than other plans. Lastly, our measure of medication adherence measured the mean days’ supply that a member had filled, which may not have fully measured whether the medications were taken, but rather only that the medications were available to be taken.

CONCLUSIONS

The findings of this study indicate that a targeted grocery benefit intervention by a payer was associated with short- to medium-term clinically important improvements to member outcomes and prescription adherence. Our findings further suggest that the intervention was not associated with negative changes in health care utilization or other outcomes.

Author Affiliations: Carelon Research (JR, WC), Wilmington, DE; Elevance Health (KE, RM, DG, SA), Indianapolis, IN.

Source of Funding: Research was conducted by Elevance Health employees as a part of their job functions.

Author Disclosures: Drs Romine and Chi are employees of Carelon Research, an independent research subsidiary owned by Elevance Health. Dr Essel, Ms Mullikin, Ms Gibson, and Dr Agrawal are employees of Elevance Health, and Dr Essel, Ms Gibson, and Dr Agrawal receive stock in Elevance Health as employees. The study population was enrolled in an Elevance Health employer-sponsored medical insurance plan.

Authorship Information: Concept and design (JR, WC, KE, RM, DG, SA); acquisition of data (JR, WC, DG, SA); analysis and interpretation of data (JR, WC, KE, SA); drafting of the manuscript (JR, KE, DG); critical revision of the manuscript for important intellectual content (JR, WC, KE, RM, SA); statistical analysis (JR); provision of patients or study materials (DG); obtaining funding (WC); administrative, technical, or logistic support (RM, DG); and supervision (WC, SA).

Address Correspondence to: Jeff Romine, PhD, Carelon Research, 123 Justison St, Ste 200, Wilmington, DE 19801. Email: jeff.romine@carelon.com.

REFERENCES

1. Micha R, Peñalvo JL, Cudhea F, Imamura F, Rehm CD, Mozaffarian D. Association between dietary factors and mortality from heart disease, stroke, and type 2 diabetes in the United States. JAMA. 2017;317(9):912-924. doi:10.1001/jama.2017.0947

2. Bremner JD, Moazzami K, Wittbrodt MT, et al. Diet, stress and mental health. Nutrients. 2020;12(8):2428. doi:10.3390/nu12082428

3. Godos J, Currenti W, Angelino D, et al. Diet and mental health: review of the recent updates on molecular mechanisms. Antioxidants (Basel). 2020;9(4):346. doi:10.3390/antiox9040346

4. Lewandowska-Pietruszka Z, Figlerowicz M, Mazur-Melewska K. The history of the intestinal microbiota and the gut-brain axis. Pathogens. 2022;11(12):1540. doi:10.3390/pathogens11121540

5. Mozaffarian D, Blanck HM, Garfield KM, Wassung A, Petersen R. A food is medicine approach to achieve nutrition security and improve health. Nat Med. 2022;28(11):2238-2240. doi:10.1038/s41591-022-02027-3

6. Thomas O, Badaracco C, Akabas S. Why current ‘food is medicine’ solutions are falling short. Health Affairs Forefront. May 12, 2023. Accessed October 24, 2023. https://www.healthaffairs.org/content/forefront/five-food-problems-why-current-food-medicine-solutions-falling-short

7. Executive summary: Biden-Harris administration national strategy on hunger, nutrition, and health. The White House. September 27, 2022. Accessed October 17, 2023. https://web.archive.org/web/20231001143854/https://www.whitehouse.gov/briefing-room/statements-releases/2022/09/27/executive-summary-biden-harris-administration-national-strategy-on-hunger-nutrition-and-health/

8. Berkowitz SA, Terranova J, Hill C, et al. Meal delivery programs reduce the use of costly health care in dually eligible Medicare and Medicaid beneficiaries. Health Aff (Millwood). 2018;37(4):535-542. doi:10.1377/hlthaff.2017.0999

9. Berkowitz SA, Terranova J, Randall L, Cranston K, Waters DB, Hsu J. Association between receipt of a medically tailored meal program and health care use. JAMA Intern Med. 2019;179(6):786-793. doi:10.1001/jamainternmed.2019.0198

10. Budhu JA. Food is medicine, so health insurance should cover it. The Hill. August 18, 2023. Accessed October 18, 2023. https://thehill.com/opinion/healthcare/4157888-food-is-medicine-health-insurance-should-cover-it

11. Downer S, Berkowitz SA, Harlan TS, Olstad DL, Mozaffarian D. Food is medicine: actions to integrate food and nutrition into healthcare. BMJ. 2020;369:m2482. doi:10.1136/bmj.m2482

12. Hager K, Cudhea FP, Wong JB, et al. Association of national expansion of insurance coverage of medically tailored meals with estimated hospitalizations and health care expenditures in the US. JAMA Netw Open. 2022;5(10):e2236898. doi:10.1001/jamanetworkopen.2022.36898

13. Hager K, Shi P, Li Z, et al. Evaluation of a produce prescription program for patients with diabetes: a longitudinal analysis of glycemic control. Diabetes Care. 2023;46(6):1169-1176. doi:10.2337/dc22-1645

14. Hager K, Du M, Li Z, et al. Impact of produce prescriptions on diet, food security, and cardiometabolic health outcomes: a multisite evaluation of 9 produce prescription programs in the United States. Circ Cardiovasc Qual Outcomes. 2023;16(9):e009520. doi:10.1161/CIRCOUTCOMES.122.009520

15. Gany F, Melnic I, Wu M, et al. Food to overcome outcomes disparities: a randomized controlled trial of food insecurity interventions to improve cancer outcomes. J Clin Oncol. 2022;40(31):3603-3612. doi:10.1200/JCO.21.02400

16. Little M, Rosa E, Heasley C, Asif A, Dodd W, Richter A. Promoting healthy food access and nutrition in primary care: a systematic scoping review of food prescription programs. Am J Health Promot. 2022;36(3):518-536. doi:10.1177/08901171211056584

17. Hager K, Sabatino M, Williams J, et al. Medicaid nutrition supports associated with reductions in hospitalizations and ED visits in Massachusetts, 2020-23. Health Aff (Millwood). 2025;44(4):413-421. doi:10.1377/hlthaff.2024.01409

18. Berkowitz SA, Archibald J, Yu Z, et al. Medicaid spending and health-related social needs in the North Carolina Healthy Opportunities Pilots program. JAMA. 2025;333(12):1041-1050. doi:10.1001/jama.2025.1042

19. Lumpkin JR, Taylor LH, Hattori A, Jedele JM. Impact of food delivery and health coaching on outcomes and costs of care: a payer’s perspective. NEJM Catal Innov Care Deliv. 2023;4(4). doi:10.1056/CAT.22.0351

20. Hess A, Passaretti M, Coolbaugh S. Fresh Food Farmacy. Am J Health Promot. 2019;33(5):830-832. doi:10.1177/0890117119845711d

21. Doyle J, Alsan M, Skelley N, Lu Y, Cawley J. Effect of an intensive food-as-medicine program on health and health care use: a randomized clinical trial. JAMA Intern Med. 2024;184(2):154-163. doi:10.1001/jamainternmed.2023.6670

22. Lee H, Harris KM, Gordon-Larsen P. Life course perspectives on the links between poverty and obesity during the transition to young adulthood. Popul Res Policy Rev. 2009;28(4):505-532. doi:10.1007/s11113-008-9115-4

23. Levine JA. Poverty and obesity in the U.S. Diabetes. 2011;60(11):2667-2668. doi:10.2337/db11-1118

24. Ludwig J, Sanbonmatsu L, Gennetian L, et al. Neighborhoods, obesity, and diabetes—a randomized social experiment. N Engl J Med. 2011;365(16):1509-1519. doi:10.1056/NEJMsa1103216

25. Morland K, Diez Roux AV, Wing S. Supermarkets, other food stores, and obesity: the Atherosclerosis Risk in Communities Study. Am J Prev Med. 2006;30(4):333-339. doi:10.1016/j.amepre.2005.11.003

26. Cullen D, Cobb R, Sylwestrzak G, et al. Supporting low-income workers to address nutritional needs. NEJM Catal Innov Care Deliv. 2023;4(8). doi:10.1056/CAT.22.0405

27. American Community Survey 5-year data profile (2009-2024). US Census Bureau. Accessed May 22, 2024. https://www.census.gov/data/developers/data-sets/acs-5year.2019.html#list-tab-1806015614

28. Food Access Research Atlas. US Department of Agriculture Economic Research Service. Accessed May 22, 2024. https://www.ers.usda.gov/data-products/food-access-research-atlas/

29. Callaway B, Sant’Anna PHC. Difference-in-differences with multiple time periods. J Econom. 2021;225(2):200-230. doi:10.1016/j.jeconom.2020.12.001

30. Sant’Anna PHC, Zhao J. Doubly robust difference-in-differences estimators. J Econom. 2020;219(1):101-122. doi:10.1016/j.jeconom.2020.06.003

31. Rios-Avila F, Sant’Anna PHC, Callaway B. CSDID: Stata module for the estimation of difference-in-difference models with multiple time periods. Accessed March 6, 2024. https://ideas.repec.org/c/boc/bocode/s458976.html