1. Introduction

Overview

The New York City Farmers’ Markets are an essential part of the city’s public health and food access ecosystem. These markets serve as community hubs, offering fresh produce and often participating in nutrition assistance programs. Understanding how these markets operate across boroughs and whether they promote equitable access to healthy food options is crucial, particularly for policy initiatives aimed at food insecurity and nutrition outreach.

Key Variables

For this analysis, I focus on the following key variables from the NYC Farmers’ Markets dataset:

  • accepts_ebt_bin: Whether the market accepts EBT (Electronic Benefit Transfer), recoded into binary form (1 = Yes, 0 = No).
  • health_bucks_bin: Whether the market distributes Health Bucks (1 = Yes, 0 = No).
  • open_year_round_bin: Whether the market operates year-round (1 = Yes, 0 = No).
  • Borough: The NYC borough in which the market is located.

Research Questions

  1. Are markets that distribute Health Bucks more likely to accept EBT?
  2. Is there a relationship between year-round market operation and EBT acceptance?
  3. Does borough location influence whether a market accepts EBT?

Hypothesis

  • H1: Markets that distribute Health Bucks are more likely to accept EBT.
  • H2: Markets open year-round are more likely to accept EBT.
  • H3: Borough location has a significant association with EBT acceptance rates.

Statistical Analysis

To investigate these research questions, I use logistic regression models, treating accepts_ebt_bin as the dependent variable. Predictor variables include: - distributes_health_bucks - open_year_round - borough

The analysis includes: - Data cleaning and transformation - Descriptive statistics using datasummary_skim() - Regression modeling with glm() - Simulation and interpretation of average marginal effects using the clarify package - Visualization and formatted tables with gt and modelsummary

Load data

farmers_markets <- read.csv("C:/Users/susha/Downloads/NYC_Farmers_Markets_20241107.csv")

View data

head(farmers_markets)
##     Borough                                              Market.Name
## 1 Manhattan                  Harvest Home East Harlem Farmers Market
## 2  Brooklyn Seeds in the Middle - Flatbush - Hillel Plaza Farm Stand
## 3     Bronx                                 Morris Heights Farmstand
## 4     Bronx                                           170 Farm Stand
## 5     Bronx                    JBOLC Garden Community Farmers Market
## 6    Queens                                         Perez Farm Stand
##                              Street.Address Community.District Latitude
## 1                   E. 104th St. & 3rd Ave.                111 40.79030
## 2                Flatbush/Nostrand Triangle                314 40.63262
## 3          University Ave. and W. 179th St.                205 40.85457
## 4                        1406 Townsend Ave.                204 40.84014
## 5              Sedgwick Ave. & Goulden Ave.                207 40.88265
## 6 134-20 Jamaica Ave., by the Axel Building                409 40.70236
##   Longitude Days.of.Operation Hours.of.Operations           Season.Begin
## 1 -73.94563          Thursday           8AM - 3PM 06/15/2023 12:00:00 AM
## 2 -73.94744         Wednesday           4PM - 7PM 01/01/2023 12:00:00 AM
## 3  40.85457         Wednesday          11AM - 3PM 03/27/2023 12:00:00 AM
## 4 -73.91659         Wednesday           2PM - 6PM 07/12/2023 12:00:00 AM
## 5 -73.88656          Saturday          10AM - 3PM 06/17/2023 12:00:00 AM
## 6 -73.81853         Wednesday          10AM - 4PM 06/14/2023 12:00:00 AM
##               Season.End Accepts.EBT Distributes.Health.Bucks. Open.Year.Round
## 1 11/16/2023 12:00:00 AM         Yes                       Yes              No
## 2 12/31/2023 12:00:00 AM         Yes                       Yes              No
## 3 11/29/2023 12:00:00 AM         Yes                       Yes              No
## 4 11/22/2023 12:00:00 AM         Yes                       Yes              No
## 5 10/28/2023 12:00:00 AM         Yes                       Yes              No
## 6 11/01/2023 12:00:00 AM         Yes                       Yes              No
##   Cooking.Demonstrations          Location.Point Zip.Code
## 1                     No   (40.7903, -73.945635)    10029
## 2                     No  (40.63262, -73.947439)    11210
## 3                     No  (40.854567, 40.854567)    10453
## 4                    Yes (40.840138, -73.916591)    10452
## 5                     No (40.882647, -73.886562)    10468
## 6                     No  (40.70236, -73.818531)    11418
colnames(farmers_markets)
##  [1] "Borough"                   "Market.Name"              
##  [3] "Street.Address"            "Community.District"       
##  [5] "Latitude"                  "Longitude"                
##  [7] "Days.of.Operation"         "Hours.of.Operations"      
##  [9] "Season.Begin"              "Season.End"               
## [11] "Accepts.EBT"               "Distributes.Health.Bucks."
## [13] "Open.Year.Round"           "Cooking.Demonstrations"   
## [15] "Location.Point"            "Zip.Code"
table(farmers_markets$Accepts.EBT)
## 
##  No TBD yes Yes 
##   7   3   1 152

Standardizing ‘Accepts.EBT’ column

farmers_markets <- farmers_markets %>%
  mutate(accepts_ebt = case_when(
    Accepts.EBT %in% c("Yes", "yes") ~ 1,   
    Accepts.EBT %in% c("No", "TBD") ~ 0,    
    TRUE ~ NA_real_))
farmers_markets$accepts_ebt <- as.factor(farmers_markets$accepts_ebt)
table(farmers_markets$accepts_ebt)
## 
##   0   1 
##  10 153

Select Predictor Variables

farmers_markets <- farmers_markets %>%
  mutate(
    open_year_round = ifelse(Open.Year.Round == "Yes", 1, 0), 
    open_year_round = as.factor(open_year_round),
    distributes_health_bucks = ifelse(Distributes.Health.Bucks. == "Yes", 1, 0),
    distributes_health_bucks = as.factor(distributes_health_bucks),
    borough = as.factor(Borough)
  )
table(farmers_markets$distributes_health_bucks)
## 
##   0   1 
##  18 145
table(farmers_markets$open_year_round)
## 
##   0   1 
## 134  29
table(farmers_markets$accepts_ebt)
## 
##   0   1 
##  10 153
table(farmers_markets$borough)
## 
##         Bronx      Brooklyn     Manhattan        Queens Staten Island 
##            31            61            42            24             5

2. Methods

Model Overview:

We created three logistic regression models:

  1. Model 1: Considers only the borough of the market.

  2. Model 2: Adds open_year_round to see if year-round operation impacts EBT acceptance.

  3. Model 3: Includes distributes_health_bucks to assess its impact on EBT acceptance.

# Model 1: 
model1 <- glm(accepts_ebt ~ borough, data = farmers_markets, family = binomial)

# Model 2: 
model2 <- glm(accepts_ebt ~ borough + open_year_round, data = farmers_markets, family = binomial)

# Model 3: 
model3 <- glm(accepts_ebt ~ borough + open_year_round + distributes_health_bucks, data = farmers_markets, family = binomial)

Likelihood Ratio Tests (LRT):

  • Model 1 vs Model 2: The addition of open_year_round significantly improved the model (p-value = 0.0336).
  • Model 2 vs Model 3: Adding distributes_health_bucks also improved the model significantly (p-value < 0.0001).
  • Model 1 vs Model 3: This comparison confirmed the importance of both open_year_round and distributes_health_bucks in predicting EBT acceptance.
anova(model1, model2, test = "Chisq")
## Analysis of Deviance Table
## 
## Model 1: accepts_ebt ~ borough
## Model 2: accepts_ebt ~ borough + open_year_round
##   Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
## 1       158     67.165                       
## 2       157     62.652  1   4.5132  0.03363 *
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
anova(model2, model3, test = "Chisq")
## Analysis of Deviance Table
## 
## Model 1: accepts_ebt ~ borough + open_year_round
## Model 2: accepts_ebt ~ borough + open_year_round + distributes_health_bucks
##   Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
## 1       157     62.652                          
## 2       156     18.867  1   43.785 3.665e-11 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
anova(model3, model1, test = "Chisq")
## Analysis of Deviance Table
## 
## Model 1: accepts_ebt ~ borough + open_year_round + distributes_health_bucks
## Model 2: accepts_ebt ~ borough
##   Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
## 1       156     18.867                          
## 2       158     67.165 -2  -48.298 3.252e-11 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Interpretation

models <- list(model1, model2, model3)

# Create a table of model summaries
modelsummary(models)
(1) (2) (3)
(Intercept) 19.566 20.474 19.372
(1931.478) (3144.316) (6899.764)
boroughBrooklyn -16.604 -17.722 -19.777
(1931.478) (3144.316) (6899.764)
boroughManhattan -17.001 -18.277 -20.470
(1931.478) (3144.316) (6899.764)
boroughQueens -17.957 -19.027 -20.065
(1931.478) (3144.316) (6899.764)
boroughStaten Island -0.000 -0.211 -19.718
(5182.699) (8240.835) (22234.772)
open_year_round1 18.139 19.154
(3182.055) (7145.938)
distributes_health_bucks1 22.618
(3734.209)
Num.Obs. 163 163 163
AIC 77.2 74.7 32.9
BIC 92.6 93.2 54.5
Log.Lik. -33.583 -31.326 -9.433
F 0.756 0.521 0.037
RMSE 0.23 0.23 0.14

3. Results

The following models were fitted:

  • Model 1: Only borough as the predictor (AIC = 77.2, BIC = 92.6)
  • Model 2: Borough + open_year_round (AIC = 74.7, BIC = 93.2)
  • Model 3: Borough + open_year_round + distributes_health_bucks (AIC = 32.9, BIC = 54.5)

Key Coefficients Interpretation:

  • Borough: Markets in Bronx, Brooklyn, Manhattan, and Queens are significantly less likely to accept EBT compared to Staten Island (reference category).
  • Open Year Round: Markets that operate year-round are more likely to accept EBT, with a positive coefficient in both Model 2 and Model 3.
  • Distributes Health Bucks: The distribution of Health Bucks significantly increases the likelihood of accepting EBT. This is shown to be a key factor in Model 3.

Model Fit Comparison:

  • AIC and Log-Likelihood values improve from Model 1 to Model 3, confirming that the addition of open_year_round and distributes_health_bucks significantly improves the model’s explanatory power.
  • Model 3 has the lowest AIC and BIC, indicating the best fit.

Working with clarify

# Simulate coefficients for model3
sim_coefs <- sim(model3)

# Compute AME for 'distributes_health_bucks'
sim_est <- sim_ame(sim_coefs, var = "distributes_health_bucks", contrast = "rd", verbose = FALSE)

summary(sim_est)
##          Estimate     2.5 %    97.5 %
## E[Y(0)]  5.78e-01  1.40e-01  6.85e-01
## E[Y(1)]  1.00e+00  2.22e-16  1.00e+00
## RD       4.22e-01 -4.93e-01  6.98e-01
sim_df <- as.data.frame(summary(sim_est))

kable(sim_df, digits = 3, caption = "Simulated AME Summary") %>%
  kable_styling(full_width = FALSE)
Simulated AME Summary
Estimate 2.5 % 97.5 %
E[Y(0)] 0.578 0.140 0.685
E[Y(1)] 1.000 0.000 1.000
RD 0.422 -0.493 0.698

4. Key Takeaways

  • Borough Impact: The likelihood of a farmers’ market accepting EBT varies significantly by borough. Markets in boroughs like Bronx and Brooklyn are less likely to accept EBT, while Staten Island markets are more likely.
  • Year-Round Operation: Markets that operate year-round have a higher probability of accepting EBT, suggesting that continuous access to the market may be linked to EBT acceptance.
  • Health Bucks Distribution: The availability of Health Bucks is a major positive predictor for EBT acceptance. This shows that financial incentives for low-income consumers are an important factor in the decision to accept EBT.

5. Conclusion

Based on the analysis, I found that several factors significantly influence whether a farmers’ market accepts EBT in New York City. The borough in which the market is located was one of the most notable factors, with markets in boroughs like Brooklyn, Manhattan, and Queens being less likely to accept EBT compared to Staten Island. Additionally, markets that operate year-round and those that distribute Health Bucks show a significantly higher likelihood of accepting EBT. This suggests that location, the operational model, and financial incentives play an important role in EBT adoption at farmers’ markets.

These findings help answer my research questions by highlighting how both geographic and operational factors contribute to EBT acceptance. The hypothesis that certain boroughs would be less likely to accept EBT was supported, and the idea that year-round operation and Health Bucks distribution would increase EBT acceptance was also confirmed. This analysis provides insight into the factors that could be targeted to improve EBT access at farmers’ markets, ultimately helping to enhance food security and access to healthy food for low-income communities across NYC.

Policy Implications:

  • Targeted interventions in specific boroughs, especially Bronx and Brooklyn, could help increase the adoption of EBT at farmers’ markets.
  • Expanding year-round market operations and Health Bucks distribution may further improve accessibility to fresh produce for low-income communities.

Future Research:

Future studies could look into the impact of other socio-economic factors (e.g., average income, educational levels) or assess the role of government policies in encouraging farmers’ markets to adopt EBT.

6. References

In this assignment, the following key source was referenced to provide context, methodology, and data analysis techniques:

New York City Open Data. (2023). Farmers Markets Dataset. Retrieved from https://opendata.cityofnewyork.us/ - Used as the primary data source for the analysis of farmers’ markets in New York City and their EBT acceptance status.