LawFirmHW10/13Revision
Jonah Dratfield
2025-10-10
Introduction
I am a data analyst working for a law firm that analyzes NYC parking violation data. In the following analysis, I explore whether different agencies issue meaningfully different fine amounts, whether drivers from different states generally pay meaningfully different fine amounts, and whether certain counties tend to issue meaningfully different fine amounts. You can look through the dataset I am using here
endpoint<-"https://data.cityofnewyork.us/resource/nc67-uf89.json"
resp <- GET(endpoint, query = list(
"$limit" = 99999,
"$order" = "issue_date DESC"
))
cameradata <- fromJSON(content(resp, as = "text"), flatten = TRUE)
cameradata$payment_amount <- as.numeric(gsub("[^0-9.]", "", cameradata$payment_amount))Issuing Agency Analysis
ggplot(cameradata, aes(x = issuing_agency, y = payment_amount)) +
geom_boxplot() +
coord_flip()
favstats(payment_amount ~ issuing_agency, data = cameradata) %>%
arrange(desc(mean))## issuing_agency min Q1 median Q3 max
## 1 HEALTH DEPARTMENT POLICE 243.81 243.810 243.81 243.8100 243.81
## 2 SEA GATE ASSOCIATION POLICE 190.00 190.000 190.00 190.0000 190.00
## 3 FIRE DEPARTMENT 180.00 180.000 180.00 180.0000 180.00
## 4 NYS OFFICE OF MENTAL HEALTH POLICE 0.00 180.000 180.00 190.0000 210.00
## 5 ROOSEVELT ISLAND SECURITY 0.00 135.000 180.00 190.0000 246.68
## 6 PORT AUTHORITY 0.00 180.000 180.00 190.0000 242.76
## 7 NYS PARKS POLICE 0.00 45.000 180.00 190.0000 242.58
## 8 PARKS DEPARTMENT 0.00 90.000 180.00 190.0000 245.28
## 9 TAXI AND LIMOUSINE COMMISSION 125.00 125.000 125.00 125.0000 125.00
## 10 HEALTH AND HOSPITAL CORP. POLICE 0.00 0.000 180.00 190.0000 245.64
## 11 POLICE DEPARTMENT 0.00 0.000 180.00 190.0000 260.00
## 12 CON RAIL 0.00 0.000 95.00 228.8875 243.87
## 13 DEPARTMENT OF TRANSPORTATION 0.00 50.000 75.00 125.0000 690.04
## 14 TRAFFIC 0.00 65.000 115.00 115.0000 245.79
## 15 OTHER/UNKNOWN AGENCIES 0.00 40.115 80.23 120.3450 160.46
## 16 TRANSIT AUTHORITY 0.00 0.000 75.00 125.0000 190.00
## 17 SUNY MARITIME COLLEGE 65.00 65.000 65.00 65.0000 65.00
## 18 NYC OFFICE OF THE SHERIFF 0.00 28.750 57.50 86.2500 115.00
## 19 DEPARTMENT OF SANITATION 0.00 0.000 65.00 105.0000 115.00
## 20 LONG ISLAND RAILROAD 0.00 0.000 0.00 0.0000 0.00
## mean sd n missing
## 1 243.81000 NA 1 0
## 2 190.00000 0.00000 2 0
## 3 180.00000 NA 1 0
## 4 161.33333 65.99423 15 0
## 5 149.16083 90.57967 24 0
## 6 147.35792 82.58394 48 0
## 7 143.86176 89.24158 34 0
## 8 128.47736 78.92728 144 0
## 9 125.00000 NA 1 0
## 10 124.71373 98.60130 51 0
## 11 123.93855 88.00388 214 0
## 12 112.62000 124.87146 6 0
## 13 99.52822 82.88394 87273 0
## 14 94.59362 44.47453 12091 0
## 15 80.23000 113.46235 2 0
## 16 78.00000 82.05181 5 0
## 17 65.00000 NA 1 0
## 18 57.50000 81.31728 2 0
## 19 56.78571 48.26239 14 0
## 20 0.00000 NA 1 0anova_model <- aov(payment_amount ~ issuing_agency, data = cameradata)
summary(anova_model)## Df Sum Sq Mean Sq F value Pr(>F)
## issuing_agency 19 937675 49351 7.858 <2e-16 ***
## Residuals 99910 627464684 6280
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 69 observations deleted due to missingnesssupernova(anova_model)## Analysis of Variance Table (Type III SS)
## Model: payment_amount ~ issuing_agency
##
## SS df MS F PRE p
## ----- --------------- | ------------- ----- --------- ----- ----- -----
## Model (error reduced) | 937675.432 19 49351.339 7.858 .0015 .0000
## Error (from model) | 627464683.951 99910 6280.299
## ----- --------------- | ------------- ----- --------- ----- ----- -----
## Total (empty model) | 628402359.383 99929 6288.488The model sum of squares, 937,675.432, is a miniscule proportion of the total sum of squares. Therefore, it only accounts for a very small fraction of the variance. F(19, 99,910) = 7.86, p < .001. This is a very small p value, so it is statistically significant. The PRE is .0015, so it accounts for less than 1% of the variance. Consequently, it is probably not worthwhile for the law firm to take into account issuing agency when creating their marketing strategy (unless they have some underlying philosophical or ideological reason for doing so, divorced from these statistical analyses).
Driver State Analysis
threestatefilter <- cameradata %>%
filter(state %in% c("NY", "NJ", "CT"))
view(threestatefilter)
ggplot(threestatefilter, aes(x = state, y = payment_amount)) +
geom_boxplot() +
coord_flip()
favstats(payment_amount ~ state, data = threestatefilter) %>%
arrange(desc(mean))## state min Q1 median Q3 max mean sd n missing
## 1 NJ 0 50 75 115 682.35 101.5746 89.97170 8654 3
## 2 NY 0 50 75 125 690.04 101.0902 80.93015 79541 10
## 3 CT 0 50 75 100 276.57 80.6627 46.07849 1457 2anova_model <- aov(payment_amount ~ state, data = threestatefilter)
summary(anova_model)## Df Sum Sq Mean Sq F value Pr(>F)
## state 2 602716 301358 45.48 <2e-16 ***
## Residuals 89649 594098897 6627
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 15 observations deleted due to missingnesssupernova(anova_model)## Analysis of Variance Table (Type III SS)
## Model: payment_amount ~ state
##
## SS df MS F PRE p
## ----- --------------- | ------------- ----- ---------- ------ ----- -----
## Model (error reduced) | 602716.142 2 301358.071 45.475 .0010 .0000
## Error (from model) | 594098896.889 89649 6626.944
## ----- --------------- | ------------- ----- ---------- ------ ----- -----
## Total (empty model) | 594701613.031 89651 6633.519The model sum of squares, 602,716.142, is a small proportion of the total sum of squares. Therefore, state only accounts for a small fraction of the overall variance in payment amount. F(2, 89,649) = 45.48, p < .001. This is clearly statistically significant due to the low p-value. However, the PRE is .0010, meaning that the model explains less than 0.1% of the variance. Consequently, it is probably not worthwhile for the marketing firm to emphasize state in their campaign.
County Analysis
cameradata <- cameradata %>%
mutate(county = case_when(
county %in% c("K", "BK", "Kings") ~ "Kings County",
county %in% c("Q", "QN", "Qns") ~ "Queens County",
county %in% c("BX", "B", "Bronx", "BRONX") ~ "Bronx County",
county %in% c("R", "ST", "SI", "RICH") ~ "Richmond County",
county %in% c("NY", "N", "MN") ~ "New York County",
TRUE ~ county
))
ggplot(cameradata, aes(x = county, y = payment_amount)) +
geom_boxplot() +
coord_flip()
favstats(payment_amount ~ county, data = cameradata) %>%
arrange(desc(mean))## county min Q1 median Q3 max mean sd n missing
## 1 Richmond County 0 50 125 180 250.00 114.53669 77.55385 1349 0
## 2 Kings County 0 50 75 115 690.04 110.88983 126.20448 16112 0
## 3 Bronx County 0 65 75 145 245.64 99.65870 67.53373 247 0
## 4 New York County 0 50 75 115 281.80 97.62502 62.55866 23479 0
## 5 Queens County 0 50 50 100 283.03 83.46501 60.08515 17366 0anova_model <- aov(payment_amount ~ county, data = cameradata)
summary(anova_model)## Df Sum Sq Mean Sq F value Pr(>F)
## county 4 6702478 1675619 233.4 <2e-16 ***
## Residuals 58548 420413471 7181
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 41446 observations deleted due to missingnesssupernova(anova_model)## Analysis of Variance Table (Type III SS)
## Model: payment_amount ~ county
##
## SS df MS F PRE p
## ----- --------------- | ------------- ----- ----------- ------- ----- -----
## Model (error reduced) | 6702477.756 4 1675619.439 233.352 .0157 .0000
## Error (from model) | 420413471.103 58548 7180.663
## ----- --------------- | ------------- ----- ----------- ------- ----- -----
## Total (empty model) | 427115948.859 58552 7294.643The model sum of squares, 6,702,477.756, is a small proportion of the total sum of squares. It, therefore, only accounts for a small fraction of the variance. F(4, 58,548) = 233.35, p < .001. This, once again, is a very a small p-value and is statistically significant. PRE is 1.57%, so 1.57% of the overall variance is explained by county. It is, therefore, likely not a meaningful metric for the law firm to use in their marketing campaign.
Overall Conclusion
From a statistical standpoint, I can not, in good conscience, recommend that the law firm use any of the three variables explored in this analysis as central components of their marketing campaign. Though all of these variables are statistically significant, none of them account for more than 2% of the overall variance. That being said, the firm, if it had to use one of these variables, should choose county (since it is the only one that accounts for more than 1% of the variance). Violation type (as explored in a previous analysis) accounts for 33% of the variance, so is the best metric for the firm to use in their campaign.
```