Some stats about the number of POIs by category

skim(df_poi_cnt_by_cat)

Variable type: character

Variable type: numeric

ggplot(df_restaurant_cnt) +
  geom_histogram(aes(x = num_pois), bins = 20) + 
  labs(title = "Histogram POIs in Restaurant Category", x = "Number of POIs", y = "Count") +
  theme_minimal()

ggplot(df_shopping_cnt) +
  geom_histogram(aes(x = num_pois), bins = 20) + 
  labs(title = "Histogram POIs in Shopping Category", x = "Number of POIs", y = "Count") +
  theme_minimal()

From the stats and histogram, there are many categories that have small number. Let take a look at top 10 categories by business type.

Restaurants

ggplot(df_restaurant_cnt %>% 
         arrange(desc(num_pois)) %>% 
         slice(1:10), 
       aes(y = reorder(categories_title, num_pois), 
           x = num_pois)) +
  geom_bar(stat = "identity") + 
  labs(title = "Number of POIs in Restaurant Category", x = "Categories", y = "Number of POIs") +
  theme_minimal()

The top 10 of restaurant business POIs are shown above. I notice that there is a parks category which might be due to restaurants also provide parking lots for customers so in Yelp’s data they also record them like this. In further analysis, we should remove this category to prevent misunderstanding.

Shopping

ggplot(df_shopping_cnt %>% 
         arrange(desc(num_pois)) %>% 
         slice(1:10), 
       aes(y = reorder(categories_title, num_pois), 
           x = num_pois)) +
  geom_bar(stat = "identity") + 
  labs(title = "Number of POIs in Shopping Category", x = "Categories", y = "Number of POIs") +
  theme_minimal()

For the shopping business POIs, the top most is woman’s clothing which more than 40 points out of 398 shopping POIs.

Distribution of rating by price

# Restaurant
ggplot(city_poi_restaurant_sf %>% filter(!is.na(price))) +
  geom_density(aes(x = rating, fill = price), alpha = 0.3) +
  labs(title = "Rating Distribution of Restaurant Business POIs by Price") +
  theme_minimal()

## Warning: Groups with fewer than two data points have been dropped.

## Warning in max(ids, na.rm = TRUE): no non-missing arguments to max; returning
## -Inf

# Shopping
ggplot(city_poi_shopping_sf %>% filter(!is.na(price))) +
  geom_density(aes(x = rating, fill = price), alpha = 0.3) +
  labs(title = "Rating Distribution of Shopping Business POIs by Price") +
  theme_minimal()

For the restaurant POIs, there are less than 2 data points for $$$ price so the plot is not shown. For $ price restaurant, the rating distribution is concentrated in the lower range of the rating scale, peaking between 1.5 and 2.5. This indicates that lower-priced restaurants tend to receive lower ratings overall. For the middle price tier $$, the density shows a peak between ratings of 2.5 and 3. This suggests that moderately priced restaurants are more likely to receive higher ratings compared to the lowest-priced category.

For the shopping POIs, Similar to the restaurant category, shopping businesses with higher prices $$$ tend to receive more favorable ratings, showing a strong association between price and quality perception. Moderately priced shopping businesses also tend to receive favorable ratings, but with a slightly more diverse rating distribution compared to the highest price tier.

Load Household Income at Census Tract Level

tract_data <- suppressMessages(
  get_acs(geography = 'tract',
          state = state,
          county = county,
          variables = c(hhincome = "B19013_001",
                        pop = "B01003_001"),
          year = 2022,
          survey = 'acs5', 
          geometry = TRUE,
          output = 'wide'))

## Warning: • You have not set a Census API key. Users without a key are limited to 500
## queries per day and may experience performance limitations.
## ℹ For best results, get a Census API key at
## http://api.census.gov/data/key_signup.html and then supply the key to the
## `census_api_key()` function to use it throughout your tidycensus session.
## This warning is displayed once per session.

Convert Tract CRS to EPSG4326

epsg_id <- 4326
tract_data <- tract_data %>% 
  st_transform(crs = epsg_id)

Filter POIs by each tract

tract_pois <- st_join(tract_data, city_poi_sf, join=st_intersects)
print(paste0("Number of all POIs: ", nrow(city_poi_sf)))

## [1] "Number of all POIs: 708"

print(paste0("Number of records after spatial join: ", nrow(tract_pois)))

## [1] "Number of records after spatial join: 717"

There are some tracts don’t have POIs.

Remove tracts that don’t have POIs

tract_pois_clean <- tract_pois %>%
  filter(!is.na(id))

print(paste0("Number of records after cleaning: ", nrow(tract_pois_clean)))

## [1] "Number of records after cleaning: 708"

Aggregate rating and count number of POIs group by business category and tract

tract_pois_clean_agg <- tract_pois_clean %>%
  st_drop_geometry() %>%
  group_by(GEOID, main_category, hhincomeE, popE) %>%
  summarise(num_pois = n_distinct(id), med_rating = median(rating), avg_rating = mean(rating)) %>%
  ungroup()

## `summarise()` has grouped output by 'GEOID', 'main_category', 'hhincomeE'. You
## can override using the `.groups` argument.

Plot Scatter Plots between Household Income, Population and rating and number of POIs

ggplot(tract_pois_clean_agg, aes(x=hhincomeE, y=avg_rating, color=main_category)) +
  geom_point() +
  labs(title="Household Income vs Avg Rating")+
  theme_minimal()

ggplot(tract_pois_clean_agg, aes(x=hhincomeE, y=num_pois, color=main_category)) +
  geom_point() +
  labs(title="Household Income vs Number of POIs")+
  theme_minimal()

ggplot(tract_pois_clean_agg, aes(x=popE, y=avg_rating, color=main_category)) +
  geom_point() +
  labs(title="Population vs Avg Rating")+
  theme_minimal()

ggplot(tract_pois_clean_agg, aes(x=popE, y=num_pois, color=main_category)) +
  geom_point() +
  labs(title="Population vs Number of POIs")+
  theme_minimal()

From 4 scatter plots above there is no association between the household income to avg. rating and number of POIs, and between population to avg. rating and number of POIs. Because the sample size is small (only 19 census tracts) and the POIs categories may not be relevant. Some others POIs categories may exhibit relationship between these variables.