Latent Dirichlet allocation model (LDA) is a textual analysis method used to understand the most important terms in a text. Before applying LDA model text data needed to be preprocessed, like using the Inverse Document Frequency (IDF) which refers to the inverse fraction of documents in the words collection which has specific term. The more this term appears, the lower the IDF score is. In this way, they get IDF score for words and by sorted them, stop words can be collected. The number of topics as K in LDA model which is modifiable, to cluster articles into K types. After brief experimentation, we settled on K = 6 due since it gave us both a good amount of keywords and is easy to interpret.

LDA allows sets of observations to be explained by unobserved groups that explain why some parts of the data are similar. If these observations are words that are collected into text files, it posits that each text file is a mixture of a small number of topics and that each word's presence is attributable to one of the document's topics. As mentioned before, we use this model to determine the important keywords in articles about LinkNYC and broadband access.

Hypothesis testing is a statistical inference method used to determine the probability that a given hypothesis is true. Z-test is a commonly used statistical test used for dealing with big samples. Z-test assumes that data follows a normal distribution.

We use a Two-sample Z-score for comparing two means to analyze if there are statistically significant differences between areas with LinkNYC kiosks and areas without LinkNYC kiosks. We calculate Z-score with equation below. We examine whether LinkNYC launch has impacted city services using the same approach.

Two-sample Z-score for comparing two means equation:

Clustering analysis is an unsupervised learning method that groups similar observations into a number of clusters. We used K-means clustering to group LinkNYC PUMAs as described in Section 2.1.3 into clusters using LinkNYC Median number of EBT-related calls done in that area and percentage of households. Tables containing descriptive statistics for both variables can be found in the figure below.We can the clusters depicted by colors as well on a corresponding map.

We find that the following areas share the same profile based on our two factors, ordered from lowest to highest means:

• Cluster 2
  • Manhattan - Lower Manhattan (less Lower East Side), All Midtown Manhattan, Upper East Side , Upper West Side
  • Brooklyn - Northwestern Brooklyn
  • Queens - (Northwestern Queens, parts of Central Queens)
  • All Staten Island
• Cluster 0
  • Manhattan - Lower East Side, Morningside Heights, West Harlem
  • Brooklyn - Northern Brooklyn, Southern Brooklyn, parts of Central and Eastern Brooklyn
  • Queens - Northeastern Queens
  • The Bronx - Northwest Bronx
• Cluster 3:
  • Manhattan - Washington Heights, East Harlem
  • Queens - Central Queens, Southeastern Queens
• Cluster 1:
  • Brooklyn - parts Eastern Brooklyn,
  • Bronx - all parts except Northwest Bronx

Based on the time series analysis and ARMA model, we looked at a possible relationship between the number LinkNYC kiosks and unemployment rates. We noticed a downward trend of monthly unemployment rate in New York as seen in the Figure 1 below (The Time Series analysis of unemployment change over time in New York). Blue line gives the change of rate of unemployment over time The red line is the trend of the data seen through the ARMA model. Then we discovered a reverse trend in the number of LinkNYC kiosk activations.

From the Figure 2 below (Changes in unemployment and LinkNYC station activation), we can see that the number LinkNYC of new LinkNYC station increases month-over-month in the timeframe of Late 2015 to Late 2018. Although, there might be some correlation between these two variables, we could not conclude a causal relationship between the variables. There is a number of external factors that explains the correlation, such as the steady growth of the US economy and predetermined incremental increase in LinkNYC stations in the city.

Pearson’s correlation coefficient: