Executive Summary
-We attempt to analyze how gentrified areas experience different shifts in employment by industry, income, and commuting patterns.
-First, construct a model that tests for if a tract has gentrified from 2010-2019. We use several criteria to analyze these tracts, namely they must have had higher increases in incomes or the percent of college educated residents, compared to the metro area as a whole.
-Construct a Resident’s Ratio to analyze Residential Area vs Work Area Characteristics from LODES data, ie number of residents in a given industry compared to number of jobs in the area in that industry.
-We find that gentrified areas see a fast increase in jobs in the service industries, faster than the rate of residents moving into these areas in these industries, so they must start “importing” labor in these sectors.
-This increased demand for service sector jobs is bringing more commuters to the region, however these type of jobs typically are lower quality. It is interesting to consider what forces may push people to commute into these jobs- even though business is expanding in these sectors into the region, clearly there is not as much labor mobility allowing these workers to live near where they work.
-Looking at regions that saw an influx in residential jobs, we see that production and professional jobs saw notable increases in their Residents Ratio. Cathederal Park became a bedroom community for Professional Services, and Fishtown/Kensington for Production jobs. However, we see that the service industry did not change in Cathederal Park and was already an area where workers commuted in for the Service industry, and Fishtown/Kensington’s resident’s ratio for the Service industry dropped from above to below one. This means that it used to be a bedroom community for the industry, but is now a commuter community.
-Areas that remained commuter communities saw more mixed results. The Cecil B Moore South/ Chinatown area saw a jump in production residents, however it kept a small ratio in the professional industry. The service industry ratio increased, and it is approaching a value of one which means this area could soon become a bedroom community for workers in the service industry. In contrast, East Falls saw decreases in ratios across the board. Notably, it had an extremely small Production ratio that got even smaller, and the professional ratio dropped by almost half. It had a larger service ratio than even Cathedral Park of Kensington that also fell, however this was simply because more jobs moved in than residents (the number of residents increased). We see that East Falls this is a community that has most of its residents working in the professional services, however this is still only about a third of the total professional jobs that exist in the region. This means that professional jobs are moving into the area faster than actual residents are, and so the new jobs must be occupied by commuters.

Literature Review
To determine our best model of analysis, we conducted literature review on empirical papers focusing on gentrification in the labor market. We drew our main inspiration from two papers: Meltzer & Ghorbani “Does gentrification increase employment opportunities in low-income neighborhoods?” (2017), and Freeman & Braconi “Gentrification and Displacement in New York City in the 1990s” (2004). Both authors define similar but unique criteria for “gentrified” tracts before conducting analysis on the impacts of gentrification on the labor market in their specific geographies and time periods.

Meltzer & Ghorbani utilize LODES employment data, the Neighborhood Change Database, and National Establishment Time Series data to analyze changes in total “local” jobs. They create several “live-work” buffers ranging from 1/3 to 2 mile radii to capture employment levels surrounding gentrified tracts, and consider these buffers in the largest metropolitan statistical areas in the US. Controlling for MSA and state-year fixed effects, they run a series of ordinary least squares regressions to analyze the impact of a gentrification flag variable on total local jobs. Their results show a larger increase in commerical activity in gentrified areas compared to non, and that smaller live-work zones see high jobs losses compared to larger live work zones that see job gains. Conclude that gentrification has a negative impact on jobs in the immediate area, however larger zones of analysis make findings less consistent. Saw increase in goods producing and high wage jobs in the one-mile buffer, and that job gains in the 1-2 mile raidus seemed to offset the losses in the immediate live-work zones.

Freeman & Braconi was one of the most commonly cited papers we came across in our background research, namely for the method they employed to identify gentrified tracts.

Methodology
Combining the approach of Meltzer and Freeman, we construct a series of criteria for our analysis. Tracts must meet all criteria to gentrifier, but those that only meet the first two could be considered to have had the potential to gentrify but did not.

\(Measure \ Ratios: \displaystyle \frac{Tract Level Measure}{MSA Level Measure}\)

Even if a tract meets 1 and 2, if 3 and 4 are not met this means that there was not an exceptional increase in income or educational level, or positive change in home values, which would go against our assumptions about gentrification. We conduct statistically significance tests to filter out tracts that did not experience a statistically significant change in these measures.

After we select these tracts, we generate a 100 meter buffer for each tract. We utilize this buffer to include block level LODES data that may not be inside of the tract, but realistically contains jobs that effect the labor force of our selected tract.

Our resulting buffer zone as follows:

After selecting these gentrified buffers, we calculate the yearly changes in median income and education levels for all gentrified tracts and compare it to the MSA level changes in these measures:

Our results go along with our assumptions about gentrification: both education level and median home value increased much more dramatically in the gentrified tracts compared to the MSA. In addition, both measures start at below the MSA level value and jumped to far above, illustrating a change from a disadvantaged to a “gentrified” population in these areas. Notably, the education level measure surpassed the MSA level before median home value did, possibly suggesting that increases in home values are lagged by an influx of college educated residents.

LODES Employment Data

After selecting our gentrified buffers, we bring in block level LODES data to assign to each buffer. We aggregate the block level data to buffer level, and remove industries that have exceptionally low (less than 50) employment numbers for the entire buffer.

Our first step is to analyze the difference in jobs by income level for all gentrified tracts compared to Philadelphia. We first calculate the share of each income bin for all of Philadelphia, and see how this share changed over the analysis period. The bins are categorized as follows: * 1) Extremely Poor: wage less than $1,250 a month ($15,000 annual gross salary) * 2) Poor: wage between $1,250 and $3,333 a month ($15,000 to $39,996 annual gross salary) * 3) Standard: wage between $3,333 and above a month ($39,996 and above annual gross salary)

These bins are notably lack specifictity, as the standard bin is very broad.

We see that for all of Philadelphia, the Standard bin is growing in share mostly due to a strong decrease in the Poor bin. Notably, the Poor bin decreased greater than the Very Poor bin. This suggests that income gains are happening largely in those making the Poor wages, with less growth and mobility in the Very Poor.

We repeat the same analysis for the gentrified areas:

## Warning: attribute variables are assumed to be spatially constant throughout all
## geometries

We see that compared to all of Philadelphia, gentrified tracts had a much higher percent of jobs in the “Poor” category, 38% in 2010, but this bin decreased by the same rate as in all of Philly. Gentrified also tracts had a lower decrease in percent of Extremely Poor compared to all of Philadelphia, and a lower growth in Standard wages.

We see that gentrified tracts had more Poor and Very Poor jobs compared to all of Philadelphia both at the start and end of the gentrification period, however the changes in these two categories were comprable if not smaller than the changes for all of Philadelphia.In fact, while the direction of the changes was the same as in all of Philadelphia, the magnitudes for all three categories were smaller. While the dynamics of the intial period match our assumptions about gentrification, we would have expected to see larger increases in the Standard wage jobs and rapid decline in the Poor and Very Poor jobs. We can’t for certain rule this out, however, due to the lack of specificty noted above in regards to the bin levels for wages.

## Warning: attribute variables are assumed to be spatially constant throughout all
## geometries
## Warning: attribute variables are assumed to be spatially constant throughout all
## geometries

Location Quotients

In addition to income levels, we can also look at location quotients (LQ). A LQ measures the specialization of an industry in a given geography, and is calculated similar to our measure ratios: 

$Location \ Quotient:\displaystyle \frac{Portion \ of \ employment \ in \ industry \ i, \ for \ region \ n}{Portion \ of \ employment \ in \ indutry \ i\  for \ MSA}$

Where we are calculated a ratio of local employment by industry and comparing it to that ratio of the national level. A greater LQ indicates higher concentration or specialization of a given industry. </br>

Top 3 LQs did not change: Utilities, Manufacturing, and Education Manufacturing was the only one had that a positive change in the LQ Utilities, Arts and Construction actually had the biggest decreases Notably, Art went from 1.19 to 0.76.

Resident Job Ratios
Finally, we focus a great deal of our analysis on a measure we construct called “Resident Job Ratios”. We construct this ratio to express the difference of jobs by industry of the residents on a region, to the total jobs that exist in a given region regardless by establishment location. LODES employs two measures for their employment data: Residential Area Characteristics and Work Area Characteristics. By calculating a ratio of the two, we can see the ratio of local to commuter jobs and better understand the labor market dynamics of a given region. Our ratio looks as follows;

$Resident \ Jobs\  Ratio: \displaystyle \frac{Residents \ in \ industry \ i \ in \ region \ n}{total \ jobs \ in \ industry \ i \ in \ region n}$

First, we look at the total RAC/WAC ratio for all industries in each of these areas. We group the regions into three categories based on their RAC/WAC ratios in 2010 and 2019: * 1) Stayed below one: these regions started with more jobs than residents, and stayed this way * 2) Jumped from below to above one: these regions had more jobs than residents in 2010, but their ratios jumped so they now have more residents than jobs. * 3) Stayed above one: these regions started with more jobs than residents, and stayed this way.

RAC and WAC totals in each area: 

rac.wac.totals <- rac.wac.totals %>%
  mutate("Category" = ifelse(ratio.19 > 1 & ratio.10 < 1, "Jumped from below to above 1", "na"),
         "Category" = ifelse(ratio.19 < 1 & ratio.10 < 1, "Stayed below 1", Category),
         "Category" = ifelse(ratio.19 > 1 & ratio.10 > 1, "Started & Stayed above 1", Category),
  ) %>%
  left_join(both.buff, by = "names") %>%
  st_as_sf()

tm_view()
## $tm_layout
## $tm_layout$style
## [1] NA
## 
## 
## attr(,"class")
## [1] "tm"
tm_shape(rac.wac.totals) +
  tm_polygons(col = "Category") +
  tm_text("names")