Description
Goal: In this assignment, you will compute the Townsend Material Deprivation Index for
census tracts in New York County (i.e., Manhattan) using American Community Survey
(ACS) data (administered by the U.S. Census Bureau).
The Townsend index is constructed from four variables: % unemployment among people
16 and over (unemp), % of overcrowded households defined as homes with more than one
person per room (oc), % of housing which is rented (rent), and % of households without a
vehicle (car). A higher percentage of any of these variables indicates more deprivation in a
geographic area (Note: you should not convert the percentage variables into decimal form).
Let geographic region be denoted by the subscript i and variable by j. To compute the
index, follow the four steps described next.
1For more information: (a) Townsend, P., Phillimore, and P., Beattie, A. (1988). Health and deprivation:
inequalities and the north. Croom Helm, London. (b) Carstairs, V. and Morris, R. (1991). Deprivation
and health in Scotland. Aberdeen University Press, Aberdeen. (c) Jarman, B. (1983). Identification of
underprivileged areas. BMJ 286 1705-1709. (d) Jordon, H., Roderick, P., and Martin, D. (2004). The index
of multiple deprivation 2000 and accessibility effects on health. Journal of Epidemiology and Community
Health. 58 250-257.
1
(a) transform the variables to reduce skewness (note: log is natural log):
ti,unemp = log(xi,unemp + 1)
ti,oc = log(xi,oc + 1)
ti,rent = log(xi,rent + 1)
ti,car =
√
xi,car
(b) compute geographic region mean tj and standard deviation sj
for each variable j.
(c) standardize variables for each region i and variable j: zij =
tij−tj
sj
(d) compute sum of standardized variables for each region i: T ownsendi =
P4
j=1 zij
You will have one Townsend index value for each region i. These values are on a relative
scale; that is, the actual number is meaningless, just the ranking is important. A negative
score indicates a less deprived region; a score of 0 indicates roughly the average level of
deprivation; a positive score indicates a more deprived region.
1. What is a census tract? How many census tracts are in New York County? (Provide
the citations for references used.)
2. Describe one advantage and one disadvantage of computing estimates after combining
5-years of data.
3. Download the ACS data for 5-year estimates spanning from 2011-2015. (This means
2015 5-year estimates, not a combination of 1-year estimates for 2011, 1-year estimates
for 2012, etc.) for for all New York County census tracts for the following variables
using American FactFinder (link; further instructions can be found on Blackboard in
the file “Downloading Map and ACS Data”). Table DP03 contains selected economic
characteristics and Table DP04 includes selected housing characteristics. Each row in
the table represents a single census tract in New York County.
• unemployment: Table DP03, variable HC03 VC07
• housing tenure (whether house is rented or owned): Table DP04, variable HC03 VC66
• no vehicles: Table DP04, variable HC03 VC85
• low occupancy: Table DP04, variable HC03 VC113 (You will have to transform
this variable to get % overcrowded to use in the index)
Clean the data and merge the tables into one data frame, each row representing a census
tract, each column representing one of the Townsend variables (keep the geography
columns).
Page 2 of 3
For each variable, construct a histogram and compute the following summary statistics:
mean, median, standard deviation, maximum, and minimum. Describe the shape of
each histogram.
4. How many observations are missing for each variable? What percentage of census tracts
do not have complete data? Is this a problem for our analysis? Justify your answer.
(Note: do not delete tracts with missing data.)
5. Construct scatterplots of the four variables. Are they linearly related? Now, transform
the variables as given in step (a), adding the transformed variables to your data frame.
Make another scatter plot matrix with the transformed variables. Are they linearly
related? Construct a correlation matrix of the transformed variables and describe your
results.
6. Compute the Townsend index value for each census tract. Add the index to your data
frame. For how many census tracts are you able to compute the Townsend index? Why
does this number not equal the total number of census tracts?
7. Identify which census tract is the most deprived and which is the least deprived (give
the census tract number and deprivation index level). Based on your results, would you
like to live in the least deprived census tract? Justify your answer.
8. The ACS data includes not only estimates but their margins of error which we ignored
in our calculations. What are the implications?
9. Construct a map color-coded by the deprivation index value quintile. Each quintile
(i.e., 20%) should be assigned a different color from least to most deprived. Download
the shape files for New York state census tracts for 2015 from the U.S. Census Bureau
website (link; further instructions can be found on Blackboard in the file “Downloading
Map and ACS Data”). Extract the tracts for New York County only. Include a legend
and plot title. Describe the patterns you see, especially in relation to what you know
about neighborhoods in New York City. What does the large rectangle in the middle of
the map represent?
10. In which census tract is 140 W. 62nd St. (where we have class)? What is the deprivation
level rank (where a rank of 1 is the most deprived)? Mark it on the map (use the
computer, not by hand) and add it to your legend. (Provides citations for references.)
11. New York County is an urban county, however New York state has roughly 22 counties
classified as rural (e.g., Allegany, Essex, Otsego, Sullivan). Would it make sense to compute the Townsend index values for all census tracts within New York state combined?
Why or why not?
Page 3 of 3
