PAF 516 Final Project: Economic Hardship Index Sensitivity Analysis

Student Instructions

1 Overview

The final project asks you to expand a baseline Economic Hardship Index (EHI) by adding one to four additional indicators and analyze how the spatial patterns, hot spot clusters, and policy implications change. This is fundamentally an exercise in measurement sensitivity analysis — a core skill in applied policy research. You are not writing new code. You are uncommenting pre-built options, re-rendering the dashboard, and writing substantive analysis of what changes and why.

The instructor dashboard uses a 3-variable baseline EHI (poverty rate, unemployment rate, median household income). Your student dashboard includes four additional variables that are commented out by default. By uncommenting one or more, you expand the index and observe how the spatial story shifts.

2 Why Index Construction Matters for Policy

2.1 Composite indices are unavoidably political

Every composite index encodes a theory of what matters. The decision to include unemployment but not educational attainment, or poverty but not housing cost burden, determines which communities appear “hardship-affected” and which do not. These are not neutral technical choices — they are value judgments with real distributional consequences. Two researchers studying the same county with different index compositions will identify different hot spots, recommend different interventions, and direct resources to different neighborhoods.

2.2 Multidimensional disadvantage

Amartya Sen’s Capability Approach and Martha Nussbaum’s development framework argue that poverty and disadvantage are inherently multidimensional. A household can be above the federal poverty line but face severe transportation barriers, food insecurity, and housing instability simultaneously. Single-indicator measures — including the federal poverty rate alone — systematically under-identify disadvantage by collapsing a complex, multi-dimensional phenomenon into a single number. Composite indices attempt to address this, but the specific dimensions chosen still constrain what the index can “see.”

2.3 Real-world stakes

This is not an academic exercise. The CDFI Fund’s Community Development Financial Institutions program, HUD’s Community Development Block Grants, and the Biden administration’s Justice40 initiative all use composite indices to allocate billions in federal resources. The EPA’s EJScreen tool determines which communities qualify for environmental justice investments. In every case, index composition determines who gets funding and who does not. A community that scores high on poverty but low on the other included dimensions may be excluded from programs that could transform residents’ lives.

2.4 The tyranny of indicators

Cathy O’Neil’s Weapons of Math Destruction documents how composite scores, once embedded in policy systems, can be gamed, can systematically exclude the most vulnerable, and can create feedback loops that entrench the very inequalities they claim to measure. When an index becomes a policy target, actors optimize for the measured dimensions while neglecting unmeasured ones. Your sensitivity analysis directly engages this problem: by changing which dimensions are measured, you observe how the “map of hardship” shifts — and which communities appear or disappear from the policy radar.

3 What You Are Doing

3.1 This is essentially a no-code assignment

The baseline dashboard (instructor version) uses a 3-variable Economic Hardship Index and is already rendered and available for you to view. Your student dashboard is identical code but with four additional variables commented out in the index-config chunk.

All you do is remove the # from one to four lines and re-render. That is the only code change required.

Everything else — all Census API pulls, all maps, all LISA spatial clustering analysis, all trajectory detection, all value boxes, all policy statistics — updates automatically based on your expanded index. This is intentional: it lets you focus on interpreting substantive changes in spatial patterns rather than debugging code.

3.2 The four additional components

Option	Variable	What It Measures	ACS Table
A	`renter_rate`	% households that rent (housing cost pressure)	B25003
B	`no_hs_rate`	% adults 25+ without a high school diploma	B06009
C	`snap_rate`	% households receiving SNAP/food stamps	B22003
D	`no_vehicle_rate`	% households with no vehicle available	B08201

Each component is automatically z-score standardized and averaged into the composite index alongside the three baseline variables. The direction of each variable is pre-configured (higher values = more hardship for all four options).

4 Step-by-Step Instructions

Download the student QMD file using the link at the bottom of this page
Open the file in RStudio
Find the STUDENT CONFIGURATION block (around lines 60–100). Look for the section bordered by ╔══════╗ characters

Uncomment the indicator lines by removing the # from lines inside the STUDENT_COMPONENTS <- c(...) block. The simplest approach is to uncomment all four:

STUDENT_COMPONENTS <- c(
    "renter_rate",       # Option A — Renter Burden
    "no_hs_rate",        # Option B — Low Educational Attainment
    "snap_rate",         # Option C — Food Insecurity (SNAP)
    "no_vehicle_rate"    # Option D — Transportation Disadvantage
)

You may also experiment by uncommenting fewer options to see how different combinations change the results. If you leave no_vehicle_rate (the last item) commented out, make sure the line above it does not end with a trailing comma — R will throw an error. For example, this is correct (no comma after "snap_rate"):

STUDENT_COMPONENTS <- c(
    "renter_rate",       # Option A — Renter Burden
#   "no_hs_rate",        # Option B — Low Educational Attainment
    "snap_rate"          # Option C — Food Insecurity (SNAP)
#   "no_vehicle_rate"    # Option D — Transportation Disadvantage
)

Save the file
Delete the _cache/ folder next to the QMD file if one exists. This is only an issue if you are rendering multiple times and changing your component selection between renders — cached data from a previous render will prevent your new variables from being pulled from the Census API. If this is your first render, there is no cache folder and you can skip this step
Click Render in RStudio (or use the terminal: quarto render Final_Project.qmd)
Wait approximately 5–10 minutes for Census API pulls to complete. The national county pull is cached and will be fast; the state and local tract pulls will re-run with your expanded variable list
Open the rendered HTML and visually compare each tab against the instructor dashboard

5 Comparison Guide

When comparing your expanded-index dashboard to the instructor baseline, focus on these specific questions for each tab:

National Context tab:

Does your county’s national rank change? By how much?
Do any Arizona counties change relative position in the state ranking table?
Does the dot plot show Arizona counties shifting left or right relative to the U.S. extremes?

Arizona in Focus tab:

Does the spatial distribution of high-hardship tracts change visually?
Which counties gained or lost hot spot status in the LISA map?
Does the Moran’s I value (spatial clustering strength) increase or decrease?

Clusters & Trajectories tab:

Do the Persistent HH corridors stay in the same locations or shift?
Does the Sankey diagram show different quintile mobility patterns?
Do the “Tracts Improved” and “Tracts Worsened” percentages change meaningfully?

Policy Implications tab:

Do the auto-computed statistics (persistent hot spots, emerging hot spots) change?
Would your policy recommendations differ based on the expanded index results?

6 What You Submit

Completed answers in the Index Sensitivity tab (3 reflection questions)
Completed policy recommendations in the Policy Implications tab (3 cards)
Your RPubs link — the public URL of your published dashboard (see below)

You do not submit a separate policy brief or HTML file. Everything is inside the dashboard, and you publish it to RPubs.

7 Publishing to RPubs

After rendering your dashboard, you need to publish it online using RPubs so the instructor can view it:

Open the rendered HTML in RStudio (it should open automatically after rendering, or click the .html file in the Files pane)
Click the Publish button (the blue publish icon in the top-right corner of the viewer pane)
If prompted to install the rsconnect package, click Yes to install it
Select RPubs as the publishing destination
Create a free RPubs account at rpubs.com if you do not already have one — you can sign up with your ASU email
Name your publication something like PAF 516 Final Dashboard (or any descriptive title)
Click Publish — your dashboard will upload and you will be redirected to its public URL
Copy the RPubs URL (it will look something like https://rpubs.com/yourusername/your-dashboard-name)
Submit the RPubs link on Canvas — paste the URL into the Canvas assignment submission so the instructor can view your published dashboard

8 Yellowdig Discussion

After completing and publishing your dashboard, post a reflection on Yellowdig describing your experience. Your post should address:

What you learned from experimenting with adding new variables to the Economic Hardship Index — which components did you add, and how did the spatial patterns change?
Implications — what surprised you about how the index shifted when you expanded its composition? Did communities appear or disappear from the hardship map?
Meaningful lessons — what does this exercise tell you about how composite indices work in practice, and why measurement choices matter for policy decisions?

This is an open-ended reflection. There is no minimum word count, but your post should demonstrate genuine engagement with the sensitivity analysis you performed.

9 Advanced Option (Optional — Not Required)

For students comfortable with R who want to explore a different geography, it is technically possible to change the target county. This requires:

Changing TARGET_STATE and TARGET_COUNTY (3 lines in the county-config chunk)
Deleting all _cache/ folders so data re-pulls for the new geography
Debugging potential tidycensus county name ambiguity errors (e.g., “Harris” in Texas matches both Harris County and Harrison County — you would need to use the full county name or FIPS code)
Accepting that the Arizona-specific bounding boxes in the mapgl calls will need manual adjustment

Recommendation: Stick with Arizona/Maricopa County and focus your analytical energy on index sensitivity. The measurement question is the point of the assignment — not geographic novelty.