Assignment 1: Census Data Quality for Policy Decisions

Evaluating Data Reliability for Algorithmic Decision-Making

Author

Sujan Kakumanu

Published

September 29, 2025

Assignment Overview

Scenario

You are a data analyst for the Ohio Department of Human Services. The department is considering implementing an algorithmic system to identify communities that should receive priority for social service funding and outreach programs. Your supervisor has asked you to evaluate the quality and reliability of available census data to inform this decision.

Drawing on our Week 2 discussion of algorithmic bias, you need to assess not just what the data shows, but how reliable it is and what communities might be affected by data quality issues.

Learning Objectives

Apply dplyr functions to real census data for policy analysis
Evaluate data quality using margins of error
Connect technical analysis to algorithmic decision-making
Identify potential equity implications of data reliability issues
Create professional documentation for policy stakeholders

Setup

library(tidyverse)
library(tidycensus)
library(knitr)

census_api_key(Sys.getenv("MY_CENSUS_API_KEY"))

my_state <- "OH"

State Selection: I have chosen Ohio for this analysis because it is my home state! I spent most of my life there and although I am excited for my new chapter in Philadelphia, the state of Ohio will always have a place in my heart.

Part 2: County-Level Resource Assessment

2.1 Data Retrieval

Data: - Geography: county level - Variables: median household income (B19013_001) and total population (B01003_001)
- Year: 2022 - Survey: acs5 - Output format: wide

county_data <- get_acs(
  geography = "county",
  variables = c(
    median_income = "B19013_001",   # Median household income
    total_pop = "B01003_001"       # Total population
  ),
  state = my_state,
  year = 2022,
  output = "wide"
)

# Clean the county names to remove state name and "County" 
words_to_remove <- c("County", "Ohio", ', ', ' ')
pattern <- paste0("\\b(", paste(words_to_remove, collapse = "|"), ")\\b")

county_data <- county_data %>%
  mutate(county_name = str_remove_all(NAME, pattern))

# Display the first few rows
glimpse(county_data)

Rows: 88
Columns: 7
$ GEOID          <chr> "39001", "39003", "39005", "39007", "39009", "39011", "…
$ NAME           <chr> "Adams County, Ohio", "Allen County, Ohio", "Ashland Co…
$ median_incomeE <dbl> 46234, 58976, 62254, 53663, 48750, 75231, 56943, 66677,…
$ median_incomeM <dbl> 4072, 2172, 2954, 2186, 3695, 2795, 2417, 3077, 1604, 3…
$ total_popE     <dbl> 27509, 102087, 52522, 97666, 61276, 46263, 66554, 43715…
$ total_popM     <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA,…
$ county_name    <chr> "Adams", "Allen", "Ashland", "Ashtabula", "Athens", "Au…

2.2 Data Quality Assessment

Your Task: Calculate margin of error percentages and create reliability categories.

Requirements: - Calculate MOE percentage: (margin of error / estimate) * 100 - Create reliability categories: - High Confidence: MOE < 5% - Moderate Confidence: MOE 5-10%
- Low Confidence: MOE > 10% - Create a flag for unreliable estimates (MOE > 10%)

Hint: Use mutate() with case_when() for the categories.

# Calculate MOE percentages
county_data <- county_data %>%
  mutate(
      median_income_moe_pct = (median_incomeM / median_incomeE) * 100,
  )

county_data <- county_data %>%
  mutate(
    median_income_moe_cat = case_when(
      median_income_moe_pct > 10 ~ "Low Confidence: MOE > 10%",
      median_income_moe_pct >= 5 & median_income_moe_pct <= 10 ~ "Moderate Confidence: MOE 5-10%",
      TRUE ~ "High Confidence: MOE < 5%"
    ),
  )

# Summary showing count of counties in each reliability category
moe_cat_summary_table <- county_data %>%
  count(median_income_moe_cat, name = "number_of_counties")

moe_cat_summary_table

# A tibble: 3 × 2
  median_income_moe_cat          number_of_counties
  <chr>                                       <int>
1 High Confidence: MOE < 5%                      57
2 Low Confidence: MOE > 10%                       2
3 Moderate Confidence: MOE 5-10%                 29

2.3 High Uncertainty Counties

Workflow: - Sort by MOE percentage (highest first) - Select the top 5 counties - Display: county name, median income, margin of error, MOE percentage, reliability category - Format as a professional table using kable()

# Create table of top 5 counties by MOE percentage
top_moe_counties <- county_data %>%
   slice_max(median_income_moe_pct, n = 5) %>%
    select(county_name, median_incomeE, median_incomeM, median_income_moe_pct, median_income_moe_cat)

kable(
  top_moe_counties,
  caption = "Top 5 Counties in Ohio by Median Income Margin of Error (MOE)",
  col.names = c("County Name", "Median Income Estimate $", "Median Income MOE $", "Median Income MOE %", "Median Income MOE Category")
)

Top 5 Counties in Ohio by Median Income Margin of Error (MOE)
County Name	Median Income Estimate $	Median Income MOE $	Median Income MOE %	Median Income MOE Category
Vinton	50967	6264	12.290306	Low Confidence: MOE > 10%
Noble	51547	5566	10.797913	Low Confidence: MOE > 10%
Monroe	55381	5415	9.777722	Moderate Confidence: MOE 5-10%
Adams	46234	4072	8.807371	Moderate Confidence: MOE 5-10%
Pike	50832	4258	8.376613	Moderate Confidence: MOE 5-10%

Data Quality Commentary:

These results are not surprising given that the counties are among the least populous in the state (Vinton, Noble, and Monroe have less than 15,000 people according to 2024 estimates). Since the ACS relies on sample data, rather than the total enumeration provided by the Census, small population groups will result in small sample sizes.

In this scenario: Any algorithmic system should account for the data inaccuracies arising from Ohio’s smaller counties.

Part 3: Neighborhood-Level Analysis

3.1 Focus Area Selection

Selection: Counties that represent different reliability levels (e.g., 1 high confidence, 1 moderate, 1 low confidence) to compare how data quality varies.

selected_counties <- county_data %>%
  filter(county_name %in% c("Cuyahoga", "Marion", "Vinton"))

# Display the selected counties with their key characteristics
# Show: county name, median income, MOE percentage, reliability category
kable(
  caption = "Selected Counties (1 in each reliability category)",
  selected_counties %>% select(county_name, median_incomeE, median_income_moe_pct, median_income_moe_cat),
  col.names = c("County Name", "Median Income Estimate $", "Median Income MOE %", "Median Income MOE Category")
)

Selected Counties (1 in each reliability category)
County Name	Median Income Estimate $	Median Income MOE %	Median Income MOE Category
Cuyahoga	60074	1.163565	High Confidence: MOE < 5%
Marion	55106	8.313069	Moderate Confidence: MOE 5-10%
Vinton	50967	12.290306	Low Confidence: MOE > 10%

Comment on the output: Cuyahoga county is one of the most populous counties in Ohio, so the sample sizes used lent to its low MOE of ~1.16%. Although this can not be extrapolated to all counties, it is interesting to see a decline of median income as the MOE increases among the selected counties.

3.2 Tract-Level Demographics

Geography: tract level
Variables: white alone (B03002_003), Black/African American (B03002_004), Hispanic/Latino (B03002_012), total population (B03002_001)
Use the same state and year as before
Output format: wide

race_vars <- c(
  white    = "B03002_003",
  black    = "B03002_004",
  hispanic = "B03002_012",
  total    = "B03002_001"
)

# Based on https://www.census.gov/programs-surveys/geography/guidance/geo-identifiers.html,
# GEOIDs for counties are made of two digit state codes and three digit county codes. This is visible
# in how GEOID in county_data stays constant for the first two digits, and is unique for the last 3.
# We can parse this and create a list:
selected_counties <- selected_counties %>%
  mutate(county_code = str_sub(GEOID, 3, 5))

county_codes <- selected_counties %>% pull(county_code)

tract_demo_selected_counties = get_acs(
  geography = "tract",
  state     = my_state,
  county    = county_codes,
  variables = race_vars,
  output    = "wide"
)

# Calculate percentage of each group using mutate()
# Create percentages for white, Black, and Hispanic populations
tract_demo_selected_counties <- tract_demo_selected_counties %>%
  mutate(
    pct_white    = 100 * whiteE    / totalE,
    pct_black    = 100 * blackE    / totalE,
    pct_hispanic = 100 * hispanicE / totalE
  )

# Add readable tract and county name columns using str_extract() or similar
tract_demo_selected_counties <- tract_demo_selected_counties %>%
  mutate(
    tract_name  = str_extract(NAME, "^[^;]+"),
    county_name = str_extract(NAME, "(?<=;\\s).*")
  )

# We can clean up the county name as we did before:
words_to_remove <- c("County", "Ohio", '; ', ' ')
pattern <- paste0("\\b(", paste(words_to_remove, collapse = "|"), ")\\b")
tract_demo_selected_counties <- tract_demo_selected_counties %>%
  mutate(county_name = str_remove_all(county_name, pattern))

3.3 Demographic Analysis

# Find the tract with the highest percentage of Hispanic/Latino residents
# Hint: use arrange() and slice() to get the top tract
top_hispanic <- tract_demo_selected_counties %>%
  slice_max(order_by = pct_hispanic, n = 1)

top_hispanic %>%
  select(county_name, tract_name, pct_white, pct_black, pct_hispanic) %>%
  kable(
    caption = "Details of Census Tract with Highest Percentage Hispanic Residents",
    col.names = c("County Name", "Tract", "% White", "% Black", "% Hispanic")
  )

Details of Census Tract with Highest Percentage Hispanic Residents
County Name	Tract	% White	% Black	% Hispanic
Cuyahoga	Census Tract 1029	18.89952	19.13876	61.48325

# Calculate average demographics by county using group_by() and summarize()
# Show: number of tracts, average percentage for each racial/ethnic group

selected_counties_summary <- tract_demo_selected_counties %>%
  group_by(county_name) %>%
  summarize(
    n_of_tracts      = n(),
    avg_pct_white = mean(pct_white,    na.rm = TRUE),
    avg_pct_black = mean(pct_black,    na.rm = TRUE),
    avg_pct_hisp  = mean(pct_hispanic, na.rm = TRUE),
  )

# Create a nicely formatted table of your results using kable()
selected_counties_summary %>%
  kable(
    caption = "Average Demographics for Selected Ohio Counties",
    digits = 2,
    col.names = c("County Name", "Number of Tracts", "Avg % White", "Avg % Black", "Avg % Hispanic")
  )

Average Demographics for Selected Ohio Counties
County Name	Number of Tracts	Avg % White	Avg % Black	Avg % Hispanic
Cuyahoga	428	51.12	34.98	6.77
Marion	18	86.13	4.75	3.50
Vinton	3	96.42	0.49	0.07

Part 4: Comprehensive Data Quality Evaluation

4.1 MOE Analysis for Demographic Variables

Workflow: - Calculate MOE percentages for each demographic variable - Flag tracts where any demographic variable has MOE > 15% - Create summary statistics

# I noticed some of the estimates or margins were 0s, or NaNs, so I'm accounting for that here as well:
tract_demo_selected_counties <- tract_demo_selected_counties %>%
  mutate(
    white_moe_pct    = if_else(whiteE    > 0, (whiteM    / whiteE), NA_real_) * 100,
    black_moe_pct    = if_else(blackE    > 0, (blackM    / blackE), NA_real_) * 100,
    hispanic_moe_pct = if_else(hispanicE > 0, (hispanicM / hispanicE), NA_real_) * 100
  )

# Create a flag for tracts with high MOE on any demographic variable
# Use logical operators (| for OR) in an ifelse() statement
tract_demo_selected_counties <- tract_demo_selected_counties %>%
  mutate(
    high_moe_flag = ifelse(
      white_moe_pct > 15 | black_moe_pct > 15 | hispanic_moe_pct > 15,
      TRUE,
      FALSE
    )
  )

# Create summary statistics showing how many tracts have data quality issues

tract_demo_selected_counties %>%
  group_by(county_name) %>%
  summarize(
    total_tracts    = sum(!is.na(high_moe_flag), na.rm = TRUE),
    high_moe_tracts = sum(high_moe_flag, na.rm = TRUE),
  ) %>%
  kable(
    caption = "Tract Data Quality by County",
    digits = 0,
    col.names = c("County Name", "Number of Tracts (excluding those without data)", "Tracts with Quality Issues (>15% MOE)")
  )

Tract Data Quality by County
County Name	Number of Tracts (excluding those without data)	Tracts with Quality Issues (>15% MOE)
Cuyahoga	421	421
Marion	18	18
Vinton	2	2

I noticed that there were some tracts with no data, so the sum of tracts excludes those.

4.2 Pattern Analysis

# Group tracts by whether they have high MOE issues
# Calculate average characteristics for each group:
# - population size, demographic percentages

# Use group_by() and summarize() to create this comparison
# Create a professional table showing the patterns

tract_demo_selected_counties %>%
  group_by(high_moe_flag) %>%
  summarize(
    avg_pct_white  = mean(pct_white, na.rm = TRUE),
    avg_pct_black  = mean(pct_black, na.rm = TRUE),
    avg_pct_hisp   = mean(pct_hispanic, na.rm = TRUE),
    n_tracts       = n(),
    .groups = "drop"
  ) %>%
  kable(
    caption = "Average Demographics by High MOE Status",
    digits = 1,
    col.names = c(
      "High MOE Flag",
      "Avg % White",
      "Avg % Black",
      "Avg % Hispanic",
      "Number of Tracts"
    )
  )

Average Demographics by High MOE Status
High MOE Flag	Avg % White	Avg % Black	Avg % Hispanic	Number of Tracts
TRUE	52.6	33.7	6.6	441
NA	98.9	0.0	0.0	8

This summary was simple, but when I did a visual inspection of the data, I see there’s some patterns that this table analysis is obscuring. I see that the MOE for Hispanic residents is above 15% for all tracts! I’ll do a second summary table, this time showing

tract_demo_selected_counties %>%
  summarize(
    n_tracts          = n(),
    white_high_moe    = sum(white_moe_pct    > 15, na.rm = TRUE),
    black_high_moe    = sum(black_moe_pct    > 15, na.rm = TRUE),
    hispanic_high_moe = sum(hispanic_moe_pct > 15, na.rm = TRUE),
    pct_white_high    = 100 * white_high_moe    / n_tracts,
    pct_black_high    = 100 * black_high_moe    / n_tracts,
    pct_hispanic_high = 100 * hispanic_high_moe / n_tracts
  ) %>%
  kable(
    caption = "Tracts with MOE > 15% by Demographic Group",
    digits = 1,
    col.names = c(
      "Total Tracts",
      "White High MOE Tracts",
      "Black High MOE Tracts",
      "Hispanic High MOE Tracts",
      "% of Tracts with White High MOE",
      "% of Tracts with Black High MOE",
      "% of Tracts with Hispanic High MOE"
    )
  )

Tracts with MOE > 15% by Demographic Group
Total Tracts	White High MOE Tracts	Black High MOE Tracts	Hispanic High MOE Tracts	% of Tracts with White High MOE	% of Tracts with Black High MOE	% of Tracts with Hispanic High MOE
449	306	416	403	68.2	92.7	89.8

tract_demo_selected_counties %>%
  group_by(county_name) %>%
  summarize(
    n_tracts          = n(),
    white_high_moe    = sum(white_moe_pct    > 15, na.rm = TRUE),
    black_high_moe    = sum(black_moe_pct    > 15, na.rm = TRUE),
    hispanic_high_moe = sum(hispanic_moe_pct > 15, na.rm = TRUE),
    pct_white_high    = 100 * white_high_moe    / n_tracts,
    pct_black_high    = 100 * black_high_moe    / n_tracts,
    pct_hispanic_high = 100 * hispanic_high_moe / n_tracts,
    .groups = "drop"
  ) %>%
  kable(
    caption = "County-Level Breakdown of Tracts with MOE > 15%, by Demographic Group",
    digits = 1,
    col.names = c(
      "County",
      "Total Tracts",
      "White High MOE Tracts",
      "Black High MOE Tracts",
      "Hispanic High MOE Tracts",
      "% of Tracts with White High MOE",
      "% of Tracts with Black High MOE",
      "% of Tracts with Hispanic High MOE"
    )
  )

County-Level Breakdown of Tracts with MOE > 15%, by Demographic Group
County	Total Tracts	White High MOE Tracts	Black High MOE Tracts	Hispanic High MOE Tracts	% of Tracts with White High MOE	% of Tracts with Black High MOE	% of Tracts with Hispanic High MOE
Cuyahoga	428	298	402	383	69.6	93.9	89.5
Marion	18	8	12	18	44.4	66.7	100.0
Vinton	3	0	2	2	0.0	66.7	66.7

Pattern Analysis: A larger share of census tracts flagged as High MOE (>15%) are driven primarily by high median income margins of error among Black or Hispanic populations. The last chart highlights that, with both communities having a higher MOE as a percentage of tracts being marked.

For Marion and Vinton counties, this is potentially explained by Black and Hispanic populations being a smaller percentage of the total. This would make the sample sizes for both communities fairly small in these counties. In Cuyahoga county, a visual inspection reveals tracts being flagged when they have small percentages of Hispanic, Black, or White population.

Part 5: Policy Recommendations

5.1 Analysis Integration and Professional Summary

Executive Summary:

Across the analyses, a pattern emerges: the MOE for demographic data is highly sensitive to both the population size of census tracts and the demographic composition within them. Small-population counties, such as Vinton County, show particularly large MOEs due to limited sample sizes in the ACS. Similarly, tracts with highly diverse populations tend to have inflated MOEs across multiple demographic groups, making nearly every group susceptible to high uncertainty in the estimates.

In examining equity implications, Black and Hispanic populations face the highest risk of unreliable data. While the largest MOE observed for White populations is around 200%, Black and Hispanic populations frequently have MOEs in the hundreds of percent and, in extreme cases, as high as 2000%. This suggests that algorithmic decisions relying on this demographic data can be highly unreliable for these communities.

We can classify the root causes of these data issues into two categories. First, small sample sizes in low-population tracts and counties amplify MOEs for all estimates. Second, demographic variability can take the form of both extremely small minority populations and highly heterogeneous tracts. This further inflates MOE. An example is Census Tract 1711.02, which is nearly 87% Black but flagged as “High MOE” because the MOE for the White population exceeds 40%.

To address these systematic issues, decision-makers should incorporate demographic composition into analyses, weighting MOEs appropriately when interpreting the demographic data. Underrepresented communities should be prioritized for follow-up investigations at the tract or county level to better understand them. By acknowledging and adjusting for these data limitations, we can improve the equity and reliability of policy decisions.

6.3 Specific Recommendations

# Create a summary table using your county reliability data
# Include: county name, median income, MOE percentage, reliability category

# Add a new column with algorithm recommendations using case_when():
# - High Confidence: "Safe for algorithmic decisions"
# - Moderate Confidence: "Use with caution - monitor outcomes"  
# - Low Confidence: "Requires manual review or additional data"

# Format as a professional table with kable()

county_data %>%
  mutate(
    algorithm_recommendation = case_when(
      median_income_moe_cat == "High Confidence: MOE < 5%"  ~ "Safe for algorithmic decisions",
      median_income_moe_cat == "Moderate Confidence: MOE 5-10%" ~ "Use with caution - monitor outcomes",
      median_income_moe_cat == "Low Confidence: MOE > 10%" ~ "Requires manual review or additional data",
      TRUE ~ "Check data"
    )
  ) %>%
  select(county_name, median_incomeE, median_income_moe_pct, median_income_moe_cat, algorithm_recommendation) %>%
  kable(
    caption = "County-Level Median Income Reliability and Algorithmic Recommendations",
    digits = 1,
    col.names = c(
      "County Name",
      "Median Income",
      "MOE % Median Income",
      "Reliability Category",
      "Algorithm Recommendation"
    )
  )

County-Level Median Income Reliability and Algorithmic Recommendations
County Name	Median Income	MOE % Median Income	Reliability Category	Algorithm Recommendation
Adams	46234	8.8	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Allen	58976	3.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Ashland	62254	4.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Ashtabula	53663	4.1	High Confidence: MOE < 5%	Safe for algorithmic decisions
Athens	48750	7.6	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Auglaize	75231	3.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Belmont	56943	4.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Brown	66677	4.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Butler	77062	2.1	High Confidence: MOE < 5%	Safe for algorithmic decisions
Carroll	59872	5.9	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Champaign	70486	5.5	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Clark	58954	2.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Clermont	79573	2.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Clinton	64210	7.1	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Columbiana	55473	4.0	High Confidence: MOE < 5%	Safe for algorithmic decisions
Coshocton	52048	5.9	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Crawford	52486	4.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Cuyahoga	60074	1.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Darke	60237	6.5	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Defiance	69302	4.8	High Confidence: MOE < 5%	Safe for algorithmic decisions
Delaware	123995	1.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Erie	65171	4.4	High Confidence: MOE < 5%	Safe for algorithmic decisions
Fairfield	82969	3.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Fayette	56773	6.2	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Franklin	71070	1.3	High Confidence: MOE < 5%	Safe for algorithmic decisions
Fulton	71453	4.5	High Confidence: MOE < 5%	Safe for algorithmic decisions
Gallia	55533	6.4	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Geauga	97162	3.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Greene	81243	3.0	High Confidence: MOE < 5%	Safe for algorithmic decisions
Guernsey	53901	7.7	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Hamilton	68249	1.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Hancock	67006	4.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Hardin	55876	5.1	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Harrison	54056	6.7	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Henry	71616	5.9	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Highland	60522	5.0	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Hocking	59007	4.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Holmes	72987	4.1	High Confidence: MOE < 5%	Safe for algorithmic decisions
Huron	64144	4.1	High Confidence: MOE < 5%	Safe for algorithmic decisions
Jackson	56549	7.8	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Jefferson	53124	4.4	High Confidence: MOE < 5%	Safe for algorithmic decisions
Knox	71246	5.2	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Lake	76835	1.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Lawrence	51846	5.2	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Licking	78505	2.8	High Confidence: MOE < 5%	Safe for algorithmic decisions
Logan	69125	4.8	High Confidence: MOE < 5%	Safe for algorithmic decisions
Lorain	67272	1.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Lucas	57265	1.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Madison	77062	4.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Mahoning	54279	2.0	High Confidence: MOE < 5%	Safe for algorithmic decisions
Marion	55106	8.3	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Medina	89968	2.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Meigs	46255	6.9	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Mercer	73278	4.5	High Confidence: MOE < 5%	Safe for algorithmic decisions
Miami	71440	3.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Monroe	55381	9.8	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Montgomery	61942	1.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Morgan	51056	7.8	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Morrow	70412	6.4	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Muskingum	56810	4.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Noble	51547	10.8	Low Confidence: MOE > 10%	Requires manual review or additional data
Ottawa	69515	5.0	High Confidence: MOE < 5%	Safe for algorithmic decisions
Paulding	65331	5.7	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Perry	62899	7.3	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Pickaway	67600	4.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Pike	50832	8.4	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Portage	69796	3.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Preble	66355	5.9	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Putnam	79453	4.0	High Confidence: MOE < 5%	Safe for algorithmic decisions
Richland	56557	3.5	High Confidence: MOE < 5%	Safe for algorithmic decisions
Ross	58048	4.1	High Confidence: MOE < 5%	Safe for algorithmic decisions
Sandusky	60814	5.0	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Scioto	46360	3.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Seneca	62476	3.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
Shelby	73502	4.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Stark	63130	1.8	High Confidence: MOE < 5%	Safe for algorithmic decisions
Summit	68360	1.3	High Confidence: MOE < 5%	Safe for algorithmic decisions
Trumbull	53537	3.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Tuscarawas	61953	3.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Union	104496	2.9	High Confidence: MOE < 5%	Safe for algorithmic decisions
VanWert	64841	5.4	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes
Vinton	50967	12.3	Low Confidence: MOE > 10%	Requires manual review or additional data
Warren	103128	2.3	High Confidence: MOE < 5%	Safe for algorithmic decisions
Washington	59053	4.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Wayne	70320	2.7	High Confidence: MOE < 5%	Safe for algorithmic decisions
Williams	60632	3.6	High Confidence: MOE < 5%	Safe for algorithmic decisions
Wood	70537	3.2	High Confidence: MOE < 5%	Safe for algorithmic decisions
Wyandot	68552	7.6	Moderate Confidence: MOE 5-10%	Use with caution - monitor outcomes

Key Recommendations:

Counties suitable for immediate algorithmic implementation: Most Ohio counties are appropriate for immediate algorithmic use; however, some caveats apply. In counties with small or underrepresented populations, certain communities may face higher margins of error, and in highly diverse counties, careful attention or additional algorithmic analysis is needed.
Counties requiring additional oversight: Adams, Athens, Carroll, Champaign, Clinton, Coshocton, Darke, Fayette, Gallia, Guernsey, Hardin, Harrison, Henry, Highland, Jackson, Knox, Lawrence, Marion, Meigs, Monroe, Morgan, Morrow, Paulding, Perry, Pike, Preble, Sandusky, Van Wert, Wyandot.

These counties can be further categorized based on population and demographic characteristics. In counties with very small populations, targeted local outreach may be sufficient to understand community needs. In larger or more diverse counties, more granular analysis at sub-county levels may be necessary to ensure accurate insights and equitable decision-making.

Counties needing alternative approaches: Noble, Vinton.

These counties have some of the lowest populations in Ohio, making traditional ACS estimates highly uncertain. Given the small population size, targeted local outreach—allowing for larger effective sample sizes than the ACS—may be the most feasible approach to accurately assess community needs.

Questions for Further Investigation

The relationship between demographics and population size in determining MOE warrants further investigation. The current analysis, which applies MOE cutoffs based on either demographic or county-level data, involves tradeoffs and may exclude or underrepresent certain populations. Developing a cutoff that incorporates both population size and demographic characteristics could provide a more accurate and equitable assessment, rather than relying single, uniform metrics.

Technical Notes

Data Sources: - U.S. Census Bureau, American Community Survey 2018-2022 5-Year Estimates - Retrieved via tidycensus R package in September of 2025 (final run conducted on September 28th 2025)

Reproducibility: - All analysis conducted in R version 4.5.1 (2025-06-13) - Census API key required for replication - Complete code and documentation available at: https://musa-5080-fall-2025.github.io/portfolio-setup-kakumanus/

Methodology Notes: Nothing of note in this analysis.

Limitations: For algorithmic decision-making at the state government level, this analysis serves as an initial assessment, providing key considerations rather than providing a final solution. Further exploration is needed, particularly because relying on ACS-based estimates can inadvertently exclude populations—especially in counties where demographic characteristics make sampling approaches less reliable.