Third-Place Decline and County Loneliness Trends Study

ISEF Category: Behavioral and Social Sciences

Subcategory: Sociology and Anthropology  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A county can have plenty of people and still feel lonely. That is the puzzle behind this project. You will test whether places like cafés, libraries, and community centers, the spots where people bump into each other without planning it, line up with lower loneliness proxies in CDC data. If the pattern holds, you have a strong social-science story built from public numbers.

What Is It?

Think of a third place as the social version of a charging station. Home is one plug, work or school is another, and cafés, libraries, and community centers are the spots where casual contact happens in between. This project asks whether counties with more of those hangout spaces tend to show less loneliness on CDC survey measures.

Yelp gives you a way to count some of those places, and BRFSS gives you a public health signal that can stand in for loneliness. A proxy is a stand-in, not the feeling itself. You are comparing patterns across counties and years to see whether social space density and loneliness move together.

Why This Is a Good Topic

This is a good science fair topic because you can turn a broad social question into a clean data test. You can measure one thing, count third places, and compare it with a public loneliness proxy at the county level. That connects to a real problem, social isolation, and lets you practice data cleaning, normalization, and regression without needing a wet lab.

Research Questions

• How does the per capita count of cafés relate to county loneliness proxy scores?
• What is the effect of libraries per capita on loneliness proxy scores after adjusting for population density?
• Does the relationship between community center density and loneliness differ between urban and rural counties?
• To what extent do changes in third-place counts from 2010 to 2023 track changes in loneliness proxies within the same county?
• Which third-place type, cafés, libraries, or community centers, has the strongest association with loneliness proxies?
• How does the strength of the association change when counties are grouped by income or age structure?

Basic Materials

• Laptop or desktop computer with reliable internet.
• Spreadsheet software such as Google Sheets or Excel.
• County-level Yelp-derived place count dataset.
• CDC BRFSS loneliness proxy dataset and codebook.
• County FIPS lookup table.
• Notes document for tracking category choices and data cleaning rules.

Advanced Materials

• Python with pandas, statsmodels, and geopandas.
• R with tidyverse, sf, and ggplot2.
• QGIS for county maps and spatial joins.
• U.S. Census county shapefiles and demographic covariates.
• Access to a university workstation or server for larger merges.
• Statistical package with fixed-effects or spatial regression tools.

Software & Tools

• Python: Cleans county files, merges datasets, and runs correlation and regression models.
• R: Builds plots, tests model variants, and formats tables for your report.
• QGIS: Maps county patterns and checks whether nearby counties look similar.
• OpenRefine: Fixes messy place names, categories, and county labels before analysis.
• Google Sheets: Helps you inspect columns, spot outliers, and draft quick charts.

Experiment Steps

1. Define the loneliness proxy you will treat as your outcome and write down why it fits the BRFSS data you have.
2. Choose which Yelp business categories count as third places, then freeze that rule before you merge anything.
3. Match counties with FIPS codes and decide how you will handle missing, merged, or tiny counties.
4. Plan a per capita normalization and one or two sensitivity checks, such as log scaling or excluding outlier counties.
5. Add control variables that could explain the pattern, such as population density, income, age mix, or urbanicity.
6. Predefine the year-by-year and region-by-region comparisons you will run so you can test whether the pattern stays stable.

Common Pitfalls

• Using raw business counts instead of per capita counts, which makes large counties look more connected just because they are large.
• Combining cafés, libraries, and community centers without checking whether each category behaves differently.
• Letting Yelp category changes or county code mismatches break your merge and silently drop records.
• Calling the association a cause-and-effect result when the design only supports correlation.
• Skipping year-to-year checks, which hides changes in Yelp coverage, BRFSS sampling, or county demographics over time.

What Makes This Competitive

A strong version of this project goes past a simple scatter plot. It compares different third-place types, adds county controls, and tests whether the result survives models that handle year and county differences. You can also split the sample by urban, rural, or income group and see where the pattern is strongest. That kind of design shows real judgment about measurement and causation limits.

Project Variations

• Compare cafés only with loneliness proxies to see whether one social venue type carries most of the signal.
• Replace county totals with per capita third-place density and test whether the pattern gets stronger in fast-growing counties.
• Add a spatial map and test whether neighboring counties show similar loneliness and third-place patterns.

Learn More

• CDC BRFSS: Find survey documentation, questionnaires, and public-use data on the CDC BRFSS page.
• CDC PLACES: Find county-level health estimates and methods notes on CDC's PLACES pages.
• U.S. Census Bureau, American Community Survey: Find county population and demographic tables on the Census website.
• USDA ERS Rural-Urban Continuum Codes: Find county rurality classifications on the USDA ERS website.
• PubMed: Search review articles on loneliness, social isolation, social capital, and built environments.
• QGIS Documentation: Find free guides for mapping counties and joining shapefiles on the QGIS website.