Mobile Dashboard Under Construction.
Evaluating policy transfer via similarity analysis and causal inference.
Public policymaking is hard. Each day, local elected officials and policy analysts across the country devise programs to improve the social and economic conditions of their communities, but face the obstacle that the success of a policy hinges on a panoply of factors that are challenging to predict. In this context, historical precedent is the ultimate guide: an intervention that was effective in achieving an outcome in one community is likely to also be successful in other, similar communities. But what does “similar” mean?

polis is a tool that allows users to compute how similar jurisdictions are to each other based on customizable metrics, such as GDP trends or income distribution, with the aim of identifying “sister communities” in which policy outcomes are more likely to transfer. The tool supports the following types of questions:
  1. Similarity: Which communities are similar to mine, based on a user-defined set of social and economic variables, weighted by importance?
  2. Similar-But-Different-Outcome: Which communities are similar to mine according to a set of variables, but achieved a different outcome than mine on some dimension?
  3. Counterfactual Analysis: If a particular policy intervention that was missed in my community had actually been made, what would its impact have been, based on similar communities' experiences?
In addition, the tool supports questions of general data exploration (Explorer). Check out examples and a tutorial for each of these functionalities below, or try it out here.

Similarity Tool Tutorial
The similarity tool identifies similar jurisdictions for a selected base jurisdiction. Hover over the features below to highlight them.
County-Metro Selector: Toggles whether the analyzed jurisdictions are counties or metropolitan statistical areas.
Base Jurisdiction Input: Selects a jurisdiction w.r.t. which similarity scores are calculated.
Variable Set Selector: Toggles view between showing the basic variable set, containing just five core comparison variables, and the advanced variable set, containing every variable.
Importance Weights: Selects an importance level for each possible comparison variable. Each slider has five weights: The highest weight indicates a variable is very important for identifying similar counties, the middle weight indicates it is somewhat important, and the lowest weight indicates it is not used in the computation.
Map-Plot View: Toggles whether to display the map, where brighter colors indicate higher similarity scores, or plots of individual variables.
Region Filter: Selects a region of the US over which to constrain the similarity computation.
Ranking Panel: Shows the top 20 most similar jurisdictions based on the selected settings. Hover on a jurisdiction name to highlight it on the map, click to zoom in on it, and click again to reset the zoom.
Examples
The following are sample questions that the similarity tool can answer. Click to open the similarity tool with the pre-filled query.
Q: Which counties are most similar to Wayne County, MI (Detroit) based on racial demographics?
  1. St. Louis, MO
  2. Pulaski, AR
  3. Pitt, NC
  4. Pike, AL
  5. Guilford, NC
  6. Chatham, GA
  7. Richland, SC
  8. East Baton Rouge, LA
  9. Lincoln, LA
  10. Iberia, LA
Open in Similarity Tool ↗
Q: Which counties are most similar to San Francisco County, CA based on population trends?
  1. Baltimore, MD
  2. Middlesex, NJ
  3. Worcester, MA
  4. Montgomery, PA
  5. Ventura, CA
  6. Essex, NJ
  7. Macomb, MI
  8. New Haven, CT
  9. El Paso, TX
  10. Pierce, WA
Open in Similarity Tool ↗
Q: Which Midwestern counties are most similar to Bernalillo County, NM (Albuquerque) based on urbanization, industry composition, and GDP trends?
  1. Lancaster, NE
  2. Douglas, KS
  3. Sangamon, IL
  4. Sedgwick, KS
  5. DeKalb, IL
  6. Monroe, IL
  7. Pennington, SD
  8. Peoria, IL
  9. St. Louis, MN
  10. Greene, MO
Open in Similarity Tool ↗
Similar But Different Outcome Tool Tutorial
The similar-but-different-outcome tool identifies jurisdictions that were similar to the base jurisdiction in the past, but then diverged along some dimension. In particular, it is useful for finding places that achieved a desirable outcome – such as higher GDP or lower unemployment – that one's own jurisdiction has not achieved, despite previously having similar trajectories.

The feature is activated by clicking the up arrow (hover to view) beside the Alternative Paths title at the bottom-left of the similarity tool. This updates the map so that only jurisdictions that have achieved the desired outcome – e.g. GDP above the 85th percentile – are highlighted, and others removed. The heatmap colors indicate the similarity score of each of these different-outcome jurisdictions with respect to the base jurisdiction.
Map-Plot View: Toggles whether to show the plot view or the map view. This plot shows that Adams County, PA experienced a similar GDP trend to many counties for some time, but then fell off while the others continued to improve.
Year Slider: Selects the year up until which the base jurisdiction and another jurisdiction are similar.
Outcome Variable Selector: Selects outcome dimension along which the base jurisdiction and other jurisdictions diverge.
Outcome Percentile Range: Selects the percentile range into which other jurisdictions must fall with respect to the chosen outcome variable. The red vertical line on the slider indicates the base jurisdiction's percentile for the outcome variable. For example, Adams County (A.C.), PA falls into the 34th percentile for GDP.
Examples
The following are sample questions that the similar-but-different-outcome tool can answer. Click to open the similarity tool with the pre-filled query.
Q: Which counties are most similar to Tulsa, OK based on sci/tech sector employment before 2019, but have a larger sci/tech sector today?
  1. Ramsey, MN
  2. Douglas, NE
  3. Baltimore, MD
  4. Brevard, FL
  5. Plymouth, MA
  6. Somerset, NJ
  7. Virginia Beach, VA
  8. Union, NJ
  9. Pierce, WA
  10. Kent, MI
Open in Similarity Tool ↗
Q: Which Southeast counties are most similar to Claiborne, LA based on GDP and population before 2015, but have higher GDP today?
  1. Hoke, NC
  2. Twiggs, GA
  3. McDuffie, GA
  4. Obion, TN
  5. Calhoun, AR
  6. Wilkinson, GA
  7. Taylor, GA
  8. Pulaski, VA
  9. Liberty, FL
  10. Page, VA
Open in Similarity Tool ↗
Q: Which counties are most similar to Brunswick, VA based on GDP and racial demographics before 2019, but have lower GDP today?
  1. Wilcox, AL
  2. Macon, GA
  3. Ware, GA
  4. Macon, MO
  5. King, TX
  6. Evans, GA
  7. Montmorency, MI
  8. Montgomery, GA
  9. Vilas, WI
  10. Haines Borough, AK
Open in Similarity Tool ↗
Counterfactual Tool Tutorial
The counterfactual tool combines similarity analysis with historical policy intervention data to predict what would have happened in a particular jurisdiction if a policy had been implemented in the past. In other words, it considers the counterfactual scenario. The tool allows the following variables to be set.
Jurisdiction Input: Selects a jurisdiction of interest.
Outcome Variable Selector: Selects an outcome variable of interest, e.g. unemployment rate.
Intervention Variable Selector: Selects intervention type of interest, e.g. U.S. Department of Commerce spending.
Intervention Year Slider: Selects the year when the hypothetical intervention occurred.
Intervention Level Selector: Selects the hypothetical intervention variable's value as a percentile compared to values for that intervention across the country. The red vertical line on the slider indicates the base jurisdiction's actual percentile for the intervention variable.
Map-Plot View: Toggles whether to display the plot, which shows the predicted impact on the selected county, or the map, which displays the predicted impact of the intervention in all counties in the same state, for comparison.
Examples
The following are sample questions that the counterfactual tool can answer. Click to open the similarity tool with the pre-filled query.
Q: What is the predicted GDP change in Schuykill County, PA in 2018 if the Department of Commerce awarded 5 billion dollars in federal funding to the county in 2015, instead of 0 dollars? A: +2.68 (Chain-Type GDP).
Open in Counterfactual Tool ↗
Q: What is the predicted change in number of science/tech jobs in Tulsa County, OK in 2021 if the Department of Commerce awarded 107 million dollars in federal funding to the county in 2018, instead of 6 million dollars? A: +1,428 jobs.
Open in Counterfactual Tool ↗
Q: What is the predicted number of transportation jobs in Suffolk County, MA in 2021 if the Department of Transportation awarded 320 million dollars in federal funding to the county in 2018, instead of 23 million dollars? A: +2,225 jobs.
Open in Counterfactual Tool ↗
Explorer Tool Tutorial
The explorer tool visualizes the raw datasets used to perform the calculations in the similarity, similar-but-different-outcome, and counterfactual tools. By explicitly showing the jurisdictions with the highest and lowest values for each variable (i.e. the extrema of the dataset), it provides key context in which the results from the other tools can be interpreted.
Variable Selector: Selects the variable to display on the map.
Ranking Order Toggle: Selects whether the ranking panel orders jurisdictions by variable value from highest-to-lowest (default) or lowest-to-highest.
Examples
The following are sample questions that the explorer tool can answer. Click to open the explorer tool with the pre-filled query.
Q: Which US counties have the largest percentage of residents making more than 150k dollars in annual income?
  1. Loudoun, VA: 52%
  2. Santa Clara, CA: 47%
  3. San Mateo, CA: 46%
  4. Marin, CA: 44%
  5. San Francisco, CA: 43%
  6. Howard, MD: 43%
  7. Nassau, NY: 42%
  8. Arlington, VA: 42%
  9. Morris, NJ: 41%
  10. Hunterdon, NJ: 40%
Open in Explorer Tool ↗
Q: Which US counties have the highest Puerto Rican population by percentage?
  1. Osceola, FL
  2. Hampden, MA
  3. Bronx, NY
  4. Cumberland, NJ
  5. Orange, FL
  6. Lehigh, PA
  7. Hartford, CT
  8. Berks, PA
  9. Polk, FL
  10. Montgomery, NY
Open in Explorer Tool ↗
Q: Which Southeast counties have the highest rates of child poverty?
  1. Sunflower, MS
  2. Issaquena, MS
  3. Concordia, LA
  4. Coahoma, MS
  5. Sharkey, MS
  6. Greene, AL
  7. East Carroll, LA
  8. Perry, AL
  9. Quitman, MS
  10. Wilcox, AL
Open in Explorer Tool ↗
Interested? We’d love to hear from you.
polis is a research tool under very active development, and we are eager to hear feedback from users in the policymaking and public administration spaces to accelerate its benefit.

If you have feature requests, recommendations for new data sources, tips for how to resolve missing data issues, find bugs in the tool (they certainly exist!), or anything else, please do not hesitate to contact us at polis@basis.ai.

To stay up to date on our latest features, you can subscribe to our mailing list. In the near-term, we will send out a notice about our upcoming batch of improvements (including performance speedups, support for mobile, and more comprehensive tutorials), as well as an interest form for users who would like to work closely with us on case studies to make the tool most useful in their work.

Lastly, we emphasize that this website is still in beta testing, and hence all predictions should be taken with a grain of salt.

Acknowledgments: polis was built by Basis, a non-profit AI research organization dedicated to creating automated reasoning technology that helps solve society's most intractable problems. To learn more about us, visit https://basis.ai.