Skip to content

Analyze the margin of error in U.S. census data

License

Notifications You must be signed in to change notification settings

datadesk/census-error-analyzer

Repository files navigation

census-error-analyzer

Analyze the margin of error in U.S. census data

Installation

pipenv install census-error-analyzer

Usage

Import the library.

import census_error_analyzer

Test statistical difference

Are two values, considering their respective margins of error, statistically different? The Census Bureau advises that this test be conducted for all comparisons. This test answers the question and returns True or False.

Accepts two lists, each expected to be a pair with a value and its margin of error.

us_medianage, us_medianage_moe = 37.9, 0.1
nyc_medianage, nyc_medianage_moe = 38.4, 0.1
census_error_analyzer.is_statistically_different(
    (us_medianage, us_medianage_moe), (nyc_medianage, nyc_medianage_moe)
)
True

The precise difference can also be accessed. According to the Census Bureau, values greater than 1.0 can be considered to be statistically significant.

census_error_analyzer.statistical_difference(
    (us_medianage, us_medianage_moe), (nyc_medianage, nyc_medianage_moe)
)
3.535533905932737

Get statistical range

The minimum and maximum values in an estimate's statistical range given its margin of error. Expects two arguments: The estimate first. The margin of error second.

census_error_analyzer.statistical_range(us_medianage, us_medianage_moe)
37.8, 38.0

Convert to alternative confidence levels

The margins of error published by the Census are at the 90% confidence level. They can be converted to the 95% and 99% levels using tools in this library.

census_error_analyzer.convert_to_95_percent_confidence(3778)
4501.446808510638
census_error_analyzer.convert_to_99_percent_confidence(3778)
5925.373860182372

References

This module was designed to conform with the Census Bureau's April 18, 2018, presentation "Using American Community Survey Estimates and Margin of Error."

Prior to publication, the code was reviewed by Brian Dumbacher, a mathematical statistician in the U.S. Census Bureau's Economic Statistical Methods Division.

Links