How We Weight Our Polls

Online polls often suffer from sampling bias—certain demographics are over or underrepresented compared to the actual population. For example, younger people may be more likely to participate in online polls, while older adults might be underrepresented.

To correct for this, we use demographic weighting, a statistical technique that adjusts the influence of each response to match the known demographic composition of the U.S. population based on Census data.

We use an iterative proportional fitting method called raking(also known as RIM weighting or rake weighting). This algorithm:

1. Starts by assigning each response a weight of 1.0
2. Compares the demographic distribution of responses to Census targets
3. Iteratively adjusts weights for each demographic dimension (age, gender, region) until the weighted sample matches targets
4. Applies weight caps (minimum 0.33, maximum 3.0) to prevent any single response from having too much influence
5. Normalizes weights so they sum to the total number of responses

This process typically converges in 5-10 iterations, producing weights that simultaneously satisfy all demographic targets.

PolitiPolls is designed with privacy in mind:

• No personally identifying information is collected or stored with poll responses
• Demographics are categorical—we collect age ranges and regions, not exact ages or addresses
• Browser fingerprinting is used only to prevent duplicate responses, not to track users
• IP addresses are hashed and never stored in plain text
• All demographics are optional—users can skip providing any demographic information

While we strive for accuracy, it's important to understand the limitations of online polling:

• Self-selection bias: People who choose to participate may differ systematically from those who don't
• Digital divide: Not everyone has equal access to the internet, potentially underrepresenting certain groups
• Voluntary demographics: When users skip demographic questions, we can't weight their responses, which may introduce bias
• Sample size: Smaller polls may have higher margins of error due to fewer responses

PolitiPolls results should be viewed as indicators of public sentiment rather than definitive measurements. For critical decisions, we recommend consulting professional polling organizations with rigorous sampling methods.