Calculating Scores in Smaller Areas

In many cases, it is critical to understand public sentiment at a division level (e.g., council district or police precinct). Sometimes, the sample size at the division level is relatively small, and the sample composition may not be representative of the division, leading to low reliability from cycle to cycle. In addition, some areas may not have enough sample to provide scores at all. To address these issues and to deliver the most precise and actionable results to our clients, we use a combination of division-level rake weighting and smoothing. 

Smoothing

Smoothing is a process used for quarterly and monthly clients that have divisions where the average expected sample size per division is less than 150. In some cases, this means that the larger division level is not smoothed but a smaller subdivision is. Scores at the city level are never smoothed. 

As an example, a city with a population of 1 million is conducting a monthly Blockwise survey. They have five precincts which each contain three police districts, giving us a total of 15 police districts across the city. The target sample size for the city is 2000. We would therefore expect, on average, 400 responses for each of the five precincts and about 133 responses in each of the 15 police districts. In this case, smoothing would be implemented for the police districts but not the precincts or the city. 

Smoothing combines data from the current cycle and the previous cycle at the division level. In the example above, we would combine June and July monthly Blockwise data to create July  police district scores. Then, in August, we would combine July and August data to create those scores. 

The dashboard will indicate when and at which levels smoothing has been implemented. For example, the screenshot below shows the summary page for geographic area E10. During the selected cycle, 48 submissions were collected. 

However, when we go into the Analysis tab, we see that the analytic sample is 100, which includes the 48 responses collected in July and 52 responses collected in June.

Rake weighting

After we have combined cycles in the smoothing step, we adjust the data using data from the U.S. Census. This process, called rake weighting, ensures that no demographic group is overrepresented or underrepresented in the data. In some cases, rake weighting is done at the division level.