Ethnic, Gender and Religion Targeting
Roger Lee has been credited by his peers for bringing accurate ethnic voter targeting to modern American politics. Roger Lee and The Roger Lee Group work with a firm of ethnologists and gender-identity specialists to offer Democratic campaigns and progressive organizations an enhanced ability to identify voters by ethnicity, gender and even religion.

This methodology—which is used by major corporations including AT&T, Bank of America and Sears—increased the accuracy of ethnic coding by 116% in a random sampling of major Democratic voter files across four states. This system codes over 175 classifications of ethnicity. In addition to those ethnic groups traditionally coded (e.g. African American, Hispanic and Vietnamese) it codes emerging groups such as Arab, Armenian, Native American and Indian.

Unlike the methods of voter-file firms and other list vendors, this method is not simply surname-based. It analyzes first name, last name, unique prefixes and suffixes and geography, as well as applying over 2,000 expert system rules based on the study of the origin and history of proper names. This methodology expands the pool of ethnic voters and at the same time increases the accuracy of identifying them. It also improves targeting by gender and additionally offers the ability to target by religion. It represents a significant increase in accuracy that can enhance any voter file, fundraising list or any other name and address list.

A Test of State Party Voter Files
After gaining permission from four major Democratic voter-file firms, this methodology was used to compare the accuracy of our ethnic coding against the ethnic coding methods used by these voter-file vendors. These tests were conducted on voter files in California, New Jersey, Pennsylvania and Texas. Overall, this non-surname system improved the voter files’ ethnic coding by 116%. In a sampling of specific ethnic groups, the voter files were improved by:

• 42% for African Americans;
• 92% for Chinese;
• 98% for Jewish and;
• 66% for Hispanics.

Identifying the “Uncoded” Majority of Voters
This system allowed us to find most of the individuals that voter files remit to a large “uncoded” category. The four voter files we analyzed averaged 68% uncoded for ethnicity. This methodology was able to positively identify over 90% of those voters who were previously uncoded. The analysis also includes coding for assimilated or unassimilated names, helping us to determine the voters’ first language—English or the language of their country of origin.

Improved Targeting by Gender
This methodology increases the accuracy of gender matches to 99% of the file because it incorporates unusual spellings, ethnic names and shortened nicknames. In contrast, many list vendors still use variations of “name your baby” book lists—the pink book is for girls and the blue book is for boys. The problems with this method are similar to those found with ethnic surname dictionaries: gender dictionaries are not up to date; they were inaccurate or incomplete from the start; and they don’t take into account unusual spellings, ethnic names, or shortened nicknames.

The Ability to Target by Religion
This system also identifies voters by religion, including Protestant, Catholic, Jewish, Islamic, Eastern Orthodox, Greek Orthodox, Shinto, Buddhist and Hindu. This is an entirely new capability for Democratic campaigns. For example, in the study of voter files we found over 232,000 Catholic voters in the Pittsburgh area that would not have been identified on a standard voter file.

Voter Files Can “Feel” Right and Still Be Wrong
Our analysis found that voter files often ballpark the right percentage for each voter category, but then code the wrong people within that category. For example, in Bergen County, New Jersey, 6% of the voter file was Hispanic as coded by the vendor, which “felt” like the correct percentage to people familiar with the area.

But when we examined the file name by name, 35% of supposedly Hispanic names were clearly a different ethnicity, while another 40% of Hispanics were either uncoded or coded incorrectly. By subtracting the 35% wrong names and adding 40% more correct names, the overall file remained at around 6% Hispanic, but we had improved the files’ accuracy by 75%. The lesson we learned is that a file can be quite inaccurate even when the percentage of voters in a given category “feels” right.

Identifying African Americans Outside Cluster Areas
In a recent Wall Street Journal article, Harvard demographer and economist Edward Glaeser documented a strong trend of African Americans moving to less segregated communities within the United States. As African Americans in the vast majority of metropolitan areas become more integrated into suburban culture, the traditionally inaccurate manner of locating African American voters solely by cluster analysis has become even less effective.

This methodology can—for the first time ever—accurately identify African Americans who live outside of cluster areas. This is an exceptionally important advancement. It means that Democratic campaigns can increase the number of African Americans identified in a given state or district by as much as 20%. This methodology will not only expand the universe of African American voters, it will more accurately identify them.

Identifying African Americans has always been especially problematic with voter files, and the major vendors do not all use the same system. Generally, African Americans are identified by selecting precincts or areas where African Americans are clustered, and then by removing surnames for groups who may not be African American. Some voter-file vendors remove only Hispanic and Asian names, while others remove more ethnic categories. Regardless of the method used, the surname-basis for selecting ethnicity is so riddled with errors that targeting African Americans has been highly inaccurate in the past.

Our system looks at the first name first, which can be the best way to positively identify African American voters. Then, it recognizes that the vast majority of African Americans have names from one of seven ethnic categories: English, Scotch, Welsh, Irish, French, Arabic, or African. This methodology can increase the accuracy of identifying African Americans in cluster areas by as much as 30 percent. Then it can do something no other system has ever done—locate up to 20% more African Americans who live outside of cluster areas, using geocoding and sophisticated rules for selecting African American names. These individuals would not be located otherwise.

Reaching Hispanic Voters Who Are Truly Hispanic
Voter files frequently contain individuals who do not consider themselves Hispanic. List providers who use surname systems have already discovered the problems inherent in that approach. Many people who have Hispanic surnames are not Hispanic. For example, “Zvi Garcia” is likely Jewish. Because this system incorporates first names into its approach, it can discern a Sephardic Jew from a Hispanic.

This methodology can also predict the likelihood of whether or not a household is Spanish speaking. This knowledge will obviously have a major impact on a campaign’s communications strategy.

Finally, because the methodology uses unique first names in its analysis, it will identify Hispanic women heads of household who do not have Hispanic surnames. For example, Lupe Sanchez who marries John Smith and is registered to vote as Lupe Smith would never be found with a traditional surname system.