Q: How does the random selection process work? What role does the Legislature play?
The Michigan Constitution outlines a specific process for the random selection of the final 13 commissioners. The Secretary of State must randomly select commissioners from the pool of eligible applicants. This process was completed between June and August 2020. There are a three primary steps in that process.
Step 1: Following the closure of the application period, the Secretary of State's Office must randomly select 200 semifinalist applicants and provide those applications to the Michigan Legislature for review. Of the 200 randomly selected applicants, 60 people must affiliate with the Republican Party, 60 people must affiliate with the Democratic Party and 80 must not affiliate with either of those political parties. The Secretary of State's Office was required to use statistical weighting methods to ensure the pool of 200 semifinalists mirrors the geographic and demographic makeup of the state, as specified by the state constitution.
Step 2: In July 2020 the minority and majority leaders in the Michigan Senate and the Speaker of the House and minority leader in the Michigan House removed 20 applicants from the semifinalist pool, as permitted by the State constitution. You can review those strikes here.
Step 3: In August 2020, from the remaining pool received from the Legislature, the Secretary of State's Office randomly selected four people who affiliate with the Republican Party, four people who affiliate with the Democratic Party and five people who do not affiliate with either major political party to serve on commission.
Q: Is the selection process truly random?
Yes. The Secretary of State plays a purely administrative role in the random selection process. Applicants drawn in various stages of the selection process will are selected at random, and the random-ness of that selection will be verified by external, independent firms as well as public observers.
Q: What did the Department of state use to define the demographic and geographic makeup of the state for the purposes of the random selection?
We utilized data from the 2018 American Community Survey 5-Year Estimate, which is available to the general public via the American Community Survey Data Profiles online search tool. Michigan demographic information can be viewed here, and detailed tables with data for race and ethnicity, age and sex, and geographic distribution are also available.
The variables were defined as follows:
Hispanic, Latino, or Spanish origin?
Yes - 5.0%
No - 95.0%
White - 78.52%
Black or African American - 13.81%
American Indian or Alaska Native - 0.53%
Asian - 3.06%
Other - 1.23%
Two or more races - 2.85%
Males - 49.2%
Females - 50.8%
18-34 years old - 28.8%
35-54 years old - 32.4%
55+ years old - 38.8%
Southeast Michigan - 35.4% of the total state population
Eaton, Ingham, Livingston, Oakland, Macomb, Jackson, Washtenaw, Hillsdale, Lenawee, Monroe
Wayne County - 17.7% of the total state population
West Michigan - 23.8% of the total state population
Oceana, Newaygo, Mecosta, Isabella, Muskegon, Montcalm, Ottawa, Kent, Ionia, Allegan, Barry, Van Buren, Kalamazoo, Calhoun, Berrien, Cass, St. Joseph, Branch
East Central Michigan - 13.6% of the total state population
Midland, Bay, Huron, Gratiot, Saginaw, Tuscola, Sanilac, Clinton, Shiawassee, Genesee, Lapeer, St. Clair
Northern Michigan - 6.4% of the total state population
Emmet, Cheboygan, Charlevoix, Presque Isle, Antrim, Otsego, Montmorency, Alpena, Leelanau, Grand Traverse, Benzie, Kalkaska, Crawford, Oscoda, Alcona, Manistee, Wexford, Missaukee, Roscommon, Ogemaw, Iosco, Mason, Lake, Osceola, Arenac, Clare, Gladwin
Upper Peninsula - 3.1% of the total state population
Gogebic, Ontonagon, Houghton, Keweenaw, Baraga, Iron, Marquette, Dickinson, Menominee, Alger, Delta, Schoolcraft, Luce, Mackinac, Chippewa
Q: How large a deviation from the target population proportions did the State accept in the pool of 200 semi-finalists?
The Department's independent contractor, Rehmann LLC, provided the Department with a recommendation based on practical field tests of the software application and preliminary test data. Accordingly, the statistical tolerances for the random selection of 200 semifinalists were set to the following:
Sex: 5.0% tolerance
Race: 3.0% tolerance
Hispanic: 2.0% tolerance
Age: 5.0% tolerance
Region: 5.0% tolerance
Additionally, at least one applicant from each demographic and geographic sub-category (including all racial groups and all geographic regions) must be included in the group of 200 semifinalists.
You can see the final geographic and demographic makeup of the 200 semifinalists.
Q: Are the relevant selection demographics based on the total population or the voting age population (18 and over)? If the latter is used, does it consider all voting age people or only citizens?
The Michigan Constitution requires that the Secretary of State's office use "accepted statistical weighting methods to ensure that the pools, as closely as possible, mirror the geographic and demographic makeup of the state." Accordingly, the demographics for race and ethnicity, sex, and geographic area are based on the total population in Michigan. However, in order to avoid a weighting imbalance and propose nearly equal weights for eligible age categories, we did not factor people age 17 and under into the weighted categories for age.
Q: Are Hispanic, Latino, and Spanish origin responses considered a "race" variable? Are they calculated together with or separate from non-Hispanic racial categories?
On the U.S. Census and the American Community Survey, "Hispanic, Latino, or Spanish origin" is a separate question from "Race." They are separate categories and our proposed method treats them as separate variables as well. In other words, the categories for race include both Hispanic and non-Hispanic individuals. For example, a person could be both Hispanic and black, or non-Hispanic and black, and they would both be identified as "black" for the racial category. Hispanic identity is captured in a separate question, which makes it a separate variable.
Q: How did the State plan combine the requirement for the partisan breakdown on the Commission (30%/30%/40%) with the rule that the selection process should "use accepted statistical weighting methods to ensure that the pools, as closely as possible, mirror the geographic and demographic makeup of the state"?
The Michigan Constitution requires that the entire pool of 200 "semifinalists" mirror the geographic and demographic makeup of the state. There is no requirement or expectation that each of the individual partisan groups (30%/30%/40%) mirror the geographic and demographic makeup of the state individually - only that the combined pool of 200 semifinalists collectively mirror the state.
Accordingly, each demographic factor was separately weighted and considered, without any cross tabulation of political party affiliation. The formula randomly selected 60 applicants who affiliate with the Democratic Party, 60 applicants who affiliate with the Republican Party, and 80 applicants who do not affiliate with either party, until the group of 200 meets the demographic and geographic criteria to reflect the state as a whole. That group of 200 semifinalists was referred to the legislature. That will be the group of 200 semifinalists referred to the legislature.
Q: Did the State use raking to apply the demographic weights to the group of applicants prior to selecting semifinalists? If so, what is the exact function in the visual basic program?
Yes, raking was performed using basic Excel functions. "Raking" is a statistical method for weighting data, commonly used in polls or surveys (for example, see Pew Research Center's explanation). Each application received a weighting factor for each demographic variable. The formula is a simple ratio of the percentage in the population of applications to the target percentage from the U.S. Census.
For more explanation, view the archived livestream of random selection.
Q: What if the random selection fails to meet stated criteria?
In order to maintain the statistical randomness of the process, the software application operated in a fully automated manner. Once the applications were loaded and the tolerances set for mirroring "as closely as possible" the geographic and demographic makeup of the state, the computer drew a random selection of 200 applications - 60 who affiliate with the Democratic Party, 60 who affiliate with the Republican Party, and 80 who do not affiliate with either party. Once that sample was within the stated tolerances, the program output the results. In other words, in this automated process individual applicants were never discarded. Every applicant had the opportunity to be selected until all 200 met the stated criteria.
Q: Will the State disclose the software and the underlying computer code used in the random selection process so it can be reviewed and validated by outside groups?
A copy of the Visual basic for Applications code in an Excel file can be made available upon request.