Closed mzwu closed 1 year ago
South region summary:
SMC: 25,000 sampled plans of 29 districts on 4,410 units
`adapt_k_thresh`=0.985 • `seq_alpha`=0.95
`est_label_mult`=1 • `pop_temper`=0.03
Plan diversity 80% range: 0.43 to 0.98
Sampling diagnostics for SMC run 1 of 2 (12,500 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 3,590 (28.7%) 22.4% 0.30 7,958 (101%) 8
Split 2 3,433 (27.5%) 28.5% 0.33 4,841 ( 61%) 6
Split 3 3,209 (25.7%) 32.2% 0.38 4,846 ( 61%) 5
Split 4 2,902 (23.2%) 25.7% 0.40 4,833 ( 61%) 6
Split 5 2,745 (22.0%) 24.8% 0.42 4,865 ( 62%) 6
Split 6 2,369 (18.9%) 34.4% 0.43 4,671 ( 59%) 4
Split 7 2,141 (17.1%) 42.0% 0.45 4,757 ( 60%) 3
Split 8 2,128 (17.0%) 31.9% 0.45 4,843 ( 61%) 4
Split 9 1,303 (10.4%) 30.5% 0.46 4,791 ( 61%) 4
Split 10 2,144 (17.2%) 28.8% 0.47 4,790 ( 61%) 4
Split 11 2,046 (16.4%) 27.4% 0.48 5,002 ( 63%) 4
Split 12 2,264 (18.1%) 33.5% 0.51 5,024 ( 64%) 3
Split 13 2,382 (19.1%) 24.6% 0.53 5,196 ( 66%) 4
Split 14 1,653 (13.2%) 23.1% 0.53 5,319 ( 67%) 4
Split 15 1,953 (15.6%) 17.6% 0.62 5,078 ( 64%) 5
Split 16 1,876 (15.0%) 26.5% 0.65 5,176 ( 66%) 3
Split 17 1,835 (14.7%) 24.7% 0.65 5,135 ( 65%) 3
Split 18 921 (7.4%) 22.8% 0.67 5,330 ( 67%) 3
Split 19 1,337 (10.7%) 21.7% 0.70 4,799 ( 61%) 3
Split 20 1,826 (14.6%) 29.7% 0.70 4,518 ( 57%) 2
Split 21 1,991 (15.9%) 19.1% 0.69 5,048 ( 64%) 3
Split 22 2,508 (20.1%) 23.3% 0.67 5,475 ( 69%) 2
Split 23 2,694 (21.6%) 14.0% 0.72 5,723 ( 72%) 3
Split 24 2,162 (17.3%) 10.9% 0.73 5,471 ( 69%) 3
Split 25 2,804 (22.4%) 11.7% 0.73 5,129 ( 65%) 2
Split 26 2,892 (23.1%) 10.8% 0.69 5,238 ( 66%) 2
Split 27 3,114 (24.9%) 7.9% 0.72 4,917 ( 62%) 2
Split 28 2,245 (18.0%) 5.3% 0.76 4,438 ( 56%) 2
Resample 2,009 (16.1%) NA% 9.42 4,396 ( 56%) NA
Sampling diagnostics for SMC run 2 of 2 (12,500 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 3,573 (28.6%) 19.6% 0.31 7,911 (100%) 9
Split 2 3,413 (27.3%) 28.3% 0.33 4,785 ( 61%) 6
Split 3 3,176 (25.4%) 32.7% 0.38 4,805 ( 61%) 5
Split 4 2,941 (23.5%) 37.8% 0.41 4,789 ( 61%) 4
Split 5 2,818 (22.5%) 45.4% 0.42 4,763 ( 60%) 3
Split 6 2,603 (20.8%) 28.2% 0.43 4,789 ( 61%) 5
Split 7 2,480 (19.8%) 33.3% 0.44 4,784 ( 61%) 4
Split 8 2,477 (19.8%) 31.5% 0.45 4,884 ( 62%) 4
Split 9 2,158 (17.3%) 30.3% 0.47 4,827 ( 61%) 4
Split 10 2,043 (16.3%) 28.9% 0.49 4,896 ( 62%) 4
Split 11 1,798 (14.4%) 35.8% 0.50 4,956 ( 63%) 3
Split 12 1,892 (15.1%) 45.5% 0.50 4,945 ( 63%) 2
Split 13 2,365 (18.9%) 19.8% 0.52 4,982 ( 63%) 5
Split 14 2,137 (17.1%) 23.6% 0.53 5,226 ( 66%) 4
Split 15 1,856 (14.8%) 28.6% 0.62 5,160 ( 65%) 3
Split 16 2,141 (17.1%) 36.9% 0.62 5,115 ( 65%) 2
Split 17 2,006 (16.0%) 19.1% 0.67 5,183 ( 66%) 4
Split 18 2,092 (16.7%) 11.9% 0.67 5,136 ( 65%) 6
Split 19 1,867 (14.9%) 16.3% 0.73 5,313 ( 67%) 4
Split 20 1,831 (14.6%) 19.9% 0.73 5,112 ( 65%) 3
Split 21 2,219 (17.8%) 13.5% 0.72 5,310 ( 67%) 4
Split 22 2,357 (18.9%) 15.7% 0.71 5,332 ( 67%) 3
Split 23 2,871 (23.0%) 13.1% 0.74 5,257 ( 67%) 3
Split 24 2,574 (20.6%) 7.5% 0.72 5,473 ( 69%) 4
Split 25 1,926 (15.4%) 7.4% 0.72 5,314 ( 67%) 3
Split 26 2,396 (19.2%) 6.7% 0.74 5,022 ( 64%) 3
Split 27 1,712 (13.7%) 10.5% 0.80 5,004 ( 63%) 2
Split 28 1,367 (10.9%) 5.2% 0.78 4,647 ( 59%) 3
Resample 840 (6.7%) NA% 9.68 3,501 ( 44%) NA
• Watch out for low effective samples, very low acceptance rates (less than 1%), large std. devs. of the log weights (more than 3
or so), and low numbers of unique plans. R-hat values for summary statistics should be between 1 and 1.05.
Bay area summary:
SMC: 25,000 sampled plans of 15 districts on 2,213 units
`adapt_k_thresh`=0.985 • `seq_alpha`=0.5
`est_label_mult`=1 • `pop_temper`=0.05
Plan diversity 80% range: 0.51 to 0.77
Sampling diagnostics for SMC run 1 of 2 (12,500 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 10,660 (85.3%) 12.8% 0.45 7,876 (100%) 7
Split 2 10,175 (81.4%) 14.1% 0.61 7,544 ( 95%) 6
Split 3 10,015 (80.1%) 20.0% 0.67 7,360 ( 93%) 4
Split 4 9,956 (79.6%) 24.5% 0.70 7,333 ( 93%) 3
Split 5 9,786 (78.3%) 23.0% 0.73 7,340 ( 93%) 3
Split 6 9,677 (77.4%) 16.1% 0.75 7,325 ( 93%) 4
Split 7 9,342 (74.7%) 19.3% 0.80 7,283 ( 92%) 3
Split 8 9,339 (74.7%) 25.0% 0.82 7,254 ( 92%) 2
Split 9 9,617 (76.9%) 15.5% 0.80 7,217 ( 91%) 3
Split 10 8,010 (64.1%) 19.4% 0.82 7,223 ( 91%) 2
Split 11 7,803 (62.4%) 17.3% 0.86 6,984 ( 88%) 2
Split 12 7,467 (59.7%) 15.0% 0.92 6,645 ( 84%) 2
Split 13 6,841 (54.7%) 15.7% 1.26 5,513 ( 70%) 2
Split 14 4,097 (32.8%) 21.3% 0.89 5,817 ( 74%) 2
Resample 1,344 (10.7%) NA% 10.18 3,589 ( 45%) NA
Sampling diagnostics for SMC run 2 of 2 (12,500 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 10,623 (85.0%) 14.9% 0.45 7,849 ( 99%) 6
Split 2 10,211 (81.7%) 20.9% 0.61 7,446 ( 94%) 4
Split 3 10,033 (80.3%) 25.8% 0.66 7,387 ( 93%) 3
Split 4 9,765 (78.1%) 33.8% 0.70 7,357 ( 93%) 2
Split 5 9,708 (77.7%) 22.9% 0.74 7,288 ( 92%) 3
Split 6 9,646 (77.2%) 21.2% 0.77 7,326 ( 93%) 3
Split 7 9,426 (75.4%) 14.7% 0.79 7,251 ( 92%) 4
Split 8 9,485 (75.9%) 13.2% 0.80 7,262 ( 92%) 4
Split 9 9,401 (75.2%) 15.5% 0.81 7,191 ( 91%) 3
Split 10 8,947 (71.6%) 19.5% 0.80 7,153 ( 91%) 2
Split 11 8,610 (68.9%) 11.7% 0.83 7,026 ( 89%) 3
Split 12 8,278 (66.2%) 14.9% 0.86 6,646 ( 84%) 2
Split 13 7,229 (57.8%) 13.5% 0.90 6,138 ( 78%) 2
Split 14 4,816 (38.5%) 19.4% 0.93 5,907 ( 75%) 2
Resample 2,233 (17.9%) NA% 10.43 3,674 ( 46%) NA
• Watch out for low effective samples, very low acceptance rates (less than 1%), large std. devs. of the log weights (more than 3
or so), and low numbers of unique plans. R-hat values for summary statistics should be between 1 and 1.05.
Looks great to me!
This looks fabulous! Thanks @mzwu!
Redistricting requirements
In California, according to the California Constitution Article XXI, districts must:
Algorithmic Constraints
We enforce a maximum population deviation of 0.5%. We use a pseudo-county constraint to limit the county and municipality splits. We add VRA constraints encouraging Hispanic VAP and Asian VAP majorities in districts.
Data Sources
Data for California comes from the ALARM Project's 2020 Redistricting Data Files. Data for the 2010 California enacted congressional map comes from All About Redistricting.
Pre-processing Notes
Islands were connected to their nearest point within county on the mainland.
Simulation Notes
We sample 25,000 districting plans in each cluster across 2 independent runs of the SMC algorithm. We next sample 50,000 districting plans for California across 2 independent runs of the SMC algorithm for the remainder. We then thin the sample to down to 5,000 plans. To balance county and municipality splits, we create pseudocounties for use in the county constraint. These are counties are Alameda County, Contra Costa County, Fresno County, Kern County, Los Angeles County, Orange County, Riverside County, Sacramento County, San Bernardino County, San Diego County, San Francisco County, San Joaquin County, San Mateo County, Santa Clara County, and Ventura County, which are larger than a congressional district in population. A small population tempering value was used for each cluster to avoid losing diversity at the final step based on initial runs.
1. Clustering Procedure
First, we run partial SMC in two pieces: the south and the Bay Area. The counties in each cluster are:
We sample in each of these regions with a population deviation of 0.5%. We sample 28 districts in the southern region and 14 districts in the Bay Area. Because each cluster will have leftover population, we apply an additional constraint that incentivizes leaving any unassigned areas on the edge of these clusters to avoid discontiguities. For each cluster, we add VRA constraints encouraging Hispanic VAP and Asian VAP concentrations in districts, in line with the enacted plan.
2. Combination Procedure
Then, these partial map simulations are combined to run statewide simulations. We sample 11 districts in the remainder.
Validation
Checklist
TODO
lines from the template code have been removedenforce_style()
to format my coderedist_map
andredist_plans
objects, and summary statistics) have been editedAdditional Notes
Histograms:
Dot plots:
@CoryMcCartan @christopherkenny