Closed christopherkenny closed 2 years ago
CoryMcCartan
commented
2 years ago Looks good. While we wait for the final plan, let's over-generate (maybe 6k instead of 5k) and discard those few with minority VAP < 50%, then subsample to a final 5k. MO-1 is an MMD if not a Black MD and seems reasonable to enforce that.
christopherkenny
commented
2 years ago In Missouri, districts must:
We enforce a maximum population deviation of 0.5%. We apply a basic county constraint to be in line with the splits in the plan, though there is no legal requirement. We add a VRA constraint targeting one BVAP opportunity district.
Data for Missouri comes from the ALARM Project's 2020 Redistricting Data Files.
No manual pre-processing decisions were necessary.
We sample 6,000 districting plans for Missouri and subset to 5,000 which contain at least one majority minority district. We use a standard algorithmic county constraint. No special techniques were needed to produce the sample.
@CoryMcCartan
CoryMcCartan
commented
2 years ago Great!
christopherkenny
commented
2 years ago In Missouri, districts must:
We enforce a maximum population deviation of 0.5%. We apply a basic county constraint to be in line with the splits in the plan, though there is no legal requirement. We add a VRA constraint targeting one BVAP opportunity district.
Data for Missouri comes from the ALARM Project's 2020 Redistricting Data Files.
No manual pre-processing decisions were necessary.
We sample 20,000 districting plans for New Jersey across two independent runs of the SMC algorithm, and then thin the sample to down to 5,000 plans. We use a standard algorithmic county constraint. No special techniques were needed to produce the sample.
@CoryMcCartan
CoryMcCartan
commented
2 years ago also, summary(plans) output?
christopherkenny
commented
2 years ago In Missouri, districts must:
We enforce a maximum population deviation of 0.5%. We apply a basic county constraint to be in line with the splits in the plan, though there is no legal requirement. We add a VRA constraint targeting one BVAP opportunity district.
Data for Missouri comes from the ALARM Project's 2020 Redistricting Data Files.
No manual pre-processing decisions were necessary.
We sample 10,000 districting plans for Missouri across two independent runs of the SMC algorithm, and then thin the sample to down to 5,000 plans. We use a standard algorithmic county constraint. No special techniques were needed to produce the sample.
SMC: 5,000 sampled plans of 8 districts on 4,604 units
`adapt_k_thresh`=0.985 • `seq_alpha`=0.95
`est_label_mult`=1 • `pop_temper`=0
Plan diversity 80% range: 0.47 to 0.73
R-hat values for summary statistics:
pop_overlap total_vap plan_dev comp_edge comp_polsby pop_hisp pop_white pop_black
1.011086 1.003021 1.004126 1.018469 1.001638 1.005015 1.018586 1.013368
pop_aian pop_asian pop_nhpi pop_other pop_two vap_hisp vap_white vap_black
1.002968 1.013584 1.007631 1.002935 1.008035 1.006255 1.010121 1.011582
vap_aian vap_asian vap_nhpi vap_other vap_two pre_16_rep_tru pre_16_dem_cli uss_16_rep_blu
1.009598 1.012810 1.007415 1.009529 1.020016 1.012339 1.007639 1.012887
uss_16_dem_kan gov_16_rep_gre gov_16_dem_kos atg_16_rep_haw atg_16_dem_hen sos_16_rep_ash sos_16_dem_smi uss_18_rep_haw
1.014356 1.013625 1.011240 1.011824 1.012727 1.011692 1.014208 1.013098
uss_18_dem_mcc pre_20_rep_tru pre_20_dem_bid gov_20_rep_par gov_20_dem_gal atg_20_rep_sch atg_20_dem_fin sos_20_rep_ash
1.009841 1.008009 1.011305 1.009734 1.012051 1.010966 1.009091 1.010426
sos_20_dem_fal arv_16 adv_16 arv_18 adv_18 arv_20 adv_20 county_splits
1.008248 1.012360 1.010968 1.013098 1.009841 1.009935 1.009855 1.001761
muni_splits ndv nrv ndshare e_dvs pr_dem e_dem pbias
1.007331 1.008315 1.011801 1.011479 1.011471 1.000000 1.000457 1.006232
egap
1.000828
Sampling diagnostics for SMC run 1 of 2 (5,000 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 2,981 (59.6%) 12.3% 0.48 3,197 (101%) 13
Split 2 2,528 (50.6%) 18.1% 0.52 2,644 ( 84%) 8
Split 3 867 (17.3%) 20.2% 0.56 2,668 ( 84%) 6
Split 4 1,907 (38.1%) 25.0% 0.58 2,571 ( 81%) 4
Split 5 2,237 (44.7%) 26.5% 0.54 2,674 ( 85%) 3
Split 6 1,753 (35.1%) 20.2% 0.55 2,649 ( 84%) 3
Split 7 2,902 (58.0%) 9.2% 0.57 2,306 ( 73%) 2
Resample 2,697 (53.9%) NA% 0.67 2,774 ( 88%) NA
Sampling diagnostics for SMC run 2 of 2 (5,000 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 2,975 (59.5%) 14.7% 0.48 3,151 (100%) 11
Split 2 2,222 (44.4%) 20.0% 0.53 2,707 ( 86%) 7
Split 3 914 (18.3%) 28.6% 0.56 2,649 ( 84%) 4
Split 4 2,386 (47.7%) 28.8% 0.53 2,519 ( 80%) 3
Split 5 2,548 (51.0%) 15.5% 0.53 2,761 ( 87%) 6
Split 6 2,330 (46.6%) 17.1% 0.57 2,673 ( 85%) 4
Split 7 2,625 (52.5%) 7.2% 0.56 2,392 ( 76%) 3
Resample 2,410 (48.2%) NA% 0.70 2,744 ( 87%) 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.
christopherkenny
commented
2 years ago In Missouri, districts must:
We enforce a maximum population deviation of 0.5%. We apply a basic county constraint to be in line with the splits in the plan, though there is no legal requirement. We add a VRA constraint targeting one BVAP opportunity district.
Data for Missouri comes from the ALARM Project's 2020 Redistricting Data Files.
No manual pre-processing decisions were necessary.
We sample 10,000 districting plans for Missouri across two independent runs of the SMC algorithm, subset to plans with at least 30% BVAP in the most Black district, and then thin the sample to down to 5,000 plans. The subsetting by BVAP removes around 2% of sampled plans. We use a standard algorithmic county constraint. No special techniques were needed to produce the sample.
SMC: 5,000 sampled plans of 8 districts on 4,604 units
`adapt_k_thresh`=0.985 • `seq_alpha`=0.95
`est_label_mult`=1 • `pop_temper`=0
Plan diversity 80% range: 0.45 to 0.73
R-hat values for summary statistics:
vap_black pop_overlap total_vap plan_dev comp_edge comp_polsby pop_hisp
1.003587 1.016695 1.012983 1.000903 1.017739 1.012518 1.022481
pop_white pop_black pop_aian pop_asian pop_nhpi pop_other pop_two
1.015122 1.005551 1.004617 1.006272 1.007635 1.002684 1.011791
vap_hisp vap_white vap_aian vap_asian vap_nhpi vap_other vap_two
1.009834 1.021428 1.008917 1.004242 1.008639 1.001351 1.016830
pre_16_rep_tru pre_16_dem_cli uss_16_rep_blu uss_16_dem_kan gov_16_rep_gre gov_16_dem_kos atg_16_rep_haw
1.021180 1.004011 1.042427 1.003088 1.027639 1.002789 1.040899
atg_16_dem_hen sos_16_rep_ash sos_16_dem_smi uss_18_rep_haw uss_18_dem_mcc pre_20_rep_tru pre_20_dem_bid
1.002163 1.048891 1.002186 1.039801 1.004271 1.020286 1.003561
gov_20_rep_par gov_20_dem_gal atg_20_rep_sch atg_20_dem_fin sos_20_rep_ash sos_20_dem_fal arv_16
1.032736 1.003141 1.048776 1.003026 1.048498 1.003306 1.037797
adv_16 arv_18 adv_18 arv_20 adv_20 county_splits muni_splits
1.002967 1.039801 1.004271 1.040133 1.003197 1.004967 1.015247
ndv nrv ndshare e_dvs e_dem pbias egap
1.003244 1.036002 1.021201 1.021707 1.008580 1.038742 1.005395
Sampling diagnostics for SMC run 1 of 2 (5,000 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 2,948 (59.0%) 12.4% 0.47 3,183 (101%) 13
Split 2 1,959 (39.2%) 17.8% 0.53 2,680 ( 85%) 8
Split 3 1,484 (29.7%) 11.3% 0.57 2,622 ( 83%) 11
Split 4 1,927 (38.5%) 15.4% 0.57 2,527 ( 80%) 7
Split 5 2,122 (42.4%) 11.8% 0.54 2,636 ( 83%) 8
Split 6 2,520 (50.4%) 13.9% 0.56 2,602 ( 82%) 5
Split 7 1,885 (37.7%) 7.2% 0.56 2,366 ( 75%) 3
Resample 1,629 (32.6%) NA% 0.72 2,656 ( 84%) NA
Sampling diagnostics for SMC run 2 of 2 (5,000 samples)
Eff. samples (%) Acc. rate Log wgt. sd Max. unique Est. k
Split 1 3,002 (60.0%) 14.7% 0.47 3,167 (100%) 11
Split 2 2,117 (42.3%) 22.7% 0.52 2,703 ( 86%) 6
Split 3 1,756 (35.1%) 17.8% 0.55 2,697 ( 85%) 7
Split 4 2,101 (42.0%) 21.0% 0.55 2,630 ( 83%) 5
Split 5 2,043 (40.9%) 22.0% 0.58 2,628 ( 83%) 4
Split 6 2,330 (46.6%) 12.4% 0.57 2,639 ( 83%) 6
Split 7 3,029 (60.6%) 6.0% 0.55 2,451 ( 78%) 4
Resample 2,842 (56.8%) NA% 0.67 2,816 ( 89%) 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
> plans %>% subset_sampled() %>% group_by(draw, chain) %>%
+ summarize(mmds = sum(vap_black/total_vap > 0.3 & ndshare > 0.5), .groups = 'drop') %>%
+ count(mmds)
mmds n
1 1 5000@CoryMcCartan 10th time is the charm
Redistricting requirements
In Missouri, districts must:
Interpretation of requirements
We enforce a maximum population deviation of 0.5%. We apply a basic county constraint to be in line with the splits in the plan, though there is no legal requirement. We add a VRA constraint targeting one BVAP opportunity district.
Data Sources
Data for Missouri comes from the ALARM Project's 2020 Redistricting Data Files.
Pre-processing Notes
No manual pre-processing decisions were necessary.
Simulation Notes
We sample 5,000 districting plans for Missouri. We use a standard algorithmic county constraint. No special techniques were needed to produce the sample.
Validation
Checklist
TODOlines from the template code have been removedenforce_style()to format my coderedist_mapandredist_plansobjects, and summary statistics) have been edited@CoryMcCartan
Additional Notes:
Here is a performance plot for BVAP.