What’s a Ward Endorsement worth?

In 2017, about 6 points

With ballot positions decided, candidates are starting to vie for coveted ward endorsements. How many votes are they really worth?

Two years ago, I did a simplistic analysis of the Court of Common Pleas, where I found that judicial candidates received 0.9 more percent of the vote in wards where they were endorsed. In a race where candidates win with 4.3 percent of the vote, that effect is huge (and larger than even ballot position).

There were a number of caveats to that analysis: I only had endorsements from a few, systematically different wards, and I didn’t do anything to identify causality–we know that candidates do better in wards where they are endorsed, but we don’t know if the endorsements cause that increase, or if the ward leaders were endorsing candidates who would have done well there anyway.

Let’s do better.

In 2017, Max Marin at Philadelphia Weekly undertook the herculean effort of tracking down endorsements in 62 of Philadelphia’s 66 wards. Let’s use that, and do some spatial econometrics.

View code

library(tidyverse)
library(sp)
library(rgeos)
library(rgdal)
library(sf)
source("../../admin_scripts/util.R")

df_major <- safe_load("../../data/processed_data/df_major_2017_12_01.Rda")

df_major$WARD_DIVSN <- with(df_major, paste0(WARD16, DIV16))

df_major <- df_major %>%
  filter(
    election == "primary" & CANDIDATE != "Write In" & PARTY == "DEMOCRATIC"
  )
df_major <- df_major %>%
  group_by(WARD_DIVSN, OFFICE, year) %>%
  mutate(pct_vote = VOTES / sum(VOTES))

df_major <- df_major %>%
  filter(OFFICE %in% c("COUNCIL AT LARGE", "DISTRICT ATTORNEY"))


bg_17_acs <- read.csv("../../data/census/acs_2013_2017_phila_bg_race_income.csv")
bg_17_acs <- bg_17_acs %>% 
  mutate(Geo_FIPS = as.character(Geo_FIPS)) %>%
  select(
    Geo_FIPS, pop, pop_nh_white, pop_nh_black, pop_nh_asian, pop_hisp, pop_median_income_2017
  )

sp_divs <- readOGR("../../data/gis/2016/2016_Ward_Divisions.shp", verbose = FALSE)
sp_divs <- spChFIDs(sp_divs, as.character(sp_divs$WARD_DIVSN))
sp_divs <- spTransform(sp_divs, CRS("+init=EPSG:4326"))

library(tigris)
options(tigris_use_cache = TRUE)
bg_shp <- block_groups(42, 101, year = 2015)
bg_shp <- spChFIDs(bg_shp, as.character(bg_shp$GEOID))
bg_shp <- spTransform(bg_shp, CRS(proj4string(sp_divs)))

sp_divs$bg <- over(
  gCentroid(sp_divs, byid = TRUE), 
  bg_shp
)$GEOID

sp_divs@data <- sp_divs@data %>%
  left_join(bg_17_acs, by = c("bg"="Geo_FIPS"))

df_major <- df_major %>%
  left_join(sp_divs@data) %>%
  mutate(
    pct_wht = pop_nh_white / pop,
    pct_blk = pop_nh_black / pop,
    pct_asian = pop_nh_asian/ pop,
    pct_hisp = pop_hisp / pop
  )

In the 2017 DA race, no single candidate monopolized the endorsements; O’Neill led the way with 11 endorsements, largely in the Northeast.

View code

endorsements <- read_csv("da_2017_endorsements.csv")
endorsements$ward <- sprintf("%02d", endorsements$ward)

da_results <- df_major %>% 
  filter(election == "primary" & year == 2017 & OFFICE == "DISTRICT ATTORNEY") %>%
  mutate(
    last_name = gsub(
      "^.*\\s([A-Z])([A-Z]+)

View code

quot;, "\\U\\1\\L\\2", CANDIDATE, perl = TRUE ) ) %>% group_by(WARD_DIVSN) %>% mutate(total_votes = sum(VOTES)) %>% group_by() %>% mutate(pvote = VOTES / total_votes) da_results$last_name <- with( da_results, ifelse( last_name == "Neill", "O'Neill", ifelse(last_name == "Shabazz", "El-Shabazz", last_name) ) ) da_results %>% group_by(WARD16, last_name) %>% summarise(votes = sum(VOTES)) %>% group_by(WARD16) %>% mutate( ward_votes = sum(votes), pvote = votes/ ward_votes ) %>% left_join( endorsements %>% mutate(is_endorsed = TRUE), by = c("WARD16" = "ward", "last_name" = "endorsement") ) %>% mutate( is_endorsed = replace(is_endorsed, is.na(is_endorsed), FALSE) ) %>% group_by(last_name, is_endorsed) %>% summarise( pct_vote = 100 * weighted.mean(pvote, w = ward_votes), total_votes = sum(ward_votes), n_wards = n() ) %>% group_by(last_name) %>% summarise( pct_vote_overall = weighted.mean(pct_vote, w = total_votes), wards_endorsed = ifelse(any(is_endorsed), n_wards[is_endorsed], 0), turnout_endorsed = ifelse(any(is_endorsed), total_votes[is_endorsed], 0), pct_vote_notendorsed = pct_vote[!is_endorsed], pct_vote_endorsed = ifelse(any(is_endorsed), pct_vote[is_endorsed], NA) ) %>% arrange(desc(pct_vote_overall)) %>% knitr::kable( digits = 0, format = "html", format.args = list(big.mark = ","), col.names = c( "Candidate", "Citywide % of vote", "Number of ward endorsements", "Turnout in endorsed wards", "% of vote in un-endorsed wards", "% of vote in endorsed wards" ) )

Candidate	Citywide % of vote	Number of ward endorsements	Turnout in endorsed wards	% of vote in un-endorsed wards	% of vote in endorsed wards
Krasner	38	9	28,700	36	46
Khan	20	8	23,485	18	32
Negrin	14	10	20,794	13	21
El-Shabazz	12	7	19,504	11	17
Untermeyer	8	9	16,200	7	17
O’Neill	6	11	21,520	4	20
Deni	2	0	0	2	NA

Krasner won by over 5,000 votes (18%), despite receiving the typical number of ward endorsements. The endorsements that he did receive came from wards with the highest turnout, but part of that is reverse causality: the places that he energized turned out big.

Naively, candidates did about 11 percentage points better in wards where they were endorsed than in wards where they weren’t. BUT. This suffers from the same lack of causal identification as the Judicial Analysis above: we don’t know if they did better because of the endorsements, or if they were just endorsed in wards where they would have done well anyway.

How can we do better? Let’s use something I noticed in last week’s post on District 7: the strength of boundaries.

The strongest ward endorsements can have visible effects in divisions just across the street from each other.

View code

wards <- readOGR("../../data/gis/2016","2016_Wards", verbose=FALSE) %>%
  spTransform(CRS(proj4string(sp_divs))) 

ggwards <- fortify(spChFIDs(wards, sprintf("%02d", wards$WARD)))

bbox <- sp_divs[substr(sp_divs$WARD_DIVSN,1,2) %in% c("10", "50", "09"),] %>% 
  gUnionCascaded() %>% 
  bbox()


bbox <- rowMeans(bbox) + 1.2 * sweep(bbox, 1, rowMeans(bbox))

polygon_in_bbox <- function(p) {
  coords <- p@Polygons[[1]]@coords
  any(
    coords[,1] > bbox[1,1] &
      coords[,1] < bbox[1,2] &
      coords[,2] > bbox[2,1] &
      coords[,2] < bbox[2,2] 
  )
}
sp_divs$in_bbox <- sapply(sp_divs@polygons, polygon_in_bbox) 

ggdivs <- fortify(spChFIDs(sp_divs, as.character(sp_divs$WARD_DIVSN)))

ggdivs <- ggdivs %>%
  left_join(
    sp_divs@data %>% select(WARD_DIVSN, in_bbox),
    by = c("id" = "WARD_DIVSN")
  ) %>%
  left_join(
    da_results %>% filter(last_name %in% c("Khan", "Krasner", "El-Shabazz")),
    by = c("id" = "WARD_DIVSN")
  )

ward_centroids <- gCentroid(wards, byid=TRUE) %>% as.data.frame()
ward_centroids$ward <- wards$WARD

ggplot(
  ggdivs %>% filter(in_bbox),
  aes(x=long, y=lat)
) +
  geom_polygon(aes(fill = 100 * pvote, group=group), color = NA) +
  geom_polygon(data = ggwards, aes(group=group), fill = NA, color = "white") +
  geom_text(data = ward_centroids, aes(x=x, y=y, label=ward), color = "white") +
  facet_wrap(~last_name) +
  scale_fill_viridis_c("% of vote") +
  theme_map_sixtysix() +
  coord_map(xlim=bbox[1,], ylim=bbox[2,]) +
  theme(
    legend.position = "bottom",
    legend.direction = "horizontal"
  ) +
  ggtitle("Percent of the Vote in Northwest Wards", "2017 DA Race")

Wards 10 and 50 endorsed Krasner, Ward 9 endorsed Khan, and Ward 22 endorsed El-Shabazz. You can immediately see the strength of 10 and 50’s endorsements: Krasner did better in divisions inside the boundary of 10 and 50 than he did just across the street. Same for 9, maybe, where Khan did well. And El-Shabazz did better in 22, though there isn’t an obvious boundary effect.

I’ll use this intuition to measure the effect across all boundaries in the whole city. To isolate the causal effect of the wards, I’ll limit the analysis to only compare divisions that are across the street from each other but happen to be divided by a ward boundary, and where different candidates were endorsed. This will ensure that we’re comparing divisions apples-to-apples, where the only thing that’s different is the ward endorsement.

I’ll go one step farther, and control for the census demographics of the block groups that the division sits in, in case a ward boundary happens to also serve as an emergent boundary (dissertation plug). I measure how each candidate’s vote correlated with the race and ethnicity of the neighborhood and subtract that out, leaving a measure of how much better or worse that candidate did than expected. It’s that “residual” that I will compare across boundaries.

View code

da_fit <- lm(
  pvote ~
    CANDIDATE * pct_wht + 
    CANDIDATE * pct_blk + 
    CANDIDATE * pct_hisp,
    # CANDIDATE * log(pop_median_income_2017),
  data = da_results
)

da_results$predicted <- predict(da_fit, newdata = da_results)
da_results$resid <- with(da_results, pvote - predicted)

neighbors <- st_intersection(st_as_sf(sp_divs), st_as_sf(sp_divs))
neighbors <- neighbors %>%
  filter(WARD != WARD.1)

neighbors <- neighbors %>%
  mutate(geometry_type = st_geometry_type(geometry)) %>%
  filter(!geometry_type %in% c("POINT", "MULTIPOINT"))


neighbors <- neighbors %>% 
  mutate(
    WARD.0 = sprintf("%02d", asnum(WARD)),
    WARD.1 = sprintf("%02d", asnum(WARD.1))
  ) %>%
  left_join(
    endorsements %>% rename(endorsement.0 = endorsement),
    by = c("WARD.0" = "ward")
  ) %>%
  left_join(
    endorsements %>% rename(endorsement.1 = endorsement),
    by = c("WARD.1" = "ward")
  ) 

neighbors <- neighbors %>% 
  left_join(
    da_results %>% 
      select(WARD_DIVSN, last_name, total_votes, pvote, resid) %>%
      rename(total_votes.0 = total_votes, pvote.0 = pvote, resid.0 = resid),
    by = c("WARD_DIVSN" = "WARD_DIVSN", "endorsement.0" = "last_name")
  ) %>% 
  left_join(
    da_results %>% 
      select(WARD_DIVSN, last_name, total_votes, pvote, resid) %>%
      rename(total_votes.1 = total_votes, pvote.1 = pvote, resid.1 = resid),
    by = c("WARD_DIVSN.1" = "WARD_DIVSN", "endorsement.0" = "last_name")
  )

To correctly measure wards’ individual strength, I fit a random effects model, which simultaneously estimates the average effect of all wards’ endorsements and how much each individual ward varies from that.

View code

library(lme4)

df0 <- neighbors %>% filter(endorsement.0 != endorsement.1)

fit_lmer <- function(neighbor_df){
  re_fit <- lmer(
    resid.0 - resid.1 ~ (1 | WARD.0),
    data = neighbor_df,
    weights = neighbor_df %>%
      with(pmin(total_votes.0, total_votes.1))
  )

  re <- ranef(re_fit)$WARD.0 
  re <- re %>%
    mutate(
      ward = row.names(re),
      effect = re_fit@beta + `(Intercept)`
    )

  return(
    list(
      fit = re_fit,
      re = re
    )
  )
}

fit_baseline <- fit_lmer(df0)

n_boot <- 200
bs_list <- vector(mode = "list", length = n_boot)
for(b in 1:n_boot){

  sample_divs = sample(unique(df0$WARD_DIVSN), replace = TRUE)

  #if(b %% floor(n_boot / 10) == 0) print(b)
  df_samp <- data.frame(WARD_DIVSN = sample_divs) %>% left_join(df0)
  bs_fit <- fit_lmer(df_samp)
  bs_list[[b]] <- bs_fit
}

fixef_ci <- quantile(
  sapply(bs_list, function(x) fixef(x$fit)),
  c(0.025, 0.975)
)
cat(paste0(
  "Average Effect of a Ward Endorsement:\n",
  sprintf(
    "%0.1f (%0.1f, %0.1f)",
    fixef(fit_baseline$fit)["(Intercept)"] * 100,
    fixef_ci[1] * 100,
    fixef_ci[2] * 100
  )
))

## Average Effect of a Ward Endorsement:
## 5.8 (5.0, 6.9)

The average Ward endorsement was worth 5.8 percentage points in the 2017 DA race. This is about half of the 11 percentage point gap we saw in the naive analysis above; it turns out the other half was because of wards endorsing candidates that the voters already supported.

But some wards are much more important than others.

How does each ward’s endorsement stack up? The table below sorts the wards by order of the vote effect, which is the percentage effect of the endorsement times the 2017 primary turnout.

View code

ranef_ci <- bind_rows(
  lapply(bs_list, function(x) x$re), 
  .id = "sim"
) %>%
  group_by(ward) %>%
  summarise(
    p025 = quantile(effect, 0.025),
    p975 = quantile(effect, 0.975)
  )

fit_baseline$re %>% 
  select(ward, effect) %>% 
  left_join(ranef_ci) %>%
  mutate(
    ci = sprintf("(%0.1f, %0.1f)", 100 * p025, 100*p975)
  ) %>%
  left_join(
    da_results %>% 
      group_by(WARD16, last_name) %>%
      summarise(
        pvote = 100 * weighted.mean(pvote, w = total_votes),
        total_votes = sum(total_votes)
      ) %>%
      inner_join(
        endorsements, 
        by = c("WARD16" = "ward", "last_name" = "endorsement")
      ),
    by = c("ward" = "WARD16")
  ) %>%
  mutate(
    pvote = round(pvote, 0),
    effect = round(100 * effect, 0),
    vote_effect = round(effect/100 * total_votes)
  ) %>%
  rename(endorsement = last_name) %>%
  select(ward, endorsement, pvote, effect, ci, total_votes, vote_effect) %>%
  arrange(desc(vote_effect)) %>% 
  DT::datatable(
    rownames=FALSE,
    colnames = c("Ward", "Endorsee", "% of Vote in Ward", "Endorsement Effect at Boundary","CI", "Ward Votes", "Vote Effect of Endorsement")
  )

Ward	Endorsee	% of Vote in Ward	Endorsement Effect at Boundary	CI	Ward Votes	Vote Effect of Endorsement
10	Krasner	50	14	(8.0, 19.5)	3,719	521
09	Khan	37	12	(6.7, 17.5)	4,264	512
30	Khan	39	14	(8.5, 17.9)	3,403	476
52	Negrin	17	10	(1.1, 20.0)	3,768	377
36	Negrin	18	9	(4.8, 13.6)	3,932	354
56	Untermeyer	26	14	(9.9, 19.5)	2,346	328
50	Krasner	56	6	(-0.7, 15.6)	5,094	306
61	El-Shabazz	24	12	(4.7, 19.3)	2,547	306
40	O’Neill	15	8	(4.8, 14.3)	3,591	287
42	Krasner	45	20	(14.6, 25.2)	1,270	254
38	Negrin	29	10	(3.3, 21.0)	2,507	251
01	O’Neill	13	7	(3.7, 9.1)	2,954	207
03	Untermeyer	22	8	(3.0, 14.3)	2,312	185
63	O’Neill	24	8	(2.4, 16.0)	1,920	154
19	Negrin	50	26	(16.4, 40.6)	589	153
57	O’Neill	26	8	(4.5, 13.0)	1,719	138
65	O’Neill	26	8	(5.9, 11.0)	1,644	132
23	O’Neill	20	10	(5.2, 18.7)	1,284	128
05	Khan	30	2	(-2.5, 7.2)	5,927	119
60	Negrin	15	5	(2.0, 8.2)	2,350	118
07	Negrin	55	21	(6.2, 36.9)	548	115
21	Khan	34	2	(-3.3, 8.7)	5,383	108
31	Khan	23	5	(0.4, 9.0)	2,076	104
24	Untermeyer	11	7	(4.4, 11.0)	1,437	101
51	Untermeyer	13	4	(1.8, 9.8)	2,386	95
46	El-Shabazz	9	2	(-0.9, 5.3)	4,515	90
16	Untermeyer	21	9	(1.7, 13.0)	965	87
12	Krasner	42	3	(-0.7, 7.8)	2,627	79
27	Krasner	70	4	(-3.2, 11.4)	1,978	79
41	Khan	24	7	(3.9, 11.0)	975	68
48	Untermeyer	13	4	(1.8, 7.8)	1,623	65
64	O’Neill	22	7	(-0.8, 13.8)	795	56
58	Untermeyer	15	2	(-6.2, 7.2)	2,606	52
14	Negrin	12	5	(2.8, 7.9)	986	49
34	Krasner	34	1	(-3.6, 5.2)	4,900	49
06	Krasner	47	3	(-10.0, 10.1)	1,605	48
39	O’Neill	25	1	(-1.3, 5.9)	4,462	45
43	Negrin	23	4	(-0.1, 8.4)	1,091	44
25	Untermeyer	14	5	(-1.5, 11.7)	801	40
55	O’Neill	18	3	(-2.8, 7.9)	1,305	39
54	O’Neill	11	4	(0.8, 6.5)	720	29
62	O’Neill	23	2	(-1.8, 5.8)	1,126	23
45	Khan	22	2	(-1.6, 5.5)	891	18
32	El-Shabazz	27	1	(-3.3, 6.7)	1,727	17
33	Khan	26	3	(-1.4, 9.3)	566	17
35	Untermeyer	11	1	(-1.9, 5.6)	1,724	17
04	El-Shabazz	31	0	(-3.1, 4.5)	2,110	0
44	Krasner	37	0	(-5.4, 7.7)	1,416	0
47	El-Shabazz	15	-1	(-5.0, 5.9)	749	-7
29	Negrin	23	-2	(-5.5, 4.2)	1,331	-27
15	Negrin	18	-1	(-5.1, 3.4)	3,692	-37
49	El-Shabazz	19	-3	(-7.1, 0.3)	2,348	-70
22	El-Shabazz	13	-2	(-4.5, 1.3)	5,508	-110
08	Krasner	42	-3	(-6.8, 0.6)	6,091	-183

The most important ward in 2017 was Ward 10, which gave Krasner a 14 percentage point boost on a turnout of 3,719, meaning an estimated bump of 507 votes. (The exact order of the rankings has a lot of uncertainty. Don’t take them as gospel.) Those three Northwest wards we looked at above, 10, 9 and 50, were all in the top seven, with 10 and 9 making up first and second place, largely on the back of their high turnout.

What this means for May

This analysis is specific to the 2017 DA race in a number of ways. I expect ward endorsements to have more importance in low-information races, and all of the races this time around–City Council At Large, Judicial, and Commissioner–will be lower-information than the 2017 DA.

Consider the simplistic analysis I did for the 2017 Court of Common Please. That analysis found that endorsed candidates performed 0.9 percentage points better, in a race that took 4.3% of the vote to win. That estimate is the analog to the 11 point DA effect in the first table. We found that half of the 11 DA points was actually causal, so 0.45 points is a naive guess of the effect in judicial races.

But there are two more changes. First, taking half of the effect is almost certainly too conservative for judges. There are few pre-existing preferences among voters, so much less of that correlation will be “wards endorsing candidates that are already popular”. The causal part will be higher.

But second, the wards that I had data for in that analysis are all the wards with the strongest endorsement effect in this one: 9, 30, 52, and 50 were all among the 18 wards I had data for. So that estimate might be higher, too, than if we had data for every ward.

We end up in between. The Ward endorsements–especially in the top wards on the chart–are effective but not decisive. They are powerful enough that they likely decide close judicial races, but not enough to have changed 2017’s DA race.

Appendix: Ward Map

View code

ggplot(ggwards, aes(x = long, y=lat)) +
  geom_polygon(aes(group=group), fill = strong_green, color = "white") +
  geom_text(data = ward_centroids, aes(x=x, y=y, label=ward), color = "white") +
  theme_map_sixtysix() +
  coord_map() +
  ggtitle("Philadelphia's Wards")

March 25, 2019March 26, 2019

The neighborhoods (or wards) that decide District 7

Could Maria Lose?

Maria Quiñones-Sánchez is facing a significant challenge for the second election in a row. The three term councilmember was challenged by Manny Morales four years ago, and eked out a 53.5% – 46.5% win, a margin of only 868 votes. She’s being challenged again this year by state Rep and ward leader Angel Cruz. What should we expect?

The 7th district is decidedly different from Jannie Blackwell’s 3rd or Kenyatta Johnson’s 2nd. Those heavily-segregated districts had clear racial coalitions that swung differently from year to year.

North Philly’s 7th is segregated, but more homogenously so. It’s Philadelphia’s most Hispanic district, with a predominantly Black section of Frankford in the Northeast and the beginnings of White Kensington gentrification expanding up in the South. In the manner of Philadelphia’s Hispanic neighborhoods, it has very low turnout.

But the vote this May won’t split along racial lines. Instead, what matters is the Wards.

Four years ago, three of the district’s ward leaders coordinated to support Quiñones-Sánchez’s challenger, with heavy involvement from Johnny Doc’s Local 98, including more than $100,000 in contributions and over 1,000 “voter assist requests” for help in the voting booth.

The leaders are coordinating a challenge again, with a candidate Cruz who has apparently less baggage than Morales.

(I can’t quite do justice to the story of this race. Check out the links above and Billy Penn’s summary. )

The challenge seems stronger this year, with eight of the twelve ward leaders in the district supporting Cruz in the party endorsement. But Quiñones-Sánchez held it off last year. What should we expect?

District 7’s voting blocks

The 7th Council district covers North Philly, with 6th and 9th Streets serving as the Western border.

View code

library(tidyverse)
library(rgdal)
library(rgeos)
library(sp)
library(ggmap)

sp_council <- readOGR("../../../data/gis/city_council/Council_Districts_2016.shp", verbose = FALSE)
sp_council <- spChFIDs(sp_council, as.character(sp_council$DISTRICT))

sp_divs <- readOGR("../../../data/gis/2016/2016_Ward_Divisions.shp", verbose = FALSE)
sp_divs <- spChFIDs(sp_divs, as.character(sp_divs$WARD_DIVSN))
sp_divs <- spTransform(sp_divs, CRS(proj4string(sp_council)))

load("../../../data/processed_data/df_major_2017_12_01.Rda")

ggcouncil <- fortify(sp_council) %>% mutate(council_district = id)
ggdivs <- fortify(sp_divs) %>% mutate(WARD_DIVSN = id)

View code

## Need to add District result election from 2015
raw_d2 <-  read.csv("../../../data/raw_election_data/2015_primary.csv") 
raw_d2 <- raw_d2 %>% 
  filter(OFFICE == "DISTRICT COUNCIL-7TH DISTRICT-DEM") %>%
  mutate(
    WARD = sprintf("%02d", asnum(WARD)),
    DIV = sprintf("%02d", asnum(DIVISION))
  )

load('../../../data/gis_crosswalks/div_crosswalk_2013_to_2016.Rda')
crosswalk_to_16 <- crosswalk_to_16 %>% group_by() %>%
  mutate(
    WARD = sprintf("%02s", as.character(WARD)),
    DIV = sprintf("%02s", as.character(DIV))
  )

d2 <- raw_d2 %>% 
  left_join(crosswalk_to_16) %>%
  group_by(WARD16, DIV16, OFFICE, CANDIDATE) %>%
  summarise(VOTES = sum(VOTES * weight_to_16)) %>%
  mutate(PARTY="DEMOCRATIC", year="2015", election="primary")
df_major <- bind_rows(df_major, d2)

View code

races <- tribble(
  ~year, ~OFFICE, ~office_name,
  "2015", "MAYOR", "Mayor",
  "2015", "DISTRICT COUNCIL-7TH DISTRICT-DEM", "Council 7th District",
  "2016", "PRESIDENT OF THE UNITED STATES", "President",
  "2017", "DISTRICT ATTORNEY", "District Attorney"
) %>% mutate(election_name = paste(year, office_name))

candidate_votes <- df_major %>% 
  filter(election == "primary" & PARTY == "DEMOCRATIC") %>%
  inner_join(races %>% select(year, OFFICE)) %>%
  mutate(WARD_DIVSN = paste0(WARD16, DIV16)) %>%
  group_by(WARD_DIVSN, OFFICE, year, election) %>%
  mutate(
    total_votes = sum(VOTES),
    pvote = VOTES / sum(VOTES)
  ) %>% 
  group_by()
  
turnout_df <- candidate_votes %>%
  filter(!grepl("COUNCIL", OFFICE)) %>% 
  group_by(WARD_DIVSN, OFFICE, year, election) %>%
  summarise(total_votes = sum(VOTES)) %>%
  left_join(
    sp_divs@data %>% select(WARD_DIVSN, AREA_SFT)
  )

turnout_df$AREA_SFT <- asnum(turnout_df$AREA_SFT)

View code

get_labpt_df <- function(sp){
  mat <- sapply(sp@polygons, slot, "labpt")
  df <- data.frame(x = mat[1,], y=mat[2,])
  return(
    cbind(sp@data, df)
  )
}

ggplot(ggcouncil, aes(x=long, y=lat)) +
  geom_polygon(
    aes(group=group),
    fill = strong_green, color = "white", size = 1
  ) +
  geom_text(
    data = get_labpt_df(sp_council),
    aes(x=x,y=y,label=DISTRICT)
  ) +
  theme_map_sixtysix() +
  coord_map() +
  ggtitle("Council Districts")

View code

DISTRICT <- "7"
sp_district <- sp_council[row.names(sp_council) == DISTRICT,]

bbox <- sp_district@bbox
## expand the bbox 20%for mapping
bbox <- rowMeans(bbox) + 1.2 * sweep(bbox, 1, rowMeans(bbox))

if(file.exists("map_cache.Rda")){
  load("map_cache.Rda")
} else {
    basemap <- get_map(bbox, maptype="toner-lite", filename="map_cache.png")
    save(basemap, file="map_cache.Rda")
}

district_map <- ggmap(
  basemap, 
  extent="normal", 
  base_layer=ggplot(ggcouncil, aes(x=long, y=lat, group=group)),
  maprange = FALSE
) 
## without basemap:
# district_map <- ggplot(ggcouncil, aes(x=long, y=lat, group=group))

district_map <- district_map +
  theme_map_sixtysix() +
  coord_map(xlim=bbox[1,], ylim=bbox[2,])


sp_divs$council_district <- over(
  gCentroid(sp_divs, byid = TRUE), 
  sp_council
)$DISTRICT

polygon_in_bbox <- function(p) {
  coords <- p@Polygons[[1]]@coords
  any(
    coords[,1] > bbox[1,1] &
      coords[,1] < bbox[1,2] &
      coords[,2] > bbox[2,1] &
      coords[,2] < bbox[2,2] 
  )
}

sp_divs$in_bbox <- sapply(
  sp_divs@polygons,
  polygon_in_bbox
)

ggdivs <- ggdivs %>% 
  left_join(
    sp_divs@data %>% select(WARD_DIVSN, in_bbox)
  )

district_map +
  geom_polygon(
    aes(alpha = (id == DISTRICT)),
    fill="black",
    color = "grey50",
    size=2
  ) +
  scale_alpha_manual(values = c(`TRUE` = 0.2, `FALSE` = 0), guide = FALSE) +
  ggtitle(sprintf("Council District %s", DISTRICT))

The district has the lowest turnout in the city. Philadelphia’s Hispanic neighborhoods have very low turnout, and this district is the most Hispanic. Curiously, not only does the neighborhood have low turnout in Presidential elections, but it then has disproportionately lower turnout in municipal elections even given that: even residents who vote for President are less likely to vote in other years.

View code

# hist(turnout_df$total_votes / turnout_df$AREA_SFT, breaks = 1000)

turnout_df <- turnout_df %>%
  left_join(races)

district_map +
  geom_polygon(
    data = ggdivs %>%
      filter(in_bbox) %>%
      left_join(turnout_df, by =c("id" = "WARD_DIVSN")),
    aes(fill = pmin(total_votes / AREA_SFT, 0.0005) * 5280^2)
  ) +
  scale_fill_viridis_c(
    "Votes per mile", 
    labels=scales::comma, 
    guide=guide_colorbar(label.theme=element_text(angle=90, size = 10), label.hjust=1)
  ) +
  geom_polygon(
    fill=NA,
    color = "white",
    size=1
  ) +
  facet_wrap(~ election_name) +
  expand_limits(fill=0) +
  ggtitle(
    "Votes per mile in the Democratic Primary", 
    sprintf("Council District %s", DISTRICT)
  ) +
  theme(legend.position = "bottom", legend.direction = "horizontal")

The district as a whole cast 25,000 votes in the 2016 Presidential primary, 12,000 last time Quninones-Sanchez ran, and only 6,000 in the the 2017 District Attorney race. They did not see the Krasner surge.

Demographically, the district is Philadelphia’s most heavily hispanic, though with Frankford in its Northeast being predominantly Black:

View code

bg_17_acs <- read.csv("../../../data/census/acs_2013_2017_phila_bg_race_income.csv")
bg_17_acs <- bg_17_acs %>% 
  mutate(Geo_FIPS = as.character(Geo_FIPS)) %>%
  select(Geo_FIPS, pop, pop_nh_white, pop_nh_black, pop_nh_asian, pop_hisp, pop_median_income_2017)

library(tigris)
options(tigris_use_cache = TRUE)
bg_shp <- block_groups(42, 101, year = 2015)
bg_shp <- spChFIDs(bg_shp, as.character(bg_shp$GEOID))
bg_shp <- spTransform(bg_shp, CRS(proj4string(sp_divs)))

bg_shp$in_bbox <- sapply(
  bg_shp@polygons,
  polygon_in_bbox
)

gg_bgs <- fortify(bg_shp)
gg_bgs <- gg_bgs %>%
  left_join(bg_shp@data[,c("GEOID", "ALAND", "in_bbox")], by = c("id" = "GEOID")) %>%
  left_join(bg_17_acs, by = c("id" = "Geo_FIPS"))

district_map +
  geom_polygon(
    data = gg_bgs %>%
      filter(in_bbox) %>%
      gather("key", "race_pop",pop_nh_white, pop_nh_black, pop_nh_asian, pop_hisp) %>%
      mutate(
        pct_pop = 100 * race_pop / pop,
        race = c(
          pop_hisp = "Hispanic", pop_nh_white="NH White", pop_nh_black="Black", pop_nh_asian="Asian"
        )[key]
      ),
    aes(fill = pct_pop)
  ) + 
  geom_polygon(
    fill=NA,
    color = "white",
    size=1
  ) +
  facet_wrap(~race) +
  scale_fill_viridis_c("Percent of\n Population") +
  theme(legend.position = "right") +
  ggtitle(sprintf("Race and Ethnicity of District %s", DISTRICT))

The district is less obviously politically split than other parts of the city. When it does split, it often does so for Latino candidates. Below are the results from the last race for the 7th and three other recent, compelling Democratic Primary races: 2015 Mayor, 2016 President, and 2017 District Attorney. The maps below show the vote for the top two candidates in District 2 (except for City Council in 2015, where I use Helen Gym and Isaiah Thomas, who were 4th and 5th in the district, and 5th and 6th citywide.)

View code

candidate_votes <- candidate_votes %>%
  left_join(sp_divs@data %>% select(WARD_DIVSN, council_district))

## Choose the top two candidates in district 3
## Except for city council, where we choose Gym and Thomas
# candidate_votes %>%
#   group_by(OFFICE, year, CANDIDATE) %>%
#   summarise(
#     city_votes = sum(VOTES),
#     district_votes = sum(VOTES * (council_district == DISTRICT))
#   ) %>%
#   arrange(desc(city_votes)) %>%
#   filter(OFFICE == "MAYOR")

candidates_to_compare <- tribble(
  ~year, ~OFFICE, ~CANDIDATE, ~candidate_name, ~row,
  "2015", "DISTRICT COUNCIL-7TH DISTRICT-DEM", "MANNY MORALES", "Manny Morales", 1,
  "2015", "DISTRICT COUNCIL-7TH DISTRICT-DEM", "MARIA QUINONES SANCHEZ", "Maria Quiñones-Sánchez", 2,
  "2015", "MAYOR", "JIM KENNEY", "Jim Kenney",  2,
  "2015", "MAYOR", "NELSON DIAZ", "Nelson Diaz", 1,
  "2016", "PRESIDENT OF THE UNITED STATES", "BERNIE SANDERS", "Bernie Sanders", 2,
  "2016", "PRESIDENT OF THE UNITED STATES", "HILLARY CLINTON", "Hillary Clinton", 1,
  "2017", "DISTRICT ATTORNEY", "LAWRENCE S KRASNER", "Larry Krasner", 2,
  "2017", "DISTRICT ATTORNEY", "RICH NEGRIN","Rich Negrin", 1
)

candidate_votes <- candidate_votes %>%
  left_join(races) %>%
  left_join(candidates_to_compare)

vote_adjustment <- function(pct_vote, office){
  ifelse(office == "COUNCIL AT LARGE", pct_vote * 4, pct_vote)
}

district_map +
  geom_polygon(
    data = ggdivs %>%
      filter(in_bbox) %>%
      left_join(
        candidate_votes %>% filter(!is.na(row))
      ),
    aes(fill = 100 * vote_adjustment(pvote, OFFICE))
  ) +
  scale_fill_viridis_c("Percent of Vote") +
  theme(
    legend.position =  "bottom",
    legend.direction = "horizontal",
    legend.justification = "center"
  ) +
  geom_polygon(
    fill=NA,
    color = "white",
    size=1
  ) +
  geom_label(
    data=candidates_to_compare %>% left_join(races),
    aes(label = candidate_name),
    group=NA,
    hjust=0, vjust=1,
    x=bbox[1,1],
    y=bbox[2,2]
  ) +
  facet_grid(row ~ election_name) +
  theme(strip.text.y = element_blank()) +
  ggtitle(
    sprintf("Candidate performance in District %s", DISTRICT) 
    # "Percent of vote (times 4 for Council, times 1 for other offices)"
  )

The district split for Mayor and for District Attorney along racial lines: the densest latino neighborhoods voted for Diaz and Negrin, while the rest of the district voted for the citywide winners Kenney and Krasner. However, the cleavage was geographically different for Quiñones-Sánchez v Morales, presumably because both candidates were latinx. That’s true this time around, too.

Just because there wasn’t a racial split doesn’t mean the district voted uniformly. There indeed were stark variations in how Quiñones-Sánchez did, but the reason isn’t obviously clear. She did well in the Northwest of the district, especially West of 6th St, and in the Northeast of the district except for South of the Boolevard and East of Oxford.

What gives? Those are all Ward boundaries.

View code

wards <- readOGR("../../../data/gis/2016","2016_Wards", verbose=FALSE) %>%
  spTransform(CRS(proj4string(sp_divs)))

ggwards <- fortify(wards)

wards_centers <- sapply(wards@polygons, slot, "labpt") %>% t
wards_centers <- as.data.frame(wards_centers)
names(wards_centers) <- c("x", "y")

wards@data <- cbind(wards@data, wards_centers)


district_map + 
  geom_polygon(data = ggwards, fill = NA, color = strong_red, size= 2) +
  geom_polygon(
    data = ggcouncil %>% filter(council_district == 7), 
    fill=NA, 
    color = "black", 
    size = 2
  ) +
  geom_text(
    data = wards@data,
    aes(x=x, y=y, label=WARD),
    group = NA,
    color = strong_red,
    fontface="bold"
  ) +
  ggtitle("The wards of District 7")

Quiñones-Sánchez’s performance is a potent example of the power of Ward endorsements: she performed very differently in different wards, in some cases on literally the other side of the street. That thin sliver in the Northwest of the District where she did exceptionally well was the 43rd Ward. She received large percentages in the 23rd and 54th in the Northwest, too, but that region where she did poorly, South of the Boolevard and East of Oxford, exactly lines up with the boundaries to the 62nd.

So the question for the 2019 race becomes how the ward endorsements will shake out and how powerful each ward is, both in terms of ability to swing the vote and typicaly turnout. Below are measures of their strength. I’ve also pulled in the Ward leaders’ endorsements from the DCC vote.

View code

## Get vote-weighted populations
div_centroids <- gCentroid(sp_divs[sp_divs$council_district == DISTRICT,], byid=TRUE)
div_centroids$WARD_DIVSN <- attr(div_centroids@coords, "dimnames")[[1]]
div_centroids$bg_GEOID <- over(div_centroids, bg_shp)$GEOID
div_centroids@data <- left_join(div_centroids@data, bg_17_acs, by = c("bg_GEOID" = "Geo_FIPS"))

district_7_results <- df_major %>%
  filter(
    year == 2015 & grepl("7TH DISTRICT", OFFICE) & CANDIDATE != "Write In"
  ) %>%
  mutate(WARD_DIVSN = paste0(WARD16, DIV16)) %>%
  select(WARD_DIVSN, CANDIDATE, VOTES) %>%
  spread(CANDIDATE, VOTES) %>%
  mutate(
    total_votes = (`MARIA QUINONES SANCHEZ` + `MANNY MORALES`),
    p_quinones_sanchez = `MARIA QUINONES SANCHEZ` / total_votes
  )

div_centroids@data <- left_join(div_centroids@data, district_7_results)

ward_pops <- div_centroids@data %>%
  mutate(ward = substr(WARD_DIVSN, 1, 2)) %>%
  group_by(ward) %>%
  summarise(
    p_quinones_sanchez = 100 * weighted.mean(p_quinones_sanchez, w = total_votes),
    pct_nh_white = 100 * weighted.mean(pop_nh_white / pop, w = total_votes),
    pct_nh_black = 100 * weighted.mean(pop_nh_black / pop, w = total_votes),
    pct_nh_asian = 100 * weighted.mean(pop_nh_asian / pop, w = total_votes),
    pct_hisp = 100 * weighted.mean(pop_hisp / pop, w = total_votes),
    council_votes = sum(total_votes)
  )

ward_results <- df_major %>%
  mutate(WARD_DIVSN = paste0(WARD16, DIV16)) %>%
  inner_join(
    tribble(
      ~year, ~election, ~PARTY, ~OFFICE,
      "2016", "primary", "DEMOCRATIC", "PRESIDENT OF THE UNITED STATES",
      "2017", "primary", "DEMOCRATIC", "DISTRICT ATTORNEY"
    )
  ) %>%
  inner_join(div_centroids@data) %>%
  group_by(year, WARD16) %>%
  summarise(total_votes = sum(VOTES)) %>%
  group_by() %>%
  group_by(year) %>%
  mutate(pct_of_year = 100 * total_votes / sum(total_votes)) %>%
  gather("key", "value", total_votes, pct_of_year) %>%
  unite("key", key, year) %>%
  spread(key, value)
    
ward_pops <- ward_pops %>% 
  left_join(
    ward_results %>% rename(ward = WARD16)
  ) %>%
  group_by() %>%
  mutate(pct_of_year_2015 = 100 * council_votes / sum(council_votes))

ward_leaders <- tribble(
  ~ward, ~leader, ~endorsement,
  "07", "Angel Cruz", "Cruz",
  "18", "Theresa Alicea", "Cruz",
  "19", "Carlos Matos","Cruz",
  "23", "Timothy Savage","Quiñones-Sánchez",
  "25", "Thomas Johnson","Cruz",
  "31", "Margaret Rzepski","Cruz",
  "33", "Donna Aument","Cruz",
  "42", "Sharon Vaughn","Quiñones-Sánchez",
  "43", "Emilio Vazquez","Quiñones-Sánchez",
  "49", "Shirley Gregory","Cruz",
  "54", "Alan Butkovitz","Quiñones-Sánchez",
  "62", "Margaret Tartaglione","Cruz"
)

ward_pops <- ward_pops %>% left_join(ward_leaders)

knitr::kable(
  ward_pops %>%
    select(ward, council_votes, pct_of_year_2015, p_quinones_sanchez, leader, endorsement, pct_hisp, pct_nh_white, pct_nh_black, total_votes_2016, pct_of_year_2016, total_votes_2017, pct_of_year_2017) %>%
    arrange(desc(council_votes)),
    digits=0, 
    format.args=list(big.mark=','),
    col.names=c("Ward", "Votes for the 7th in 2015 Primary", "Pct of District", "% for Quiñones-Sánchez",  "Leader", "Endorsement", "% Hispanic", "% White", "% Black", "Votes in 2016 Primary", "Pct of District", "Votes in 2017 Primary", "Pct of District")
  )

Ward	Votes for the 7th in 2015 Primary	Pct of District	% for Quiñones-Sánchez	Leader	Endorsement	% Hispanic	% White	% Black	Votes in 2016 Primary	Pct of District	Votes in 2017 Primary	Pct of District
23	2,108	17	71	Timothy Savage	Quiñones-Sánchez	28	19	46	4,021	16	1,284	22
62	1,774	14	33	Margaret Tartaglione	Cruz	31	22	41	3,546	14	913	16
19	1,683	14	37	Carlos Matos	Cruz	77	6	15	2,817	11	589	10
07	1,655	14	49	Angel Cruz	Cruz	80	8	11	3,340	13	549	9
33	1,455	12	57	Donna Aument	Cruz	63	10	18	3,337	13	566	10
42	1,177	10	63	Sharon Vaughn	Quiñones-Sánchez	64	8	16	2,581	10	543	9
43	1,124	9	65	Emilio Vazquez	Quiñones-Sánchez	66	3	29	2,259	9	445	8
18	611	5	51	Theresa Alicea	Cruz	45	29	16	1,340	5	527	9
54	311	3	66	Alan Butkovitz	Quiñones-Sánchez	17	18	49	906	4	196	3
25	133	1	59	Thomas Johnson	Cruz	47	33	15	350	1	77	1
49	117	1	72	Shirley Gregory	Cruz	28	0	72	192	1	44	1
31	98	1	63	Margaret Rzepski	Cruz	23	57	8	265	1	129	2

The ward with the most votes for the 7th in 2015 also voted hard for Quiñones-Sánchez: the 23rd. Her 71% dominance of the 2,108 votes cast gave her an 886 vote edge, more than what she won the entire District by. That ward, which includes the district’s predominantly Black neighborhoods, turns out stronger than the rest. It represented 17% of the district’s votes in 2015, and then a whopping 22% of the votes in low-turnout 2017.

The next three most dominant wards were 62, 19, and 7 and those went for Morales with 63, 67, and 51% of the vote. (Notice that 62 is actually split by Council Districts; the number above is only for District 7). Those are the three wards led by State Senator Margaret Tartaglione, Carlos Matos, and Angel Cruz himself, the ward leaders that organized the challenge. Quiñones-Sánchez won all of the other Wards, but those three were enough to make the race close.

We can simplify the table above by combining all of the wards whose leaders supported Quiñones-Sánchez and those whose leaders supported Cruz.

View code

get_line <- function(x_total_votes, y_total_votes){
  ## solve p_x t_x+ p_y t_y > 50
  tot <- x_total_votes + y_total_votes
  tx <- x_total_votes / tot
  ty <- y_total_votes / tot

  slope <- -tx / ty
  intercept <- 50 / ty  # use 50 since proportions are x100
  c(intercept, slope)
}

endorsement_summary <- ward_pops %>%
  group_by(endorsement) %>%
  summarise(
    p_quinones_sanchez = weighted.mean(p_quinones_sanchez, w = council_votes),
    council_votes = sum(council_votes),
    total_votes_2016 = sum(total_votes_2016),
    total_votes_2017 = sum(total_votes_2017)
  )

candidate_results = with(
  endorsement_summary,
  tribble(
    ~candidate, 
    ~p_in_mqs_wards,
    ~p_in_challenger_wards, 
    ~total_votes_in_mqs_wards, 
    ~total_votes_in_challenger_wards,
    "Quiñones-Sánchez", 
    p_quinones_sanchez[endorsement == "Quiñones-Sánchez"],
    p_quinones_sanchez[endorsement == "Cruz"],
    council_votes[endorsement == "Quiñones-Sánchez"],
    council_votes[endorsement == "Cruz"],
    "Morales", 
    100 - p_quinones_sanchez[endorsement == "Quiñones-Sánchez"],
    100 - p_quinones_sanchez[endorsement == "Cruz"],
    council_votes[endorsement == "Quiñones-Sánchez"],
    council_votes[endorsement == "Cruz"]
  )
) %>%
  mutate(
    votes_in_mqs_wards = p_in_mqs_wards * total_votes_in_mqs_wards / 100,
    votes_in_challenger_wards = p_in_challenger_wards * total_votes_in_challenger_wards / 100
  )

knitr::kable(
  candidate_results %>% select(
    candidate,
    p_in_mqs_wards,
    votes_in_mqs_wards,
    p_in_challenger_wards,
    votes_in_challenger_wards
  ),
  digits=0, 
  format.args=list(big.mark=','),
  col.names=c("Candidate", "Percent in MQS-Endorsed Wards", "Votes in MQS-Endorsed Wards", "Percent in Cruz-Endorsed Wards",  "Votes in Cruz-Endorsed Wards")
)

Candidate	Percent in MQS-Endorsed Wards	Votes in MQS-Endorsed Wards	Percent in Cruz-Endorsed Wards	Votes in Cruz-Endorsed Wards
Quiñones-Sánchez	67	3,174	45	3,383
Morales	33	1,546	55	4,143

The four wards that backed Quiñones-Sánchez constituted only 39% of the votes in 2015, but she won them 2:1. The other eight wards combined for 61% of the vote, but Morales only won them 5:4. Quiñones-Sánchez’s wards represented more of the votes in 2017, boding well for this year, but it’s also likely that Cruz will do better in his wards than Morales did.

View code

line_2017 <- with(
  endorsement_summary,
  get_line(
    total_votes_2017[endorsement == "Quiñones-Sánchez"],
    total_votes_2017[endorsement == "Cruz"]
  )
)

line_2015 <- with(
  endorsement_summary,
  get_line(
    council_votes[endorsement == "Quiñones-Sánchez"],
    council_votes[endorsement == "Cruz"]
  )
)

library(ggrepel)
ggplot(
  candidate_results,
  aes(
    x=p_in_mqs_wards,
    y=p_in_challenger_wards
  )
) +
  geom_point() +
  geom_text_repel(aes(label=candidate)) +
  geom_abline(
    intercept = c(line_2015[1], line_2017[1]),
    slope = c(line_2015[2], line_2017[2]),
    linetype="dashed"
  ) +
  coord_fixed() +
  scale_x_continuous(
    "Percent in wards that endorsed Quiñones-Sánchez",
    breaks = seq(0,100,10)
  ) +
  scale_y_continuous(
    "Percent in wards that endorsed Cruz",
    breaks = seq(0, 100, 10)
  ) +
  annotate(
    geom="text",
    label=paste(c(2015, 2017), "turnout"),
    x=c(10, 8),
    y=c(
      line_2015[1] + 10 * line_2015[2],
      line_2017[1] + 8 * line_2017[2]
    ),
    hjust=0,
    vjust=-0.2,
    angle = atan(c(line_2015[2], line_2017[2])) / pi * 180,
    color="grey40"
  )+
  annotate(
    geom="text",
    x = 80,
    y=75,
    label="Candidate wins",
    fontface="bold",
    color = strong_green
  ) +
  geom_hline(yintercept = 50, color="grey50") +
  geom_vline(xintercept = 50, color="grey50")+
  expand_limits(x=100, y=30)+
  theme_sixtysix() +
  ggtitle(
    "The strength of District 7's wards in 2015",
    "Candidates to the top-right of dashed lines win."
  )

Looking to May

So what are we to make of this race? Keep your eyes on the mobilization behind the Ward endorsements, especially the top turnout wards. Cruz will presumably do even better in his own ward than Morales did, so Quiñones-Sánchez’s success will likely hinge on continued high turnout in the 23rd and trying to consolidate the votes of the rest.

She managed to do that just well enough four years ago to eke out a 9 point victory. This year she faces a candidate without homophobic Facebook posts who also happens to be a Ward leader. If the results in every district were the same as in 2015 except Angel Cruz managed to win 77% of the vote in his own ward, the race would be an exact tie.

It’ll be close.