Milwaukee-Geo-Crosswalks

This repo creates crosswalks between common Milwaukee geographies, e.g census tracts and neighborhood.

Download the crosswalks as CSVs

• Census blocks (2010) to voting wards (2018)
• Census blocks (2010) to neighborhoods
• Census tracts (2017) to neighborhoods
• Voting wards (2018) to neighborhoods
• Census block groups (2017) to neighborhoods
• Census block groups (2017) to voting wards (2018)

What is this?

When working with spatial data, the info we want is often available for a different kind of geography than the areas we care about. For example, some Census tracts are located entirely within neighborhoods, but others straddle two or more. The crosswalk files I’ve provided above allow data users to conveniently (dis)aggregate data from one geography into another.

Method

A quick way of creating a crosswalk is just to intersect the land areas of the two geographies. If 45% of geography A’s land lies within geography B, then 45% of geography A’s data is assigned to geography B. My method is a bit more sophisticated. I take the centroid of every residential parcel in the City of Milwaukee and intersect it with the data source geography. Then I calculate the proportion of bedrooms in geography A which are also in geography B.

How it works

Here is what the crosswalks look like.

There is 1 row for every target geography/data geography combination–in this case Census blocks to neighborhoods. The first line tells us that there are 154 bedrooms in both Alcott Park and Census block 550790195001005. These 154 bedrooms make up 100% of the bedrooms in the Census block, and they make up 9.4% of the bedrooms in Alcott Park.

neighborhood	block	bedrooms	pct.of.block	pct.of.neighborhood
ALCOTT PARK	550790195001005	154	1	0.0939597
ALCOTT PARK	550790195001006	238	1	0.1452105
ALCOTT PARK	550790195001007	113	1	0.0689445
ALCOTT PARK	550790195001008	109	1	0.0665040
ALCOTT PARK	550790195002008	133	1	0.0811470
ALCOTT PARK	550790195002009	138	1	0.0841977

Here are the 2010 decennial census population counts for each block.

block	pop
550790001011003	139
550790001011000	77
550790001011001	235
550790001011002	73
550790001011004	865
550790001012006	156

Merge them. Calculate the population in each neighborhood/block combo by multiplying the pct.of.block adjustment factor by the population of the block. Group by neighborhood, then summarize.

blocks.to.neighborhoods %>%
  inner_join(total.pop) %>%
  mutate(pop.in.neighborhood = pop * pct.of.block) %>%
  group_by(neighborhood) %>%
  summarise(pop = sum(pop.in.neighborhood)) %>%
  arrange(desc(pop))

## # A tibble: 186 x 2
##    neighborhood                pop
##    <chr>                     <dbl>
##  1 BAY VIEW                 16613 
##  2 LINCOLN VILLAGE          13062 
##  3 OLD NORTH MILWAUKEE      12226.
##  4 HISTORIC MITCHELL STREET 12016 
##  5 MUSKEGO WAY              11932 
##  6 HARAMBEE                 11815 
##  7 RIVERWEST                11520 
##  8 SILVER SPRING            11441 
##  9 LOWER EAST SIDE          10333.
## 10 MORGANDALE                9040 
## # … with 176 more rows

The above process works as long as you are dealing with a population count. If you want to (dis)aggregate a summary value (like median age), you can use the pct.of.neighborhood value to calculate a weighted mean.

blocks.to.neighborhoods %>%
  inner_join(median.age) %>%
  group_by(neighborhood) %>%
  summarise(median_age = weighted.mean(x = age, w = pct.of.neighborhood, na.rm = T)) %>%
  arrange(desc(median_age))

## # A tibble: 186 x 2
##    neighborhood     median_age
##    <chr>                 <dbl>
##  1 FREEDOM VILLAGE        70.8
##  2 WHISPERING HILLS       65.8
##  3 HILLSIDE               64.4
##  4 BRYNWOOD               61.3
##  5 GRANVILLE WOODS        54.4
##  6 VALLEY FORGE           52.0
##  7 MACK ACRES             51.3
##  8 MILL VALLEY            48.0
##  9 GOLDEN GATE            47.4
## 10 COPERNICUS PARK        46.6
## # … with 176 more rows

How accurate is it?

The smaller the data source geography, the more likely that it will fit entirely within the target geography. The table below shows how each crosswalk performs. Each cell shows the percent of bedrooms matched at the degree of overlap indicated in the first column

• 99.1% of bedrooms in Census blocks fell entirely within a single neighborhood.
• 99.9% of bedrooms Census blocks fell entirely within a single ward.
• Tracts are much more likely to straddle geographies. Just 27.8% of tract-bedrooms fell within a single neighborhood. 31.5% of bedrooms in tracts are assigned to a neighborhood with less than 50% overlap.
• Wards can be merged to neighborhoods more accurately than tracts, but less accurately than blocks. 37.1% of wards lie entirely within a single neighborhood.
• Blockgroups fit within neighborhoods considerably better than tracts do. Of course, block group data also comes with a considerably higher margin of error.

geo overlap	blocks to neighborhoods	blocks to wards	tracts to neighborhoods	wards to neighborhoods	blockgroups to wards	blockgroups to neighborhoods
100%	99.13	99.93	27.83	37.14	35.53	60.82
98-100	0.08	0.00	5.96	4.95	2.55	4.34
90-98	0.21	0.04	4.58	4.76	4.98	2.87
80-90	0.09	0.00	4.55	6.96	5.06	5.00
70-80	0.15	0.00	7.40	8.79	8.69	4.34
60-70	0.07	0.01	6.12	9.61	8.60	4.90
50-60	0.09	0.00	12.49	5.98	10.05	5.96
less than 50	0.18	0.01	31.09	21.80	24.54	11.77