Filtering On Columns

SELECT count(*) FROM ipums_db
  WHERE YEAR = 1910

SELECT * FROM ipums_db
  WHERE YEAR = 1910
  LIMIT 10

SELECT YEAR, SAMPLE, STATEFIP, SEX, AGE, LIT, MARST FROM ipums_db
  WHERE YEAR = 1910
  AND STATEFIP = 48
  LIMIT 10

Visualizing Our Query

From here, I may want to take this data and visualize it. Unfortunately, SQL really isn’t the best place to do that, but I can just pass this data to R or Python with Quarto and start going to work on some nice charts. In my case, I’ve tried to line up these two distributions and compare them by decade and gender for a simple visual comparison. While imagining the visualization in my mind, I decided to do some extra transformation on the 253 rows I pulled. It was simple enough to quickly do that within R and pass it to our chart for the comparison.

At a quick glance, we can some similarities in population size before age 30. Once 30 hits, the gaps start to widen. We could explore that more, but that’s not this post! So I’ll stop elaborating on this visualization for now.

library(tidyverse)

chart_query_df <- as_query |> 
  mutate(decade = cut(age, breaks = c(seq(0, max(age) + 9, by = 10), Inf), 
                      right = FALSE, labels = FALSE, include.lowest = TRUE),
         decade = decade-1,
         decade_lbl = if_else(decade==0,
                              paste0(decade, "-", decade+1, "0"),
                              paste0(decade, "0-", decade+1, "0"))) |> 
  group_by(decade, decade_lbl, gender) |> 
  summarize(count = sum(count))

chart_query_df |> 
  ggplot() +
  aes(x=reorder(decade_lbl, decade), y = count, fill = gender) +
  geom_col(alpha = 0.8, position = position_dodge()) +
  geom_col(data = chart_query_df |> filter(decade >= 30), position = position_dodge(),) + 
  scale_y_continuous(labels = scales::label_comma(scale = .001, suffix = "K"),
                     n.breaks = 6) +
  scale_fill_manual(values = c(Male = "#005f86", Female = "#bf5700")) + 
  mrworthingtonR::theme_mrw(base_family = "Graphik",
                            subtitle_size = 18) +
  theme(plot.title = ggtext::element_markdown(margin = ggplot2::margin(b = 5), hjust = 0),
        panel.grid.major.y = element_line(color = alpha("#8d8d8d", 0.2), linetype = 1)) +
  labs(title = "In 1910, Texas <span style='color:#005f86'>Men</span> Seemed to Live Longer than <span style='color:#bf5700'>Women</span> After Age 30",
       subtitle = "Distributions Among Texan Ages, by Gender | 1910 Complete US Census Count",
       y = "Population Count",
       x = "Age Group",
       caption = "Source: IPUMS 1910 Complete-Count US Census File (CSV)")

First of all, {tidycensus}

I didn’t actually learn about tidycensus through sharing the last post, but it’s another incredible asset in my own spatial workflow. Part of that is because I went to grad school for public policy in the US and working with data from the US Census Bureau was…

1. Something I wanted to be good at

2. A slight nightmare to work with. Especially if it involved making maps or doing any kind of spatial analysis.

It’s hard to overstate how much Kyle Walker’s tidycensus addressed both of those in the best way possible. Let me show you how. First things first, let’s load the packages.

```{r load-packages}
#| code-summary: "**The Code** | Package Setup"
library(tidyverse)     # Loads core tidyverse tools
library(tidycensus)    # Loads tools to pull census data
library(viridis)       # Loads color palettes I use w/ {rdeck}
library(rdeck)         # Loads rdeck library
library(sf)            # Loads core spatial tools
options(tigris_use_cache = TRUE) # Caches data we pull from tidycensus to speed up recalls
```

```{r get-acs-spatial}
#| code-fold: show
#| code-summary: "**The Code** | {tidycensus} Call"

travis <- get_acs(
  state = "TX",
  county = "Travis",
  geography = "tract",
  variables = "B19013_001",
  geometry = TRUE,
  year = 2020
) |> 
  st_transform(crs = st_crs(4326))
```

Alright, so what happened? Well, we got data for every single census tract’s Median Household Income in Travis County. A snapshot of that data is shown in the output below, but wait… what’s that st_transform() thing about in the chunk above?

Simple feature collection with 6 features and 5 fields
Geometry type: MULTIPOLYGON
Dimension:     XY
Bounding box:  xmin: -97.77649 ymin: 30.18394 xmax: -97.644 ymax: 30.32954
Geodetic CRS:  WGS 84
        GEOID                                     NAME   variable estimate
1 48453002201 Census Tract 22.01, Travis County, Texas B19013_001    57803
2 48453000304  Census Tract 3.04, Travis County, Texas B19013_001    80000
3 48453001307 Census Tract 13.07, Travis County, Texas B19013_001    76009
4 48453002105 Census Tract 21.05, Travis County, Texas B19013_001    40811
5 48453002413 Census Tract 24.13, Travis County, Texas B19013_001    45462
6 48453000801  Census Tract 8.01, Travis County, Texas B19013_001    83958

So, now, I’ve got my spatial data and want to map it. Well, one thing I thought would be great is trying to replicate a ggplot2-style map. I know I could use ggiraph (another amazing package I’ll write about at some point), but I wanted to make something that feels like a ggplot2 map made with geom_sf(), but interactive. I also learned that rdeck was partly inspired by ggplot2. To get that feel with rdeck, you can set the map_style argument to NULL and controller=FALSE get something similar to that ggplot2-style map.

```{r tidycensus-map}
#| column: page
#| fig-height: 7
#| code-summary: "**The Code** | {tidycensus} ggplot-style  Map"

travis_map <- rdeck(map_style = NULL,
      initial_bounds = travis,
      controller = FALSE,
      theme = "light") |> 
  add_polygon_layer(data = travis,
                    visibility_toggle = TRUE,
                    name = "Median Household Income (by Census Tract)",
                    opacity = 0.9,
                    get_polygon = geometry,
                    tooltip = estimate,
                    pickable = TRUE,
                    auto_highlight = TRUE,
                    get_fill_color = scale_color_linear(
                      col = estimate,
                      palette = magma(10)
                    ))

travis_map
```

Annotating{rdeck} with the {rdeck} editor

One really amazing thing I didn’t realize is that rdeck has an editor built-in. The editor can be turned on by setting editor=TRUE in the main rdeck() function. Miles McBain pointed out that you can use this with shiny for filtering spatial data and any linked data in-app. That sounds very useful, but I also think you can use it for creating custom annotations to spatial maps if you want to call out a particular area on maps you regularly make.

This sounds simple enough, but spatial annotations can be pretty difficult without a tool like this where you can see the area you’re trying to highlight on a map. Take a look at the GIF below to see how it works.

In short, you click on the tool you want (in my case a polygon), then you can draw with the tool, and once you’re done, rdeck saves it as a “.geojson” file which can be read back into R using read_sf() and now you can add that spatial object onto your map as an annotation. The code chunk below does just that. Pretty great, huh?

```{r annotated-map}
#| column: page
#| fig-height: 7
#| code-summary: "**The Code** | Annotated Map"

custom_shape <- read_sf("rdeck.geojson")

travis_map |> 
  add_polygon_layer(data = custom_shape,
                    name = "Highlighted Area",
                    opacity = 0.1,
                    get_polygon = geometry,
                    get_fill_color = "#5da5da")
```

Anyway, enjoy digging more into rdeck (and tidycensus). Hope you find new stuff and write about it! Doesn’t even have to be a blog. Could just be twitter.