Framed!

class: center, middle, inverse, title-slide

# Framed!
## Reproducing 150 Year Old Charts
### Susan VanderPlas Ryan Gorluch Heike Hofmann

---

background-image: url(images/StatisticalAtlasCover.png)
background-position: 50% 50%
background-size: 100%
class: center, fullscale

---
class:center

# Slides

[http://srvanderplas.github.io/Presentations/JSM2018/Framed.html](http://srvanderplas.github.io/Presentations/JSM2018/Framed.html)

# Project Repository

[https://github.com/srvanderplas/Statistical_Atlas](https://github.com/srvanderplas/Statistical_Atlas)

---
layout:true
# Introduction

- Statistical Atlas published for the 9th, 10th, and 11th census    
(1870, 1880, 1890)

- Charts created by hand and reproduced using lithography

---

- Goal: Show the composition of the country
 - Maps
 
.center[<img src="images/DistributionOfWealth.png" width = "50%" alt = "Map of Distribution of Wealth, 1870 Statistical Atlas" style = "margin-top:-30px;"/>]

---

- Goal: Show the composition of the country
 - Abstract Charts:
 - age pyramids
.center[<img src="images/AgePyramid.png" width = "90%" alt = "Age Pyramid, 1870 Statistical Atlas"/>]

---
- Goal: Show the composition of the country
    - Abstract Charts:
        - age pyramids
        - ring charts

.center[<img src="images/RingChart.jpg" width = "90%" alt = "Ring Chart, 1870 Statistical Atlas"/>]

---
- Goal: Show the composition of the country
    - Abstract Charts:
        - age pyramids
        - ring charts
        - framed mosaic and spine plots

---
layout: false
background-image: url(images/Church_Plate31.png)
background-position: 50% 0%
background-size: 100%
class: center, fullscale

---
background-image: url(images/ChurchAccommodations_USIntroLegend.png)
background-position: 50% 50%
background-size: 95%
class: center

???

- Spine plots
- Percent of religious sittings by denomination for each state
- Border: percent of unaccommodated population over the age of 10
- Shows top 4 denominations in each state, top 8 denominations overall (technical constraints)
- Denominations ordered from largest to smallest within states - comparisons are hard

---
background-image: url(images/Occupation_Plate32.png)
background-position: 50% 0%
background-size: 100%
class: center, fullscale

---
background-image: url(images/Occupation_USIntroLegend.png)
background-position: 50% 50%
background-size: 95%
class: center

???

- Mosaic plot - a novel type of plot in 1874
- Georg von Mayr didn't publish mosaic plots until 1877 (Friendly, 2002)
- Proportion of males and females in each occupation shown in the interior of the chart
- Proportion of unaccounted-for persons in the outer band (not separated by gender)

---
# Statistical Archaeology

### Reproduce the plots with modern methods

Data Sources:

- National Historical Geographic Information System 
- Original 1870 Census tables
- Unaccomodated population estimates using 1% microsample data from Integrated Public Use Microdata Series (IPUMS).

- Pixel measurements (by hand) of the area of each chart (for comparison)

---
## Occupation

.center[<img src="images/Occupation_Framed_Redone.png" width = "100%" alt="Re-created framed and unframed mosaic plots"/>]

- Frame cuts the uncounted population into quarters visually
- No segmentation of the uncounted population - 97% of uncounted population is female

---
## Occupation

---
class: bottom
## Church Accommodation

.center[<img src="images/Church_Framed_Orig.png" width = "80%" src="Church Accommodations Plots, for 3 states, selected from Plate 32 of the 1870 Statistical Atlas"/>]

.center[<img src="images/Church_Framed_Redone.png" width = "80%" src="Re-imagined version of Church Accommodations Plots, for 3 states"/>]

---
## Church Accommodation

.center[<img src="images/Church_Framed_Redone.png" width = "80%" src="Re-imagined version of Church Accommodations Plots, for 3 states"/>]

.center[<img src="images/Church_Unframed_Redone.png" width = "80%" src="Re-imagined version of Church Accommodations Plots, for 3 states, without frame"/>]

---
## Church Accommodation

.center[Whoops. The lithographer screwed up.]

---
# Perception of Framed Plots

- Frame cuts the unaccommodated population into quarters - hard to estimate the size

- Difficult to compare between states, as there's no aligned axis due to the frame

- Estimation requires a two-stage process: 
    1. Area of the main plot that is accommodated
    2. Area of the center region devoted to a specific category

- Primary comparisons are on the x-axis; the frame disrupts the ease of these comparisons

---
# Frames in Practice

- 98 Participants (32 from reddit.com/r/samplesize, 63 from Amazon Mechanical Turk)

- Each participant evaluated 18 plots 
    - Region size: small, medium, large
    - Pie, Mosaic, Spine plots
    - Framed or Unframed charts

- 25 seconds per plot

---
# Frames in Practice

.center[<img src="images/Alabama-pie_without_frame1.png" width = "28%" style="margin-right:10px;"/>
<img src="images/Maine-mosaic_without_frame1.png" width = "28%" style="margin-right:10px;"/>
<img src="images/Michigan-spine_without_frame1.png" width = "28%"/>]

.center[<img src="images/Alabama-pie_with_frame1.png" width = "28%" style="margin-right:10px;"/>
<img src="images/Maine-mosaic_with_frame1.png" width = "28%" style="margin-right:10px;"/>
<img src="images/Michigan-spine_with_frame1.png" width = "28%"/>]

---
# Results

---
# Conclusions

- Estimation of the framed portion of plots is significantly worse than the same area in an unframed plot

- Framed plots are not great for accurately communicating the data
 - They don't appear in the 1880 or 1890 census
 - Pie charts largely replaced mosaic and spine plots

- The Census Bureau and assocciated institutions did a great job of preserving the data necessary to reproduce the results and identify flaws in the analysis process

---
# Conclusions

.center[## 150 years later, we can use completely different technology to get the same charts]

How's that for reproducibility?