Digitizing Qualitative GIS Data with qualmap
This package implements a process for converting qualitative GIS data from an exercise where respondents are asked to identify salient locations on a map. This article focuses primarily on the use of the software to digitize these data.
Motivation and Approach
Qualitative GIS outputs are notoriously difficult to work with because individuals’ conceptions of space can vary greatly from each other and from the realities of physical geography themselves.
qualmap builds on a semi-structured approach to qualitative GIS data collection. Respondents use a specially designed basemap that allows them free reign to identify geographic features of interest and makes it easy to convert their annotations into digital map features. This is facilitated by including on the basemap a series of polygons, such as neighborhood boundaries or census geography, along with an identification number that can be used by
qualmap. A circle drawn on the map can therefore be easily associated with the features that it touches or contains.
qualmap provides a suite of functions for entering, validating, and creating
sf objects based on these hand drawn clusters and their associated identification numbers. Once the clusters have been created, they can be summarized and analyzed either within R or using another tool.
This approach provides an alternative to either unstructured qualitative GIS data, which are difficult to work with empirically, and to digitizing respondents’ annotations as rasters, which require a sophisticated workflow. This semi-structured approach makes integrating qualitative GIS with existing census and administrative data simple and straightforward, which in turn allows these data to be used as measures in spatial statistical models.
More details on the package and how it fits into the broader ecosystem of qualitative GIS are available in a pre-print on SocArXiv. All data associated with the pre-print are also available on Open Science Framework, and the code are available via Open Science Framework and GitHub.
An article describing
qualmap’s approach to qualitative GIS has been published in Cartographica. All data associated with the article are also available on Open Science Framework, and the code are available via Open Science Framework and GitHub. Please cite the paper if you use areal in your work!
The easiest way to get
qualmap is to install it from CRAN:
You can install the development version of
qualmap from Github with the
Note that installations that require
sf to be built from source will require additional software regardless of operating system. You should check the
sf package website for the latest details on installing dependencies for that package. Instructions vary significantly by operating system.
qualmap is built around a number of fundamental principles. The primary data objects created by
qm_combine() are long data rather than wide. This is done to facilitate easy, consistent data management. The package also implements simple features objects using the
sf package. This provides a modern interface for working with spatial data in
qualmap implements six core verbs for working with mental map data:
qm_define() - create a vector of feature id numbers that constitute a single “cluster”
qm_validate() - check feature id numbers against a reference data set to ensure that the values are valid
qm_preview() - plot cluster on an interactive map to ensure the feature ids have been entered correctly (the preview should match the map used as a data collection instrument)
qm_create() - create a single cluster object once the data have been validated and visually inspected
qm_combine() - combine multiple cluster objects together into a single tibble data object
qm_summarize() - summarize the combined data object based on a single qualitative construct to prepare for mapping
The order that these functions are listed here is the approximate order in which they should be utilized. Data should be defined, validated and previewed, and then cluster objects should be created, combined, and summarized.
All of the main functions except
qm_combine() rely on two key arguments:
ref - a reference object. This should be an
sf object that contains a master list of features that appear in your study. This could a
sf object representing all census tracts in a city or county, for example, or a tessellated grid covering the extent of a city.
key - the name of geographic id variable in the
ref object to match input values to. This could be a FIPS code, the
GEOID variable in most census data, or the
OBJECTID of a tessellated grid. Values entered into
Additionally, a number of the initial functions have a third essential argument:
value - the name of the cluster created using
To begin, you will need a simple features object containing the polygons you will be matching respondents’ data to. Census geography polygons can be downloaded via
tigris, and other polygon shapefiles can be read into
R using the
Here is an example of preparing data downloaded via
library(dplyr) # data wrangling
library(sf) # simple features objects
library(tigris) # access census tiger/line data
<- tracts(state = "MO", county = 510)
stLouis <- st_as_sf(stLouis)
stLouis <- mutate(stLouis, TRACTCE = as.numeric(TRACTCE))stLouis
We download the census tract data for St. Louis, which come in
sp format, using the
tracts() function from
tigris. We then use the
st_as_sf() function to convert these data to a simple features object and convert the
TRACTCE variable to numeric format.
If you want to use your own base data instead, you can use the
st_read() function from
sf to bring them into
Once we have a reference data set constructed, we can begin entering the tract numbers that constitute a single circle on the map or “cluster”. We use the
qm_define() function to input these id numbers into a vector:
<- qm_define(118600, 119101, 119300)cluster1
We can then use the
qm_validate() function to check each value in the vector and ensure that these values all match the
key variable in the reference data:
> qm_validate(ref = stLouis, key = TRACTCE, value = cluster1)
qm_validate() returns a
TRUE value, all data are matches. If it returns
FALSE, at least one of the input values does not match any of the
key variable values. In this case, our
key is the
TRACTCE variable in the
sf object we created earlier.
Once the data are validated, we can preview them interactively using
qm_preview(), which will show the features identified in the given vector in red on the map:
qm_preview(ref = stLouis, key = TRACTCE, value = cluster1)
Create Cluster Object
A cluster object is tibble data frame that is “tidy” - each feature in the reference data is a row. Cluster objects also contain metadata about the cluster itself: the respondent’s identification number from the study, a cluster identification number, and a category that describes what the cluster represents. Clusters are created using
> cluster1_obj <- qm_create(ref = stLouis, key = TRACTCE, value = cluster1, rid = 1, cid = 1, category = "positive")
# A tibble: 3 x 5
RID CID CAT TRACTCE COUNT* <int> <int> <chr> <dbl> <dbl>
1 1 1 positive 119300 1.00
2 1 1 positive 118600 1.00
3 1 1 positive 119101 1.00
Combine and Summarize Multiple Clusters
Once several cluster objects have been created, they can be combined using
qm_combine() to produce a tidy tibble formatted data object:
> clusters <- qm_combine(cluster1_obj, cluster2_obj, cluster3_obj)
# A tibble: 9 x 5
RID CID CAT TRACTCE COUNT<int> <int> <chr> <dbl> <dbl>
1 1 1 positive 119300 1.00
2 1 1 positive 118600 1.00
3 1 1 positive 119101 1.00
4 1 2 positive 119300 1.00
5 1 2 positive 121200 1.00
6 1 2 positive 121100 1.00
7 1 3 negative 119300 1.00
8 1 3 negative 118600 1.00
9 1 3 negative 119101 1.00
Since the same census tract appears in multiple rows as part of different clusters, we need to summarize these data before we can map them. Part of
qualmap’s opinionated approach revolves around clusters representing only one construct. When we summarize, therefore, we also subset our data so that they represent only one phenomenon. In the above example, there are both “positive” and “negative” clusters. We can use
qm_summarize() to extract only the “positive” clusters and then summarize them so that we have one row per census tract:
> pos <- qm_summarize(ref = stLouis, key = TRACTCE, clusters = clusters,
+ category = "positive", geometry = TRUE, use.na = FALSE)
106 features and 7 fields
Simple feature collection with : POLYGON
geometry type: XY
dimension: xmin: -90.32052 ymin: 38.53185 xmax: -90.16657 ymax: 38.77443
bboxepsg (SRID): 4269
: +proj=longlat +ellps=GRS80 +towgs84=0,0,0,0,0,0,0 +no_defs
STATEFP COUNTYFP TRACTCE GEOID NAME NAMELSAD positive geometry1 29 510 112100 29510112100 1121 Census Tract 1121 0 POLYGON ((-90.30445 38.6328...
2 29 510 116500 29510116500 1165 Census Tract 1165 0 POLYGON ((-90.24302 38.5975...
3 29 510 110300 29510110300 1103 Census Tract 1103 0 POLYGON ((-90.24032 38.6643...
4 29 510 103700 29510103700 1037 Census Tract 1037 0 POLYGON ((-90.29877 38.6028...
5 29 510 103800 29510103800 1038 Census Tract 1038 0 POLYGON ((-90.32052 38.5941...
6 29 510 104500 29510104500 1045 Census Tract 1045 0 POLYGON ((-90.29432 38.6209...
7 29 510 106100 29510106100 1061 Census Tract 1061 0 POLYGON ((-90.29005 38.6705...
8 29 510 105500 29510105500 1055 Census Tract 1055 0 POLYGON ((-90.28601 38.6589...
9 29 510 105200 29510105200 1052 Census Tract 1052 0 POLYGON ((-90.29481 38.6473...
10 29 510 105300 29510105300 1053 Census Tract 1053 0 POLYGON ((-90.29705 38.6617...
qm_summarize() function has an options to return
NA values instead of
0 values for features not included in any clusters (when
use.na = TRUE), and can return a non-
sf tibble of valid features instead of the
sf object (when
geometry = FALSE).
Mapping Summarized Data
Finally, we can use the
geom_sf() geom from
ggplot2 to map our summarized data, highlighting areas most discussed as being “positive” parts of St. Louis in our hypothetical study:
geom_sf(data = qualData, mapping = aes(fill = positive)) +
qualmap output are
sf objects, they will work with any of the spatial packages that also support
- If you are new to
R itself, welcome! Hadley Wickham’s R for Data Science is an excellent way to get started with data manipulation in the tidyverse, which
areal is designed to integrate seamlessly with.
- If you are new to spatial analysis in
R, we strongly encourage you check out the excellent new Geocomputation in R by Robin Lovelace, Jakub Nowosad, and Jannes Muenchow.
- If you have questions about using
areal, you are encouraged to use the RStudio Community forums. Please create a
reprex before posting. Feel free to tag Chris (
@chris.prener) in any posts about
- If you think you’ve found a bug, please create a
reprex and then open an issue on GitHub.