The goal of orgsurveyr is to facilitate the use of Thomas Lin Pedersen's excellent tidygraph and ggraph packages with organisational information used by many Human Resources departments, in particular employee surveys.
You can install orgsurveyr from github with:
# install.packages("devtools")
devtools::install_github("ukgovdatascience/orgsurveyr", build_vignettes = TRUE)There is also infrastructure within the repo to install and run orgsurveyr within Docker. See the section at the end of this help file for details but essentially you clone the repo and type make run. More help is available with make help.
An organisation is a very basic type of network known as a tree. The tidygraph package lets us simulate and represent a tree structure as follows:
tg <- tidygraph::create_tree(13,3)
tg
#> # A tbl_graph: 13 nodes and 12 edges
#> #
#> # A rooted tree
#> #
#> # Node Data: 13 x 0 (active)
#> #
#> # Edge Data: 12 x 2
#> from to
#> <int> <int>
#> 1 1 2
#> 2 1 3
#> 3 1 4
#> # ... with 9 more rowsThe ggraph package can be used to plot networks with the familiar ggplot2 syntax. Below the simulated organisation is plotted as a dendrogram:
ggraph(tg, 'dendrogram') + geom_edge_diagonal() +
geom_node_point(size=5) + theme_bw()
The orgsurveyr package provides detailed vignettes, convenience functions, shiny apps and example data to help HR analysts make use of the ggraph and tidygraph packages in the analysis of organisation data.
A realistic organisation can be simulated and plotted:
set.seed(1236)
tg2 <- create_realistic_org(4,3, prob=0.3)
plot_org(tg2)
Individual level data also be simulated first by simulating the number of individuals in unit:
tg3 <- simulate_unit_size(tg2)
tg3
#> # A tbl_graph: 38 nodes and 37 edges
#> #
#> # A rooted tree
#> #
#> # Node Data: 38 x 4 (active)
#> unit_id org_depth is_leaf unit_size
#> <chr> <int> <lgl> <dbl>
#> 1 1 0 FALSE 3
#> 2 3 1 FALSE 2
#> 3 4 1 FALSE 3
#> 4 5 1 FALSE 3
#> 5 10 2 FALSE 3
#> 6 11 2 FALSE 2
#> # ... with 32 more rows
#> #
#> # Edge Data: 37 x 2
#> from to
#> <int> <int>
#> 1 1 2
#> 2 1 3
#> 3 1 4
#> # ... with 34 more rowsFollowed by simulating the individuals and any additional variables. The commands can be piped as per tidyverse convention:
tg3_individuals_df <- tg3 %>%
simulate_individuals_df() %>%
mutate(test_var2 = purrr::map_dbl(individual_id, ~rnorm(1, 20,3)))
tg3_individuals_df
#> # A tibble: 146 x 5
#> individual_id individual_name unit_id test_var test_var2
#> <chr> <chr> <chr> <dbl> <dbl>
#> 1 1 1_1 1 9.97 22.8
#> 2 2 1_2 1 16.0 22.0
#> 3 3 1_3 1 11.8 19.5
#> 4 4 3_1 3 13.0 26.3
#> 5 5 3_2 3 9.67 19.1
#> 6 6 4_1 4 12.5 18.5
#> 7 7 4_2 4 11.4 19.6
#> 8 8 4_3 4 12.0 16.8
#> 9 9 5_1 5 14.0 23.4
#> 10 10 5_2 5 13.2 20.4
#> # ... with 136 more rowsThe individual data is aggregated at the unit level. This can be either be just those individuals associated with an organisational unit or (more usefully) a cumulative aggregation involving all individuals that are part of that unit and any units below it in the organisation:
summary_df <- calc_summary_df(tg=tg3, df=tg3_individuals_df,
selected_vars=c('test_var', 'test_var2'), is_cumulative=TRUE)
#> Using wide data frame format for individual variables
summary_df
#> # A tibble: 76 x 3
#> unit_id metric_id value
#> <chr> <chr> <dbl>
#> 1 1 test_var 10.2
#> 2 1 test_var2 20.3
#> 3 10 test_var 8.85
#> 4 10 test_var2 20.7
#> 5 11 test_var 9.84
#> 6 11 test_var2 20.1
#> 7 13 test_var 10.8
#> 8 13 test_var2 20.8
#> 9 14 test_var 9.03
#> 10 14 test_var2 19.4
#> # ... with 66 more rowsThe aggregated data can be plotted on the dendrogram:
plot_org(tg3, fill_var = 'test_var', summary_df)
Larger organisations can be represented as starburst plots:
set.seed(1236)
tg4 <- create_realistic_org(5,5, prob=0.3) %>%
simulate_unit_size()
tg4_individuals_df <- tg4 %>%
simulate_individuals_df() %>%
mutate(test_var2 = purrr::map_dbl(individual_id, ~rnorm(1, 20,3)))
tg4_summary_df <- calc_summary_df(tg=tg4, df=tg4_individuals_df,
selected_vars=c('test_var', 'test_var2'), is_cumulative=TRUE)
#> Using wide data frame format for individual variables
plot_org(tg4, fill_var = 'test_var', df = tg4_summary_df, is_circular = TRUE, is_dendrogram = FALSE) +
scale_fill_gradientn(colours=RColorBrewer::brewer.pal(11, 'PiYG'))
An interactive visualisation in shiny can be launched:
orgviz(tg=tg3, df=summary_df)The git repository contains infrastructure for installing and running orgsurveyr within Docker. This is useful if you quickly and easily want to try out the package before installing it, and is also useful for trying out the shiny apps.
First clone the git repository to your local machine:
git clone https://github.com/ukgovdatascience/orgsurveyr.git
Then from within the repo type make run. This will:
- Build the Docker image
- Start a container
- Open RStudio and Shiny server in your default browser (localhost:8789 and localhost:3839 respective)
To stop and remove the Docker container type make remove. More help is available by typing make help.
- Data Imaginist - Thomas Lin Pedersen's blog
- 'Tidying up your network analysis with tidygraph and ggraph': RStudio::conf presentation from Thomas Lin Pedersen
- useR 2018 Interview with Thomas Lin Pedersen