Exploratory Data Analysis (EDA) is highly visual and can be a motivating entry point into data science and analysis.
burro attempts to make EDA accessible to a larger audience by exposing datasets as a simple Shiny App that can be shared via
shinyapps.io or other Shiny hosts. You can see an example here: https://tladeras.shinyapps.io/nhanes_explore/
burro as an introductory tool for EDA by using it in “data scavenger hunts”, where groups of students are given specific questions to answer about the data, and then have to show their fellow students the answer and how the discovered it. Looking at the data together is vital to building understanding of the data together.
By concentrating on the data visualization first,
burro apps let us have conversations about the data, and hopefully motivate students to learn more tools of EDA such as
burro is currently only on github and not on CRAN yet. To install it, run the following.
burro expects a dataset as a
data.table. The dataset should have at least 2 numeric variables and two categorical variables.
burro requires an outcome variable, which should be categorical/factor. It’s on my list of things to do to make
burro adaptive to the data passed into it, but it currently is pretty inflexible about these two things.
An optional (though helpful) requirement is to have a data dictionary which has a column called
variableNames that defines each variable in the dataset.
Here we make a
burro app using the
explore_data option for the NHANES (National Health and Nutrition Examination Survey) data. We specify our covariates, and our outcome of interest (
Depressed, the number of depressive episodes).
You can see the
burro app for the
NHANES data here: https://tladeras.shinyapps.io/nhanes_explore/
library(burro) #make sure that NHANES package is installed library(NHANES) data(NHANES) data_dict <- readr::read_csv(system.file("nhanes/data_dictionary.csv", package="burro")) ##specify outcome variable here outcome <- c("Depressed") ## specify covariates here (including outcome variable) covars <- c("Gender", "Age", "SurveyYr", "Race1", "Race3" ,"MaritalStatus", "BMI", "HHIncome", "Education", "BMI_WHO", "BPSysAve", "TotChol", "Depressed", "LittleInterest", "SleepHrsNight", "SleepTrouble", "TVHrsDay", "AlcoholDay", "Marijuana", "RegularMarij", "HardDrugs") explore_data(dataset=NHANES, covariates=covars, data_dictionary=data_dict, outcome_var=outcome)
We can examine the
biopics dataset using
burro. We specify our
outcome_var to be
subject_sex, so we can examine everything through the facet of gender.
library(burro) library(fivethirtyeight) data(biopics) explore_data(biopics, outcome_var = "subject_sex")
library(ggplot2) data(diamonds) burro::explore_data(diamonds, outcome_var="cut")
burro was partially developed with funding from Big Data to Knowledge (BD2K) and a National Library of Medicine T15 Training Grant supplement for the development of data science curricula.
burro hex sticker uses clipart designed by Freepik.