Meet the datarium datasets • datarium

datarium ships 23 small, curated data sets for teaching classical statistical inference in R. The idea is simple: one data set per test. Each data set is built to illustrate exactly one analysis — a t-test, an ANOVA, a chi-square test, a regression — so you can go straight from “which test do I need?” to a worked example, and every one is used in the free Datanovia biostatistics lessons.

The complete map is always one keystroke away in R:

library(datarium)
?datarium

The promise: one data set per test

Every data set names its analysis in its title and shows the canonical fit in its @examples. So the workflow is always the same — load it, and run the test its help page demonstrates:

library(datarium)

# One-sample t-test: does the average mouse weight differ from 25 g?
data("mice")
t.test(mice$weight, mu = 25)
#> 
#>  One Sample t-test
#> 
#> data:  mice$weight
#> t = -8.1045, df = 9, p-value = 1.995e-05
#> alternative hypothesis: true mean is not equal to 25
#> 95 percent confidence interval:
#>  18.78346 21.49654
#> sample estimates:
#> mean of x 
#>     20.14

# Poisson regression: infection counts by treatment and age
data("infections")
glm(count ~ treatment + age, family = poisson, data = infections)
#> 
#> Call:  glm(formula = count ~ treatment + age, family = poisson, data = infections)
#> 
#> Coefficients:
#>      (Intercept)  treatmenttreated               age  
#>          0.66807          -0.83175           0.02395  
#> 
#> Degrees of Freedom: 99 Total (i.e. Null);  97 Residual
#> Null Deviance:       171.3 
#> Residual Deviance: 75.84     AIC: 383.2

Meet the datasets

Comparing means

Data set	In one line	Test
`mice`	Do 10 mice weigh more than 25 g?	one-sample t-test
`mice2`	The same 10 mice, weighed before and after a treatment.	paired-samples t-test
`genderweight`	Do women and men differ in weight?	two-samples (independent) t-test

Analysis of variance

Data set	In one line	Test
`jobsatisfaction`	Job-satisfaction score by gender and education level.	two-way ANOVA
`headache`	Migraine pain score across three treatments, by gender and risk.	three-way ANOVA
`heartattack`	Cholesterol across three drugs, by gender and risk.	three-way ANOVA
`selfesteem`	Self-esteem measured at three time points.	one-way repeated-measures ANOVA
`selfesteem2`	Self-esteem across a control and a diet trial, at three time points.	two-way repeated-measures ANOVA
`weightloss`	Weight loss under diet × exercise, over time.	three-way repeated-measures ANOVA
`anxiety`	Anxiety over time in three exercise groups.	two-way mixed ANOVA
`depression`	A depression treatment followed over four time points.	two-way mixed ANOVA
`performance`	Performance at two time points, by gender and stress.	three-way mixed ANOVA
`stress`	Stress score by treatment and exercise, adjusting for age.	two-way ANCOVA

Categorical data and proportions

Data set	In one line	Test
`properties`	Which type of buyer buys which type of property?	chi-square test of independence
`housetasks.raw`	Who in the couple does which household task?	chi-square test of independence
`titanic.raw`	Who survived the Titanic, by class, sex and age?	categorical descriptive statistics
`antismoking`	Smoking status before and after an anti-smoking video.	McNemar’s test
`taskachievment`	Success or failure on three tasks by the same people.	Cochran’s Q test
`renalstone`	Does renal-stone risk rise with age?	Cochran-Armitage trend test

Regression

Data set	In one line	Test
`marketing`	Sales versus YouTube, Facebook and newspaper budgets.	linear regression
`infections`	Infection counts by treatment and age.	Poisson regression
`heartdisease`	Real four-site cardiology data, with missing values to clean first.	logistic regression

Time series

Data set	In one line	Test
`AirPassengersDf`	Monthly international airline passengers, 1949–1960.	time series

Real vs simulated

datarium is honest about provenance. A handful of data sets are recovered from real, published sources — titanic.raw from base R’s Titanic, AirPassengersDf from base R’s AirPassengers, housetasks.raw from the factoextra housetasks table, renalstone from Hazra and Gogtay (2016), and heartdisease from the UCI Heart Disease repository (Janosi et al., 1989). The rest are simulated for teaching: each such data set’s @source states so plainly, and for the simulated ones the score columns are documented as an arbitrary scale with no real-world units, so nothing is over-claimed.

heartdisease is the deliberate exception to how tidy the simulated sets are: it ships close to raw — four hospitals’ data stacked together, integer-coded categories, and heavy, uneven missing values (one site even records a cholesterol of 0 to mean “not measured”). It is there to teach the step the clean sets skip — inspect and clean the data before you model it.

Where to go next

?datarium — the same one-data-set-per-test map, inside R.
Each data set’s own help page (e.g. ?anxiety) — variables, source, and a runnable example of its analysis.
The Datanovia biostatistics lessons — a worked tutorial for each test, using these data sets.