Statistical Data Types

stype (pronounced stipe) is an R package for statistical data types. It depends heavily upon the vctrs package to:

  • implement S3 classes aligned with statistical language: binary, continuous, count, non-negative continuous, time to event, right-censored, etc;
  • create variable-level metadata useful for sharing information about a variable across applications;
  • add automatically generated summary statistics to the variable metadata.

Statistical types

The stype package provides classes that enforce (run-time) safety for types common to many statistical analyses such as v_binary, v_continuous, v_count, and v_nominal. For example, binary data can be represented in R in at least three ways: a logical, a factor with two levels, or a numeric using just 0 and 1. Which representation should one use? The latter two do not guarantee that certain binary operations are closed in a mathematical sense; e.g., c(0, 1, 0, 1) + 1:4 returns c(1, 3, 3, 5). Such behavior is not possible with v_binary. Similarly, count data can be represented by an integer in R but without the restriction of being non-negative. The v_count constructor enforces positivity.

Contextual information

Each instance of stype objects contain 2 attributes that users may find useful: context and data_summary. A context can be used to specify project-specific metadata. It is an S4 object containing slots such as short_label, long_label, description, security_type, tags, and purpose. A purpose, for example, can be used to define a variable’s role in a study design such as “outcome”, “identifier”, “covariate”, or “exposure”. This kind of contextual information is invaluable in data pipelines.

Summary statistics

A stype vector also contains a data_summary object, which is automatically generated and contain summary statistics about the data. All objects contain the following statistics:

  • n: number of observations
  • has_missing: an indicator of whether the variable has missing data
  • n_nonmissing: the number of nonmissing
  • n_missing: the number of missing
  • proportion_missing: the proportion missing
  • is_constant: an indicator of whether all the values are the same

Each type has additional summary statistics relevant to its data. For example, v_continuous contains the mean, standard deviation, min, max, and various quantiles. The data_summary is updated whenever a variable is subset or two vectors of the type are combined.

The package also prints certain attributes, for example:

> stype::v_binary(c(TRUE, FALSE, TRUE))
<binary[3]>
[1] 1 0 1
Proportion = 0.667
> stype::v_binary(c(TRUE, FALSE, TRUE, NA))
<binary[4]>
[1]  1  0  1 NA
Proportion = 0.667; Missing = 1.000