add_blow: Bind the weighted log odds to a tidy dataset
In scottfrechette/funcyfrech: Frechette Functions

add_blow

R Documentation

Bind the weighted log odds to a tidy dataset

Description

Calculate and bind the log odds ratio, weighted by a Dirichlet prior, of a tidy dataset to the dataset itself. The weighted log odds ratio is added as a column named zeta, with optional columns log_odds, variance, odds, and prob. This functions supports non-standard evaluation through the tidyeval framework.

Usage

add_blow(
  df,
  group,
  feature,
  n,
  topic = NULL,
  .prior = c("empirical", "uninformative", "tidylo"),
  .compare = c("dataset", "groups"),
  .k_prior = 0.1,
  .alpha_prior = 1,
  .complete = FALSE,
  .log_odds = FALSE,
  .se = FALSE,
  .odds = FALSE,
  .prob = FALSE,
  .sort = FALSE
)

Arguments

`df`	A tidy dataset with one row per feature and set
`group`	Column of groups between which to compare features, such as documents for text data
`feature`	Column of features for identifying differences, such as words or bigrams with text data
`n`	Column containing feature-set counts
`topic`	(Optional) topic to compare groups within
`.prior`	Whether prior should be based on g-prior from empirical Bayes, uninformed with set alpha (uninformed), or total frequency count from tidylo implementation
`.compare`	Whether to compare group-feature to entire dataset or against all other groups
`.k_prior`	Penalty term for informed prior
`.alpha_prior`	Frequency of each feature for uninformed prior
`.complete`	Whether to complete all topic-group-feature combinations
`.log_odds`	Whether to include point estimate log odds
`.se`	Whether to include standard error of estimate
`.odds`	Whether to include odds of seeing feature within group
`.prob`	Whether to include probability for feature within group
`.sort`	Whether to sort by largest zeta

Details

The arguments group, feature, n, and topic are passed by expression and support quasiquotation; you can unquote strings and symbols. Grouping is preserved but ignored.

The dataset must have exactly one row per topic-group-feature combination for this calculation to succeed. Read Monroe, Colaresi, and Quinn (2017) for more on the weighted log odds ratio.