vwr (version 0.3.0)

ldknn: Run the ldknn algorithm

Description

The ldknn algorithm is used to detect bias in the composition of a lexical decison task, using k-nearest neighbor classification and the Levenshtein distance metric.

Usage

ldknn(stimuli, types, reference, k = 1, method='levenshtein', parallel = FALSE)

Arguments

stimuli

character strings corresponding to the stimuli in the experiment.

types

factor corresponding to the type of each stimulus in the experiment.

reference

a character string giving the reference level. Must be a level of the factor in types

k

a value for the k parameter. Set to 1 by default.

method

• "levenshtein": uses levenshtein.distance to calculate distances
• 'levenshtein.damerau': uses levenshtein.damerau.distance to calculate distances

parallel

with parallel=TRUE, ldknn will run in parallel an multiple cores. The number of parallel processes is specified by detectCores(logical = FALSE).

Value

A list with class ldknn.run.

data

A data frame containing the results of the run. stimulus gives the stimulus values, type gives the types of the stimuli, p gives the probability for a reference.level response for that stimulus.

reference level

The reference level used for the simulation.

Odds

The odds, z value, and p value for a reference level response, resulting from a logistic regression in which the probabilities generated by the ldknn algorithm are used to predict stimulus types.

plot and print methods are available for objects of class ld1nn.run

Details

Combining k nearest neighbor classification with the Levenshtein distance produces an algorithm which can be described as follows. For an experiment containing a number of stimuli, which can be words or nonwords:

1. Compute the Levenshtein distances between the currently presented stimulus and all previously presented stimuli.
2. Identify the previously presented stimuli that are at the k nearest distances from the current stimulus.
3. Compute the probability of a word response for the given stimulus based on the relative frequency of words among the nearest neighbors.

References

Keuleers, E., & Brysbaert, M. (2011). Detecting inherent bias in lexical decision experiments with the LD1NN algorithm. The Mental Lexicon, 6(1), 34–52.

See Also

levenshtein.distance, levenshtein.damerau.distance

Examples

data(english.words)
data(basque.words)
# set up a mock experiment: English stimuli are words, Basque stimuli are nonwords
experiment<-data.frame(stimulus=c(sample(english.words,500),
 sample(basque.words,500)),
 type=factor(rep(c('Word','Nonword'),each=500),levels=c('Word','Nonword')))
# randomize the trials
experiment<-experiment[sample(1:1000,1000),]
# run the ldknn algorithm
results<-ldknn(experiment$stimulus,experiment$type,'Word')
print(results)
plot(results)