Takes data that has been summarized into time-bins by make_time_sequence_data(), finds adjacent time
bins that pass some test-statistic threshold, and assigns these adjacent bins into groups (clusters).
Output is ready for a cluster permutation-based analyses (Maris & Oostenveld, 2007). Supports t.test,
wilcox.test, (g)lm, and (g)lmer. Also includes support for
the "bootstrapped-splines" test (see ?make_boot_splines_data and
the divergence vignette for more info).
By default, this function uses 'proportion-looking' (Prop) as the DV, which can be changed
by manually specifying the formula.
make_time_cluster_data(data, ...)
# S3 method for time_sequence_data
make_time_cluster_data(
data,
predictor_column,
aoi = NULL,
test,
threshold = NULL,
formula = NULL,
treatment_level = NULL,
...
)The output of the make_time_sequence_data function
Any other arguments to be passed to the selected 'test' function (e.g., paired, var.equal, etc.)
The column name containing the variable whose test statistic you are interested in.
Which AOI should be analyzed? If not specified (and dataframe has multiple AOIs), then AOI should be a predictor/covariate in your model (so `formula` needs to be specified).
What type of test should be performed in each time bin? Supports
t.test, (g)lm, or (g)lmer. Also includes experimental support for
the "bootstrapped-splines" test (see ?make_boot_splines_data and
the divergence vignette
for more info). Does not support wilcox.test.
Time-bins with test-statistics greater than this amount will be grouped into clusters.
What formula should be used for test? Optional (for all but (g)lmer), if unset
uses Prop ~ [predictor_column]
If your predictor is a factor, regression functions like `lm` and `lmer` by default will treatment-code it. One option is to sum-code this predictor yourself before entering it into this function. Another is to use the `treatment_level` argument, which specifies the level of the predictor. For example, you are testing a model where `Target` is a predictor, which has two levels, 'Animate' and 'Inanimate'. R will code 'Animate' as the reference level, and code 'Inanimate' as the treatment level. You'd therefore want to set `treatment_level = Inanimate`.
The original data, augmented with information about clusters. Calling summary on this data will
describe these clusters. The dataset is ready for the analyze_time_clusters method.
make_time_cluster_data(time_sequence_data): Make data for time cluster analysis
if (FALSE) {
data(word_recognition)
data <- make_eyetrackingr_data(word_recognition,
participant_column = "ParticipantName",
trial_column = "Trial",
time_column = "TimeFromTrialOnset",
trackloss_column = "TrackLoss",
aoi_columns = c('Animate','Inanimate'),
treat_non_aoi_looks_as_missing = TRUE )
response_window <- subset_by_window(data, window_start_time = 15500, window_end_time = 21000,
rezero = FALSE)
# identify clusters in the sequence data using a t-test with
# threshold t-value of 2
# (note: t-tests require a summarized dataset)
response_time <- make_time_sequence_data(response_window, time_bin_size = 500, aois = "Animate",
predictor_columns = "Sex",
summarize_by = "ParticipantName")
time_cluster_data <- make_time_cluster_data(data = response_time,
predictor_column = "Sex",
aoi = "Animate",
test = "t.test",
threshold = 2
)
# identify clusters in the sequence data using an lmer() random-effects
# model with a threshold t-value of 1.5.
# random-effects models don't require us to summarize
response_time <- make_time_sequence_data(response_window, time_bin_size = 500, aois = "Animate",
predictor_columns = "Sex")
# but they do require a formula to be specified
time_cluster_data <- make_time_cluster_data(data = response_time,
predictor_column = "SexM",
aoi = "Animate",
test = "lmer",
threshold = 1.5,
formula = LogitAdjusted ~ Sex + (1|Trial) + (1|ParticipantName)
)
}