diff --git a/docs/articles/eyetrackingR-comparison.html b/docs/articles/eyetrackingR-comparison.html index 793a196..b4e6e33 100644 --- a/docs/articles/eyetrackingR-comparison.html +++ b/docs/articles/eyetrackingR-comparison.html @@ -285,15 +285,15 @@

CPA in {eyetrackingR}
 system.time({
   clust_analysis <- analyze_time_clusters(
-df_timeclust,
-within_subj = TRUE,
-paired = TRUE,
-samples = 150,
-quiet = TRUE
+    df_timeclust,
+    within_subj = TRUE,
+    paired = TRUE,
+    samples = 150,
+    quiet = TRUE
   )
 })
 #>    user  system elapsed 
-#>   19.48    0.14   36.24
+#> 13.22 0.10 39.03

This simulates a null distribution of cluster-mass statistics. The output, when printed, is essentially the output of summary(df_timeclust) with p-values (the @@ -332,7 +332,7 @@

CPA in {jlmerclusterperm} jlmerclusterperm_setup() }) #> user system elapsed -#> 0.04 0.01 26.13 +#> 0.04 0.01 32.76

In {jlmerclusterpm}, CPA can also be conducted in two steps:

    @@ -381,7 +381,7 @@

    CPA in {jlmerclusterperm} cpa <- clusterpermute(spec, threshold = threshold_t, nsim = 150) }) #> user system elapsed -#> 0.05 0.00 16.29 +#> 0.03 0.00 16.05

    The same kinds of information are returned:

     cpa
@@ -473,7 +473,7 @@ 
    Performanceclusterpermute(spec, threshold = threshold_t, nsim = 1000)
 })
 #>    user  system elapsed 
-#>    0.00    0.04   12.34
    +#> 0.06 0.05 11.44

    While eyetrackingR does have an option for parallelization, it has limited support and is platform dependent. In contrast, jlmerclusterperm leverages Julia’s built-in @@ -484,14 +484,14 @@

    Performancejlmerclusterperm_setup(restart = TRUE, verbose = FALSE) }) #> user system elapsed -#> 0.06 0.01 22.89 +#> 0.14 0.01 23.20

    10,000 simulations with 7 threads:

     system.time({
   clusterpermute(spec, threshold = threshold_t, nsim = 10000)
 })
 #>    user  system elapsed 
-#>    0.10    0.16   52.57
    +#> 0.29 0.99 54.15 diff --git a/vignettes/articles/eyetrackingR-comparison.Rmd b/vignettes/articles/eyetrackingR-comparison.Rmd index 966c691..6f2569b 100644 --- a/vignettes/articles/eyetrackingR-comparison.Rmd +++ b/vignettes/articles/eyetrackingR-comparison.Rmd @@ -131,47 +131,47 @@ In `{eyetrackingR}`, CPA is conducted in two steps: 1) Prepare data for CPA with `make_time_cluster_data()`: ```{r} -df_timeclust <- make_time_cluster_data(response_time, - test = "t.test", paired = TRUE, - predictor_column = "Target", - threshold = threshold_t -) + df_timeclust <- make_time_cluster_data(response_time, + test = "t.test", paired = TRUE, + predictor_column = "Target", + threshold = threshold_t + ) ``` This step computes the timewise statistics from the data and identifies the empirical clusters, which can be inspected with a `plot()` and `summary()` method: ```{r} -plot(df_timeclust) + plot(df_timeclust) ``` ```{r} -summary(df_timeclust) + summary(df_timeclust) ``` 2) Run the permutation test on the cluster data with `analyze_time_clusters()`: ```{r} -system.time({ - clust_analysis <- analyze_time_clusters( - df_timeclust, - within_subj = TRUE, - paired = TRUE, - samples = 150, - quiet = TRUE - ) -}) + system.time({ + clust_analysis <- analyze_time_clusters( + df_timeclust, + within_subj = TRUE, + paired = TRUE, + samples = 150, + quiet = TRUE + ) + }) ``` This simulates a null distribution of cluster-mass statistics. The output, when printed, is essentially the output of `summary(df_timeclust)` with p-values (the `Probability` column) ```{r} -clust_analysis + clust_analysis ``` The null distribution (and the extremety of the empirical clusters in that context) can be visualized with a `plot()` method: ```{r} -plot(clust_analysis) + plot(clust_analysis) ``` ## CPA in `{jlmerclusterperm}` @@ -194,45 +194,45 @@ In `{jlmerclusterpm}`, CPA can also be conducted in two steps: We first specify the formula, data, and grouping columns of the data. Instead of a paired t-test, we specify a linear model with the formula `Prop ~ Target`. ```{r} -spec <- make_jlmer_spec( - formula = Prop ~ Target, - data = response_time, - subject = "ParticipantName", - trial = "Target", - time = "TimeBin" -) + spec <- make_jlmer_spec( + formula = Prop ~ Target, + data = response_time, + subject = "ParticipantName", + trial = "Target", + time = "TimeBin" + ) ``` The output prepares the data for CPA by subsetting it and applying the contrast coding schemes, among other things: ```{r} -spec + spec ``` 2) Run the permutation test with the specification using `clusterpermute()` ```{r, message = FALSE, warning = FALSE} -system.time({ - cpa <- clusterpermute(spec, threshold = threshold_t, nsim = 150) -}) + system.time({ + cpa <- clusterpermute(spec, threshold = threshold_t, nsim = 150) + }) ``` The same kinds of information are returned: ```{r} -cpa + cpa ``` The different pieces of information are available for further inspection using `tidy()`, which returns the dataframe underlying the summary: ```{r} -null_distribution <- tidy(cpa$null_cluster_dists) -null_distribution + null_distribution <- tidy(cpa$null_cluster_dists) + null_distribution ``` ```{r} -empirical_clusters <- tidy(cpa$empirical_clusters) -empirical_clusters + empirical_clusters <- tidy(cpa$empirical_clusters) + empirical_clusters ``` In contrast to `{eyetrackingR}`, `{jlmerclusterperm}` does not provide a custom `plot()` method. However, the same kinds of plots can be replicated with a few lines of ggplot code: