Data Cleaning

# Load data
load(file = "./data/Single_Trial_Data.rda")


# For word categorization analysis, exclude missing responses, responses with wrong key, and RT outliers 
single_trial_data_priming <- single_trial_data %>%
  dplyr::filter(
      gng_response_type != "miss" &
      gng_response_type != "wrong_key" &
      word_accuracy     != "miss" &
      word_accuracy     != "wrong_key" &
      (is.na(gng_rt_invalid)  | gng_rt_invalid  == FALSE) &
      (is.na(word_rt_outlier) | word_rt_outlier == FALSE)
  ) # (14130 of 15480 trials left)


# Create numeric word accuracy variable (1 = correct, 0 = incorrect)
single_trial_data_priming <- single_trial_data_priming %>%
  dplyr::mutate(word_accuracy_numeric = ifelse(word_accuracy == "correct", 1, 0))


# Make categorical variables factors
single_trial_data_priming$gng_response_type <- factor(single_trial_data_priming$gng_response_type, levels = c("SH", "FH", "FA", "IR"))
single_trial_data_priming$word_valence      <- factor(single_trial_data_priming$word_valence)
single_trial_data_priming$participant_id    <- factor(single_trial_data_priming$participant_id)

Trials were excluded from all analyses if RT in the go/no-go task was shorter than 100 ms or longer than 700 ms, or if the word categorization RT was more than three median absolute deviations (Leys et al., 2013) above or below a participant’s median RT computed per condition. We further discarded trials in which responses in the go/no-go task were missing or performed with one of the word categorization keys. In the analysis of word categorization RT, only trials with correct word categorization responses were considered. In the analysis of word categorization accuracy, trials with correct or incorrect word categorization were included, but not trials with missing categorization responses or responses performed with the key to be pressed to go stimuli.

# Calculate percentage of excluded trials
excluded_trials <- single_trial_data %>%
  dplyr::group_by(participant_id) %>%
  dplyr::summarize(
    below_100_ms   = sum(!is.na(gng_rt_invalid) & gng_rt_invalid != FALSE & gng_rt < 100) / length(participant_id) * 100,
    above_700_ms   = sum(!is.na(gng_rt_invalid) & gng_rt_invalid != FALSE & gng_rt > 700) / length(participant_id) * 100,
    gng_misses     = sum(gng_response_type == "miss")      / length(participant_id) * 100,
    gng_wrong_key  = sum(gng_response_type == "wrong_key") / length(participant_id) * 100,
    outlier_pos    = sum(!is.na(word_rt_outlier) & word_rt_outlier != FALSE & word_valence == "pos") / length(participant_id) * 100,
    outlier_neg    = sum(!is.na(word_rt_outlier) & word_rt_outlier != FALSE & word_valence == "neg") / length(participant_id) * 100,
    word_misses    = sum(word_accuracy == "miss")          / length(participant_id) * 100,
    word_wrong_key = sum(word_accuracy == "wrong_key")     / length(participant_id) * 100
  ) %>% 
  # Calculate M and SD of the variables
  dplyr::summarise_each(list(mean,sd), -participant_id) %>%
  dplyr::ungroup()


# Create dataframe with rows for M and SD for display
table_excluded_trials <- as.data.frame(rbind(as.numeric(excluded_trials[,c(1:8)]),   # M
                                             as.numeric(excluded_trials[,c(9:16)])), # SD
                                       row.names = c("M", "SD"))


# Display percentage of excluded trials
my_table_template(table_excluded_trials,
  caption = "Excluded Trials (in %)", row_names = TRUE,
  col_names = c("< 100 ms", "> 700 ms", "Misses", "Wrong key", "Outlier pos", "Outlier neg", "Misses", "Wrong key"),
  header_above_config = c(" " = 1, "Go/no-go trials" = 4, "Word categorization trials" = 4)
)
Excluded Trials (in %)
Go/no-go trials
Word categorization trials
< 100 ms > 700 ms Misses Wrong key Outlier pos Outlier neg Misses Wrong key
M 0.41 1.12 1.01 0.18 2.89 3.40 0.13 0.11
SD 0.47 2.17 1.55 0.19 1.12 1.56 0.25 0.23



Descriptive Statistics

Means and CIs

This table corresponds to Table S6 in the supplemental material.

# Calculate descriptive statistics for RT per condition
descriptive_statistics_rt <- summarySEwithinO(
  data          = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
  measurevar    = "word_rt",
  withinvars    = c("gng_response_type", "word_valence"),
  idvar         = "participant_id", 
  conf.interval = .95
) %>%
  # Reorder factor levels for plot
  dplyr::mutate(
    gng_response_type = factor(gng_response_type, levels = c("SH", "FH", "FA", "IR")),
    word_valence      = factor(word_valence, levels = c("pos", "neg"))
  )


# Calculate descriptive statistics for accuracy per condition
descriptive_statistics_accuracy <- summarySEwithinO(
  data          = single_trial_data_priming,
  measurevar    = "word_accuracy_numeric",
  withinvars    = c("gng_response_type", "word_valence"),
  idvar         = "participant_id", 
  conf.interval = .95
) %>%
  # Multiply numeric values by 100 to obtain values in percent
  dplyr::mutate_if(is.numeric, list(~ . * 100)) %>%
  # Reorder factor levels for plot
  dplyr::mutate(
    gng_response_type = factor(gng_response_type, levels = c("SH", "FH", "FA", "IR")),
    word_valence      = factor(word_valence, levels = c("pos", "neg"))
  )


# Make a nice merged table for display
descriptive_statistics <- full_join(descriptive_statistics_rt, descriptive_statistics_accuracy,
  by = c("gng_response_type", "word_valence")
) %>%
  # Format confidence interval column
  dplyr::mutate(
    ci_rt       = paste0("[", round(word_rt - ci.x, digits = 0), 
                         ", ", round(word_rt + ci.x, digits = 0), "]"),
    ci_accuracy = paste0("[", round(word_accuracy_numeric - ci.y, digits = 2), 
                         ", ", round(word_accuracy_numeric + ci.y, digits = 2), "]")
  ) %>%
  # Round RT means to zero decimals
  dplyr::mutate_at("word_rt", round, digits = 0) %>%  
  # Select columns to be displayed
  dplyr::select(c("gng_response_type", "word_valence", "word_rt", "ci_rt", "word_accuracy_numeric", "ci_accuracy"))


# Display descriptive statistics for RT and accuracy (and rearrange order of rows)
my_table_template(descriptive_statistics[c(8, 7, 4, 3, 2, 1, 6, 5), ],
  caption = "Word Categorization Behavioral Performance",
  col_names = c("Preceding response type", "Word valence", "M", "95% CI", "M", "95% CI"),
  header_above_config = c(" " = 2, "RT (ms)" = 2, "Accuracy (%)" = 2),
  footnote_config = c(general = "Confidence intervals are adjusted for within-participant designs as described by Morey (2008).")
)
Word Categorization Behavioral Performance
RT (ms)
Accuracy (%)
Preceding response type Word valence M 95% CI M 95% CI
SH pos 579 [574, 584] 94.33 [93.43, 95.23]
SH neg 603 [598, 607] 93.07 [92.11, 94.03]
FH pos 569 [564, 575] 94.14 [93.02, 95.26]
FH neg 608 [602, 614] 92.93 [91.74, 94.12]
FA pos 743 [724, 763] 72.37 [69.17, 75.57]
FA neg 636 [622, 649] 90.71 [88.63, 92.79]
IR pos 698 [689, 708] 94.69 [93.53, 95.85]
IR neg 689 [681, 698] 96.13 [95.11, 97.15]
Note:
Confidence intervals are adjusted for within-participant designs as described by Morey (2008).


Plot

This figure corresponds to Figure 2 in the manuscript.

# Create plot RT
plot_rt <- ggplot(descriptive_statistics_rt, 
                  aes(x = gng_response_type, y = word_rt, fill = word_valence)) +
  geom_bar(stat = "identity", position = position_dodge(), colour = "black", size = 0.25) +
  geom_errorbar(aes(ymax = word_rt + ci, ymin = word_rt - ci), 
                position = position_dodge(width = 0.9), width = 0.2, size = 0.3) +
  my_figure_theme +
  labs(x = "Preceding Response Type", y = "RT (ms)") +
  coord_cartesian(ylim = c(400, 800)) +
  scale_y_continuous(expand = c(0, 0)) +
  scale_fill_manual(values = my_figure_colors, name = "Word Valence", labels = c("Pos", "Neg")) +
  geom_signif(
    y_position = c(785, 785, 785), xmin = c(0.75, 1.75, 2.75), xmax = c(1.25, 2.25, 3.25),
    annotation = c("**", "***", "***"), tip_length = 0.01, vjust = 0.5, textsize = 4, size = 0.3
  )


# Create plot accuracy
plot_accuracy <- ggplot(descriptive_statistics_accuracy, 
                        aes(x = gng_response_type, y = word_accuracy_numeric, fill = word_valence)) +
  geom_bar(stat = "identity", position = position_dodge(), colour = "black", size = 0.25) +
  geom_errorbar(aes(ymax = word_accuracy_numeric + ci, ymin = word_accuracy_numeric - ci),
                position = position_dodge(width = 0.9), width = 0.2, size = 0.3) +
  my_figure_theme +
  labs(x = "Preceding Response Type", y = "Accuracy (%)") +
  coord_cartesian(ylim = c(50, 100)) +
  scale_y_continuous(expand = c(0, 0)) +
  scale_fill_manual(values = my_figure_colors, name = "Word Valence", labels = c("Pos", "Neg")) +
  geom_signif(
    y_position = c(98.2), xmin = c(2.775), xmax = c(3.225),
    annotation = c("***"), tip_length = 0.01, vjust = 0.5, textsize = 4, size = 0.3
  )


# Create common legend for plots (function from http://www.sthda.com/english/wiki/wiki.php?id_contents=7930#add-a-common-legend-for-multiple-ggplot2-graphs)
get_legend <- function(myggplot) {
  tmp      <- ggplot_gtable(ggplot_build(myggplot))
  leg      <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
  legend   <- tmp$grobs[[leg]]
  return(legend)
}
legend    <- get_legend(plot_rt)


# Remove previous legends from plots
plot_rt       <- plot_rt       + theme(legend.position = "none")
plot_accuracy <- plot_accuracy + theme(legend.position = "none")


# Arrange plots
figure_priming_behav <- ggdraw() +
  draw_plot(plot_rt,       x =  0,   y = .2,  width = .5, height = 0.7) +
  draw_plot(plot_accuracy, x = .5,   y = .2,  width = .5, height = 0.7) +
  draw_plot(legend,        x = 0.25, y = 0.1, width = .5, height = 0)   +
  draw_plot_label(c("A", "B"), c(0, 0.5), c(1, 1), size = 15)


# Save plot
# ggsave("figure_priming_behav.tiff",width = 16, height = 10 , units = "cm", dpi=600, compression = "lzw")


# Display plot
figure_priming_behav
Note. (A) RT and (B) accuracy in the word categorization task are shown as a function of preceding response type and word valence. Error bars represent 95% confidence intervals adjusted for within-participant designs as described by Morey (2008). Asterisks refer to significant differences based on linear mixed-effects model analysis.

Note. (A) RT and (B) accuracy in the word categorization task are shown as a function of preceding response type and word valence. Error bars represent 95% confidence intervals adjusted for within-participant designs as described by Morey (2008). Asterisks refer to significant differences based on linear mixed-effects model analysis.


Fast hit time

To obtain a sufficient number of false alarms in the go/no-go task, an individually calibrated RT limit was used. The individual RT limit was calculated in calibration blocks. Participants completed three calibration blocks that were each followed by two experimental blocks. For the first two experimental blocks, the RT limit was set to 80% of the participant’s mean RT for correct responses in the preceding calibration block. For the subsequent experimental blocks, the RT limit was set to 90% of the mean RT in the preceding calibration block.

# Calculate FH time
FH_time <- single_trial_data_priming %>%
  dplyr::filter(gng_response_type == "FH" | gng_response_type == "SH") %>%
  dplyr::group_by(participant_id) %>%
  dplyr::summarize(
    FH_1 = mean(gng_rt[trial <= 28])                 - (mean(gng_rt[trial <= 28])                 * 20) / 100,
    FH_2 = mean(gng_rt[trial >= 173 & trial <= 200]) - (mean(gng_rt[trial >= 173 & trial <= 200]) * 10) / 100,
    FH_3 = mean(gng_rt[trial >= 345 & trial <= 372]) - (mean(gng_rt[trial >= 345 & trial <= 372]) * 10) / 100
  ) %>%
  dplyr::ungroup()


FH_time_mean <- round((mean(FH_time$FH_1) + mean(FH_time$FH_2) + mean(FH_time$FH_3)) / 3, digits = 0)
FH_time_sd   <- round((sd(FH_time$FH_1)   + sd(FH_time$FH_2)   + sd(FH_time$FH_3))   / 3, digits = 0)

FH time: M = 281 ms, SD = 41 ms


(G)LMM Analyses

To evaluate whether word categorization was faster and more accurate for words that were affectively congruent to the preceding action in the go/no-go task, word categorization RT and accuracy were modeled using a linear mixed-effects model (LMM) and a binomial generalized linear mixed-effects model (GLMM), respectively. In both models, response type of the preceding go/no-go response (slow hit, fast hit, false alarm, inhibited response) and word valence (positive, negative) were specified as fixed factors and participants and word stimuli as random factors.

Fixed effects were coded using sliding difference contrasts, such that the intercept reflects the grand mean across all conditions and differences in means of adjacent factor levels (e.g., false alarm minus fast hit) are tested. The random-effects structure for each model was determined based on the procedure proposed by Bates, Kliegl, et al. (2015). We started with the maximal random-effects structure, including random intercepts for participants and word stimuli, as well as random slopes for all fixed factors and their interactions. If the model with the maximal random-effects structure would not converge, correlations of the random terms were set to zero. We performed a principal components analysis on the random-effects variance–covariance estimates to determine the number of components supported by the data and removed random effects explaining zero variance to prevent overparametrization (Matuschek et al., 2017).

# Define contrasts (sliding difference contrasts)
contrasts(single_trial_data_priming$gng_response_type) <- contr.sdif(4)
contrasts(single_trial_data_priming$word_valence)      <- contr.sdif(2)


# Add contrasts as numerical covariates via model matrix*
model_matrix <- model.matrix(~ gng_response_type * word_valence, single_trial_data_priming)


# Attach the model matrix (8 columns) to the dataframe
single_trial_data_priming[, (ncol(single_trial_data_priming) + 1):(ncol(single_trial_data_priming) + 8)] <- model_matrix


# Assign descriptive names to the contrasts
names(single_trial_data_priming)[(ncol(single_trial_data_priming) - 7):ncol(single_trial_data_priming)] <- c(
  "Grand Mean", "FH_SH", "FA_FH", "IR_FA", "pos_neg", "FH_SH:pos_neg", "FA_FH:pos_neg", "IR_FA:pos_neg"
  )


# *Note: For the random effects, we needed to enter the separate random effect terms in the models to enable
# double-bar notation (||). This allows fitting a model that sets correlations of the random terms to zero.


RT

LMM analysis of word categorization RT was conducted on inverse-transformed RT values to meet the assumption of normally distributed residuals. The appropriate transformation was determined using the Box–Cox procedure (Box & Cox, 1964).

# Determine transformation of the dependent variable (RT) by estimating optimal lambda using Box–Cox procedure
bc <- boxcox(word_rt ~ 1, data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ])

optlambda <- bc$x[which.max(bc$y)]

The optimal lambda was -0.83, suggesting that inverse transformation is most appropriate. 

# Density plot and q-q plot for raw RT values
par(mfrow = c(2, 2)) # arrange plots
plot(density(single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ]$word_rt),
  main = "Raw RT: Density  Plot"
)
qqnorm(single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ]$word_rt,
  main = "Raw RT: Q-Q Plot", pch = 1, frame = FALSE
)


# Density plot and q-q plot for inverse-transformed RT values
plot(density(single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ]$word_rt_inverse),
  main = "Inverse RT: Density  Plot"
)
qqnorm(single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ]$word_rt_inverse,
  main = "Inverse RT: Q-Q Plot", pch = 1, frame = FALSE
)

par(mfrow = c(1, 1)) # reset layout

We additionally performed a parallel analysis on untransformed (i.e., raw) RTs using a GLMM in which RT data were modeled with an identity link function and an inverse Gaussian distribution. This procedure was recommended by Lo and Andrews (2015), as nonlinear data transformations may systematically alter patterns of interaction effects (Balota et al., 2013).


Model specification

LMM
#### 1) Run model with maximal random-effects structure
LMM_rt_max <- lmer(word_rt_inverse ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg | participant_id) +
  (1 + FH_SH + FA_FH + IR_FA | word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = FALSE,
control = lmerControl(optimizer = "bobyqa")
)


# Check model output
summary(LMM_rt_max) # Singular fit


#### 2) Run zero-correlation parameter model by using || syntax to reduce model complexity
LMM_rt_red1 <- lmer(word_rt_inverse ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH + IR_FA || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = FALSE,
control = lmerControl(optimizer = "bobyqa")
)


# Check model output
summary(LMM_rt_red1) # Singular fit


# Check PCA of random-effects variance-covariance estimates
summary(rePCA(LMM_rt_red1)) # For word stimuli there is one term that does not explain variance (< 0.5%)


# Check which random effect explains near zero variance
print(VarCorr(LMM_rt_red1), comp = "Variance") # It is IR - FA for word stimuli

As the model with the maximal random-effects structure did not converge, correlation parameters between random terms were set to zero. One random effect (IR − FA for word stimuli) explaining zero variance was removed to prevent overparametrization.


GLMM
#### 1) Run model with maximal random-effects structure
GLMM_rt_max <- glmer(word_rt ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg | participant_id) +
  (1 + FH_SH + FA_FH + IR_FA | word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
family = inverse.gaussian(link = "identity"),
control = glmerControl(optimizer = "bobyqa")
)


# Check model output
summary(GLMM_rt_max) # Model does not converge


#### 2) Run zero-correlation parameter model by using || syntax to reduce model complexity
GLMM_rt_red1 <- glmer(word_rt ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH + IR_FA || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
family = inverse.gaussian(link = "identity"),
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 2e6))
)


# Check model output
summary(GLMM_rt_red1) # Model does not converge


# Check PCA of random-effects variance-covariance estimates
summary(rePCA(GLMM_rt_red1)) # All terms explain variance


# Check which random effect explains least variance
print(VarCorr(GLMM_rt_red1), comp = "Variance") # It is FA - FH for word stimuli

As the model with the maximal random-effects structure did not converge, correlation parameters between random terms were set to zero. One random effect (FA − FH for word stimuli) explaining the least variance was removed to achieve convergence.


Run final model

LMM

This table corresponds to Table 2 in the manuscript.

#### 3) Run final model without random terms explaining zero variance (i.e., remove IR_FA for word stimuli)
LMM_rt_final <- lmer(word_rt_inverse ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = TRUE, 
control = lmerControl(optimizer = "bobyqa")
)


# Check model output
# summary(LMM_rt_final) # Model does converge, no singular fit


# Re-check PCA of random-effects variance-covariance estimates
# summary(rePCA(LMM_rt_final)) # All terms explain variance


# Display results (fixed effects)
tab_model(LMM_rt_final,
  dv.labels = "RT", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value",  p.val = "satterthwaite"
)


# Display random effects
print("Random effects:")
print(VarCorr(LMM_rt_final), digits = 1, comp = "Std.Dev.")
  RT
Predictors b 95 % CI t value p value
Intercept -1.68 -1.77 – -1.59 -36.38 <0.001
FH - SH -0.02 -0.04 – 0.01 -1.27 0.214
FA - FH 0.23 0.18 – 0.28 9.12 <0.001
IR - FA 0.02 -0.04 – 0.07 0.63 0.536
Pos - Neg -0.01 -0.05 – 0.03 -0.34 0.739
FH - SH x Pos - Neg -0.02 -0.07 – 0.03 -0.90 0.375
FA - FH x Pos - Neg 0.34 0.27 – 0.40 10.14 <0.001
IR - FA x Pos - Neg -0.24 -0.30 – -0.19 -8.31 <0.001
Observations 13092 
[1] "Random effects:"
 Groups           Name          Std.Dev.
 word             FA_FH         0.01    
 word.1           FH_SH         0.03    
 word.2           (Intercept)   0.05    
 participant_id   IR_FA:pos_neg 0.11    
 participant_id.1 FA_FH:pos_neg 0.14    
 participant_id.2 FH_SH:pos_neg 0.10    
 participant_id.3 pos_neg       0.08    
 participant_id.4 IR_FA         0.14    
 participant_id.5 FA_FH         0.13    
 participant_id.6 FH_SH         0.06    
 participant_id.7 (Intercept)   0.25    
 Residual                       0.30

Analysis of word categorization RT revealed a significant main effect of response type. Following false alarms, participants responded slower to words than following fast hits, suggesting that false alarms were associated with PES.
Crucially, the analysis further yielded the predicted interaction between response type and word valence, which was significant for false alarms compared to fast hits and for inhibited responses compared to false alarms but not for fast hits compared to slow hits. These interactions indicate that the effect of word valence on word categorization RT was different for false alarm trials compared to fast hit trials and to inhibited response trials, respectively.


GLMM

This table corresponds to Table S7 in the supplemental material.

#### 3) Run final model without random term with least variance to achieve convergence (i.e., remove FA - FH for word stimuli)
GLMM_rt_final <- glmer(word_rt ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + IR_FA || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
family = inverse.gaussian(link = "identity"),
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 2e6))
)


# Check model output
# summary(GLMM_rt_final) # Model does converge, no singular fit 


# Re-check PCA of random-effects variance-covariance estimates
# summary(rePCA(GLMM_rt_final)) # All terms explain variance


# Display results (fixed effects)
tab_model(GLMM_rt_final,
  dv.labels = "RT", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value"
)


# Display random effects
print("Random effects:")
print(VarCorr(GLMM_rt_final), digits = 1, comp = "Std.Dev.")
  RT
Predictors b 95 % CI t value p value
Intercept 749.11 725.89 – 772.32 63.24 <0.001
FH - SH -8.78 -21.51 – 3.96 -1.35 0.177
FA - FH 121.14 99.99 – 142.29 11.23 <0.001
IR - FA -4.25 -30.71 – 22.21 -0.32 0.753
Pos - Neg 14.79 -3.29 – 32.87 1.60 0.109
FH - SH x Pos - Neg -10.32 -28.72 – 8.09 -1.10 0.272
FA - FH x Pos - Neg 136.80 119.54 – 154.07 15.53 <0.001
IR - FA x Pos - Neg -106.35 -126.62 – -86.07 -10.28 <0.001
Observations 13092 
[1] "Random effects:"
 Groups           Name          Std.Dev.
 word             IR_FA         11.421  
 word.1           FH_SH         11.341  
 word.2           (Intercept)   12.236  
 participant_id   IR_FA:pos_neg 53.087  
 participant_id.1 FA_FH:pos_neg 48.288  
 participant_id.2 FH_SH:pos_neg 29.630  
 participant_id.3 pos_neg       19.006  
 participant_id.4 IR_FA         41.642  
 participant_id.5 FA_FH         41.968  
 participant_id.6 FH_SH         17.535  
 participant_id.7 (Intercept)   36.253  
 Residual                        0.007

The GLMM analysis yielded the same pattern of results concerning the predicted interaction between response type and word valence as the LMM analysis of inverse-transformed RT data.


LMM with valence-to-key mapping

An additional analysis was performed in which the mapping of valence to the horizontally arranged response keys was included as fixed factor in the LMM on word categorization RT. Thereby, we aimed to examine whether word categorization RT differed between the counterbalanced assignment of valence to keys (e.g., in our sample of right-handers, the right response key might have been associated with positive valence and the left with negative valence, leading to overall faster word categorization when positive valence was assigned to the right response key) and whether accounting for variance related to valence assignment to keys might affect the findings of the action-based priming effect.

This table corresponds to Table S8 in the supplemental material.

# Create variable for mapping of valence to keys
single_trial_data_priming$key_mapping <- as.numeric(single_trial_data_priming$participant_id) %% 2 # odd participant IDs: pos is left


# Relabel values 
single_trial_data_priming[single_trial_data_priming$key_mapping == 0,]$key_mapping <- "pos_right"
single_trial_data_priming[single_trial_data_priming$key_mapping == 1,]$key_mapping <- "pos_left"


# Define contrasts (sliding difference contrasts)
single_trial_data_priming$key_mapping            <- factor(single_trial_data_priming$key_mapping)
contrasts(single_trial_data_priming$key_mapping) <- contr.sdif(2)


# Add labels for LMM table
labels_mapping <- c(labels,
  "key_mapping2-1" = "Pos Right - Pos Left",
  "gng_response_type2-1:key_mapping2-1" = "FH - SH x Pos Right - Pos Left", 
  "gng_response_type3-2:key_mapping2-1" = "FA - FH x Pos Right - Pos Left",
  "gng_response_type4-3:key_mapping2-1" = "IR - FA x Pos Right - Pos Left",
  "word_valence2-1:key_mapping2-1" = "Pos - Neg x Pos Right - Pos Left",    
  "gng_response_type2-1:word_valence2-1:key_mapping2-1" = "FH - SH x Pos - Neg x Pos Right - Pos Left", 
  "gng_response_type3-2:word_valence2-1:key_mapping2-1" = "FA - FH x Pos - Neg x Pos Right - Pos Left", 
  "gng_response_type4-3:word_valence2-1:key_mapping2-1" = "IR - FA x Pos - Neg x Pos Right - Pos Left", 
  "gng_response_typeSH:key_mapping2-1" = "SH: Pos Right - Pos Left",
  "gng_response_typeFH:key_mapping2-1" = "FH: Pos Right - Pos Left",
  "gng_response_typeFA:key_mapping2-1" = "FA: Pos Right - Pos Left",
  "gng_response_typeIR:key_mapping2-1" = "IR: Pos Right - Pos Left",
  "gng_response_typeSH:key_mappingpos_left:word_valence2-1" = "SH: Pos Left: Pos - Neg",
  "gng_response_typeFH:key_mappingpos_left:word_valence2-1" = "FH: Pos Left: Pos - Neg",
  "gng_response_typeFA:key_mappingpos_left:word_valence2-1" = "FA: Pos Left: Pos - Neg",
  "gng_response_typeIR:key_mappingpos_left:word_valence2-1" = "IR: Pos Left: Pos - Neg",
  "gng_response_typeSH:key_mappingpos_right:word_valence2-1" = "SH: Pos Right: Pos - Neg",
  "gng_response_typeFH:key_mappingpos_right:word_valence2-1" = "FH: Pos Right: Pos - Neg",
  "gng_response_typeFA:key_mappingpos_right:word_valence2-1" = "FA: Pos Right: Pos - Neg",
  "gng_response_typeIR:key_mappingpos_right:word_valence2-1" = "IR: Pos Right: Pos - Neg"
)


# Run model including assignment of valence to keys as factor
LMM_rt_mapping <- lmer(word_rt_inverse ~ gng_response_type * word_valence * key_mapping +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = TRUE, 
control = lmerControl(optimizer = "bobyqa")
)


# Check model output
# summary(LMM_rt_mapping) # Model does converge, no singular fit


# Re-check PCA of random-effects variance-covariance estimates
# summary(rePCA(LMM_rt_mapping)) # All terms explain variance


# Display results (fixed effects)
tab_model(LMM_rt_mapping,
  dv.labels = "RT", pred.labels = labels_mapping, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value",  p.val = "satterthwaite", wrap.labels = 80
)


# Display random effects
print("Random effects:")
print(VarCorr(LMM_rt_mapping), digits = 1, comp = "Std.Dev.")
  RT
Predictors b 95 % CI t value p value
Intercept -1.68 -1.77 – -1.59 -36.87 <0.001
FH - SH -0.02 -0.04 – 0.01 -1.25 0.220
FA - FH 0.24 0.18 – 0.29 9.07 <0.001
IR - FA 0.02 -0.04 – 0.07 0.60 0.555
Pos - Neg -0.01 -0.05 – 0.03 -0.35 0.731
Pos Right - Pos Left 0.12 -0.06 – 0.30 1.35 0.188
FH - SH x Pos - Neg -0.02 -0.07 – 0.03 -0.89 0.380
FA - FH x Pos - Neg 0.34 0.27 – 0.40 10.23 <0.001
IR - FA x Pos - Neg -0.24 -0.30 – -0.19 -8.40 <0.001
FH - SH x Pos Right - Pos Left -0.01 -0.06 – 0.05 -0.22 0.829
FA - FH x Pos Right - Pos Left 0.00 -0.10 – 0.10 0.00 0.997
IR - FA x Pos Right - Pos Left -0.02 -0.13 – 0.09 -0.37 0.716
Pos - Neg x Pos Right - Pos Left 0.03 -0.03 – 0.09 1.10 0.279
FH - SH x Pos - Neg x Pos Right - Pos Left 0.01 -0.08 – 0.10 0.20 0.844
FA - FH x Pos - Neg x Pos Right - Pos Left 0.11 -0.02 – 0.24 1.73 0.092
IR - FA x Pos - Neg x Pos Right - Pos Left -0.12 -0.23 – -0.01 -2.06 0.047
Observations 13092 
[1] "Random effects:"
 Groups           Name          Std.Dev.
 word             FA_FH         0.01    
 word.1           FH_SH         0.03    
 word.2           (Intercept)   0.05    
 participant_id   IR_FA:pos_neg 0.11    
 participant_id.1 FA_FH:pos_neg 0.14    
 participant_id.2 FH_SH:pos_neg 0.10    
 participant_id.3 pos_neg       0.08    
 participant_id.4 IR_FA         0.14    
 participant_id.5 FA_FH         0.13    
 participant_id.6 FH_SH         0.06    
 participant_id.7 (Intercept)   0.25    
 Residual                       0.30

RT did not differ significantly between the counterbalanced assignment of valence to keys. Furthermore, this analysis yielded the same results concerning the priming effects following slow hits, fast hits, and false alarms (see below for following up the interactions).


Follow up interactions

LMM

We followed up interactions by reparametrizing the fixed-effects part of the model with the effect of word valence nested within each response type. Except for the nesting, the model was specified identically to the non-nested model, therefore yielding identical results in terms of response type main effects and model fit indices.

This table corresponds to Table 2 in the manuscript.

#### 4) Run model with the effect of word valence nested within each response type
LMM_rt_nested <- lmer(word_rt_inverse ~ gng_response_type / word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = TRUE,
control = lmerControl(optimizer = "bobyqa")
)


# Display results (fixed effects)
tab_model(LMM_rt_nested,
  dv.labels = "RT", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value",  p.val = "satterthwaite"
)
  RT
Predictors b 95 % CI t value p value
Intercept -1.68 -1.77 – -1.59 -36.38 <0.001
FH - SH -0.02 -0.04 – 0.01 -1.27 0.214
FA - FH 0.23 0.18 – 0.28 9.12 <0.001
IR - FA 0.02 -0.04 – 0.07 0.63 0.536
SH: Pos - Neg -0.10 -0.16 – -0.04 -3.35 0.001
FH: Pos - Neg -0.12 -0.17 – -0.07 -4.52 <0.001
FA: Pos - Neg 0.22 0.16 – 0.28 7.15 <0.001
IR: Pos - Neg -0.03 -0.09 – 0.03 -0.87 0.388
Observations 13092


Following false alarms, participants categorized negative words faster than positive words. Following hits (fast hits and slow hits), participants categorized positive words faster than negative words.


GLMM

We followed up interactions by reparametrizing the fixed-effects part of the model with the effect of word valence nested within each response type. Except for the nesting, the model was specified identically to the non-nested model, therefore yielding comparable results in terms of response type main effects and model fit indices.

This table corresponds to Table S7 in the supplemental material.

#### 4) Run model with the effect of word valence nested within each response type
GLMM_rt_nested <- glmer(word_rt ~ gng_response_type / word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + IR_FA || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
family = inverse.gaussian(link = "identity"),
control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun = 2e6))
)


# Display results (fixed effects)
tab_model(GLMM_rt_nested,
  dv.labels = "RT", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value"
)
  RT
Predictors b 95 % CI t value p value
Intercept 749.10 728.64 – 769.57 71.74 <0.001
FH - SH -8.78 -21.65 – 4.09 -1.34 0.181
FA - FH 121.14 93.26 – 149.02 8.52 <0.001
IR - FA -4.25 -23.55 – 15.04 -0.43 0.666
SH: Pos - Neg -19.29 -40.01 – 1.43 -1.82 0.068
FH: Pos - Neg -29.60 -46.41 – -12.79 -3.45 0.001
FA: Pos - Neg 107.20 92.29 – 122.10 14.10 <0.001
IR: Pos - Neg 0.85 -13.64 – 15.34 0.12 0.908
Observations 13092


Following false alarms, participants categorized negative words faster than positive words. Following hits (fast hits; slow hits as statistical trend), participants categorized positive words faster than negative words.


LMM with valence-to-key mapping

We followed up interactions by reparametrizing the fixed-effects part of the model with the effect of word valence nested within each response type and valence-to-key mapping. Except for the nesting, the model was specified identically to the non-nested model, therefore yielding identical results in terms of response type main effects and model fit indices.

This table corresponds to Table S8 in the supplemental material.

# Run model including mapping of valence to keys as factor with the effect of word valence nested within each response type and key mapping
LMM_rt_mapping_nested <- lmer(word_rt_inverse ~ gng_response_type / key_mapping / word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH || word),
data = single_trial_data_priming[single_trial_data_priming$word_accuracy == "correct", ],
REML = TRUE,
control = lmerControl(optimizer = "bobyqa")
)


# Display results (fixed effects)
tab_model(LMM_rt_mapping_nested,
  dv.labels = "RT", pred.labels = labels_mapping, show.stat = TRUE, show.icc = FALSE, show.r2 = FALSE, 
  show.re.var = FALSE, show.ngroups = FALSE, string.est = "b", string.stat = "t value", 
  string.ci = "95 % CI", string.p = "p value",  p.val = "satterthwaite", wrap.labels = 80
)
  RT
Predictors b 95 % CI t value p value
Intercept -1.68 -1.77 – -1.59 -36.87 <0.001
FH - SH -0.02 -0.04 – 0.01 -1.25 0.220
FA - FH 0.24 0.18 – 0.29 9.07 <0.001
IR - FA 0.02 -0.04 – 0.07 0.60 0.555
SH: Pos Right - Pos Left 0.13 -0.06 – 0.32 1.37 0.180
FH: Pos Right - Pos Left 0.12 -0.06 – 0.31 1.32 0.195
FA: Pos Right - Pos Left 0.12 -0.06 – 0.31 1.31 0.200
IR: Pos Right - Pos Left 0.10 -0.09 – 0.30 1.03 0.307
SH: Pos Left: Pos - Neg -0.10 -0.17 – -0.02 -2.53 0.014
FH: Pos Left: Pos - Neg -0.12 -0.19 – -0.06 -3.59 0.001
FA: Pos Left: Pos - Neg 0.15 0.07 – 0.23 3.79 <0.001
IR: Pos Left: Pos - Neg -0.03 -0.11 – 0.05 -0.70 0.489
SH: Pos Right: Pos - Neg -0.10 -0.17 – -0.02 -2.53 0.013
FH: Pos Right: Pos - Neg -0.12 -0.18 – -0.05 -3.31 0.001
FA: Pos Right: Pos - Neg 0.28 0.20 – 0.36 6.88 <0.001
IR: Pos Right: Pos - Neg -0.02 -0.10 – 0.05 -0.61 0.542
Observations 13092


The priming effects were present for both valence-to-key mappings.


Accuracy

The GLMM models the probability of a correct word categorization and estimates reflect log-odds ratios for a correct response.


Model specification

#### 1) Run model with maximal random-effects structure
GLMM_acc_max <- glmer(word_accuracy_numeric ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg | participant_id) +
  (1 + FH_SH + FA_FH + IR_FA | word),
data = single_trial_data_priming,
family = binomial,
control = glmerControl(optimizer = "bobyqa")
)


# Check model output
summary(GLMM_acc_max) # Model does not converge


#### 2) Run zero-correlation parameter model by using || syntax to reduce model complexity
GLMM_acc_red1 <- glmer(word_accuracy_numeric ~ gng_response_type * word_valence +
  (1 + FH_SH + FA_FH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH + IR_FA || word),
data = single_trial_data_priming,
family = binomial,
control = glmerControl(optimizer = "bobyqa")
)


# Check model output
summary(GLMM_acc_red1) # Singular fit


# Check PCA of random-effects variance-covariance estimates
summary(rePCA(GLMM_acc_red1)) # For participants there is one term that does not explain variance (< 0.5%)


# Check which random effect explains near zero variance
print(VarCorr(GLMM_acc_red1), comp = "Variance") # It is FA - FH for participants

As the model with the maximal random-effects structure did not converge, correlation parameters between random terms were set to zero. One random effect (FA − FH for participants) explaining zero variance was removed to prevent overparametrization.


Run final model

This table corresponds to Table 2 in the manuscript.

#### 3) Run final model without random terms explaining zero variance (i.e., remove FA_FH for participants)
GLMM_acc_final <- glmer(word_accuracy_numeric ~ gng_response_type * word_valence +
  (1 + FH_SH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH + IR_FA || word),
data = single_trial_data_priming,
family = binomial,
control = glmerControl(optimizer = "bobyqa")
)


# Check model output
# summary(GLMM_acc_final) # Model does converge, no singular fit 


# Re-check PCA of random-effects variance-covariance estimates
# summary(rePCA(GLMM_acc_final)) # All terms explain variance


# Display results (fixed effects)
tab_model(GLMM_acc_final,
  dv.labels = "Accuracy", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, 
  show.r2 = FALSE, show.re.var = FALSE, show.ngroups = FALSE, string.stat = "z value", 
  string.ci = "95 % CI", string.p = "p value"
)


# Display random effects
print("Random effects:")
print(VarCorr(GLMM_acc_final), digits = 2, comp = "Std.Dev.")
  Accuracy
Predictors Odds Ratios 95 % CI z value p value
Intercept 21.31 14.99 – 30.30 17.04 <0.001
FH - SH 0.95 0.78 – 1.16 -0.49 0.623
FA - FH 0.39 0.31 – 0.49 -8.05 <0.001
IR - FA 4.00 2.94 – 5.44 8.82 <0.001
Pos - Neg 0.57 0.38 – 0.86 -2.68 0.007
FH - SH x Pos - Neg 1.05 0.69 – 1.59 0.21 0.834
FA - FH x Pos - Neg 0.12 0.06 – 0.23 -6.35 <0.001
IR - FA x Pos - Neg 4.59 2.33 – 9.05 4.40 <0.001
Observations 14130 
[1] "Random effects:"
 Groups           Name          Std.Dev.
 word             IR_FA         0.14    
 word.1           FA_FH         0.13    
 word.2           FH_SH         0.17    
 word.3           (Intercept)   0.53    
 participant_id   IR_FA:pos_neg 1.08    
 participant_id.1 pos_neg:FA_FH 1.20    
 participant_id.2 FH_SH:pos_neg 0.47    
 participant_id.3 pos_neg       0.68    
 participant_id.4 IR_FA         0.38    
 participant_id.5 FH_SH         0.15    
 participant_id.6 (Intercept)   0.86

The analysis yielded an interaction between response type and word valence, which was significant for false alarms compared to fast hits and for inhibited responses compared to false alarms. These interactions indicate that the effect of word valence on categorization accuracy was different for false alarm trials compared to fast hit trials and to inhibited response trials, respectively.


Follow up interactions

We followed up interactions by reparametrizing the fixed-effects part of the model with the effect of word valence nested within each response type. Except for the nesting, the model was specified identically to the non-nested model, therefore yielding identical results in terms of response type main effects and model fit indices.

This table corresponds to Table 2 in the manuscript.

#### 4) Run model with the effect of word valence nested within each response type
GLMM_acc_nested <- glmer(word_accuracy_numeric ~ gng_response_type / word_valence +
  (1 + FH_SH + IR_FA + pos_neg + FH_SH:pos_neg + FA_FH:pos_neg + IR_FA:pos_neg || participant_id) +
  (1 + FH_SH + FA_FH + IR_FA || word),
data = single_trial_data_priming,
family = binomial,
control = glmerControl(optimizer = "bobyqa")
)


# Display results (fixed effects)
tab_model(GLMM_acc_nested,
  dv.labels = "Accuracy", pred.labels = labels, show.stat = TRUE, show.icc = FALSE, 
  show.r2 = FALSE, show.re.var = FALSE, show.ngroups = FALSE, string.stat = "z value", 
  string.ci = "95 % CI", string.p = "p value"
)
  Accuracy
Predictors Odds Ratios 95 % CI z value p value
Intercept 21.31 14.99 – 30.30 17.04 <0.001
FH - SH 0.95 0.78 – 1.16 -0.49 0.623
FA - FH 0.39 0.31 – 0.49 -8.05 <0.001
IR - FA 4.00 2.94 – 5.44 8.82 <0.001
SH: Pos - Neg 1.09 0.64 – 1.85 0.32 0.750
FH: Pos - Neg 1.14 0.67 – 1.94 0.48 0.630
FA: Pos - Neg 0.14 0.08 – 0.25 -6.62 <0.001
IR: Pos - Neg 0.63 0.33 – 1.22 -1.36 0.173
Observations 14130


Following false alarms, participants categorized negative words more accurately than positive words.


ANOVAs

To facilitate comparison with previously reported results obtained using a similar task and aggregation-based analyses (Aarts et al., 2012, 2013), word categorization RT and accuracy were additionally analyzed with Greenhouse–Geisser corrected repeated-measures analyses of variance (ANOVAs) including the within-participant factors response type (slow hit, fast hit, false alarm, inhibited response) and word valence (positive, negative). As can be seen below, the ANOVAs yielded the same results as obtained with mixed-effects modeling with respect to all main effects, interaction effects, and all pairwise comparisons. This analysis corresponds to page 10 in the supplemental material.

# Due to the afex package, contrasts are automatically set to effect-coding (contr.sum). Afex package 
# also checks sphericity assumptions and automatically corrects for any violations if necessary.


# Get data in correct format for ANOVAs (aggregate within participants per condition)
data_anova <- single_trial_data_priming %>%
  dplyr::group_by(participant_id, gng_response_type, word_valence) %>%
  dplyr::summarize(
    word_rt_inverse              = mean(word_rt_inverse[word_accuracy == "correct"]),
    percentage_correct_responses = sum(word_accuracy_numeric) / length(participant_id) * 100
  ) %>%
  dplyr::ungroup()


# ANOVA RT
anova_rt <- aov_ez(
  id     = "participant_id", 
  dv     = "word_rt_inverse", 
  data   = data_anova,
  within = c("gng_response_type", "word_valence")
)


# ANOVA accuracy
anova_accuracy <- aov_ez(
  id     = "participant_id", 
  dv     = "percentage_correct_responses", 
  data   = data_anova,
  within = c("gng_response_type", "word_valence")
)


# Display ANOVA results
knitr::kables(
  list(
    my_table_template(nice(anova_rt, MSE = FALSE, sig_symbols = rep("", 4)), 
                      header_above_config = c(" " = 1, "RT" = 4)),
    my_table_template(nice(anova_accuracy, MSE = FALSE, sig_symbols = rep("", 4))[, c(2:5)], 
                      header_above_config = c("Accuracy" = 4))
  ),
  caption = "ANOVAs"
) %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), font_size = 12, full_width = F, position = "left") %>%
  footnote(general = "Ges = generalized eta squared, the recommended effect size for repeated-measures designs (Bakeman, 2005).")
ANOVAs
RT
Effect df F ges p.value
gng_response_type 1.80, 52.33 64.06 .17 <.0001
word_valence 1, 29 0.03 <.0001 .87
gng_response_type:word_valence 2.21, 64.13 40.73 .06 <.0001
Accuracy
df F ges p.value
1.42, 41.09 29.36 .16 <.0001
1, 29 11.46 .05 .002
1.19, 34.41 20.15 .15 <.0001
Note:
Ges = generalized eta squared, the recommended effect size for repeated-measures designs (Bakeman, 2005).


Pairwise comparisons

Significant main effects and interactions were followed up with two-tailed paired t tests with Holm–Bonferroni p value adjustments.

# Use multivariate model for all follow-up tests to adequately control for violations of sphericity
afex_options(emmeans_model = "multivariate")


# Pairwise t tests
pairwise_rt_m  <- summary(pairs(emmeans(anova_rt, ~gng_response_type), adjust = "holm"))
pairwise_rt_i  <- summary(pairs(emmeans(anova_rt, "word_valence", by = "gng_response_type"), adjust = "holm"))
pairwise_acc_m <- summary(pairs(emmeans(anova_accuracy, ~gng_response_type), adjust = "holm"))
pairwise_acc_i <- summary(pairs(emmeans(anova_accuracy, "word_valence", by = "gng_response_type"), adjust = "holm"))


# Add Cohen's dz (CIs for d could be added if needed, as it can be returned by the "t_to_d" function)
pairwise_rt_m$cohens_dz  <- round(t_to_d(pairwise_rt_m$t.ratio,  pairwise_rt_m$df,  paired = TRUE)[1], digits = 2)
pairwise_rt_i$cohens_dz  <- round(t_to_d(pairwise_rt_i$t.ratio,  pairwise_rt_i$df,  paired = TRUE)[1], digits = 2)
pairwise_acc_m$cohens_dz <- round(t_to_d(pairwise_acc_m$t.ratio, pairwise_acc_m$df, paired = TRUE)[1], digits = 2)
pairwise_acc_i$cohens_dz <- round(t_to_d(pairwise_acc_i$t.ratio, pairwise_acc_i$df, paired = TRUE)[1], digits = 2)


# Display results main effect response type
knitr::kables(
  list(
    my_table_template(pairwise_rt_m, digits = c(0, 2, 2, 0, 2, 3, 2), 
                      header_above_config = c(" " = 1, "RT" = 6)),
    my_table_template(pairwise_acc_m[, c(2:7)], digits = c(2, 2, 0, 2, 3, 2), 
                      header_above_config = c("Accuracy" = 6))
  ),
  caption = "Main Effect Response Type"
) %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), font_size = 12, full_width = F, position = "left") %>%
  footnote(general = "P values are adjusted with Holm–Bonferroni method.")


# Display results interaction response type x word valence
knitr::kables(
  list(
    my_table_template(pairwise_rt_i, digits = c(0, 0, 2, 2, 0, 2, 3, 2), 
                      header_above_config = c(" " = 2, "RT" = 6)),
    my_table_template(pairwise_acc_i[, c(3:8)], digits = c(2, 2, 0, 2, 3, 2), 
                      header_above_config = c("Accuracy" = 6))
  ),
  caption = "Interaction Response Type x Word Valence"
) %>%
  kable_styling(bootstrap_options = c("striped", "condensed"), font_size = 12, full_width = F, position = "left") %>%
  footnote(general = "P values are adjusted with Holm–Bonferroni method.")
Main Effect Response Type
RT
contrast estimate SE df t.ratio p.value cohens_dz
SH - FH 0.02 0.01 29 1.25 0.443 0.23
SH - FA -0.24 0.03 29 -7.07 0.000 -1.31
SH - IR -0.24 0.02 29 -11.18 0.000 -2.08
FH - FA -0.25 0.03 29 -8.79 0.000 -1.63
FH - IR -0.25 0.02 29 -15.40 0.000 -2.86
FA - IR 0.00 0.03 29 -0.06 0.953 -0.01
Accuracy
estimate SE df t.ratio p.value cohens_dz
0.38 0.64 29 0.59 0.562 0.11
9.71 1.58 29 6.16 0.000 1.14
-1.75 0.75 29 -2.35 0.051 -0.44
9.34 1.69 29 5.52 0.000 1.02
-2.13 0.83 29 -2.56 0.048 -0.48
-11.47 1.98 29 -5.80 0.000 -1.08
Note:
P values are adjusted with Holm–Bonferroni method.
Interaction Response Type x Word Valence
RT
contrast gng_response_type estimate SE df t.ratio p.value cohens_dz
neg - pos SH 0.10 0.03 29 3.91 0.001 0.73
neg - pos FH 0.12 0.02 29 5.28 0.000 0.98
neg - pos FA -0.23 0.04 29 -5.73 0.000 -1.06
neg - pos IR 0.03 0.02 29 1.69 0.103 0.31
Accuracy
estimate SE df t.ratio p.value cohens_dz
-1.39 1.16 29 -1.20 0.240 -0.22
-1.04 1.07 29 -0.97 0.339 -0.18
19.82 4.43 29 4.47 0.000 0.83
1.33 0.96 29 1.38 0.177 0.26
Note:
P values are adjusted with Holm–Bonferroni method.



References

Aarts, K., De Houwer, J., & Pourtois, G. (2012). Evidence for the automatic evaluation of self-generated actions. Cognition, 124(2), 117-127. https://doi.org/10.1016/j.cognition.2012.05.009

Aarts, K., De Houwer, J., & Pourtois, G. (2013). Erroneous and correct actions have a different affective valence: Evidence from ERPs. Emotion, 13(5), 960-973. https://doi.org/10.1037/a0032808s

Bakeman, R (2005). Recommended effect size statistics for repeated measures designs. Behavior Research Methods, 37(3), 379–384. https://doi.org/10.3758/BF03192707

Balota, D. A., Aschenbrenner, A. J., & Yap, M. J. (2013). Additive effects of word frequency and stimulus quality: The influence of trial history and data transformations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39(5), 1563-1571. https://doi.org/10.1037/a0032186

Bates, D., Kliegl, R., Vasishth, S., & Baayen, H. (2015). Parsimonious mixed models. arXiv. https://arxiv.org/abs/1506.04967v2

Box, G. E., & Cox, D. R. (1964). An analysis of transformations. Journal of the Royal Statistical Society: Series B (Methodological), 26(2), 211-243. https://doi.org/10.1111/j.2517-6161.1964.tb00553.x

Leys, C., Ley, C., Klein, O., Bernard, P., & Licata, L. (2013). Detecting outliers: Do not use standard deviation around the mean, use absolute deviation around the median. Journal of Experimental Social Psychology, 49(4), 764-766. https://doi.org/10.1016/j.jesp.2013.03.013

Lo, S., & Andrews, S. (2015). To transform or not to transform: Using generalized linear mixed models to analyse reaction time data. Frontiers in Psychology, 6, Article 1171. https://doi.org/10.3389/fpsyg.2015.01171

Matuschek, H., Kliegl, R., Vasishth, S., Baayen, H., & Bates, D. (2017). Balancing Type I error and power in linear mixed models. Journal of Memory and Language, 94, 305-315. https://doi.org/10.1016/j.jml.2017.01.001

Morey, R. (2008). Confidence intervals from normalized data: A correction to Cousineau (2005). Tutorials in Quantitative Methods for Psychology, 4(2), 61-64. https://doi.org/10.20982/tqmp.04.2.p061


Session Info

sessionInfo()
R version 3.6.1 (2019-07-05)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19042)

Matrix products: default

locale:
[1] LC_COLLATE=German_Germany.1252  LC_CTYPE=German_Germany.1252   
[3] LC_MONETARY=German_Germany.1252 LC_NUMERIC=C                   
[5] LC_TIME=German_Germany.1252    

attached base packages:
[1] stats     graphics  grDevices datasets  utils     methods   base     

other attached packages:
 [1] plyr_1.8.6        effectsize_0.3.2  afex_0.25-1       ez_4.4-0         
 [5] emmeans_1.4.2     sjPlot_2.8.4      MASS_7.3-51.4     lmerTest_3.1-0   
 [9] lme4_1.1-21       Matrix_1.2-17     cowplot_1.0.0     ggsignif_0.6.0   
[13] papaja_0.1.0.9842 ggplot2_3.3.2     kableExtra_1.1.0  knitr_1.29       
[17] dplyr_0.8.3      

loaded via a namespace (and not attached):
 [1] nlme_3.1-140        insight_0.9.0       webshot_0.5.1      
 [4] httr_1.4.1          numDeriv_2016.8-1.1 tools_3.6.1        
 [7] backports_1.1.5     R6_2.4.1            sjlabelled_1.1.3   
[10] mgcv_1.8-28         colorspace_1.4-1    withr_2.1.2        
[13] tidyselect_0.2.5    curl_4.2            compiler_3.6.1     
[16] performance_0.4.3   rvest_0.3.6         xml2_1.2.2         
[19] labeling_0.3        bayestestR_0.7.2    scales_1.1.0       
[22] mvtnorm_1.0-11      readr_1.3.1         stringr_1.4.0      
[25] digest_0.6.22       foreign_0.8-71      minqa_1.2.4        
[28] rmarkdown_1.16      rio_0.5.16          pkgconfig_2.0.3    
[31] htmltools_0.4.0     highr_0.8           rlang_0.4.8        
[34] readxl_1.3.1        rstudioapi_0.10     farver_2.0.1       
[37] generics_0.0.2      zip_2.0.4           car_3.0-5          
[40] magrittr_1.5        parameters_0.8.2    Rcpp_1.0.5         
[43] munsell_0.5.0       abind_1.4-5         lifecycle_0.2.0    
[46] stringi_1.4.3       yaml_2.2.0          carData_3.0-3      
[49] grid_3.6.1          parallel_3.6.1      sjmisc_2.8.2       
[52] forcats_0.4.0       crayon_1.3.4        lattice_0.20-38    
[55] ggeffects_0.15.1    haven_2.2.0         splines_3.6.1      
[58] sjstats_0.17.8      hms_0.5.2           pillar_1.4.6       
[61] boot_1.3-22         estimability_1.3    codetools_0.2-16   
[64] reshape2_1.4.3      glue_1.3.1          evaluate_0.14      
[67] data.table_1.12.6   renv_0.12.0         modelr_0.1.5       
[70] vctrs_0.3.4         nloptr_1.2.1        cellranger_1.1.0   
[73] gtable_0.3.0        purrr_0.3.3         tidyr_1.0.0        
[76] assertthat_0.2.1    xfun_0.16           openxlsx_4.1.3     
[79] xtable_1.8-4        broom_0.5.2         viridisLite_0.3.0  
[82] tibble_3.0.4        ellipsis_0.3.0