todo:
- new exclusion criteria
- ideal learner stuff
- look at target reaction time?
- import the complexity norm?
Visualizing raw trial looking time
d_no_target <- d %>%
filter(trial_stimulus_type != "target")Looking time distribution
RAW
data_with_demog %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram(bins = 90) +
xlim(0, 9000)## Warning: Removed 72 rows containing non-finite values (stat_bin).## Warning: Removed 2 rows containing missing values (geom_bar).
data_with_demog %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram() +
scale_x_log10()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
AFTER EXCLUSION
d %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram() +
scale_x_log10()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
#
# d %>%
# ggplot(aes(x = trial_looking_time)) +
# geom_histogram(bins = 90) +
# xlim(0, 6000) +
# facet_wrap(~subject)Digging through the second hump
- Something wacky is happening in background trials.
- Deviant trials are slower
- target trials are likely a mixture between hits and misses
d %>%
ggplot(aes(x = trial_looking_time,
fill = trial_pressed_space_bar)) +
geom_histogram() +
scale_x_log10() +
facet_wrap(~trial_stimulus_type)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
Zoom in on background trials. The “hump” (second mode) is there in all parts of the experiment.
d %>%
mutate(block_num = ceiling(total_trial_num / 48)) %>%
filter(trial_stimulus_type == "background") %>%
ggplot(aes(x = trial_looking_time,
fill = trial_stimulus_complexity)) +
geom_histogram() +
scale_x_log10() +
facet_wrap(~block_num)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
d %>%
filter(trial_stimulus_type == "background") %>%
group_by(subject) %>%
mutate(mlog = mean(log(trial_looking_time)),
sdlog = sd(log(trial_looking_time)),
outlier = log(trial_looking_time) > mlog + 3*sdlog |
log(trial_looking_time) < mlog - 3*sdlog) %>%
ggplot(aes(x = trial_looking_time, fill = outlier)) +
geom_histogram() +
scale_x_log10() +
facet_wrap(~subject)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
We see this pattern within subs to some extent, irrespective of outliers. Is it after target press?
No, it seems that even on background trials where there was not a target beforehand.
d %>%
group_by(subject) %>%
mutate(t_to_b = c(FALSE, diff(as.numeric(as.factor(trial_stimulus_type == "target"))) == -1) & trial_stimulus_type == "background",
d_to_b = c(FALSE, diff(as.numeric(as.factor(trial_stimulus_type == "deviant"))) == -1) & trial_stimulus_type == "background") %>%
filter(trial_stimulus_type == "background") %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram() +
scale_x_log10() +
facet_grid(d_to_b~t_to_b)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
just look at the best of trials of each individual
d %>%
group_by(subject) %>%
mutate(mlog = mean(log(trial_looking_time)),
sdlog = sd(log(trial_looking_time)),
best_trials = log(trial_looking_time) < mlog + 2*sdlog &&
log(trial_looking_time) > mlog - 2*sdlog) %>%
filter(best_trials == TRUE) %>%
ggplot(aes(x = trial_looking_time, fill = trial_stimulus_complexity)) +
geom_histogram() +
scale_x_log10() ## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## block type by complexity
only looking at trials within 2sd of the means
d %>%
group_by(subject) %>%
mutate(mlog = mean(log(trial_looking_time)),
sdlog = sd(log(trial_looking_time)),
best_trials = log(trial_looking_time) < mlog + 2*sdlog &&
log(trial_looking_time) > mlog - 2*sdlog) %>%
filter(best_trials == TRUE) %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram() +
scale_x_log10() +
facet_wrap(~block)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
doesn’t seem to make a huge difference with or without the outlier
d %>%
group_by(subject) %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram() +
scale_x_log10() +
facet_wrap(~block)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
break down by trial type and stimuli type
d %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram()+
scale_x_log10() +
facet_grid(block~trial_stimulus_type)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
ok, maybe it’s time to consider number of repetitions. maybe the second hump means participants getting very bored and they started zoning out???
library(mgcv)## Loading required package: nlme##
## Attaching package: 'nlme'## The following object is masked from 'package:lme4':
##
## lmList## The following object is masked from 'package:dplyr':
##
## collapse## This is mgcv 1.8-33. For overview type 'help("mgcv-package")'.rep_d <- d %>%
mutate(
number = 1) %>%
group_by(
subject, block, trial_stimulus_type
) %>%
mutate(num_times_stimulus_seen = cumsum(number)) %>%
filter(trial_stimulus_type == "background")
gam_d <- rep_d %>%
mutate(trial_stimulus_complexity = as.factor(trial_stimulus_complexity),
block = as.factor(block))
gam_m <- gam(log(trial_looking_time) ~ s(num_times_stimulus_seen),
data = gam_d,
method = "REML")
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## log(trial_looking_time) ~ s(num_times_stimulus_seen)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.270316 0.008667 723.4 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(num_times_stimulus_seen) 7.001 8.072 12.81 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.0181 Deviance explained = 1.93%
## -REML = 5591.4 Scale est. = 0.4234 n = 5636plot(gam_m)
note this is not a very good model
gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-0.0005601243,-3.726007e-07]
## (score 5591.437 & scale 0.4233963).
## Hessian positive definite, eigenvalue range [1.99837,2817.004].
## Model rank = 10 / 10
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(num_times_stimulus_seen) 9 7 1 0.44but still let’s give it a try: maybe 10 repetitions are the “turning point”
rep_d %>%
mutate(
reptition_type = if_else(num_times_stimulus_seen > 10, "more", "less")
) %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram()+
scale_x_log10() +
facet_grid(block~reptition_type)## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
umm this really doesn’t make any sense to me…
age really has nothing to do with it
d %>%
mutate(
m_age = mean(demog_age, na.rm = TRUE),
age_group = if_else(demog_age < m_age | demog_age == m_age, "young", "old")
) %>%
filter(!is.na(age_group)) %>%
ggplot(aes(x = trial_looking_time, fill = age_group)) +
geom_histogram() +
scale_x_log10() ## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
complexity difference
d %>%
ggplot(aes(x = trial_looking_time,
fill = trial_stimulus_complexity),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000)## Warning: Removed 94 rows containing non-finite values (stat_density).
d %>%
ggplot(aes(x = trial_looking_time,
fill = trial_stimulus_complexity),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000) +
facet_wrap(~subject)## Warning: Removed 94 rows containing non-finite values (stat_density).
block difference
d %>%
ggplot(aes(x = trial_looking_time,
fill = block),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000)## Warning: Removed 94 rows containing non-finite values (stat_density).
d %>%
ggplot(aes(x = trial_looking_time,
fill = block),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000) +
facet_wrap(~subject)## Warning: Removed 94 rows containing non-finite values (stat_density).
Complexity differences no target
d_no_target %>%
ggplot(aes(x = trial_looking_time,
fill = trial_stimulus_complexity),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000)## Warning: Removed 64 rows containing non-finite values (stat_density).
d_no_target %>%
ggplot(aes(x = trial_looking_time,
fill = trial_stimulus_complexity),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000) +
facet_wrap(~subject)## Warning: Removed 64 rows containing non-finite values (stat_density).
Block differences no target
d_no_target %>%
ggplot(aes(x = trial_looking_time,
fill = block),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000)## Warning: Removed 64 rows containing non-finite values (stat_density).
d_no_target %>%
ggplot(aes(x = trial_looking_time,
fill = block),
) +
geom_density(alpha = 0.5)+
xlim(0, 6000) +
facet_wrap(~subject)## Warning: Removed 64 rows containing non-finite values (stat_density).## Warning: Groups with fewer than two data points have been dropped.## Warning in max(ids, na.rm = TRUE): no non-missing arguments to max; returning -
## Inf
Visualizing aggregarded looking time
d_sum_individual <- d %>%
group_by(subject, block) %>%
summarise(
mean_lt = mean(trial_looking_time, na.rm = TRUE),
sd = sd(trial_looking_time, na.rm = TRUE),
n = n(),
ci_range_95 = qt(1 - (0.05 / 2), n - 1) * (sd/sqrt(n)),
ci_ub = mean_lt + ci_range_95,
ci_lb = mean_lt - ci_range_95
)## `summarise()` regrouping output by 'subject' (override with `.groups` argument)d_sum <- d %>%
group_by(block) %>%
summarise(
mean_lt = mean(trial_looking_time, na.rm = TRUE),
sd = sd(trial_looking_time, na.rm = TRUE),
n = n(),
ci_range_95 = qt(1 - (0.05 / 2), n - 1) * (sd/sqrt(n)),
ci_ub = mean_lt + ci_range_95,
ci_lb = mean_lt - ci_range_95
)## `summarise()` ungrouping output (override with `.groups` argument)aggregated
d_sum %>% ggplot(aes(x = block, y = mean_lt)) +
geom_pointrange(aes(ymin = ci_lb, ymax = ci_ub)) +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
individual
something weird happened
d_sum_individual %>%
ggplot(aes(x = block, y = mean_lt)) +
geom_pointrange(aes(ymin = ci_lb, ymax = ci_ub)) +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
d_sum_individual %>%
ggplot(aes(x = block, y = mean_lt)) +
geom_pointrange(aes(ymin = ci_lb, ymax = ci_ub)) +
facet_wrap(~subject) +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
weird ones
We can see three very weird ones: SS1604513317537 SS1604515995769 SS1604516882157
weird <- c("SS1604513317537",
"SS1604515995769",
"SS1604516660396")
d_sum_individual %>%
filter(subject %in% weird) %>%
ggplot(aes(x = block, y = mean_lt)) +
geom_pointrange(aes(ymin = ci_lb, ymax = ci_ub)) +
facet_wrap(~subject) +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
d %>%
filter(subject %in% weird) %>%
ggplot(aes(x = trial_looking_time)) +
geom_histogram(bins = 90) +
facet_wrap(~subject) +
xlim(0, 6000)+
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))## Warning: Removed 2 rows containing non-finite values (stat_bin).## Warning: Removed 6 rows containing missing values (geom_bar).
Excluding weird ones
d_no_weird <- d %>%
filter(!(subject %in% weird))
d_no_weird_sum <- d_no_weird %>%
group_by(block) %>%
summarise(
mean_lt = mean(trial_looking_time, na.rm = TRUE),
sd = sd(trial_looking_time, na.rm = TRUE),
n = n(),
ci_range_95 = qt(1 - (0.05 / 2), n - 1) * (sd/sqrt(n)),
ci_ub = mean_lt + ci_range_95,
ci_lb = mean_lt - ci_range_95
)## `summarise()` ungrouping output (override with `.groups` argument)d_no_weird_sum %>% ggplot(aes(x = block, y = mean_lt)) +
geom_pointrange(aes(ymin = ci_lb, ymax = ci_ub)) +
theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust=1))
Demographic
SES
education == “Some high school” ~ 1, education == “High school diploma” ~ 2, education == “Associate Degree/Technical certification” ~ 3, education == “Bachelor’s Degree” ~ 4, education == “Master’s Degree” ~ 5, education == “Doctorate/Professional degree” ~ 6, education == “Other” ~ NA_real_
d %>%
distinct(subject, .keep_all = TRUE) %>%
ggplot(aes(x = demog_education)) +
geom_histogram()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 4 rows containing non-finite values (stat_bin).
null_m <- lmer(trial_looking_time ~ 1 + (1|subject),
data = filter(d, !is.na(demog_education)))
edu_m <- lmer(trial_looking_time ~ demog_education + (1|subject),
data = filter(d, !is.na(demog_education)))
summary(edu_m)## Linear mixed model fit by REML ['lmerMod']
## Formula: trial_looking_time ~ demog_education + (1 | subject)
## Data: filter(d, !is.na(demog_education))
##
## REML criterion at convergence: 139982.2
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -0.940 -0.136 -0.087 -0.027 32.089
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 272043 521.6
## Residual 8451381 2907.1
## Number of obs: 7448, groups: subject, 38
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 174.8 490.0 0.357
## demog_education 197.5 124.5 1.587
##
## Correlation of Fixed Effects:
## (Intr)
## demog_edctn -0.983anova(null_m, edu_m)## refitting model(s) with ML (instead of REML)## Data: filter(d, !is.na(demog_education))
## Models:
## null_m: trial_looking_time ~ 1 + (1 | subject)
## edu_m: trial_looking_time ~ demog_education + (1 | subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## null_m 3 140013 140034 -70004 140007
## edu_m 4 140012 140040 -70002 140004 2.5686 1 0.109Age
d %>%
distinct(subject, .keep_all = TRUE) %>%
ggplot(aes(x = demog_age)) +
geom_histogram()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 4 rows containing non-finite values (stat_bin).
null_m <- lmer(trial_looking_time ~ 1 + (1|subject),
data = filter(d, !is.na(demog_age)))
age_m <- lmer(trial_looking_time ~ demog_age + (1|subject),
data = filter(d, !is.na(demog_age)))
summary(age_m)## Linear mixed model fit by REML ['lmerMod']
## Formula: trial_looking_time ~ demog_age + (1 | subject)
## Data: filter(d, !is.na(demog_age))
##
## REML criterion at convergence: 139990.2
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -0.938 -0.139 -0.086 -0.025 32.086
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 293140 541.4
## Residual 8451383 2907.1
## Number of obs: 7448, groups: subject, 38
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 853.827 286.498 2.980
## demog_age 2.296 7.309 0.314
##
## Correlation of Fixed Effects:
## (Intr)
## demog_age -0.945anova(null_m, age_m)## refitting model(s) with ML (instead of REML)## Data: filter(d, !is.na(demog_age))
## Models:
## null_m: trial_looking_time ~ 1 + (1 | subject)
## age_m: trial_looking_time ~ demog_age + (1 | subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## null_m 3 140013 140034 -70004 140007
## age_m 4 140015 140043 -70003 140007 0.104 1 0.7471number of background repetition on looking time
full, aggregated, by complexity
full_aggregated <- d %>%
mutate(
number = 1) %>%
group_by(
subject, block, trial_stimulus_type
) %>%
mutate(num_times_stimulus_seen = cumsum(number))
full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = trial_stimulus_complexity)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10))## `geom_smooth()` using formula 'y ~ x'
full, aggregated, by blocks
full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = block)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10))## `geom_smooth()` using formula 'y ~ x'
no weird, aggregated, by complexity
excluded_sum <- full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
filter(!(subject %in% weird))
excluded_sum %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = trial_stimulus_complexity)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10))## `geom_smooth()` using formula 'y ~ x'
no weird, aggregated, by blocks
excluded_sum %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = block)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10))## `geom_smooth()` using formula 'y ~ x'
individual, by complexity
full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = trial_stimulus_complexity)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10)) +
facet_wrap(~subject)## `geom_smooth()` using formula 'y ~ x'
individual, by block
full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
ggplot(
aes(y = log(trial_looking_time),
x = num_times_stimulus_seen,
color = block)
) +
geom_point(aes(alpha = 0.2), size = 3, shape = ".") +
guides(alpha = FALSE) +
labs(color = "Stimulus Complexity") +
ylab("Mean Looking Time (log)") +
xlab("Number of Stimulus Reptitions") +
geom_smooth(method = "lm") +
theme(axis.text = element_text(size = 10)) +
facet_wrap(~subject)## `geom_smooth()` using formula 'y ~ x'
Model
linear
block
null_m <- lmer(trial_looking_time ~ 1 + (1|subject),
data = d)
basic_m <- lmer(trial_looking_time ~ block + (1|subject),
data = d)
summary(basic_m)## Linear mixed model fit by REML ['lmerMod']
## Formula: trial_looking_time ~ block + (1 | subject)
## Data: d
##
## REML criterion at convergence: 151943
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.108 -0.127 -0.067 -0.016 33.208
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 568448 754
## Residual 13515244 3676
## Number of obs: 7888, groups: subject, 42
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 939.86 142.80 6.582
## blockall_simple 233.59 116.02 2.013
## blockmixed_complex_deviant 34.13 116.79 0.292
## blockmixed_simple_deviant 64.13 118.66 0.540
##
## Correlation of Fixed Effects:
## (Intr) blckl_ blckmxd_c_
## blckll_smpl -0.406
## blckmxd_cm_ -0.409 0.497
## blckmxd_sm_ -0.397 0.489 0.490anova(null_m, basic_m)## refitting model(s) with ML (instead of REML)## Data: d
## Models:
## null_m: trial_looking_time ~ 1 + (1 | subject)
## basic_m: trial_looking_time ~ block + (1 | subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## null_m 3 151999 152020 -75996 151993
## basic_m 6 152000 152042 -75994 151988 4.7832 3 0.1884stimuli repetition
null_m <- lmer(log(trial_looking_time) ~ 1 + (1|subject),
data = excluded_sum)
rep_m <- lmer(log(trial_looking_time) ~ num_times_stimulus_seen + (1|subject),
data = excluded_sum)
summary(rep_m)## Linear mixed model fit by REML ['lmerMod']
## Formula: log(trial_looking_time) ~ num_times_stimulus_seen + (1 | subject)
## Data: excluded_sum
##
## REML criterion at convergence: 9108.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3206 -0.4424 -0.1691 0.1589 9.2370
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.1305 0.3613
## Residual 0.3031 0.5505
## Number of obs: 5434, groups: subject, 39
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 6.3491570 0.0598278 106.12
## num_times_stimulus_seen -0.0043827 0.0007391 -5.93
##
## Correlation of Fixed Effects:
## (Intr)
## nm_tms_stm_ -0.222anova(null_m, rep_m)## refitting model(s) with ML (instead of REML)## Data: excluded_sum
## Models:
## null_m: log(trial_looking_time) ~ 1 + (1 | subject)
## rep_m: log(trial_looking_time) ~ num_times_stimulus_seen + (1 | subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## null_m 3 9133.5 9153.3 -4563.7 9127.5
## rep_m 4 9100.4 9126.8 -4546.2 9092.4 35.056 1 3.204e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1stimuli repetition and block interaction
interaction_m <- lmer(log(trial_looking_time) ~ num_times_stimulus_seen * block + (1|subject),
data = excluded_sum)
summary(interaction_m)## Linear mixed model fit by REML ['lmerMod']
## Formula: log(trial_looking_time) ~ num_times_stimulus_seen * block + (1 |
## subject)
## Data: excluded_sum
##
## REML criterion at convergence: 9147.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -3.3465 -0.4503 -0.1655 0.1712 9.2622
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.1306 0.3614
## Residual 0.3028 0.5502
## Number of obs: 5434, groups: subject, 39
##
## Fixed effects:
## Estimate Std. Error t value
## (Intercept) 6.342578 0.065381 97.010
## num_times_stimulus_seen -0.005310 0.001475 -3.601
## blockall_simple 0.073070 0.043044 1.698
## blockmixed_complex_deviant -0.013373 0.043044 -0.311
## blockmixed_simple_deviant -0.033754 0.043133 -0.783
## num_times_stimulus_seen:blockall_simple -0.001964 0.002085 -0.942
## num_times_stimulus_seen:blockmixed_complex_deviant 0.002592 0.002085 1.243
## num_times_stimulus_seen:blockmixed_simple_deviant 0.003113 0.002094 1.487
##
## Correlation of Fixed Effects:
## (Intr) nm_t__ blckl_ blckmxd_c_ blckmxd_s_ nm___:_
## nm_tms_stm_ -0.406
## blckll_smpl -0.329 0.617
## blckmxd_cm_ -0.329 0.617 0.500
## blckmxd_sm_ -0.328 0.615 0.499 0.499
## nm_tms_s_:_ 0.287 -0.707 -0.872 -0.436 -0.435
## nm_tms_stmls_sn:blckmxd_c_ 0.287 -0.707 -0.436 -0.872 -0.435 0.500
## nm_tms_stmls_sn:blckmxd_s_ 0.286 -0.704 -0.434 -0.434 -0.871 0.498
## nm_tms_stmls_sn:blckmxd_c_
## nm_tms_stm_
## blckll_smpl
## blckmxd_cm_
## blckmxd_sm_
## nm_tms_s_:_
## nm_tms_stmls_sn:blckmxd_c_
## nm_tms_stmls_sn:blckmxd_s_ 0.498anova(null_m, interaction_m)## refitting model(s) with ML (instead of REML)## Data: excluded_sum
## Models:
## null_m: log(trial_looking_time) ~ 1 + (1 | subject)
## interaction_m: log(trial_looking_time) ~ num_times_stimulus_seen * block + (1 |
## interaction_m: subject)
## Df AIC BIC logLik deviance Chisq Chi Df Pr(>Chisq)
## null_m 3 9133.5 9153.3 -4563.7 9127.5
## interaction_m 10 9100.9 9166.9 -4540.4 9080.9 46.575 7 6.755e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1GAM
only focusing on looking time at the background
library(mgcv)
gam_d <- full_aggregated %>%
filter(trial_stimulus_type == "background") %>%
mutate(trial_stimulus_complexity = as.factor(trial_stimulus_complexity),
block = as.factor(block))number of repetition
gam_m <- gam(trial_looking_time ~ s(num_times_stimulus_seen),
data = gam_d,
method = "REML")
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## trial_looking_time ~ s(num_times_stimulus_seen)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 917.30 56.23 16.31 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(num_times_stimulus_seen) 1.057 1.111 0.32 0.643
##
## R-sq.(adj) = -0.000115 Deviance explained = 0.00725%
## -REML = 55035 Scale est. = 1.7822e+07 n = 5636plot(gam_m,
pages = 1,
se = TRUE,
shade = TRUE)
gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-0.0001684747,-5.863759e-05]
## (score 55035.44 & scale 17821988).
## Hessian positive definite, eigenvalue range [0.001528264,2817].
## Model rank = 10 / 10
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(num_times_stimulus_seen) 9.00 1.06 1.01 0.71concurvity(gam_m, full = TRUE)## para s(num_times_stimulus_seen)
## worst 1.113898e-26 1.124554e-26
## observed 1.113898e-26 5.090767e-31
## estimate 1.113898e-26 3.587588e-29number of repetition (with log?)
gam_m <- gam(log(trial_looking_time) ~ s(num_times_stimulus_seen),
data = gam_d,
method = "REML")
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## log(trial_looking_time) ~ s(num_times_stimulus_seen)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.270316 0.008667 723.4 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F p-value
## s(num_times_stimulus_seen) 7.001 8.072 12.81 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.0181 Deviance explained = 1.93%
## -REML = 5591.4 Scale est. = 0.4234 n = 5636plot(gam_m,
pages = 1,
se = TRUE,
shade = TRUE)
gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-0.0005601243,-3.726007e-07]
## (score 5591.437 & scale 0.4233963).
## Hessian positive definite, eigenvalue range [1.99837,2817.004].
## Model rank = 10 / 10
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(num_times_stimulus_seen) 9 7 1.02 0.95concurvity(gam_m, full = TRUE)## para s(num_times_stimulus_seen)
## worst 1.113898e-26 1.124554e-26
## observed 1.113898e-26 6.851835e-29
## estimate 1.113898e-26 3.587588e-29number of repetition by stimulus complexity
gam_m <- gam(trial_looking_time ~ s(num_times_stimulus_seen,
by = trial_stimulus_complexity),
data = gam_d,
method = "REML")
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## trial_looking_time ~ s(num_times_stimulus_seen, by = trial_stimulus_complexity)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 917.31 56.21 16.32 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 1.036 1.070 3.789
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 1.360 1.634 0.739
## p-value
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 0.0503 .
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 0.3521
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.000686 Deviance explained = 0.111%
## -REML = 55027 Scale est. = 1.7808e+07 n = 5636plot(gam_m,
pages = 1,
se = TRUE,
shade = TRUE)
gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-0.01454695,0.01786747]
## (score 55027.21 & scale 17807718).
## Hessian positive definite, eigenvalue range [0.0142575,2816.482].
## Model rank = 19 / 19
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 9.00 1.04 1
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 9.00 1.36 1
## p-value
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 0.14
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 0.19concurvity(gam_m, full = TRUE)## para
## worst 1.207005e-05
## observed 1.207005e-05
## estimate 1.207005e-05
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex
## worst 6.119153e-06
## observed 3.148278e-08
## estimate 5.550473e-07
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple
## worst 5.950974e-06
## observed 1.834908e-07
## estimate 5.413397e-07number of repetition by stimulus complexity (log?)
gam_m <- gam(log(trial_looking_time) ~ s(num_times_stimulus_seen,
by = trial_stimulus_complexity),
data = gam_d,
method = "REML")
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## log(trial_looking_time) ~ s(num_times_stimulus_seen, by = trial_stimulus_complexity)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.270357 0.008666 723.6 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 5.158 6.269 4.719
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 6.463 7.606 10.020
## p-value
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 5.71e-05 ***
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.0185 Deviance explained = 2.05%
## -REML = 5597 Scale est. = 0.42323 n = 5636plot(gam_m,
pages = 1,
se = TRUE,
shade = TRUE)
gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 7 iterations.
## Gradient range [-0.0002553821,-2.062741e-10]
## (score 5597.005 & scale 0.423228).
## Hessian positive definite, eigenvalue range [0.8220389,2816.504].
## Model rank = 19 / 19
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 9.00 5.16 0.98
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 9.00 6.46 0.98
## p-value
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex 0.12
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple 0.09 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1concurvity(gam_m, full = TRUE)## para
## worst 1.207005e-05
## observed 1.207005e-05
## estimate 1.207005e-05
## s(num_times_stimulus_seen):trial_stimulus_complexitycomplex
## worst 6.119153e-06
## observed 1.516816e-06
## estimate 5.550473e-07
## s(num_times_stimulus_seen):trial_stimulus_complexitysimple
## worst 5.950974e-06
## observed 1.684933e-06
## estimate 5.413397e-07number of repetition by block
gam_m <- gam(trial_looking_time ~ s(num_times_stimulus_seen, by = block),
data = gam_d,
method = "REML")
plot(gam_m, all.terms = TRUE, pages = 1,
shade = TRUE,
rug = TRUE,
se = TRUE)
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## trial_looking_time ~ s(num_times_stimulus_seen, by = block)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 917.45 56.21 16.32 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F
## s(num_times_stimulus_seen):blockall_complex 1.345 1.612 4.193
## s(num_times_stimulus_seen):blockall_simple 1.009 1.018 1.646
## s(num_times_stimulus_seen):blockmixed_complex_deviant 1.064 1.126 0.121
## s(num_times_stimulus_seen):blockmixed_simple_deviant 1.050 1.097 0.180
## p-value
## s(num_times_stimulus_seen):blockall_complex 0.0444 *
## s(num_times_stimulus_seen):blockall_simple 0.1971
## s(num_times_stimulus_seen):blockmixed_complex_deviant 0.7547
## s(num_times_stimulus_seen):blockmixed_simple_deviant 0.6922
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.000711 Deviance explained = 0.15%
## -REML = 55014 Scale est. = 1.7807e+07 n = 5636gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 6 iterations.
## Gradient range [-0.01500944,0.01919385]
## (score 55014.2 & scale 17807260).
## Hessian positive definite, eigenvalue range [0.00108231,2815.481].
## Model rank = 37 / 37
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(num_times_stimulus_seen):blockall_complex 9.00 1.35 1.01 0.68
## s(num_times_stimulus_seen):blockall_simple 9.00 1.01 1.01 0.62
## s(num_times_stimulus_seen):blockmixed_complex_deviant 9.00 1.06 1.01 0.61
## s(num_times_stimulus_seen):blockmixed_simple_deviant 9.00 1.05 1.01 0.71concurvity(gam_m, full = TRUE)## para s(num_times_stimulus_seen):blockall_complex
## worst 4.313787e-05 6.092671e-06
## observed 4.313787e-05 5.126272e-06
## estimate 4.313787e-05 4.633189e-06
## s(num_times_stimulus_seen):blockall_simple
## worst 5.199012e-06
## observed 2.897025e-06
## estimate 2.704647e-06
## s(num_times_stimulus_seen):blockmixed_complex_deviant
## worst 1.180826e-05
## observed 3.280353e-06
## estimate 2.557590e-06
## s(num_times_stimulus_seen):blockmixed_simple_deviant
## worst 2.003918e-05
## observed 5.720030e-06
## estimate 7.292117e-06number of repetition by block (log?)
gam_m <- gam(log(trial_looking_time) ~ s(num_times_stimulus_seen, by = block),
data = gam_d,
method = "REML")
plot(gam_m, all.terms = TRUE, pages = 1,
shade = TRUE,
rug = TRUE,
se = TRUE)
summary(gam_m)##
## Family: gaussian
## Link function: identity
##
## Formula:
## log(trial_looking_time) ~ s(num_times_stimulus_seen, by = block)
##
## Parametric coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.27042 0.00867 723.3 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Approximate significance of smooth terms:
## edf Ref.df F
## s(num_times_stimulus_seen):blockall_complex 2.463 3.068 4.672
## s(num_times_stimulus_seen):blockall_simple 6.443 7.587 8.832
## s(num_times_stimulus_seen):blockmixed_complex_deviant 3.332 4.136 4.069
## s(num_times_stimulus_seen):blockmixed_simple_deviant 2.302 2.869 1.355
## p-value
## s(num_times_stimulus_seen):blockall_complex 0.00267 **
## s(num_times_stimulus_seen):blockall_simple < 2e-16 ***
## s(num_times_stimulus_seen):blockmixed_complex_deviant 0.00228 **
## s(num_times_stimulus_seen):blockmixed_simple_deviant 0.21364
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## R-sq.(adj) = 0.0176 Deviance explained = 2.02%
## -REML = 5603.3 Scale est. = 0.4236 n = 5636gam.check(gam_m)
##
## Method: REML Optimizer: outer newton
## full convergence after 7 iterations.
## Gradient range [5.879741e-11,5.306652e-06]
## (score 5603.29 & scale 0.4235977).
## Hessian positive definite, eigenvalue range [0.04763123,2815.503].
## Model rank = 37 / 37
##
## Basis dimension (k) checking results. Low p-value (k-index<1) may
## indicate that k is too low, especially if edf is close to k'.
##
## k' edf k-index p-value
## s(num_times_stimulus_seen):blockall_complex 9.00 2.46 1 0.46
## s(num_times_stimulus_seen):blockall_simple 9.00 6.44 1 0.37
## s(num_times_stimulus_seen):blockmixed_complex_deviant 9.00 3.33 1 0.47
## s(num_times_stimulus_seen):blockmixed_simple_deviant 9.00 2.30 1 0.46concurvity(gam_m, full = TRUE)## para s(num_times_stimulus_seen):blockall_complex
## worst 4.313787e-05 6.092671e-06
## observed 4.313787e-05 3.784249e-06
## estimate 4.313787e-05 4.633189e-06
## s(num_times_stimulus_seen):blockall_simple
## worst 5.199012e-06
## observed 1.156366e-06
## estimate 2.704647e-06
## s(num_times_stimulus_seen):blockmixed_complex_deviant
## worst 1.180826e-05
## observed 1.124140e-06
## estimate 2.557590e-06
## s(num_times_stimulus_seen):blockmixed_simple_deviant
## worst 2.003918e-05
## observed 1.607399e-05
## estimate 7.292117e-06within subject difference