Main visual precision statistics

library(tidyverse)
library(here)
library(lmerTest)
library(MuMIn)
library(lme4)
library(dotenv)
library(broom)
library(broom.mixed)
library(effects)
library(emmeans)

env_file = here(".env")
if (file.exists(env_file)) {
  load_dot_env(file = env_file)
  PROJECT_VERSION <- Sys.getenv("PROJECT_VERSION")
  if (PROJECT_VERSION == "") PROJECT_VERSION <- "main"
} else {
  PROJECT_VERSION <- "main"
}

Warning in readLines(file): incomplete final line found on
'/Users/tspri/Documents/visual-precision/.env'

VALID_SECTIONS <- c("sample1", "sample2")
DATA_FOLDER    <- if (PROJECT_VERSION %in% c(VALID_SECTIONS, "main")) "main" else PROJECT_VERSION
SECTION_FILTER <- if (PROJECT_VERSION %in% VALID_SECTIONS) PROJECT_VERSION else NULL
source(here("analysis/lmer_helpers.R"))

Load data

trial_metadata <- read.csv(here("data","metadata","level-trialtype_data.csv"))
trial_summary_data <- read.csv(here("data", DATA_FOLDER, "processed_data", "level-trials_data.csv"))
all_similarities <- read.csv(here("data", "embeddings", "similarities-all_data.csv"))
cdi_data <- read.csv(here("data", "main", "data_to_analyze", "level-participant_source-cdi_added-percentile_data.csv"))
usable_trials <- trial_summary_data |>
  # excluding possible scam participant
  filter(exclude_participant_insufficient_data == 0 & trial_exclusion == 0 & exclude_participant == 0 & SubjectInfo.subjID != "PH2RNZ") |> left_join(cdi_data, by=c("SubjectInfo.subjID"="local_id"))
trial_metadata_with_sims <- trial_metadata |> left_join(all_similarities, by=c("Trials.targetImage"="text1", "Trials.distractorImage"="text2"))
layerwise <- read.csv(here("data/embeddings/similarities-layerwise_data.csv"))

# Merging with similarity information and mean-centering main effects
trials_with_effect_vars <- usable_trials |>
  left_join(trial_metadata_with_sims) |>
  mutate(age_in_months = SubjectInfo.testAge/30,
         aoa_difference = AoA_Est_target - AoA_Est_distractor) |>
  left_join(layerwise, by=c("Trials.targetImage"="text1", "Trials.distractorImage"="text2"))

Joining with `by = join_by(Trials.trialID, Trials.targetImage,
Trials.distractorImage, Trials.imagePair, section)`

Warning in left_join(usable_trials, trial_metadata_with_sims): Detected an unexpected many-to-many relationship between `x` and `y`.
ℹ Row 1 of `x` matches multiple rows in `y`.
ℹ Row 1 of `y` matches multiple rows in `x`.
ℹ If a many-to-many relationship is expected, set `relationship =
  "many-to-many"` to silence this warning.

# Filter by section if PROJECT_VERSION specifies a particular sample
if (!is.null(SECTION_FILTER)) {
  trials_with_effect_vars <- trials_with_effect_vars |> filter(section == SECTION_FILTER)
}

Sanity check - making sure all participants have at least 16 trials and that we have 83 participants

low_trial_count <- trials_with_effect_vars |> distinct(SubjectInfo.subjID,Trials.trialID) |> summarize(n=n(),.by=SubjectInfo.subjID) |> filter(n < 25)
nrow(trials_with_effect_vars |> distinct(SubjectInfo.subjID))

[1] 181

tidy_model <- function(main_effect){
  table_data <- tidy(main_effect, effects = "fixed") %>%
  mutate(
    #p.value = 2 * (1 - pt(abs(statistic),df)),  # Calculate p-values for lmer - just using default calculated ones
    p.value.condensed = case_when(
      p.value < .001 ~ "<.001", 
      p.value < .01 ~ "<.01",
      p.value < .05 ~ "<.05",
      TRUE ~ sprintf("%.3f", p.value)),
    term = case_when(
      term == "(Intercept)" ~ "Intercept",
      term == "scale(age_in_months)" ~ "Age (scaled)",
      #term == "scale(image_similarity)" ~ "Target-distractor image embedding similarity (scaled)",
      TRUE ~ term
    )
  ) %>%
  rename(
    Predictor = term,
    "b" = estimate,
    "SE" = std.error,
    "t" = statistic,  # Note: changed from z to t for lmer
    "p" = p.value.condensed,
    "p.full" = p.value
  ) %>%
  mutate(across(c("b", "SE", "t"), ~round(., 2)))
  return(table_data)
}

Main mixed-effects models

Model 1 and 2: These are the models we said we’d run in our pre-reg

prereg_text_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) + scale(age_in_months) + scale(AoA_Est_target) 
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)


summary(prereg_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) + scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + (scale(text_similarity) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20594

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.97699 -0.62682 -0.02847  0.67589  3.00951 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.034331 0.18529       
                    scale(text_similarity) 0.010530 0.10262  0.09 
 Trials.targetImage (Intercept)            0.009169 0.09575       
 Trials.imagePair   (Intercept)            0.002015 0.04489       
 Residual                                  0.932066 0.96544       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                          Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)              -0.014924   0.024518  62.564248  -0.609  0.54492    
scale(text_similarity)   -0.036497   0.018178  41.230422  -2.008  0.05126 .  
scale(age_in_months)      0.077383   0.017921 185.103921   4.318 2.56e-05 ***
scale(AoA_Est_target)    -0.079130   0.019207  42.740461  -4.120  0.00017 ***
scale(MeanSaliencyDiff)   0.008517   0.016900  64.141883   0.504  0.61600    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E
scl(txt_sm)  0.054                     
scl(g_n_mn)  0.005 -0.002              
scl(AA_Es_)  0.061  0.033  0.004       
scl(MnSlnD)  0.009  0.010 -0.003  0.023

#Swapping text similarity with image similarity:
prereg_image_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) +  scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(image_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)
summary(prereg_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) + scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + (scale(image_similarity) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + +(1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20598.9

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.02372 -0.63447 -0.02949  0.67498  2.94717 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.034289 0.18517       
                    scale(image_similarity) 0.006833 0.08266  0.30 
 Trials.targetImage (Intercept)             0.009219 0.09602       
 Trials.imagePair   (Intercept)             0.002352 0.04850       
 Residual                                   0.935294 0.96711       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                         Estimate Std. Error        df t value Pr(>|t|)    
(Intercept)              -0.01226    0.02468  61.14054  -0.497 0.621232    
scale(image_similarity)  -0.02758    0.01801  34.40858  -1.532 0.134736    
scale(age_in_months)      0.07703    0.01779 185.99796   4.329 2.44e-05 ***
scale(AoA_Est_target)    -0.07331    0.01947  43.67387  -3.764 0.000495 ***
scale(MeanSaliencyDiff)   0.00935    0.01696  60.68111   0.551 0.583456    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E
scl(mg_sml)  0.037                     
scl(g_n_mn)  0.005 -0.003              
scl(AA_Es_)  0.056 -0.111  0.004       
scl(MnSlnD)  0.009  0.008 -0.003  0.019

multimodal sim

multimodal_effect <- lmer(scale(corrected_target_looking) ~ scale(multimodal_similarity) + scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(multimodal_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)
summary(multimodal_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(corrected_target_looking) ~ scale(multimodal_similarity) +  
    scale(age_in_months) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(multimodal_similarity) | SubjectInfo.subjID) + (1 |  
    Trials.targetImage) + +(1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20588.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.02572 -0.63698 -0.02779  0.67952  3.16079 

Random effects:
 Groups             Name                         Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)                  0.034421 0.18553       
                    scale(multimodal_similarity) 0.013200 0.11489  0.09 
 Trials.targetImage (Intercept)                  0.010017 0.10008       
 Trials.imagePair   (Intercept)                  0.002947 0.05429       
 Residual                                        0.928857 0.96377       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                               Estimate Std. Error         df t value Pr(>|t|)
(Intercept)                   -0.013644   0.025437  60.255450  -0.536 0.593665
scale(multimodal_similarity)  -0.011927   0.019339  71.959084  -0.617 0.539353
scale(age_in_months)           0.074753   0.017880 185.062967   4.181 4.48e-05
scale(AoA_Est_target)         -0.077331   0.020036  42.042564  -3.860 0.000385
scale(MeanSaliencyDiff)        0.009165   0.017329  58.514227   0.529 0.598875
                                
(Intercept)                     
scale(multimodal_similarity)    
scale(age_in_months)         ***
scale(AoA_Est_target)        ***
scale(MeanSaliencyDiff)         
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E
scl(mltmd_)  0.048                     
scl(g_n_mn)  0.005 -0.001              
scl(AA_Es_)  0.061  0.069  0.004       
scl(MnSlnD)  0.006 -0.035 -0.003  0.015

Models from 1st prereg

text_similarity_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) +
                    + (scale(text_similarity) | SubjectInfo.subjID) 
                    + (1 | Trials.imagePair)
                    + (1 | Trials.targetImage), 
                    data = trials_with_effect_vars)
image_similarity_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +
                    + (scale(image_similarity) | SubjectInfo.subjID) 
                    
                    + (1 | Trials.imagePair)
                    + (1 | Trials.targetImage), 
                    data = trials_with_effect_vars)
summary(text_similarity_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) +  
    +(scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.imagePair) +  
    (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20602.7

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.96343 -0.63367 -0.02719  0.67335  3.05039 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.034452 0.18561       
                    scale(text_similarity) 0.010925 0.10452  0.09 
 Trials.targetImage (Intercept)            0.014382 0.11993       
 Trials.imagePair   (Intercept)            0.003897 0.06242       
 Residual                                  0.931329 0.96505       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                  -0.010287   0.027751  63.415667
scale(text_similarity)                       -0.032669   0.020426  41.286370
scale(age_in_months)                          0.076943   0.017986 184.373973
scale(text_similarity):scale(age_in_months)  -0.007322   0.013556 159.684999
                                            t value Pr(>|t|)    
(Intercept)                                  -0.371    0.712    
scale(text_similarity)                       -1.599    0.117    
scale(age_in_months)                          4.278 3.03e-05 ***
scale(text_similarity):scale(age_in_months)  -0.540    0.590    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__)
scl(txt_sm)  0.054              
scl(g_n_mn)  0.004 -0.001       
scl(_):(__) -0.001  0.002  0.074

summary(image_similarity_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    +(scale(image_similarity) | SubjectInfo.subjID) + (1 | Trials.imagePair) +  
    (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20604.8

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.00914 -0.63420 -0.02824  0.66996  2.93301 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.034590 0.18598       
                    scale(image_similarity) 0.006747 0.08214  0.31 
 Trials.targetImage (Intercept)             0.013765 0.11733       
 Trials.imagePair   (Intercept)             0.003616 0.06014       
 Residual                                   0.934849 0.96688       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                   -0.008025   0.027302  61.779850
scale(image_similarity)                       -0.032764   0.019617  32.383183
scale(age_in_months)                           0.075156   0.017929 185.355320
scale(image_similarity):scale(age_in_months)  -0.014834   0.012716 192.203413
                                             t value Pr(>|t|)    
(Intercept)                                   -0.294    0.770    
scale(image_similarity)                       -1.670    0.105    
scale(age_in_months)                           4.192 4.28e-05 ***
scale(image_similarity):scale(age_in_months)  -1.167    0.245    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__)
scl(mg_sml)  0.034              
scl(g_n_mn)  0.004 -0.003       
scl(_):(__) -0.002 -0.002  0.106

What about this similarity effect per section?

image_similarity_effect_sample1 <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                     + (1 | SubjectInfo.subjID)
                     + (1|Trials.targetImage), 
                    data = trials_with_effect_vars |> filter(section=="sample1"))
summary(image_similarity_effect_sample1)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    (1 | SubjectInfo.subjID) + (1 | Trials.targetImage)
   Data: filter(trials_with_effect_vars, section == "sample1")

REML criterion at convergence: 13853.8

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.89029 -0.62951 -0.01588  0.66561  2.94060 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.03544  0.1882  
 Trials.targetImage (Intercept) 0.02000  0.1414  
 Residual                       0.93863  0.9688  
Number of obs: 4936, groups:  SubjectInfo.subjID, 91; Trials.targetImage, 24

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                    -0.01295    0.03789   39.93406
scale(image_similarity)                        -0.06269    0.01937  333.55728
scale(age_in_months)                            0.05751    0.02407   89.55031
scale(image_similarity):scale(age_in_months)   -0.02695    0.01379 4847.01196
                                             t value Pr(>|t|)   
(Intercept)                                   -0.342  0.73440   
scale(image_similarity)                       -3.236  0.00133 **
scale(age_in_months)                           2.389  0.01899 * 
scale(image_similarity):scale(age_in_months)  -1.955  0.05067 . 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__)
scl(mg_sml) -0.001              
scl(g_n_mn)  0.005 -0.003       
scl(_):(__)  0.000 -0.004  0.005

image_similarity_effect_sample2 <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                     + (1 | SubjectInfo.subjID)
                     + (1|Trials.targetImage), 
                    data = trials_with_effect_vars |> filter(section=="sample2"))
summary(image_similarity_effect_sample2)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    (1 | SubjectInfo.subjID) + (1 | Trials.targetImage)
   Data: filter(trials_with_effect_vars, section == "sample2")

REML criterion at convergence: 6773

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-2.9586 -0.6503 -0.0386  0.6961  2.9211 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.03370  0.18358 
 Trials.targetImage (Intercept) 0.00683  0.08264 
 Residual                       0.95143  0.97541 
Number of obs: 2397, groups:  SubjectInfo.subjID, 90; Trials.targetImage, 24

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                  -7.052e-03  3.287e-02  3.452e+01
scale(image_similarity)                       3.578e-04  2.352e-02  6.222e+01
scale(age_in_months)                          1.017e-01  2.785e-02  8.922e+01
scale(image_similarity):scale(age_in_months) -4.856e-04  1.977e-02  2.298e+03
                                             t value Pr(>|t|)    
(Intercept)                                   -0.215 0.831385    
scale(image_similarity)                        0.015 0.987910    
scale(age_in_months)                           3.651 0.000439 ***
scale(image_similarity):scale(age_in_months)  -0.025 0.980406    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__)
scl(mg_sml) -0.006              
scl(g_n_mn)  0.010 -0.007       
scl(_):(__) -0.009  0.003  0.000

BV-image model

Trying out image only models

bv_image_effect <- lmer(scale(corrected_target_looking) ~ scale(dinov3.babyview_image_similarity) * scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                     + (scale(dinov3.babyview_image_similarity) | SubjectInfo.subjID) 
                    + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)
dinov3_effect <- lmer(scale(corrected_target_looking) ~ scale(dinov3_image_similarity) * scale(age_in_months)
                      + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(dinov3_image_similarity) | SubjectInfo.subjID) 
                    + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)
summary(bv_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(dinov3.babyview_image_similarity) *  
    scale(age_in_months) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(dinov3.babyview_image_similarity) | SubjectInfo.subjID) +  
    (1 | Trials.imagePair) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20572.4

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.05378 -0.62602 -0.02099  0.66687  2.95854 

Random effects:
 Groups             Name                                    Variance Std.Dev.
 SubjectInfo.subjID (Intercept)                             0.034156 0.18481 
                    scale(dinov3.babyview_image_similarity) 0.014714 0.12130 
 Trials.targetImage (Intercept)                             0.009761 0.09880 
 Trials.imagePair   (Intercept)                             0.003445 0.05869 
 Residual                                                   0.925480 0.96202 
 Corr 
      
 0.29 
      
      
      
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                               Estimate
(Intercept)                                                   -0.013392
scale(dinov3.babyview_image_similarity)                       -0.008564
scale(age_in_months)                                           0.075656
scale(AoA_Est_target)                                         -0.075136
scale(MeanSaliencyDiff)                                        0.009411
scale(dinov3.babyview_image_similarity):scale(age_in_months)  -0.036722
                                                             Std. Error
(Intercept)                                                    0.025588
scale(dinov3.babyview_image_similarity)                        0.021756
scale(age_in_months)                                           0.017849
scale(AoA_Est_target)                                          0.020280
scale(MeanSaliencyDiff)                                        0.017226
scale(dinov3.babyview_image_similarity):scale(age_in_months)   0.014567
                                                                     df t value
(Intercept)                                                   60.597982  -0.523
scale(dinov3.babyview_image_similarity)                       47.645377  -0.394
scale(age_in_months)                                         186.920351   4.239
scale(AoA_Est_target)                                         43.503350  -3.705
scale(MeanSaliencyDiff)                                       56.039841   0.546
scale(dinov3.babyview_image_similarity):scale(age_in_months) 201.901395  -2.521
                                                             Pr(>|t|)    
(Intercept)                                                  0.602621    
scale(dinov3.babyview_image_similarity)                      0.695584    
scale(age_in_months)                                         3.53e-05 ***
scale(AoA_Est_target)                                        0.000593 ***
scale(MeanSaliencyDiff)                                      0.586999    
scale(dinov3.babyview_image_similarity):scale(age_in_months) 0.012482 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(3.__) sc(__) s(AA_E s(MSD)
scl(dn3.__)  0.030                              
scl(g_n_mn)  0.004  0.003                       
scl(AA_Es_)  0.062 -0.167    0.004              
scl(MnSlnD)  0.007  0.037   -0.003  0.010       
s(3.__):(__  0.004  0.001    0.104 -0.001  0.005

summary(dinov3_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(corrected_target_looking) ~ scale(dinov3_image_similarity) *  
    scale(age_in_months) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(dinov3_image_similarity) | SubjectInfo.subjID) + (1 |  
    Trials.imagePair) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20597.1

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-2.9928 -0.6380 -0.0220  0.6782  2.9895 

Random effects:
 Groups             Name                           Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)                    0.034399 0.18547       
                    scale(dinov3_image_similarity) 0.010495 0.10245  0.11 
 Trials.targetImage (Intercept)                    0.009844 0.09922       
 Trials.imagePair   (Intercept)                    0.002738 0.05233       
 Residual                                          0.931247 0.96501       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                      Estimate Std. Error
(Intercept)                                          -0.012932   0.025204
scale(dinov3_image_similarity)                       -0.018569   0.019811
scale(age_in_months)                                  0.075138   0.017910
scale(AoA_Est_target)                                -0.074122   0.019919
scale(MeanSaliencyDiff)                               0.007917   0.017321
scale(dinov3_image_similarity):scale(age_in_months)  -0.020186   0.013803
                                                            df t value Pr(>|t|)
(Intercept)                                          61.903809  -0.513 0.609702
scale(dinov3_image_similarity)                       39.425296  -0.937 0.354299
scale(age_in_months)                                184.292558   4.195 4.23e-05
scale(AoA_Est_target)                                43.593734  -3.721 0.000564
scale(MeanSaliencyDiff)                              56.589956   0.457 0.649361
scale(dinov3_image_similarity):scale(age_in_months) 175.327761  -1.462 0.145412
                                                       
(Intercept)                                            
scale(dinov3_image_similarity)                         
scale(age_in_months)                                ***
scale(AoA_Est_target)                               ***
scale(MeanSaliencyDiff)                                
scale(dinov3_image_similarity):scale(age_in_months)    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(3__) sc(__) s(AA_E s(MSD)
scl(dnv3__)  0.033                             
scl(g_n_mn)  0.005 -0.001                      
scl(AA_Es_)  0.055 -0.104   0.004              
scl(MnSlnD)  0.009  0.089  -0.003  0.009       
s(3__):(__)  0.000  0.002   0.063 -0.001  0.003

r.squaredGLMM(prereg_image_effect)

            R2m        R2c
[1,] 0.01286489 0.06551206

r.squaredGLMM(prereg_text_effect)

            R2m        R2c
[1,] 0.01365907 0.06960221

r.squaredGLMM(bv_image_effect)

            R2m        R2c
[1,] 0.01318324 0.07521137

Image similarity is trending towards significance.

Checking if text similarity and image similarity are differently correlated with AoA

trials_with_effect_vars_metadata <- trials_with_effect_vars |> distinct(Trials.imagePair, text_similarity, image_similarity, dinov3.babyview_image_similarity)
cor.test(trial_metadata$AoA_Est_target, trial_metadata$text_similarity)

    Pearson's product-moment correlation

data:  trial_metadata$AoA_Est_target and trial_metadata$text_similarity
t = -0.098581, df = 94, p-value = 0.9217
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.2102248  0.1907073
sample estimates:
        cor 
-0.01016738 

cor.test(trial_metadata$AoA_Est_target, trial_metadata$image_similarity)

    Pearson's product-moment correlation

data:  trial_metadata$AoA_Est_target and trial_metadata$image_similarity
t = 1.4599, df = 94, p-value = 0.1477
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.05317716  0.33925485
sample estimates:
      cor 
0.1488962 

cor.test(trials_with_effect_vars_metadata$image_similarity, trials_with_effect_vars_metadata$dinov3.babyview_image_similarity)

    Pearson's product-moment correlation

data:  trials_with_effect_vars_metadata$image_similarity and trials_with_effect_vars_metadata$dinov3.babyview_image_similarity
t = 2.4265, df = 30, p-value = 0.02147
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 0.06562208 0.66045987
sample estimates:
      cor 
0.4050477 

cor.test(trials_with_effect_vars_metadata$text_similarity, trials_with_effect_vars_metadata$dinov3.babyview_image_similarity)

    Pearson's product-moment correlation

data:  trials_with_effect_vars_metadata$text_similarity and trials_with_effect_vars_metadata$dinov3.babyview_image_similarity
t = 1.396, df = 30, p-value = 0.173
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.1113072  0.5484361
sample estimates:
      cor 
0.2469723 

cor.test(trial_metadata$MeanSaliencyDiff, trial_metadata$image_similarity, method="pearson")

    Pearson's product-moment correlation

data:  trial_metadata$MeanSaliencyDiff and trial_metadata$image_similarity
t = -4.3596e-16, df = 94, p-value = 1
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.2004859  0.2004859
sample estimates:
         cor 
-4.49655e-17 

Well image similarity has a higher r but both are still insignificant.

visual saliency without AoA

vs_image_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair)
                    , 
                    data = trials_with_effect_vars)

vs_text_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair)
                    , 
                    data = trials_with_effect_vars)
summary(vs_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | SubjectInfo.subjID) + (1 |  
    Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20620

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.93995 -0.63458 -0.02431  0.67458  2.99094 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.034184 0.18489 
 Trials.targetImage (Intercept) 0.014095 0.11872 
 Trials.imagePair   (Intercept) 0.003747 0.06121 
 Residual                       0.941821 0.97047 
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                  -7.637e-03  2.748e-02  6.128e+01
scale(image_similarity)                      -3.163e-02  1.880e-02  2.755e+01
scale(age_in_months)                          7.536e-02  1.790e-02  1.851e+02
scale(MeanSaliencyDiff)                       1.086e-02  1.898e-02  6.378e+01
scale(image_similarity):scale(age_in_months) -1.580e-02  1.120e-02  7.158e+03
                                             t value Pr(>|t|)    
(Intercept)                                   -0.278    0.782    
scale(image_similarity)                       -1.682    0.104    
scale(age_in_months)                           4.210 3.98e-05 ***
scale(MeanSaliencyDiff)                        0.572    0.569    
scale(image_similarity):scale(age_in_months)  -1.411    0.158    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(mg_sml) -0.017                     
scl(g_n_mn)  0.004 -0.004              
scl(MnSlnD)  0.008  0.024 -0.003       
scl(_):(__) -0.004 -0.004 -0.008 -0.003

summary(vs_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (scale(text_similarity) | SubjectInfo.subjID) +  
    (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20608.5

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.96951 -0.63506 -0.02455  0.67307  3.05429 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.034463 0.18564       
                    scale(text_similarity) 0.010909 0.10445  0.09 
 Trials.targetImage (Intercept)            0.014716 0.12131       
 Trials.imagePair   (Intercept)            0.003928 0.06268       
 Residual                                  0.931334 0.96506       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                  -0.010218   0.027898  62.443958
scale(text_similarity)                       -0.032418   0.020492  41.121220
scale(age_in_months)                          0.076913   0.017987 184.366336
scale(MeanSaliencyDiff)                       0.010696   0.019177  64.684385
scale(text_similarity):scale(age_in_months)  -0.007326   0.013553 159.715698
                                            t value Pr(>|t|)    
(Intercept)                                  -0.366    0.715    
scale(text_similarity)                       -1.582    0.121    
scale(age_in_months)                          4.276 3.05e-05 ***
scale(MeanSaliencyDiff)                       0.558    0.579    
scale(text_similarity):scale(age_in_months)  -0.541    0.590    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(txt_sm)  0.054                     
scl(g_n_mn)  0.004 -0.002              
scl(MnSlnD)  0.009  0.020 -0.003       
scl(_):(__) -0.001  0.002  0.074 -0.001

Just a sanity check that adding our saliency metric as a covariate does not affect our similarity effects.

Alternate window analyses

Only using first instance of an item

first_instance_target <- trials_with_effect_vars |>
  group_by(SubjectInfo.subjID, Trials.targetImage) |>
  arrange(Trials.ordinal, .by_group = TRUE) |>
  slice(1) |>
  ungroup()

second_instance_target <- trials_with_effect_vars |>
  group_by(SubjectInfo.subjID, Trials.targetImage) |>
  arrange(Trials.ordinal, .by_group = TRUE) |>
  slice(2) |>
  ungroup()

primary_targets <- trials_with_effect_vars |>
  filter(Trials.trialType %in% c("easy", "hard"))
  
first_instance_primary_target <- first_instance_target |>
  filter(Trials.trialType %in% c("easy", "hard"))
  
first_instance_image_pair <- trials_with_effect_vars |>
  group_by(SubjectInfo.subjID, Trials.imagePair) |>
  arrange(Trials.ordinal, .by_group = TRUE) |>
  slice(1) |>
  ungroup()


first_instance_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = first_instance_target)

second_instance_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = second_instance_target)

first_instance_pt_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = first_instance_primary_target)

boundary (singular) fit: see help('isSingular')

pt_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) + scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = primary_targets)

first_instance_image_pair_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = first_instance_image_pair)

summary(first_instance_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | Trials.ordinal) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: first_instance_target

REML criterion at convergence: 10836.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.94659 -0.64069 -0.02467  0.67624  2.84944 

Random effects:
 Groups             Name        Variance  Std.Dev.
 SubjectInfo.subjID (Intercept) 0.0286845 0.16936 
 Trials.ordinal     (Intercept) 0.0008575 0.02928 
 Trials.targetImage (Intercept) 0.0102297 0.10114 
 Residual                       0.9528808 0.97616 
Number of obs: 3832, groups:  
SubjectInfo.subjID, 181; Trials.ordinal, 57; Trials.targetImage, 48

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                  -1.337e-03  2.551e-02  4.917e+01
scale(image_similarity)                      -3.728e-02  1.968e-02  1.153e+02
scale(age_in_months)                          8.575e-02  2.021e-02  1.825e+02
scale(MeanSaliencyDiff)                       6.147e-03  2.104e-02  5.785e+01
scale(image_similarity):scale(age_in_months) -5.694e-04  1.551e-02  3.703e+03
                                             t value Pr(>|t|)    
(Intercept)                                   -0.052   0.9584    
scale(image_similarity)                       -1.894   0.0608 .  
scale(age_in_months)                           4.243 3.51e-05 ***
scale(MeanSaliencyDiff)                        0.292   0.7712    
scale(image_similarity):scale(age_in_months)  -0.037   0.9707    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(mg_sml) -0.004                     
scl(g_n_mn)  0.005  0.000              
scl(MnSlnD)  0.001 -0.083 -0.005       
scl(_):(__) -0.002 -0.010  0.002 -0.003

summary(second_instance_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | Trials.ordinal) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: second_instance_target

REML criterion at convergence: 6885.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.89044 -0.63490 -0.01476  0.66275  2.74153 

Random effects:
 Groups             Name        Variance  Std.Dev.
 SubjectInfo.subjID (Intercept) 0.0203529 0.142664
 Trials.ordinal     (Intercept) 0.0000315 0.005612
 Trials.targetImage (Intercept) 0.0125075 0.111837
 Residual                       0.9613449 0.980482
Number of obs: 2429, groups:  
SubjectInfo.subjID, 181; Trials.ordinal, 51; Trials.targetImage, 34

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                   3.285e-03  3.061e-02  3.149e+01
scale(image_similarity)                      -4.247e-02  2.418e-02  1.259e+02
scale(age_in_months)                          6.675e-02  2.320e-02  1.642e+02
scale(MeanSaliencyDiff)                       6.965e-03  2.676e-02  4.309e+01
scale(image_similarity):scale(age_in_months) -3.154e-02  2.001e-02  2.383e+03
                                             t value Pr(>|t|)   
(Intercept)                                    0.107  0.91519   
scale(image_similarity)                       -1.756  0.08146 . 
scale(age_in_months)                           2.877  0.00454 **
scale(MeanSaliencyDiff)                        0.260  0.79591   
scale(image_similarity):scale(age_in_months)  -1.576  0.11509   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(mg_sml) -0.004                     
scl(g_n_mn)  0.003 -0.032              
scl(MnSlnD)  0.003  0.050 -0.022       
scl(_):(__) -0.019 -0.009  0.003 -0.003

summary(first_instance_pt_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | Trials.ordinal) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: first_instance_primary_target

REML criterion at convergence: 3968

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.81542 -0.65735 -0.02088  0.68711  2.80514 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.033698 0.18357 
 Trials.ordinal     (Intercept) 0.000000 0.00000 
 Trials.targetImage (Intercept) 0.009633 0.09815 
 Residual                       0.944792 0.97200 
Number of obs: 1400, groups:  
SubjectInfo.subjID, 181; Trials.ordinal, 41; Trials.targetImage, 16

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                   9.061e-04  3.827e-02  1.778e+01
scale(image_similarity)                      -3.518e-02  3.173e-02  3.600e+02
scale(age_in_months)                          1.055e-01  2.935e-02  1.804e+02
scale(MeanSaliencyDiff)                      -6.914e-03  3.663e-02  3.221e+01
scale(image_similarity):scale(age_in_months)  5.557e-02  2.572e-02  1.359e+03
                                             t value Pr(>|t|)    
(Intercept)                                    0.024  0.98137    
scale(image_similarity)                       -1.109  0.26826    
scale(age_in_months)                           3.594  0.00042 ***
scale(MeanSaliencyDiff)                       -0.189  0.85147    
scale(image_similarity):scale(age_in_months)   2.161  0.03090 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(mg_sml) -0.001                     
scl(g_n_mn)  0.000  0.011              
scl(MnSlnD)  0.002  0.415  0.004       
scl(_):(__)  0.007 -0.035  0.001 -0.008
optimizer (nloptwrap) convergence code: 0 (OK)
boundary (singular) fit: see help('isSingular')

summary(pt_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) + scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | Trials.ordinal) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: primary_targets

REML criterion at convergence: 10324.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.01446 -0.63271 -0.01787  0.67285  3.00036 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.05203  0.22811 
 Trials.ordinal     (Intercept) 0.00629  0.07931 
 Trials.targetImage (Intercept) 0.01426  0.11942 
 Residual                       0.92026  0.95930 
Number of obs: 3678, groups:  
SubjectInfo.subjID, 181; Trials.ordinal, 54; Trials.targetImage, 16

Fixed effects:
                          Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)              4.007e-03  4.068e-02  2.892e+01   0.098 0.922216    
scale(image_similarity) -2.722e-02  1.990e-02  1.775e+03  -1.368 0.171532    
scale(age_in_months)     8.129e-02  2.339e-02  1.882e+02   3.475 0.000634 ***
scale(MeanSaliencyDiff)  1.341e-02  2.777e-02  8.923e+01   0.483 0.630389    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__)
scl(mg_sml) -0.003              
scl(g_n_mn)  0.002 -0.005       
scl(MnSlnD)  0.016  0.397 -0.002

summary(first_instance_image_pair_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(MeanSaliencyDiff) + (1 | Trials.ordinal) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: first_instance_image_pair

REML criterion at convergence: 7919.2

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.83546 -0.63270 -0.03685  0.67349  2.88785 

Random effects:
 Groups             Name        Variance  Std.Dev.
 SubjectInfo.subjID (Intercept) 0.0220093 0.14836 
 Trials.targetImage (Intercept) 0.0083699 0.09149 
 Trials.ordinal     (Intercept) 0.0006849 0.02617 
 Residual                       0.9655488 0.98262 
Number of obs: 2790, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.ordinal, 40

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                  -1.602e-03  2.628e-02  2.901e+01
scale(image_similarity)                      -1.466e-02  2.113e-02  1.319e+02
scale(age_in_months)                          6.809e-02  2.166e-02  1.807e+02
scale(MeanSaliencyDiff)                      -2.423e-02  2.256e-02  5.723e+01
scale(image_similarity):scale(age_in_months)  2.736e-03  1.843e-02  2.624e+03
                                             t value Pr(>|t|)   
(Intercept)                                   -0.061  0.95181   
scale(image_similarity)                       -0.694  0.48900   
scale(age_in_months)                           3.143  0.00195 **
scale(MeanSaliencyDiff)                       -1.074  0.28729   
scale(image_similarity):scale(age_in_months)   0.149  0.88195   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(MSD)
scl(mg_sml) -0.004                     
scl(g_n_mn)  0.002  0.001              
scl(MnSlnD)  0.000 -0.032 -0.009       
scl(_):(__) -0.003 -0.002 -0.006  0.007

Random order effects

main_image_effect_ordinal <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (1 | Trials.ordinal)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = trials_with_effect_vars)
summary(main_image_effect_ordinal)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + (1 | Trials.ordinal) +  
    (1 | SubjectInfo.subjID) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20601.5

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.91229 -0.63888 -0.02136  0.67423  2.92802 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.034043 0.18451 
 Trials.ordinal     (Intercept) 0.005326 0.07298 
 Trials.targetImage (Intercept) 0.009602 0.09799 
 Residual                       0.938186 0.96860 
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.ordinal, 61; Trials.targetImage, 48

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                  -9.915e-03  2.649e-02  8.440e+01
scale(image_similarity)                      -2.887e-02  1.453e-02  2.770e+02
scale(age_in_months)                          7.549e-02  1.786e-02  1.855e+02
scale(AoA_Est_target)                        -7.735e-02  1.876e-02  4.942e+01
scale(MeanSaliencyDiff)                       7.089e-03  1.695e-02  8.116e+01
scale(image_similarity):scale(age_in_months) -1.605e-02  1.120e-02  7.176e+03
                                             t value Pr(>|t|)    
(Intercept)                                   -0.374 0.709099    
scale(image_similarity)                       -1.987 0.047922 *  
scale(age_in_months)                           4.226 3.73e-05 ***
scale(AoA_Est_target)                         -4.124 0.000142 ***
scale(MeanSaliencyDiff)                        0.418 0.676873    
scale(image_similarity):scale(age_in_months)  -1.433 0.151987    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E s(MSD)
scl(mg_sml) -0.018                            
scl(g_n_mn)  0.005 -0.006                     
scl(AA_Es_)  0.049 -0.147  0.005              
scl(MnSlnD)  0.009  0.024 -0.003  0.031       
scl(_):(__) -0.004 -0.005 -0.008  0.004 -0.003

Order does not explain much variance.

Target image order effects

ranked_trials <- trials_with_effect_vars |>
  group_by(SubjectInfo.subjID, Trials.targetImage) |>
  arrange(Trials.ordinal, .by_group = TRUE) |>
  mutate(
    slice_num = row_number(),
    order = case_when(
      slice_num == 1 ~ -0.5,
      slice_num == 2 ~ 0.5,
      TRUE ~ NA
    )
  )
main_image_effect_ranked <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)*(scale(order))
                                 
                    + scale(AoA_Est_target)
                    + scale(age_in_months)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = ranked_trials)
# |> filter(Trials.trialType %in% c("easy", "hard")
summary(main_image_effect_ranked)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * (scale(order)) +  
    scale(AoA_Est_target) + scale(age_in_months) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: ranked_trials

REML criterion at convergence: 17499.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.07639 -0.63464 -0.01797  0.67917  2.99934 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.039456 0.19863 
 Trials.targetImage (Intercept) 0.009212 0.09598 
 Residual                       0.923143 0.96080 
Number of obs: 6261, groups:  SubjectInfo.subjID, 181; Trials.targetImage, 48

Fixed effects:
                                       Estimate Std. Error         df t value
(Intercept)                          -1.930e-02  2.396e-02  9.466e+01  -0.806
scale(image_similarity)              -2.420e-02  1.576e-02  1.934e+02  -1.536
scale(order)                          2.534e-03  1.277e-02  5.076e+03   0.198
scale(AoA_Est_target)                -7.732e-02  1.874e-02  4.822e+01  -4.125
scale(age_in_months)                  7.870e-02  1.898e-02  1.878e+02   4.145
scale(image_similarity):scale(order) -3.513e-03  1.282e-02  3.688e+03  -0.274
                                     Pr(>|t|)    
(Intercept)                          0.422538    
scale(image_similarity)              0.126269    
scale(order)                         0.842711    
scale(AoA_Est_target)                0.000145 ***
scale(age_in_months)                 5.13e-05 ***
scale(image_similarity):scale(order) 0.784038    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) scl(r) s(AA_E sc(__)
scl(mg_sml) -0.022                            
scale(ordr)  0.043  0.017                     
scl(AA_Es_)  0.060 -0.166 -0.030              
scl(g_n_mn)  0.006 -0.010  0.000  0.007       
scl(mg_):()  0.014  0.057  0.010 -0.022 -0.009

Z-scoring embeddings

Z-scoring embeddings led to significant image similarity even with AoA as a covariate in section 1 – todo to include section 2!

#main_image_effect_zscored <- lmer(scale(corrected_target_looking) ~ scale(image_sim_zscore)*scale(age_in_months)
#                    + scale(AoA_Est_target)
#                    + scale(MeanSaliencyDiff)
#                    + (1 | SubjectInfo.subjID) 
#                    + (1|Trials.targetImage)
#                    + ( 1 | section), 
#                    data = trials_with_effect_vars)
#summary(main_image_effect_zscored)

Baseline window as a covariate

baseline_covariate_looking_text <- lmer(scale(mean_target_looking_critical_window) ~ scale(age_in_months)*scale(text_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(mean_target_looking_baseline_window)
                    + (scale(text_similarity) || SubjectInfo.subjID) 
                    
                    + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)

baseline_covariate_looking_image <- lmer(scale(mean_target_looking_critical_window) ~ scale(age_in_months)*scale(image_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(mean_target_looking_baseline_window)
                    + (scale(image_similarity) || SubjectInfo.subjID) 
                    
                     + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)

summary(baseline_covariate_looking_text)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_critical_window) ~ scale(age_in_months) *  
    scale(text_similarity) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    scale(mean_target_looking_baseline_window) + (scale(text_similarity) ||  
    SubjectInfo.subjID) + (1 | Trials.imagePair) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 19523.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.89416 -0.70760  0.06795  0.74158  2.46232 

Random effects:
 Groups               Name                   Variance Std.Dev.
 SubjectInfo.subjID   scale(text_similarity) 0.010448 0.10221 
 SubjectInfo.subjID.1 (Intercept)            0.043751 0.20917 
 Trials.targetImage   (Intercept)            0.025709 0.16034 
 Trials.imagePair     (Intercept)            0.004319 0.06572 
 Residual                                    0.793779 0.89094 
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                 -3.991e-03  3.251e-02  5.933e+01
scale(age_in_months)                         9.788e-02  1.879e-02  1.853e+02
scale(text_similarity)                      -5.154e-02  2.116e-02  3.700e+01
scale(AoA_Est_target)                       -1.078e-01  2.696e-02  4.193e+01
scale(MeanSaliencyDiff)                      5.596e-02  2.186e-02  6.442e+01
scale(mean_target_looking_baseline_window)   2.844e-01  1.081e-02  7.248e+03
scale(age_in_months):scale(text_similarity) -1.338e-02  1.278e-02  1.567e+02
                                            t value Pr(>|t|)    
(Intercept)                                  -0.123 0.902708    
scale(age_in_months)                          5.208 5.06e-07 ***
scale(text_similarity)                       -2.436 0.019784 *  
scale(AoA_Est_target)                        -3.999 0.000253 ***
scale(MeanSaliencyDiff)                       2.559 0.012852 *  
scale(mean_target_looking_baseline_window)   26.305  < 2e-16 ***
scale(age_in_months):scale(text_similarity)  -1.046 0.296955    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) scl(_) s(AA_E s(MSD) s(____
scl(g_n_mn)  0.005                                   
scl(txt_sm)  0.040 -0.002                            
scl(AA_Es_)  0.070  0.002  0.050                     
scl(MnSlnD)  0.012 -0.001  0.052  0.033              
scl(mn____) -0.008 -0.006  0.008  0.011 -0.031       
scl(__):(_) -0.001  0.033  0.003 -0.001 -0.001  0.006

summary(baseline_covariate_looking_image)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_critical_window) ~ scale(age_in_months) *  
    scale(image_similarity) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    scale(mean_target_looking_baseline_window) + (scale(image_similarity) ||  
    SubjectInfo.subjID) + (1 | Trials.imagePair) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 19527

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.84668 -0.71660  0.06526  0.74741  2.50800 

Random effects:
 Groups               Name                    Variance Std.Dev.
 SubjectInfo.subjID   scale(image_similarity) 0.008000 0.08944 
 SubjectInfo.subjID.1 (Intercept)             0.043897 0.20952 
 Trials.targetImage   (Intercept)             0.026224 0.16194 
 Trials.imagePair     (Intercept)             0.006546 0.08091 
 Residual                                     0.795085 0.89168 
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                   9.786e-04  3.368e-02  5.774e+01
scale(age_in_months)                          9.684e-02  1.876e-02  1.855e+02
scale(image_similarity)                      -2.556e-02  2.262e-02  3.182e+01
scale(AoA_Est_target)                        -1.019e-01  2.768e-02  4.329e+01
scale(MeanSaliencyDiff)                       5.986e-02  2.234e-02  5.975e+01
scale(mean_target_looking_baseline_window)    2.836e-01  1.082e-02  7.253e+03
scale(age_in_months):scale(image_similarity) -2.078e-02  1.226e-02  1.925e+02
                                             t value Pr(>|t|)    
(Intercept)                                    0.029 0.976919    
scale(age_in_months)                           5.162 6.26e-07 ***
scale(image_similarity)                       -1.130 0.266863    
scale(AoA_Est_target)                         -3.683 0.000637 ***
scale(MeanSaliencyDiff)                        2.680 0.009507 ** 
scale(mean_target_looking_baseline_window)    26.215  < 2e-16 ***
scale(age_in_months):scale(image_similarity)  -1.695 0.091717 .  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) scl(_) s(AA_E s(MSD) s(____
scl(g_n_mn)  0.005                                   
scl(mg_sml) -0.021 -0.003                            
scl(AA_Es_)  0.067  0.003 -0.063                     
scl(MnSlnD)  0.009 -0.001  0.051  0.022              
scl(mn____) -0.009 -0.007 -0.001  0.010 -0.032       
scl(__):(_) -0.003 -0.008 -0.001  0.002 -0.002  0.004

We’re seeing some interesting effects. although these models are still singular fits.

Predicting critical window looking

critical_looking_text <- lmer(scale(mean_target_looking_critical_window) ~ scale(age_in_months)
                    + scale(text_similarity)
                    + scale(image_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair)
                    , 
                    data = trials_with_effect_vars)

critical_looking_image <- lmer(scale(mean_target_looking_critical_window) ~ scale(age_in_months)*scale(image_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(image_similarity) | SubjectInfo.subjID) 
                    
                     + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)

critical_looking_bvimage <- lmer(scale(mean_target_looking_critical_window) ~ scale(age_in_months)*scale(dinov3.babyview_image_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(dinov3.babyview_image_similarity) | SubjectInfo.subjID) 
                     + (1|Trials.imagePair)
                    + (1|Trials.targetImage), 
                    data = trials_with_effect_vars)

summary(critical_looking_text)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_critical_window) ~ scale(age_in_months) +  
    scale(text_similarity) + scale(image_similarity) + scale(AoA_Est_target) +  
    scale(MeanSaliencyDiff) + (scale(text_similarity) | SubjectInfo.subjID) +  
    (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20172.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.56376 -0.70034  0.07018  0.76896  2.57906 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.044814 0.21169       
                    scale(text_similarity) 0.009775 0.09887  0.11 
 Trials.targetImage (Intercept)            0.047526 0.21801       
 Trials.imagePair   (Intercept)            0.008202 0.09056       
 Residual                                  0.867418 0.93135       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                          Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)               0.002925   0.040795  53.604077   0.072  0.94311    
scale(age_in_months)      0.102716   0.019174 185.582385   5.357 2.49e-07 ***
scale(text_similarity)   -0.066431   0.031013  40.142055  -2.142  0.03831 *  
scale(image_similarity)   0.017433   0.030303  32.815614   0.575  0.56901    
scale(AoA_Est_target)    -0.115995   0.035417  43.198158  -3.275  0.00209 ** 
scale(MeanSaliencyDiff)   0.068726   0.027486  60.566323   2.500  0.01513 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) scl(t_) scl(m_) s(AA_E
scl(g_n_mn)  0.004                              
scl(txt_sm)  0.081  0.000                       
scl(mg_sml) -0.060 -0.002 -0.579                
scl(AA_Es_)  0.082  0.002  0.109  -0.107        
scl(MnSlnD)  0.009 -0.001  0.026   0.056   0.029

summary(critical_looking_image)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_critical_window) ~ scale(age_in_months) *  
    scale(image_similarity) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(image_similarity) | SubjectInfo.subjID) + (1 | Trials.imagePair) +  
    (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20171.3

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.52800 -0.70337  0.06512  0.77031  2.51122 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.04494  0.2120        
                    scale(image_similarity) 0.00895  0.0946   0.18 
 Trials.targetImage (Intercept)             0.04679  0.2163        
 Trials.imagePair   (Intercept)             0.01132  0.1064        
 Residual                                   0.86713  0.9312        
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                              Estimate Std. Error        df
(Intercept)                                    0.01012    0.04166  53.99728
scale(age_in_months)                           0.10005    0.01918 186.35864
scale(image_similarity)                       -0.02251    0.02731  30.47090
scale(AoA_Est_target)                         -0.10777    0.03548  44.27658
scale(MeanSaliencyDiff)                        0.07429    0.02782  58.63251
scale(age_in_months):scale(image_similarity)  -0.02184    0.01284 190.52108
                                             t value Pr(>|t|)    
(Intercept)                                    0.243  0.80891    
scale(age_in_months)                           5.217 4.83e-07 ***
scale(image_similarity)                       -0.824  0.41632    
scale(AoA_Est_target)                         -3.037  0.00399 ** 
scale(MeanSaliencyDiff)                        2.670  0.00980 ** 
scale(age_in_months):scale(image_similarity)  -1.700  0.09069 .  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) scl(_) s(AA_E s(MSD)
scl(g_n_mn)  0.003                            
scl(mg_sml) -0.003 -0.002                     
scl(AA_Es_)  0.072  0.002 -0.045              
scl(MnSlnD)  0.008 -0.001  0.072  0.023       
scl(__):(_) -0.002  0.071 -0.001  0.002 -0.002

summary(critical_looking_bvimage)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_critical_window) ~ scale(age_in_months) *  
    scale(dinov3.babyview_image_similarity) + scale(AoA_Est_target) +  
    scale(MeanSaliencyDiff) + (scale(dinov3.babyview_image_similarity) |  
    SubjectInfo.subjID) + (1 | Trials.imagePair) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20177.9

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.44575 -0.71606  0.07505  0.76890  2.36372 

Random effects:
 Groups             Name                                    Variance Std.Dev.
 SubjectInfo.subjID (Intercept)                             0.044373 0.21065 
                    scale(dinov3.babyview_image_similarity) 0.006814 0.08255 
 Trials.targetImage (Intercept)                             0.047823 0.21868 
 Trials.imagePair   (Intercept)                             0.011666 0.10801 
 Residual                                                   0.869645 0.93255 
 Corr 
      
 0.24 
      
      
      
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                               Estimate
(Intercept)                                                    0.009560
scale(age_in_months)                                           0.100380
scale(dinov3.babyview_image_similarity)                        0.002956
scale(AoA_Est_target)                                         -0.109151
scale(MeanSaliencyDiff)                                        0.076334
scale(age_in_months):scale(dinov3.babyview_image_similarity)  -0.014967
                                                             Std. Error
(Intercept)                                                    0.042043
scale(age_in_months)                                           0.019120
scale(dinov3.babyview_image_similarity)                        0.031604
scale(AoA_Est_target)                                          0.036147
scale(MeanSaliencyDiff)                                        0.028118
scale(age_in_months):scale(dinov3.babyview_image_similarity)   0.012677
                                                                     df t value
(Intercept)                                                   54.692369   0.227
scale(age_in_months)                                         186.866177   5.250
scale(dinov3.babyview_image_similarity)                       36.987059   0.094
scale(AoA_Est_target)                                         45.395663  -3.020
scale(MeanSaliencyDiff)                                       59.478842   2.715
scale(age_in_months):scale(dinov3.babyview_image_similarity) 203.969559  -1.181
                                                             Pr(>|t|)    
(Intercept)                                                   0.82097    
scale(age_in_months)                                         4.11e-07 ***
scale(dinov3.babyview_image_similarity)                       0.92598    
scale(AoA_Est_target)                                         0.00414 ** 
scale(MeanSaliencyDiff)                                       0.00867 ** 
scale(age_in_months):scale(dinov3.babyview_image_similarity)  0.23913    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) s(3.__ s(AA_E s(MSD)
scl(g_n_mn)  0.003                            
scl(dn3.__) -0.022  0.001                     
scl(AA_Es_)  0.076  0.002 -0.137              
scl(MnSlnD)  0.005 -0.001  0.094  0.013       
s(__):(3.__  0.003  0.062  0.001  0.000  0.003

r.squaredGLMM(critical_looking_image)

            R2m      R2c
[1,] 0.02881404 0.139908

r.squaredGLMM(critical_looking_text)

            R2m       R2c
[1,] 0.03137678 0.1406644

Similar predictions to our original model.

Adding window type as a covariate

trials_window_type_separated <- trials_with_effect_vars |>
  pivot_longer(cols=c(mean_target_looking_critical_window, mean_target_looking_baseline_window), names_to="window_type", values_to="target_looking") |>
  mutate(window_type = str_replace(window_type, "mean_target_looking_", "")) |>
  mutate(trial_window_c = case_when(
    window_type=="critical_window" ~ 0.5,
    window_type=="baseline_window" ~ -0.5))

window_type_looking_text <- lmer(scale(target_looking) ~ scale(age_in_months) + trial_window_c*scale(text_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID) 
                    + (1|Trials.imagePair)
                    + (1|Trials.targetImage) , 
                    data = trials_window_type_separated)

window_type_looking_image <- lmer(scale(target_looking) ~ scale(age_in_months) + trial_window_c*scale(image_similarity)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (scale(image_similarity) | SubjectInfo.subjID) 
                    + (1 | Trials.imagePair)
                    + (1|Trials.targetImage) , 
                    data = trials_window_type_separated)

summary(window_type_looking_image)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(target_looking) ~ scale(age_in_months) + trial_window_c *  
    scale(image_similarity) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(image_similarity) | SubjectInfo.subjID) + (1 | Trials.imagePair) +  
    (1 | Trials.targetImage)
   Data: trials_window_type_separated

REML criterion at convergence: 40691.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.50970 -0.72798  0.03098  0.76572  2.24146 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.017409 0.13194       
                    scale(image_similarity) 0.007153 0.08457  0.05 
 Trials.targetImage (Intercept)             0.049790 0.22314       
 Trials.imagePair   (Intercept)             0.012356 0.11116       
 Residual                                   0.910677 0.95429       
Number of obs: 14666, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                         Estimate Std. Error         df t value
(Intercept)                             2.177e-02  4.011e-02  5.136e+01   0.543
scale(age_in_months)                    5.565e-02  1.265e-02  1.877e+02   4.401
trial_window_c                          1.848e-01  1.576e-02  1.426e+04  11.728
scale(image_similarity)                -9.152e-03  2.589e-02  2.975e+01  -0.353
scale(AoA_Est_target)                  -6.568e-02  3.533e-02  4.600e+01  -1.859
scale(MeanSaliencyDiff)                 6.152e-02  2.682e-02  5.733e+01   2.294
trial_window_c:scale(image_similarity) -5.044e-02  1.576e-02  1.426e+04  -3.200
                                       Pr(>|t|)    
(Intercept)                             0.58960    
scale(age_in_months)                   1.81e-05 ***
trial_window_c                          < 2e-16 ***
scale(image_similarity)                 0.72626    
scale(AoA_Est_target)                   0.06941 .  
scale(MeanSaliencyDiff)                 0.02550 *  
trial_window_c:scale(image_similarity)  0.00138 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) trl_w_ scl(_) s(AA_E s(MSD)
scl(g_n_mn)  0.002                                   
tril_wndw_c  0.000  0.000                            
scl(mg_sml) -0.018 -0.002  0.000                     
scl(AA_Es_)  0.076  0.002  0.000 -0.022              
scl(MnSlnD)  0.007 -0.001  0.000  0.093  0.024       
trl_wn_:(_)  0.000  0.000  0.000  0.000  0.000  0.000

summary(window_type_looking_text)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(target_looking) ~ scale(age_in_months) + trial_window_c *  
    scale(text_similarity) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.imagePair) +  
    (1 | Trials.targetImage)
   Data: trials_window_type_separated

REML criterion at convergence: 40702.3

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.48354 -0.72790  0.03524  0.76569  2.24071 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.017660 0.13289       
                    scale(text_similarity) 0.007139 0.08449  0.05 
 Trials.targetImage (Intercept)            0.049700 0.22293       
 Trials.imagePair   (Intercept)            0.010621 0.10306       
 Residual                                  0.911536 0.95474       
Number of obs: 14666, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                        Estimate Std. Error         df t value
(Intercept)                            1.758e-02  3.949e-02  5.151e+01   0.445
scale(age_in_months)                   5.589e-02  1.274e-02  1.865e+02   4.387
trial_window_c                         1.848e-01  1.577e-02  1.420e+04  11.722
scale(text_similarity)                -3.775e-02  2.441e-02  3.156e+01  -1.546
scale(AoA_Est_target)                 -7.005e-02  3.513e-02  4.536e+01  -1.994
scale(MeanSaliencyDiff)                5.693e-02  2.643e-02  5.875e+01   2.154
trial_window_c:scale(text_similarity) -3.617e-02  1.577e-02  1.420e+04  -2.294
                                      Pr(>|t|)    
(Intercept)                             0.6580    
scale(age_in_months)                  1.92e-05 ***
trial_window_c                         < 2e-16 ***
scale(text_similarity)                  0.1320    
scale(AoA_Est_target)                   0.0522 .  
scale(MeanSaliencyDiff)                 0.0354 *  
trial_window_c:scale(text_similarity)   0.0218 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) sc(__) trl_w_ scl(_) s(AA_E s(MSD)
scl(g_n_mn)  0.002                                   
tril_wndw_c  0.000  0.000                            
scl(txt_sm)  0.052 -0.002  0.000                     
scl(AA_Es_)  0.081  0.001  0.000  0.069              
scl(MnSlnD)  0.013 -0.001  0.000  0.084  0.034       
trl_wn_:(_)  0.000  0.000  0.000  0.000  0.000  0.000

The interaction makes sense, plotting the predictions here to try to understand what’s going on

trials_window_type_separated$predicted <- predict(window_type_looking_image)

# Plot interaction effect
ggplot(trials_window_type_separated, aes(x = image_similarity, y = predicted, color = factor(window_type))) +
  geom_point(alpha = 0.5) +   # Add points for raw data
  geom_smooth(method = "lm", se = TRUE) +  # Add regression lines
  labs(title = "Interaction Between Trial Window & Image Similarity",
       x = "Scaled Image Similarity",
       y = "Predicted Target Looking",
       color = "Trial Window") +
  theme_minimal()

`geom_smooth()` using formula = 'y ~ x'

This is also making me wonder whether we see any signal in how infants look at images in the baseline window. ## Baseline window looking

baseline_looking_image <- lmer(scale(mean_target_looking_baseline_window) ~ scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(image_similarity)
                    + (scale(MeanSaliencyDiff) | SubjectInfo.subjID)  + (1 | Trials.targetImage) + (1 | Trials.imagePair), 
                    data = trials_with_effect_vars)
summary(baseline_looking_image)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(mean_target_looking_baseline_window) ~ scale(AoA_Est_target) +  
    scale(MeanSaliencyDiff) + scale(image_similarity) + (scale(MeanSaliencyDiff) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20528.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.20883 -0.72786 -0.00671  0.72745  2.14487 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr  
 SubjectInfo.subjID (Intercept)             0.008954 0.09463        
                    scale(MeanSaliencyDiff) 0.021476 0.14655  -0.15 
 Trials.targetImage (Intercept)             0.051543 0.22703        
 Trials.imagePair   (Intercept)             0.011403 0.10678        
 Residual                                   0.922124 0.96027        
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                         Estimate Std. Error        df t value Pr(>|t|)  
(Intercept)              0.029990   0.040578 39.936899   0.739   0.4642  
scale(AoA_Est_target)   -0.026152   0.036875 41.314545  -0.709   0.4822  
scale(MeanSaliencyDiff)  0.063010   0.031011 71.724224   2.032   0.0459 *
scale(image_similarity)  0.002259   0.026862 26.258803   0.084   0.9336  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) s(AA_E s(MSD)
scl(AA_Es_)  0.076              
scl(MnSlnD) -0.002  0.024       
scl(mg_sml) -0.023 -0.045  0.073

Can’t get this to not be not singular but still fun to see that saliency is predictive.

Using a shorter critical window

short_image_effect <- lmer(scale(corrected_target_looking_short) ~ scale(image_similarity)*scale(age_in_months)
                    + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + (1 | SubjectInfo.subjID) 
                    + (1|Trials.targetImage)
                    , 
                    data = trials_with_effect_vars)

short_text_effect <- lmer(scale(corrected_target_looking_short) ~ scale(text_similarity)*scale(age_in_months)
                    + scale(AoA_Est_target)
                    + (1 | SubjectInfo.subjID)
                    + (1 | Trials.targetImage)
                    , 
                    data = trials_with_effect_vars)
summary(short_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(corrected_target_looking_short) ~ scale(text_similarity) *  
    scale(age_in_months) + scale(AoA_Est_target) + (1 | SubjectInfo.subjID) +  
    (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20615.4

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.91257 -0.66352 -0.00421  0.67659  2.82547 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.031824 0.17839 
 Trials.targetImage (Intercept) 0.008876 0.09421 
 Residual                       0.945244 0.97224 
Number of obs: 7333, groups:  SubjectInfo.subjID, 181; Trials.targetImage, 48

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                 -1.690e-02  2.270e-02  9.584e+01
scale(text_similarity)                      -3.000e-02  1.439e-02  2.659e+02
scale(age_in_months)                         8.272e-02  1.757e-02  1.820e+02
scale(AoA_Est_target)                       -7.928e-02  1.812e-02  5.255e+01
scale(text_similarity):scale(age_in_months)  4.693e-03  1.123e-02  7.301e+03
                                            t value Pr(>|t|)    
(Intercept)                                  -0.745    0.458    
scale(text_similarity)                       -2.085    0.038 *  
scale(age_in_months)                          4.708 4.94e-06 ***
scale(AoA_Est_target)                        -4.374 5.80e-05 ***
scale(text_similarity):scale(age_in_months)   0.418    0.676    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E
scl(txt_sm)  0.039                     
scl(g_n_mn)  0.005 -0.003              
scl(AA_Es_)  0.054  0.022  0.005       
scl(_):(__) -0.002 -0.002  0.048 -0.003

summary(short_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(corrected_target_looking_short) ~ scale(image_similarity) *  
    scale(age_in_months) + scale(AoA_Est_target) + scale(MeanSaliencyDiff) +  
    (1 | SubjectInfo.subjID) + (1 | Trials.targetImage)
   Data: trials_with_effect_vars

REML criterion at convergence: 20619.8

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.87755 -0.65901 -0.00658  0.67649  2.82919 

Random effects:
 Groups             Name        Variance Std.Dev.
 SubjectInfo.subjID (Intercept) 0.03178  0.17828 
 Trials.targetImage (Intercept) 0.00875  0.09354 
 Residual                       0.94518  0.97220 
Number of obs: 7333, groups:  SubjectInfo.subjID, 181; Trials.targetImage, 48

Fixed effects:
                                               Estimate Std. Error         df
(Intercept)                                    -0.01375    0.02263   92.05088
scale(image_similarity)                        -0.02463    0.01439  275.41749
scale(age_in_months)                            0.08257    0.01754  182.15059
scale(AoA_Est_target)                          -0.07315    0.01826   51.14233
scale(MeanSaliencyDiff)                         0.01971    0.01660   82.19682
scale(image_similarity):scale(age_in_months)   -0.01630    0.01122 7178.23961
                                             t value Pr(>|t|)    
(Intercept)                                   -0.608 0.544762    
scale(image_similarity)                       -1.712 0.088069 .  
scale(age_in_months)                           4.706 4.98e-06 ***
scale(AoA_Est_target)                         -4.006 0.000201 ***
scale(MeanSaliencyDiff)                        1.187 0.238595    
scale(image_similarity):scale(age_in_months)  -1.453 0.146271    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(_) sc(__) s(AA_E s(MSD)
scl(mg_sml) -0.020                            
scl(g_n_mn)  0.005 -0.006                     
scl(AA_Es_)  0.056 -0.149  0.006              
scl(MnSlnD)  0.010  0.019 -0.004  0.030       
scl(_):(__) -0.004 -0.006 -0.008  0.004 -0.003

Our effects are not significant with the shorter window.

AoA analyses

aoa_text_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) * scale(AoA_Est_target)
                    + scale(aoa_difference)
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)

summary(aoa_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) *  
    scale(AoA_Est_target) + scale(aoa_difference) + scale(MeanSaliencyDiff) +  
    (scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20600.1

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.95203 -0.63120 -0.03283  0.66917  2.99048 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.034329 0.18528       
                    scale(text_similarity) 0.010346 0.10171  0.08 
 Trials.targetImage (Intercept)            0.010004 0.10002       
 Trials.imagePair   (Intercept)            0.001607 0.04009       
 Residual                                  0.929165 0.96393       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                                    Estimate
(Intercept)                                                       -1.625e-02
scale(text_similarity)                                            -3.817e-02
scale(age_in_months)                                               7.238e-02
scale(AoA_Est_target)                                             -9.261e-02
scale(aoa_difference)                                              2.506e-02
scale(MeanSaliencyDiff)                                            1.035e-02
scale(text_similarity):scale(age_in_months)                       -9.491e-03
scale(text_similarity):scale(AoA_Est_target)                       3.305e-02
scale(age_in_months):scale(AoA_Est_target)                        -4.970e-02
scale(text_similarity):scale(age_in_months):scale(AoA_Est_target)  1.537e-02
                                                                  Std. Error
(Intercept)                                                        2.463e-02
scale(text_similarity)                                             1.808e-02
scale(age_in_months)                                               1.801e-02
scale(AoA_Est_target)                                              2.422e-02
scale(aoa_difference)                                              2.134e-02
scale(MeanSaliencyDiff)                                            1.735e-02
scale(text_similarity):scale(age_in_months)                        1.344e-02
scale(text_similarity):scale(AoA_Est_target)                       1.576e-02
scale(age_in_months):scale(AoA_Est_target)                         1.161e-02
scale(text_similarity):scale(age_in_months):scale(AoA_Est_target)  1.087e-02
                                                                          df
(Intercept)                                                        6.590e+01
scale(text_similarity)                                             4.160e+01
scale(age_in_months)                                               1.849e+02
scale(AoA_Est_target)                                              3.589e+01
scale(aoa_difference)                                              7.362e+01
scale(MeanSaliencyDiff)                                            6.459e+01
scale(text_similarity):scale(age_in_months)                        1.605e+02
scale(text_similarity):scale(AoA_Est_target)                       7.028e+01
scale(age_in_months):scale(AoA_Est_target)                         6.872e+03
scale(text_similarity):scale(age_in_months):scale(AoA_Est_target)  7.178e+03
                                                                  t value
(Intercept)                                                        -0.660
scale(text_similarity)                                             -2.112
scale(age_in_months)                                                4.020
scale(AoA_Est_target)                                              -3.824
scale(aoa_difference)                                               1.174
scale(MeanSaliencyDiff)                                             0.597
scale(text_similarity):scale(age_in_months)                        -0.706
scale(text_similarity):scale(AoA_Est_target)                        2.097
scale(age_in_months):scale(AoA_Est_target)                         -4.279
scale(text_similarity):scale(age_in_months):scale(AoA_Est_target)   1.414
                                                                  Pr(>|t|)    
(Intercept)                                                       0.511710    
scale(text_similarity)                                            0.040737 *  
scale(age_in_months)                                              8.48e-05 ***
scale(AoA_Est_target)                                             0.000504 ***
scale(aoa_difference)                                             0.244019    
scale(MeanSaliencyDiff)                                           0.552894    
scale(text_similarity):scale(age_in_months)                       0.481246    
scale(text_similarity):scale(AoA_Est_target)                      0.039616 *  
scale(age_in_months):scale(AoA_Est_target)                        1.90e-05 ***
scale(text_similarity):scale(age_in_months):scale(AoA_Est_target) 0.157287    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(t_) sc(__) s(AA_E scl(_d) s(MSD) sc(_):(__) s(_):(A
scl(txt_sm)  0.048                                                        
scl(g_n_mn)  0.005 -0.002                                                 
scl(AA_Es_)  0.071 -0.028   0.005                                         
scl(_dffrn) -0.036  0.073  -0.003 -0.588                                  
scl(MnSlnD)  0.014 -0.002  -0.004  0.075 -0.079                           
scl(_):(__) -0.001  0.002   0.072 -0.003  0.004  -0.002                   
s(_):(AA_E_  0.009 -0.093  -0.004  0.011  0.119   0.088  0.006            
s(__):(AA_E  0.005 -0.003   0.069  0.000  0.005  -0.006  0.046     -0.005 
s(_):(__):( -0.002  0.005   0.043  0.001 -0.005  -0.002  0.017      0.001 
            s(__):
scl(txt_sm)       
scl(g_n_mn)       
scl(AA_Es_)       
scl(_dffrn)       
scl(MnSlnD)       
scl(_):(__)       
s(_):(AA_E_       
s(__):(AA_E       
s(_):(__):(  0.113

#Swapping text similarity with image similarity:
aoa_image_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) * scale(AoA_Est_target)
                      + scale(aoa_difference)
                    + scale(MeanSaliencyDiff)
                    + (scale(image_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)
summary(aoa_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) *  
    scale(AoA_Est_target) + scale(aoa_difference) + scale(MeanSaliencyDiff) +  
    (scale(image_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20610.4

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.99700 -0.63956 -0.03454  0.67244  2.91689 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.034283 0.18516       
                    scale(image_similarity) 0.006627 0.08141  0.28 
 Trials.targetImage (Intercept)             0.009989 0.09995       
 Trials.imagePair   (Intercept)             0.002562 0.05062       
 Residual                                   0.932965 0.96590       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                                     Estimate
(Intercept)                                                        -1.514e-02
scale(image_similarity)                                            -2.216e-02
scale(age_in_months)                                                6.884e-02
scale(AoA_Est_target)                                              -8.607e-02
scale(aoa_difference)                                               1.597e-02
scale(MeanSaliencyDiff)                                             8.576e-03
scale(image_similarity):scale(age_in_months)                       -3.462e-03
scale(image_similarity):scale(AoA_Est_target)                       6.786e-03
scale(age_in_months):scale(AoA_Est_target)                         -5.091e-02
scale(image_similarity):scale(age_in_months):scale(AoA_Est_target)  1.325e-02
                                                                   Std. Error
(Intercept)                                                         2.543e-02
scale(image_similarity)                                             1.976e-02
scale(age_in_months)                                                1.797e-02
scale(AoA_Est_target)                                               2.641e-02
scale(aoa_difference)                                               2.294e-02
scale(MeanSaliencyDiff)                                             1.736e-02
scale(image_similarity):scale(age_in_months)                        1.342e-02
scale(image_similarity):scale(AoA_Est_target)                       1.795e-02
scale(age_in_months):scale(AoA_Est_target)                          1.170e-02
scale(image_similarity):scale(age_in_months):scale(AoA_Est_target)  1.201e-02
                                                                           df
(Intercept)                                                         5.697e+01
scale(image_similarity)                                             3.176e+01
scale(age_in_months)                                                1.892e+02
scale(AoA_Est_target)                                               3.440e+01
scale(aoa_difference)                                               6.387e+01
scale(MeanSaliencyDiff)                                             5.756e+01
scale(image_similarity):scale(age_in_months)                        2.234e+02
scale(image_similarity):scale(AoA_Est_target)                       6.061e+01
scale(age_in_months):scale(AoA_Est_target)                          7.135e+03
scale(image_similarity):scale(age_in_months):scale(AoA_Est_target)  5.064e+03
                                                                   t value
(Intercept)                                                         -0.595
scale(image_similarity)                                             -1.121
scale(age_in_months)                                                 3.830
scale(AoA_Est_target)                                               -3.259
scale(aoa_difference)                                                0.696
scale(MeanSaliencyDiff)                                              0.494
scale(image_similarity):scale(age_in_months)                        -0.258
scale(image_similarity):scale(AoA_Est_target)                        0.378
scale(age_in_months):scale(AoA_Est_target)                          -4.353
scale(image_similarity):scale(age_in_months):scale(AoA_Est_target)   1.103
                                                                   Pr(>|t|)    
(Intercept)                                                        0.554051    
scale(image_similarity)                                            0.270518    
scale(age_in_months)                                               0.000174 ***
scale(AoA_Est_target)                                              0.002519 ** 
scale(aoa_difference)                                              0.489010    
scale(MeanSaliencyDiff)                                            0.623179    
scale(image_similarity):scale(age_in_months)                       0.796696    
scale(image_similarity):scale(AoA_Est_target)                      0.706664    
scale(age_in_months):scale(AoA_Est_target)                         1.36e-05 ***
scale(image_similarity):scale(age_in_months):scale(AoA_Est_target) 0.270017    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(m_) sc(__) s(AA_E scl(_d) s(MSD) sc(_):(__) s(_):(A
scl(mg_sml) -0.025                                                        
scl(g_n_mn)  0.005 -0.005                                                 
scl(AA_Es_)  0.096 -0.267   0.007                                         
scl(_dffrn) -0.072  0.275  -0.004 -0.654                                  
scl(MnSlnD)  0.014 -0.013  -0.002  0.072 -0.089                           
scl(_):(__) -0.003  0.001   0.055  0.001 -0.002  -0.004                   
s(_):(AA_E_ -0.144  0.316  -0.002 -0.183  0.216  -0.009  0.006            
s(__):(AA_E  0.007 -0.002   0.062  0.002  0.004  -0.004 -0.147     -0.009 
s(_):(__):(  0.000  0.007  -0.088 -0.005  0.000  -0.007  0.304      0.005 
            s(__):
scl(mg_sml)       
scl(g_n_mn)       
scl(AA_Es_)       
scl(_dffrn)       
scl(MnSlnD)       
scl(_):(__)       
s(_):(AA_E_       
s(__):(AA_E       
s(_):(__):( -0.081

aoa_text_diff_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) * scale(aoa_difference)
                    + scale(MeanSaliencyDiff)
                    + (scale(text_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)

summary(aoa_text_diff_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) *  
    scale(aoa_difference) + scale(MeanSaliencyDiff) + (scale(text_similarity) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20619.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.01267 -0.63083 -0.02533  0.67596  3.03472 

Random effects:
 Groups             Name                   Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.034631 0.1861        
                    scale(text_similarity) 0.010994 0.1049   0.09 
 Trials.targetImage (Intercept)            0.010759 0.1037        
 Trials.imagePair   (Intercept)            0.005257 0.0725        
 Residual                                  0.930371 0.9646        
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                                    Estimate
(Intercept)                                                       -1.175e-02
scale(text_similarity)                                            -3.759e-02
scale(age_in_months)                                               7.672e-02
scale(aoa_difference)                                             -2.404e-02
scale(MeanSaliencyDiff)                                            1.714e-02
scale(text_similarity):scale(age_in_months)                       -7.080e-03
scale(text_similarity):scale(aoa_difference)                       3.935e-02
scale(age_in_months):scale(aoa_difference)                        -3.038e-02
scale(text_similarity):scale(age_in_months):scale(aoa_difference)  7.406e-03
                                                                  Std. Error
(Intercept)                                                        2.716e-02
scale(text_similarity)                                             2.101e-02
scale(age_in_months)                                               1.801e-02
scale(aoa_difference)                                              1.819e-02
scale(MeanSaliencyDiff)                                            1.789e-02
scale(text_similarity):scale(age_in_months)                        1.357e-02
scale(text_similarity):scale(aoa_difference)                       1.750e-02
scale(age_in_months):scale(aoa_difference)                         1.182e-02
scale(text_similarity):scale(age_in_months):scale(aoa_difference)  1.184e-02
                                                                          df
(Intercept)                                                        6.196e+01
scale(text_similarity)                                             3.995e+01
scale(age_in_months)                                               1.843e+02
scale(aoa_difference)                                              4.984e+01
scale(MeanSaliencyDiff)                                            4.727e+01
scale(text_similarity):scale(age_in_months)                        1.594e+02
scale(text_similarity):scale(aoa_difference)                       6.466e+01
scale(age_in_months):scale(aoa_difference)                         7.015e+03
scale(text_similarity):scale(age_in_months):scale(aoa_difference)  7.077e+03
                                                                  t value
(Intercept)                                                        -0.433
scale(text_similarity)                                             -1.789
scale(age_in_months)                                                4.260
scale(aoa_difference)                                              -1.322
scale(MeanSaliencyDiff)                                             0.958
scale(text_similarity):scale(age_in_months)                        -0.522
scale(text_similarity):scale(aoa_difference)                        2.249
scale(age_in_months):scale(aoa_difference)                         -2.571
scale(text_similarity):scale(age_in_months):scale(aoa_difference)   0.625
                                                                  Pr(>|t|)    
(Intercept)                                                         0.6667    
scale(text_similarity)                                              0.0811 .  
scale(age_in_months)                                              3.26e-05 ***
scale(aoa_difference)                                               0.1923    
scale(MeanSaliencyDiff)                                             0.3429    
scale(text_similarity):scale(age_in_months)                         0.6025    
scale(text_similarity):scale(aoa_difference)                        0.0279 *  
scale(age_in_months):scale(aoa_difference)                          0.0102 *  
scale(text_similarity):scale(age_in_months):scale(aoa_difference)   0.5318    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(t_) sc(__) scl(_d) s(MSD) sc(_):(__) s(_):(_) s(__):
scl(txt_sm)  0.058                                                         
scl(g_n_mn)  0.004 -0.001                                                  
scl(_dffrn) -0.002  0.052   0.000                                          
scl(MnSlnD)  0.003 -0.005  -0.003 -0.026                                   
scl(_):(__) -0.001  0.002   0.074  0.001  -0.002                           
scl(t_):(_) -0.034 -0.063   0.002  0.244   0.116  0.000                    
scl(__):(_)  0.001  0.001   0.003  0.007  -0.002 -0.004     -0.004         
s(_):(__):(  0.003 -0.001  -0.005 -0.005   0.002  0.012      0.003    0.186

aoa_image_diff_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) * scale(aoa_difference)
                    + scale(MeanSaliencyDiff)
                    + (scale(image_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)
summary(aoa_image_diff_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) * scale(age_in_months) *  
    scale(aoa_difference) + scale(MeanSaliencyDiff) + (scale(image_similarity) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 20620.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.04098 -0.63664 -0.03243  0.67447  2.92277 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.034722 0.18634       
                    scale(image_similarity) 0.006824 0.08261  0.31 
 Trials.targetImage (Intercept)             0.010751 0.10369       
 Trials.imagePair   (Intercept)             0.004898 0.06998       
 Residual                                   0.933631 0.96625       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                                                     Estimate
(Intercept)                                                        -1.128e-02
scale(image_similarity)                                            -3.636e-02
scale(age_in_months)                                                7.493e-02
scale(aoa_difference)                                              -3.080e-02
scale(MeanSaliencyDiff)                                             1.417e-02
scale(image_similarity):scale(age_in_months)                       -1.474e-02
scale(image_similarity):scale(aoa_difference)                       4.446e-02
scale(age_in_months):scale(aoa_difference)                         -3.104e-02
scale(image_similarity):scale(age_in_months):scale(aoa_difference)  6.985e-03
                                                                   Std. Error
(Intercept)                                                         2.692e-02
scale(image_similarity)                                             2.034e-02
scale(age_in_months)                                                1.794e-02
scale(aoa_difference)                                               1.775e-02
scale(MeanSaliencyDiff)                                             1.777e-02
scale(image_similarity):scale(age_in_months)                        1.273e-02
scale(image_similarity):scale(aoa_difference)                       1.836e-02
scale(age_in_months):scale(aoa_difference)                          1.165e-02
scale(image_similarity):scale(age_in_months):scale(aoa_difference)  1.227e-02
                                                                           df
(Intercept)                                                         6.144e+01
scale(image_similarity)                                             3.261e+01
scale(age_in_months)                                                1.852e+02
scale(aoa_difference)                                               5.252e+01
scale(MeanSaliencyDiff)                                             4.697e+01
scale(image_similarity):scale(age_in_months)                        1.919e+02
scale(image_similarity):scale(aoa_difference)                       6.242e+01
scale(age_in_months):scale(aoa_difference)                          7.033e+03
scale(image_similarity):scale(age_in_months):scale(aoa_difference)  7.097e+03
                                                                   t value
(Intercept)                                                         -0.419
scale(image_similarity)                                             -1.788
scale(age_in_months)                                                 4.175
scale(aoa_difference)                                               -1.735
scale(MeanSaliencyDiff)                                              0.798
scale(image_similarity):scale(age_in_months)                        -1.158
scale(image_similarity):scale(aoa_difference)                        2.421
scale(age_in_months):scale(aoa_difference)                          -2.664
scale(image_similarity):scale(age_in_months):scale(aoa_difference)   0.569
                                                                   Pr(>|t|)    
(Intercept)                                                         0.67666    
scale(image_similarity)                                             0.08309 .  
scale(age_in_months)                                               4.57e-05 ***
scale(aoa_difference)                                               0.08854 .  
scale(MeanSaliencyDiff)                                             0.42906    
scale(image_similarity):scale(age_in_months)                        0.24836    
scale(image_similarity):scale(aoa_difference)                       0.01839 *  
scale(age_in_months):scale(aoa_difference)                          0.00773 ** 
scale(image_similarity):scale(age_in_months):scale(aoa_difference)  0.56925    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(m_) sc(__) scl(_d) s(MSD) sc(_):(__) s(_):(_) s(__):
scl(mg_sml)  0.033                                                         
scl(g_n_mn)  0.004 -0.003                                                  
scl(_dffrn) -0.002  0.097   0.000                                          
scl(MnSlnD)  0.006  0.002  -0.003 -0.054                                   
scl(_):(__) -0.002 -0.002   0.106  0.000  -0.003                           
scl(m_):(_) -0.058  0.007   0.001  0.093   0.027 -0.002                    
scl(__):(_)  0.000 -0.001   0.004  0.009  -0.002 -0.001     -0.006         
s(_):(__):(  0.000 -0.001  -0.003 -0.007  -0.002  0.015      0.006    0.067

CDI analyses

cdi_text_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) + scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(percentile_rank)
                    + (scale(text_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)

boundary (singular) fit: see help('isSingular')

summary(cdi_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) + scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + scale(percentile_rank) +  
    (scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 3468.6

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.09049 -0.65526 -0.00956  0.69160  2.88838 

Random effects:
 Groups             Name                   Variance  Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)            0.0193807 0.13921       
                    scale(text_similarity) 0.0009772 0.03126  1.00 
 Trials.targetImage (Intercept)            0.0000000 0.00000       
 Trials.imagePair   (Intercept)            0.0000000 0.00000       
 Residual                                  0.8423642 0.91780       
Number of obs: 1283, groups:  
SubjectInfo.subjID, 48; Trials.targetImage, 24; Trials.imagePair, 16

Fixed effects:
                          Estimate Std. Error         df t value Pr(>|t|)   
(Intercept)               -0.03466    0.03522   54.85136  -0.984  0.32930   
scale(text_similarity)    -0.01877    0.02141  255.52724  -0.877  0.38131   
scale(age_in_months)       0.08995    0.03331   49.22744   2.701  0.00947 **
scale(AoA_Est_target)     -0.06465    0.02781 1238.92017  -2.325  0.02024 * 
scale(MeanSaliencyDiff)    0.03380    0.02538 1240.73353   1.331  0.18327   
scale(percentile_rank)     0.08534    0.03225   48.25058   2.646  0.01095 * 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(t_) sc(__) s(AA_E s(MSD)
scl(txt_sm)  0.382                             
scl(g_n_mn)  0.019  0.001                      
scl(AA_Es_)  0.325  0.264   0.000              
scl(MnSlnD) -0.007 -0.019   0.011 -0.039       
scl(prcnt_)  0.012  0.012  -0.217  0.009  0.010
optimizer (nloptwrap) convergence code: 0 (OK)
boundary (singular) fit: see help('isSingular')

#Swapping text similarity with image similarity:
cdi_image_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity) + scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(percentile_rank)
                    + (scale(image_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_effect_vars)

boundary (singular) fit: see help('isSingular')

summary(cdi_image_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(image_similarity) + scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + scale(percentile_rank) +  
    (scale(image_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_effect_vars

REML criterion at convergence: 3467.4

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-3.10295 -0.65389  0.00107  0.70012  2.87627 

Random effects:
 Groups             Name                    Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept)             0.015747 0.12549       
                    scale(image_similarity) 0.002819 0.05309  0.74 
 Trials.targetImage (Intercept)             0.000000 0.00000       
 Trials.imagePair   (Intercept)             0.000000 0.00000       
 Residual                                   0.840163 0.91660       
Number of obs: 1283, groups:  
SubjectInfo.subjID, 48; Trials.targetImage, 24; Trials.imagePair, 16

Fixed effects:
                          Estimate Std. Error         df t value Pr(>|t|)  
(Intercept)               -0.02365    0.03284   51.96532  -0.720   0.4747  
scale(image_similarity)   -0.03231    0.02448   45.28833  -1.320   0.1935  
scale(age_in_months)       0.08869    0.03314   45.44219   2.676   0.0103 *
scale(AoA_Est_target)     -0.05551    0.02678 1200.60051  -2.073   0.0384 *
scale(MeanSaliencyDiff)    0.03434    0.02537 1211.36579   1.353   0.1762  
scale(percentile_rank)     0.08074    0.03213   44.78770   2.513   0.0156 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(m_) sc(__) s(AA_E s(MSD)
scl(mg_sml)  0.070                             
scl(g_n_mn)  0.021 -0.009                      
scl(AA_Es_)  0.284 -0.052  -0.002              
scl(MnSlnD)  0.000 -0.028   0.010 -0.033       
scl(prcnt_)  0.010 -0.006  -0.219  0.007  0.009
optimizer (nloptwrap) convergence code: 0 (OK)
boundary (singular) fit: see help('isSingular')

individual word knowledge

all_cdi_data <- read.csv(here("data/main/data_to_analyze/level-words_source-cdi_data.csv"))
trials_with_word_knowledge <- trials_with_effect_vars |>
  left_join(all_cdi_data |> filter(type=="produces" & word_type=="extra_words") |> transmute(local_id, word, knows_target=responded), by=c("Trials.targetImage"="word", "SubjectInfo.subjID"="local_id")) |>
  mutate(knows_target = case_when(
    knows_target == 1 ~ 0.5,
    knows_target == 0 ~ -0.5,
    TRUE ~ NA
  ))

knowledge_text_effect <- lmer(scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(knows_target)
                    + (scale(text_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_word_knowledge)

boundary (singular) fit: see help('isSingular')

summary(knowledge_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + scale(knows_target) +  
    (scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_word_knowledge

REML criterion at convergence: 3545.8

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-2.9907 -0.6458 -0.0039  0.6957  2.9250 

Random effects:
 Groups             Name                   Variance  Std.Dev.  Corr 
 SubjectInfo.subjID (Intercept)            1.973e-02 1.405e-01      
                    scale(text_similarity) 6.723e-04 2.593e-02 1.00 
 Trials.targetImage (Intercept)            0.000e+00 0.000e+00      
 Trials.imagePair   (Intercept)            4.263e-10 2.065e-05      
 Residual                                  8.350e-01 9.138e-01      
Number of obs: 1314, groups:  
SubjectInfo.subjID, 49; Trials.targetImage, 24; Trials.imagePair, 16

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                 -2.129e-02  3.493e-02  5.798e+01
scale(text_similarity)                      -2.733e-02  2.108e-02  3.340e+02
scale(age_in_months)                         7.639e-02  3.610e-02  6.023e+01
scale(AoA_Est_target)                       -3.586e-02  2.918e-02  1.298e+03
scale(MeanSaliencyDiff)                      2.183e-02  2.545e-02  1.284e+03
scale(knows_target)                          9.706e-02  3.177e-02  8.456e+02
scale(text_similarity):scale(age_in_months) -5.483e-03  2.058e-02  3.027e+02
                                            t value Pr(>|t|)   
(Intercept)                                  -0.610  0.54452   
scale(text_similarity)                       -1.297  0.19570   
scale(age_in_months)                          2.116  0.03847 * 
scale(AoA_Est_target)                        -1.229  0.21933   
scale(MeanSaliencyDiff)                       0.858  0.39114   
scale(knows_target)                           3.055  0.00232 **
scale(text_similarity):scale(age_in_months)  -0.266  0.79009   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(t_) sc(__) s(AA_E s(MSD) scl(k_)
scl(txt_sm)  0.349                                     
scl(g_n_mn) -0.043  0.044                              
scl(AA_Es_)  0.332  0.204  -0.137                      
scl(MnSlnD) -0.023  0.003   0.085 -0.102               
scl(knws_t)  0.084 -0.121  -0.378  0.347 -0.192        
scl(_):(__) -0.016 -0.004   0.307 -0.054  0.016 -0.079 
optimizer (nloptwrap) convergence code: 0 (OK)
boundary (singular) fit: see help('isSingular')

#Swapping text similarity with image similarity:
knowledge_image_effect <- lmer(scale(corrected_target_looking) ~ scale(image_similarity)  * scale(age_in_months) + scale(AoA_Est_target)
                    + scale(MeanSaliencyDiff)
                    + scale(knows_target)
                    + (scale(image_similarity) | SubjectInfo.subjID)  
                    + (1|Trials.targetImage)
                    + (1|Trials.imagePair), 
                    data = trials_with_word_knowledge)

boundary (singular) fit: see help('isSingular')

summary(knowledge_text_effect)

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: 
scale(corrected_target_looking) ~ scale(text_similarity) * scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + scale(knows_target) +  
    (scale(text_similarity) | SubjectInfo.subjID) + (1 | Trials.targetImage) +  
    (1 | Trials.imagePair)
   Data: trials_with_word_knowledge

REML criterion at convergence: 3545.8

Scaled residuals: 
    Min      1Q  Median      3Q     Max 
-2.9907 -0.6458 -0.0039  0.6957  2.9250 

Random effects:
 Groups             Name                   Variance  Std.Dev.  Corr 
 SubjectInfo.subjID (Intercept)            1.973e-02 1.405e-01      
                    scale(text_similarity) 6.723e-04 2.593e-02 1.00 
 Trials.targetImage (Intercept)            0.000e+00 0.000e+00      
 Trials.imagePair   (Intercept)            4.263e-10 2.065e-05      
 Residual                                  8.350e-01 9.138e-01      
Number of obs: 1314, groups:  
SubjectInfo.subjID, 49; Trials.targetImage, 24; Trials.imagePair, 16

Fixed effects:
                                              Estimate Std. Error         df
(Intercept)                                 -2.129e-02  3.493e-02  5.798e+01
scale(text_similarity)                      -2.733e-02  2.108e-02  3.340e+02
scale(age_in_months)                         7.639e-02  3.610e-02  6.023e+01
scale(AoA_Est_target)                       -3.586e-02  2.918e-02  1.298e+03
scale(MeanSaliencyDiff)                      2.183e-02  2.545e-02  1.284e+03
scale(knows_target)                          9.706e-02  3.177e-02  8.456e+02
scale(text_similarity):scale(age_in_months) -5.483e-03  2.058e-02  3.027e+02
                                            t value Pr(>|t|)   
(Intercept)                                  -0.610  0.54452   
scale(text_similarity)                       -1.297  0.19570   
scale(age_in_months)                          2.116  0.03847 * 
scale(AoA_Est_target)                        -1.229  0.21933   
scale(MeanSaliencyDiff)                       0.858  0.39114   
scale(knows_target)                           3.055  0.00232 **
scale(text_similarity):scale(age_in_months)  -0.266  0.79009   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(t_) sc(__) s(AA_E s(MSD) scl(k_)
scl(txt_sm)  0.349                                     
scl(g_n_mn) -0.043  0.044                              
scl(AA_Es_)  0.332  0.204  -0.137                      
scl(MnSlnD) -0.023  0.003   0.085 -0.102               
scl(knws_t)  0.084 -0.121  -0.378  0.347 -0.192        
scl(_):(__) -0.016 -0.004   0.307 -0.054  0.016 -0.079 
optimizer (nloptwrap) convergence code: 0 (OK)
boundary (singular) fit: see help('isSingular')

All similarities

library(lme4)
library(lmerTest)   # adds p-values to lmer summaries
library(dplyr)
library(purrr)

sim_cols <- c(
  "clip_image_similarity", "clip_text_similarity", "clip_multimodal_similarity",
  "clip.hf_image_similarity", "cvcl_image_similarity", "cvcl_text_similarity",
  "cvcl_multimodal_similarity", "dinov2_image_similarity",
  "dino_say_vitb14_image_similarity", "dino_imagenet100_vitb14_image_similarity",
  "dinov3.babyview_image_similarity", "dinov3_image_similarity",
  "layer1_image_similarity", "layer12_image_similarity"
)
# or: sim_cols <- grep("_similarity$", names(trials_with_effect_vars), value = TRUE)

fit_sim_model <- function(sim_col, data, REML = TRUE) {
  data$sim <- data[[sim_col]]
  lmer(scale(corrected_target_looking) ~ scale(sim) * scale(age_in_months) +
         scale(AoA_Est_target) + scale(MeanSaliencyDiff) +
         (scale(sim) | SubjectInfo.subjID) +
         (1 | Trials.targetImage) +
         (1 | Trials.imagePair),
       data = data, REML = REML)
}

run_sim_models <- function(data, sim_cols, REML = TRUE) {
  models <- map(set_names(sim_cols), function(col) {
    tryCatch(fit_sim_model(col, data, REML),
             error = function(e) { message(col, " failed: ", conditionMessage(e)); NULL })
  }) |> compact()

  summary_tbl <- imap_dfr(models, function(m, col) {
    co <- coef(summary(m))
    co <- co[grepl("scale\\(sim\\)", rownames(co)), , drop = FALSE]
    tibble(
      model    = col,
      term     = if_else(grepl(":", rownames(co)), "similarity x age", "similarity"),
      estimate = co[, "Estimate"],
      se       = co[, "Std. Error"],
      t        = co[, "t value"],
      p        = co[, "Pr(>|t|)"],
      singular = isSingular(m),
      AIC      = AIC(m),
      n        = nobs(m)
    )
  })

  list(models = models, summary = summary_tbl)
}

results <- run_sim_models(trials_with_effect_vars, sim_cols)

Warning in checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, : Model failed to converge with max|grad| = 0.00216622 (tol = 0.002, component 1)
  See ?lme4::convergence and ?lme4::troubleshooting.

Warning in checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, : Model failed to converge with max|grad| = 0.00456806 (tol = 0.002, component 1)
  See ?lme4::convergence and ?lme4::troubleshooting.

results$summary |> filter(term == "similarity") |> arrange(p)

# A tibble: 14 × 9
   model               term  estimate     se      t      p singular    AIC     n
   <chr>               <chr>    <dbl>  <dbl>  <dbl>  <dbl> <lgl>     <dbl> <int>
 1 clip_text_similari… simi… -0.0365  0.0182 -2.01  0.0512 FALSE    20624.  7333
 2 layer1_image_simil… simi… -0.0421  0.0214 -1.97  0.0545 FALSE    20604.  7333
 3 layer12_image_simi… simi… -0.0418  0.0218 -1.92  0.0624 FALSE    20610.  7333
 4 dino_say_vitb14_im… simi… -0.0373  0.0207 -1.80  0.0786 FALSE    20607.  7333
 5 dino_imagenet100_v… simi… -0.0334  0.0194 -1.73  0.0933 FALSE    20630.  7333
 6 clip_image_similar… simi… -0.0275  0.0180 -1.53  0.135  FALSE    20628.  7333
 7 clip.hf_image_simi… simi… -0.0275  0.0180 -1.53  0.135  FALSE    20628.  7333
 8 cvcl_multimodal_si… simi…  0.0243  0.0192  1.26  0.210  FALSE    20630.  7333
 9 dinov2_image_simil… simi… -0.0240  0.0190 -1.26  0.216  FALSE    20631.  7333
10 cvcl_text_similari… simi… -0.0229  0.0201 -1.14  0.260  FALSE    20631.  7333
11 dinov3_image_simil… simi… -0.0186  0.0198 -0.937 0.354  FALSE    20621.  7333
12 clip_multimodal_si… simi… -0.0118  0.0194 -0.612 0.543  FALSE    20618.  7333
13 dinov3.babyview_im… simi… -0.00856 0.0218 -0.394 0.696  FALSE    20596.  7333
14 cvcl_image_similar… simi… -0.00236 0.0193 -0.122 0.903  FALSE    20638.  7333

summary(results$models$clip_text_similarity)   # full summary for any one model

Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: scale(corrected_target_looking) ~ scale(sim) * scale(age_in_months) +  
    scale(AoA_Est_target) + scale(MeanSaliencyDiff) + (scale(sim) |  
    SubjectInfo.subjID) + (1 | Trials.targetImage) + (1 | Trials.imagePair)
   Data: data

REML criterion at convergence: 20600.5

Scaled residuals: 
     Min       1Q   Median       3Q      Max 
-2.97702 -0.62924 -0.02646  0.67574  3.00822 

Random effects:
 Groups             Name        Variance Std.Dev. Corr 
 SubjectInfo.subjID (Intercept) 0.034301 0.18521       
                    scale(sim)  0.010717 0.10352  0.08 
 Trials.targetImage (Intercept) 0.009158 0.09570       
 Trials.imagePair   (Intercept) 0.002010 0.04484       
 Residual                       0.932047 0.96543       
Number of obs: 7333, groups:  
SubjectInfo.subjID, 181; Trials.targetImage, 48; Trials.imagePair, 32

Fixed effects:
                                  Estimate Std. Error         df t value
(Intercept)                      -0.014895   0.024507  62.528136  -0.608
scale(sim)                       -0.036545   0.018202  41.422498  -2.008
scale(age_in_months)              0.076703   0.017963 184.351081   4.270
scale(AoA_Est_target)            -0.079105   0.019199  42.730944  -4.120
scale(MeanSaliencyDiff)           0.008526   0.016895  64.157751   0.505
scale(sim):scale(age_in_months)  -0.007371   0.013517 160.017925  -0.545
                                Pr(>|t|)    
(Intercept)                      0.54552    
scale(sim)                       0.05123 .  
scale(age_in_months)            3.12e-05 ***
scale(AoA_Est_target)            0.00017 ***
scale(MeanSaliencyDiff)          0.61553    
scale(sim):scale(age_in_months)  0.58629    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Correlation of Fixed Effects:
            (Intr) scl(s) sc(__) s(AA_E s(MSD)
scale(sim)   0.054                            
scl(g_n_mn)  0.005 -0.001                     
scl(AA_Es_)  0.061  0.032  0.004              
scl(MnSlnD)  0.009  0.010 -0.003  0.023       
scl(s):(__) -0.001  0.002  0.071 -0.002 -0.002

results$summary |> filter(term == "similarity x age") |> arrange(p)

# A tibble: 14 × 9
   model              term  estimate     se      t       p singular    AIC     n
   <chr>              <chr>    <dbl>  <dbl>  <dbl>   <dbl> <lgl>     <dbl> <int>
 1 dino_say_vitb14_i… simi… -0.0399  0.0142 -2.80  0.00560 FALSE    20607.  7333
 2 dinov3.babyview_i… simi… -0.0367  0.0146 -2.52  0.0125  FALSE    20596.  7333
 3 layer12_image_sim… simi… -0.0306  0.0149 -2.06  0.0410  FALSE    20610.  7333
 4 dino_imagenet100_… simi… -0.0201  0.0131 -1.54  0.125   FALSE    20630.  7333
 5 dinov3_image_simi… simi… -0.0202  0.0138 -1.46  0.145   FALSE    20621.  7333
 6 cvcl_text_similar… simi… -0.0171  0.0126 -1.35  0.178   FALSE    20631.  7333
 7 clip.hf_image_sim… simi… -0.0150  0.0127 -1.18  0.239   FALSE    20628.  7333
 8 clip_image_simila… simi… -0.0150  0.0127 -1.18  0.239   FALSE    20628.  7333
 9 dinov2_image_simi… simi… -0.0129  0.0130 -0.987 0.325   FALSE    20631.  7333
10 clip_multimodal_s… simi…  0.0120  0.0142  0.849 0.397   FALSE    20618.  7333
11 layer1_image_simi… simi… -0.00826 0.0146 -0.565 0.573   FALSE    20604.  7333
12 clip_text_similar… simi… -0.00737 0.0135 -0.545 0.586   FALSE    20624.  7333
13 cvcl_image_simila… simi… -0.00670 0.0125 -0.537 0.592   FALSE    20638.  7333
14 cvcl_multimodal_s… simi…  0.00734 0.0149  0.493 0.623   FALSE    20630.  7333