In response to reviewer 1.

ME_DATA_PATH <- here("data/0_metaanalysis_data.csv")
ma_raw <- read_csv(ME_DATA_PATH) %>%
  select(1:4, 6:8, 10:16, 18,19,21, 27:29, 31:34, 48:50, 59) %>%
  mutate_if(is.character, as.factor)

AVG_MONTH <- 30.43688
ma_c <- ma_raw %>%
  filter(!is.na(d_calc)) %>%
  mutate(mean_age = mean_age_1/AVG_MONTH,
         year = as.numeric(str_sub(short_cite, -5, -2)),
         condition_type = as.factor(ifelse(infant_type == "typical" & ME_trial_type == "FN", "TFN", 
                                 ifelse(infant_type == "typical" & ME_trial_type == "NN", "TNN", 
                                        as.character(infant_type))))) %>%
  filter(mean_age < 144)   #  (12 yo), excludes MR population
  # filter(!(study_ID %in% c("williams2009", "frank1999"))) # for checking that holds when exclude dissertations [it does]

Outliers

overall_mean <- mean(ma_c$d_calc)
overall_sd <- sd(ma_c$d_calc)

plus2 <- overall_mean + (2*(overall_sd))
minus2 <- overall_mean - (2*(overall_sd))

plus3 <- overall_mean + (3*(overall_sd))
minus3 <- overall_mean - (3*(overall_sd))

twosd_outliers <- ma_c %>%
  filter(d_calc > plus2 | d_calc < minus2) %>%
  data.frame()

threesd_outliers <- ma_c %>%
  filter(d_calc > plus3 | d_calc < minus3) %>%
  data.frame()

There are 8 conditions that are 2sds beyond the mean; There are 3 conditions that are 3sds beyond the mean:

kable(threesd_outliers)
study_ID short_cite long_cite peer_reviewed expt_num response_mode dependent_measure infant_type infant_type2 group_name_1 n_1 mean_age_1 same_infant x_1 SD_1 t d num_trials object_stimulus N_AFC mean_production_vocab ME_trial_type lab_group data_source d_notes d_calc d_var_calc es_method mean_age year condition_type
estis2015 Estis & Beverly (2015) Estis, J. M., & Beverly, B. L. (2015). Children with SLI exhibit delays resolving ambiguous reference. Journal of child language, 42(1), 180-195. yes 1 behavior target_selection NT SLI experimental 8 2617.571 25 0.960 0.070 NA NA 10 objects N_AFC-2 NA FN estis paper NA 6.571429 2.823980 group_means_one 85.99998 2015 NT
frank2016 Frank et al. (2016) Frank, M. C., Sugarman, E., Horowitz, A., Lewis, M. L., & Yurovsky, D. (2016). Using tablets to collect data from young children. Journal of Cognition and Development,17(1), 1-17. yes 1 behavior target_selection typical typical experimental 20 1636.320 39 0.970 0.060 NA 7.833333 8 digital N_AFC-2 NA FN framkm paper NA 7.833333 1.584028 group_means_one 53.76110 2016 TFN
gollek2016 Gollek & Doherty (2016) Gollek, C.& Doherty, M.J. (2016). Metacognitive developments in word learning: Mutual exclusivity and theory of mind. Journal of Experimental Child Psychology. 148, 51-69. yes 3 behavior target_selection typical typical disambiguation - younger 15 1278.349 42 0.986 0.052 NA NA 5 objects N_AFC-2 NA FN gollek paper NA 9.346154 2.978353 group_means_one 42.00000 2016 TFN

Testing for heteroskedasticity

There’s heteroskedasticity if we include the outliers; it goes away if you exclude them.

Including outliers:

model <- lm(d_calc ~ mean_age, data = ma_c, weights = n_1)
olsrr::ols_test_breusch_pagan(model)
## 
##  Breusch Pagan Test for Heteroskedasticity
##  -----------------------------------------
##  Ho: the variance is constant            
##  Ha: the variance is not constant        
## 
##                Data                
##  ----------------------------------
##  Response : d_calc 
##  Variables: fitted values of d_calc 
## 
##          Test Summary           
##  -------------------------------
##  DF            =    1 
##  Chi2          =    11.94575 
##  Prob > Chi2   =    0.0005477213
olsrr::ols_test_f(model)
## 
##  F Test for Heteroskedasticity
##  -----------------------------
##  Ho: Variance is homogenous
##  Ha: Variance is not homogenous
## 
##  Variables: fitted values of d_calc 
## 
##        Test Summary        
##  --------------------------
##  Num DF     =    1 
##  Den DF     =    144 
##  F          =    3.12629 
##  Prob > F   =    0.07915652

Excluding outliers (3sd beyond mean):

model <- lm(d_calc ~ mean_age, data = ma_c  %>% filter(d_calc > plus3 | d_calc < minus3), weights = n_1)
olsrr::ols_test_breusch_pagan(model)
## 
##  Breusch Pagan Test for Heteroskedasticity
##  -----------------------------------------
##  Ho: the variance is constant            
##  Ha: the variance is not constant        
## 
##                Data                
##  ----------------------------------
##  Response : d_calc 
##  Variables: fitted values of d_calc 
## 
##         Test Summary         
##  ----------------------------
##  DF            =    1 
##  Chi2          =    0.1313952 
##  Prob > Chi2   =    0.7169899
olsrr::ols_test_f(model)
## 
##  F Test for Heteroskedasticity
##  -----------------------------
##  Ho: Variance is homogenous
##  Ha: Variance is not homogenous
## 
##  Variables: fitted values of d_calc 
## 
##       Test Summary        
##  -------------------------
##  Num DF     =    1 
##  Den DF     =    1 
##  F          =    8.231854 
##  Prob > F   =    0.2135048