A Bayesian meta-analysis of parafoveal preview effects during reading
For any questions or issues, please contact Martin Vasilev (mvasilev@bournemouth.ac.uk).
This file contains the most important information in order to repeat the analyses. Due to the vast amount of data and analyses (as well was my lack of experience when I started this), the scripts are spread over multiple functions and files. Here, I explain how the analysis was done.
If you are interested in the literature search and study selection: please see Documents/all_papers.xls. The file is heavily commented and is generally self-contained.
Before you start: Install JAGS: http://mcmc-jags.sourceforge.net/; then, in Rstudio, install the package 'rjags': install.packages("rjags")
-
The most basic scripts are N1.R (for N+1 studies) and N2.R (for N+2 studies). This is where all the information extracted from the papers is coded. The scripts are a bit clumsy and take long to execute- I have provided them for completeness. However, the data which the two scripts give are already available in the folder "Data".
-
check_ES.R does multiple things, but it generally subsets the data and prepares it for the analyses (it also does the SD imputation). If you want to repeat the funnel plots, set impute to FALSE in the beginning. Please note that it is technically not needed to run the scripts as the data is also already available.
-
pub_bias.R generates code, graphs etc. related to publication bias (however see the above comment about imputation).
-
get_bench.R does the baseline (or benchmark) analyses of fixation durations when parafoveal preview is allowed (i.e. valid preview).
-
N1_Bmeta_final.R and N2_Bmeta_final.R contain the core analyses for N+1 and N+2 preview effects. Both files are generally commented. Please note that to generate the graphs, additional functions are used (see /Plots/). Due to multiple issues that I had with plotting software such as ggplot2, I generate some parts of the graphs in external software (Microsoft Publisher); the files are generally available in the same folder, but if something is missing, please contact me for them.
-
N1_B_Reg.R contains the meta-regression results for N+1 studies (N+2 meta-regression analyses are in N2_Bmeta_final.R).
-
Additional functions/ scripts:
- N1_probs.R generates N+1 probability plot
- CohensD custom-made function for checking the coded statistics for mistakes and double-checking of whether the correct values were coded.
- JModel.R generates JAGS models (as txt files) by taking some input values. This function was created to minimise the code that was repeated over and over again.
- JReg.R Same as above, but it generates meta-regression models
- sensitivity_analysis.R performs a sensitivity analysis using the leave-one-out method. Please note: depending on your machine, N+1 sensitivity analyses can take A LOT of time (up to 2 hrs).