Effect size estimation, significance testing and the file-drawer problem