A decade after the publication of Visible Learning: A Synthesis of Over 800 Meta-Analyses Relating to Achievement, the influence of John Hattie's statistical work continues to grow. This paper critically examines one aspect of Hattie's 'synthesis' of education meta-analyses, his summation of research data for whole language literacy teaching. Of the four meta-analyses reported as being on whole language in Visible Learning, one is in fact a survey of research on language experience. Hattie uncritically accepted another meta-analysis, which reported that whole language triggers dramatic student unlearning, despite ample reasons to question its findings. On the other hand Hattie, revisiting the sole meta-analysis, reported finding a positive effect size for whole language in an unusual manner, with selected research studies removed to produce an effect size of zero. This paper argues that a prior bias on the part of the author of Visible Learning -- against whole language and in favour of synthetic phonics and direct instruction -- likely explains this interpretation of the research data on whole language. Potential implications for basing pedagogical practice and public policy decisions on aggregated effect sizes from meta-analyses are discussed.