I work first hand in a research community and I've never seen a meta analysis published that was based solely on unpublished works. However, most of the good on

I work first hand in a research community and I've never seen a meta analysis published that was based solely on unpublished works. However, most of the good ones consist of a mix of published and unpublished works to ascertain whether there's a real effect or whether there's been a publishing bias that doesn't reflect the total body of research. A good meta analysis will take the time to discuss their methods for selecting published works and any efforts they made to solicit unpublished works such as contacting prominent authors in the field, getting on mailing lists, etc. The good ones also tend to discuss how they analyzed the papers for selection, discuss the paper stats (i.e. we started with 200 published papers and 20 unpublished ones, after winnowing this got reduced to 130 published and 10 unpublished), and how they collapsed results to calculate effect sizes etc. Meta analyses should include as much of the published literature as possible for the research question they're attempting to answer AND they should seek out unpublished works. The meta analysis will consist primarily of published works, the unpublished works are collected to counter the presence of positive result publishing bias. This provides a fuller picture of the research field than simply relying on published works. After they've collected the articles, the researchers analyze the collected works to ascertain which ones actually address the research question(s) they're attempting to answer, they toss everything else.