I ran an analysis on 500,000+ recent records to see how often Lazy Scholar finds at least one free full text. This value is 36-41%. Let me break that down a bit.
36% reflects the percent of free full texts that are definitively free to view to the user, but does not include results from Google Scholar. This is because in practice, I’ve noticed that some full text links on Google Scholar are indexed incorrectly and not really freely available. If I include such links from Google Scholar in the analysis, that number if 41%. In other words, Lazy Scholar finds a free full text somewhere between 36-41% of the time.
If I look at Google Scholar itself as a source, it provides a free full text on 18% of articles that Lazy Scholar users open. This isn’t far from this estimate of Google Scholar showing 24% of the literature as free to access (by either an open access version or an archived version somewhere), and the lower number probably reflects the fact that Lazy Scholar searches tend to be from more recent papers which may not have free versions indexed yet. Interestingly, Google Scholar only found full texts 4.4% of the time when other sources that Lazy Scholar uses couldn’t locate one (in 0.9% of these cases a DOI couldn’t be found, so some other sources couldn’t be checked).
By design, Lazy Scholar attempts to find a free full text on any page that it recognizes as a scientific abstract or paper. It does so by looking for a DOI or by matching the url to a large list. Doing so without a DOI is more difficult. If I restrict the analysis to only papers where a DOI is available, Lazy Scholar finds a free full text 49-54% of the time. This estimate is similar to Unpaywall’s 47% estimate that was done with records containing a DOI.
In the upcoming version launching next week, I’ve added a tweak which may increase success a bit, so I’ll share some more data in the coming months.