Another Google Scholar Purge

Google Scholar has again gone through a process of forgetting or purging a number of citations to my papers.  This time last week it said I had 1,589 citations to my work.  Today, 1,170 (both figures down from 2,130 in January last year).  Google Scholar is fairly opaque in how it does things, and I don't know what records it deleted and why, but it seems unfortunately very unreliable.   As a record of my papers, it's useful.  For citations?  It's becoming increasingly less so.  Anyone else noticed the same with them?

After noticing (some years ago) strangely fluctuating numbers coming from Google Scholar I started periodically recording the number of citations it reported for me on a spreadsheet.  Hence I can present the graph attached here.  The latest downward line is smaller than the one I noticed mid-way through last year, but still takes the number of citations back to that of around 5 years ago!

Google vs ISI

Following occasional posts in the past (here, here, here) about the functioning of different publication citation databases,  I thought I'd look at the latest state of play by comparing Google Scholar with ISI.  I didn't want to be overwhelmed with a unwieldy data set, so I restricted my analysis to set of papers with a modest citation profile (i.e. those co-authored by me).  The graph below shows my papers, indexed in order according to decreasing number of citations as judged by ISI:

Looking at the most cited papers, it doesn't look like there's a terribly strong correlation between the two databases, though the general trend across the whole slew of publications looks more reasonable with a few notable inconsistencies.  

As researchers and Universities get ready for the 2021 Research Excellence Framework (sic), they will be forced to play a game in which citation databases will party inform the panels.  Two possible database options seem to disagree with each other quite radically.  I don't know if the Physics sub-panel will count citations.  The published draft guidelines say:

 

and that 

 

which is quite soon.  Still, despite (probably) making use of citation data in deciding what universities' research to fund, they add, for clarity:

hmm.

update: I can't help playing around a bit with the data.  If I group the publications by journal, then it looks like PRC (Physical Review C) is measured most erratically between the two databases, with data from JPG (Journal of Physics G) also showing quite a few difference of opposite sign. In the plot, the thick black bars show the difference between the two databases for each paper.  Journal name abbreviations are shown where there are at least two papers in that journal.

update 2: I see from the REF website that ISI is to be the chosen database of REF2021