A short post on the correlation between shares and readership and why this metric (along with AVE) shouldn’t be used.
Upfront, AVE has been thoroughly discredited, and rightly so. In fact PR Week claims 75% of PRs no longer use these (albeit this is from a survey, so less trustworthy).
But for some reason readership seems to persist among several companies that I encounter – possibly because it’s an easy, impressive stat to give (Meltwater etc give) and it’s not been as discredited? Even though (in at least one case I’ve seen – for the Giro d’Italia bike race – it added up to more than the planet’s population.
Anyway, before I start there’s a bit of a precursor. A short while back, when planning for a big product launch, an agency’s new client* had said they measured PR by the combined readership of the coverage.
As far as the agency could see, the end company did it because its last agency had. The company initially advised against it, highlighting that proposed campaign was more focused – concentrating on niche B2B titles in countries (US / Far East) that would reach the engineers the company needed, rather than (as the old agency had) the consumer in the UK, who obviously wouldn’t have any say in the buying decision of its technology.
But the agency needed proof that reach would be a pointless measure, so we took the coverage list (with readership) for the old agency and correlated it with shares.
Long story short – it had a tiny correlation: 0.0027 in fact (Pearson**). And this was for all coverage of the client, when you took into consideration the coverage the old agency said it got, this dropped further to 0.0008.
To put this number in context, a perfect correlation is 1 (or -1). And 0 represents no correlation
Reach is even worse than AVE
I recently ran the figures again, this time for a similar tech company, taking 2.5-years of coverage – 9175 items in total.
Again, using the Pearson correlation formula this gave a correlation of 0.1100. Significantly better than the above, but still in no ways a correlation.
But, wire coverage is included in this and has been shown to be invisible except to those actively searching for it on Google – it doesn’t get shared and doesn’t get seen. So I also eliminated this from the figures, this reduced it to 1963 articles.
This time (strangely) the correlation decreased: to 0.0886***.
But when the same correlation was done for AVE vs shares, the co-efficients were 0.1492 (all coverage) and 0.1323 (wire coverage removed).
In short – if this data set (and I acknowledge it’s only one data set) is a representation… really, REALLY, don’t use readership as a substitute for AVE.
*I won’t mention the tech firm, or the old agency. But can recommend the new agency, drop me an email if you want its name.
**Pearson was used rather than Spearman as it’s a metric variable rather than a ranked variable.
***Need to investigate why it decreased… another blog when I have time, perhaps.