Forum Moderators: DixonJones
My Webtrends referrers shows the following:
1.http://www.mysite.com 27 %
2.No Referrer 19 %
3.the rest are the search engines and sites linking to me.
I throw up a theory which, though not provable, is (probably) a reasonable working model when trying to gauge the worth of referral sites, and perhaps why “click throughs” measured by the referral site do not match the arrivals as measured at the site getting the click throughs.
Theory is that actual referrals from named sites are approx twice the recorded numbers
To reach this conclusion, I have assumed
1. The site is cacheable, which means that around 25% of users never get logged (see the previous thread for an explanation). No one has actually proved this figure, but it appears to be the accepted ball park figure.
2. One has to make an assumption about the number of “real” “no referrers”. In other words write ins, book marks, email clicks, etc. My research shows that no one really has a clue what this is. For the sake of argument let us call this 20%. It appears to me reasonable (and makes the mathematics easy to follow, and to implement!)
3. Under these assumptions the Log File totals for visits to mysite.com is inflated by 25% to allow for caching. Then 20% of this inflated total removed to allow for “real” “no referrers”. Which takes one back to the number you started with on the Log File.
This means that (providentially!) the Log File numbers for visits represents approximately the number of visitors that have arrived from named referrer sites.
4. You therefore get an approximation for the “real” number of referrals from each referral site by dividing out the “mysite.com” and “no referrers” in the Log Files among the named referrer sites in proportion to the named sites contribution to the total.
5. Assume that the “mysite.com” and “no referrers” in the Log Files are statistically from the same spread of named referrers as the ones that Log Files actually measure (one can argue the validity of this assumption, but it appears to be reasonable)
6. I looked at the Log File analysis of 20 individual hotel sites that I manage. Interestingly the results for all the hotel sites were very similar to (+ or – 5%)
“mysite.com + no referrer” = 50%
named referrer = 50%
This is not necessarily true for all sites, but is probably true for most hotel sites
7. So to “normalise” the statistics, one multiplies all “named referrer” sites by 2 in order to get an approximation.
8. So if one had a typical hotel site with Log File analysis of
mysite.com 12,611
no referrer 7,245
referrerA 2727
referrerB 2118
referrerC 2116
referrerD 1945
referrerE 645
referrerF 362
referrerG 345
referrerH 311
referrerI 231
referrerJ 225
All other smaller referrers 9489
Total Log Files 40,370
9. My conjecture therefore is that a more accurate evaluation of these figures means that, using a multiplier of 2
Total Log Files around 50,000 (adding 25% to analysed figure). And this is broken down into
No Referrers 10,000 (20% of total)
referrerA 5454
referrerB 4236
referrerC 4232
referrerD 3890
referrerE 1290
referrerF 724
referrerG 690
referrerH 622
referrerI 462
referrerJ 450
etc, etc
10. Its not rocket science, and the assumptions are debatable, but I think it gives a clearer idea as to what is really coming from named referral sites in Log File Analysis. And is a simple to use formula for evaluating the volume of referrals.
I invite anyone to come up with a better model for answering this thorny question on Log File Analysis that seems to be swept under the carpet most times it is raised.
I have to give this one another read!
