randall@uvaarpa.virginia.edu (Randall Atkinson) (09/21/89)
In article <18401@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes: >Greg has often written to me that he doesn't feel there is any significant >accuracy to the readers/machine figures that one can derive from Brian's >arbitron figures. One thing is clear, the readers/machine figure is >exact for arbitron sites. Until last month beginning who knows when, edison.cho.ge.com was reporting wildly inaccurrate counts of people reading newsgroups. The reason was that the expire we used didn't have the options that arbitron expected us to use, so our active file misled arbitron about which articles were recent. This has been fixed somewhat, but now arbitron reports the correct number of newsreaders but 0 users on the system. The arbitron data is not exact for arbitron sites. QED Nevertheless, I think that the arbitron and inpaths data are useful benchmarks as long as we recognise upfront that they are not accurate. I do trust them for broad trends and for propogation and flow information. The point about large sites running rrn and sending in misleading "who reads what" statistics is very true. This university (UVa) primarily uses rrn to a single machine and so any reports which might be sent from here are going to be much lower than reality with respect to the "reader" data. DISCLAIMER: I don't speak for the corporation which is the University of Virginia; these are my own views.
chuq@Apple.COM (Chuq Von Rospach) (09/22/89)
>Nevertheless, I think that the arbitron and inpaths data are useful >benchmarks as long as we recognise upfront that they are not accurate. >I do trust them for broad trends and for propogation and flow information. Agreed. Arbitron has generaly shown readership of rec.mag.otherrealms in the 3500-5000 range. I've done independent surveys where, by checking the percentage of returns in an area with known levels of readership and extrapolating those out to the entire survey response, my readership is closer to 11,000 to 15,000. That's at least a 50% difference between arbitron and my numbers. My numbers may be a little high. They may also be a little low (+- 5% or so). It's impossible to tell. All I do know is that Brian's numbers are a lot different and significantly lower, well beyond his margin of error. What I think you can tell from Brian's numbers are relative things -- I don't think the relative number of readers will change. I don't think the relative interest in a group would change. etc. Absolute numbers, though, they aren't. -- Chuq Von Rospach <+> Editor,OtherRealms <+> Member SFWA/ASFA chuq@apple.com <+> CI$: 73317,635 <+> [This is myself speaking. I am not Appl Segmentation Fault. Core dumped.
brad@looking.on.ca (Brad Templeton) (09/22/89)
In article <34921@apple.Apple.COM> chuq@Apple.COM (Chuq Von Rospach) writes: > >Agreed. Arbitron has generaly shown readership of rec.mag.otherrealms in the >3500-5000 range. I've done independent surveys where, by checking the >percentage of returns in an area with known levels of readership and >extrapolating those out to the entire survey response, my readership is >closer to 11,000 to 15,000. OtherRealms, as you know, is a very special case, because it is 12 articles over 3 days once every N months. Many sites are down to 1 week expire now. (Hell, I used to run 1 day expire on most groups) and so it makes sense that a once/month arbitron report would be off by a factor of 4. In fact, didn't Chuq once deliberately arrange to expire R.M.O well beyond the default time, in part to avoid having to deal with back-issue requests, but also in part to make sure it showed up on the Arbitron surveys? Actually, my personal opinion is that the survey results are too high, by as much as a factor of 5 in some cases. 71,000 for rec.humor.funny? Not a chance in the world. But that's because we have no really strong reliable way to find out what percentage of total netreaders the 719 arbitron sites represent. Brian says 4.5% based on the fact that there are 15,000 machines in net maps. A survey I did said that about 21% of all USENET articles are posted by readers at arbitron sites. The real answer lies between. But all this is not important in calculating readers per site, unless you feel that the readers/site figure for arbitron sites is extremely atypical. NNTP doesn't affect this figure, other than to plump it up, since adding reports for sites that read by NNTP would add thousands of single-user sites that can't possibly increase the readers/site figure over 1. -- Brad Templeton, Looking Glass Software Ltd. -- Waterloo, Ontario 519/884-7473
david@indetech.com (David Kuder) (09/24/89)
In article <19376@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes: >71,000 for rec.humor.funny? Not a chance in the world. If you want a better number why not ask for it. Just post a demand in your playground/newsgroup. Ask every reader to send you e-mail to some address that has a counter on it. >But all this is not important in calculating readers per site, unless >you feel that the readers/site figure for arbitron sites is extremely >atypical. NNTP doesn't affect this figure, other than to plump it >up, since adding reports for sites that read by NNTP would add >thousands of single-user sites that can't possibly increase the >readers/site figure over 1. Well, the problem is "feel". I may feel that the readers/site from Brian's data is bogus while you feel otherwise. We won't be able agree because it isn't great data. The selection mechanism is poor, the data collected overlooks at one mechanism for reading news, the definition of site is strange (any thing that ever showed up in a "Path:", we used to send out workstation names here while running arbitron from our server, what's that do to the numbers), the timeliness of the data is queer (we last sent in an arbitron about 3 months ago, this will probably be the first month since then that we aren't counted), and there are all the things Brian lists himself. Not that any of this makes the data less interesting. It is interesting. It just isn't good enough to prove anything. You, Mr. Survey, ought to know better than putting trust in anything collected from the net. -- David A. Kuder david@indetech.com 415 438-2003 {sun,sharkey,pacbell}!indetech!david
vnend@phoenix.Princeton.EDU (D. W. James) (10/03/89)
In article <19376@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes:
)But all this is not important in calculating readers per site, unless you
)feel that the readers/site figure for arbitron sites is extremely
)atypical. NNTP doesn't affect this figure, other than to plump it up,
)since adding reports for sites that read by NNTP would add thousands of
)single-user sites that can't possibly increase the readers/site figure over 1.
I disagree. NNTP *does* affect it, if you view readership/site
as based on the *server*, rather than reader machine (which is much like
claiming that it should be based on per terminal...) NNTP *significantly*
affects your (questionable) method of evaluating the success of groups
if you view it that way.
Another factor that is becoming more and more important (in
terms of readers and posters) on the net is BITNET sites, for which
there isn't an ARBITRON analogue. Here at Princeton PUCC is usually
right behind phoenix in terms of posted volume, and it certainly
has more accounts than Phoenix. So, I agree that the stats are
only valid as comparisons between groups, not for use as
absolute numbers (which really throws your "stat" out the
window.)
--
Later Y'all, Vnend Ignorance is the mother of adventure.
SCA event list? Mail? Send to:vnend@phoenix.princeton.edu or vnend@pucc.bitnet
Anonymous posting service (NO FLAMES!) at vnend@ms.uky.edu
"First, they stood guard over us. Then, they sat guard over us. Then they wandered off to find some corn plasters and we escaped."