reid@glacier.ARPA (Brian Reid) (03/11/86)
I now have data from 0.4% of the network for newsgroup readership. If you don't think these numbers are fair, then please run "arbitron -m" on your system (I've posted it twice in the last couple of days). Summary: 2738 users at 23 sites; 465 of them (16.98%) are netreaders. +-- number of people who read this group | +-- percentage of the total user population who read this group | | +-- percentage of newsreaders who read this group. V V V 103 3.76 22.15 net.sources 95 3.47 20.43 net.general 93 3.40 20.00 net.announce 88 3.21 18.92 net.jokes 80 2.92 17.20 mod.sources 79 2.89 16.99 mod.newprod 75 2.74 16.13 net.jobs 70 2.56 15.05 mod.os.unix 69 2.52 14.84 net.arch 67 2.45 14.41 net.announce.newusers 66 2.41 14.19 mod.movies 65 2.37 13.98 mod.unix 61 2.23 13.12 mod.std 60 2.19 12.90 net.audio 59 2.15 12.69 net.rumor 55 2.01 11.83 net.usenix 54 1.97 11.61 net.unix-wizards 53 1.94 11.40 net.unix 51 1.86 10.97 net.sources.bugs 51 1.86 10.97 net.singles 51 1.86 10.97 mod.os 50 1.83 10.75 net.lang 50 1.83 10.75 net.graphics 49 1.79 10.54 net.sources.games 48 1.75 10.32 net.movies 47 1.72 10.11 mod.recipes 46 1.68 9.89 net.rec.nude 46 1.68 9.89 net.lang.c++ 45 1.64 9.68 net.lan 44 1.61 9.46 net.announce.arpa-internet 43 1.57 9.25 net.news 43 1.57 9.25 net.consumers 42 1.53 9.03 net.wanted 42 1.53 9.03 net.sources.mac 42 1.53 9.03 net.micro.mac 42 1.53 9.03 net.lang.c 42 1.53 9.03 net.cooks 41 1.50 8.82 net.women 41 1.50 8.82 net.invest 41 1.50 8.82 net.bugs 41 1.50 8.82 net.bicycle 40 1.46 8.60 net.micro 40 1.46 8.60 net.lang.mod2 40 1.46 8.60 net.emacs 39 1.42 8.39 net.med 39 1.42 8.39 mod.graphics 38 1.39 8.17 net.travel 38 1.39 8.17 net.mail 37 1.35 7.96 net.rec 37 1.35 7.96 net.legal 37 1.35 7.96 net.auto 37 1.35 7.96 mod.computers.workstations 36 1.31 7.74 net.net-people 36 1.31 7.74 net.lsi 36 1.31 7.74 mod.std.c 35 1.28 7.53 net.social 35 1.28 7.53 net.news.group 35 1.28 7.53 net.misc 35 1.28 7.53 net.jokes.d 34 1.24 7.31 net.taxes 34 1.24 7.31 mod.test 34 1.24 7.31 mod.computers.pyramid 33 1.21 7.10 net.micro.pc 33 1.21 7.10 net.games 33 1.21 7.10 net.astro 32 1.17 6.88 net.sf-lovers 32 1.17 6.88 net.music 32 1.17 6.88 net.dcom 32 1.17 6.88 net.books 32 1.17 6.88 mod.politics 32 1.17 6.88 mod.human-nets 32 1.17 6.88 mod.compilers 31 1.13 6.67 net.research 31 1.13 6.67 net.rec.ski 31 1.13 6.67 net.mail.headers 31 1.13 6.67 net.bugs.4bsd 31 1.13 6.67 net.auto.tech 31 1.13 6.67 mod.std.mumps 31 1.13 6.67 mod.protocols 31 1.13 6.67 mod.computers.sun 31 1.13 6.67 mod.computers.laser-printers 31 1.13 6.67 mod.computers.ibm-pc 30 1.10 6.45 net.micro.68k 30 1.10 6.45 net.crypt 30 1.10 6.45 net.ai 30 1.10 6.45 mod.telecom 29 1.06 6.24 net.periphs 29 1.06 6.24 net.decus 29 1.06 6.24 net.columbia 29 1.06 6.24 mod.protocols.kermit 29 1.06 6.24 mod.general 28 1.02 6.02 net.sci 28 1.02 6.02 net.news.adm 28 1.02 6.02 net.followup 28 1.02 6.02 mod.computers.macintosh 27 0.99 5.81 net.tv 27 0.99 5.81 net.puzzle 27 0.99 5.81 net.lang.pascal 27 0.99 5.81 net.astro.expert 27 0.99 5.81 net.analog 26 0.95 5.59 net.space 26 0.95 5.59 net.rec.photo 26 0.95 5.59 net.news.stargate 26 0.95 5.59 net.music.classical 26 0.95 5.59 net.flame 26 0.95 5.59 mod.protocols.tcp-ip 25 0.91 5.38 net.wines 25 0.91 5.38 net.os 25 0.91 5.38 net.news.sa 25 0.91 5.38 net.news.b 25 0.91 5.38 net.math 25 0.91 5.38 net.lang.prolog 25 0.91 5.38 net.lang.lisp 25 0.91 5.38 mod.techreports 24 0.88 5.16 net.video 24 0.88 5.16 net.text 24 0.88 5.16 net.lang.st80 24 0.88 5.16 net.bugs.uucp 24 0.88 5.16 net.bio 24 0.88 5.16 mod.ai 23 0.84 4.95 net.mag 23 0.84 4.95 net.games.video 23 0.84 4.95 mod.newslists 22 0.80 4.73 net.wanted.sources 22 0.80 4.73 net.games.hack 22 0.80 4.73 mod.music 21 0.77 4.52 net.rec.wood 21 0.77 4.52 net.ham-radio 20 0.73 4.30 net.startrek 20 0.73 4.30 net.physics 20 0.73 4.30 net.music.folk 20 0.73 4.30 net.kids 20 0.73 4.30 net.database 20 0.73 4.30 mod.std.unix 20 0.73 4.30 mod.computers.apollo 19 0.69 4.09 net.nlang 19 0.69 4.09 net.news.config 19 0.69 4.09 net.lang.ada 19 0.69 4.09 net.cse 19 0.69 4.09 net.college 19 0.69 4.09 net.bugs.v7 19 0.69 4.09 mod.ber 18 0.66 3.87 net.music.gdead 18 0.66 3.87 net.micro.amiga 18 0.66 3.87 net.internat 18 0.66 3.87 mod.singles 18 0.66 3.87 mod.risks 18 0.66 3.87 mod.protocols.appletalk 18 0.66 3.87 mod.motss 18 0.66 3.87 mod.map 17 0.62 3.66 net.wobegon 17 0.62 3.66 net.veg 17 0.62 3.66 net.tv.drwho 17 0.62 3.66 net.micro.16k 17 0.62 3.66 net.math.symbolic 17 0.62 3.66 net.comics 17 0.62 3.66 mod.legal 16 0.58 3.44 net.politics 16 0.58 3.44 net.pets 16 0.58 3.44 net.news.newsite 16 0.58 3.44 net.lang.apl 16 0.58 3.44 net.info-terms 16 0.58 3.44 net.garden 16 0.58 3.44 net.eunice 16 0.58 3.44 mod.computers.vax 16 0.58 3.44 mod.computers.sequent 15 0.55 3.23 net.music.synth 15 0.55 3.23 net.lang.forth 15 0.55 3.23 net.lang.f77 15 0.55 3.23 net.games.rogue 15 0.55 3.23 net.games.chess 15 0.55 3.23 mod.vlsi 14 0.51 3.01 net.sources.d 14 0.51 3.01 net.nlang.india 14 0.51 3.01 net.motss 14 0.51 3.01 net.games.trivia 14 0.51 3.01 net.cycle 14 0.51 3.01 net.cog-eng 14 0.51 3.01 net.bugs.usg 14 0.51 3.01 net.aviation 13 0.47 2.80 net.religion.jewish 13 0.47 2.80 net.railroad 13 0.47 2.80 net.misc.coke 13 0.47 2.80 net.micro.apple 13 0.47 2.80 net.games.frp 13 0.47 2.80 net.games.emp 13 0.47 2.80 net.games.board 13 0.47 2.80 mod.os.os9 13 0.47 2.80 mod.computers.gould 12 0.44 2.58 net.religion 12 0.44 2.58 net.poems 12 0.44 2.58 net.news.notes 12 0.44 2.58 net.micro.atari16 12 0.44 2.58 net.micro.atari 12 0.44 2.58 net.challenger 11 0.40 2.37 net.nlang.africa 11 0.40 2.37 net.ham-radio.packet 11 0.40 2.37 mod.rec.guns 10 0.37 2.15 net.works 10 0.37 2.15 net.test 10 0.37 2.15 net.sport.baseball 10 0.37 2.15 net.sport 10 0.37 2.15 net.micro.cpm 10 0.37 2.15 net.micro.att 10 0.37 2.15 net.bugs.2bsd 9 0.33 1.94 net.philosophy 9 0.33 1.94 net.origins 9 0.33 1.94 net.micro.trs-80 9 0.33 1.94 net.micro.hp 9 0.33 1.94 net.micro.cbm 9 0.33 1.94 net.abortion 9 0.33 1.94 mod.computers.ridge 8 0.29 1.72 net.roots 8 0.29 1.72 net.games.pbm 8 0.29 1.72 mod.politics.arms-d 7 0.26 1.51 net.sport.hoops 7 0.26 1.51 net.sport.football 7 0.26 1.51 net.religion.christian 7 0.26 1.51 net.rec.boat 6 0.22 1.29 net.tv.soaps 6 0.22 1.29 net.suicide 6 0.22 1.29 net.rec.bridge 6 0.22 1.29 net.movies.sw 6 0.22 1.29 net.micro.6809 6 0.22 1.29 net.math.stat 6 0.22 1.29 net.games.go 5 0.18 1.08 net.sport.hockey 5 0.18 1.08 net.rec.scuba 5 0.18 1.08 net.nlang.celts 5 0.18 1.08 net.micro.ti 5 0.18 1.08 net.micro.atari8 5 0.18 1.08 net.bio.slime 5 0.18 1.08 mod.mac 4 0.15 0.86 net.women.only 4 0.15 0.86 net.vvs 4 0.15 0.86 net.rec.skydive 4 0.15 0.86 net.rec.birds 3 0.11 0.65 net.nlang.greek 3 0.11 0.65 net.news.map 3 0.11 0.65 net.micro.432 3 0.11 0.65 mod.map.uucp 2 0.07 0.43 net.usoft 2 0.07 0.43 net.rec.disc 2 0.07 0.43 net.politics.theory -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
woods@hao.UUCP (Greg Woods) (03/12/86)
> I now have data from 0.4% of the network for newsgroup readership. > If you don't think these numbers are fair, then please run "arbitron -m" > on your system (I've posted it twice in the last couple of days). If you expect this to be a standard for the network, you are going to need to post versions which do not depend on the Bourne shell. We don't have a working version of it here, at least not one that will parse your program. We are a backbone site. Therefore you cannot hold your data up as a standard for the whole net unless you are prepared to do a lot more work. In our case, a csh version would suffice, but I'm sure there are other sites (V6, micros, etc.) that couldn't handle that either. So none of this 'if you don't like my data, run my program' stuff, at least not until you have a program everyone can run. --Greg -- {ucbvax!hplabs | decvax!noao | mcvax!seismo | ihnp4!seismo} !hao!woods CSNET: woods@ncar.csnet ARPA: woods%ncar@CSNET-RELAY.ARPA "If the game is lost, we're all the same; No one left to place or take the blame"
reid@glacier.ARPA (Brian Reid) (03/12/86)
Baloney, Greg. Leaving out one site means nothing. Leaving out a backbone site means nothing. For that matter, leaving out 100 sites or 20% of the backbone, means nothing. If a significant fraction of the net (perhaps more than 30%) cannot run the Bourne shell, then it is perhaps worth worrying about making a version of this data-gathering scheme that uses some other shell. As far as I am concerned, this data is already pretty useful with only 1% of the net reporting in. By the time 30% of the net reports in, almost any 30%, I believe that the statistical quality of the readership data will be so much better than any other metric ever applied to the network that nit-picky objections about how this or that site can't/hasn't report in will be vacant. The biggest single problem with the network in its 6 years of existence has been that the loud, angry users get all the attention. What I am doing is collecting data from and about people who would otherwise never respond. I now have information from 1000 people's .newsrc files, and complaints from 7 people that my survey is unfair because it didn't handle their wierd special case properly (can't run csh, didn't find my .newsrc because I keep it in a funny place, doesn't count because I use home-made shell scripts for reading news and they don't update .newsrc, etc.) I have no interest in hearing from all of the hackers. I want to hear from and about the people who would never dream of arguing about things like this, but who are net readers. I claim that my shell script, even if it won't run on your machine, is picking up that data. This data is not perfect. I'll grant that. It might not even be accurate to within 20%. But it is 100 times more accurate than any other data anybody else has. Let's collect this round of data, and look at it, and then talk about ways of making marginal improvements on the data-gathering techniques. Brian Reid Stanford (soon to be Brian Reid DEC Western Research Laboratory) -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
woods@hao.UUCP (Greg Woods) (03/14/86)
> Baloney, Greg. Leaving out one site means nothing. Leaving out a backbone site > means nothing. For that matter, leaving out 100 sites or 20% of the backbone, > means nothing. FIne. As long as you make this clear to everyone who would interpret your results, then I have no problem with it. > If a significant fraction of the net (perhaps more than 30%) > cannot run the Bourne shell, then it is perhaps worth worrying about making > a version of this data-gathering scheme that uses some other shell. And assuming, of course, that there is nothing you would like to find out about that is related it *any* way to running or not running the Bourne shell. > reporting in. By the time 30% of the net reports in, almost any 30%, This is a typical fallacy. It certainly DOES matter which 30%. After all, the soapbox groups that hao no longer carries account for nearly that much of the net traffic. > I believe > that the statistical quality of the readership data will be so much better > than any other metric ever applied to the network True but irrelevant. All the other 'metrics' have been virtually non- existent. Better than nothing isn't saying much. > The biggest single problem with the network in its 6 years of existence > has been that the loud, angry users get all the attention. Not always true. Would you describe Spaf as loud and angry? :-) > What I am doing > is collecting data from and about people who would otherwise never respond. I applaud your effort and I support you 100%. Just don't go too far and uphold your data as representing the whole network when you have a pitiful 2% of the data. A lot can happen in 98%. You first have to demonstrate that there isn't a correlation between those who do/don't respond to your survey (for whatever reason(s)) and whatever it is you are trying to observe. > I now have information from 1000 people's .newsrc files, and complaints from > 7 people that my survey is unfair because it didn't handle their wierd > special case properly ...and God knows how many more who didn't complain or whose complaint or even a legitimate response got lost in the morass of uucp mail. I also do not consider inability to run one particular shell a 'weird special case'. You are clearly biased, and that bias is likely to be reflected in any results you come up with. At least post a C program for Chrissake, if you claim to represent a network of totally varied UNIX sites. C is about the only thing close to a standard that exists. And even that has it's problems.... > I have no interest in > hearing from all of the hackers. I want to hear from and about the people who > would never dream of arguing about things like this, but who are net readers. If you are claiming to represent the whole network, why does it make any effing difference what YOU are interested in? The hackers are a significant portion of the readership! > I claim that my shell script, even if it won't run on your machine, is picking > up that data. I challenge you to demonstrate that this is the case. I congratulate you on starting the effort. No one else has even bothered to try, and you deserve credit for that. But let's not get carried away. Any data gathering scheme that depends on anything more than a C compiler that can compile STANDARD C (i.e. no 20-character identifiers) can hardly be considered as representing this entire network. > This data is not perfect. I'll grant that. It might not even be accurate > to within 20%. But it is 100 times more accurate than any other data anybody > else has. Let's collect this round of data, and look at it, and then talk > about ways of making marginal improvements on the data-gathering techniques. We need more than 'marginal' improvements. 1.4% is PITIFUL. I do grant you that it's orders of magnitude better than anything we've had previously, but that doesn't justify holding it up as representing the entire net. P.S. Are you and I the only ones in on this? What does everyone else think? Any concrete suggestions from the statistician types out there as to how we might actually go about collecting a representative sample? > Brian Reid > Stanford -- {ucbvax!hplabs | decvax!noao | mcvax!seismo | ihnp4!seismo} !hao!woods CSNET: woods@ncar.csnet ARPA: woods%ncar@CSNET-RELAY.ARPA "If the game is lost, we're all the same; no one left to place or take the blame; Will we leave this place an empty stone, or a shining ball of earth, we can call our home"
flaps@utcs.uucp (Alan J Rosenthal) (03/20/86)
Umm.. maybe I'm missing something obvious here but it seems to me that there is a big problem with these statistics.. I don't read nearly all of the groups in my .newsrc. My .newsrc contains the most number of groups I've EVER managed to catch up in. But they are in a careful order, and I rarely get to the end while reading. So, for example, I rarely read net.unix-wizards but am always subscribed to it. So wouldn't I be counted for that too? Alan J Rosenthal {linus|decvax}!utzoo!utcs!flaps, {ihnp4|allegra}!cbosgd!utcs!flaps
reid@glacier.ARPA (Brian Reid) (03/25/86)
In article <1165@utcs.uucp> flaps@utcs.UUCP (Alan J Rosenthal) writes: >Umm.. maybe I'm missing something obvious here but it seems to me that there >is a big problem with these statistics.. I don't read nearly all of the >groups in my .newsrc. My .newsrc contains the most number of groups I've >EVER managed to catch up in. But they are in a careful order, and I rarely >get to the end while reading. So, for example, I rarely read >net.unix-wizards but am always subscribed to it. So wouldn't I be counted >for that too? No. The definition of "Person X reads group Y" is that person X's .newsrc file shows that he has read (or marked as read) at least one message in group Y that has not yet been expired. If "expire" is run with standard options, this means that person X has read at least one message in that newsgroup in the last two weeks. If your site expires more rapidly than two weeks, then a more stringent test will be applied. This works by looking in the (2.10.2 and newer) active file to find the "oldest unexpired message" and the "most recent message" numbers. Let N be the number of the highest-numbered message in the .newsrc file that is shown as having been read; you will be counted as reading the group if N is greater than or equal to the oldest unexpired message number, less than or equal to the most recent message number, and if the most recent message number is greater than the oldest unexpired message number. This algorithm could, of course, yield an erroneously large number of subscribers for a group because of the "catch up" commands. However, what the surveys show is that far LESS people read the newsgroups than anybody had previously thought. For example, only about 3% of the sampled population even bothers to list net.unix-wizards in their .newsrc at all. -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
eppstein@garfield.columbia.edu (David Eppstein) (03/25/86)
In article <5641@glacier.ARPA> reid@glacier.UUCP (Brian Reid) writes: > ... The definition of "Person X reads group Y" is that person X's .newsrc > file shows that he has read (or marked as read) at least one message in > group Y that has not yet been expired. If "expire" is run with standard > options, this means that person X has read at least one message in that > newsgroup in the last two weeks. If your site expires more rapidly than two > weeks, then a more stringent test will be applied. This brings up a problem with this survey that I have been wondering about for a week or so. Some groups have very low volume, such that it is possible for no articles to be current in the group when the survey is run. If that were the case, the survey would show no readers when in fact many people may read the group. As a practical example, one of my favorite groups happens to be net.roots. We expire news weekly, and in many weeks there are no messages in net.roots. Yet I would object very strenuously to being called a non-reader of the group. A possible solution would be to also count people who have caught up entirely with a group even if there is no message left unexpired in that group. This would still have some inaccuracies, but they would perhaps be counterbalanced by the people who haven't yet bothered to unsubscribe to the groups because they are so low volume. Perhaps the survey already does this, but there is no indication of that in the included description. -- David Eppstein, eppstein@cs.columbia.edu, seismo!columbia!cs!eppstein (note that the garfield in my headers is *not* the one in the UUCP map)
wmartin@brl-smoke.ARPA (Will Martin ) (03/26/86)
One other query about the poll: What effect, if any, is there from hosts that do not permit access to all newsgroups? I have no idea if this is a very rare situation, or one relatively common and becoming commoner, or somewhere in-between... For example, on this host, access is limited to groups considered "work-related" or somehow justifiable on government- owned equipment. However, to be a "good neighbor" and to act as a full-fleged participant in the network community, all groups are allowed in and passed on downstream -- it is just that protections are set so that the local user community cannot get to the "restricted" groups. While this situation is sad, that is the way life is, considering GAO and Army Audit inspections and the regulations on computer usage. Now, if the "arbitron" polling is a test of popularity, I would like it know that I really like and would read certain of the groups that I cannot get to. They are still in my .newsrc, but I've shifted them to the end, and they show up in an "rn" "L" listing as so many postings unread. I can't read them now, but I would if I could. Does some similar situation exist at many other net sites? Or is this so unusual that there is no point in paying attention to it, since it would be an extreme anomaly? (As far as I know, "arbitron" has not been run at this host, so it could not have affected the so-far published results. But what would be the effect on the statistics if this was common?) One thing that makes me wonder this is the relatively low rating of "net.consumers". I can't read that now, but it was the first group in my .newsrc back when all groups were available. Since *everyone* buys consumer goods, I can't imagine *anyone* having access to this group yet not reading it. But it seems that it is down below other groups of much less interest. Could it be that there are some artificial skewings of the data due to "work-related" groups being accessible on more machines (or by a larger group of users on some single machines) than these "personal-interest" kinds of groups like net.consumers? This is not necessarily a wrong or bad thing, now -- if people cannot get to the groups, even if they would want to, it is probably more realistic to rate the groups by the real number of people getting to them, instead of the number of people who *want* to get to them (which would include both those that could and those that could not). Anyway, I haven't seen any mention of this factor, so I thought I'd bring it up. Regards, Will Martin
reid@glacier.ARPA (Brian Reid) (03/28/86)
In article <1425@garfield.columbia.edu> (David Eppstein) writes: >This brings up a problem with this survey that I have been wondering >about for a week or so. Some groups have very low volume, such that >it is possible for no articles to be current in the group when the >survey is run. If that were the case, the survey would show no >readers when in fact many people may read the group. The first version of arbitron that I posted took care of this. That is why, for example, mod.movies shows up as being so popular. It counts news readers for groups in which there is no recent traffic. I thought this was OK, but a problem arose. Some sites do not get all groups, and they incorrectly have the unsubscribed groups still present in their active files. While technically this is an installation error for netnews (you shouldn't have a group in your active file unless you actually exchange that group with your news partner), we all know how fruitless it is to get people to fix bugs in their news installations. If some site does not get net.abortion, and never did, then it will look as though everybody reads it. It is in the area of low-readership, low-volume groups that the measurement is least accurate. Although a few people have flamed me for the measurement not being as accurate as they would like, nobody has come up with a better technique, so I propose to keep on doing this until we figure out better things to measure. -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
gam@amdahl.UUCP (G A Moffett) (03/29/86)
In article <5784@glacier.ARPA> reid@glacier.UUCP (Brian Reid) writes: > Although a few people have flamed me for the measurement > not being as accurate as they would like, nobody has come up with a better > technique, so I propose to keep on doing this until we figure out better > things to measure. Indeed, this survey of the net provides us with our first major collection of data regarding netnews readership, and also challenges us to decide what methods we will use in the future to more accurately measure it. So: don't flame, think! -- Gordon A. Moffett ...!{ihnp4,seismo,hplabs}!amdahl!gam Inferior people should not be employed.
ben@catnip.UUCP (Bennett Broder) (03/31/86)
In article <5784@glacier.ARPA>, reid@glacier.UUCP writes: > In article <1425@garfield.columbia.edu> (David Eppstein) writes: > I thought this was OK, but a problem arose. Some sites do not get all > groups, and they incorrectly have the unsubscribed groups still present in > their active files. While technically this is an installation error for > netnews (you shouldn't have a group in your active file unless you actually > exchange that group with your news partner), we all know how fruitless it is > to get people to fix bugs in their news installations. If some site does not > get net.abortion, and never did, then it will look as though everybody reads > it. Sorry Brian, but as a site that does not get all groups, I can tell you that there is a problem with your suggestion. If I remove net.abortion from my active file, every time an article comes into my machine with net.abortion in the newsgroups line (because it was crossposted with a group my site does receive) inews will put an error message in the errlog file. And since my .profile checks the errlog file when I login, this would be very annoying. Other than confusing your arbitron script, can you think of any reason *not* to have net.abortion in my active file? -- Ben Broder {ihnp4,decvax} !hjuxa!catnip!ben {houxm,topaz}/