Traffic Metrics - Handle with Care
Not everything that can be counted counts and not everything that counts can be counted.
- Albert Einstein
Traffic is the key mechanism of making money as a webmaster, so its importance can't be over-estimated. It needs to be measured, analyzed, studied, and managed-otherwise you don't know how you're doing, whether it's time to make changes, or evaluate the effectiveness of marketing, content, and design efforts. "Useful Web site statistics are not just numbers; they're tools for making decisions and taking actions. They have a single purpose-to make your Web site more effective in achieving its goals." [Elms 2000]
But, as common expressions such as "lying with statistics" or "garbage in-garbage out" suggest, counting and analysis are not always as straightforward as they may seem-even though webmasters often have a certain degree of blind faith in these statistics. Complacency is involved even at the most basic level: "Units resemble sports officials: the only time you really pay attention to them is when something stupid happens." [Steve Mirsky. Scientific American August 2000]
A Case in Point
Validity and reliability are the key attributes of a measure or statistic. Validity refers to the ability of an indicator to accurately measure what it's supposed to measure. Reliability refers to the quality of consistently producing the same results. If a thermometer gives the right temperature, that's validity. If it does it the same every time-high/low, Summer/Winter, indoors/outdoors-that's reliability. Although good data will have both, sources, measures, and instruments may have both qualities, just one, or neither (and in varying degrees).
An interesting IBM White Paper by Brian Tomz illustrates one way in which your seemingly valuable traffic data may be misleading or incomplete-seriously challenged in both the validity and reliability departments. Analysis of Web logs can lead to many inaccuracies since the majority of surfers come from large proxies, such as company networks or large ISPs. What's wrong with that? If you consider how Web logs are generated (see sidebar) there can be a lot wrong with it.
Two scenarios are highly likely: (1) many distinct visitors will visit your Web site from the same proxy and IP address; and (2) it is likely that a visitor's IP address could change during a single visit. Thus, visit calculations are skewed in both directions. If the first scenario is overwhelmingly common on your site, your visit numbers will significantly underestimate site traffic. Multiple distinct visits will be reported as a single visit. The opposite is true in the second scenario. If a visitor's IP address changes during his visit, the visit will be broken and reported as multiple visits.
CNET's Theresa Elms adds a further issue here, "Pages served by a proxy server may not appear in your access logs, even though the user perceives an interaction with what appears to be your Web site. If enough pages are served from the proxy server instead of directly from your site, a user may appear to have ended his visit (or session), when in fact he is still browsing content you created. Distortions in your visitor counts, page views, and session durations appear as a result." [Elms 2000]
Tomz goes on to say that proxy caching can also have a serious effect on your visit and traffic calculations. A visit session containing ten page views that is a result of a visitor going through a proxy may show anywhere from 0-10 page requests on the hosting server. Thus, potentially large amounts of data never make it beyond the proxy servers. Your traffic software must then do the best job that it can in trying to piece together a visit with only a fraction of the puzzle.
Does this mean that your visit numbers are meaningless? "No, not necessarily. Visit information calculated in this way is useful for trending and getting general estimates. Further, for many webmasters, it is their only currently available option for visit calculations." But you must keep the limitations in mind.
Unique visitor calculations are not an option when dealing with a Web log in its standard format. No information in the data lends itself to identifying individuals. A few traffic analysis packages will provide unique visitor calculations based on standard Web logs. However, this calculation simply reflects a count of unique IP address and user agent combinations. For the same reasons mentioned above, this data can be misleading. Thus, unique, new and repeat visitor calculations can only be accurately surmised with the addition of a unique identifier such as cookies or registered user IDs.
Not All Data Is Useful
Adult Webmasters' Connor Young says there are two commonly used methods of obtaining detailed stats about the activity of your Web site:
Log analyzers that collect stats after the fact and live scripts that collect stats as you go. Log analyzers are extremely accurate and usually generate statistics at one or two previously specified times throughout the day. Live scripts can be somewhat less accurate - as the code must be executed each time a visitor hits a page - but they provide up-to-the-minute stats and usually are easier to install than log analyzers. [Young 2003]
Log files contain a lot of data and log analyzer programs can generate even more from the basic stats. Heaping masses of traffic data can easily be generated automatically. There are three basic options for analyzing your site's log files:
- hosting service will often provide reporting, which may or may not be detailed enough
- install your own traffic analysis software
- outsource this task to a service provider that specializes in traffic reporting.
However, it's been estimated that maybe 90% of this analysis it is not especially useful. To use traffic data effectively, webmasters need to understand the nature of these measurements and concentrate on the 10% of it that is really useful.
eMage eMarketing's "The Metrics for Success" explains that the most basic site statistic-a hit-is also the most useless: "A hit is any request for a file a server receives. That includes images, sound files, and anything else that may appear on a page. One request for a page can result in many, many hits. Hits do not provide an accurate picture of the activity on a site and should never be a part of a marketing discussion."
Similarly, a number of other common indexes are really not terribly useful, except in special cases. These include: page views, browser breakdown, geographic demographics, and errors. Of course, it depends what you want you know. If you want to optimize your pages for the most popular browsers, or are developing a geo-targeting strategy, then it's whole different story. Nevertheless, as is often the case, these measures have inherent limitations-page views don't distinguish between unique and repeat visitors, and the geographic index doesn't tell you where your surfer is, only where their ISP is.
According to eMage, the really useful stats are visitors and unique visitors, referrers, site paths, entry and exit pages, as well as search engine referrers and keywords. Although "information is power" and it's desirable to have as much of it as possible, it's counter-productive to bog yourself down in a morass of useless or mildly useful data. This can easily obscure the real picture that your data is trying to tell you. Better to concentrate on the important indicators.
Conclusion
The upshot of all this is that webmasters need to understand the nature of their traffic metrics and the procedures used to collect them-otherwise they may find themselves waving a printout full of meaningless numbers in a fool's paradise, while basing important business decisions on them. Often, the key is to understand that a given metric necessarily has limitations-but you understand them and may take the data with a grain of salt:
You can't take measurements without standard units of measure. So what about all those log analyzers, network monitors, third-party site analysis services, and other products and services that count and report Web site statistics? They do what you must also do-use a combination of identifiers that, taken together, probably distinguish the items of interest. [Elms 2000]
Furthermore, just reliability in a metric can lend it utility even when it's seriously flawed from the validity point of view. Consider a broken barometer. Although it still goes up and down consistently, it never gives the right numbers. But, who cares if it's 1010 millibars today or not? The important thing with a barometer is whether it's going up or down and how fast-the relative values are important. You may not know what the precise barometric pressure is, but you still know if it's going to rain. Same with a broken thermometer-if it's always off by 10% then it even has validity so long as you always add 10% degrees to what it says.
So, all is not lost. Webmasters can work usefully with the kind of site activity data currently at their disposal. Interpretation, multiple indicators, and relative trends will disclose at least the broad outline of the picture. But an informed critical eye is important. Don't take a number at face value just because it appears to be "a number"-be sure you know what it actually means and how it came to be. Otherwise, you are likely to be bamboozled by your own traffic stats-making good decisions but based on flawed data, making for an inevitable "garbage in, garbage out" scenario. And that's not what you want.
References
Teresa Elms. "Web Statistics Primer," CNET Builder.com 11/23/00. http://builder.cnet.com/webbuilding/pages/Servers/Statistics/index.html
"The Metrics for Success," eMage eMarketing.
http://www.emage-emarketing.com/063002.html.
Brian Tomz. "Misleading visit and visitor metrics: The effects of proxies on traffic metrics and the benefits of unique identifiers," IBM White Paper http://surfaid.dfw.ibm.com/web/home/whitepapers/metrics.html.
Connor Young. "Get stats and step into the light," Adult Webmaster Magazine 05/28/03. http://www.theadultwebmaster.com/articles/030528_stats.phtml.
Web Logs
Most Web traffic analysis tools use Web logs as their data input. A Web log contains records about every interaction between a site visitor and that Web server. Thus, every time a visitor makes an un-cached request from that Web server, a record is written to the Web log. A standard Web log record contains several components: an IP address; a time stamp; a status code; the requested resource; the number of bytes transferred; the referring resource; a user agent string and potentially many additional fields. Traffic analysis software crunches these Web logs in an attempt to report the information in a format that will facilitate meaningful analysis.
Generally speaking, Web traffic analysis tools calculate a visit, by reconstructing hits in a Web log. The software scans the Web logs to find records with common IP addresses and user agent string combinations. If the matches make sense chronologically, the series of hits form a visit. Other rules, such as ending a visit after a sustained period of inactivity, are often implemented as well.
Source: Brian Tomz. "The effects of proxies on traffic metrics and the benefits of unique identifiers," IBM White Paper www.-106.ibm.com/developerworks.
Getting It Straight - A Traffic Metricon
- Committed Visitor Index
Ratio of Page Views to Visits for only very long visits, a very important metric for most sites because it combines Page Views and Time. If you are getting long Visits with light Page Views, in most cases something is wrong, or your pages are very long and content heavy. [Novo 2003]
- Committed Visitor Share
Percent of very long visits, similar in nature to Heavy User Share, only using time-based visits instead of Page Views. If you want visitors to stick around for a long time, you want to see this metric move higher. If you want them to be able to do their business efficiently and move on, you want this metric moving lower over time. [Novo 2003]
- Committed Visitor Volume
Percent of total Page Views on the site viewed by Visitors with very long visit behavior. For an advertising-driven site, this is a metric bound to get attention, because it speaks to the overall quality of a page view. If you have 100,000 page views and only 1% of them can be attributed to visitors who stick around for a while, in most cases you are attracting the wrong audience. [Novo 2003]
- Conversion Ratio
When we talk about conversion ratios, we are talking about simple math. When a surfer visits a site, goes through the tour and then purchases a membership, that surfer has then been converted into a buyer. Let's say that I send 1000 people to a sponsor. Out of those 1,000 people, five people purchase a membership to the site. Therefore, the ratio of people that purchased a membership is 5:1000 which would be written as 1:200. (Divide both sides by 5). [Muffy 2002]
- Cookies
Using cookies enables webmasters to log even more detailed information about how individual users are accessing a site, but as Internet privacy becomes more and more legislated this particular use of cookies will be curtailed if not banned. [eMage eMarketing]
- Heavy User Share
Percent of Visits involving very high page view counts. If you are targeting your visitors properly (and the site is easy to use) this metric should increase over time. If for some reason you don't want people to view a lot of pages in a session, you want this number to decrease. [Novo 2003]
- Hit
"Hits" are server calls from your browser software requesting a specific file; so if your Web page (1 hit to load) has 3 pictures (3 more hits) and 2 banners (2 more hits) a "hit counter" may show "6 hits" for what is really 1 'unique' visitor having 1 page view. If that surfer left your page and then returned, you'd have 1 unique visitor generating 12 hits (2 visits to your page) with 2 page views. Get it? [Yagielowicz 2002a]
- Hit Counter
Tool used to track the number of surfers that click onto your site. Usually provides other useful stats like the number of return visitors, geographic locations, weekly-monthly totals, and so on. [Cozy Frog]
- Impression
A misused term, often referring to "page views" when in fact it should mean the number of times an advertisement was shown. If two identical banners are on a page, that single page view would have generated two banner impressions for a given sponsor. If the banners were different (not promoting the same sponsor), you would have one page view generating one banner impression for each of two sponsors. [Yagielowicz 2002a]
- Page View
A page view means just that. Once again, it is not a true indication of how many different people are visiting your Web site, but it is a good way to judge how "sticky" (the ability to retain the interest of visitors) your Web site is. [Bloch 2002]
Quality
Conversions are most heavily influenced by two factors: the quality of your traffic, and the quality of the sponsor's offer. The quality of your traffic is determined by its source (TGP traffic will convert differently from Search Engine traffic, for instance) and by how well you target it to the sponsor. [Yagielowicz 2002b]
- Raws
If a surfer clicks a banner and visit's the sponsor, that is 1 "raw'"click, which at this point equals 1 "unique" click (or visitor). If the surfer then leaves the sponsor's site, but for one reason or another clicks the banner again and returns, then those 2 "raw" clicks now still equal only 1 "unique" click, as they represent the actions of one individual surfer. [Yagielowicz 2002b]
- Ratio
When a webmaster says they got a 50:1000 ratio, they mean that out of 1000 people who clicked on his sponsor banner, fifty surfers paid for memberships. Fifty out one thousand is a pretty high sales ratio. Unfortunately a ratio of three sales out of one thousand is a more realistic norm. [Cozy Frog]
- Referrers
URLs denoting the portals or Web sites through which another site is reached. [eMage eMarketing]
- Reject Rate: All Pages
Percent Visitors who viewed any page once and then left, this metric frequently ties to broad navigation or design issues. While focusing on top entry pages is more important in the short term because that is where the traffic is happening, this more global metric is likely to point to global design flaws in navigation or page layout. When you make global design changes, pay attention to this one - you want it to be forever falling. [Novo 2003]
- Reject Rate: Home Page
Percent of Visitors who requested the Home page and then left. If you have time to track only one thing, track this one (assuming your home page is the top entry page.) If you have other high volume entry pages, they should be tracked instead of or in addition to the Home page. [Novo 2003]
- Repeat Visitor Share
Percent Unique Visitors coming back over time. This can vary dramatically based on the Session Length variable and the period of time you are running you report for. Most sites want their visitors to come back, so the Repeat Rate is something you want to see rise over time. [Novo 2003]
- Scanning Visitor Index
Pages scanned in a one minute or less visit. The closer this index is to 1 page, the less useful people are finding your site - they're just reading one page and leaving. This could signal navigation problems (or poor targets from marketing). If you make significant navigation or design changes, look to this index to rise. If it falls, you have hurt your usability and should look for ways to improve it. [Novo 2003]
- Scanning Visitor Volume
Percent of total site page views completed in visits of one minute or less. Depending on what your site is for, you might want this percentage to rise or fall over time. For a content site, you probably want it to fall. For a commerce site, it depends on the business. [Novo 2003]
- Sign-up Ratio
Proportion of surfers that dish out cash to view your site, as opposed to the number of visitors that visit your site. Example: If 23 out of every 100 surfers that visit your adult pay site sign up for a membership, your Sign-up Ratio would be 23:100. [Cozy Frog]
- Take Rate
Percent of Unique Visitors engaging in a particular key action activity. This is a measure of how compelling your offerings are to the audience you get at the site, and how well you are marketing to them. [Novo 2003]
- Traffic
When someone says that their site receives a lot of traffic, what they mean is that their site gets a lot of visitors. Traffic is what drives the industry and is what can make or break a Web site. In order to be successful a Web site must have high amounts of high quality traffic or visitors. Webmasters must figure out how to get traffic, how to keep it, or how to send it somewhere else that will make them money. [Muffy 2002]
- Unique Visitor
A unique visitor is someone with a unique IP address who is entering a Web site for the first time that day (or some other specified period). When you divide the number of visitors by the number of page views, this can give an excellent indication of whether traffic is transient or is staying on your site. [Bloch 2002]
- Uniques
The number of visitors to your site. If four individuals visit your site on any given day and two of them click onto your site six times the same day, it will be referred to as four unique visitors for that day. [Cozy Frog]
- Visitor Engagement Index
This is Sessions divided by Unique Visitors, indicating the tendency for multiple sessions on the part of users. Unlike "repeat visitors," this metric gives you a feel for the "intensity" of repeating behavior. If you have a very targeted audience of the same people who come back over and over again, the index will be well over 1. If you have no repeat visitors, it will be very close to 1, meaning almost every visitor has one session. [Novo 2003]
- Visitors
Number of persons who visit a site. An individual who visits a site three times in one day is typically counted as three visitors. [eMage eMarketing]
Sources
Kath Blackwell. "Traffic terminology: what does it mean... and what should it mean to you?" Klixxx Magazine (April, 2002). http://www.klixxxx.com.
Michael Bloch. "Hits, uniques, and page views," XBIZ 12/28/02. http://xbiz.com/articles/index.php?article_idp=418.
"Industry Ebonics," Cozy Frog. http://www.cozyfrog.com/guides/ebonics/index.asp.
"The Metrics for Success," eMage eMarketing.
http://www.emage-emarketing.com/063002.html.
Muffy. "Online Porn Glossary," 2002. Webmaster Vault. http://www.webmastervault.com/tip-glossary1.shtml.
Dawn Elizabeth Yagielowicz. "Traffic terms 101," XBIZ 04/31/02. http://xbiz.com/articles/index.php?article_idp=196
Stephen Yagielowicz. "Understanding conversion ratio," XBIZ 01/28/02. http://xbiz.com/articles/index.php?article_idp=127.
Jim Novo. Drilling Down Project. 2003. http://www.jimnovo.com/metrics-definitions.htm.
|
Printer
Friendly Version