50 Pointers on Web Analytics

Economic analysis depends on data - often imperfect. Google Analytics (GA), FeedBurner, AddThis, FeedJIT, Facebook Pages, AdBrite, and AdSense (AS) provide different views. The metrics often don't match. Here is the complex flow. Below is a dump of got-cha's.

Realtime Reporting

FeedJIT, AS, MyBlogLog from Yahoo report immediately. Others are delayed - usually daily.

GA for Impressions, Visits, and Visitors

• GA tracks who views which page. 
  
• Impression=Page; Visit=Session; Visitor=People/Device.
• A page is requested, downloaded, and rendered. What is counted?
• Web logs count requests.
• An impression is a rendered page. Caching eliminates downloads and vary stats. 
• Filtering (e.g. excluding the owner's views) can vary stats.
• GA is slow. If the user leaves before full rendering, the impression does not count. Undercount could be 10% or more.
• A visitor (i.e. unique visitor or person) has a cookie that identifies the visitor over time. 
• The cookie tracks a device. If the device is used by many, the visitor count is inaccurate. Windows does allow many user accounts, but few use this feature.
• If the cookie is deleted, visitors are over counted. 
• Wide adoption of OpenID will help, maybe.
• A visit or session counts impressions over the session duration. A session ends if there is no activity from the user for a fixed time period, like 20 minutes. 
• If the user views one page, goes idle or visits elsewhere for an hour, and views a second page - this is counted as two visits with 1 page per visit. 
• If the user views a page, goes to another site, and returns within 20 minutes - this is one session, 2 pages, and an inflated visit duration. 
• The term hits should be avoided since early web analytics counted every file. Modern pages mash dozens, if not hundreds, of files per page. Hits inflate rapidly.

GA Traffic Sources

• Direct Traffic - can be one of the following:

1. Domain typed into the browser's address box
2. Links in Outlook email, PDF, Word, or other similar document
3. Click on a built-in RSS reader of a browser or desktop application
4. Browser's referring option has been turned off - mobile device with WAP?

• Referring Site 

1. Links from another site. 
   
2. Includes clicks on RSS feeds, which duplicates the reach number from Feedburner
   

• Search Engines - source by portal and keyword. Is this biased toward Google? Who defined the standard?
  
  
• Ad Campaigns - sources of visits by ad words, campaigns, and versions

Sequential Goals versus Deep Linking

Early websites try to guide users from the home page through a series of pages. GA's goals feature supports this analysis. 

Modern websites have dozens of links per page and thousands of entry points. The user path is non-sequential and goals become an irrelevant metric. 

Further, the common practice to cross link and share conversations creates deep links as opposed to home page links. The Flypaper effect also increases deep links. 

Subscribers, Reach, and RSS Robots

Readers can subscribe using RSS feeds. Feedburner tracks these uses.

• Rather than pages, RSS feeds a set of pages formatted as XML. 
• Subscriber approximates visitor; Reach approximates impression.
• Subscriber does not count subscriptions. It counts requests for the XML feed. The count depends on the frequency of use of the RSS readers. A reader can subscribe to hundreds of feeds.
• RSS cannot track subscription deletions.
  
• Subscriber is not a visitor. A subscriber may view zero, one, or many pages of a feed using the local RSS reader.
• Reach is a click that brings readers to a web site. This duplicates GA counts of referral and direct sources.
• Feedburner uses cookies to count visitors. Thus cookie problems compound errors.
• Multiple subscriptions from one user (e.g. Yahoo mail, myYahoo, iGoogle) count as one subscriber.
• Feedburner recycles the day at 11pm, PST. There is no means to set the local time. Thus, Feedburner stats by day will be substantially from GA and AS.
• Subscribers do not include robotic subscriptions from Facebook, FriendFeed, and others.
• Hits from robots don't distinguish multiple subscriptions at Facebook, FriendFeed, etc.
• Feedburner inter-site link tags increase the reach count.
• Feedburner tags copied to another site also inflate the reach. It duplicates GA stats.

The FeedBurner Stats service for websites provides the following information for your blog or site:

• Visitor summary and detail
• Page summary and detail
• Visitor and referral trends
• Inbound referral traffic breakdown
• Outbound click breakdown

AdSense Counts Ad Metrics

What's the difference between ad, ad unit, and page impressions?

Our advanced reports now provide the option to view your performance by page, ad unit, or individual ad. We've highlighted the differences between each report below.

• Page reports will show an impression every time a user views a page displaying Google ads. We'll only report one page impression no matter how many Google ads are displayed on a page. For example, if a page with one half banner and one vertical banner is viewed once, we'll display one page impression.
• Ad unit reports will show an impression every time a user views a Google ad unit on your page. For example, if a page with one half banner and one vertical banner is viewed once, we'll display two ad unit impressions. 
• Link units link a category to a list of text ads.
• Individual ad reports will show an impression for each individual ad that's shown in any ad unit. For example, if a page with one half banner and one vertical banner is viewed once, it will generate three ad impressions.

Because ad and ad unit reports will show a higher number of impressions than reports on the page level, you can expect to see a lower eCPM when viewing reports by ad or ad unit.

• Many ads display per page. AS calls them individual ad impressions. 
• Also many adNets can serve one page.
• Ad units spec the box for an ad. This is similar to panels used by widgets. 
• Channels associate a name with an ad unit. Adbrite calls them zones. 
• Channels can track the ads on many websites using one AS account.
• Channels not integrated with Blogger and Feedburner ad units.
• Browsers like Safari and Firefox disable adNet cookies. This defeats BT and over counts visitors.
• AS click counts, CPC, and page impressions convert to eCPM.
• Robotic fraud detection counts clicks, but don't report the CPC. Thus, clicks can have zero revenues.
• AS also delivers CPM ads at fractional pennies. Thus, penny eCPM's show revenues from CPM ads. 
• Reports don't show transparency of CPC versus CPM ads. 
• Ad buyers bid maximum CPC. Pay one penny more than the next highest CPC.
• Minimum bid is one penny per click.
• Complex algorithms to maximize CPC bids, CPM bids, CPA bids, text versus image, and other options. Do they benefit affiliates?
• Ad buyers asked to improve their CTR ratings to receive more impressions. Can they?

Emerging Web Analytics

• FeedJIT and MyBlogLog report real-time outbound clicks. GA and Feedburner do not.
• Facebook profile, newsfeed, and applications
• Video viewing patterns - abandon points
• Game playing - 
• Widget viewing as superset of Feedburner RSS widgets
• Yahoo and Microsoft provide similar web analytic services. So does Hitbox.

Conclusion

Too much data. Overlapping. Incomplete. Needs changing rapidly. 

"Blind men with an elephant." e.g. Google reports $4 billion per quarter mostly from CPC ads; WSJ claims 500 million clicks per month; implied CPC is $2.00+. Efficient Frontier says average CPC is less than $0.60. Huh?

Additions, corrections, comments welcomed.

---

What is your web analytics communication strategy?

Judah’s recent post titled “what does your web analytics team look like” reminded me of something that has been on my mind a lot since I presented my Web Analytics: A Day a Month webcast for the American Marketing Association last month. As I travel the world talking about web analytics to companies of all shapes and sizes, one thing I’m struck by is the number of differences in how companies approach sharing web analytic data and information.

It’s not as if there is any one “right” way to communicate about web analytics, but it is clear that there are many, many wrong ways to do it. But rather than dwell on wrongness, I prefer to focus on rightness so here are a few thoughts on developing a clear strategy for communicating web analytics.

This post may seem pretty basic to many of you, but if it does I would encourage you to ask yourself these questions:

• What decisions are web analytics driving in your organization?
• Are those decisions largely tactical or are they truly strategic?
• Do you feel like most people in your organization understand web data?
• Are you producing reports that are going under-used, unused, or are flat out being ignored?

Ed: From Web Analytics Demystified. Oxymoron.

Who Cares How Many Subscribers You’ve Got?

To truly understand and address what’s driving your success it is necessary to understand the web of relationships between the different determiners which lead to the outcomes that you are looking for. The diagram below shows the network of measurable items which make up these relationships, showing how each is interconnected.