Archive for July, 2008

GA: Why do pages refer to themselves?

Posted on July 29, 2008 by Robbin

Content - Navigation

About a week ago, I read a post by Avinash that answered GA questions; but when I got to the part about the navigation report (see screen shot, left), I just didn’t agree. The question was, “Navigation summary question – why is previous and next page often the same as the page you are viewing? ” Like this report on the left: Notice that 6.23% of pages that lead to the index page come from the index page, and 6.23% of pages that come from the index page go to itself. A little strange, no?

Why I was suspicious of the original answer.

In his post, Avinash wrote that someone at GA explained what caused this peculiar beharior. Here is how he described it — basically, it is about viewers that look at a regular tagged page and then look at a picture on the page in larger format (which isn’t tagged). Here is the example he gives:

Visitor Action One (view): /avinash/2007/09/rethink-web-analytics-introducing-web-analytics-20.html
Result: javascript hit generated (data collected)

Visitor Action Two (click): http://www.kaushik.net/avinash/wp-content/uploads/2007/09/web_analytics_1.0.png
Result: NO javascript hit generated (no data collected)

Visitor Action Three (back): /avinash/2007/09/rethink-web-analytics-introducing-web-analytics-20.html
Result: javascript hit generated

Visitor Action Four (click): http://www.kaushik.net/avinash/wp-content/uploads/2007/09/web_analytics_2.0.png
Result: NO javascript hit generated

Visitor Action Five (back): /avinash/2007/09/rethink-web-analytics-introducing-web-analytics-20.html
Result: javascript hit generated

To Google Analytics (or any other Analytics tool), it will look like this:

1) /avinash/2007/09/rethink-web-analytics-introducing-web-analytics-20.html - javascript hit generated

2) /avinash/2007/09/rethink- web-analytics-introducing-web-analytics-20.html- javascript hit generated

3) /avinash/2007/09/rethink-web-analytics-introducing-web-analytics-20.html - javascript hit generated

</Avinash>

This sounded plausible, but too neat. Much too neat for me. What if someone got to one of those pictures – one of those untagged .png pages – and decided to leave the site altogether? If just a single person bailed out, that would make the percentages different. In order for this explanation to work, every single person would have to exhibit the identical behavior – they would all have to look at two pictures and come back to the same page. It has to be perfectly symmetrical, and it is in the hands of thousands of humans to do it the same way.

Do you believe that? I didn’t. But I didn’t know the answer.

The Truth According to John (aka Google Analytics Gang Signing)

So yesterday, I was working with John and Jonathan here at LunaMetrics. “Did you see Avinash’s post a week ago?” I asked them, “Those numbers are WAY too clean. How could a page refer to itself and then refer to itself again every single time?”

John thought to himself for a couple of minutes and then said, “Oh, I get it. Here is what happens. Whenever the page is viewed twice in a row – like a page reload — the whole thing automatically works.” He put his hands together in the configuration on the left. Jonathan nodded wisely. I looked at them like they were nuts.

But ultimately, I understood what he meant:

If a page precedes itself, it also follows itself. That’s what John meant with his fingers — on one side of the report, we see a page preceding itself, on the other side of the report, we see the page following itself. It is just the same story, told twice.

The key is, you can’t think of that report like a clickstream when it involves the same page more than once. Once you stop thinking about it that way, it becomes intelligible. The page is the same no matter which of the columns of the report it appears in, and the numbers have to match exactly because of that.

Still lost? I know that some of you are sitting there nodding your heads, while others are saying, “What is she talking about?” So for the latter crowd, let me describe it in a different way. I hope you won’t mind if I use numbers instead of percentages, just to make this clearer.

Let’s say that Page A refers to itself via a page reload 100 times. And let’s say that the website has only one page — Page A. The report would look like this — in a conceptual way:

Notice how we get 200 pageviews in the middle of the page (and we know that that’s how many there are.) Notice how the number of pageviews on the left and on the right are symmetrical. And notice how these are two identical pictures, which meet in the middle — just like the picture of John’s hands above.

So I think I have run out of ways to explain this problem. It is sometimes caused by a reload, and sometimes caused by part of the explanation that Avinash gave. But it never requires thousands of people to exhibit the identical behavior.

And in closing, John wanted me to show off that he is really known for his good looks and not for his gang signs, so here is he is.

Robbin

View Comments (19 Responses) | Categories: Google Analytics

The Dark Knight: A movie-lover's lesson in Web Analytics

Posted on July 18, 2008 by Robbin

My daughter really wants to go see the opening of Mama Mia! this weekend. But, while a good play, it is not exactly what I am looking for in a movie. On the other hand, I am dying (heh) to see Heath Ledger’s posthumous appearance as the Joker in The Dark Knight.

“It got a ‘Must Go” rating on Fandango last night,” I pointed out to her. Ever the analysts daughter, she retorted, “And that was out of three people?”

Well, no, that were actually hundredds of people who succeeded in seeing it before it opened (Hmm, maybe it opened in other countries in other time zones, the way that you could get an iPhone in New Zealand almost a full day earlier than here.) But it got me to thinking. If there are only five rankings: Must Go, Go, So-So, No and Oh No! — then how does anyone ever achieve a ranking at the ends of the scales? It’s like asking someone to take a survey and they can choose a number between 1 (lousy) and 5 (awesome) — unless everyone chooses a 5, how does anyone end up with an average of 5?

“Aren’t you assuming a lot about Fandango’s algorithm?” asked John Henson, famous creator of the GA Goal Copy tool. “Maybe it’s like the Google ratings,” pointed out SEO Jim Gianoglio, “They count more if you not only rate but also write a review.”

Well, analytics to the rescue. If you click through, you can actually see the rankings in buckets (sort of like the Google Analytics loyalty charts, but without all the misleading titles):

Obviously, you don’t have to get all “fives” to get a five. So let’s expand the system and pretend that Fandango weights all answers on a scale of 1-10, and you have to get between a 9 and a 10 to score a “Must Go.” And maybe each vote gets the top of its category (so if you vote “must go,” it is worth ten points, and if you vote, “go” it is worth eight points. We would have (in my made-up algorithm):

45*2, 26*4, 63*6, 96*8 and 991*10

all of which gets divided by the number of votes, 1221. For a weighted average, i.e. raking of 9.21376 (OK, that is a little overly precise given that I don’t know the algorithm.)

Late note: After publishing, I realized that this (made up) algorithm only works at the high end. What if you had a lot of Oh No! and a scattering of other rankings — if you gave a “two” to an “oh no!” ranking, you could never get a movie to rank, overall, as an “oh no!” So probably it is more of a sliding scale — but the concept is the same.

Well anyway, that is your web analytics movie lesson. Enjoy the weekend. Comment when you see the movie and tell me if I should go.

View Comments (4 Responses) | Categories: Web Analytics

Conversion and PPC: Can you start small?

Posted on July 13, 2008 by Robbin

While some customers (and friends!) are ready to go out and spend gobs of money on their pay per click campaigns, I don’t usually hear that. More often, I hear, “I’d like to start very small, learn what works and what doesn’t, and then roll out in a large way.” It sounds like a great idea, but it doesn’t usually seem to work — in my opinion. Traci Scharf, our pay per click (PPC) specialist, disagrees, so after I write, she is going to do a rebuttal (and you’ll get to see both sides of the problem.)

Oh, before I get into this in a big way, let me not forget:  our next Google Analytics training is August 12 in Washington DC, it costs $285, and you can read more and register here.

Robbin’s opinion: It’s hard to start small with Pay Per Click:
So you want to start small with your pay per click campaigns and roll out after you know what works? Here has been my experience:

To make the numbers easy, let’s say that a click for the customer we are working with is $1.00, and his conversion rate for visits to the site is .5%. May be he wants to spend a million dollars eventually, but up front, he is starting with $100/day. He starts wtih 500 keywords and multiple ads.

If a click costs $1.00, and his budget is $100/day * 30 days, he has a monthly budget of $3000, i.e. 3000 clicks/month. With a .5% conversion rate, that’s 15 orders.

Those 15 orders will be spread over many keywords. There may be two keywords that got three conversion each, and three keyword that got two orders each, and another three keywords that got one each. Some of those may be branded keywords, too, like “LunaMetrics conversion rate.” When someone does a branded search, they are already looking for you (a topic for a different post.)

So what can we learn from this? My answer: just about nothing. We don’t have enough data to be able to say, “This keyword does well,” or “This keyword does poorly.” If everything is coming from the same AdGroup or campaign, we may be able to learn more there, but my experience has been that we generally learn what we already know — which products, or which areas of the site, draw the most visitors. Which products tend to sell the most. Whereas real learning would be, “When we use exact match on these five keywords, we have a higher quality score and get better conversion for less money, but on this AdGroup, we can’t get the kind of traffic we need with exact match, and so we need to use a different kind of match” (for example.) Or even, “These keywords suck!! We have to retool this whole campaign.” Now, that’s learning.

OK, Traci, your turn.
Traci’s Opinion: It’s Really Important to Start Small with PPC
I’ve always felt that limiting your initial spending in a PPC campaign is a smart move for most businesses. Let me give you a little analogy:

Say you want to lose weight, and someone says to you, “Hey, you can lose a lot of weight on the peanut butter diet.” You might be willing to give it a try, but probably don’t want to invest too much of your time and energy until you know whether it works for you or not. A rational decision, then, might be to try it just for three weeks, and then get on the scale and see whether your weight went up or down.

When client companies say they want to learn small, this is in effect what they’re proposing. They’re saying, “Let’s spend a set amount of money, and see if our investment is getting a positive ROI.” Because, just like a diet (“Are you losing weight or aren’t you?”), PPC is pass/fail (“Are you making money or aren’t you?”). Maybe you know that, given where your initial budget is set, you need 15 conversions to break even on your PPC campaigns over the course of three weeks. If you find that your campaigns are getting 30 conversions over the course of three weeks, you’ve learned one big important piece of information: “You’re making money – so go ahead and increase your budgets.”

But probably the most compelling reason I encourage companies to limit their initial ad spend is because they’ll want to have enough money to act on what they learn. Consider the company that runs PPC campaigns for three weeks and finds out they are getting, on average, three conversions a week, but need to be getting ten/week in order to break even. Well, assuming I’ve done my job in setting up their campaigns to drive qualified traffic to their site, we will want to look at what is going amiss with their landing pages. That will mean doing a best practices analysis on their landing pages, and then creating and testing alternate versions so that we can transform their conversion rate. However, if they’ve already blown their whole PPC budget, there is no place to go from there, except to cut losses and admit defeat – not the best strategy for getting ahead!

So the bottom line is, be leery when anyone tells you that you have to spend a lot of money in the beginning months of your PPC campaigns. Exercise the same caution you would with anything else, and remember that you can’t just throw money at PPC and expect success.

View Comments (8 Responses) | Categories: Paid Search

Advanced Filters with Fields Required/Not Required

Posted on July 7, 2008 by Jonathan

On a post way back in April on Custom Advanced Filters, Idris left a comment asking about the required/not required selection (seconded by Paul):

Hey, great articles. I am trying to do some advanced filtering, but I’m confused by the “Field X Required” option. If I say “Yes” to the requirement, which of the following two things does that mean?

a. If the regex in this field does not match, do not include this hit in the profile at all.

b. If the regex in this field does not match, skip this filter, and move on to the next, but still include these hits in the profile.

These two are obviously very different. Which does Google Analytics do?

The confusion is about what exactly is “required”. We were pretty sure we knew, but we did an experiment to confirm. It’s basically b from what Idris suggested.

Here are the details:

1. If the field is required and the regex matches, the output is written to the field you select.
2. If the field is required and the regex does not match, the output is not written to the field you select.
3. If the field is not required, the output is written to the field you select regardless of whether the regex matches.

In no case are the pageviews excluded from the profile entirely (you need an exclude filter for that). The filter just doesn’t apply if the field is required and it doesn’t match.

View Comments (2 Responses) | Categories: Filters, Google Analytics