Comments on: Relationships are a thing of the past… http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/ Web Analytics Blog - Paving the way to understanding web data as it relates to statistics and other methodologies. Thu, 28 Aug 2008 15:52:03 +0000 http://wordpress.org/?v=2.1.3 By: Robbin Steif http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-13 Robbin Steif Wed, 06 Jun 2007 00:38:15 +0000 http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-13 Hi Wendi. Wouldn't it be a whole lot easier and more valid to look at bounce rates by keyword/source? Furthermore (and I am so not a statistician, that's why I read your stuff, to learn), wouldn't you really just learn about correlation and not causation? So bounce rates increased during those weeks when you ran that new campaign but maybe it was your competitor's great new pricing, everyone was aware of it, and as soon as they landed, they saw that they couldn't get the same price? Hi Wendi. Wouldn’t it be a whole lot easier and more valid to look at bounce rates by keyword/source? Furthermore (and I am so not a statistician, that’s why I read your stuff, to learn), wouldn’t you really just learn about correlation and not causation? So bounce rates increased during those weeks when you ran that new campaign but maybe it was your competitor’s great new pricing, everyone was aware of it, and as soon as they landed, they saw that they couldn’t get the same price?

]]> By: Wendi http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-15 Wendi Wed, 06 Jun 2007 14:09:26 +0000 http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-15 Hi Robbin, All very great points and questions. In a business setting data is not always "nice" in that you can't control for everything and you may run into confounding issues but if you know that 1. you have a semi-controlled environment (you didn't just launch a redsign of your site during the same time period) and 2. there is an underlying business relationship between the two factors then correlation makes sense. Of course you don't want to compare changes in wind speed to your online SEM campaigns even though you could possibly prove some kind of relationship, but it just wouldn't make logical sense. You are right in that correlation doesn't equal causation (and never will) but knowing that there is a statistical relationship and there is a business relationship (obviously you know that running an SEM campaign should ultimately affect your site traffic otherwise you wouldn't pay) you can make inferences from those relationships. I am not suggesting to completely drop your SEM campaign but I would certianly probe more into testing the existince of these new keywords in my campaign. And, yes I would certianly suggest looking into the details as well and take a hard look at bounce rates by keyword. For smaller campaigns this may be easier to go straight to but what if you added over 2000 new keywords. I would first suggest to see whether there was an impact at all as a whole. If not then move on, but if there was a slight correlation and linear relationship then work your way down the segment chain, first by source, then ad group, then keyword so that you can isolate the possible culprit. Thanks for the comments and thought provking questions. I hope I answered some of them. Wendi Hi Robbin, All very great points and questions. In a business setting data is not always “nice” in that you can’t control for everything and you may run into confounding issues but if you know that 1. you have a semi-controlled environment (you didn’t just launch a redsign of your site during the same time period) and 2. there is an underlying business relationship between the two factors then correlation makes sense. Of course you don’t want to compare changes in wind speed to your online SEM campaigns even though you could possibly prove some kind of relationship, but it just wouldn’t make logical sense. You are right in that correlation doesn’t equal causation (and never will) but knowing that there is a statistical relationship and there is a business relationship (obviously you know that running an SEM campaign should ultimately affect your site traffic otherwise you wouldn’t pay) you can make inferences from those relationships. I am not suggesting to completely drop your SEM campaign but I would certianly probe more into testing the existince of these new keywords in my campaign.

And, yes I would certianly suggest looking into the details as well and take a hard look at bounce rates by keyword. For smaller campaigns this may be easier to go straight to but what if you added over 2000 new keywords. I would first suggest to see whether there was an impact at all as a whole. If not then move on, but if there was a slight correlation and linear relationship then work your way down the segment chain, first by source, then ad group, then keyword so that you can isolate the possible culprit.

Thanks for the comments and thought provking questions. I hope I answered some of them.
Wendi

]]> By: Coremark Analytics » Relationships Take 2 http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-16 Coremark Analytics » Relationships Take 2 Tue, 12 Jun 2007 01:33:54 +0000 http://coremarkanalytics.com/blog/2007/06/03/relationships-are-a-thing-of-the-past/#comment-16 [...] promise, I wanted to take a deeper dive into the regression data posted in my previous discussion.  To bring you back to the topic, I was discussing the relationship between the percent of [...] […] promise, I wanted to take a deeper dive into the regression data posted in my previous discussion.  To bring you back to the topic, I was discussing the relationship between the percent of […]

]]>