grep www.distilled.net-access_log -e 'Googlebot' | grep -e '404‘ grep www.distilled.net-access_log -e ' 404 ' | grep -e 'store'

Using and abusing data Setting us apart from other marketers Will Critchlow, 2011

This presentation is about skills and things to try… Understanding data is all that separates us from the savages

Some skills I firmly believe technical skills make you a better SEO

Gathering data: there are loads of places it could come from, but you should be adept at accessing it. Check out http://www.infochimps.com/

There are loads of great APIs: http://www.seomoz.org/blog/api-and-dataset-cheatsheet-building-quick-dirty-tools

Sometimes you need to go get your own: with varying levels of difficulty. Check out http://www.mozenda.com/, http://scraperwiki.com/

Rapid prototyping FTW: I’m talking *really* Heath Robinson

On statistics The modern world runs on math. Learn just a little bit

We are good at spotting (and extrapolating) trends Go, go, go

But randomness often looks like a trend Uh oh

And sometimes, we are just wrong PANIC!!1

A 1% mis-diagnosis rate can result in more errors than correct diagnoses

Pie charts are often hard to be read and can easily be made misleading http://dis.tl/m6sZLO

Watch out for charts that don’t go to zero on the y-axis http://dis.tl/iH1qgs

The same data presented in a less misleading way

Action: read “How to lie with statistics” http://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/0393310728 Old, but great

Don’t blindly mine data for patterns (you’ll definitely find something). Have a hypothesis Test on different data to that used to form the hypothesis Don’t forget that a 95% confidence interval is wrong 1 in 20 times Common statistical misconceptions / mistakes

Visual =Powerful

Geckoboard: build a culture of data - http://www.geckoboard.com

Visible data and clear targets work: we had a modest internal target to hit 2,000 followers of @distilled by our #linklove conference Target set Target ended Source: twittercounter.com

One of our clients built a tool to visualize website changes over time: http://www.reevoo.com / https://github.com/georgebrock/timelapse

Public = powerful

Coming soon: public post analytics on SEOmoz

Back to getting the data

=importxml("http://www.google.com/search?gl=us&q="&A2, "//h3[@class='r']/a/@href") We’ve written about the power of Gdocs for rapid prototyping a bunch of times – see for example: http://dis.tl/kZsv9y and http://dis.tl/jqk05J

Another xpath resource: http://developer.yahoo.com/yql/console/

Command line

Linux: native command lines make a lot of this easier. The latest Ubuntu and VirtualBox are a piece of cake to install and get using

VirtualBox: take snapshots really frequently

Unix principle: lots of small tools, piped together – great for ad-hoc or to create specifications grep * -r -e ‘UA-XXXXXX’ | sed -e ‘s/UA-\([0-9]*\)/UA-XXXXXX/g’

Inspiration: via @tomcritchlow - http://www.neilkodner.com/

Real programming

apt-get install python-virtualenv virtualenv --no-site-packages <path> source <path>/bin/activate pip install gdata http://code.google.com/p/gdata-python-client/ http://code.google.com/apis/analytics/docs/gdata/gdataExplorer.html

https://gist.github.com/967503 (~/Downloads/googleanalytics.py)

Training: I really like the peepcode videos for learning technical stuff. Recommended: command line, git, vim. http://peepcode.com/

STOP

If you start catching the bug, go learn a framework. I like django (https://docs.djangoproject.com/en/1.3/intro/tutorial01/)

Stackoverflow is your friend – everything you can’t get answers to here is a copy and paste error guaranteed(*)

Time series

You come across data that looks like this all the time Day Date Visits 1 Monday 05/04/2010 877 2 Tuesday 06/04/2010 1087 3 Wednesday 07/04/2010 1018 4 Thursday 08/04/2010 1039 5 Friday 09/04/2010 917 6 Saturday 10/04/2010 670 7 Sunday 11/04/2010 746 8 Monday 12/04/2010 1165 9 Tuesday 13/04/2010 1192 10 Wednesday 14/04/2010 1053 11 Thursday 15/04/2010 1022 12 Friday 16/04/2010 947

Your first action: eyeball the data

Your action: Fire up R - http://www.r-project.org/ > pg = read.csv("<path_to_file.csv>") > tspg=ts(pg[,3],start=1,freq=7) > plot(stl(tspg,s.window="periodic"))

Your action: decompose time series into its constituent parts

The “trend” is useful: this removes seasonality and outliers

The “remainder” is the outliers: might correspond to activity

Videos are now available of our #linklove conference in London and New Orleans: www.distilled.net/store