One day I was wondering whether anyone actually reads that blog. A thought about setting up Google Analytics crossed my mind, but legitimate concerns about users privacy made me hesitant. Then I wanted to try out Piwik, but installation and configuration requires some effort and I wasn't sure if I want to make that investment. Eventually I figured out that I should probably do whatever other R and python bloggers are doing. But first I had to learn what it is.
I quickly started drawing out solution based on
xml2. In my data storage, I obviously needed columns for blog address, community name, Google Analytics usage and Piwik usage. Some blogs might be present in both
R-bloggers and Planet Python, so maybe use two columns for community instead of one. Then I made hypothesis that companies might care about usage metrics more than individuals - to test it I would need to keep information whether blog is run by company or not. But first I would have to figure it out. Blog name might be strong indicator of that, so storing it would be good idea…
Then I stopped for a moment. I actually don't need any of these things. I want to know how many people use Google Analytics and how many people use Piwik. Popularity of other tracking/analytical services would be good to have, but I don't really know what other options there are and I guess I don't care enough to research it. So let's focus on Google Analytics and Piwik now.
I was able to collect data using these shell commands:
cd /tmp/ mkdir R python wget 'http://www.r-bloggers.com/blogs-list/' -O r.html xmllint --html --xpath '//*[@id="linkcat-3"]/ul/li/a/@href' r.html 2>/dev/null |sed -e 's:\s*href="\([^"]*\)":\1\n:g' | uniq > r-list wget 'http://planetpython.org/' -O python.html xmllint --html --xpath '//*[@id="menu"]/ul/li/ul/li/a/@href' python.html 2>/dev/null |sed -e 's:\s*href="\([^"]*\)":\1\n:g' | sort -u > python-list wget --tries=2 -T 5 -i r-list --directory-prefix=R/ wget --tries=2 -T 5 -i python-list --directory-prefix=python/
Then I could get raw numbers using following commands:
$ grep -l -i 'google-analytics\.com/' python/* |wc -l 253 $ grep -l -i 'google-analytics\.com/' R/* |wc -l 191 $ ls -1 python |wc -l 531 $ ls -1 R |wc -l 547 $ grep -l -i 'piwik\.js' python/* |wc -l 12 $ grep -l -i 'piwik\.js' R/* |wc -l 6
Finally, I calculated percentages using your ordinary calculator. Granted, this is not reproducible, but I just can't use shell arithmetic syntax or
bc without looking them up on the web. Anyway, Google Analytics is used by 35% of R bloggers and by 48% of python bloggers. I actually expected it to be more popular among R bloggers, because they usually focus on statistics and data analysis, what makes them more likely to care about their website usage statistics. Well, turns out it's not a case. Piwik usage is negligible.
Let's look up at time required by each step:
- Finding pages that list all blogs aggregated by R-bloggers and Planet Python: 5 minutes
- Figuring out XPath expressions to extract links to aggregated blogs: 5 minutes
- Evaluating viability of
xmllintoutput format is insane: 20 minutes
- Learning how to make
xmllintoutput usable, because I don't remember what was wrong with
xmlstarlet: 10 minutes
- Downloading all pages linked by R-bloggers: 90 minutes
wgetman page and figuring out how to not waste time waiting for nothing: 10 minutes
- Downloading all pages linked by Planet Python: 15 minutes
- Counting sites with Google Analytics and Piwik: 5 minutes
Total: 2 hours and 40 minutes (160 minutes), but could be reduced to just over one hour by using correct
wget flags from the start and by downloading all pages linked by both R-bloggers and Planet Python in parallel.
Had I gone with R, in the same time maybe I would have function for downloading website content in faulty network environment.
Takeaway lesson: Premature generalisation is root of quite some evil. Don't waste time writing programs that answer questions that maybe you might want to ask in the future.