October, 2007

process — 3 Comments
30
Oct 07

We shipped funnelcake01

Previously we wrote about “the funnel“, and our desire to better understand how a download becomes an active and long-term Fx user. There, we suggested a way to measure this with some degree of accuracy, and since launching the project, we are now at the point of digging through the data seeking findings. Before presenting those findings, this post is meant to highlight how we launched funnelcake (the project’s name), and how we did so without comprising users’ privacy.

One goal within our process was to tie a Firefox download to a particular date, with greater accuracy. For example, we can tell that a Firefox download of 2.0.0.6 happened sometime between July 30, 2007 (2.0.0.6 release date) and September 18, 2007 (2.0.0.7 release date). With funnelcake, we were able to tie a build to a particular date.

On October 4th from 00:00:00 – 23:59:59 PST, we included an extension with Firefox 2.0.0.7 (en-US and de) that changed three preferences relating to the funnel. These preferences where:

  • firstrun “Welcome to Firefox” URL
    • the default: http://en-us.www.mozilla.com/en-US/firefox/2.0.0.7/firstrun/
    • funnelcake: http://en-us.www.mozilla.com/en-US/firefox/2.0.0.7/firstrun/?f=1
  • whatsnew “You’ve been updated” URL
    • the default: http://en-us.www.mozilla.com/en-US/firefox/2.0.0.7/whatsnew/
    • funnelcake: http://en-us.www.mozilla.com/en-US/firefox/2.0.0.7/whatsnew/?f=1
  • update channel
    • the default: release
    • funnelcake: release-cck-mozilla01

If the build is opened, and we see hits on the differentiated URLs, we know that the build was downloaded on October 4th, 2007. This is like knowing that 2.0.0.6 builds were released between July 30th and September 18th, except for a specific date. This allows us to gain insight into when the build was released and when it is subsequently used.

The extension is listed under the addons manager as, “Mozilla Settings for October 2007″, with the description, “October 2007 settings for the automatic update feature of Firefox. This extension points Firefox to an alternate update channel corresponding to October 2007, and can be disabled or removed safely without affecting automatic updates.”

We’re still wading through the findings — and should have some results to post. Since we are not collecting anything user identifiable, we feel we’ve navigated through the necessary precautions allowing us to safely share what we find!

Uncategorized — 6 Comments
10
Oct 07

Firefox Usage – Does the Day of Week Have an Effect?

We know that internet usage (as well as Firefox usage) sees a good deal of fluctuation based on seasonality. We’ve also historically noticed that the day of the week affects usage of the internet (or number of people using the internet, depending on your definition of usage).

Thinking about this latter trend, I decided to quantify this “day of week” effect. Applying regression analysis to some numbers that give us a rough idea of daily active Fx users, I’m able to account for a few factors, including the month of the year and US holidays (“controlling for all observable variables” in statistical speak). Here’s what I found:

day_changes

If there are any stats fans reading, please let me know. I’m happy to go into further detail regarding my regression model, the regression coefficients, t-stats, etc. For everyone else, there’s an easy way to intrepret these findings. Guessing the number of daily active internet users on Fridays as 400 million, we can estimate the day-to-day change in internet usage during a typical week:

internet_pop

I wonder if comScore, Nielsen and other measurement firms see similar usage patterns…

Uncategorized — 1 Comment
4
Oct 07

Fun with Google Trends

Google trends has always fascinated me. While it is pretty unscientific, I found it interesting to compare searches on iPod, to searches on Firefox.

Firefox vs. iPod

(red is ipod, blue is firefox)

I wonder how much it would change things if we got our “News reference volume” equal to that of the iPod.

Has anyone else found interesting mozilla trends using this tool?

results — 6 Comments
1
Oct 07

Understanding Shifts in Fx Market Share

While most of us have a pretty good understanding of Fx’s market share, I thought it’d be cool to explore ways to better interpret some of the information we see from third-party sources. For example, Net Applications provides a monthly breakdown of approximately 50 different browser versions. Their overall methodology isn’t perfect, but this type of drill-down data might allow us to come upon an interesting phenomenon… or simply allow us to better understand the change in our market share over time.

It *seems* as though IE7, Safari and Opera have each enjoyed their own successes this year, but is this really the case? To answer this, I considered that a certain slice of the overall market share pie is “up for grabs” each month and then looked at how those slices have evolved since early 2007 (e.g., is Opera Mini or Safari 3.0 grabbing a disproportionate slice of that pie?).

Using this methodology, here’s what I found:

  • Firefox has performed phenomenally this year relative to the competition
  • The size of the market share pie that changes hands each month is extremely tiny (relative to my expectations)

Comparing an average of Jan-Feb numbers to Aug-Sept numbers, Firefox netted nearly 60% of new users (or those who switched browsers). The other big winner? Opera Mini claimed 23% of new users. Safari and Opera Desktop each saw gains about 1/7 the scale of Fx, and IE was essentially responsible for all loses.*

pie

The second part of the story is that even with our success, our market share number has increased by 0.8% since early this year (e.g., from 15.0% to 15.8%). All of this is a long way of saying that it’s very difficult to move the needle with respect to market share… though, relative to our competition, Fx rocks!

———–

* This methodology is based on an absolute gain of users. For those curious, below are the raw numbers that I used. The source can be found at http://marketshare.hitslink.com/report.aspx?qprid=6. When you take the numbers below and assume that 1% of market share equals X number of users, you’ll arrive at the pie chart above.

data