Evilsoft.org

A few weeks ago, a video was passed around our office that was about Google interviewing people in Times Square about what a browser is. The results were fairly appalling. For the most part, people responded with various portal and search web sites instead of something like “IE” or “Firefox”. This coincided with a conversation I was having with someone in which I posited the hypothesis that IE as a browser would bias to the political right, and other browsers (specifically firefox) would bias to the political left. My reasoning was:

1. Conservatives would be more comfortable with a browser delivered by a major corporation that has faced antitrust charges over that browser then liberals would.
2. Liberals anecdotally are more prone to counter-cultural choices, and thus would be more likely to seek out an alternative to the default browser.

The Google interviews made me realize one other thing. In testing for this kind of effect, you would need to eliminate people who didn’t know what a browser was. Clearly, a person who doesn’t even know what a browser is is highly unlikely to proactively switch from their system’s default browser. Further, if a person’s political views are correlated at all with their likelihood of understanding what a browser is, not eliminating people who don’t understand could hide real results.

The company I work for, Knewton Inc, is becoming known in certain circles of its clever usage for the vastly underused (IMHO) Amazon Mechanical Turk service. When I mentioned my contention to our guy who’s been pioneering our MTurk usage, Dahn Tamir, he suggested that we build an MTurk task and get some real data to find out whether my hypothesis had any basis in reality. The rest of this post represents my findings.

The Experiment
We posted the following poll to MTurk, allowing users the ability to do the task only once (making it a lot harder to stuff the ballot box, or, at the very least, require a hell of a lot more effort on the part of the users):

The options for political affiliation were:

• Republican
• Independant
• Democrat

The options for political philosophy were:

• Strongly Liberal
• Liberal
• Moderately Liberal
• Moderate
• Moderately Conservative
• Conservative
• Strongly Conservative
• Socialist
• Libertarian

For the web browser, we considered a person to have passed the test if he or she answered Internet Explorer or Firefox, but no others. We considered them to have failed if they answered Yahoo, Google or MSN. It was pointed out that AOL did have a browser that could be reasonably referred to as AOL, so we ignored this field. Finally, we recorded the workers’ User Agent, letting us see what browser/os they were using.

Raw Results
Observations:

• Clearly, MTurkers are going to be more likely to know what a browser is, and it shows in the data. Google found that 8% of people in Times Square knew what a browser was, but 86% of respondents here did.
• We appear to have more Democrats/Liberals than Republicans/Conservatives one would expect in the normal population.
• I only added Socialist on a lark, and I was pretty amazed to see that it still got 2.24% in a poll of Americans
• Firefox polled a lot higher here than one would expect in the wild. This should make it a lot easier to see any effects, along with the high pass rate of the test.

Party Affiliation
Total % Republican 244 23.71% Independent 397 38.58% Democrat 388 37.71%

Political Philosophy
Total % Strongly Liberal 92 8.94% Liberal 188 18.27% Moderately Liberal 169 16.42% Moderate 197 19.14% Moderately Conservative 157 15.26% Conservative 105 10.20% Strongly Conservative 34 3.30% Socialist 23 2.24% Libertarian 64 6.22%

Browsers
Total % Firefox 479 46.55% Internet Explorer 439 42.66% Safari 63 6.12% Chrome 41 3.98% Opera 7 0.68%

Browser Test Results
Total % Pass 886 86.10% Fail 143 13.90%

Test Results
Next, after pulling in the raw results, we compared browser, affiliation and philosophy to the test results.

Observations:

• Unsurprisingly, IE fared better amongst people who failed the test. Firefox did substantially better amongst people who passed the test. Safari also did better amongst people who passed the test, which was a little surprising, because we expected all default browsers to do better with people who failed. The sample, however, is small enough that the Safari factor could just be noise. It has been mentioned that this could have been caused by Safari for windows. That supposition can potentially be answered in the data.
• Democrats did worse on the test, while Republicans and Independents did basically the same. We see the likely reason why in the political philosophy section. The further left you go, the worse you tend to do on the browser test, until you hit strongly liberal. Strong Liberals reversed the trend entirely and tested better than any other group.

Test Results by Party Affiliation
Pass % Fail % Republican 87.70% 12.30% Independent 88.16% 11.84% Democrat 82.99% 17.01%

Test Results By Political Philosophy
Pass % Fail % Strongly Liberal 91.30% 8.70% Liberal 81.91% 18.09% Moderately Liberal 84.62% 15.38% Moderate 86.29% 13.71% Moderately Conservative 86.62% 13.38% Conservative 87.62% 12.38% Strongly Conservative 88.24% 11.76% Socialist 86.96% 13.04% Libertarian 89.06% 10.94%

Browser by Test Results
Pass % Fail % Internet Explorer 39.16% 64.34% Firefox 49.89% 25.87% Safari 6.21% 5.59% Chrome 4.06% 3.50% Opera 0.68% 0.70%

Browser choice by political opinion
Here we have the final analysis. Looking at how browser choice changes across the political spectrum.

Observations:

• My initial hypothesis appears to be supported by this data set. IE falls off sharply the further left you go on the spectrum from Moderate. Strong liberals use IE at half the rate of strong conservatives.
• Both Libertarians and Socialists appear to fit in on the left end of the spectrum at least where browser choice is concerned. Libertarians appear to loath IE and are nuts for Firefox. They have the highest Firefox rating of anyone with an astounding 70.18%
• Firefox’s fall as you go to the right is less pronounced as IE’s fall as you go to the left. This is probably caused by the generally higher likelyhood of picking Safari as you go leftward.
• Opera appears to have its ’stronghold’ with Socialists, who were as likely to pick it as Chrome or Safari.
• Libertarians also appear to have an exaggerated preference for Opera, Safari and Chrome. Basically, Libertarians really don’t like IE.

Browser by Party Affiliation
IE % Firefox % Safari % Chrome % Opera % Republican 45.33% 45.79% 5.14% 3.27% 0.47% Independent 36.00% 52.86% 6.57% 3.71% 0.86% Democrat 38.51% 49.38% 6.52% 4.97% 0.62%

Browser By Political Philosophy
IE % Firefox % Safari % Chrome % Opera % Strongly Liberal 25.00% 63.10% 8.33% 3.57% 0.00% Liberal 35.71% 50.65% 9.74% 2.60% 1.30% Moderately Liberal 41.96% 48.25% 4.20% 5.59% 0.00% Moderate 45.88% 44.12% 4.12% 5.88% 0.00% Moderately Conservative 47.06% 43.38% 7.35% 2.21% 0.00% Conservative 44.57% 46.74% 4.35% 3.26% 1.09% Strongly Conservative 50.00% 46.67% 0.00% 3.33% 0.00% Socalist 30.00% 55.00% 5.00% 5.00% 5.00% Libertarian 12.28% 70.18% 8.77% 5.26% 3.51%

Edit
Some have found the graph confusing, so here are two other versions. Both horizontal, showing the data in one by browser, and in the other by political philosophy. Click on them to enlarge

(enlarge)

(enlarge)

Conclusion
There appear to be a few strong correlations in data between a person’s political thinking and his or her browser choice. Still, this data is not representative of the population as a whole. It is, instead, a sample of people who are web savvy enough to use MTurk and who are motivated to take polls like this for small amounts of money. Also, the strong bias of the tech and open source communities toward libertarianism is probably causing the strong bias of libertarianism toward Firefox in this sample. Still, I thought these results were too interesting to keep to myself. Enjoy.

Notes
If you want to toy with the raw data and get access to the source I used for experimenting with the data, as well as my analysis spreadsheets, those are available here. We did this in 3 separate data gatherings, and those samples are still separated into 3 data files the raw data.