News for hackers. Stuff that matters.

mytype

No, fans of Fight Club on Facebook are much less conformist than other people on Facebook.

If you actually read the article and came to this conclusion then either I'm a bad writer or you're a bad reader.

It's very awesome. My favorite find of the last several months, if not year.

I'm curious, what are you going to do with it?

My best interpretation:

The fan profile shows that Fight Club is particularly appealing to the 18-30 crowd (makes sense). It's also a smart film. Most people who like it are either a poor college/postgraduate student or a professional making decent money.

We see this same income dichotomy with a number of "smart" likes and preferences.

We have a lot of Facebook profile info and psychology survey data, going to write some fun, OkCupid-style posts with it. If anyone wants to get involved, let me know at tim [at] mytype.com. We handle a lot of the messy stuff already (data cleaning, etc.)
We have not just Facebook movie data but obviously books, music and lots of other good stuff. Lots of relationships to explore between "likes" and personality, values, morals, which we plan to do on our blog, in the style of OkCupid. If anyone is interested in doing analysis with us, shoot me a note at tim [at] mytype.com. Would be cool to find someone else to write posts with.
Re:How WikiLeaks threatens transparency on Tuesday July 27, 2010 @12:30PM (#1551737)
This leak doesn't really provide information on what the government is doing. They are low level reports written from the field, largely recognized by people working in this space to be unreliable. The government should certainly be more open about our activities in Afghanistan and the information that drives those actions, but reports like these without context are not the way.

Let's also think about this leak from the perspective of the Afghans and Pakistanis who risk their lives to work with the United States. If we cannot be trusted to protect their identities, then we will end up with far fewer local partners.

Yes, it's a biased sample. I've already admitted that. My point is that it's not egregiously more biased than any other sample from academia or professional surveyors, and in many cases it's less biased.

Do you reject most academic psychology research because it is based on students at the college of the researcher? Certainly that's a more biased sample.

Do you reject political polls because they're based on who answers calls from random phone numbers and then does not hang up once they realize it's a poller?

Do you reject most commercial research based on paid volunteers or visitors to sites with much smaller and more biased audiences than Facebook?

I state upfront in the article that this is based on MyType users who are on Facebook. What more do you want? If you want no bias, just do math and don't believe any data based on people, written by people, spoken by people, anything having to do with people.

The bottom line point is, this is much more rigorous than much of the crap blogs and media report on. I'll take a random example that I googled for the iPad: http://techcrunch.com/2010/04/06/ipad-sentiment-analysis/. "87% of tweets indicate intent to purchase the iPad". Give me a break. Talk about bias. The sampling errors there are horrific.

I'm just trying to maintain a reasonable perspective on MyType's data, not hide any facts about the shortcomings of it. There are shortcomings, they're just not so bad to make the results "entirely unreliable".

We focused on the 11% and 3% slices, yes. But from there we looked for the most over-represented groups. The fact that the independent geeks and selfish elites are 3 and 6 times more prominent in those slices is statistically significant, and it's a statement about the strongest tendencies of owners and critics.

But yes most owners are not selfish elites, and most critics are not independent geeks. It's difficult to say any group is "mostly" any particular psychographic profile. We're highlighting correlations and tendencies, not stereotyping. Admittedly, the title doesn't convey this, but if it tried to no one here would be reading the post, it would sound like something out of an academic journal.

The study is getting top-of-front page coverage right now from wired.com
There's two explanations for that difference:

1) Forrestor's 3.8% was collected in June 2010, whereas our data was collected from March through May. Clearly the iPad is gaining momentum, as time goes on more people are planning to buy one (at least for now).

2) More importantly, Forrestor said that "no time frame was specified in the survey question". So they asked people who don't own an iPad if they intend to buy one. That's fairly different from our question, which allowed the respondent to choose from:

1) plan to buy one 2) want to play with one first 3) will wait for later versions 4) waiting for the consensus opinion

Clearly, with the variety of options, some people who might have simply answered "yes" to the "intend to buy" Forrestor question would pick 2, 3 or 4 in our question.

Does that settle that? And by the way, these are people who wanted to take a personality quiz, not answer a poll. Their motivation had nothing to do with the iPad, that question was randomly inserted into the personality quiz.

While you seem to be "amused" by a blogger who in your mind completely makes stuff up, I'm amused by a commenter who after 5 minutes of review feels confident enough to slam a study that took multiple people dozens of hours to complete.

We're not publishing this in a science journal. That's not what we're going for. It's not complete BS either though. We do normalization, we have lots of data, and our psychological measures are based on the best contemporary research. The results are worthwhile. I'll say again: I challenge anyone to point out a major flaw in our data (not my interpretation of it).

This certainly would be a good point if it were true. In reality people come to MyType to take a personality quiz. We just randomly inserted the iPad question into the personality quiz for 20,000 people. They're not particularly savvy on technology, just the average Facebook user.

We realized that as a personality typing application we have a very good platform for surveying any sort of opinion. Our user base is fairly reflective of the general US population, and none of them are coming to answer a survey about x. We plan to look at the personality traits that drive preferences in beer, politics, and much more. Subscribe to the blog if you haven't already :)

I'm the author of the post. Admittedly I got into some speculation in the post, and I tried to frame it as such, not as unequivocal facts "proven" by the data. The only thing that's solid is the data. There are a number of interpretations that can be made from it, mine is just one of them.

We labeled owners "selfish" because they scored low on measures of kindness (our official term is benevolence) and altruism. These are solid psychological measures. The "biggest sin" question is more just for fun and wasn't taken into account in summarizing the Owners and Critics.

There's a lot of data there, trying to paint a picture that is consistent with all of the data is pretty much impossible. I did my best to pull out the main theme from the data.

Our sampling rigor cannot be compared to political polls, it's true. However we do weight the sample not based on guesses but on official data about the gender, age, and personality distribution in the US. We did not use geographic weights because our respondents were more or less properly distributed.

Any poll can be attacked. Even political polls are subject to bias in who picks up the phone, who responds to the surveyor, etc. This bias is partially driven by personality type, which we are able to nearly eliminate through weighting by personality traits. How many polls can do that?

Most polling is much less random than political polls, and yet the results are still treated as worthwhile. Some firms put up surveys on sites (site audience bias), or have carefully selected paid volunteers (bias to incorporate people who are willing to do that), etc.

The fact that our results are corroborated by Forrestor and Sybase, to the extent that we can be compared, is very satisfying to me personally. Sure, we may be a few percentage points off on some things, but are the overall results invalid? No. I challenge anyone to point to any result in the study and give a reasonable explanation why that result is likely way off. The only things that are suspect to me are the major outliers, subsets like native americans for which we did not have much data. I'll give you that. Anything else? Let me know.