Alessandro’s team has extensive experience studying privacy in online social networks.  Alessandro and Ralph Gross wrote one of the earliest (and most cited) Facebook privacy papers: Imagined Communities: Awareness, Information Sharing, and Privacy on the Facebook. Last summer, the team published a truly head-turning study, showing that information gleaned from social network profiles could be used to predict social security numbers.  Most recently, Alessandro’s work was featured in Jeffrey Rosen’s New York Times Magazine article The Web Means the End of Forgetting.

I look forward to building on my current areas of research – privacy, identity and support in social networks – while being exposed to new opportunities and new challenges at CMU.  Speaking of challenges, the next challenge is a dissertation defense (later this fall) and then a move to Pittsburgh.  It has been a while since I’ve been to Pittsburgh, so I’m open to advice!

Because of this sanctioning system, we can make a reasonable inference that items that have been taken off of Craigslist have been sold.  The items that have short lifespans on Craigslist are desirable – they are a good value, priced properly – and those with long lifespans are either unwanted or improperly priced.  I’ve recently been in the market for a used car (cough, a minivan), so I’ve been collecting information about the cars offered on Craigslist and their lifespans on the service. By looking at prices and lifespans (and a few other variables), can we automatically identify cars that offer the greatest value?

What follows are some charts from a simple survival analysis of the last 30 days of Honda Odyssey sales on Craigslist in Raleigh/Durham.  The de-duped dataset includes 55 cars (out of about 130 posts). Before you read much into the data, many of the variables I explored (mileage, model year, etc.) weren’t significant predictors of “hazard” (that is, sale). If you were able to get this data on a larger scale, it does seem likely you’d be able to identify patterns of value. That said, there is a lot of randomness is a car’s quality once it has been driven, so the value of such a model-based approach would only be in prioritizing potentially under-priced cars.

n.b.: You could also do this sort analysis on want-ads. Want-ads have a great sanctioning system, as it is pointess to pay for an ad after you’ve sold your car.

p.s.: Perhaps what is charming about Craigslist is that there isn’t any meaningful historical data. This likely generates more variability in price, leading to the perception that you can find great deals (which you can!).

However, we’ve gotten feedback from translators and users in other countries that translations wind up being too confusing when people have not specified a sex on their profiles. People who haven’t selected what sex they are frequently get defaulted to the wrong sex entirely in Mini-Feed stories. For this reason, we’ve decided to request that all Facebook users fill out this information on their profile.

Just today, I discovered (via the R Bloggers news feed) an video on the use of R in corporations like Google and Facebook.  The representative of the Facebook data team talked about some exploratory data analysis they did in 2007.  The finding?  “If a user comes on more than once and is willing to give Facebook a very basic piece of information – their gender – that seems to be the strongest predictor of whether they will stay on the site.”

I’m not looking to stir up any controversy.  Rather, I think it is an interesting example of analytics-based development, of research informing design.  Of course, the challenge of translating research into practice is immense.  Are there critical differences between individuals that share gender and those that don’t?  Did a forced gender-selection process invalidate the predictive model?  Was the controversy over gender selection worth the predicted benefit?  Perhaps Facebook’s 500 million users owe more to gender selection than we can imagine.

Anyway, the video has some age on it, but I did enjoy hearing about Facebook’s use of R (the other analytic examples provided are cited in the “Maintained Relationships on Facebook” report, plus there are a few ICWSM papers, I believe).  You can find the full video here (doesn’t look like embed is supported).

Recently, I ran a survey of first-year students at UNC that tested preferences toward compensation.  The survey offered a dual-tier lottery compensation: Participants were entered to win an iPod touch or their choice of three gift certificates (See [2] for more on dual-tier incentives).  The three gift card choices were iTunes, Amazon, or a popular on-campus cafe, in the amount of ten dollars.  Response to the survey was good, by email-solicitations standards, at 31% (n~1200).  Males were slightly underrepresented, as is commonly the case.

So, what gift cards did my students prefer?  Clearly, the students preferred gift cards to iTunes (n=442) and Amazon (n=442) over the local cafe (n=131).  And we don’t really need any significance tests to see that the difference between iTunes and Amazon is a wash (p=.8406).

When conducting surveys, we’re not always interested in a large homogeneous population.  Sometimes we’re interested in sub-populations, such as certain genders, ages, or ethnicities.    Breaking the perferences out by gender, visual inspection indicates that female students prefer iTunes over Amazon, while male students prefer Amazon over iTunes.  Since neither population comes close to preferring the local cafe, I will focus on the difference between iTunes and Amazon for the rest of the analysis (i.e. drop the people who prefer the Local Cafe).

Of the students that selected Amazon or iTunes, we see that 53% of female students prefer iTunes, 47% Amazon.  Of males, 58% prefer Amazon, 42% iTunes.  The Chi-square test indicates a relationship between gender and preference (p=.001), and within-gender Chi-square goodness of fit tests indicate that while the female student preference difference is insignificant (.0922), the male preference towards Amazon is significant (p=.0064).

To test some higher order interactions, I employed a logistic regression model to test the effects of gender and a few other covariates.  First, since much of my sample is from NC, I tested to see if NC residency might contribute towards a preference.  In this model, gender remained significant, but NC residence was not significant (p=.828).  Next, looked to see if GPA might be a factor in preference.  Gender remained significant, and GPA’s p-value was low (p=.081), but not close to significance (directionality was higher GPA’s towards Amazon).

In the last two models, I looked at ethnicity and age.  In the ethnicity model, gender is significant, and only one ethnicity is significant.  Compared to other ethnicities, students who self-report as Asian demonstrate a preference towards Amazon (OR=.158, p=.000).  With age, gender again remained significant, but 19 year old students (compared to 18 year old students) seem to prefer iTunes (OR 1.49, p=.004).  Notably, a gender by age interaction was not significant, however.

To briefly review, it seems that among my population, the anecdotal preference towards iTunes is just that: anecdotal.  This is good news for me, because it is much more complicated to process iTunes gift cards than Amazon gift cards.  Some final notes: This is not really a proper experiment – such an experiment would use completely randomized solicitation.  Also, the presence of the third category (Local Cafe) is potentially troubling if being a fan of a Local Cafe also correlates to, say, being an iTunes fan or an Amazon fan.  Caveat emptor, blog post, not peer reviewed, etc.

1.  I don’t have a citation for this, but I do monitor to a number of email lists that frequently offer research solicitations.  YMMV.

2. See Li, Kaiwen (2006).  Student Preference for Survey Incentive.  UC Davis Student Affairs Research & Information Tech Report.

Finally, I promise that Amazon has not compensated me in any way, say, by sending me a bunch of gift certificates or a Nikon 12-24mm DX lens or anything like that.

Ms Stone says Freedom and other such programs are “a first step”, since anyone who installs and uses one of them is admitting that there is a problem, and “something needs to shift”. But the next step is to go beyond a software crutch, Ms Stone says, and to learn to change one’s behaviour without the need for full-screen modes and internet-disabling utilities.

In the blog post, she expands on the general concept:

I’m not opposed to using technologies to support us in reclaiming our attention. But I prefer passive, ambient, non-invasive technologies over parental ones. Consider the Toyota Prius. The Prius doesn’t stop in the middle of a highway and say, “Listen to me, Mr. Irresponsible Driver, you’re using too much gas and this car isn’t going to move another inch until you commit to fix that.” Instead, a display engages us in a playful way and our body implicitly learns to shift to use less gas.

With technologies like Freedom, we re-assign the role of tyrant to the technology. The technology dictates to the mind. The mind dictates to the body. Meanwhile, the body that senses and feels, that turns out to offer more wisdom than the finest mind could even imagine, is ignored.

I’d suggest reading the whole post – it’s good and very thought provoking – but I take issue with the central premise of Stone’s argument, that it’s just a matter of time until we “create personal technologies that are prosthetics for our beings.”

Here’s my argument:  There’s no question that Freedom is a tyrant: but Freedom doesn’t control you, it controls technology.  And I have to believe that to many industry insiders, this is an uncomfortable direction for technology to take.

It is not controversial to claim that the dominant ideology of computing in the modern era has been “bigger, better, faster.”  In fact, this ideology – the connection between technological progress and advancement as a civilization – has stuctured the way we think about ourselves and other societies for hundreds of years.  In the epilogue to his excellent book Machines as the Measures of Men, Michael Adas writes:

The long-standing assumption that technological innovation was essential to progressive social development came to be viewed in terms of a necessary association between mechanization and modernity.  As Richard Wilson has argued, in American thinking, the “machine and all of its manifestations – as an object, a process, and ultimately a symbol – became the fundamental fact of modernism.”

Since the origins of the computing industry, Ruth Schwartz Cowan argues in A Social History of Technology, the focus has been squeezing productivity out of  machines and operators.  This logic of practice was inscribed to the industry “because the government [the dominant early contractor of the computing industry], fighting the protracted cold war with the Soviet Union, believed that it would need better and better computation facilities…”

This constant drive towards efficiency has many rewards: Transistors that are orders of magnitude cheaper than ones produced just years prior, Terabyte disks that sit on desktops, and the iDevices that I so covet.  My argument does not downplay the value of such advances, and to do so would be foolish.

Rather, I argue that the drive towards bigger, better, faster has left us with devices that are out of sync with our work patterns.  To address the growing divergence between our devices and work practice, we’ve constructed and attempted to empiricize the concept of multi-tasking.  Multi-tasking, as we now know, has decreasing marginal effectiveness as task complexity increases.  Multi-tasking fails most those who need it most.

Flipping through the last ten years of CHI, CSCW, and GROUP proceedings, we see an array of systems built to support multi-tasking, to facilitate remote work, to prostheticise our beings.  In these technologies we see the march towards progress, efficiency: bigger, better, faster.

Freedom joins these technologies in the march towards progress and efficiency, but with a different value set: smaller, better, slower.

In the past five or ten years, the devices we use for work have exploded in complexity.  No longer a word processor or spreadsheet, our computers are now televisions, game machines, and – most importantly – a portal to an always-on channel of social exchange.  Yet because these changes have been realized in code as opposed to form, we think of the device as static.  A computer is just a computer.  Rather, I see devices that are increasingly beginning to fail the market, with disastrous consequences for productivity, progress, and self-worth.

Freedom has always been about control.  It was first designed to reclaim space – to return the pre-internet state of a coffee shop that has suddenly gone wi-fi.  Only through extensive use have I realized that Freedom is about pushing back at the device itself, a device that has failed the work market in a drive toward progress.

In closing, Linda Stone asks “What tools, technologies, and techniques will it take for personal technologies to become prosthetics of our full human potential?”  First, we must understand that we, humans, are not the problem.  Second, we must reconsider our relationships with our devices, and examine with open minds where our devices have failed us.  Third, we must change the ideology of the productivity industry, moving away from bigger, better and faster and towards smaller, better, and slower.

Of course, this is easier said than done.  And it will almost certainly come from outside industry, which is constrained by its dominant logic of practice.  But I can’t help but think that we’re at the beginning of something big.

I developed Anti-Social because of a problem I ran into consistently with Freedom – I loved being offline, but found myself frustrated when I needed to look up a citation or a new article when Freedom was running. Anti-Social allows you to tune out the social parts of the web – Twitter, Facebook, etc. – while allowing you access to research materials, Google, and other invaluable resources. I’ve been using it for the past few weeks while working on an R&R – Anti-Social allowed me to remain in focused writing mode, while allowing me to research as I revised the manuscript.

Together, Freedom and Anti-Social represent an emergent computing phenomena I’ve been calling “80% computing.” By taking problems that are socially or computationally hard (e.g. changing habits, reducing compulsive surfing), and providing imperfect solutions, I’ve found there’s an interesting spot in the market. I wonder what other highly complex problems (e.g. productivity) we could solve with 80% solutions?  If we move away from perfection as a computational standard, and allow individuals to adapt their practice to imperfect technologies, we may be able to develop some very simple solutions to very challenging problems.

Along those lines, the Economist recently profiled my software in a wonderful article. I’ll quote at length:

“CLEAR your screen and clear your mind.” That is the philosophy behind a new wave of dedicated software utilities, and special modes in word-processing packages and other applications, that do away with distractions to enable you to get on with your work. The problem with working on a computer, after all, is that computers provide so many appealing alternatives to doing anything useful: you can procrastinate for hours, checking e-mail, browsing social-networking sites or keeping up with Twitter.

But in its severity and simplicity, Freedom (for Macintosh and Windows) may be the ultimate tool to ward off distractions: the virtual equivalent of retiring to a remote getaway, or going on a writers’ retreat, to get things done.

But fans of Freedom are not concerned by such philosophical niceties; they use it because it makes them more productive. Peter Sagal, the host of the American public radio show “Wait Wait…Don’t Tell Me!”, is one such fan. He has no trouble writing to a strict deadline at work. But outside work, “I simply can’t resist the call of a website or an RSS feeder or now my Twitter feed. I simply can’t do it,” he says. Before he started using Freedom he managed to write a book, but only by unplugging his cable modem to cut off his internet access. “But that was too easy to plug back in,” he says. The internet, he grumbles, has “murdered” his ability to do extracurricular creative work, such as writing books, plays and screenplays.

Hardware and software are usually sold on the basis that they can do more, do things faster or have whizzy new features. There is clearly a place for products that are simple to use and hide complexity—a hallmark of Apple’s products. It is perhaps more surprising that there also seems to be demand for products that disable features. But for people trying to get things done, a hobbled computer may in fact be more useful than a fully functional one, for an hour or two at least. Temporarily worse can, in some ways, be better.

Artwork from the Economist.

Of note, the New York Post also ran an article that prominently featured Freedom and Anti-Social. The title of the article was a classic Post headline: Fatal Distraction.

I should close with the following. First, I am aware that spending time writing anti-procrastination software is actually meta-procrastination. Second, Anti-Social really is great. Check it out. It is a revelation to be on the un-social Internet. Finally, I’m waiting for Peter Sagal to come and ask me for a percentage of my sales. He is simply too kind with his advocacy of Freedom!

The lecture by Carter Butts will be of particular interest to individuals looking to use R for social network analysis.  Butts is the author of a number of network analysis packages for R (many of which come bundled in the amazing statnet package).

Network Analysis with statnet for Individual, Organizational, and International Relations Applications by Carter Butts, University of California-Irvine

• Lecture Part 1
• Lecture Part 2

Advanced Network Analysis by Matthew O. Jackson, Stanford University

• Lecture Part 1
• Lecture Part 2

If you find these materials useful, you might also wish to check out Steve Goodreau and David Hunter’s tutorial Advanced Social Network Analysis Using R and statnet available at the Complexity and Social Networks blog.

Social plugins let you see what your friends have liked, commented on or shared on sites across the web. All social plugins are extensions of Facebook and are specifically designed so none of your data is shared with the sites on which they appear.

According to Mashable, over 50,000 plugins have been installed since the rollout.  Seeing one’s Facebook friends suddenly start showing up on third party sites has raised privacy concerns, which Facebook quickly addressed in a blog post, stating “Because [third party sites] have given Facebook this “real estate” on their sites, they do not receive or interact with the information that is contained or transmitted there.”

Here’s the rub.  By giving “real estate” to Facebook, third party sites have created a one-way mirror, allowing Facebook to peer in on what we’re doing.  If you’re logged in to Facebook, and you visit a third party page with a social plugin, Facebook knows where you’ve been.  The mechanism is simple – cookies and referrals – and it will allow Facebook to create personalized behavioral profiles that, combined with the information we articulate in Facebook, will be tremendously valuable.

To explore the privacy implications of Facebook’s social plugins, I visited the websites of the top 15 U.S. online news destinations (based on some 2009 Nielsen data), and a few honorable mentions.  I then selected a news story from the front page, and loaded the page.  I checked to see if social plugins were enabled, if the Facebook cookie was called, and if the referring page was sent to Facebook (basically, did the site identify you to Facebook, and share the page you were on).

I found that of the top 15 online news destinations, 9 were sharing information with Facebook (MSNBC, CNN, CBS, ABC, Fox News, Washington Post, and the Tribune, McClatchy and Gannett Companies[1]).  Notably, The New York Times, BBC, Yahoo News, AOL News, and Google News did not share information.  I then checked a few favorites of mine: NPR (yes), Drudge (no), Huffington Post (yes), and Politico (no).  I’ve included all of the details on a spreadsheet, embedded below or html version.

According to Nielsen, the 9 news organizations sharing information with Facebook account for over 177,161,000 monthly unique visitors.  Granted, not all of these views will go to social plugin enabled pages, and not all visitors will be logged-in Facebook users.  But with 400 million users, it is safe to assume that a substantial proportion of that information will go to Facebook.  If you stay logged in to Facebook, it is increasingly likely that Facebook will know what news you read.

My beef here isn’t necessarily with Facebook; Google and other behavioral-targeting firms have very similar SOP’s.  Rather, I’m uncomfortable that so many news organizations felt comfortable sharing the news-reading behaviors of their customers that just so happen to be logged in to Facebook.  And really, what do they get for trading this tremendously valuable asset?  I get to see that a random friend liked an article?

I think it is time that someone wrote a Firefox plugin that specifically manages the Facebook cookie, only allowing it to be accessed when someone is on Facebook proper.  Clearly, we can’t trust third parties – even reputable news organizations – to protect our data.  Here’s the spreadsheet from my analysis:

Note: For media conglomerates (Tribune, McClatchy, Gannett) I visited the flagship outlet (Chicago Trib, Sac Bee, and USA Today, respectively).

I’m loathe to trust bloggers getting survey data right, so I requested a copy of the report from Edison Research (available here).  At first glance, the data looks good – the research was conducted by Arbitron, it employs a landline/mobile random digit dialing (RDD) frame, with about 1,750 people age 12 and older interviewed.  “National probability” studies of this sort are generally considered valid for population estimates.

Without getting into too much detail, a study’s validity is dependent on the sampling method and sample size (among many other things).  In terms of method, RDD is not a true equal-probability of selection method, but both industry and academia consider it “good enough” when the sample is weighted to known totals.  As for size, a sample of 1750 people allows us to make claims about a large population at an error rate of about plus or minus 3 percent.

Let’s cut to the chase: Where did the Edison Research interpretation go wrong?  In the report, Tom Webster states:

The percentage of Twitter users who are African-American currently stands at roughly 25%, which is approximately double the percentage of African-Americans in the current U.S. population. Indeed, many of the “trending topics” on Twitter on a typical day are reflective of African-American culture, memes and topics.

From this, we are to believe that of all Twitter users, 25% are African-American.  Not only is this surprising considering current population estimates, but also because Twitter is a global service.  Let’s explore how Edison got to this 25 percent number (conveniently rounded up from 24 percent).

In the phone interview, Edison asked all respondents 12+ (n=1750) if they “currently ever use[d] Twitter.”  7% of respondents said yes, approximately 123 people.  Of those 123, Edison then asked how often they used Twitter.  85% of those respondents (105 people) indicated they used Twitter at least once a month, and were thus recoded as “Monthly Twitter Users.”  Herein lies the problem: It was from these 105 individuals (not the 1750 total respondents) that Edison based its estimates of Twitter use.

Let’s return to sampling error.  Because random samples are asymptotically efficient, a sample of 1750 can speak to a population of hundreds of millions almost as well as a sample of 2000, 3000, or even 5000.  But a sample of 105 people speaking to the very large userbase (self reported at 100 million) of Twitter? Not so efficient.  The margins of error are approximately +/- 10% at an alpha of .05, +/- 12.5 at an alpha of .01.  And these margins assume true equal probability of selection, and no nonresponse bias.  With weighting for proportionality, it is almost certain these margins increase substantially (1).

Let’s explore what this means practically.  First, Edison Research can’t speak to all Twitter users, because all Twitter users weren’t potentially included in the sample.  Edison can, however, speak to USA Twitter use, from its sample of 105 monthly users.  If we assume that only 5 million Twitter users in the USA use the service every month, Edison is still using 105 people to speak about these 5 million people (the margins of error don’t change).  Unfortunately, this is highly unreliable.

The American Community Survey finds that approximately 13.1% of the US population self identifies as Black or African American.  At an alpha of .05, the range of potentially true estimates of African-American Twitter use in the US is actually anywhere from 14% to 34%.  At an alpha of .01, this estimate ranges anywhere from 11% to almost 38%, causing us to reject the hypothesis that the estimate is not attributable to sampling error or random effects.  If we then include weights in our estimates of error (likely the case because Edison’s sample over-represents people under 24), the growth in error causes us to fail to reject the null hypothesis at the .05 level as well.  We just can’t trust that the demographics of Twitter actually do vary from current population estimates.

Is Twitter “disproportionately” African American, White, Hispanic, or Green?  The simple fact is that from this data, we can’t say so with confidence.  If Edison had been a little more forthcoming with their sample sizes, it might be more likely that the blogger/journalist who reported these data would have sensed something wrong.  But I wouldn’t bank on it, because it seems like Edison Research was pushing this spin from the get-go.

A final note: as I was researching/considering this piece, it was interesting to see the “spin” being placed on this “fact” around the blogosphere.  Of course, you had your standard racist comments/tweets of the “there goes the neighborhood” variety, but there also appeared to be a large swath of users who were heralding this as a point of pride.  Before you examine my subconscious racist motives for examining this question, please just know I like getting surveys right.  And if Edison wanted to get this right, they could start by giving us a topline cross-tab of ethnicity, Twitter use, and the respective margins of error.

Ugh, footnotes on a blog!

1. Research consistently demonstrates a negatively correlated relationship between age and nonresponse; young users are more likely to under-respond, increasing their odds of being weighted in a population (and increasing their margins of error).  Research is mixed on the relationship between ethnicity and nonresponse.