The “friendly world syndrome” induced by simple filtering rules
I’ve written previously about how filtered activity streams [edit: i.e. news feeds] can lead to biased views of behaviors in our social neighborhoods. Recent conversations with two people writing popular-press books on related topics have helped me clarify these ideas. Here I reprise previous comments on filtered activity streams, aiming to highlight how they apply even in the case of simple and transparent personalization rules, such as those used by Twitter.
—
Birds of a feather flock together. Once flying together, a flock is also subject to the same causes (e.g., storms, pests, prey). Our friends, family, neighbors, and colleagues are more similar to us for similar reasons (and others). So we should have no illusions that the behaviors, attitudes, outcomes, and beliefs of our social neighborhood are good indicators of those of other populations — like U.S. adults, Internet users, or homo sapiens of the past, present, or future. The apocryphal Pauline Kael quote “How could Nixon win? No one I know voted for him” suggests both the ease and error of this kind of inference. I take it as a given that people’s estimates of larger populations’ behaviors and beliefs are often biased in the direction of the behaviors and beliefs in their social neighborhoods. This is the case with and without “social media” and filtered activity streams — and even mediated communication in general.
That is, even without media, our personal experiences are not “representative” of the American experience, human experience, etc., but we do (and must) rely on it anyway. One simple cognitive tool here is using “ease of retrieval” to estimate how common or likely some event is: we can estimate how common something is based on how easy it is to think of. So if something prompts someone to consider how common a type of event is, they will (on average) estimate the event as more common if it is more easy to think of an example of the event, imagine the event, etc. And our personal experiences provide these examples and determine how easy they are to bring to mind. Both prompts and immediately prior experience can thus affect these frequency judgments via ease of retrieval effects.
Now this is not to say that we should think as ease of retrieval heuristics as biases per se. Large classes and frequent occurrences are often more available to mind than those that are smaller or less frequent. It is just that this is also often not the case, especially when there is great diversity in frequency among physical and social neighborhoods. But certainly we can see some cases where these heuristics fail.
Media are powerful sources of experiences that can make availability and actual frequency diverge, whether by increasing the biases in the direction of projecting our social neighborhoods onto larger population or in other, perhaps unexpected directions. In a classic and controversial line of research in the 1970s and 80s, Gerbner and colleagues argued that increased television-watching produces a “mean world syndrome” such that watching more TV causes people to increasingly overestimate, e.g., the fraction of adult U.S. men employed in law enforcement and the probability of being a victim of violent crime. Their work did not focus on investigating heuristics producing these effects, but others have suggested the availability heuristic (and related ease of retrieval effects) as at work. So even if my social neighborhood has fewer cops or victims of violent crime than the national average, media consumption and the availability heuristic can lead me to overestimate both.
Personalized and filtered activity streams certainly also affect us through some of the same psychological processes, leading to biases in users’ estimates of population-wide frequencies. They can aIso bias inference about our own social neighborhoods. If I try to estimate how likely a Facebook status update by a friend is to receive a comment, this estimate will be affected by the status updates I have seen recently. And if content with comments is more likely to be shown to me in my personalized filtered activity stream (a simple rule for selecting more interesting content, when there is too much for me to consume it all), then it will be easier for me to think of cases in which status updates by my friends do receive comments.
In my previous posts on these ideas, I have mainly focused on effects on beliefs about my social neighborhood and specifically behaviors and outcomes specific to the service providing the activity stream (e.g., receiving comments). But similar effects apply for beliefs about other behaviors, opinions, and outcomes. In particular, filtered activity streams can increase the sense that my social neighborhood (and perhaps the world) agrees with me. Say that content produced by my Facebook friends with comments and interaction from mutual friends is more likely to be shown in my filtered activity streams. Also assume that people are more likely to express their agreement in such a way than substantial disagreement. As long as I am likely to agree with most of my friends, then this simple rule for filtering produces an activity stream with content I agree with more than an unfiltered stream would. Thus, even if I have a substantial minority of friends with whom I disagree on politics, this filtering rule would likely make me see less of their content, since it is less likely to receive (approving) comments from mutual friends.
I’ve been casually calling this larger family of effects this the “friendly world syndrome” induced by filtered activity streams. Like the mean world syndrome of the television cultivation research described above, this picks out a family of unintentional effects of media. Unlike the mean world syndrome, the friendly world syndrome includes such results as overestimating how many friends I have in common with my friends, how much positive and accomplishment-reporting content my friends produce, and (as described) how much I agree with my friends.1
Even though the filtering rules I’ve described so far are quite simple and appealing, they still are more consistent with versions of activity streams that are filtered by fancy relevance models, which are often quite opaque to users. Facebook News Feed — and “Top News” in particular — is the standard example here. On the other hand, one might think that these arguments do not apply to Twitter, which does not apply any kind of machine learning model estimating relevance to filtering users’ streams. But Twitter actually does implement a filtering rule with important similarities to the “comments from mutual friends” rule described above. Twitter only shows “@replies” to a user on their home page when that user is following both the poster of the reply and the person being replied to.2 This rule makes a lot of sense, as a reply is often quite difficult to understand without the original tweet. Thus, I am much more likely to see people I follow replying to people I follow than to others (since the latter replies are encountered only from browsing away from the home page. I think this illustrates how even a straightforward, transparent rule for filtering content can magnify false consensus effects.
One aim in writing this is to clarify that a move from filtering activity streams using opaque machine learning models of relevance to filtering them with simple, transparent, user-configurable rules will likely be insufficient to prevent the friendly world syndrome. This change might have many positive effects and even reduce some of these effects by making people mindful of the filtering.3 But I don’t think these effects are so easily avoided in any media environment that includes sensible personalization for increased relevance and engagement.
- This might suggest that some of the false consensus effects observed in recent work using data collected about Facebook friends could be endogenous to Facebook. See Goel, S., Mason, W., & Watts, D. J. (2010). Real and perceived attitude agreement in social networks. Journal of Personality and Social Psychology, 99(4), 611-621. doi:10.1037/a0020697 [↩]
- Twitter offers the option to see all @replies written by people one is following, but 98% of users use the default option. Some users were unhappy with an earlier temporary removal of this feature. My sense is that the biggest complaint was that removing this feature removed a valuable means for discovering new people to follow. [↩]
- We are investigating this in ongoing experimental research. Also see Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H., & Simons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61(2), 195-202. doi:10.1037/0022-3514.61.2.195 [↩]
Search queries in referrer headers: Technical knowledge, privacy, and the status quo
I have been fascinated by Christopher Soghoian‘s complaint to the FTC about Google’s practices of including search query information in the HTTP referrer header.
In summary, Google has taken proactive efforts to ensure that Web site owners that get visitors from Google search receive the search terms entered by Google’s users. Meanwhile, Google has agreed that search query data is personally sensitive information and that it does not disclosure this information, except under specific, limited circumstances; this is reflected in its privacy policy. Note that Google has not just let the URL do the work, but has specifically worked to make the referrer header include search terms (and additional information) when it has adopted techniques that would otherwise prevent these disclosures from being made. (For a fuller summary, see his blog post and this WSJ article. Or this article at Search Engine Land.)
I am not going to discuss the ethics and legal issues in this particular case. Instead, I just want to draw attention to how this issue reveals the importance of technical knowledge in thinking about privacy issues.
A common response from people working in the Internet industry is that Soghoian is a non-techie that has suddenly “discovered” referrer headers. For example, Danny Sullivan writes “former FTC employee discovers browsers sends referrer strings, turns it into google conspiracy”. (Of course, Soghoian is actually technically savvy, as reading the complaint to the FTC makes clear.)
What’s going on here? Folks with technical knowledge perceive search query disclosure as the status quo (though I bet most don’t often think about the consequences of clicking on a link after a sensitive search).
But how would most Internet users be aware of this? Certainly not through Google’s statements, or through warnings from Web browsers. One of the few ways I think users might realize this is happening is through query-highlighting — on forums, mailing list archives, and spammy pages. So a super-rational user who cares to think about how that works, might guess something like this is going on. But I doubt most users would actively work out the mechanisms involved. Futhermore, their observations likely radically underdetermine the mechanism anyway, since it is quite reasonable that a Web browser could do this kind of highlighting directly, especially for formulaic sites, like forums. Even casual use of Web analytics software (such as Google Analytics) may not make it clear that this per-user information is being provided, since aggregated data could reasonably be used to present summaries of top search queries leading to a Web site.1
This should be a reminder why empirical studies of privacy attitudes and behaviors are useful: us techie folks often have severe blind spots. I don’t know that this is just a matter of differences in expectations, but rather involves differences in preferences. Over time, these expectations change our sense of the status quo, from which we can calibrate our preferences and intentions.
Google has worked to ensure that referrer headers continue to include search query information — even as it adopts techniques that would make this not happen simply by the standard inclusion of the URL there.2 A difference in beliefs about the status quo puts these actions by Google in a different context. For us techies, that is just maintaining the status quo (which may seem more desirable, since we know it’s the industry-wide standard). For others, it might seem more like Google putting advertisers and Web site owners above its promises to its users about their sensitive data.
Economic imperialism and causal inference
And I, for one, welcome our new economist overlords…
Readers not in academic social science may take the title of this post as indicating I’m writing about the use of economic might to imperialist ends.1 Rather, economic imperialism is a practice of economists (and acolytes) in which they invade research territories that traditionally “belong” to other social scientific disciplines.2 See this comic for one way you can react to this.3
Economists bring their theoretical, statistical, and research-funding resources to bear on problems that might not be considered economics. For example, freakonomists like Levitt study sumo wrestlers and the effects of the legalization of abortion on crime. But, hey, if the Commerce Clause means that Congress can legislate everything, then, for the same reasons, economists can — no, must — study everything.
I am not an economist by training, but I have recently had reason to read quite a bit in econometrics. Overall, I’m impressed.4 Economists have recently taken causal inference — learning about cause and effect relationships, often from observational data — quite seriously. In the eyes of some, this has precipitated a “credibility revolution” in economics. Certainly, papers in economics and (especially) econometrics journals consider threats to the validity of causal inference at length.
On the other hand, causal inference in the rest of the social sciences is simultaneously over-inhibited and under-inhibited. As Judea Pearl observes in his book Causality, lack of clarity about statistical models (that social scientists often don’t understand) and causality has induced confusion about distinctions between statistical and causal issues (i.e., between estimation methods and identification).5
So, on the one had, many psychologists stick to experiments. Randomized experiments are, generally, the gold standard for investigating cause–effect relationships, so this can and often does go well. However, social psychologists have recently been obsessed with using “mediation analysis” to investigate the mechanisms by which causes they can manipulate produce effects of interest. Investigators often manipulate some factors experimentally and then measure one or more variables they believe fully or partially mediate the effect of those factors on their outcome. Then, under the standard Baron & Kenny approach, psychologists fit a few regression models, including regressing the outcome on both the experimentally manipulated variables and the simply measured (mediating) variables. The assumptions required for this analysis to identify any effects of interest are rarely satisfied (e.g., effects on individuals are homogenous).6 So psychologists are often over-inhibited (experiments only please!) and under-inhibited (mediation analysis).
Likewise, in more observational studies (in psychology, sociology, education, etc.), investigators are sometimes wary of making explicit causal claims. So instead of carefully stating the causal assumptions that would justify different causal conclusions, readers are left with phrases like “suggests” and “is consistent with” followed by causal claims. Authors then recommend that further research be conducted to better support these causal conclusions. With these kinds of recommendations awaiting, no wonder that economists find the territory ready for taking: they can just show up with econometrics tools and get to work on hard-won questions that “rightly belong to others”.
- Well, if economists have better funding sources, this might apply in some sense. [↩]
- For arguments in favor of economic imperialism, see Lazear, E.P. (1999). Economic imperialism. NBER Working Paper No. 7300. [↩]
- Or see this comic for imperialism by physicists. [↩]
- At least by the contemporary literature on what I’ve been reading on — IVs, encouragement designs, endogenous interactions, matching estimators. But it is true that in some of these areas econometrics has been able to fruitfully borrow from work on potential outcomes in statistics and epidemiology. [↩]
- Econometricians have made similar observations. [↩]
- For a bit on this topic, see the discussion and links to papers here. [↩]
Homophily and peer influence are messy business
Some social scientists have recently been getting themselves into trouble (and limelight) claiming that they have evidence of direct and indirect “contagion” (peer influence effects) in obesity, happiness, loneliness, etc. Statisticians and methodologists — and even science journalists — have pointed out their troubles. In observational data, peer influence effects are confounded with those of homophily and common external causes. That is, people are similar to other people in their social neighborhood because ties are more likely to form between similar people, and many external events that could cause the outcome are localized in networks (e.g., fast food restaurant opens down the street).
Econometricians1 have worked out the conditions necessary for peer influence effects to be identifiable.2 Very few studies have plausibly satisfied these requirements. But even if an investigator meets these requirements, it is worth remembering that homophily and peer influence are still tricky to think about — let along produce credible quantitative estimates of.
As Andrew Gelman notes, homophily can depend on network structure and information cascades (a kind of peer influence effect) to enable the homophilous relationships to form. Likewise, the success or failure of influence in a relationship can affect that relationship. For example, once I convert you to my way of thinking — let’s say, about climate change, we’ll be better friends. To me, it seems like some of the downstream consequences of our similarity should be attributed to peer influence. If I get fat and so you do, it could be peer influence in many ways: maybe that’s because I convinced you that owning a propane grill is more environmentally friendly (and then we both ended up grilling a lot more red meat). Sounds like peer influence to me. But it’s not that me getting fat caused you to.
Part of the problem here is looking only at peer influence effects in a single behavior or outcome at once. I look forward to the “clear thinking and adequate data” (Manski) that will allow us to better understand these processes in the future. Until then: scientists, please at least be modest in your claims and radical policy recommendations. This is messy business.
- They do statistics but speak a different language than big “S” statisticians — kind of like machine learning folks. [↩]
- For example, see Manski, C. F. (2000). Economic analysis of social interactions. Journal of Economic Perspectives, 14(3):115–136. Economists call peer influence effects endogenous interactions and contextual interactions. [↩]
Aardvark’s use of Wizard of Oz prototyping to design their social interfaces
The Wall Street Journal’s Venture Capital Dispatch reports on how Aardvark, the social question asking and answering service recently acquired by Google, used a Wizard of Oz prototype to learn about how their service concept would work without building all the tech before knowing if it was any good.
Aardvark employees would get the questions from beta test users and route them to users who were online and would have the answer to the question. This was done to test out the concept before the company spent the time and money to build it, said Damon Horowitz, co-founder of Aardvark, who spoke at Startup Lessons Learned, a conference in San Francisco on Friday.
“If people like this in super crappy form, then this is worth building, because they’ll like it even more,” Horowitz said of their initial idea.
At the same time it was testing a “fake” product powered by humans, the company started building the automated product to replace humans. While it used humans “behind the curtain,” it gained the benefit of learning from all the questions, including how to route the questions and the entire process with users.
This is a really good idea, as I’ve argued before on this blog and in a chapter for developers of mobile health interventions. What better way to (a) learn about how people will use and experience your service and (b) get training data for your machine learning system than to have humans-in-the-loop run the service?
My friend Chris Streeter wondered whether this was all done by Aardvark employees or whether workers on Amazon Mechanical Turk may have also been involved, especially in identifying the expertise of the early users of the service so that the employees could route the questions to the right place. I think this highlights how different parts of a service can draw on human and non-human intelligence in a variety of ways — via a micro-labor market, using skilled employees who will gain hands-on experience with customers, etc.
I also wonder what UIs the humans-in-the-loop used to accomplish this. It’d be great to get a peak. I’d expect that these were certainly rough around the edges, as was the Aardvark customer-facing UI.
Aardvark does a good job of being a quite sociable agent (e.g., when using it via instant messaging) that also gets out of the way of the human–human interaction between question askers and answers. I wonder how the language used by humans to coordinate and hand-off questions may have played into creating a positive para-social interaction with vark.