Posts filed under 'research methods'
Amazon Mechanical Turk is a platform and market for human intelligence tasks (HITs) that are submitted by requesters and completed by workers (or “turkers”). Each HIT is associated with a payment, often a few cents. This post covers some basics of Mechanical Turk and shows its lack of designed-in support for dynamic reprioritization is problematic for some uses. I also mention some other factors that influence latency and throughput.
With mTurk one can create a HIT that asks someone to rate some search results for a query, evaluate the credibility of a Wikipedia article, draw a sheep facing left, enter names for a provided color, annotate a photo of a person with pose information, or create a storyboard illustrating a new product idea. So Mechanical Turk can be used in many ways for basic research, building a training set for machine learning, or actually enabling a (perhaps prototype) service in use through a kind of Wizard-of-Oz approach. Additionally, I’ve used mTurk to code images captured by participants in a lab experiment (more on this in another post or article).
When creating HITs, a requester can specify a QuestionForm (QF) (e.g., via command line tools or an SDK) that is then presented to the worker by Amazon. This can include images, free text answers, multiple choice, etc. Additionally one can embed Flash or Java objects in it. But the easiest way of creating HITs is to use a QF and not create a Java or Flash application of one’s own. This is especially true for HITs that are handled well by the basic question form. The other option is to create an ExternalQuestion (EQ), which is hosted on one’s own server and is displayed in an iFrame. This provides greater freedom but requires additional development and it is you that host the page (though you can do so through Amazon’s S3). QF HITs (without embeds) also offer a familiar interface to workers (though it is possible to create a more efficient, custom interface by e.g. making all the targets larger). So when possible, it is often preferable to use a QF rather than an EQ.
For some of the uses of mTurk for powering a service, it can be important to minimize latency for specific HITs, including prioritizing particular new HITs over previously created HITs. For example, after some HIT has not been completed for a specific period after creation, it may still be important to complete it, but when it is completed may become less important. This can happen easily if there the value of a HIT being completed has a sharp drop off after some time.
This should be done while maintaining high throughput; that is, you don’t want to reduce the rate at which your HITs are completed. When there are more HITs of the same type, workers can check a box to immediately start the next HIT of the same type when they submit the current one (see screenshot). Workers will often complete many HITs of the same type in a row. So throughput can drop substantially if any workers run out of HITs of the same type at any point: they may switch to another HIT type, or if they do your HITs once more appear there will be a delay. As we’ll see, these two requirements don’t seem to be well met by the platform — or at least certain uses of it.
Mechanical Turk does not provide a mechanism for prioritizing HITs of the same type, so without deleting all but particular high-priority HITs of that type, there is not a way to ensure that some particular HIT gets done before the rest. And deleting the other HITs would hurt throughput and increase latency for any new high-priority HITs added int he near future (since workers won’t simply start these once they finish their previous HITs).
EQs allow one to avoid this problem. Unlike with QF HITs (without Flash and Java embeds), one does not have to specify the full content of the HIT in advance. When a worker accepts an EQ HIT, you can dynamically serve up the HIT you want to depending on changing priorities. But this means that you can’t take advantage of e.g. the simplicity of creating and managing data from QF HITs. So though there are ways of coping, adding dynamic reprioritization to Mechanical Turk would be a boon for time-sensitive uses.
There are, of course, other factors that influence latency and throughput on mTurk when (EQ) HITs are reprioritized. Here are a few:
- HIT and sub-tasks duration. How long does it take for workers to complete a HIT, which may be composed of multiple sub-tasks? A worker cannot be assigned a new HIT until they complete (or reject) the previous one. This can be somewhat avoided by creating longer HITs that are subdivided into dynamically selected sub-tasks. This can be done with an EQ HIT or an embedded Flash or Java application in a QF HIT. But the sub-task duration is always a limiting factor, unless one is willing to force abortion of the current sub-task, replacing it will still in progress (with an EQ, Flash, or Java).
- Available workers. How many workers are logged into mTurk and completing task? How many are currently switching HIT types? This can vary with the time of day.
- Appeal of your HITs. How much do workers like your HITs — are they fun? How much do you pay for how much you ask? How many of their completed assignments do you approve?
- Reliability. How accurate or precise must your results be? How many workers do you need to complete a HIT before you have reliable results? Do other workers need to complete meta-HITs before the data can be used?
July 24th, 2008
Much of current human-computer interaction (HCI) research focuses on novice users in “walk-up and use” scenarios. I can think of three major causes for this:
- A general shift from examining non-discretionary use to discretionary use
- How much easier it is to find (and not train) study participants unfamiliar with a system than experts (especially with a system that is only a prototype)
- The push from practitioners in the direction, especially with the advent of the Web, where new users just show up at your site, often deep-linked
This focus sometimes comes in for criticism, especially when #2 is taken as a main cause of the choice.
On the other hand, some research threads in HCI continue to focus on expert use. As I’ve been reading a lot of research on both human performance modeling and situated & embodied approaches to HCI, it has been interesting to note that both instead have (comparatively) a much bigger focus on the performance and experience of expert and skilled use.
Grudin’s “Three Faces of Human-Computer Interaction” does a good job of explaining the human performance modeling (HPM) side of this. HPM owes a lot to human factors historically, and while The Psychology of Human-Computer Interaction successfully brought engineering-oriented cognitive psychology to the field, it was human factors, said Stuart Card, “that we were trying to improve” (Grudin 2005, p. 7). And the focus of human factors, which arose from maximizing productivity in industrial settings like factories, has been non-discretionary use. Fundamentally, it is hard for HPM to exist without a focus on expert use because many of the differences — and thus research contributions through new interaction techniques — can only be identified and are only important for use by experts or at least trained users. Grudin notes:
A leading modeler discouraged publication of a 1984 study of a repetitive task that showed people preferred a pleasant but slower interaction technique—a result significant for discretionary use, but not for modeling aimed at maximizing performance.
Situated action and embodied interaction approaches to HCI, which Harrison, Tatar, and Senger (2007) have called the “third paradigm of HCI”, are a bit different story. While HPM research, like a good amount in traditional cognitive science generally, contributes to science and design by assimilating people to information processors with actuators, situated and embodied interaction research borrows a fundamental concern of ethnomethodology, focusing on how people actively make behaviors intelligible by assimilating them to social and rational action.
There are at least three ways this motivates the study of skilled and expert users:
- Along with this research topic comes a methodological concern for studying behavior in context with the people who really do it. For example, to study publishing systems and technology, the existing practices of people working in such a setting of interest are of critical importance.
- These approaches emphasize the skills we all have and the value of drawing on them for design. For example, Dourish (2001) emphasizes the skills with which we all navigate the physical and social world as a resource for design. This is not unrelated to the first way.
- These approaches, like and through their relationships to the participatory design movement, have a political, social, and ethical interest in empowering those who will be impacted by technology, especially when otherwise its design — and the decision to adopt it — would be out of their control. Non-discretionary use in institutions is the paradigm prompting situation for this.
I don’t have a broad conclusion to make. Rather, I just find it of note and interesting that these two very different threads in HCI research stand out from much other work as similar in this regard. Some of my current research is connecting these two threads, so expect more on their relationship.
References
Dourish, P. (2001). Where the Action Is: The Foundations of Embodied Interaction. MIT Press.
Grudin, J. (2005). Three Faces of Human-Computer Interaction. IEEE Ann. Hist. Comput. 27, 4 (Oct. 2005), 46-62.
Harrison, S., Tatar, D., and Senger, P. (2007). The Three Paradigms of HCI. Extended Abstracts CHI 2007.
May 27th, 2008
As I’ve blogged before, I spoke at the Texting 4 Health conference on the topic of research methods for mobile messaging. One method I covered was an interesting use of Wizard of Oz techniques for designing mobile services. I’ve since started getting some of this material in writing for the Texting 4 Health book. Here is a taste of that material, minus the health-specific focus and examples.
——-
Just like the famous Wizard of Oz, one can simulate something impressive with a just a humble person behind the curtain — and use this simulation to inform design decisions. When using a Wizard of Oz technique to study a prototype, a human “wizard” carries out functions that, in a deployed application or service, would be handled by a computer. This can allow evaluating a design without fully building what can be expensive back-end parts of the system (Kelley 1984). The technique is often used in recognition-based interfaces, but it also has traditional applications to identifying usability problems and carrying out experiments in which the interaction is systematically manipulated.
Wizard of Oz techniques are well suited to prototyping mobile services, especially those using mobile messaging (SMS, MMS, voice messaging). When participants send a request, a wizard reads or listens to it and chooses the appropriate response, or just creates it on-the-fly. Since all user actions in mobile messaging are discrete messages and (depending on the application) the user can often tolerate a short delay, a few part-time wizards, such as you and a colleague, can manage a short field trial. As you’ll see, this can be used for purposes beyond many traditional uses of a Wizard of Oz.
Probing photo consumption needs with realistic motivations
One use for this technique in designing a mobile messaging service is a bit like a diary study. In designing an online and mobile photography service, we wanted to better understand what photos people wanted to view and what prompted these desires. Instead of just making diary entries, participants actually made voice requests to the system for photos – and received a mobile message with photos fitting the request in return. We didn’t need to first build a robust system that could do this; a few of us served as wizards, listening to the request, doing a couple manual searches, and choosing which photos to return on demand. Though this can be done with a normal voice call, we used a mobile client application that also recorded contextual information not available via a normal voice call (e.g. location), so that participants could make context-aware requests as they saw fit (e.g. “I want too see photos of this park”)
In this case, we didn’t plan to (specifically) create a voice-based photo search system; instead, like a diary study, this technique served as a probe to understand what we should build. As a probe it provided realistic motivations for submitting requests, as the request would actually be fulfilled. This design research, in additional to other interviews and a usability study, informed our creation of Zurfer, a mobile application that supports exploring and conversing around personalized, location-aware channels of photos.
It is great if the Wizard of Oz prototype is quite similar to what you later build, but it can yield valuable insights even if not. Sometimes it is precisely these insights that can lead you to substantially change your design.
This study design can apply in designing many mobile services. As in our photos study, participants can be interviewed about the trigger for the requests (why did they want that media or information) and how satisfied they were with the (human-created) responses.
References
Kelley, J.F. (1984). An iterative design methodology for user-friendly natural language office information applications. In ACM Trans. Inf. Syst., vol. 2, pp. 26-41.
May 5th, 2008
Two personal-professional narratives that I’ve been somewhat familiar with for a while have recently highlighted for me the significance of riskful decisions and thinking in academia. I think the stories are interesting on their own, but they also emphasize some questions and concerns for the functioning of scholarly inquiry.
The first is about the American philosopher Donald Davidson, whose work has long been of great interest to me (and was the topic of my undergraduate Honors thesis). The second is about Cliff Nass (Clifford Nass), Professor of Communication at Stanford, an advisor and collaborator. The major published source I draw on for each of these narratives is an interview: for Davidson’s story, it is an interview by Ernest Lepore (2004), a critic and expositor of Davidson’s philosophy; for Cliff Nass, it is an interview by Tamara Adlin (2007). After sharing these stories, I’ll discuss some similarities and briefly discuss risk-taking in decisions and thinking.
Donald Davidson is considered one of the most important and influential philosophers of the past 60 years, and he is my personal favorite. Davidson is often described as a highly systematic philosopher — uncharacteristically so for 20th century philosophy, in that his contributions to several areas of philosophy (philosophy of language, mind, and action, semantics, and epistemology) are deeply connected in their method and the proposed theories. He is the paradigmatic programmatic philosopher of the 20th century.
Despite this, Davidson’s philosophical program did not emerge until relatively late in his career. The same is true of his publications in general. Only after accepting a tenure track position at Stanford in 1951 (which was then still up-and-coming, though quickly, in philosophy) did he begin to publish (nothing was even in the “pipeline” previous to this). This began under the wing of the younger Patrick Suppes, with whom Davidson co-authored a book (1957) on decision theory. His first philosophical article appears in 1963 (which he authored alone only through an unexpected death). As Davidson puts it in an interview with Ernest Lepore, “I was very inhibited so far as publication was concerned” and was worried “that the minute I actually published something, everyone was going to jump on me” (Davidson 2004).
Then Davidson published “Actions, Reasons and Causes” (1963), twelve years after joining the Stanford faculty. It argues against the late-Wittgensteinian dogma that reasons are not also causes. It is only with this paper that there was a publication by Davidson that drew significant attention from the community (beginning with a presentation of the paper at a meeting of the American Philosophical Association). This paper has been hugely influential and alone identified Davidson as an important thinker in the field, though he was surprised the reception was not as overwhelming as he had thought: “I didn’t realize that if you publish, as far as I can tell, no one was going to pay any attention.” Many responses, both positive and critical, did eventually come, and Davidson went on to publish many highly influential papers, reaching the height of his immense scholarly influence in the 1970s and 1980s.
Clifford Nass is widely known researcher in the psychology of human-computer interaction (HCI). With Byron Reeves, he wrote The Media Equation (1996), which presents research carried out at Stanford University on how people respond in mediated interactions (e.g. with computers and televisions) by overextending social rules normally applied to other people. This hints at the (here simplified) straight, bold line of Nass’s research program: take a finding from social psychology, replace the second human with a computer, see if you get the same results. This exact strategy has been modified and expanded from, but the general consistency of Nass’s program over many years is striking for HCI: unlike in psychology, for example, in HCI there are many investigators seeking low-hanging fruit and quickly moving on to new projects.
Nass likes to refer to his “accidental PhD”, as he hadn’t intended to get a PhD in sociology. After working for a year at Intel, he was planning to matriculate in a electrical engineering PhD program, but an unexpected death postponed that. “[J]ust to bide my time and to have some flexibility, I ended up doing a sociology degree,” says Nass. He did his dissertation on the role of pre-processing jobs in labor, taking an approach that was radical in its elimination of a role for people and that connected with contemporary research by social science outsiders doing “sociocybernetics”. With such a dissertation topic (and the dissertation itself unfinished), finding a job did not seem easy at the outset: “It’s a nutty topic. I was going to be in trouble getting jobs. I had published stuff and was doing work and all that, but my dissertation was so weird” (Adlin 2007).
There was, however, a bit of luck, well taken advantage of by Nass: the Stanford Communication Department was under construction and looking to hire some folks doing weird work. So when Nass interviewed, impressing both them and the Sociology Department, he got the job, despite knowing nothing about Communication as a discipline and having been to no conferences in the field. After beginning at Stanford, Nass was seeking a research program, as clearly there was something wrong, at least when it came to getting it accepted for academic publication, with his previous work: “I was having a terrible time getting my work accepted. In fact, to this day I’ve still never published anything off my dissertation, 20-odd years later. Because again, no field could figure out who owned the material. I got reviews like, ‘This work is offensive.’”
But Nass couldn’t settle on any normal research program. He wanted to examine how people might treat computers socially. Getting funding for this work wouldn’t have been easy, but he got a grant that the grant administrator described as the 1 of 35 given that they chose to give to the “weirdest project that was proposed”. It wasn’t all easy from there, of course. For example, it took some time to design and carry out successful experiments in this program — and even longer to get the results published. But this risk-taking in distributing this grant helped enable the work to continue.
Cliff Nass is very clear about the role riskful decisions, in admissions, hiring, and funding, played in his success:
I was very lucky. I fear that those times are gone. I really do fear to a tremendous degree that the risk-taking these people were willing to do for me, to give me an opportunity, are gone. I try to remember that. [...]
I benefited from the willingness of people to say, “We’re just going to roll the dice here.”
Of course, it isn’t just Cliff who got lucky; in a sense we all did. His work has been an important influence in HCI and has contributed to our stores of both generalizable knowledge and new lenses for approaching how we get on in the world.
What does it mean for academic research, and science generally, if this choice and ability to take these risks evaporates? There is incredible competition for academic positions now, more so in some fields than others. And the best tool in getting a job is a whole list of publications accepted in important, mainstream journals in the field. There is a lot written about the competition for academic jobs and criteria for wading through applicants to sometimes a safe option. There are case studies of families of disciplines; for example, a study of the biosciences argues that market forces are failing to create sufficient job prospects for young investigators (Freeman et al. 2001).
I won’t review them all here. Instead I suggest an article for general readers from The New York Times about state and regional colleges’ use of non-tenure track positions, which has an impact of the institutions’ bottom line and flexibility (Finder 2007). This is part of a wider trend in how tenure is used that also impacts the academic freedom and resources that scholars have to pursue new research (Richardson 1999).
Enabling riskful thinking
Hans Ulrich Gumbrecht argues that “riskful thinking” is central to the value of the humanities and arts in academia. He defines riskful thinking as investigation that can’t be expected to produce results interpretable as easy answers, but that instead is likely to produce or highlight complex and confusing phenomena and problems. But I think that this is more broadly true. Riskful thinking is critical to interdisciplinary and pre-paradigmatic sciences, or disciplines long doing normal science but in need of a shake-up. These are situations where compelling phenomena can become paradigmatic cases for study and powerful vocabularies can allow formulating new problems and theories.
What threatens riskful thinking, and how can we enable it? What is so great about riskful thinking anyway, and what makes some riskful thinking so successful, while much of it is likely to fail? At Nokia Research Center in Palo Alto, our lab head John Shen champions the importance of risk taking in industry research, but also argues that risk-taking is often misunderstood and that it is only some kinds of risk-taking that are most important to cultivate in industry research.
—
Finally, a list of Davidson-Nass similarities, just for fun:
- Both were hired to tenure track positions at Stanford, where they first did and published highly influential work
- Both are easily and widely seen as highly programmatic, having defined a clear research program challenging to currently popular approaches and beliefs in their fields
- Both had great difficulty finding early, publishable success with their research programs, even after ceasing their early work (Davidson: Plato, empirical decision theory; Nass: information processing models of the labor force)
- Both had other draws and distractions (Davidson: business school, teaching plane identification in WWII; Nass: being a professional magician, working at Intel)
- Both produced dissertations viewed by others in the discipline as odd (Davison: Quine “was a little mystified by my writing on this. He never talked to me about it.”; Nass: “my PhD thesis was so bizarre”)
References
Adlin, T. (2007). An interview with Cliff Nass. UX Pioneers. http://www.adlininc.com/uxpioneers/home_popular_row_2/interview_cliff_nass
Davidson, D. (1963). Actions, Reasons, and Causes. Journal of Philosophy, 60(23), 685-700.
Davidson, D., & Suppes, P. (1957). Decision Making: An Experimental Approach. Stanford University Press.
Finder, A. (2007, November 20). Decline of the Tenure Track Raises Concerns. The New York Times.
Freeman, R., Weinstein, E., Marincola, E., Rosenbaum, J., & Solomon, F. (2001). Careers: Competition and Careers in Biosciences. Science, 294(5550), 2293-2294.
Lepore, E. (2004). Interview with Donald Davidson. In Problems of Rationality, Oxford University Press, 2004, pp. 231-266.
Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. In Proc. of CHI 1994. ACM Press.
Reeves, B., & Nass, C. (1996). The media equation: how people treat computers, television, and new media like real people and places. Cambridge University Press.
Richardson, J. T. (1999). Tenure in the New Millenium. National Forum, 79(1), 19-23.
Sanford, J. (2000, November 17). ‘Elementary pleasures’ and ‘riskful thinking’ matter to Gumbrecht. Stanford Report.
April 29th, 2008
As I blogged already, I attended and spoke at the first Texting 4 Health conference at Stanford University last week. You can see my presentation slides at SlideShare here, and the program, with links to the slides for most speakers is here.
The conference was very interesting, and there was quite the mix of participants — both speakers and others. There were medical school faculty, business people, people from NGOs and foundations, technologists, representatives of government agencies and centers, futurists, and social scientists. Everyone had something to learn — I know I did. This also made it somewhat difficult as a speaker because it is hard to know how best to reach, inform, and hold the interest of such a diverse audience: what is common ground with some is entirely new territory with others.
I think my favorite session was “Changing Health Behavior via SMS”. The methods used by the panelists to evaluate their interventions were both interesting to reflect on and good tools for persuading me of the importance and effectiveness of their work. One of my reflections was about what factors to vary in doing experiments on health interventions: there is (reasonable) focus on having a no-SMS control condition, and there are very few studies with manipulations of dimensions more fine-grained. Of course, the field is young and I understand how important true controls are in medical domains, but I think that real progress in understanding mobile messaging and designing effective interventions will require looking at more subtle and theoretically valuable manipulations.
You can see other posts about the conference here and here. And the conference Web site is also starting a blog to watch in the future.
March 7th, 2008
On February 29th I’m speaking at Texting 4 Health, a conference at Stanford University about using mobile messaging for health interventions and research. I’ll be talking about mobile messaging research methods I’ve used to study mobile persuasive technology. Like Mobile Persuasion 2007, it will feature a fast-paced, single-track program with time to meet and talk with participants from health, technology, policy, and research communities.
February 11th, 2008
Yes, that spells ASSIST.
Check out this call for proposals from DARPA (also see Wired News). This research program is designed to create and evaluate systems that use sensors to capture soldiers’ experiences in the field, thus allowing for (spatially and temporally) distant review and analysis of this data, as well as augmenting their abilities while still in the field.
I found it interesting to consider differences in requirements between this program and others that would apply some similar technologies and involve similar interactions — but for other purposes. For example, two such uses are (1) everyday life recording for social sharing and memory and (2) rich data collection as part of ethnographic observation and participation.
When doing some observation myself, I strung my cameraphone around my neck and used Waymarkr to automatically capture a photo every minute or so. Check out the results from my visit to a flea market in San Francisco.
Photos of two ways to wear a cameraphone from Waymarkr. Incidentally, Waymarkr uses the cell-tower-based location API created for ZoneTag, a project I worked on at Yahoo! Research Berkeley.
Also, for a use more like (1) in a fashion context, see Blogging in Motion. This project (for Yahoo! Hack Day) created a “auto-blogging purse” that captures photos (again using ZoneTag) whenever the wearer moves around (sensed using GPS).
January 15th, 2008
Diary studies are widely used in human-computer interaction research, but also in user experience research as practiced in product R&D groups. Bolger, Davis, & Rafaeli (2003) is a good review of diary research methods from a Psychology perspective. It gives practical guidance in what research questions are suited to these methods, design decisions, tools, and analysis.
Though it covers state-of-the-art technology used for these methods, I think the argument below for the taxonomy of methods used in this paper needs revision in light of new diary methods, e.g. those made possible by using context-aware devices for signaling participants. Here is the argument for the two-way taxonomy (p. 588):
Diary studies have often been classified into the three categories of interval-, signal-, and event-contingent protocols (e.g., Wheeler & Reis 1991). The interval-contingent design, the oldest method of daily event recording, requires participants to report on their experiences at regular, predetermined intervals. Signal-contingent designs rely on some signaling device to prompt participants to provide diary reports at fixed, random, or a combination of fixed and random intervals. Event-contingent studies, arguably the most distinct design strategy, require participants to provide a self-report each time the event in question occurs. This design enables the assessment of rare or specialized occurrences that would not necessarily be captured by fixed or random interval assessments.
As we see it, diary studies serve one of two major purposes: the investigation of phenomena as they unfold over time, or the focused examination of specific, and often rare, phenomena. It appears to us that the three-way classification blends this conceptual distinction with the technological issue of signaling. Instead, we incorporate interval- and signal-contingent designs into a single category, which we call time-based designs.
This argument to collapse the taxonomy does not account for methods in which participants are signaled based on factors other than time. For example, diary studies can include signaling participants to create an entry based on events that are automatically detected by the system: this occurs when the system is immediately aware of the event because it is an interaction with the system (e.g. the participant has just completed a phone call) or because it can infer an appropriate change in state (e.g. the participant has just moved from one place to another, as detected by readings from GPS).
Bolger, N., Davis, A., & Rafaeli, E. (2003). Diary Methods: Capturing Life as it is Lived. Annual Review of Psychology, 54(1), 579-616.
Wheeler, L., & Reis, H. (1991). Self-recording of everyday life events: origins, types, and uses. Journal of personality, 59(3), 339-354.
January 9th, 2008