Deborah Estrin PhD, Founding Director, Center for Embedded Networked Sensing (CENS)

In her UCLA bio Deborah Estrin is listed first as a Professor of Computer Science with a joint appointment in Electrical Engineering but she is also the Founding Director of the Center for Embedded Networked Sensing (CENS) where she and her team are leveraging mobile phones as sensing and data collection tools.

I recently had the opportunity to hear Deborah speak about Mobile Sensing at a Mobile Tech Salon in New York.  After listening to her explain how a smartphone in the hands of an average person can gather valuable scientific data, I knew she’d have a unique perspective to share on the use of mobile phones.

After a series of email messages, Deborah and I connected to talk about how smartphones are powerful sensing & data collection tools, the 3 types of mobile sensing projects she’s currently working on, and lessons she’s learned about implementing mobile data collection campaigns.  Here’s a condensation of our early morning conversation.

Q: Why are smartphones in the hands of an average person an attractive means of data collection?

A. What other device do so many people carry around because it’s important to them? A mobile phone is attractive because of its close proximity to people and smartphones serve multiple purposes which is what makes it so low cost and scalable.  It’s also a technology that continues to improve (for example, the whole mobile app store phenomenon).

So, it really has to do with it being a device that we can leverage because of the economies of scale.  Since a lot of people have them, you don’t have to justify the cost of the additional measurement or data campaign, entirely because most of the cost of the device and infrastructure was put in place for other reasons.

Q: Can you talk a bit about how you’re leveraging average people to collect data for your various projects?

A: We really do three types of data collection with mobile phones.  Some of what we do is more in the area of traditional citizen science, where there is some data that a scientist or somebody wants and we tap into average people to help collect it.  So in that case, we tap into the general public because to get that kind of density of observation is otherwise quite expensive, maybe infeasible.

One of our applications has to do with the monitoring of ecosystems, particularly what’s known as phenology –the stages of plants or invasive species — that you can’t view from remote sensing because they’re below canopy and at a smaller spatial scale.   To have people constantly going out and monitoring all the micro climates, and what’s growing where, is a very labor intensive thing.  However, if you can get people to do it with their mobile phone during the course of where they are walking in their neighborhood or hiking or what have you, you can leverage the density of people to make observations that are on a large scale.

The second and third things we do with mobile phones are much more directly related to why we tap into the general public.  The second type involves all sorts of data collection campaigns that people themselves want to do.  For example, people may want to collect different types of community data.  It might be high school kids documenting the state of the facilities on their campus,  or where they’re being harassed by police.   Or, it might be a community documenting unsafe intersections or lack of safe paths to school for young children.  In that context it’s people organizing themselves as communities and doing data campaigns of their own.  So who else would they tap into if not themselves?

The 3rd class of things we do is related to health and wellness.  With this type of project, people are collecting data about themselves to help themselves.  In this context there is no one else to tap into but the individual.

Q: Are there certain mobile phone features/capabilities that are better sensing/data collection tools than others?

A: A phone with the following capabilities makes it a really powerful platform:

  • GPS-Enabled so that everything can be automatically, at least to some extent, geocoded.
  • A nice touch screen interface.  It makes it really easy for people to do quick tagging and answering of questions.
  • A good imager.
  • Programmability.  By this I mean, the whole model that Apple launched and that Android has fallen on that makes it completely easy to download mobile applications to your phone.  That is really important because then you don’t have to wait for your service provider to officially sanction your App, which was common practice before the release of the iPhone.  An App store makes the whole notion of a phone being programmable, very easy.
  • SMS if you’re working in a context where you also want to be sure to get to people who don’t yet have smartphones.

Q:  Can any community Mobile App, like a girl scout App for example, be downloaded through an App store?

A: If a girl scout troop creates a new app, it might or might not get through the Apple App Store but you can post anything on the Android App store.  You can also post an app for an Android phone at some URL and people can download it.  There’s a phase of difference between iPhone and Android. IPhone is more controlled. You also get a more controlled experience, meaning every iPhone App looks good.  We choose to do a lot of development and exploration on Android because it allows for early and rapid prototyping and things like that.

Q: What types of data can be best collected via mobile phones, in terms of what can be usefully used, processed and have conclusions drawn from?

A. At this point, it’s hard to say.  It’s clearly best suited to observations that scale in the realm of what people see or experience in the moment. Photo Voice and Experience Sampling are what this is about.  Photo Voice has been used from the 80’s to teach people how to use systematic photography as a way of documenting things that need to be changed or for asset management and such.  Essentially, you’re allowing people to do local, in the moment, in the context observation for things you want that local, in the moment, in the context, observation.

Q: Are there other features or abilities of the mobile phone that you use during a data collection campaign?

A: We make use of the combination of the GPS and the accelerometer to capture what’s called actigraphy which is basically your mobility information. There are pedometers and high-end accelerometer devices that you could wear on your waist and while those might be more accurately and precisely calibrated, you can use your phone to get a pretty good view of generally how much are you moving around vs. driving vs. sitting throughout the day.  So on the phone itself we use accelorameter and GPS.

We’re also starting to do things related to broad-scale sleep studies that have to do with people’s sleep rhythms and sleep disruptions.  We collect the data partially through self-report, which involves asking people a few questions via their phone to assess the quality of their sleep.  We generally ask when they went to sleep, when they got up and how many times during the night did they get up.   To corroborate that information, we look at things like when did the person last touch their phone. Maybe they said they went to bed at 11pm and got up at 7am but the data we pull off the phone shows that in the middle of the night, the phone moved and that there were 3 text messages sent.

We corroborate the data in this way not so much to keep people from lying, it’s to help people with recall because even short sleep disruptions actually have effect. The accelerometer can play a key role and so can the phone itself.  Although it’s not true for all, specifically the elderly, the phone is the center of a person’s communication life. The amount of interaction going on with the phone, day and night, gives you another source of information that can be fused with activity like actigraphy and self-reports.

Q: Are there common mistakes people make when planning or executing participatory sensing projects and would you have any advice for someone just starting out?

A. Yes.

  1. Have some sense in your mind for the end-to-end goal, including how you are going to analyze the data and have that in place before you collect any data.
  2. Pretest with a few small numbers.  Just like any survey, you always pretest it because there are ways that you can ask questions that’s not asking the thing you want.  A pretest gives you a pretest of the data capture technique and it gives you a pretest of your basic analysis.
  3. You need to focus on problems that you’re going to get enough data for it to be representative. Start with things that lend themselves to small not big.  By this I mean, you want something that scales down, that still works when you’re at a smaller number.  Not something that is only useful when you get to really large numbers because it’s hard then to get to the large number.  It’s hard to get viral if it’s not doing something useful even at the smaller number.

Q: Are there any lessons you’ve learned in terms of the type of people to recruit to collect data with mobile phones?

A: Ah, you!  It doesn’t matter what their family income is.  It does not matter what their ethnicity is.  This is completely anecdotal, but through our experience there is no digital divide around economics, or race, or gender.  There is a digital divide, in this context, around age. You start to hit some gender and other demographic divisions when you get to middle age segments of the population.  And if you’re sampling the elderly, targeting their informal care givers might be the place to start.

Q: Do you train these citizens or do they pretty much have what they need to know from the start?

A: In our experience to date, it requires very little training.  However, if a person or group is creating a campaign, they need some training.  If they’re engaging other community members in the campaign, the other community members need very little training because the stuff is so easy to use (like Apps that you download).   Speaking of groups, you asked me about lessons and so one of them is:  if you just target individuals as individuals, I think it’s less successful than individuals as part of groups and communities.  The idea of targeting groups for data collection also sort of addresses that other issue of getting a density of coverage.  With groups you can plan a data campaign and have enough people, plus there’s a social aspect to it as well.

Post to Twitter

{ 19 comments… read them below or add one }