so here we are

I am back in Dar es Salaam, back at Twaweza, and back working on mobile data gathering. Before giving you an overview about the things that will be happening in the weeks to come, I will first use this post to recapitulate where we are coming from.

During August/September last year, we went into the field to draw a stratified sample of 550 respondents in Dar es Salaam, conducting face-to-face household interviews. Our extensive survey covered quite an impressively broad range of topics such as family composition, work situation, political orientations, infrastructure, health issues and children’s education (actually, given that the survey took about 1,5h to complete, I was somewhat surprised to see that almost all people took the time to answer all of our questions and that non-response rates were minimal). After completion of the questionnaire, respondents were asked if they owned or had access to a mobile phone. If so, we invited them to participate in follow-up mobile surveys. Out of those respondents that we interviewed, 76% had their own mobile phone and another 12,5% had access to the phone of another household member.

In the weeks after the baseline survey was finished, an infrastructure was set up by DataVision to call respondents for short surveys (10-15 questions) and transfer small credit top-ups after the interview as incentives. Initially, we also had planned to use USSD and IVR (and even a web application), but the technical difficulties in putting these in place seemed more serious than expected, and we resorted to simply calling respondents during the weekends.

The mobile survey system was up and running by January 2011, and in the following 18 weeks, respondents were called weekly and a rich body of real-time data were gathered, mainly about the quality of public services such as education, health, garbage collection, water and electricity. After some substantial drop-out in the first weeks, the panel size quickly stabilized around 330. Each weekend, respondents were called by the same interviewers, contributing to   sense of commitment so crucial for preventing attrition in panels.

Why people drop out
To get a better understanding of the reasons for non-contact / non-response, a field re-visit to nonparticipating respondents was organized four weeks into the mobile data gathering. Interviews revealed that the main factors that contributed to panel drop-out were the widespread sharing of handhelds, mobiles that were lost or stolen, the use of multiple SIM cards, lack of reliable electricity to charge phone and the fact that some respondents had only resided in the city temporarily.

About three weeks ago, the data gathering was temporarily paused as the whole project was transferred from Twaweza to the World Bank. Interested in exploring the potential of mobile surveys for independent third party monitoring, the World Bank will be running the panel for six months, conducting biweekly surveys on topics related to its activities in Dar es Salaam.

So this is where we are.

But there is more to come:
To correct the bias in our mobile sample created by the lack of phone access in baseline respondents, we will be heading back into the field shortly to hand out 50 mobile phones. Also, we now seem to be ready to throw in two new technologies as data gathering channels: USSD and IVR. More on all this very shortly.

Posted in call center, IVR, nonresponse, sampling, USSD | Leave a comment

and…we’re back

After many months of silence on this blog, things are about to change.

Since its launch in August 2010, this platform has worked very successful in drawing attention to what we were doing and in establishing new connections with interested and interesting people. At the same time, this blog had unfortunately somewhat dried up after I had returned to my job at the University of Rotterdam, in October last year – also because I had not been able to keep involved close enough to report back on our progress.

But indeed, things are very much about to change.

In just a couple of weeks, I will pick up my work at Twaweza on the mobile survey project again. And I can’t wait to get back to that strange city of Dar es Salaam, talk to those people who have worked hard to keep this thing up and running, get their first-hand experiences with the day-to-day handling of the mobile panel waves (the first wave was launched after I had already left), to dive into the wealth of data produced, analyze dropout patterns, look into technology issues to be solved, and all of the other things that make up this exciting project.

So don’t despair, stick with me, great things are bound to happen .

Posted in Uncategorized | 2 Comments

Breaking the silence

Not much has happened on this blog during the last two weeks. First, because I spent some of that time travelling through the north of this amazing country. Second, and more importantly, because the kick-off of the mobile survey phase had to be postponed. Some of the things we are doing – in terms of technology – have not been done before, which is what makes this whole beast an exciting one but also sometimes tricky to predict. Especially the technical implementation of some of the data gathering modes  has proven to be more challenging than expected. By now, however, all of the more serious issues have been resolved and we will be able to start re-approaching respondents very soon. So for now, some more patience is required.

Posted in Uncategorized | 1 Comment

Finishing baseline & preparing mobile follow-ups

Having completed our 550 face-to-face baseline interviews and preparing the mobile follow-up phase, it seems about time for some first data-poking. At this point, things still are rather preliminary, mainly because we so far have entered the baseline data only for two out of the three districts of Dar es Salaam. Also, if you had a look at my earlier posts you will know that our sampling procedure (non-proportional stratified) requires weighting. But with coordinating fieldwork, solving lots of small practical issues and all that, we just haven’t had the time yet to put together the weights. So here you go. So much for the smallprint.

With heavy traffic being a major problem in the city, part of our questionnaire dealt with public transportation and the time people need to get to work.

At this point, we are looking at 265 adult respondents from two districts: Temeke & Ilala. Generally, respondents were very willing to answer our questions, and their attitude was almost always perceived as cooperative by the interviewers. As I mentioned in an earlier post, this was also very much due to the help of the balozi who introduced the interviewer to the head of the household and expressed his/her support for the survey, thus reducing fears and suspicion amongst respondents.

The mean age is 35, and women are overrepresented with 59%. In 42.3% of the cases had we selected the head of the household, in 31.8% his/her spouse followed by another 13.1% adult children of household heads.

Garbage dump in western outskirts of Dar es Salaam.

The face-to-face baseline survey focusses on the provision of a number of public services such as water, sewage, garbage disposal, schools and health facilities (click here for the translated questionnaire). Even though this blog is about the mobile phone follow-up part, I will first throw some of these numbers (and some graphs) at you – to get a general idea of the lives of the people that make up our sample.

Looking at education levels, 11.7% of our respondents never went to school and 23.4% had attained any form of education beyond primary school (which goes up to about 14 years). 29.3% reported to be unemployed at the time of the interview.

Lack of cash income / food during previous month (n=265, percentages on y-axis)

For the majority of respondents the main source of water is either a covered bore hole / well (32.8%) or a neighbour’s tap (32.5%). 6.4% have running water in their own building. When it comes to garbage disposal, about half of the respondents report that their garbage (sometimes) gets picked up, 30% bury it and 10% burn it themselves. As a toilet, 63.4% use a pit latrine which is mostly shared with other households and 23.0% have a flush toilet. 48.3% of the households own a television set, 74% have chairs and 83.4% have a table. 9.1% have a car, 8.7% have a computer and 2.6% a washing machine.

The first rains come to Pugu, Dar es Salaam

A clear majority of respondents (72.1%) own a mobile phone themselves, and another 18.1% have access to the phone of another household member or friend. Of those with a mobile phone, about 60% had credit at the time of the interview. This leaves us with 9.8% without phone ownership or access – constituting the most obvious and serious source of selection bias for the mobile follow-up. Particularly, since we can already see that this group is indeed somewhat different from the rest: less educated, more often female and slightly older. Most of these differences are not (yet) significant, but this is one of the things we will continue to monitor very carefully when the rest of the data comes in – to estimate the size of the problem (bias) but also think about possible solutions (e.g. handing out phones).

Frequency of media usage (n=265, percentages on y-axis)

One thing that is giving us a bit of a headache at the moment is that only very few respondents actually have the skills and technology to use WAP to answer our mobile surveys. To be sure, this is not problematic in itself. After all, we are not favouring one technology over another just because it looks fancier on paper. In an ideal situation, respondents are approached through a channel which they are familiar and comfortable with, to decrease data entry errors and drop-out rates. However, as part of our interest (and my interest in particular) is in the methodological aspects of it all, we would like to have a decently sized group of WAP respondents so we can compare their response behaviour – or lack thereof – with those using other technologies. But again, as we are still awaiting about half of the data, much of which coming from better-off areas, there is no reason to be overly pessimistic about this.

Right now we are ironing out the last issues regarding our different mobile data gathering channels – WAP, IVR, USSD, Call Center – solving technical problems and rounding off negotiations with mobile providers.

If all this runs smoothly, we will start our first round of mobile data gathering at the end of this week – so things are bound to get pretty exciting here once again.

Posted in call center, IVR, nonresponse, poverty, sampling, USSD, WAP | 1 Comment

All about context

Before coming to Tanzania, I had read about the spread of mobiles in developing countries; I had listened to people giving talks about ICT’s potential for fostering change and fighting poverty; and I had discussed ICTs for development and that strange notion of “leapfrogging” with my students.

But still. Before I came here, I really didn’t have a sense of how rapid and unprecedented the adoption of mobile phone technology has been and what this means for peoples’ lives in a country like Tanzania.

Up until 2004, when about 70% of the Western world was already using mobile phones, no more than 5% of Tanzanians did. But then, starting in 2005 (when a mobile licensing framework opened up the market) the numbers increased dramatically and haven’t stopped doing so since. Currently, the total “mobile penetration rate” (calculated as the number of active SIM cards / total population) is somewhere around 40%, with the amount of active SIM cards growing with about 60% each year. In the first half of 2009, almost 1,000 new SIM cards were bought per day.  On average, it is estimated that mobile phone users spend 15% of their income on airtime and there is reason to assume that this percentage is actually higher amongst the poor (about one third of the population lives below the national poverty line of 500Tshs or $0.33 per day) .

To put this into some perspective: this is a country where 6% of households own a television set, about 1% of people have a PC, and – think about this – on average 15% have access to electricity (this drops to 2% for rural areas). Landlines never played a big role and there is no reason to believe that this is about to change – telephone lines per 100 people actually dropped from 0.5 to 0.3 between 2000 and 2008. As to the Internet: it is true that the arrival of undersea fiber optic cables in East Africa has improved bandwidth access tremendously, but this currently is of no significance to the overwhelming majority of the population (percentage of internet users: 1%).

So by any standard, it is the mobile phone that has become the medium of choice to keep contact, to socialize and to communicate with relatives too far away for travel. And not just in the cities, but increasingly also in the rural areas. Particularly here, phone sharing communities have become a common phenomenon, and some have started to run micro scale call centers, renting out their mobile phones. But there is more.

Mobiles are now also used to pay bills and transfer money by those who are excluded from formal financial services. In Kenya – where in 2007, only 19% of the population had access to banks – mobile phone provider Safaricom came up with M-Pesa, a system that allows people to transfer cash to anywhere in the country. Migrant workers use it to send money to their families back home,  farmers use it to buy cattle, employers use it to pay wages. And it’s big indeed: in 2009, the amount of money that was transferred through M-Pesa was equivalent to 11% of Kenya’s GDP. It has now extended its services to allow for money transfer to and from the UK and is now also operating in Tanzania, Afghanistan and South Africa. It’s the first time that those described as the “unbanked”, have a way to get cash from A to B in a safe and cheap manner. And all you need is a phone.

When driving through the country, what also strikes me is the amount of mobile phone shops everywhere. No matter where you are, you will always find a place where you can get an airtime voucher from all the four major mobile providers (no contracts, all prepaid of course). Vouchers are available in very small units and because calls between networks are disproportionately more expensive than within networks, it is quite common for people to have two, three, sometimes four different sim cards, i.e. numbers (and for those that are better-off this often also means two or three phones).

ICT indicators for Tanzania (World Bank data). As always when we use quantitative indicators to describe the world, it is worth looking at what is counted and why. Note that in this graph, the mobile penetration is displayed, i.e. the total number of subscriptions (active SIM cards) per 100 people. It should be understood that this does not necessarily tell us what percentage of people are actually using mobile phones, mainly due to the practice of phone sharing and the usage of multiple numbers – the former leading to an underestimation and the latter to an overestimation of the actual figure. Also, it has been argued that when comparing these numbers internationally, what should be taken into account is the demographic structure of the countries at hand. In Tanzania, 45% of the population is younger than 15 years (Netherlands: 18%). Since we don’t expect young children to use mobile phones, a good point can be made that the number of active SIM cards should actually be divided only by size of the adult population, obviously leading to very different figures altogether.

So what to take home from all this? First, mobile phone usage will continue to grow in East Africa, as it will in most other developing regions.  And second, people will use it in the manners and for the purposes that make sense in their lives. There should be no expectation that somehow the developing world is to follow into the footsteps of the developed countries, just every now and then boldly “leapfrogging” a step along the way.

And phone manufacturers have long understood the need to tailor to this. The cheapest Nokia model that you can get here (the 1012 which just about everyone– including myself – seem to own) looks like a regular low-end phone, similar to the first/second generation Nokia devices that were sold in Europe: sturdy case, small monocolor display, no WAP, no WLAN, no radio, and a battery life that will push smartphone users into episodes of sentimental weeping.

But spending some time going through its menus, there is more: it allows you to assign simple icons to contacts in your phone book (girl with short hair, girl with long hair, guy with beard, car, palm tree, you get point), to help those who can’t read to find the right number; it has multiple phone books and allows to set a cost/time limit for the next call, making phone sharing very easy; and the best of it all: it comes with a little torch (so far my excitement and joy of this has not been adequately shared, so please feel free to do so).

The International Telecommunications Union (ITU) has called mobile phones the most rapidly adopted technology in history, and they might have a point (after all, they should know).  However, we surely shouldn’t expect the adoption of mobile phones to taken on the same shape and color all across the world. After all, the process of appropriating technology to people’s lives has always above all been characterized by much creativity and thus a whole lot of unpredictability.

Posted in mobile phone uses, poverty | 1 Comment

A word on technologies

We have been in the field for about two weeks and finished interviewing in the district of Temeke a couple of days ago (with Ilala & Kinondoni still to go). Once the face-to-face baseline interviews are rounded off, we will approach all respondents again through their mobile phones for the short weekly set of questions. In an earlier post, I gave a brief  overview of the four technologies that we will be using for the mobile phone follow-up surveys: Call Centre, USSD, WAP and IVR (voice-based menu), and the procedure to assign respondents to these different technologies. In this post I will share some experiences regarding those technologies in the run-up phase to the actual mobile data gathering.

This is the procedure we will follow to contact the respondents of the different groups during the weekend. In the text message on Friday morning, respondents are asked to complete the questionnaire before Sunday evening. The incentive (300, 400 or 500 Tshs) will be transferred to the respondent's phone after filling in the survey. However, the IVR and WAP group will receive the first 100 Tshs as an advance alreay on Friday morning so that lack of credit will not keep them from beeping the IVR system (call and hang up) / browsing to the WAP survey page.

As I explained earlier, our aim is to end up with about the same amount of respondents in each technology condition in order to nicely estimate the effects of the data gathering channel on panel mortality, data quality, etc. (just like we randomized the amount of incentives to test effects there). So far, we have met with two possible difficulties here.

First, in the areas that we surveyed until now, the penetration rates of WAP phones and the proportion of respondents able to use them are much lower than expected. The good news is that since we are now moving from the poorer into the better-off areas of the city, the WAP rates in our sample will probably increase in the weeks to come. However, if WAP skills / availability prove to be mainly a function of household welfare, estimating the method effects here will become a whole lot trickier. But it’s still too early to draw any conclusions about this as yet.

On a more technical note, we have made a switch in technology for the WAP surveys. Initially, we were planning to use an application that would be downloaded from our server and run on the respondent’s phone. Our colleagues at Data Vision had developed a beta version of the application and it all looked pretty neat in the testing phase. However, the application would have to be downloaded again each week (to update the questions) and also the installation process opened up room for things to go wrong, especially for respondents who are less technological savvy. We therefore decided to simply have respondents navigate to a WAP (WML) website with a form where the questions can be answered (very much like a regular website with questions and bullet list / pull-down menus).

The second possible obstacle that we are encountering is that it is still unclear when and under what conditions we will be able to use USSD (Unstructured Supplementary Service Data, see the first few pages of this document for a useful introduction and also have a look at the USSD paragraph of this earlier post).

In my opinion, USSD is an extremely promising technology for what we are trying to do. Firstly, and possibly most importantly, mobile phone users in East Africa know and widely use USSD for paying bills and making financial transactions (see for example M-Pesa). Secondly, USSD will work on any mobile phone without requiring installation or configuration. Furthermore, it is cheap and the communication channel can be established by both the user and the application. As things look now, the bottleneck seems to be that you need to partner up with the mobile providers, since USSD connections are always connections to the home network servers. However, things aren’t looking too bad on this front either.

When searching for more on the uses of USSD for similar projects (there is not a lot), I was happy to see that also Johan Hellström in his insightful report on “The Innovative Use of Mobile Applications in East Africa” mentions USSD as having great potential (especially for those simple request-response applications like ours) and also concludes that the technology has not been used to its full mainly because “its implementation requires close collaboration with the operator”. What seems to be absolutely vital here, is to get mobile providers on board in a very early stage.

For now, we are still quite confident that our group of WAP respondents will fill up eventually and that USSD will be up and running once we start approaching our mobile phone panel.

Posted in call center, IVR, USSD, WAP | Leave a comment

On the map

Click to load in Google Maps

To keep myself occupied while impatiently waiting for the first wave of baseline survey data to appear in my inbox yesterday, I put together this map of the wards of Dar es Salaam that are included in our sample (see our sampling procedure). In each selected ward, between 15 and 30 interviews are conducted, depending on the number of selected mtaas within each ward. Click on the image to load the map in Google Maps.

Posted in dar es salaam, sampling, visualization | 2 Comments