Operating in four regions, the logistics of this project have been challenging.  The timing of our survey made it worse, as the wet season is a bad time for roads.  It has been difficult for our management team to get to our field offices, and it has been difficult for our surveyors to get to some respondents.

Bole was the most difficult location to get to.  Normally it’s a 5 hour ride on Metro Bus.  Shortly after our survey began, they closed the road between Tamale and Bole, and routed transportation through Kantampo.  This meant that to get to Bole, instead of a 5 hour ride, it was now an 8 hour ride, with two bus transfers.  And the bus ALWAYS left late.

Salaga had the most problems with inaccessible respondents.  At one point, the survey team left to travel to a community, got nearly there before the road became impassable, and had to turn back.  A wasted day.  Salaga also had some respondents accessible only by boat, but high waters made the journey too dangerous.  Since these respondents would likely be unable to get the treatment we are testing anyway, we replaced them in the sample.

Salaga wasn’t always a breeze to get to, either.  The road seemed to disintegrate a little more every trip I made there.  The last trip there, Metro Bus wasn’t running because of road conditions, so I took a tro-tro.  I estimate about 10% of the road was water on that trip.  We were traveling at night, and there was a rain storm.  Since the tro-tro had no glass in the windows, I spent the duration of the storm holding a plastic sheet up to shield myself and fellow passengers from the rain.  After the rain stopped, the challenges continued, as all that rain flooded the road.  At times it seemed like we were driving in a river.  At one point, we had to get out of the tro-tro and trek through 100 meters of knee-deep water, so that the tro-tro could go through empty to avoid getting stuck in the mud.  A few of the Ghanaians took great fun from telling me “Welcome to Africa!”  Ironically, because I had traveled that road eight times in the past month, I was probably less surprised by the road conditions than they were. 

Donald Marron recently bloggedabout a new economics paper on gender arbitrage by multinationals in South Korea. The idea behind gender arbitrage is that discrimination in hiring against a particular group, like women or minorities, creates opportunities for non-discriminating employers to hire talented people for a lower wage. When non-discriminating employers take advantage of this, it should eventually erase the gap in wages between the disadvantaged group and the rest of the labor market. This paper found that multinational corporations have been able to benefit from discrimination against women in the labor market that drives down wages for educated women.  In Korea, working women earn only 63% of what working men do. (Not all of this is due to discrimination.) The paper found that among multinationals, a 10 percentage point increase in the number of women in local management positions led to a 1 percentage point increase in return on assets.  Marron points out that the fact that companies that hire more women have a hire profit margin means that there is still room for more arbitrage-- implying that discrimination is still resulting in lower wages for women compared with men who have the same skills and abilities.

As unfortunate as it is that women in Korea are being paid less than they are worth, from the perspective of both women and employers in northern Ghana, this is an enviable problem.  In Ghana as a whole, about 20% of adult males have secondary education or higher; only about 10% of adult females have that level of educational attainment (source: GLSS 5), and the gender gap is most pronounced in the Northern Region.  Traditional views of gender roles still prevent girls from having access to education at the same rate as boys.  (Girls may also have a higher opportunity cost of education: girls are often more economically valuable than boys, because they can assist with child-rearing and food processing, or work as maids, at an age where boys are still too young to be much help with farm work.)  The result of this is that it is difficult to find qualified female candidates for jobs requiring a high level of education.

This is especially apparent to employers like me, who actually have a bias in favor of female employees.  Since the majority of the respondents in my survey were female, I wanted to hire female surveyors because they are more likely to put female respondents at ease.  Despite actively recruiting female candidates, posting notices encouraging women to apply, and asking the field managers to try to achieve a balance in the number of male and female surveyors we hired, we received few applications from female candidates, and less than a quarter of the surveyors we hired ended up being female.

One thing my survey sought to measure was the respondent’s perceived risk of falling ill.  We expected this to be tricky, because many of the respondents are not highly number literate, and most have no concept of probability.  This was easily the most challenging part of our survey to design.

We tried to model our question after surveys that had been done in other developing economies, as detailed in Delevande, Gine and McKenzie.  These surveys asked respondents to allocate piles of stones to different pots, depending on how likely they thought different amounts of rainfall were.  The advantage of this is that the respondent does not even need to be able to count, he or she can just compare quantities, and see that one is larger.  Also, the respondent can easily make marginal changes, moving a few stones to a different pot, if you change the conditions of the questions slightly.  We adapted our questions to ask respondents the probability they would get sick over the next month.

What sounded like a simple and elegant model turned out to be an incredible struggle.  The responses we got failed basic logic tests as often as not; for example, respondents would predict a higher risk of getting sick once in the next month than once in the next year. After rounds and rounds of piloting, and consulting everyone we could think of, we identified the major problems, and attempted to address them:

1.       People don’t like to say they will get sick, because they think it will then happen.  We addressed this by changing our question to ask about the probability someone as healthy as them would get sick over the next month.  This seemed to help.

2.       “Chance” and “likelihood” don’t translate well in Dagbani.  We did our best to write the questions in English so that they would be as simple as possible, and then carefully worked through the Dagbani to find the clearest translation.

3.       People have little concept of marginal changes in probability.  In northern Ghana, something is either certain to happen, certain not to happen, or may happen (50-50).  Our weather reports, forecasting 40% chance of rain, would make no sense here.  Conceptualizing one thing as more likely than another, or conceptualizing small changes in risk, is completely counter-intuitive.  We never completely solved this problem.

4.       People don’t like stones.  Natural objects, like stones or beans, are associated with witches’ fortune-telling, which people don’t like.  And here we were asking them to predict future sickness with stones.  We switched to bottle caps, which as a man-made object, are less threatening.  My friends in Tamale drank beer very diligently and enthusiastically in the weeks leading up to my survey in order to provide the 720 bottle tops I needed. 

As a result of continuing problems with #3, this section proved to be the most difficult in our survey.  It was only a few questions long, but the time it took to explain the concept, go through examples with the respondent, and work through the questions added up to nearly 30 minutes, almost a third of the total survey.

This week, my team will be completing our baseline survey.  This is quite a landmark achievement for us.   We pioneered a team structure my organization has not used in this region before, relying on talented temporary staff to serve as field managers overseeing survey operations in each of the four regions we worked in.  The project manager and I rotated between the locations, monitoring progress and paying surveyor salaries.  We administered a 1.5 hour survey to 1500 respondents, in at least 7 languages, using 60 surveyors.  We did it in a month, not counting our piloting and census.

During the survey, I have taken a hiatus from blogging, despite the fact that my survey has generated some material worth sharing.  For the next couple weeks, expect to see stories about my surveying experience:  the Survey Chronicles.