In July I had the fortune of visiting my sister in Phnom Penh. I was the instructor for a 3-week workshop on data processing using CSPro, the Census and Survey Processing System. The U.S. Census Bureau, where I work, distributes the software freely around the world, and it has been used to process census data in many countries, as well as thousands of surveys, including the Demographic and Health Surveys. I conducted my UNFPA-sponsored workshop at the National Institute of Statistics. The country conducted a census in 2008 but NIS is already preparing for the 2013 intercensal survey. They will use CSPro to enter, edit, and tabulate the data for this survey.
Cambodia is the first country where I have worked that gives participants “sitting fees.” That is, UNFPA supplements the salary of NIS staff attending the workshop. This supplement can make a big difference in take-home salary, almost doubling the salaries of some workers. I do not know whether or not this policy is ultimately necessary or good, but it seemed to ensure that I had a large group of participants. Most participants were already quite advanced using CSPro, so the level of the workshop was higher than typical for me, and the participants were very engaged. It was a delightful experience, and interacting with typical Cambodian urban workers gave me a perspective on the country different from one I would have received if I had been a Vietnam-Cambodia-Thailand backpacker.
Most countries conduct a census every 10 years, and the data from the census is of vital importance to the country. The data is used for planning purposes (NIS is located within the Ministry of Planning), for safety purposes, and for projecting future growth. For example, if a typhoon hits a part of the country, the statistics office can look at census data and quickly identify the number of people affected. Or an ambitious private company, say a bicycle retailer, can look at census data, identify the characteristics of bicycle owners, and using population projections, identify where future bicycle stores should be located. (More than 60% of Cambodian households own a bicycle.)
When creating a questionnaire, a statistics office generally relies on questions from the past census (for time series analysis), United Nations recommendations, and regional concerns. The questions on a census can tell you a lot about a country. For example, Bangladesh asks households if they own a sewing machine; Cambodia asks about ownership of a hand tractor; Zambia asks questions about albinism.
In the U.S. last year, some vocal conservatives were very wary about filling out the census form, interpreting the questions as an attempt by Barack Obama to gain unwelcome access to information about each household. These conservatives forgot that the legality of the census is enshrined in their sacred document, the Constitution; the phrasing is displayed prominently at the Census Bureau headquarters in Suitland, Maryland. There is much less trepidation about answering census questions in most other countries. In Cambodia each household was asked to give details about deaths in the household in the past year. Possible response categories included, among others: diarrhea, malaria, HIV/AIDS, and landmines. In the U.S. we can learn about causes of death from hospital records, but that is not possible in many places. Such questions may seem prying, but in countries without good administrative records, the census or a survey is the best way to get information on the country’s situation.
A user of statistical data must always be careful when drawing conclusions from data, whether it comes from a census or a survey. Error can be introduced because of bad question phrasing, respondent error, a mistake in marking the answer on the questionnaire, a misread of the answer by a keyer or scanner, and errors introduced by edits or during tabulation.
I will conclude by discussing error introduced because of the rules of a survey. In 2008 I was vacationing in the Ukraine. I drove from Kiev to the Crimea and was stopped twice by police. (I was stopped a third time when my friend was driving, coincidentally by the same police officer who had previously stopped me.) Instructed to by my Ukrainian friend, I bribed one of the police officers, giving him about $20, and we were let off with a warning.
Later that year, back in the U.S., I was selected to take part in a Bureau of Justice Statistics survey that included questions about interactions with police. One of the questions asked about intimidation by the police in the last year. One of the police officers had threatened to my Ukrainian friend that he could easily harm us (the foreigners). I asked the phone interviewer about whether I should discuss foreign experiences, and the interviewer looked at the survey rules and said that the rules indicated that all encounters should be discussed. This may be reasonable because probably 95% of people interviewed did not leave the U.S. in the previous year. But now in the survey there is a record of an American being threatened with bodily harm by the police, with the result that a bribe was given. Hopefully there is a note that this encounter occurred overseas, but based on my conversation with the interviewer, I highly doubt this.
Before jumping to conclusions based on data, make sure that the data was collected accurately and that the questions asked, and the way that enumerators asked the questions, result in proper answers from respondents.